Air Quality
Monitoring and
Forecasting
Pius Lee, Rick Saylor and Je McQueen
www.mdpi.com/journal/atmosphere
Edited by
Printed Edition of the Special Issue Published in Atmosphere
atmosphere
Books
MDPI
Air Quality Monitoring and Forecasting
Special Issue Editors
Pius Lee
Rick Saylor
Jeff McQueen
MDPI • Basel • Beijing • Wuhan • Barcelona • Belgrade
Books
MDPI
SpecialIssueEditors
PiusLee,RickSaylorandJeffMcQueen
NOAA
USA
EditorialOffice
MDPI
St.Alban‐Anlage66
Basel,Switzerland
Thised
itionisarepr
intoftheSpecialIssuepublishedonlineintheopenaccess
journalAtmosphere(ISSN2073‐4433)from2017–2018(availableat:
http://www.mdpi.com/journal/atmosphere/special_issues/air_monitoring).
Forcitationpurposes,citeeacharticleindependentlyasindicatedonthearticle
pageonlineandasindicatedbelow:
Lastname,F.M.;Lastname,F.M.
Articletitle.JournalNameYear,Articlenumber,
pagerange.
FirstEdition2018
ISBN978‐3‐03842‐839‐8(Pbk)
ISBN978‐3‐03842‐840‐4(PDF)
A
r
t
icles in this volume are Open Access and distributed under the Creative Commons Attribution
license (CC BY), which allows users to download, copy and build upon published articles even for
commercial purposes, as long as the author and publisher are properly credited, which ensures
maximum dissemination and a wider impact of our publicat
ions
. The book taken as a whole is
© 2018 MDPI, Basel, Switzerland, distributed under the terms and conditions of the
CreativeCommonslicenseCCBY‐NC‐ND(http://creativecommons.org/licenses/by‐nc‐nd/4.0/).
Books
MDPI
Table of Contents
About the Special Issue Editors ..................................... v
Pius Lee, Rick Saylor and Jeff McQueen
Air Quality Monitoring and Forecasting
doi: 10.3390/atmos9030089 ...................................... 1
Wei Lu, Tinghua Ai, Xiang Zhang and Yakun He
An Interactive Web Mapping Visualization of Urban Air Quality Monitoring Data of China
doi: 10.3390/atmos8080148 ...................................... 3
Tiancai Zhou, Jian Sun and Huan Yu
Temporal and Spatial Patterns of China’s Main Air Pollutants: Years 2014 and 2015
doi: 10.3390/atmos8080137 ...................................... 19
Baolei Lyu, Yuzhong Zhang and Yongtao Hu
Improving PM
2.5
Air Quality Model Forecasts in China Using a Bias-Correction Framework
doi: 10.3390/atmos8080147 ...................................... 34
Hui Zhao, Youfei Zheng and Ting Li
Air Quality and Control Measures Evaluation during the 2014 Youth Olympic Games in
Nanjing and its Surrounding Cities
doi: 10.3390/atmos8060100 ...................................... 49
Casey D. Bray, William Battye, Pornpan Uttamang, Priya Pillai and Viney P. Aneja
Characterization of Particulate Matter (PM
2.5
and PM
10
) Relating to a Coal Power Plant in
the Boroughs of Springdale and Cheswick, PA
doi: 10.3390/atmos8100186 ...................................... 61
Samuel D. Lightstone, Fred Moshary and Barry Gross
Comparing CMAQ Forecasts with a Neural Network Forecast Model for PM
2.5
in New York
doi: 10.3390/atmos8090161 ...................................... 74
Rodrigo Munoz-Alpizar, Radenko Pavlovic, Michael D. Moran, Jack Chen, Sylvie Gravel,
Sarah B. Henderson, Sylvain M´enard, Jacinthe Racine, Annie Duhamel, Samuel Gilbert,
Paul-Andr´e Beaulieu, Hugo Landry, Didier Davignon, Sophie Cousineau and
V´eronique Bouchet
Multi-Year (2013–2016) PM
2.5
Wildfire Pollution Exposure over North America as
Determined from Operational Air Quality Forecasts
doi: 10.3390/atmos8090179 ...................................... 90
George M. Woodall, Mark D. Hoover, Ronald Williams, Kristen Benedict, Martin Harper,
Jhy-Charm Soo, Annie M. Jarabek, Michael J. Stewart, James S. Brown, Janis E. Hulla,
Motria Caudill, Andrea L. Clements, Amanda Kaufman, Alison J. Parker, Martha Keating,
David Balshaw, Kevin Garrahan, Laureen Burton, Sheila Batka, Vijay S. Limaye,
Pertti J. Hakkinen and Bob Thompson
Interpreting Mobile and Handheld Air Sensor Readings in Relation to Air Quality Standards
and Health Effect Reference Values: Tackling the Challenges
doi: 10.3390/atmos8100182 ......................................114
iii
Books
MDPI
Daniel-Eduard Constantin, Alexis Merlaud, Mirela Voiculescu, Carmelia Dragomir,
Lucian Georgescu, Francois Hendrick, Gaia Pinardi and Michel Van Roozendael
Mobile DOAS Observations of Tropospheric NO
2
Using an UltraLight Trike and
Flux Calculation
doi: 10.3390/atmos8040078 ......................................140
Richard M´enard and Martin Deshaies-Jacques
Evaluation of Analysis by Cross-Validation. Part I: Using Verification Metrics
doi: 10.3390/atmos9030086 ......................................153
Richard M´enard and Martin Deshaies-Jacques
Evaluation of Analysis by Cross-Validation, Part II: Diagnostic and Optimization of Analysis
Error Covariance
doi: 10.3390/atmos9020070 ......................................169
Barry Baker and Li Pan
Overview of the Model and Observation Evaluation Toolkit (MONET) Version 1.0 for
Evaluating Atmospheric Transport Models
doi: 10.3390/atmos8110210 ......................................190
iv
Books
MDPI
About the Special Issue Editors
Pius Lee received his Ph.D. in Atmospheric and Ecological Science in 1994 under the supervision
of Professor Hiroshiro Kitada of the Toyohashi Technical University, Japan. Dr. Lee is an adjunct
professor in the Center for Spatial Information Science and Systems in George Mason University, VA
USA. Dr. Lee is leading the NOAA Air Resources Laboratory project supporting the research and
operational eligibility testing of the US National Air Quality Forecasting Capability (NAQFC). Dr.
Lee and Co-Editor McQueen started the NAQFC team in 2003 and has since been instrumental in
advancing the science, scope of service and accuracy of NAQFC.
Rick Saylor is a Physical Scientist with the National Oceanic and Atmospheric Administration
(NOAA) Air Resources Laboratory, Atmospheric Turbulence and Diffusion Division, in Oak Ridge,
Tennessee, and an Associate Research Professor at the University of Tennessee in the Department
of Civil and Environmental Engineering in Knoxville, Tennessee. He obtained a B.S., M.S. and
Ph.D. in Chemical Engineering at the University of Kentucky and has over 28 years of experience
as a researcher in atmospheric chemistry and air quality. His expertise includes analysis and
interpretation of field measurement data of trace gases and particulate matter and modeling of
atmospheric physics and chemistry across all scales from small-scale process models to regional- and
global-scale 3-D models.
Jeff McQueen has been NOAA/NWS National Center for Environmental Prediction Air Quality
and dispersion modeling program leader since 2003. Mr. McQueen leads efforts to develop, test,
implement and maintain the NOAA air quality and dispersion modeling systems. Mr. McQueen
and his staff have established a national capability for ozone and particulate matter prediction by
driving EPA’s CMAQ model with atmospheric fields from the NCEP North American Model (NAM)
fields. Mr. McQueen has also chaired the Weather Research and Forecasting (WRF) Model Science
Board and is a member of the WRF Model Atmospheric Chemistry and Ensemble Working Groups
and OFCM Atmospheric Transport and Dispersion Joint Action Group. He currently co-chairs the
Aerosols and Atmospheric Composition Unified Modeling Strategic Implementation Plan Working
Group for NWS. Mr. McQueen’s research interests include atmospheric modeling and use of physical
parameterizations and coupled models (e.g., boundary layer, land-surface) to quantify atmospheric
dispersion uncertainties.
v
Books
MDPI
atmosphere
Editorial
Air Quality Monitoring and Forecasting
Pius Lee
1,2,
*, Rick Saylor
1
and Jeff McQueen
3
1
NOAA/Air Resources Laboratory, NOAA Center for Weather and Climate Prediction,
College Park, MD 20740, USA; rick.saylor@noaa.gov
2
Center for Spatial Information Science and Systems, George Mason University, Fairfax, VA 22030, USA
3
NOAA/NCEP/Environmental Modeling Center, NOAA Center for Weather and Climate Prediction,
College Park, MD 20740, USA; jeff.mcqueen@noaa.gov
* Correspondence: pius.lee@noaa.gov
Received: 27 February 2018; Accepted: 28 February 2018; Published: 1 March 2018
Air quality forecasting is a vital tool for local health and air managers to make informed decisions
on mitigation measures to reduce public exposure risk. Given a forecast of impending poor air quality,
air quality managers may issue car-pooling advisories, authorize free public transportation or impose
other mitigation and warning measures. Air composition monitoring and exposure records can inform
long-term trends of major air pollutants and their health impacts. Epidemiologists use long term
composition data to understand air pollution related diseases and mortality rates to support public
health policies. This Special Issue highlights the interplay and co-benefit of air quality monitoring
and forecasting.
Public health is under a constant threat by air pollution across the world in various degrees and
manifestations. In China, rapid economic growth has resulted in increased occurrences of poor air
quality. In this special issue, Lu et al. [
1
] of Wuhan University and Zhou et al. [
2
] of Chengdu University
respectively studied urban haze and the distribution of multiple pollutants in China. Lyu et al. [
3
]
of Tsinghua University and Georgia Institute of technology advanced particulate matter forecasts in
China. Zhao et al. [
4
] of Nanjing University studied the strong response of emission controls during a
recent Youth Olympics event in Nanjing, China.
Ground truth of air constituent concentrations is determined by measurements. Woodall et al. [
5
]
of the US EPA conducted an intriguing study about hand held air composition measurement devices.
Constantin et al. [
6
] of the University of Galati, Romania, used an ultralight trike and flux calculations
to measure nitrogen dioxide vertical column density. Bray et al. [
7
] of North Carolina State University
characterized pollutants emitted from coal-fired power plants in Eastern USA. Baker and Pan [
8
]of
NOAA’s Air Resources Laboratory, developed a software tool which utilized many in-situ and surface
monitoring network measurements to evaluate forecast model performance. Lightstone et al. [
9
]of
City College of New York explored neural networks as a means for air quality forecasts. Environmental
and Climate Change Canada’s Munoz-Alpizar et al. [
10
] studied the impact wildfire pollution on
public health, and Ménard and Deshaies-Jacques [
11
,
12
] analyzed chemical data evaluations by
cross-validation statistical analysis.
It is clear that air pollution remains a global problem and that air quality monitoring, forecasting
and mitigation begins as a local effort conducted in concert with global partners. The articles selected
in this Special Issue speak volumes to this fact many times over across the globe. We thank the editing
office for their excellent support to realize this herculean achievement to collect and publish the cutting
edge articles in this issue.
Conflicts of Interest: The authors declare no conflict of interest.
Atmosphere 2018, 9, 89 1 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2018, 9,89
References
1.
Lu, W.; Ai, T.; Zhang, X.; He, Y. An Interactive Web Mapping Visualization of Urban Air Quality Monitoring
Data of China. Atmosphere 2017, 8, 148. [CrossRef]
2.
Zhou, T.; Sun, J.; Yu, H. Temporal and Spatial Patterns of China’s Main Air Pollutants: Years 2014 and 2015.
Atmosphere 2017, 8, 137. [CrossRef]
3.
Lyu, B.; Zhang, Y.; Hu, Y. Improving PM
2.5
Air Quality Model Forecasts in China Using a Bias-Correction
Framework. Atmosphere 2017, 8, 147. [CrossRef]
4.
Zhao, H.; Zheng, Y.; Li, T. Air Quality and Control Measures Evaluation during the 2014 Youth Olympic
Games in Nanjing and its Surrounding Cities. Atmosphere 2017, 8, 100. [CrossRef]
5.
Woodall, G.M.; Hoover, M.D.; Williams, R.; Benedict, K.; Harper, M.; Soo, J.-C.; Jarabek, A.M.; Stewart, M.J.;
Brown, J.S.; Hulla, J.E.; et al. Interpreting Mobile and Handheld Air Sensor Readings in Relation to Air
Quality Standards and Health Effect Reference Values: Tackling the Challenges. Atmosphere
2017
, 8, 182.
[CrossRef] [PubMed]
6.
Constantin, D.-E.; Merlaud, A.; Voiculescu, M.; Dragomir, C.; Georgescu, L.; Hendrick, F.; Pinardi, G.;
Van Roozendael, M. Mobile DOAS Observations of Tropospheric NO
2
Using an UltraLight Trike and Flux
Calculation. Atmosphere 2017, 8, 78. [CrossRef]
7.
Bray, C.D.; Battye, W.; Uttamang, P.; Pillai, P.; Aneja, V.P. Characterization of Particulate Matter (PM
2.5
and
PM
10
) Relating to a Coal Power Plant in the Boroughs of Springdale and Cheswick, PA. Atmosphere
2017
,
8, 186. [CrossRef]
8.
Baker, B.; Pan, L. Overview of the Model and Observation Evaluation Toolkit (MONET) Version 1.0 for
Evaluating Atmospheric Transport Models. Atmosphere 2017, 8, 210. [CrossRef]
9.
Lightstone, S.D.; Moshary, F.; Gross, B. Comparing CMAQ Forecasts with a Neural Network Forecast Model
for PM
2.5
in New York. Atmosphere 2017, 8, 161. [CrossRef]
10.
Munoz-Alpizar, R.; Pavlovic, R.; Moran, M.D.; Chen, J.; Gravel, S.; Henderson, S.B.; Ménard, S.; Racine, J.;
Duhamel, A.; Gilbert, S.; et al. Multi-Year (2013–2016) PM
2.5
Wildfire Pollution Exposure over North America
as Determined from Operational Air Quality Forecasts. Atmosphere 2017, 8, 179. [CrossRef]
11.
Ménard, R.; Deshaies-Jacques, M. Evaluation of analysis by cross-validation. Part I: Using verification
metrices. Atmosphere 2017, 8, 86.
12.
Ménard, R.; Deshaies-Jacques, M. Evaluation of analysis by cross-validation. Part II: Diagnostics and
optimization of analysis error covariance. Atmosphere 2017, 8, 70.
©
2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
2
Books
MDPI
atmosphere
Article
An Interactive Web Mapping Visualization of Urban
Air Quality Monitoring Data of China
Wei Lu , Tinghua Ai *, Xiang Zhang and Yakun He
School of Resource and Environmental Sciences, Wuhan University, Wuhan 430079, China;
whuluwei@whu.edu.cn (W.L.); xiangzhang@whu.edu.cn (X.Z.); hyk1990@whu.edu.cn (Y.H.)
* Correspondence: tinghua_ai@whu.edu.cn; Tel.: +86-139-0863-9199
Received: 7 July 2017; Accepted: 10 August 2017; Published: 13 August 2017
Abstract:
In recent years, main cities in China have been suffering from hazy weather, which is
gaining great attention among the public, government managers and researchers in different areas.
Many studies have been conducted on the topic of urban air quality to reveal different aspects of
the air quality problem in China. This paper focuses on the visualization problem of the big air
quality monitoring data of all main cities on a nationwide scale. To achieve the intuitive visualization
of this dataset, this study develops two novel visualization tools for multi-granularity time series
visualization (timezoom.js) and a dynamic symbol declutter map mashup layer for thematic mapping
(symadpative.js). With the two invented tools, we develops an interactive web map visualization
application of urban air quality data of all main cities in China. This application shows us significant
air pollution findings at the nationwide scale. These results give us clues for further studies on air
pollutant characteristics, forecasting and control in China. As the tools are invented for general
visualization purposes of geo-referenced time series data, they can be applied to other environmental
monitoring data (temperature, precipitation, etc.) through some configurations.
Keywords:
air quality; environmental data visualization; spatial-temporal visualization;
visual analytics
1. Introduction
Recent years, in China, frequent occurrences of hazy weather in big cities have aroused
great attention on urban air quality issues among the public, government managers and academic
researchers. Discussions about air quality of big cities (like Beijing, Wuhan, etc.) frequently appear
in social media. Since 2012, the government has adopted two new air quality monitoring technical
standards [
1
,
2
], then has built a real-time air quality reporting platform and an air quality forecasting
system. Because air pollution has detrimental effects on human health, vegetation, crops, etc., it has
great political, societal and economic impacts [
3
–
5
]. Therefore, with the increasing availability of
urban air quality data and the great environmental challenges we are facing, many studies have been
conducted to explore new approaches for understanding this big environmental monitoring dataset.
Environmental monitoring data can be described by multivariate time series observations
generated from geo-located monitoring stations. For our research topic, urban air quality monitoring
data consist of many air pollutant concentration values (such as fine particles, carbon monoxide, sulfur
dioxide, nitrogen oxides zone, etc.), which are reported hourly from monitoring stations fixed at
specific positions in a city. These geo-referenced time series data are an important study subject in the
areas of geovisualization and environmental science.
Multi-dimensional data visualization considering spatial distributions, temporal granularities and
multivariate thematic attributes has been an interesting question in geovisualization, as well as in the
environmental science area. In geovisualization, many studies have focused on systematic theories of
Atmosphere 2017, 8, 148 3 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2017, 8, 148
multivariate spatio-temporal data visualization [
6
]. Multiple static maps series, animation maps,
space-time cubes and self-organizing maps (SOM) are used to deal with spatio-temporal data
mapping [7,8]
. In parallel, basic graph charts, such as parallel coordinate plots (PCL), are used to show
multivariate thematic attributes. To achieve effective visualization of multivariate spatio-temporal
data, combinations of these methods are used; such as the visualization system for space-time and
multivariate patterns (VIS-STAMP) [
8
], which proposed a systematic way of joining those visualization
strategies in multiple visual views interactively. However, expertise is needed to use these visualization
systems, which are limited to domain experts. Thus, it requires some easy-to-use visualization
applications to face a wider range of users.
Visualization of environmental data is important in the area of environmental science [
9
].
Conventional technologies in the Geographic Information System (GIS) and statistical map making
skills are widely used to manage, analyze and visualize environmental data [
10
]. With the development
of web technologies, open source GIS standards and web mapping tools are adopted in the analysis
and presentation of environmental data-related studies not only for experts and managers, but also for
the public in general [
11
–
13
]. These studies mainly focus on the architecture of environmental data
visualization system building, and the visualization techniques are limited to conventional methods.
Therefore, more effective visualization tools need to be developed for visualization of the increasing
accumulated multi-dimensional environmental monitoring data. Currently, as urban air quality is an
urgent societal problem for many policy-makers and a study topic for scholars, air quality data are
gaining more attention.
Many studies have been conducted on mapping urban air pollutions mainly from two directions.
Firstly, for the distribution of monitoring stations is rarely homogeneous, and spatial interpolations
are used for pollution exposure maps [
14
], concentration maps of air pollution [
15
] and mapping
spatio-temporal trends of air pollution [
16
]. Secondly, conventional charts and modern visualization
tools are also being used for air quality data exploration [
17
] and visual analytics [
18
,
19
]. Our research
follows the second direction and aims at providing a web mapping application for visualizing a
whole year of air quality monitoring data of all cities in China with one dynamic and interactive
map. This visualization application provides freedom of configuration for users to explore this
multi-dimensional dataset from different angles. The major contributions of this paper are in three
aspects: (1) we develop two novel web client tools for interactive visualization of spatio-temporal
data; (2) we propose a mashup strategy for web mapping by combining and extending the function
of different visualization tools, which can be generalized to visualize other kinds of spatio-temporal
data, such as temperature, precipitation, etc.; (3) we implement an on-line interactive urban air quality
data visualization application that helps to explore and analyze a big air quality dataset and that puts
forward clues for further studies on air pollution.
2. Method and Data
Mapping spatio-temporal datasets is mapping changes of geographical features over time and
space [
7
]. Because the information load is huge, the interaction and dynamics [
20
] of spatio-temporal
visualization should be well designed. Multiple levels of spatial and temporal details are important
considerations in spatio-temporal visualization. Mostly, the methods that are proposed in the
literature for visualizing spatio-temporal data in a multi-scale perspective have focused on either the
spatial or temporal aspect, rather than integrating both views over multiple scales. Qiang [
21
] and
Van de Weghe [22]
presented a continuous spatio-temporal model for space-time analysis. However,
in the real visualization domain, space and time are recognized or recorded by discrete intervals,
such as the tiled map service and the periodic timekeeping system. The tiled map uses a quad tree
to represent multi-scale feature of space, while the timekeeping system uses the year, season, month,
day and hour structure to record time-based events. In this section, we propose a space and time
zooming method conforming to the tiled map service and timekeeping system. This method also
conforms to the “overview first, zoom and filter, then details on demand” process [23].
4
Books
MDPI
Atmosphere 2017, 8, 148
Mapping the time component onto an axis on 2D space or 3D space is conventional method for
time series visualization. The space-time cube [
24
] maps time to a 3D axis vertically, resulting in a 3D
trajectory for time series datasets. As the storygraph [
25
], it maps x and y coordinates of space to two
vertical axes and time to the horizontal axis, which can show trends in time series datasets. This paper
provides a cartographic method to encode the time component of spatio-temporal data. The method
separates the time component from the map space and uses the glyph map symbol to encode and
interact with time, which gives more freedom for time representation and interaction. Furthermore, all
symbols on the map are controlled by the dynamic map layer for displaying appropriate symbols at
different map zoom levels and view extents.
Environmental monitoring data, i.e., the air quality monitoring data have multidimensional
features, which can be structured as data cubes with a hierarchical structure [
26
]. Figures 1 and 2
illustrate the hierarchies of time and geographical information respectively. In this paper, the time
structure in the dashed polygon part is used for the visualization, as the instance on the right side of
Figure 1. For the geographical part, this paper focuses on the air quality of city points, each of which
has a semantic importance illustrated in the table on the right side of Figure 2. The importance level of
the city points influences their weight in dynamic map symbol selection when zooming and panning
the map.
Figure 1.
Hierarchical structure of the time dimension. The dashed polygon part is handled in this
paper. An instance of this structure is on the right side.
Figure 2.
Hierarchical structure of the geographical dimension. In this paper, we focus on the city
level, and the importance weight is in the table on the right side.
5
Books
MDPI
Atmosphere 2017, 8, 148
In the following subsections, the technologies and framework of the mapping application and
the design and development of two JavaScript visualization tools, timezoom.js and symadaptive.js,
are discussed. The two visualization components are mainly based on the general purpose web
mapping library leaflet.js and the data visualization library D3.js. Then, the data and their processing
for this study are presented.
2.1. Air Quality Mapping Technologies and Framework
As mentioned above, environmental monitoring data, i.e., urban air quality data are
geo-referenced, and a map is very suitable for visualizing this dataset. To achieve the purpose
of mapping such a big dataset on one dynamic web map application, we resort to some popular
visualization tools in the GIS and information visualization area. With the booming web mapping
tools, cartographers have more and more choices for their mapping works. Nevertheless, it is also a
challenge for cartographers to know the characteristics of all of these tools and to maintain their own
mapping frameworks. Roth [
27
] mentions this problem and provides a comprehensive analysis of
current web mapping technologies, which gives us some advice to cope with the continued evolution
of these technologies. To be more practical, cartographers need only three categories of tools for a web
mapping work: application frameworks, mapping libraries and visualization libraries. Application
frameworks work as the engine to drive the whole mapping application; mapping libraries as the
map engine to glue all kinds of map services and geospatial data services; visualization libraries as
the symbolization engine to visualize the non-spatial information within geographic entities. This is
a kind of software mashup framework, with which the application engine, the map engine and
the symbol engine can be selected by the map makers according to their visualization topics and
development habits.
Among all of these mapping technologies, this study adopts web.py [
28
], leaflet.js [
29
] and
D3.js [
30
] as the tool framework for the air mapping application. web.py works as the web application
framework to drive the whole mapping application; leaflet.js works as the map engine to glue map
services and data services; D3.js acts as a map symbolization engine to render the time series air quality
data of the multivariate index. These three technologies work together for the air quality mapping
application (Figure 3). This is a kind of technology mashup [
31
] working with the mapping mashup
application. Though this study focuses on air quality monitoring data visualization, the framework and
invented visualization tools can be applied to the visualization work of other similar spatio-temporal
environmental monitoring data, such as meteorological monitoring data (temperature, precipitation,
humidity, etc.).
Figure 3.
The air quality monitoring application framework. This paper proposes a mashup strategy
to make use of multiple visualization and mapping technologies for web mapping applications.
6
Books
MDPI
Atmosphere 2017, 8, 148
2.2. Time Series Map Symbol Encoding: timezoom.js
Time series data visualization mostly has two kind of routines: the linear one and the cyclic
one [
32
–
34
]. Linear ones take the perspective that time is a continuous concept represented by
a time line. Cyclic ones [
35
] regard time as a periodic concept, such as a time keeping system.
The timekeeping model is cyclic for the year, season, month and day, which originates from the
astronomical observations of our ancestors. This paper constructs the time series air quality data with
the hierarchical timekeeping system, and multiple temporal granularities are manipulated by map
symbol interactions. In each node of this leveled structure, the values of each index of air quality are
calculated as the mean value of lower level nodes. This hierarchic time structure is formulated in
Formula (1). Figure 4 shows the timezoom.js symbol structure.
Time
H
= G(Time
V
, Time
S
, Operation) (1)
In Formula (1):
Time
H
is the time hierarchy structure that will be calculated;
Time
V
is the time series values’ array, such as the 365 days of one year data value array,
{t
1
: v
1
, t
2
: v
2
, ..., t
365
: v
365
};
Time
S
is the time system for hierarchy construction, such as {
Year
,
Season
,
Month
,
Day
}. This can
be decided by users and can be extended to hours, minutes and seconds levels, as shown in
Figure 2;
Operation
is the operation for the value aggregation from the lower time level to the higher time
level. This can be any statistical method for aggregation, such as average, median, quantile, etc.
The timezoom.js symbol is a radial tree map visualization tool based on D3.js. The root node of
the hierarchy structure is at the center with leave nodes on the circumference. The values of all nodes
are encoded with a defined color scheme. The initial state of this symbol shows the whole hierarchy.
When clicking on any sector, the symbol will zoom to that sector and fold up other sectors; for example,
when you click on the Q2 sector of the symbol, other sectors will be folded up, and the outside circle of
the symbol will zoom to the days only in Q2 (Figure 4).
Figure 4.
Multi-granularity time zooming interaction (timezoom.js): The inner circle shows the
data value of the year, and Sectors Q1, Q2, Q3 and Q4 respectively indicate the four seasonal values
and sectors January–December for the monthly data of a year; each month sector is surrounded by
daily sectors. Clicking on each sector, the timezoom.js symbol will zoom to that time sector for detailed
views of its value distributions.
7
Books
MDPI
Atmosphere 2017, 8, 148
2.3. Adaptive Map Symbol Control: symadaptive.js
Zooming control means adaptively selecting appropriate symbols at different zoom scales
and view extents. Yang [
36
] proposes a strategy for multi-scale visualization of massive point
data, which selects points with a heavy algorithm on server-side and makes light-weight symbol
displacement on the client-side. Jari Korpi [
37
] evaluated point symbol clutter reduction methods for
map mashups. According to these works, in this paper, the map symbol clutter reduction is achieved
by symbol conflict detection with Rbush.js [
38
], which is a tool library based on the spatial index
algorithm, Rtree [
39
]. At each map zoom level, symbols of points in the current map view extent are
being tested for conflicts. When the symbol has conflicts with other symbols, the point that has a lower
semantic weight will be dropped (Figure 5). The semantic level of each city is decided by the level of
its administration division; see Figure 2.
Figure 5. Multi-scale space zooming control (symadaptive.js): When the symbol C is add to the map,
C will conflict with A and B. If C has a higher importance level than A and B, then C will be kept on
the map, and A and B will be removed. For the other situation, if A or B has a higher importance level
than C, C will be ignored.
In this paper, the mechanism of semantic importance-driven map symbol selection is proposed.
In practice, one can have their own semantic level fields in their data, and this symadaptive.js layer
tool can help to dynamically select appropriate symbols according to the assigned level weight field.
2.4. Data Sources and Processing
The Air Quality Index (AQI) is a number without any unit used to indicate how polluted the air is.
It is adopted by the newly-published national environment protection standard Technical Regulation
on Ambient Air Quality Index (on trial) in China. With this standard, different air pollutants have
their own concentration level ranges, and the index value of each pollutant can be calculated by
Formulas (2) and (3). The AQI is assigned the maximum value of the individual index value of all of
the reported pollutants (Table 1). There are six ranges of the AQI values, and each range is assigned a
descriptor and a visual color code, as shown in Table 2.
IAQI
P
=
IAQI
Hi
− IAQI
Lo
BP
Hi
− BP
Lo
(C
P
− BP
Lo
)+IAQI
Lo
(2)
AQ I
= max{IAQI
1
, IAQI
2
, IAQI
3
, ..., IAQI
n
} (3)
where:
IAQl
P
is the individual Air Quality Index of pollutant P,
C
P
is the concentration of pollutant P,
8
Books
MDPI
Atmosphere 2017, 8, 148
BP
Hi
is the the concentration division point of pollution P that is ≥ C
P
,
BP
Lo
is the concentration breakpoint of pollution P that is ≤ C
P
,
IAQI
Hi
is the index division point (Table 3) corresponding to BP
Hi
,
IAQI
Lo
is the index division point (Table 3) corresponding to BP
Lo
.
Table 1. Reported air index and pollutants and their descriptions.
Name Description and Unit
AQI Air Quality Index, value without unit, range 0–500
SO
2
Sulfur Dioxide, μg/m
3
NO
2
Nitrogen Dioxide, μg/m
3
CO Carbon Monoxide, mg/m
3
O
3
Ozone, μg/m
3
O
3
_8H 8-h average concentration of Ozone, μg/m
3
PM2.5 Particulate Matter, diameter less than 2.5 μm, μg/m
3
PM10 Particulate Matter, diameter less than 10 μm, μg/m
3
Table 2. Air Quality Index level divisions, descriptors and colors.
Divisions Health Influences Color Coding
0–50 Good: Satisfactory Green (0,228,0)
51–100 Moderate: Acceptable, but influential for very sensitive groups. Yellow (255,255,0)
101–150 Slightly Unhealthy: Influential for sensitive groups. Orange (255,126,0)
151–200 Unhealthy Red (255,0,0)
200–300 Very Unhealthy Purple (153,76,0)
>300 Hazardous Maroon (126,0,35)
We collect the hourly reported air quality data from the national real-time air quality reporting
system. By 2017, there were 367 cities with 1497 monitoring sites in China reporting their Air Quality
Index hourly, as shown in Figure 6. Daily air quality of cities is the object of this study. Therefore,
the hourly reported data are aggregated into the daily reported data according to the standard [
2
]. First,
the average hourly data of a city are calculated as the mean pollutant concentration of all monitoring
stations. Then, the daily value is calculated as the mean concentration values of 24 h if it has 16 h of
validated values in a day; otherwise, the daily value is invalidated. The final data are daily records
of all cities with 24-h average concentrations of 5 pollutants (SO
2
,NO
2
, CO, PM10, PM2.5) and 24-h
maximum concentrations of 2 pollutants (O
3
,O
3
-8H). All of the concentration values are converted to
an Individual Air Quality Index(IAQI) value ranging from 0–500 according to Table 3 and Formulas (2)
and (3). Thus, we can get the Air Quality Index (AQI) value and the primary pollutants of each city
each day.
Table 3. Individual Air Quality Index and corresponding pollutants’ concentration limits.
IAQI
Ranges
SO
2
μg/m
3
NO
2
μg/m
3
PM10
μg/m
3
PM2
_5
μg/m
3
CO
mg/m
3
O
3
μg/m
3
O
3
_8H
μg/m
3
00000000
50 50 40 50 35 2 160 100
100 150 80 150 75 4 200 160
150 475 180 250 115 14 300 215
200 800 280 350 150 24 400 265
300 1600 565 420 250 36 800 800
400 2100 750 500 350 48 1000 -
500 2620 940 600 500 60 1200 -
9
Books
MDPI
Atmosphere 2017, 8, 148
Figure 6.
Air Quality Monitoring Network in China. This image shows a snapshot air quality map of
monitoring stations in China, at 8:00, 12 December 2016. (refer to: http://mapviz.xyz:8080/).
3. Results and Discussion
Based on the discourse above, this study designs the air mapping application in terms of the
navigation of space, time and theme [
40
]. The whole application is built on the framework illustrated
in Figure 3, with the two invented JavaScript visualization tools: timezoom.js and symadaptive.js.
3.1. Air Quality Data Map Visualization Design
3.1.1. Spatial Navigation
Air quality data of all cities are loaded by symadaptive.js, and the time series data are used
to build the timezoom.js symbol for each city point. These geo-located glyph symbols are overlaid
on an OpenStreetMap tiled map background. The display of all symbols is under the control of
symadaptive.js. Meanwhile, all of the symbols can be hovered over and clicked to invoke the display
of detailed air quality information about the city point interacted with. The weight of each city is
decided by its administrative level: central city of region, province capitol, district capitol and county
capitol (see Figure 2). Figure 7 shows the interaction effects of zooming the map to different scales.
(a) (b)
Figure 7. Cont.
10
Books
MDPI
Atmosphere 2017, 8, 148
(c)
(d)
Figure 7.
Spatial navigation. When zooming the map, the symbols are selected dynamically at each
zoom level with the function provided by symadaptive.js. (
a
)–(
d
) illustrate map symbols at Zoom
Levels 3–6 on the OpenStreetMap tiled map background. (
a
) Map Zoom Level 3; (
b
) Map Zoom Level 4;
(c) Map Zoom Level 5; and (d) Map Zoom Level 6.
3.1.2. Temporal Navigation
The temporal component of air quality data is presented by the timezoom.js component.
The timezoom.js symbol is driven by the date hierarchy structure of air quality data. All symbols are
event connected. Therefore, when clicking on one map symbol at a specific time section, all symbols on
the map will zoom and change their appearance simultaneously to show air quality data according to
the chosen time section. For a better and efficient query, a bigger time wheel is designed for temporal
navigation on the map side panel. Figure 8 shows the interactions with the time series data by the
timezoom.js symbol.
11
Books
MDPI
Atmosphere 2017, 8, 148
(a) (b)
(c) (d)
Figure 8.
Temporal navigation. The left circle symbol shows the structure of the time symbol, and the
right is the map symbol that will be displayed on the map. The toggle buttons below are for the time
filter to focus on the time granularities of interest. (
a
) shows the full year data in a time symbol, and (
b
)
shows the zoomed symbol to Quarter 4 (October, November and December). (
c
,
d
) show the symbols
that are filtered by the month and day value.
3.1.3. Thematic Navigation
Air quality has a total description as AQI, which contains several important air quality sub-indexes,
PM10, PM2.5, SO
2
,NO
2
,O
3
_8h, O
3
, CO. We design a side panel on the map for thematic attribution
selection. When the thematic attribute changes, the map symbol of all points will change accordingly.
Thus, it is easy to switch the map view among the whole air quality map and individual pollutant
concentration maps.
To sum up, this study presents an urban air quality mapping application based on invented tools:
symadaptive.js and timezoom.js. The map of air quality starts with visualizing the data of central cities
in six major parts of China (see (Figure 9)).
12
Books
MDPI
Atmosphere 2017, 8, 148
Figure 9. Initial view of the online air quality mapping application.
3.2. Results and Analysis
In the air quality application, the concentrations of all of the pollutants are illustrated with graded
color hues under equal interval classification of all values of the year. The AQI map is rendered
with the commonly-used standard color scheme (Table 2). With the thematic navigation radio button,
we can render all air maps of different pollutants. In Figure 10 is presented a series of maps of the AQI
value and seven important pollutants’ concentrations at a nationwide scale, and the dataset can be
switched between two years, 2014 and 2015 (Figure 9). From this series of maps, several significant
findings are shown clearly. These findings provide some clues for in-depth research on the air pollutant
causality and relationships among air pollutants from a spatio-temporal view.
(a) (b)
Figure 10. Cont.
13
Books
MDPI
Atmosphere 2017, 8, 148
(c) (d)
(e) (f)
(g) (h)
Figure 10.
Air quality maps 2015 at OpenStreetMap Zoom Level 5. (
a
) is the AQI map of 2015; (
b
)–(
g
)
are the air pollutants’ concentration map in 2015 for PM10, PM2.5, SO
2
,NO
2
,O
3
and CO respectively.
(
a
) AQI map of 2015; (
b
) PM10 concentration map of 2015; (
c
) PM2.5 concentration map of 2015; (
d
)SO
2
concentration map of 2015; (
e
)NO
2
concentration map of 2015; (
f
)O
3
concentration map of 2015;
(g)O
3
8H concentration map of 2015; (h) CO concentration map of 2015.
3.3. Nationwide Air Quality Condition of China
The interactive mapping application gives us an overview of the air quality condition of China at
a nationwide scale throughout a whole year’s time (Figure 10a). First, for most of the cities in China,
it is more likely to have a good air condition during summer days than winter days. Some special
places are in the northwest part of China, where several cities have a terrible air condition throughout
the year. This situation should be given more attention in further research. Second, air quality maps of
a different time granularities show a clear ribbon pattern along the coastline and southwest part of
China where cities have better air conditions than other part of China; meanwhile, the cities in the
north part of China have the worst air condition. The possible reasons for such patterns reside in two
aspects. On the one hand, the coastline cities have better air circulation conditions for air purification
14
Books
MDPI
Atmosphere 2017, 8, 148
than hinterland areas of China. On the other hand, the cities in the southwest are less developed than
the hinterland; thus, the pollutant emission is lower. Nevertheless, these are hypotheses that need
further study. Furthermore, the patterns can provide auxiliary information for policy-makers to adopt
different measures in different cities.
3.4. Spatio-Temporal Pattern of Air Pollutants
With the thematic navigation radio button, one can have air maps of different pollutants.
In Figure 10, from this series of maps, several significant patterns are shown clearly. First, we can easily
indicate that PM10 and PM2.5 more likely contribute to the AQI value due to their similar pattern
in terms of space and time (see Figure 10b,c, PM10, PM2.5 concentration maps). Actually, the public
cares more about the PM10 and PM2.5 index in China, and they are the critical impacts of smog air
conditions. Second, NO
2
is mainly caused by automobile exhaust, which is more likely to be worse in
big cities, as shown in the NO
2
concentration maps (Figure 10e). Third, from the map of SO
2
, one can
find that the concentration of SO
2
is more likely to be serious in winter months in the north of China
(see SO
2
concentration maps in Figure 10d). This situation can be connected to the burning of coal for
heating in winter of cities in the north of China, which emits a great amount of SO
2
. Fourth, there is an
obvious situation that the value of CO is hardly serious enough to impact public health, as CO is a
deadly poisonous gas that has critical control in China (see the CO concentration maps in Figure 10h).
Unlike other pollutants, which are emitted directly into the air by some specific sources, ozone
(O
3
) is created by sunlight acting on NOx and VOC in the air. Thus, the index value of O
3
is higher on
sunny days throughout the year or in areas that have a longer sunlight duration and stronger sunlight
intensity. In the concentration map of O
3
(Figure 10f,g), we can find that most cities in China will have
higher O
3
values in Seasons 2 and 3, when sunshine is greater and the hours longer during the daytime.
In Lhasa City, Tibet, the O
3
value is high across the year because of its high elevation and thin air,
which causes the high intensity of sunlight. Some cities in the south of China and along the coastline
will show different O
3
value patterns (see the O
3
concentration maps in Figure 10f,g). This kind of
situation may be caused by unstable weather conditions in the south of China; for example, in summer
there are many rainy days in which sunlight intensity is mild.
All of the findings mentioned above are hypotheses. Through these clues, we can design further
studies on these topics and make more reliable conclusions. Moreover, these illustrations can help
environmental regulation governors and the general public to have intuitive images of the air condition
of China.
4. Conclusions
The research presented here describes a novel combination of modern mapping technologies,
with which this study develops an online mapping application of air quality of China. In this study,
an open web platform is fully used to collect the time series air quality data consistently. Then,
data visualization tools (D3.js) and web mapping tools (leaflet.js, rbush.js) are well combined to
produce a fine interactive mapping application of spatio-temporal data. From the application, we can
get a whole view of the air quality condition of China at a nationwide scale and a year time span
at multiple spatio-temporal granularities. This interactive map application clearly presents several
significant findings of air quality in China, which provide good assistance for visual air quality analysis
and clues for in-depth studies on air pollution.
The lessons we learned from this study reside in three aspects. First, there are more and more
open data about our living environment, into which we can delve and find important results for
making our living environment better. Second, as cartographers, we should make full use of new
technologies for data visualization and web mapping. A good combination of these excellent tools
can give us greater power for environmental data visualization. Third, the visual form of data is more
expressive than the raw data table, and it would give deeper insights into the data analysis. In other
15
Books
MDPI
Atmosphere 2017, 8, 148
words, nowadays, with more and more open data, fine and flexible visualization tools and the crowd
wisdom of the public, we can have a clear vision of the environment around us.
The future work is to enhance the efficiency of the timezoom.js symbol and to extend it to hour
granularities for a more detailed time level. What is more, as time goes on and the air quality data
are accumulated, a year selection mechanism should be added to the timezoom.js symbol, and the
comparison function should be enhanced. At the same time, some air quality study problems can be
defined from the previous discussions of the findings with the visualization, then our future work will
collect evidence to test our hypotheses.
Acknowledgments:
This research was supported by the National Key Research and Development Program of
China (Grant No. 2017YFB0503500), and the National Natural Science Foundation of China (Grant No. 41531180).
Author Contributions:
Wei Lu proposed the idea and did the visualization tools development, as well as the
paper writing. Tinghua Ai gave important experiment suggestions and editing of the paper. Xiang Zhang and
Yakun He provided suggestions on paper writing.
Conflicts of Interest: The authors declare no conflict of interest.
Appendix A. Mapping and Visualization Technologies Used
leaflet.js: leaflet.js is a lightweight, extensible, open-source JavaScript library for interactive maps,
which was developed by Vladimir Agafonkin with a group of dedicated contributors. Its simplicity and
extensibility provide the freedom to customize desirable plugins for specific use. For the map symbol,
the Icon class component of leaflet.js can be extended as the HTML(Hypertext Markup Language)
<div> element ,which provides the freedom of combination with other data visualization library.
D3.js: D3.js is short for Data-Driven Document, which is a JavaScript library for manipulate
DOM(Document Object Model) elements based on data. It also provides many powerful visualization
components based on HTML, SVG(Scalable Vector Graphics) and CSS(Cascading Stylesheets). Thus,
we can utilize these visualization facilities to represent data as SVG elements, which can be embedded
in a <div> element. Therefore, we can freely extend the Icon component of leaflet.js by using D3’s
visualization components.
rbush.js: rbush.js is a high-performance JavaScript R-Tree-based 2D spatial index library for points
and rectangles by Vladimir Agafonkin. It can be used for conflict detection for map symbols when
zooming. We can extend the LayerGroup class of leaflet.jsusing the collision detection function of
RBush to realize a self-declutter map layer.
web.py: web.py is a light-weight web application framework in the Python language. It is simple
and powerful, and it is an open sources software tool that can be used for whatever purpose with
absolutely no restrictions.
References
1.
MEP China. Ambient Air Quality Standards. GB 3095-2012; China Environmental Science Press:
Beijing, China, 2012.
2.
MEP China. Technical Regulation on Ambient Air Quality Index (on Trial) (HJ633-2012); China Environmental
Science Press: Beijing, China, 2012.
3.
Chen, C.; Arjomandi, M.; Balmes, J.; Tager, I.; Holland, N. Effects of Chronic and Acute Ozone Exposure
on Lipid Peroxidation and Antioxidant Capacity in Healthy Young Adults. Environ. Health Perspect.
2007, 115, 1732–1737.
4.
Sicard, P.; Lesne, O.; Alexandre, N.; Mangin, A.; Collomp, R. Air quality trends and potential health
effects—Development of an aggregate risk index. Atmos. Environ. 2011, 45, 1145–1153.
5.
Ochoa-Hueso, R.; Munzi, S.; Alonso, R.; Arróniz-Crespo, M.; Avila, A.; Bermejo, V.; Bobbink, R.; Branquinho,
C.; Concostrina-Zubiri, L.; Cruz, C.; et al. Ecological impacts of atmospheric pollution and interactions with
climate change in terrestrial ecosystems of the Mediterranean Basin: Current research and future directions.
Environ. Pollut. 2017, 227, 194–206.
16
Books
MDPI
Atmosphere 2017, 8, 148
6.
Andrienko, N.; Andrienko, G.; Gatalsky, P. Exploratory spatio-temporal visualization: An analytical review.
J. Vis. Lang. Comput. 2003, 14, 503–541.
7.
Kraak, M.J.; Ormeling, F. Cartography: Visualization of Spatial Data; Guilford Press: New York, NY, USA, 2011.
8.
Guo, D.; Chen, J.; MacEachren, A.M.; Liao, K. A visualization system for space-time and multivariate
patterns (vis-stamp). IEEE Trans. Vis. Comput. Graph. 2006, 12, 1461–1474.
9.
Rink, K.; Scheuermann, G.; Kolditz, O. Visualisation in environmental sciences. Environ. Earth Sci.
2014
, 72,
3749–3751.
10.
Grünfeld, K. Integrating spatio-temporal information in environmental monitoring data-a visualization
approach applied to moss data. Sci. Total Environ. 2005, 347, 1–20.
11.
Huang, B. Web-based dynamic and interactive environmental visualization. Comput. Environ. Urban Syst.
2003, 27, 623–636.
12.
Hugentobler, M.; Hurni, L. Web cartography with open standards-a solution to cartographic challenges of
environmental management. Environ. Model. Softw. 2010, 25, 988–999.
13.
Kulawiak, M.; Prospathopoulos, A.; Perivoliotis, L.; Kioroglou, S.; Stepnowski, A.; łuba, M. Interactive
visualization of marine pollution monitoring and forecasting data via a Web-based GIS. Comput. Geosci.
2010, 36, 1069–1080.
14.
Long, Y.; Wang, J.; Wu, K.; Zhang, J. Population Exposure to Ambient PM 2.5 at the Subdistrict Level
in China. Available online: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2486602 (accessed on
27 August 2014).
15.
Rohde, R.A.; Muller, R.A. Air pollution in China: Mapping of concentrations and sources. PLoS ONE
2015, 10, e0135749.
16.
Sicard, P.; Serra, R.; Rossello, P. Spatiotemporal trends in ground-level ozone concentrations and metrics in
France over the time period 1999–2012. Environ. Res. 2016, 149, 122–144.
17.
Huan, L.; Hong, F.; Feiyue, M. A Visualization Approach to Air Pollution Data Exploration—A Case Study
of Air Quality Index (PM2.5) in Beijing, China. Atmosphere 2016, 7, 35.
18.
Chung, K.L.; Qu, H.; Chan, W.Y.; Guo, P.; Xu, A.; Lau, K.H. Visual Analysis of the Air Pollution Problem in
Hong Kong. IEEE Trans. Vis. Comput. Graph. 2007, 13, 1408–1415.
19.
Zhang, Y.L.; Cao, F. Fine particulate matter (PM2.5) in China at a city level. Sci. Rep.
2015
, 5.
doi:10.1038/srep14884.
20.
MacEachren, A.M.; Kraak, M.J. Research challenges in geovisualization. Cartogr. Geogr. Inf. Sci.
2001, 28, 3–12.
21.
Qiang, Y.; Chavoshi, S.H.; Logghe, S.; De Maeyer, P.; Van de Weghe, N. Multi-scale analysis of linear data in
a two-dimensional space. Inf. Vis. 2013, 3, 248–265.
22.
Van de Weghe, N.; De Roo, B.; Qiang, Y.; Versichele, M.; Neutens, T.; De Maeyer, P. The continuous
spatio-temporal model (CSTM) as an exhaustive framework for multi-scale spatio-temporal analysis. Int. J.
Geogr. Inf. Sci. 2014, 28, 1047–1060.
23.
Shneiderman, B. The eyes have it: A task by data type taxonomy for information visualizations.
In Proceedings of the IEEE Symposium on Visual Languages, Boulder, CO, USA, 3–6 September 1996;
pp. 336–343.
24.
Bach, B.; Dragicevic, P.; Archambault, D.; Hurter, C.; Carpendale, S. A review of temporal data visualizations
based on space-time cube operations. In Proceedings of the Eurographics Conference on Visualization,
Wales, UK, 9–13 June 2014.
25.
Shrestha, A.; Zhu, Y.; Miller, B.; Zhao, Y. Storygraph: Telling stories from spatio-temporal data. In International
Symposium on Visual Computing; Springer: Berlin, Germany, 2013; pp. 693–702.
26.
Stolte, C.; Tang, D.; Hanrahan, P. Polaris: A system for query, analysis, and visualization of multidimensional
relational databases. IEEE Trans. Vis. Comput. Graph. 2002, 8, 52–65.
27.
Roth, R.E.; Donohue, R.G.; Sack, C.M.; Wallace, T.R.; Buckingham, T. A process for keeping pace with
evolving web mapping technologies. Cartogr. Perspect. 2015, 25–52, doi:10.14714/CP78.1273.
28. Swartz, A. Web.py Home Page. Available online: http://webpy.org/ (accessed on 4 December 2016).
29.
Agafonkin, V. Leaflet-a JavaScript library for interactive maps. Available online: http://leafletjs.com/
(accessed on 4 December 2016).
30.
Bostock, M. D3.js-Data-Driven Documents. Available online: http://d3js.org/ (accessed on 4 December 2016).
17
Books
MDPI
Atmosphere 2017, 8, 148
31.
Wood, J.; Dykes, J.; Slingsby, A.; Clarke, K. Interactive visual exploration of a large spatio-temporal dataset:
Reflections on a geovisualization mashup. IEEE Trans. Vis. Comput. Graph. 2007, 13, 1176–1183.
32.
Peuquet, D.J. It’s about time: A conceptual framework for the representation of temporal dynamics in
geographic information systems. Ann. Assoc. Am. Geogr. 1994, 84, 441–461.
33.
Aigner, W.; Miksch, S.; Schumann, H.; Tominski, C. Visualization of Time-Oriented Data; Springer: London,
UK, 2011; p. 232.
34. Munzner, T. Visualization Analysis and Design; CRC Press: Boca Raton, FL, USA, 2014.
35.
Carlis, J.V.; Konstan, J.A. Interactive visualization of serial periodic data. In Proceedings of the 11th Annual
ACM Symposium on User Interface Software and Technology, New York, NY, USA, 1–4 November 1998;
pp. 29–38.
36.
Min, Y.; Tinghua, A.; Wei, L.; Xiaoqiang, C.; Qi, Z. A Real-time Generalization and Multi-scale Visualization
Method for POI Data in Volunteered Geographic Information. Acta Geodaetica Cartogr. Sin. 2015, 44, 228.
37.
Korpi, J.; Ahonen-Rainio, P. Clutter reduction methods for point symbols in map mashups. Cartogr. J.
2013, 50, 257–265.
38.
Agafonkin, V. RBush-a high-performance JavaScript R-tree-based 2D spatial index for points and rectangles.
Available online: https://github.com/mourner/rbush (accessed on 4 December 2016).
39.
Guttman, A. R-Trees: A Dynamic Index Structure for Spatial Searching; Springer: Berlin, Germany, 1984;
Volume 14.
40.
Neumann, A. Spatial, Temporal and Thematic Navigation: Visualizing Biographies of European Artists
Using a Task Oriented User Interface Design Approach. Ph.D. Thesis, ETH, Zürich, Switzerland, 2010.
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
18
Books
MDPI
atmosphere
Letter
Temporal and Spatial Patterns of China’s Main Air
Pollutants: Years 2014 and 2015
Tiancai Zhou
1,2
, Jian Sun
2,
*and Huan Yu
1,
*
1
College of Earth Sciences, Chengdu University of Technology, Chengdu 610059, China;
ztc18108279610@163.com
2
Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences,
Beijing 100101, China
* Correspondence: sunjian@igsnrr.ac.cn (J.S.); yuhuan0622@126.com (H.Y.)
Received: 20 June 2017; Accepted: 25 July 2017; Published: 27 July 2017
Abstract:
China faces unprecedented air pollution today. In this study, a database (SO
2
,NO
2
, CO, O
3
,
PM
2.5
(particulate matter with aerodynamic diameter less than 2.5
μ
m), and PM
10
(particulate matter
with aerodynamic diameter less than 10
μ
m) was developed from recordings in 188 cities across China
in 2014 and 2015 to explore the spatial-temporal characteristics, relationships among atmospheric
contaminations, and variations in these contaminants. Across China, the results indicated that the
average monthly concentrations of air pollutants were higher from November to February than in
other months. Further, the spatial patterns of air pollutants showed that the most polluted areas
were located in Shandong, Henan, and Shanxi provinces, and the Beijing-Tianjin-Hebei region.
In addition, the average daily concentrations of air pollutants were also higher in spring and winter,
and significant relationships between the principal air pollutants (negative for O
3
and positive for
the others) were found. Finally, the results of a generalized additive model (GAM) indicated that
the concentrations of PM
10
and O
3
fluctuate dynamically; there was a consistent increase in CO
and NO
2
, and PM
2.5
and SO
2
showed a sharply decreasing trend. To minimize air pollution, open
biomass burning should be prohibited, the energy efficiency of coal should be improved, and the
full use of clean fuels (nuclear, wind, and solar energy) for municipal heating should be encouraged
from November to February. Consequently, an optimized program of urban development should
be highlighted.
Keywords:
air pollutants; dynamics; spatial patterns; generalized additive model; policy
recommendations
1. Introduction
Haze is an atmospheric effect that has become a serious global issue [
1
,
2
], as it affects species
diversity, global climate, and human health [
3
], social, and economic [
4
]. In general, the chemical
aspects of haze have been studied—specifically its physical and chemical properties, including the
elements Cd, Cr, Cu, Fe, and Mn, gaseous pollutants (O
3
,NO
x
,SO
2
,CO
x
), and inorganic aerosols
(SO
4
2
−
,NO
3
−
, and NH
4
+
[
5
–
7
]. Source apportioning has shown that haze is attributable primarily to
dust storms, biomass burning [
8
], coal consumption, and vehicle exhaust [
9
–
11
]. Further, secondary
inorganic and organic aerosols should not be neglected [
12
]. Some studies have focused on the
long-range transport mechanism of haze, which is controlled by meteorological conditions [
13
,
14
];
thus, affluent moisture, warm advection in the lower troposphere, and stable atmospheric stratification
favor the concentration of haze [
15
]. Other studies have investigated the side effects of haze, including
reduced visibility [16], adverse effects on health [17], and so on.
China’s air quality has deteriorated in recent years, and haze affects increasing areas [
18
].
Moreover, numerous studies have found that the regions of Beijing [
19
], North China [
20
], Wuhan [
21
],
Atmosphere 2017, 8, 137 19 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2017, 8, 137
and Guangzhou [
22
] have excessively high concentrations of haze. Other studies have examined
the formation and evolution of haze [
23
,
24
], and the role of meteorological conditions in haze
transportation (e.g., wind and relative humidity) [
25
–
27
]. The urbanization is another driving force in
the development of haze [
28
]. The varied characteristics of haze in different seasons have also been
reported: for example, haze in summer in East China [
29
] and Beijing [
30
], autumn in Shanghai [
31
],
and winter in Beijing [
32
,
33
], Shanghai [
34
], the China Loess Plateau [
35
], and the North China
Plain [36], all of which have relatively higher concentrations of haze.
Meanwhile, the high concentration of organic carbon in PM (particulate matter) indicates that the
main source of haze is biomass burning [
37
], and that the high concentration of NO
2
is closely related
to vehicle emissions [
34
]. Moreover, high concentrations of SO
2
and NO
2
contribute significantly in the
formation of secondary aerosols [
38
]. Thus, secondary aerosols (sulfate, nitrate, and organic matter)
are produced when combined with gaseous pollutants (NO
2
,SO
2
,O
3
, and formaldehyde) and main
atmospheric particles [
39
]. These constitute the main chemical compositions of particulate matter with
aerodynamic diameter less than 2.5
μ
m (PM
2.5
) in Hangzhou [
40
], and the heterogeneous chemical
processes promote and sustain the growth of haze [
41
,
42
]. Further, the spatial patterns and temporal
variation of PM
2.5
have been simulated [
43
], and the simulated values of meteorological conditions
(temperature, and wind speed and direction) agree with the values observed. The diurnal cycle of
land-sea breezes among the Pearl River Delta is the primary influence in the transport of particulate
matter with aerodynamic diameter less than 10 μm (PM
10
) [44].
Many studies have indeed investigated the spatiotemporal patterns of haze across China.
However, the dynamics and relationships of air pollutants during long time series (2 years) have been
insufficiently discussed. Hence, a special study was performed to more fully reveal the spatiotemporal
distribution of the principal air pollutants at present and their future variations in China. Within this
context, the objectives of this study were to: (1) analyze the dynamic and spatiotemporal distribution of
the principal air pollutants for 2014 and 2015, respectively; (2) reveal the dynamics and the relationships
between the principal air pollutants in the most polluted cities, respectively; and (3) identify the
principal variable trends in air pollutants over major Chinese cities based on a generalized additive
model (GAM). Finally, we present strategic policy recommendation for reducing the levels of air
pollutants according to specific haze characteristics in China.
2. Experiments
The main air pollutants (SO
2
,NO
2
, CO, O
3
,PM
2.5
, and PM
10
) were monitored by the continuous
air-monitoring stations (CAMS, Figure 1) covering 188 cities of China. The stations were established
based on the standard “Technical regulation for selection of ambient air quality monitoring stations
(HJ 664-2013)”, and the data were collected through the Ministry of Environmental Protection of
the People’s Republic of China (Available online: http://datacenter.mep.gov.cn/). In addition,
atmospheric contamination was compiled daily and monthly for 2014 and 2015.
In this study, we used the Inverse Distance Weighted (IDW) in ArcGIS 10.2 (ESRI, Inc., Redlands,
CA, USA) to acquire the spatial graphs, SigmaPlot 10.0 (Systat Software, Inc., Chicago, IL, USA)
was used to obtain the rose diagram, and Microsoft Office Visio (2007) (Microsoft corporation,
Redmond, WA, USA) for fishbone diagrams. The Performance Analytics module of R software
(R Core Development Team, R Foundation for Statistical Computing, Vienna, Austria) was employed
to obtain the scatterplot matrices and conduct correlation as well as regression analyses. The two-tailed
Pearson’s correlations at p = 0.05 were employed to determine the correlations between different
variables. The GAM (the non-linear relationship between the pollutants and day-of-year trends) was
modeled flexibly by the log link function, and we used the mgcv module in R to predict air pollutants
in China. The package mgcv (the gam fit function, gam function, family function were contained) of R
was used to fit GAMs specified by presenting a symbolic description of the additive predictor as well
as a description of the error distribution.
20
Books
MDPI
Atmosphere 2017, 8, 137
Figure 1. The study area.
3. Results
3.1. Dynamics of the Principal Air Pollutants
The average monthly concentrations of the principal air pollutants over China in 2014 are shown
in Figure 2 and Table S1. The highest levels of PM
2.5
,PM
10
, CO, SO
2
, and NO
2
were 106.5
μ
g/m
−3
,
154.1
μ
g/m
−3
, 1.69 mg/m
−3
, 62.0
μ
g/m
−3
, and 50.4
μ
g/m
−3
in January. The lowest PM
2.5
and PM
10
levels were 39.8
μ
g/m
−3
and 69.9
μ
g/m
−3
in September, and the lowest values of CO, SO
2
, and NO
2
were 0.93 mg/m
−3
, 19.7
μ
g/m
−3
, and 25.86
μ
g/m
−3
in July. In contrast, the highest and lowest levels
of O
3
were 131.6 μg/m
−3
in July and 61.3 μg/m
−3
in December.
A comparison of the average monthly concentrations of PM
2.5
,PM
10
, CO, SO
2
, and NO
2
indicated
that the values were higher in January, February, November, and December, followed by March, April,
May, and October, with lower concentrations from June to September. However, the opposite trend
was observed with O
3
, which had the highest average monthly levels from May to August, followed
by March, April, September, and October, with lower values in January, February, November, and
December. Further, the highest PM
2.5
,PM
10
, CO, SO
2
, and NO
2
levels were higher than the lowest
ones, respectively, indicating that the principal air pollutants were highest in January.
Similar trends were observed in 2015 (Figure 2). The highest levels of PM
2.5
,PM
10
, CO, and NO
2
were 86.8
μ
g/m
−3
, 129.3
μ
g/m
−3
, 1.60 mg/m
−3
, and 48.0
μ
g/m
−3
in December, and the highest levels
of SO
2
and O
3
were 53.3
μ
g/m
−3
in January and 130.8
μ
g/m
−3
in August (Table S1). The lowest
PM
2.5
and PM
10
levels were 36.4
μ
g/m
−3
and 65.4
μ
g/m
−3
in September, and the lowest values of
CO, SO
2
, and NO
2
were 0.83 mg/m
−3
, 15.7
μ
g/m
−3
, and 24.4
μ
g/m
−3
in July. The lowest level of O
3
was 55.7 μg/m
−3
in August (Table S1).
The average monthly PM
2.5
,PM
10
, CO, SO
2
, and NO
2
concentrations in January and December
were higher than in other months. Thus, the highest PM
2.5
,PM
10
, CO, SO
2
, and NO
2
levels were
2.38, 1.98, 1.91, 3.39, 1.97, and 2.35 times greater than the lowest ones. However, the average monthly
concentration of O
3
was higher from May to August, followed by March, April, September, and
October, with lower values found from January to December.
21
Books
MDPI
Atmosphere 2017, 8, 137
Figure 2.
The monthly dynamics of the principal air pollutants over China (2014 and 2015).
Graphs (
A
–
F
) represent the monthly concentration of PM
2.5
(
μ
g/m
−3
), PM
10
(
μ
g/m
−3
), CO (mg/m
−3
),
SO
2
(
μ
g/m
−3
), NO
2
(
μ
g/m
−3
), and O
3
(
μ
g/m
−3
) in 2014, respectively; Graphs (
G
–
L
) represent the
monthly concentration of PM
2.5
(
μ
g/m
−3
), PM
10
(
μ
g/m
−3
), CO (mg/m
−3
), SO
2
(
μ
g/m
−3
), NO
2
(μg/m
−3
), and O
3
(μg/m
−3
) in 2015, respectively.
3.2. Spatial Analysis of the Principal Air Pollutants
We obtained the average concentrations of the principal air pollutants during the four months
(January, February, November, and December), except for O
3
, the average value of which was calculated
from May to August. Following the processes described above, we measured the spatial distribution
of the principal air pollutants in 2014 (Figure 3). The highest and lowest levels of PM
2.5
and PM
10
were
191.4
μ
g/m
−3
and 296.9
μ
g/m
−3
in Baoding, and 26.4
μ
g/m
−3
and 44.5
μ
g/m
−3
in Sanya (Table S2).
The provinces of Shanxi, Shandong, Henan, and Shanxi, and the Beijing-Tianjin-Hebei region were the
most polluted. It should be noted that most areas of China were affected by PM
2.5
and PM
10
, including
central and northern China. With respect to CO, SO
2
,NO
2
, and O
3
, the distribution narrowed, and
22
Books
MDPI
Atmosphere 2017, 8, 137
was more centralized in the Beijing-Tianjin-Hebei region, and Shandong and Henan provinces. The
highest values of CO, SO
2
,NO
2
, and O
3
were 3.62 mg/m
−3
in Baoding, 151.5
μ
g/m
−3
in Yangquan,
85.2 μg/m
−3
in Baoding, and 225.2 μg/m
−3
in Wuhan (Table S2), respectively.
The air quality improved in 2015 in comparison to 2014 (Figure 3). The most polluted areas were
primarily in the provinces of Hebei, Shandong, and Henan, with the highest levels of PM
2.5
,PM
10
,
and CO (170.4
μ
g/m
−3
, 254.2
μ
g/m
−3
, and 3.57 mg/m
−3
) in Baoding (Table S2). The highest values
of SO
2
,NO
2
, and O
3
were 137.0
μ
g/m
−3
in Shenyang, 77
μ
g/m
−3
in Xingtai, and 184.6
μ
g/m
−3
in
Dezhou (Table S2), respectively. The lowest values of the principal air pollutants were observed in
Hainan province.
Figure 3.
The spatial pattern of the mean concentrations (during January, February, November, and
December) of the principal air pollutants over China (2014 and 2015). Graphs (
A
–
F
) represent the
spatial pattern of PM
2.5
,PM
10
, CO, SO
2
,NO
2
, and O
3
in 2014, respectively; Graphs (
G
–
L
) represent
the spatial pattern of PM
2.5
,PM
10
, CO, SO
2
,NO
2
, and O
3
in 2015, respectively.
23
Books
MDPI
Atmosphere 2017, 8, 137
The dynamic and spatiotemporal distribution of the principal air pollutants indicated that
the cities Baoding, Xingtai, Handan, Shijiazhuang, Hengshui, Dezhou, Tangshan, Heze, Langfang,
Liaocheng, Zibo, Laiwu, Anyang, Linyi, Yichang, Zhengzhou, and Pingdingshan were the most
polluted regions; hence, the average daily concentrations of the principal air pollutants were analyzed
in 2014 and 2015 (Figure 4). The average daily concentrations of air pollutants (SO
2
,NO
2
, CO,
PM
2.5
, and PM
10
) were also higher in the spring and winter, except for O
3
, which had the higher
average daily levels in the summer. Besides, the average daily concentrations of SO
2
,NO
2
, CO,
PM
2.5
, and PM
10
decreased at a rate of 16.2 (
μ
g/m
−3
)/year, 2.8 (
μ
g/m
−3
)/year, 0.1 (mg/m
−3
)/year,
11.8 (
μ
g/m
−3
)/year, and 17.8 (
μ
g/m
−3
)/year from 2014 to 2015, respectively. On the contrary, the
average daily concentrations of O
3
increased from 102.2 (
μ
g/m
−3
)/year to 105.4 (
μ
g/m
−3
)/year
between 2014 and 2015.
Figure 4.
The daily dynamics of the principal air pollutants in the 17 most polluted cities
(2014 and 2015).
3.3. The Dynamics of the Principal Air Pollutants in Beijing-Tianjin-Hebei during January, February,
November, and December
As for the mosted polluteted region (Beijing-Tianjin-Hebei), Figure 5 exhibits the dynamic of
the principal air pollutants during January, February, November, and December in 2014 and 2015.
In Beijing, the concentrations of PM
2.5
, CO, and NO
2
slightly increased; while the concentrations
of PM
10
,SO
2
, and O
3
decreased at a rate of 3.5 (
μ
g/m
−3
)/year, 14.7 (
μ
g/m
−3
)/year, and
3.0 (
μ
g/m
−3
)/year, respectively. All the principal air pollutants in Tianjin showed a decreasing
trend, especially for the SO
2
, which decreased at a rate of 30.9 (
μ
g/m
−3
)/year. Similarly, all the
principal air pollutants in Shijiazhuang also decreased, except for the CO, which increased at a rate
of 0.1 (mg/m
−3
)/year. In addition, the concentration of PM
10
obviously decreased at a rate of
83.1 (μg/m
−3
)/year in Shijiazhuang.
3.4. The Dynamics of the Principal Air Pollutants in Shandong, Henan, and Hebei during Autumn
In terms of the major agricultural provinces Shandong, Henan, and Hebei, the dynamics of the
principal air pollutants in Jinan and Zhengzhou during July and September are presented in Figure 6.
In Jinan, all the principal air pollutants showed a decreasing trend, especially for SO
2
, which decreased
at a rate of 18.0 (
μ
g/m
−3
)/year. However, the average concentrations of PM
2.5
,PM
10
,NO
2
, and O
3
in
Zhengzhou increased at a rate of 7.0 (
μ
g/m
−3
)/year, 14.7 (
μ
g/m
−3
)/year, 9.4 (
μ
g/m
−3
)/year, and
38.6 (
μ
g/m
−3
)/year, respectively. Fortunately, the concentration of SO
2
and CO presented a decreasing
24
Books
MDPI
Atmosphere 2017, 8, 137
trend in Zhengzhou during autumn. In addition, all the principal air pollutants in Shijiazhuang also
decreased, except for the NO
2
, which slightly increased at a rate of 0.4 (μg/m
−3
)/year.
Figure 5.
The average concentration of the principal air pollutants in Beijing-Tianjin-Hebei during
January, February, November, and December (2014 and 2015). Units: PM
2.5
(
μ
g/m
−3
), PM
10
(
μ
g/m
−3
),
CO (mg/m
−3
), SO
2
(μg/m
−3
), NO
2
(μg/m
−3
), and O
3
(μg/m
−3
).
Figure 6.
The average concentration of the principal air pollutants in Jinan, Zhengzhou, and
Shijiazhuang in autumn (2014 and 2015). Units: PM
2.5
(
μ
g/m
−3
), PM
10
(
μ
g/m
−3
), CO (mg/m
−3
),
SO
2
(μg/m
−3
), NO
2
(μg/m
−3
), and O
3
(μg/m
−3
).
3.5. The Relationships among the Principal Air Pollutants
As Figure 7 shows, the correlation coefficients among air pollutants were significant at p < 0.05 in
the most polluted 17 cities (2014). Our results indicated that there were positive correlations among
PM
2.5
,PM
10
, CO, SO
2
, and NO
2
, and negative correlations for O
3
. As for PM
2.5
, and the correlation
coefficients between PM
10
, CO, SO
2
,NO
2
, and O
3
were 0.94, 0.89, 0.74, 0.82, and
−
0.42, respectively.
High correlation coefficients among CO, SO
2
, and NO
2
(0.88 for CO and SO
2
, 0.89 for CO and NO
2
,
0.87 for SO
2
and NO
2
) were observed. We also found significant negative correlation coefficients
among O
3
, CO, SO
2
, and NO
2
(
−
0.57 for O
3
and CO,
−
0.60 for O
3
and SO
2
, and
−
0.56 for O
3
and NO
2
).
Figure 8 shows that there were significant relationships among the principal air pollutants in 2015.
The correlation coefficients between PM
2.5
and PM
10
, CO, SO
2
,NO
2
, and O
3
were 0.94, 0.93, 0.73, 0.84,
and
−
0.38. In addition, there was a slight decrease in the correlation coefficients among CO, SO
2
, and
NO
2
compared to those in 2014, with values of 0.77 for CO and SO
2
, 0.87 for CO and NO
2
, and 0.82 for
SO
2
and NO
2
. A similar decrease in the correlation coefficients among O
3
, CO, SO
2
, and NO
2
was
observed (−0.54 for O
3
and CO, −0.52 for O
3
and SO
2
, and −0.44 for O
3
and NO
2
).
25
Books
MDPI
Atmosphere 2017, 8, 137
Figure 7.
The relationships among the principal air pollutants in the 17 most polluted cities (2014).
All the numbers and red stars represent the correlation coefficients and significant relationships among
the principal air pollutants.
Figure 8.
The relationships among the principal air pollutants in the most polluted cities (2015). All the
numbers and red stars represent the correlation coefficients and significant relationships among the
principal air pollutants.
26
Books
MDPI
Atmosphere 2017, 8, 137
3.6. GAM Prediction of the Trend for Principal Air Pollutants
The GAM was employed to reveal the future dynamics of atmospheric pollutants (Figure 9), and
the non-linear relationship between the pollutants and day-of-year trends was modeled flexibly by
the log link function. The results showed that although PM
2.5
and SO
2
will decrease sharply in the
future, further observation is imperative. Unfortunately, a consistent increase in NO
2
and CO was
predicted. Furthermore, a slight increase of PM
10
was observed until it peaked between 100 < x < 150,
and then decreased slightly over the remaining days. Meanwhile, the estimated O
3
showed a slight
increase before x = 50 and then decreased until approximately x = 125, followed by a slight upward
trend during the remaining days. In summary, a small variation in PM
10
and O
3
indicated that the
simulated results were reliable and desirable.
Figure 9.
Predicted changes in the principal air pollutants together with temporal dynamics by
generalized additive model (GAM) analysis. Rugplot on the x-axis represents the DOY (day of year),
and the light blue belts indicate the credible intervals.
4. Discussion
4.1. Temporal and Spatial Patterns of China’s Principal Air Pollutants
The concentration of the principal air pollutants was higher in January and December—a phenomenon
also observed in Fuzhou [
45
] and Shanghai [
46
,
47
]. In January 2013, two severe air pollution events
happened in Beijing, during which the hourly concentration of PM
2.5
rose to 680
μ
g/m
−3
and
530
μ
g/m
−3
, respectively [
38
]. In Shanghai, the highest and lowest levels of PM
10
,NO
2
, and SO
2
were found in winter and autumn, with vehicle emissions and meteorological conditions the most
probable causes [
36
]. In the North China Plain, high emissions of atmospheric contamination, biomass
burning, and stable weather conditions contributed to the haze events in winter [
48
], and higher
energy consumption and motor vehicle emissions occurred in winter in Guangzhou [
49
]. If natural
gas is supplied for municipal heating in Beijing rather than coal, air pollutant emissions (PM
2.5
,SO
2
,
NO
X
, etc.) would decrease by 52% in winter [
50
] because the inter-transport of pollution and the
27
Books
MDPI
Atmosphere 2017, 8, 137
secondary aerosols O
3
, OC (organic carbon) and VOC (volatile organic compounds) promote pollution.
Thus, reducing the levels of air pollutants during the cold season is imperative for public health [51].
The spatial patterns of the average monthly simulation values indicated that in 2014 the most
polluted regions were the Beijing-Tianjin-Hebei region and the provinces of Shanxi, Shandong, Henan,
Hubei, and Shanxi. Industrial and domestic sources of pollution and agricultural emissions are the
chief regional contributors to pollution in these regions [
52
]. Compared to 2014, the distribution
patterns of air pollutants show a slight shift towards a smaller area in 2015, becoming more centralized
in Hebei, Shandong, and Henan provinces, where haze episodes were also more frequent. A similar
situation prevailed in the North China Plain [
53
], Guangzhou [
22
], and the Yangtze River Delta [
26
,
54
].
The stricter laws protecting air quality by government may account for this decreasing trend between
2014 and 2015 to some extent.
The significant relationships among air pollutants illustrated that there could be similar or
interacting sources for the pollutions. Many studies have been undertaken to explore the source
of large areas of haze. Biomass open burning contributes 47% of the PM
2.5
in the Yangtze River
Delta [
55
], is an important precursor for O
3
[
56
], and is a stressor on marine ecosystems [
57
]. The fact
that the inter-transport of pollution accompanies high humidity is the dominant reason for the haze in
East China [
29
,
46
] and Northern Taiwan [
27
]. Dust was a major source of pollution in eastern Inner
Mongolia [
58
], local emissions and regional transport accounted for pollution in Nanjing [
59
], coal
consumption and industry increased pollution in Beijing in winter [
32
,
50
], and industrial pollution
and vehicle emissions were the dominant local contributors to the levels of NO
2
and PM
2.5
in
Shanghai [31,34].
Specifically, for the most polluted region—the northeast of China—the mineral dust from the
deserts of western China contributes significantly to the concentration of PM [
60
,
61
]. The long-range
transport dust plumes mixed with regional pollutants aid the formation of haze episodes [
53
,
60
–
62
].
Thus, the local pollutants also have significant contributions to the widespread haze pollution [
48
].
Meanwhile, during haze episodes, the secondary inorganic pollutants evident increased in PM [
36
],
suggesting a joint effect among them [
63
]. Under unfavorable meteorological conditions, the
interaction of PM and the secondary inorganic pollutants produced a large amount of aerosols with
the characteristic of low visibility [36].
In general, the dust, municipal heating, and vehicle and local emissions are the dominant
contributors to haze in winter [
31
], and meteorological conditions contribute significantly in the
distribution, formation, and evolution of haze: for example, higher relative humidity and weaker
wind speed contribute to haze [
23
,
25
,
30
]. In addition, there is a positive relationship between visibility
and wind speed, and a negative relationship with relative humidity [
45
,
49
,
59
]. Secondary inorganic
ions were also positively correlated with stable weather conditions (higher relative humidity), which
determined the specific chemical composition of haze [
64
]. In summary, the median or highest values
and the distribution area of air pollutants showed a decreasing trend from 2014 to 2015, which may be
explained in part by the improved energy efficiency and stricter laws protecting air quality.
4.2. Policy Recommendations
The estimated values of PM
2.5
and SO
2
showed a sharp decreasing trend. However, we observed
a consistent increasing trend in CO and NO
2
, while the estimated values of PM
10
and O
3
remained
stable overall. Thus, the concentration of CO and NO
2
in China will continue to increase. As we know
already, vehicle emissions are a key factor in the high concentration of NO
2
, which has been caused
by the rapid increase in the number of vehicles and industrial parks [
31
,
34
]. Therefore, technological
innovation is imperative in coal gasification, liquefaction, and storage to improve the energy efficiency
of coal [
65
]. Indeed, replacing fossil fuels with clean fuel (natural gas, nuclear energy, etc.) for
municipal heating in winter would be an effective long-term measure to mitigate the dense haze north
of the Yellow River [
50
]. In addition, stringent measures to control particle emissions (biomass burning)
should be implemented during specific periods based on meteorological conditions [
29
,
66
], while
28
Books
MDPI
Atmosphere 2017, 8, 137
mass transit should be encouraged to avoid energy waste. Further, the central and local government
should subsidize the manufacture of hybrid, flex-fuel, and electric automobiles [
67
,
68
], and remove
market-entry barriers to promote competition among companies [
65
]. Finally, we need to improve
the proportion of the tertiary industry to optimize the economic structure that uses less energy and
produces fewer particle emissions [69,70].
As Figure 10 shows, this study highlighted the measures needed to decrease current sources of air
pollutants. Indeed, replacing fossil fuels with clean fuels (natural gas, nuclear, wind, and solar energy)
will help to address air pollution’s root causes. To transform and improve air quality, restricting
biomass burning during specific periods (according to meteorological conditions) is imperative, and it
may be necessary to implement strict environmental policies forbidding open biomass burning when
municipal heating is highest. We should also consider recycling biological feed. Equally important,
the issue of vehicle emissions requires the cooperation of government, manufacturers, retailers,
and consumers—particularly in the development and promotion of flex-fuel (hybrid) and electric
automobiles. Further, ecosystem restoration projects such as the Three-North Shelterbelt System can be
effective in defending against sandstorms while simultaneously improving the regional environment.
Remote sensing should also be employed to monitor the formation and development of haze, to record,
measure, and evaluate the sources of pollution, and monitor the trajectory of haze over the long-term.
Indeed, predicting the values and levels of air pollutants and communicating this information by
mobile messages will cement trust between individuals and governments. Further, analyzing the
structure and function of cities based on local conditions to facilitate scientific urban planning and
formulate reasonable population distribution policies should help to mitigate pollution. All of the
aforementioned underscore the essential need to share information, provide support, and increase
cooperation among different departments to achieve effective haze control. Of course, trade-offs
between environmental objectives, economic growth, energy consumption, living standards, and
funding are inevitable, but we believe that these can be balanced for the benefit and welfare of all.
Figure 10. The mind map for decision-makers.
5. Conclusions
Because particulate haze episodes over China have increased in recent years, it is imperative that
we understand pollutant pathways and propose recommendations for a policy framework. In general,
November to February demonstrated the highest concentrations of air pollutants (PM
2.5
,PM
10
, CO,
SO
2
, and NO
2
), excluding O
3
levels, which were highest from May to August. Further, the most highly
polluted areas were in the provinces of Shandong, Henan, and Shanxi, and in the Beijing-Tianjin-Hebei
29
Books
MDPI
Atmosphere 2017, 8, 137
region. Fortunately, most of the principal air pollutants presented a decreasing or relatively stable
trend in Beijing-Tianjin-Hebei during January, February, November, and December. Meanwhile, most
of the principal air pollutants also presented a decreasing trend in Jinan and Shijiazhuang in autumn.
However, in Zhengzhou, the haze events during autumn were unoptimistic. Although the conflict
between clean air and economic growth will continue, measures can be taken to mitigate the sources
of air pollutants and to use resources to control, reduce, and manage air pollution.
To improve our understanding of the formation and frequency of haze, satellite observations and
monitoring of high-risk areas are important subjects for future research. We need to understand the
specific effects of meteorological conditions on the transport mechanism of air pollutants, and the role
they play in the secondary formation process.
Supplementary Materials:
The following are available online at www.mdpi.com/2073-4433/8/8/137/s1:
Table S1: The average monthly concentration of air pollutants over China during 2014 and 2015; Table S2: The mean
concentration of air pollutants over China among four months (January, February, November, and December).
Acknowledgments:
This work was supported by the Ministry of Environmental Protection of the People’s
Republic of China. The research was funded as well by the West Light Foundation of The Chinese Academy of
Sciences, and the Open Fund of the Key Laboratory of Mountain Surface Processes and Eco-regulation.
Author Contributions:
J.S. contributed to the study design, J.S., T.Z. and H.Y. were involved in drafting the
manuscript, approving the final draft, and agree to be accountable for the work. All authors read and approved
the final manuscript.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
He, X.Y.; Hu, J.B.; Chen, W.; Li, X.Y. Haze removal based on advanced haze-optimized transformation
(AHOT) for multispectral imagery. Int. J. Remote Sens. 2010, 31, 5331–5348. [CrossRef]
2.
Han, L.J.; Zhou, W.Q.; Pickett, S.T.A.; Li, W.F.; Li, L. An optimum city size? The scaling relationship for
urban population and fine particulate (PM
2.5
) concentration. Environ. Pollut.
2016
, 208, 96–101. [CrossRef]
[PubMed]
3. Carrasco, L.R. Silver lining of Singapore’s haze. Science 2013, 341, 342–343. [CrossRef] [PubMed]
4.
Hao, Y.; Liu, Y.-M. The influential factors of urban PM
2.5
concentrations in China: A spatial econometric
analysis. J. Clean. Prod. 2016, 112, 1443–1453. [CrossRef]
5.
Malandrino, M.; Casazza, M.; Abollino, O.; Minero, C.; Maurino, V. Size resolved metal distribution in the
PM matter of the city of Turin (Italy). Chemosphere 2016, 147, 477–489. [CrossRef] [PubMed]
6.
Salinas, S.V.; Chew, B.N.; Miettinen, J.; Campbell, J.R.; Welton, E.J.; Reid, J.S.; Yu, L.Y.E.; Liew, S.C. Physical
and optical characteristics of the October 2010 haze event over Singapore: A photometric and lidar analysis.
Atmos. Res. 2013, 122, 555–570. [CrossRef]
7.
Kim, K.W.; Kim, Y.J.; Bang, S.Y. Summer time haze characteristics of the urban atmosphere of Gwangju and
the rural atmosphere of Anmyon, Korea. Environ. Monit. Assess. 2008, 141, 189–199. [CrossRef] [PubMed]
8.
Zhao, H.M.; Zhang, X.L.; Zhang, S.C.; Chen, W.W.; Tong, D.Q.; Xiu, A.J. Effects of Agricultural Biomass
Burning on Regional Haze in China: A Review. Atmosphere 2017, 8, 88. [CrossRef]
9. Szidat, S. Atmosphere. Sources of Asian Haze. Science 2009, 323, 470–471. [CrossRef] [PubMed]
10.
Begum, B.A.; Hopke, P.K. Identification of haze-creating sources from fine particulate matter in Dhaka
aerosol using carbon fractions. J. Air Waste Manag. 2013, 63, 1046–1057. [CrossRef]
11.
Pernigotti, D.; Belis, C.A.; Spano, L. Specieurope: The European data base for PM source profiles.
Atmos. Pollut. Res. 2016, 7, 307–314. [CrossRef]
12.
Huang, R.J.; Zhang, Y.; Bozzetti, C.; Ho, K.F.; Cao, J.J.; Han, Y.; Daellenbach, K.R.; Slowik, J.G.; Platt, S.M.;
Canonaco, F. High secondary aerosol contribution to particulate pollution during haze events in China.
Nature 2014, 514, 218–222. [CrossRef] [PubMed]
13.
Soleiman, A.; Ohtman, M.; Samah, A.A.; Sulaiman, N.M.; Radojevic, M. The occurrence of haze in Malaysia:
A case study in an urban industrial area. Pure Appl. Geophys. 2003, 160, 221–238. [CrossRef]
14.
Abdullaev, S.F.; Maslov, V.A.; Nazarov, B.I. Study of dust haze in arid zone. Izv. Atmos. Ocean. Phys.
2013
, 49,
276–284. [CrossRef]
30
Books
MDPI
Atmosphere 2017, 8, 137
15.
Zhang, R.H.; Li, Q.; Zhang, R.N. Meteorological conditions for the persistent severe fog and haze event over
eastern China in January 2013. Sci. China Earth Sci. 2014, 57, 26–35.
16.
Elias, T.; Haeffelin, M.; Drobinski, P.; Gomes, L.; Rangognio, J.; Bergot, T.; Chazette, P.; Raut, J.C.; Colomb, M.
Particulate contribution to extinction of visible radiation: Pollution, haze, and fog. Atmos. Res.
2009
, 92,
443–454. [CrossRef]
17.
Sun, J.; Zhou, T.C. Health risk assessment of China’s main air pollutants. BMC Public Health
2017
, 17, 212.
[CrossRef] [PubMed]
18.
Zhang, X.Y.; Wang, Y.Q.; Niu, T.; Zhang, X.C.; Gong, S.L.; Zhang, Y.M.; Sun, J.Y. Atmospheric aerosol
compositions in China: Spatial/temporal variability, chemical signature, regional haze distribution and
comparisons with global aerosols. Atmos. Chem. Phys. 2012, 12, 799.
19.
Wang, Y.; Li, Z.Q.; Zhang, Y.; Wang, Q.; Ma, J.Z. Impact of aerosols on radiation during a heavy haze event
in Beijing. IOP Conf. Ser. Earth Environ. Sci. 2014, 17, 012012.
20.
Sun, Y.L.; Jiang, Q.; Wang, Z.F.; Fu, P.Q.; Li, J.; Yang, T.; Yin, Y. Investigation of the sources and evolution
processes of severe haze pollution in Beijing in January 2013. J. Geophys. Res. Atmos.
2014
, 119, 4380–4398.
[CrossRef]
21.
Zhang, M.; Ma, Y.Y.; Gong, W.; Zhu, Z.M. Aerosol optical properties of a haze episode in Wuhan based on
ground-based and satellite observations. Atmosphere 2014, 5, 699–719. [CrossRef]
22.
Zhang, Z.L.; Wang, J.; Chen, L.H.; Chen, X.Y.; Sun, G.Y.; Zhong, N.S.; Kan, H.D.; Lu, W.J. Impact of haze and
air pollution-related hazards on hospital admissions in Guangzhou, China. Environ. Sci. Pollut. Res.
2014
, 21,
4236–4244. [CrossRef] [PubMed]
23.
Liu, X.G.; Li, J.; Qu, Y.; Han, T.; Hou, L.; Gu, J.; Chen, C.; Yang, Y.; Liu, X.; Yang, T.; et al. Formation and
evolution mechanism of regional haze: A case study in the megacity Beijing, China. Atmos. Chem. Phys.
2013
,
13, 4501–4514. [CrossRef]
24.
Quan, J.; Zhang, Q.; He, H.; Liu, J.; Huang, M.; Jin, H. Analysis of the formation of fog and haze in North
China Plain (NCP). Atmos. Chem. Phys. 2011, 11, 8205–8214. [CrossRef]
25.
Bei, N.F.; Xiao, B.; Meng, N.; Feng, T. Critical role of meteorological conditions in a persistent haze episode
in the Guanzhong basin, China. Sci. Total Environ. 2016, 550, 273–284. [CrossRef] [PubMed]
26.
Tang, L.L.; Yu, H.X.; Ding, A.J.; Zhang, Y.J.; Qin, W.; Wang, Z.; Chen, W.T.; Hua, Y.; Yang, X.X. Regional
contribution to PM
1
pollution during winter haze in Yangtze River Delta, China. Sci. Total Environ.
2016
, 541,
161–166. [CrossRef] [PubMed]
27.
Wang, S.H.; Hung, W.T.; Chang, S.C.; Yen, M.C. Transport characteristics of Chinese haze over Northern
Taiwan in winter, 2005–2014. Atmos. Environ. 2016, 126, 76–86. [CrossRef]
28.
Gabrielli, M.; Fracassetti, D.; Tirelli, A. Release of phenolic compounds from cork stoppers and its effect on
protein-haze. Food Control 2016, 62, 330–336. [CrossRef]
29.
Tao, M.H.; Chen, L.F.; Wang, Z.F.; Tao, J.H.; Su, L. Satellite observation of abnormal yellow haze clouds over
East China during summer agricultural burning season. Atmos. Environ. 2013, 79, 632–640. [CrossRef]
30.
Sun, Y.; Song, T.; Tang, G.Q.; Wang, Y.S. The vertical distribution of PM
2.5
and boundary-layer structure
during summer haze in Beijing. Atmos. Environ. 2013, 74, 413–421. [CrossRef]
31.
Wang, Y.J.; Li, L.; Chen, C.H.; Huang, C.; Huang, H.Y.; Feng, J.L.; Wang, S.X.; Wang, H.L.; Zhang, G.;
Zhou, M.; et al. Source apportionment of fine particulate matter during autumn haze episodes in Shanghai,
China. J. Geophys. Res. Atmos. 2014, 119, 1903–1914. [CrossRef]
32.
Lin, Y.C.; Hsu, S.C.; Chou, C.C.K.; Zhang, R.J.; Wu, Y.F.; Kao, S.J.; Luo, L.; Huang, C.H.; Lin, S.H.; Huang, Y.T.
Wintertime haze deterioration in Beijing by industrial pollution deduced from trace metal fingerprints and
enhanced health risk by heavy metals. Environ. Pollut. 2016, 208, 284–293. [CrossRef] [PubMed]
33.
Gui, K.; Che, H.Z.; Chen, Q.L.; An, L.C.; Zeng, Z.L.; Guo, Z.Y.; Zheng, Y.; Wang, H.; Wang, Y.Q.; Yu, J.;
et al. Aerosol optical properties based on ground and satellite retrievals during a serious haze episode in
December 2015 over Beijing. Atmosphere 2016, 7, 70. [CrossRef]
34.
Zhao, W.C.; Cheng, J.P.; Li, D.L.; Duan, Y.S.; Wei, H.P.; Ji, R.X.; Wang, W.H. Urban ambient air
quality investigation and health risk assessment during haze and non-haze periods in Shanghai, China.
Atmos. Pollut. Res. 2013, 4, 275–281. [CrossRef]
35.
Li, W.J.; Shi, Z.B.; Zhang, D.Z.; Zhang, X.Y.; Li, P.R.; Feng, Q.J.; Yuan, Q.; Wang, W.X. Haze particles
over a coal-burning region in the China Loess Plateau in winter: Three flight missions in December 2010.
J. Geophys. Res. Atmos. 2012, 117. [CrossRef]
31
Books
MDPI
Atmosphere 2017, 8, 137
36.
Zhao, X.J.; Zhao, P.S.; Xu, J.; Meng, W.; Pu, W.W.; Dong, F.; He, D.; Shi, Q.F. Analysis of a winter regional
haze event and its formation mechanism in the North China Plain. Atmos. Chem. Phys.
2013
, 13, 5685–5696.
[CrossRef]
37.
Zhu, C.S.; Cao, J.J.; Tsai, C.J.; Shen, Z.X.; Liu, S.X.; Huang, R.J.; Zhang, N.N.; Wang, P. The rural carbonaceous
aerosols in coarse, fine, and ultrafine particles during haze pollution in northwestern China. Environ. Sci.
Pollut. Res. 2016, 23, 4569–4575. [CrossRef] [PubMed]
38.
Wang, Y.S.; Yao, L.; Wang, L.L.; Liu, Z.R.; Ji, D.S.; Tang, G.Q.; Zhang, J.K.; Sun, Y.; Hu, B.; Xin, J.Y. Mechanism
for the formation of the January 2013 heavy haze pollution episode over central and eastern China. Sci. China
Earth Sci. 2014, 57, 14–25. [CrossRef]
39.
Zhu, T.; Shang, J.; Zhao, D.F. The roles of heterogeneous chemical processes in the formation of an air
pollution complex and gray haze. Sci. China Chem. 2011, 54, 145–153. [CrossRef]
40.
Jansen, R.C.; Shi, Y.; Chen, J.M.; Hu, Y.J.; Xu, C.; Hong, S.M.; Li, J.; Zhang, M. Using hourly measurements
to explore the role of secondary inorganic aerosol in PM
2.5
during haze and fog in Hangzhou, China.
Adv. Atmos. Sci. 2014, 31, 1427–1434. [CrossRef]
41.
Shen, X.J.; Sun, J.Y.; Zhang, X.Y.; Zhang, Y.M.; Zhang, L.; Che, H.C.; Ma, Q.L.; Yu, X.M.; Yue, Y.; Zhang, Y.W.
Characterization of submicron aerosols and effect on visibility during a severe haze-fog episode in Yangtze
River Delta, China. Atmos. Environ. 2015, 120, 307–316. [CrossRef]
42.
Lin, Y.F.; Huang, K.; Zhuang, G.S.; Fu, J.S.; Wang, Q.Z.; Liu, T.N.; Deng, C.R.; Fu, Q.Y. A multi-year evolution
of aerosol chemistry impacting visibility and haze formation over an Eastern Asia megacity, Shanghai.
Atmos. Environ. 2014, 92, 76–86. [CrossRef]
43.
Catalano, M.; Galatioto, F.; Bell, M.; Namdeo, A.; Bergantino, A.S. Improving the prediction of air pollution
peak episodes generated by urban transport networks. Environ. Sci. Policy 2016, 60, 69–83. [CrossRef]
44.
Chen, X.L.; Feng, Y.R.; Li, J.N.; Lin, W.S.; Fan, S.J.; Wang, A.Y.; Fong, S.K.; Lin, H. Numerical simulations on
the effect of sea-land breezes on atmospheric haze over the Pearl River Delta Region. Environ. Model. Assess.
2009, 14, 351–363.
45.
Zhang, F.W.; Xu, L.L.; Chen, J.S.; Chen, X.Q.; Niu, Z.C.; Lei, T.; Li, C.M.; Zhao, J.P. Chemical characteristics of
PM
2.5
during haze episodes in the urban of Fuzhou, China. Particuology 2013, 11, 264–272. [CrossRef]
46.
Wang, X.M.; Chen, J.M.; Cheng, T.T.; Zhang, R.Y.; Wang, X.M. Particle number concentration, size distribution
and chemical composition during haze and photochemical smog episodes in Shanghai. J. Environ. Sci.
2014
,
26, 1894–1902. [CrossRef] [PubMed]
47.
Wang, H.L.; Qiao, L.P.; Lou, S.R.; Zhou, M.; Ding, A.J.; Huang, H.Y.; Chen, J.M.; Wang, Q.; Tao, S.K.;
Chen, C.H.; et al. Chemical composition of PM
2.5
and meteorological impact among three years in urban
Shanghai, China. J. Clean. Prod. 2016, 112, 1302–1311. [CrossRef]
48.
Gao, M.; Carmichael, G.R.; Wang, Y.; Saide, P.E.; Yu, M.; Xin, J.; Liu, Z.; Wang, Z. Modeling study of the 2010
regional haze event in the North China Plain. Atmos. Chem. Phys. 2016, 16, 1673–1691. [CrossRef]
49.
Tan, J.H.; Duan, J.C.; Chen, D.H.; Wang, X.H.; Guo, S.J.; Bi, X.H.; Sheng, G.Y.; He, K.B.; Fu, J.M. Chemical
characteristics of haze during summer and winter in Guangzhou. Atmos. Res.
2009
, 94, 238–245. [CrossRef]
50.
Li, H.C.; Yang, S.Y.; Zhang, J.; Qian, Y. Coal-based synthetic natural gas (SNG) for municipal heating in China:
Analysis of haze pollutants and greenhouse gases (GHGs) emissions. J. Clean. Prod.
2016
, 112, 1350–1359.
[CrossRef]
51.
Fang, Z.Q.; Zhu, H.L.; Bao, W.Z.; Preston, C.; Liu, Z.; Dai, J.Q.; Li, Y.Y.; Hu, L.B. Highly transparent paper
with tunable haze for green electronics. Energ. Environ. Sci. 2014, 7, 3313–3319. [CrossRef]
52.
Wang, L.T.; Wei, Z.; Yang, J.; Zhang, Y.; Zhang, F.F.; Su, J.; Meng, C.C.; Zhang, Q. The 2013 severe
haze over southern Hebei, China: Model evaluation, source apportionment, and policy implications.
Atmos. Chem. Phys. 2014, 14, 3151–3173. [CrossRef]
53.
Zhang, X.L.; Huang, Y.B.; Zhu, W.Y.; Rao, R.Z. Aerosol characteristics during summer haze episodes from
different source regions over the coast city of North China Plain. J. Quant. Spectrosc. Radiat. Transf.
2013
, 122,
180–193. [CrossRef]
54.
Wang, S.; Zhou, C.; Wang, Z.; Feng, K.; Hubacek, K. The characteristics and drivers of fine particulate matter
(PM
2.5
) distribution in China. J. Clean. Prod. 2017, 142, 1800–1809. [CrossRef]
55.
Cheng, Z.; Wang, S.; Fu, X.; Watson, J.G.; Jiang, J.; Fu, Q.; Chen, C.; Xu, B.; Yu, J.; Chow, J.C.; et al. Impact
of biomass burning on haze pollution in the Yangtze River delta, China: A case study in summer 2011.
Atmos. Chem. Phys. 2014, 14, 4573–4585. [CrossRef]
32
Books
MDPI
Atmosphere 2017, 8, 137
56.
Zhang, Y.W.; Zhang, X.Y.; Zhang, Y.M.; Shen, X.J.; Sun, J.Y.; Ma, Q.L.; Yu, X.M.; Zhu, J.L.; Zhang, L.; Che, H.C.
Significant concentration changes of chemical components of PM
1
in the Yangtze River Delta area of China
and the implications for the formation mechanism of heavy haze-fog pollution. Sci. Total Environ.
2015
, 538,
7–15. [CrossRef] [PubMed]
57.
Jaafar, Z.; Loh, T.L. Linking land, air and sea: Potential impacts of biomass burning and the resultant haze
on marine ecosystems of Southeast Asia. Glob. Chang. Biol. 2014, 20, 2701–2707. [CrossRef] [PubMed]
58.
Guo, J.P.; Zhang, X.Y.; Cao, C.X.; Che, H.Z.; Liu, H.L.; Gupta, P.; Zhang, H.; Xu, M.; Li, X.W. Monitoring
haze episodes over the Yellow Sea by combining multisensor measurements. Int. J. Remote Sens.
2010
, 31,
4743–4755. [CrossRef]
59.
Kang, H.Q.; Zhu, B.; Su, J.F.; Wang, H.L.; Zhang, Q.C.; Wang, F. Analysis of a long-lasting haze episode in
Nanjing, China. Atmos. Res. 2013, 120, 78–87. [CrossRef]
60.
Tao, M.H.; Chen, L.F.; Su, L.; Tao, J.H. Satellite observation of regional haze pollution over the North China
Plain. J. Geophys. Res. Atmos. 2012, 117, D12. [CrossRef]
61.
Yang, T.; Wang, X.Q.; Wang, Z.F.; Sun, Y.L.; Zhang, W.; Zhang, B.; Du, Y.M. Gravity-current driven transport
of haze from North China Plain to Northeast China in Winter 2010-Part I: Observations. Sola
2012
, 8, 13–16.
[CrossRef]
62.
Ye, X.X.; Song, Y.; Cai, X.H.; Zhang, H.S. Study on the synoptic flow patterns and boundary layer process
of the severe haze events over the North China Plain in January 2013. Atmos. Environ.
2016
, 124, 129–145.
[CrossRef]
63.
Kirrane, E.; Svendsgaard, D.; Ross, M.; Buckley, B.; Davis, A.; Johns, D.; Kotchmar, D.; Long, T.C.; Luben, T.J.;
Smith, G.; et al. A Comparison of risk estimates for the effect of short-term exposure to PM, NO
2
and CO on
cardiovascular hospitalizations and emergency department visits: Effect size modeling of study findings.
Atmosphere 2011, 2, 688–701. [CrossRef]
64.
Han, B.; Zhang, R.; Yang, W.; Bai, Z.P.; Ma, Z.Q.; Zhang, W.J. Heavy haze episodes in Beijing during January
2013: Inorganic ion chemistry and source analysis using highly time-resolved measurements from an urban
site. Sci. Total Environ. 2016, 544, 319–329. [CrossRef] [PubMed]
65.
Zhang, X.H.; Wu, L.Q.; Zhang, R.; Deng, S.H.; Zhang, Y.Z.; Wu, J.; Li, Y.W.; Lin, L.L.; Li, L.; Wang, Y.J.;
et al. Evaluating the relationships among economic growth, energy consumption, air emissions and air
environmental protection investment in China. Renew. Sustain. Energy Rev. 2013, 18, 259–270. [CrossRef]
66.
Mu, M.; Zhang, R.H. Addressing the issue of fog and haze: A promising perspective from meteorological
science and technology. Sci. China Earth Sci. 2014, 57, 1–2. [CrossRef]
67.
Sun, J.; Wang, J.N.; Wei, Y.Q.; Li, Y.R.; Liu, M. The Haze Nightmare Following the Economic Boom in China:
Dilemma and Tradeoffs. Int. J. Environ. Res. Public Health 2016, 13, 402. [CrossRef] [PubMed]
68.
Wu, X.; Tan, L.; Guo, J.; Wang, Y.; Liu, H.; Zhu, W. A study of allocative efficiency of PM
2.5
emission rights
based on a zero sum gains data envelopment model. J. Clean. Prod. 2016, 113, 1024–1031. [CrossRef]
69.
Wang, S.S.; Zhou, D.Q.; Zhou, P.; Wang, Q.W. CO
2
emissions, energy consumption and economic growth in
China: A panel data analysis. Energy Policy 2011, 39, 4870–4875. [CrossRef]
70.
Yuan, X.L.; Mu, R.M.; Zuo, J.; Wang, Q.S. Economic development, energy consumption, and air pollution:
A critical assessment in China. Hum. Ecol. Risk Assess. 2015, 21, 781–798. [CrossRef]
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
33
Books
MDPI
atmosphere
Article
Improving PM
2.5
Air Quality Model Forecasts in
China Using a Bias-Correction Framework
Baolei Lyu
1,†
, Yuzhong Zhang
2,‡
and Yongtao Hu
3,
*
1
Department for Earth System Science, Tsinghua University, Beijing 100084, China; baoleilv@foxmail.com
2
School of Earth and Atmospheric Sciences, Georgia Institute of Technology, Atlanta, GA 30332, USA;
zhangyz.pku@gmail.com
3
School of Civil and Environmental Engineering, Georgia Institute of Technology, Atlanta, GA 30332, USA
* Correspondence: yh29@mail.gatech.edu; Tel.: +01-404-385-4558
† Current address: Huayun Sounding Meteorological Technology Corporation, Beijing 102299, China.
‡ Current address: School of Engineering and Applied Sciences, Harvard University, Cambridge,
MA 02138, USA.
Received: 11 July 2017; Accepted: 9 August 2017; Published: 13 August 2017
Abstract:
Chinese cities are experiencing severe air pollution in particular, with extremely high
PM
2.5
levels observed in cold seasons. Accurate forecasting of occurrence of such air pollution
events in advance can help the community to take action to abate emissions and would ultimately
benefit the citizens. To improve the PM
2.5
air quality model forecasts in China, we proposed a
bias-correction framework that utilized the historic relationship between the model biases and
forecasted and observational variables to post-process the current forecasts. The framework consists
of four components: (1) a feature selector that chooses the variables that are informative to model
forecast bias based on historic data; (2) a classifier trained to efficiently determine the forecast analogs
(clusters) based on clustering analysis, such as the distance-based method and the classification
tree, etc.; (3) an error estimator, such as the Kalman filter, to predict model forecast errors at monitoring
sites based on forecast analogs; and (4) a spatial interpolator to estimate the bias correction over
the entire modeling domain. One or more methods were tested for each step. We applied five
combinations of these methods to PM
2.5
forecasts in 2014–2016 over China from the operational AiMa
air quality forecasting system using the Community Multiscale Air Quality (CMAQ) model. All
five methods were able to improve forecast performance in terms of normalized mean error (NME)
and root mean square error (RMSE), though to a relatively limited degree due to the rapid changing
of emission rates in China. Among the five methods, the CART-LM-KF-AN (a Classification And
Regression Trees-Linear Model-Kalman Filter-Analog combination) method appears to have the best
overall performance for varied lead times. While the details of our study are specific to the forecast
system, the bias-correction framework is likely applicable to the other air quality model forecast
as well.
Keywords: PM
2.5
; forecast; post-processing; CMAQ
1. Introduction
The fine particulate matter with an aerodynamic diameter of less than 2.5
μ
m (PM
2.5
) is the
dominant air pollutant in most Chinese cities in recent years. In 2016, the nationwide annual
mean-observed PM
2.5
concentration was 47
μ
g/m
3
, which exceeded 35
μ
g/m
3
, the level II national
ambient air quality standards (NAQQS) of China (GB3095-2012), by more than 34%. The pollution
levels are much higher in densely populated regions [
?
]. For example, in the Beijing-Tianjin-Hebei
(BTH) area the average annual mean of PM
2.5
concentration was 71
μ
g/m
3
in 2016, twice of the
standard, which would cause severe adverse health effects [
??
]. It is critical to provide the public and
Atmosphere 2017, 8, 147 34 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2017, 8, 147
administrative agencies air pollution alerts in advance, to help citizens take timely protective actions
(e.g., wearing masks, staying indoors), and to help governments control emissions through dynamic
management actions [??].
Empirical models and deterministic models are the two major approaches to forecast air quality in
the near future. The empirical approaches usually employ statistical models that relate predictor and
explanatory variables [
??
]. These methods can be easily implemented as long as there are sufficient
observations available for training the statistical model. However, they have difficulty in forecasting
air pollution in longer-term and larger-scales, and have no means to predict pollutant compositions or
to provide emission controlling implications [
??
]. The deterministic approaches overcome the above
inabilities of statistical models by adopting chemical transport models (CTMs), e.g., the Community
Multi-scale Air Quality- model (CMAQ) [
?
], and especially, can produce forecasts in a much longer
lead time, with comparable accuracy, to meet the requirements of the dynamic management practice.
CTMs start with a detailed emissions inventory and forecasted meteorological fields, and solve a
series of mathematical equations in space and time to simulate air pollutants’ fates with the evolution
of physical and chemical processes in the atmosphere. Regional air quality forecasting systems that
are based upon CTMs have been widely established in the world to provide operational air quality
forecasts in real-time [
????
]. However, there could be significant prediction errors in these forecasts
due to emission inventory uncertainties, meteorology forecast uncertainties, and the missing physical
and chemical mechanisms in the CTMs [? ].
One way to improve the deterministic model forecasting performance is to utilize the statistical
methods-based, post-processing techniques to adjust the current forecasts. Fundamentally, these
post-processing techniques are bias correction techniques that utilize the deterministic model’s historic
errors to correct the current model forecasts (from now on, the word “model” exclusively refers to
the deterministic model). The simplest bias correction technique is the moving mean method, which
directly applies the averaged forecast bias of the previous time period to the current model forecast [
?
]. The Kalman filter has also been used to derive future bias from model’s historical performance [
???
?
]. A more advanced bias correction technique is the analog method which first clusters the historic
forecasts into resembling analogs, and then derives future bias from the historic analog members [
???
]. The analog method considers the distinction of performance levels between different analogs, which
might be related to the variations of pollution levels (which are due to various air masses and emission
events). The analog method can be further combined with the Kalman filter and other methods to
determine the ensemble bias within the same group of analog members [
?
]. These bias correction
techniques have been demonstrated to decrease model forecast errors in weather forecasts [
??
], and
O
3
[
????
] and PM
2.5
forecasts [
??
]. It is noteworthy that the previous bias correction studies on
PM
2.5
forecasts were all conducted for areas with much cleaner air than in China (e.g., US, UK, Italy,
and Portugal) [
???
], with their annual mean PM
2.5
concentrations around 10
μ
g/m
3
or below, and
with relatively small day-to-day variations. Also, most of these bias correction studies were conducted
for 1–2 day lead time model forecasts.
In this study, we propose a bias-correction framework that explores and utilizes the relationships
between model biases and predictor variables to improve nationwide PM
2.5
model forecasts in China.
Within the same framework, we tested and compared five bias-correction techniques to post-process
the 1-day, 3-day, and 5-day lead-time model forecasts from a nationwide air quality forecasting system
in China that has been in operation for over three years. The purpose of the study is to test the
existing bias-correction techniques in China, which includes very heavily polluted areas with very
large variations in PM
2.5
concentration. For example, the PM
2.5
concentration in Beijing can vary from
over 250
μ
g/m
3
to below 20
μ
g/m
3
within a couple of days. We utilized the framework to test the
Classification and Regression Trees (CART) method to determine analogs with better distinctions. The
study was also aimed to test the effectiveness of the bias-correction techniques for longer lead-time
forecasts, such as the 3-day and 5-day in advance, which can better support the dynamic air quality
management practice.
35
Books
MDPI
Atmosphere 2017, 8, 147
2. Experiments
2.1. PM
2.5
Model Forecasts
The daily PM
2.5
model forecasts from 2014 to 2016 were obtained from an operational air quality
forecasting system (Available online: www.aimayubao.com) built upon the CMAQ model (version
5.0.2), along with the Weather Research and Forecasting Model (WRF, version 3.4.1) [
?
] and an
emissions processing component. The emissions inventory used (called AiMa inventory) was compiled
and projected from various inventories and information sources and was further adjusted by utilizing
inverse modeling techniques. The CMAQ modeling is configured with the saprc07tc gas chemistry
mechanism and the aero6 aerosol module. The WRF simulation is driven by NCEP’s GFS 0.5-degree
global weather forecast products. The forecasting system produces 144 h forecasts (called AiMa
forecasts) at each cycle that covers 5 days local time. The operation of the forecasting system started on
4 February 2014. The model grids have a spatial resolution of 12 km covering the entirety of China.
The configurations of the forecasting system have not been changed since operation, ensuring the
consistency of the forecasting error distribution. The daily forecasts for the lead times of 1, 3, and 5
days were used in this study. The relative longer forecasting lead time (compared to 1–2 days lead
time in previous studies [
??
]) would enable us to evaluate whether the post-correction method is
useful for dynamic management practices [
?
], which usually requires about a 3–5 day lead time to
take actions.
2.2. PM
2.5
Observations
The air quality observational data from the Chinese air quality monitoring network was obtained
from the official real-time air quality monitoring publishing platform [
?
]. The monitoring network had
945 monitors in operation in 2014 and expanded to 1496 monitors in 2015. The network measured PM
2.5
concentrations with the widely used TEOM (Tapered Element Oscillating Microbalance) instruments.
The 24 hourly observations within a day from 0:00 through 24:00 (Beijing local time) were averaged to
calculate daily mean PM
2.5
concentrations. In case that two or more monitors are within the same grid
cell, the averages of the observations from these monitors are used. To distinguish from the original
observational data from monitors, we refer to this dataset as observations at “avg-monitors”. As a
result, the 945 monitors operational since 2014 were assigned to 545 different grid cells and all the 1496
monitors operational since 2015 were assigned to 840 grid cells (Figure
??
a). Figure
??
b illustrates the
spatial patterns of the observed PM
2.5
pollution levels in China in year 2016.
Figure 1.
(
a
) Avg-monitors used in this study. The monitors located in the same grid cell have been
averaged to derive avg-monitors. The blue dots denote the monitors operational since 2014. The
red and green dots both denote monitors operational since 2015, but the red dots represent monitors
that are used for evaluating the “final product”. (
b
) Annual mean PM
2.5
concentrations observed at
avg-monitors in 2016.
36
Books
MDPI
Atmosphere 2017, 8, 147
2.3. Bias-Correction Framework
To improve forecast performance, we propose a bias-correction framework as a post-processing
procedure for model forecasts. The framework utilizes historic data to establish relationships between
the model forecast biases and a variety of model-simulated or observed variables, and then uses these
relationships to correct the current forecast. The framework is conducted in four steps: (1) feature
selection; (2) forecast analog determination; (3) local correction estimation; and (4) correction spread.
Figure ?? illustrates the framework and the methods we tested in each step.
Figure 2.
Bias-correction framework with its four steps and the four method combinations that are
tested in this study. The fifth method tested is the 7-day Moving Average (7-Day-MA) method, which
can be regarded as a special case within the framework but is not explicitly shown here.
2.3.1. Feature Selection
In the “feature selection” step, we choose a group of variables containing information about model
biases from a pool of modeled or observed variables. By eliminating non-informative variables, we
can improve the predictability of model biases by analogs. In this study, data from 545 avg-monitors
(blue dots in Figure
??
) in 2014 and 2015 are used as historic data for selecting informative variables, or
features, at each of these avg-monitors. Only the variables that are selected in this step will be used in
the following “analog determination” step.
Here, we consider in total 19 candidate variables. Five of them are observation variables (i.e.,
the mean observed concentrations of the five criteria air pollutants, CO, NO
2
,O
3
,PM
2.5
, and SO
2
)
on the day prior to the forecasting cycle, and the rest of the candidate variables are model output,
including model forecasted concentrations of the five criteria air pollutants, forecasted daily mean
PM
2.5
composition concentrations (SO
4
,NO
3
,NH
4
, EC and OC), and four meteorological variables
(i.e., wind speed, temperature, planetary boundary layer heights, and relative humidity). We tested
two methods for feature selection: (1) a linear regression model (denoted as LM); and (2) the random
forest algorithm (denoted as RF). The LM method constructed a linear regression model with the
PM
2.5
forecast biases being the explained variable and the candidate variables being the explanatory
variables. Those explanatory variables with p-value < 0.05 [
??
] were retained as informative variables
for analog determination. With the RF method [
?
], the variables with large importance indicators
37
Books
MDPI
Atmosphere 2017, 8, 147
are selected as informative variables. We implemented these feature selection algorithm with the R
software (the lm function for LM and the Boruta function for RF) [? ].
Figure
??
shows the number counts of the avg-monitors that selected each candidate variable by
the LM and RF algorithms. On average, the LM algorithm selected about six informative variables
for each avg-monitors. In contrast, the RF algorithm selected about 15 on average. This difference
is likely an indication that RF is less effective than LM at distinguishing the information content
among variables. As we will find out later in our analysis (Table
??
), a method involving LM (e.g., the
CART-LM-KF-AN method, see definition in Figure
??
) often outperformed a method involving RF (e.g.,
CART-RF-KF-AN). This comparison highlights the importance of a proper feature selection procedure.
Figure 3.
Number counts of the avg-monitors that used each of the 19 features respectively by the
linear regression (a) and the random forest (b) based feature selection method.
Table 1.
Performance statistics, R
2
, NME, and RMSE (in ug/m
3
) after the “local correction
estimation” step.
Lead Time Metrics Raw 7-Day-MA DIST-AN
DIST-KF-
AN
CART-RF-
KF-AN
CART-LM-
KF-AN
1 day
R
2
0.46 0.45 0.46 0.48 0.48 0.49
NME 0.49 0.39 0.40 0.40 0.45 0.41
RMSE 32.2 27.0 27.0 27.2 29.7 27.5
3 day
R
2
0.38 0.35 0.33 0.37 0.38 0.39
NME 0.50 0.43 0.46 0.45 0.44 0.43
RMSE 31.4 27.8 29.0 28.0 28.2 27.5
5 day
R
2
0.34 0.31 0.28 0.31 0.33 0.34
NME 0.51 0.46 0.49 0.47 0.46 0.45
RMSE 32.4 29.1 30.3 28.8 29.5 28.7
Among the 19 variables, PM
2.5
observations on the initial day of the forecast cycle (I.PM
2.5
) and
CMAQ PM
2.5
forecast (F.PM
2.5
) were the top two most selected variables for the 1-day lead time,
indicating that they contain the most information about the short-term model forecast errors. As
expected, the I.PM
2.5
is less informative for longer lead times, resulting in a decrease in selection
38
Books
MDPI
Atmosphere 2017, 8, 147
counts by the LM algorithm for the 3 and 5 day lead times (Figure
??
). For all lead times in question,
model-forecast air pollutant variables, such as F.PM
2.5
, OC, and F.CO, and meteorological variables,
such as RH, were also frequently selected by the LM algorithm at many avg-monitors. The forecasted
OC and F.CO were frequently selected, likely because they are indicative of model biases in emissions
and transport.
2.3.2. Analog Determination
In the “analog determination” step, we search for a “forecast analog”, an ensemble of previous
model forecasts that are similar to the current forecast to be corrected [
??
]. The bias information in
the forecast analog is then used to estimate the correction to be applied in the current forecast in the
following “local correction estimation” step. The similarity between current and historic forecasts
is measured in terms of the informative variables we selected in the “feature selection” step, using
two different methods: (1) Euclidean distance between two forecasts in the feature space (denoted
as DIST) and (2) classification predicted by the CART algorithm [
?
] (denoted as CART). The CART
algorithm generates a decision tree which minimizes the total deviations within the branches of the
tree. The algorithm is widely used in remote sensing image processing for land cover classification [
??
], ecology modeling [
??
] and pattern recognition studies [
?
]. We implemented the CART calculation
with the rpart() function in the R software [? ].
2.3.3. Local Correction Estimation
In the “local correction estimation” step, we estimated the forecast bias at individual avg-monitors
based on the “forecast analog” determined in the previous “analog determination” step.
A straightforward way to estimate the bias is the Distance-based Analog (DIST-AN) method,
which takes an inverse distance weighted average of the biases in the forecast analog.
PM
cp,T+k
= PM
p,T+k
− δ
T+k
= PM
p,T+k
−
∑
M
m
=1
PM
p,t
m
−PM
o,t
m
d
t
m
,T+k
∑
M
m
=1
1
d
t
m
,T+k
(1)
where the
PM
cp,T+k
and
PM
p,T+k
, respectively, refer to the corrected and raw model forecasts of
the PM
2.5
concentrations.
δ
T+k
refers to the correction, which is calculated by the inverse distance
weighted mean forecast biases of the M analogs.
In addition to the inverse distance weighted average, the Kalman filter (KF), known for its easy
implementation, fast convergence speed, and effectiveness at eliminating data noises, is also tested in
this study to estimate forecast errors. The Kalman filter works on an ordered set of inputs. Previous
studies have used the input dataset ordered by time [
??
]. In this study, we implemented the Kalman
filter on a set of analogs ordered by distances. The Kalman filter used a dynamic weighting method to
fuse the observations and estimations at time t, as shown in the equation below:
ˆ
x
t+1|t
=
ˆ
x
t|t−1
+ K
t
y
t
−
ˆ
x
t|t−1
(2)
where the
ˆ
x
t+1|t
refers to the estimation of forecast error at the time t+1 using the information at time
t. The
ˆ
x
t|t−1
denotes the forecast error estimation at the time t. The
y
t
refers to the observed forecast
error at time t. The weighting factor K
t
was called Kalman gain, which was calculated through the
optimization of estimation and observation noises. The detailed approach for K
t
estimation can be
found in Delle Monache, Nipen, Liu, Roux, and Stull [
?
]. Depending on the methods used in the
“analog determination” step, the Kalman filter is applied in the Distance-based Kalman Filtering
Analog (DIST-KF-AN) method, the CART linear model Kalman Filtering Analog (CART-LM-KF-AN)
method, and the CART random forest Kalman Filtering Analog (CART-RF-KF-AN) method.
Additionally, we also tested the 7-day Moving Average (7-Day-MA). With this method, the
correction is computed as the average of the forecast biases in a 7-day window prior to the forecasting
39
Books
MDPI
Atmosphere 2017, 8, 147
cycle. This method is chosen for its fast computation and easy implementation. The 7-day window
length has also been used in previous studies on model post-correction [
??
]. The 7-Day-MA method
can be regarded as a special case of the analog method, in which the 7 days prior to the forecasting
cycle are the forecast analog and each day is weighted equally.
2.3.4. Correction Spread
After the “local correction estimation” is done, we further spread the estimated corrections at
avg-monitors to the entire domain including model grids containing no monitors. The corrections
were then applied to the original gridded model forecast to obtain bias-corrected forecasts over the
whole China domain. In this study, we used ordinary Kriging [
?
] to spatially interpolate the biases
from monitors to the entire domain.
2.4. Evaluation
Following the post-process framework, we tested five combinations of methods (Figure
??
). For
example, the CART-LM-KF-AN method uses the LM method for “feature selection”, the CART method
for “analog determination”, and the KF method for “local correction estimation”. Readers can refer
to Figure
??
for an illustration of how varied methods for each step are combined. To evaluate the
performance in varied steps of the procedure, we reported separately the performance statistics for
the “local correction estimation” and “correction spread” steps. The reported performance statistics
include the coefficient of determination (R
2
), the normalized mean error (NME), and the root mean
square error (RMSE) [? ].
The observation and model outputs from 4 February 2014 to 31 December 2015 are used as historic
data, with which we selected informative variables and searched for forecast analogs. In the “local
correction estimation” step, we used the model output for 2016 to estimate the bias-corrected forecasts
at the 545 individual avg-monitors. We then used the corresponding 2016 observations to evaluate
the performance after the “local correction estimation” step. Note that not all avg-monitors were
used in the “local correction estimation”. After the “correction spread” step, we then used the data
from the remaining 211 avg-monitors to evaluate the performance of the “final product” at locations
without observations. The 211 avg-monitors were so selected that they were adequately apart from
each other and from the 545 avg-monitors that were used in the “local correction estimation”. These
211 avg-monitors are marked as red dots in Figure
??
. The performance evaluation was conducted
separately for 1-day, 3-day, and 5-day lead forecasts.
3. Results
3.1. Performance in Estimating Local Corrections
Although up to the “local correction estimation” step we only computed the local corrections
at locations with observations, these local corrections were crucial for the performance of the final
product. Therefore, in this study, we applied five different methods (Figure
??
) and evaluated them
at avg-monitors. Figure
??
presents the performance of the raw PM
2.5
model forecasts in predicting
observations at avg-monitors in 2016. The annual mean biases of model forecasts in 2016 were generally
negative in the southern and northwestern part of China, while they were positive (mostly 0–10
μ
g/m
3
)
in the middle part of China, regardless of the forecast lead-times. According to the distributions of
statistical metrics of R
2
and NME, the raw model forecast performed much better in North China and
East China than in other regions, while it performed much worse in West China. Spatial patterns
with geographical divisions in biases and other performance statistical metrics, i.e., R
2
, NME, and
RMSE here, indicate potential non-uniform uncertainties in the emission rates estimation among
regions and varied prediction errors in meteorological forecast fields over different terrains. We do
observe a significant degradation in the performance of predicting meteorological variables at the
surface in regions with complex terrains (results not shown). However, as the lead-time increases, the
40
Books
MDPI
Atmosphere 2017, 8, 147
performance of the raw PM
2.5
model forecasts only degrade slightly (Figure
??
and Table
??
), implying
that the error does not grow much in predicting meteorology during the forecasting period of 144 h.
Figure 4.
Performance of the raw PM
2.5
model forecasts in predicting observations at avg-monitors in
2016: (
a
,
b
,
c
) annual mean biases for the 1, 3, and 5-day lead times and (
d
,
e
,
f
) annual mean R
2
, NME,
and RMSE for the 1-day lead-time.
Table
??
summarizes the performance statistics of each method for different lead times and
Table
??
summarizes the percentage changes of the performance statistics with respect to the raw
model forecasts. Compared with the raw model forecasts, all the bias-correction methods are able to
decrease the NME and the RMSE at all the three lead times. The reductions in the NME are 7.4–19.3%,
7.3–14.4%, and 4.5–12.2%, and the reductions in RMSE are 7.8–16.1%, 7.5–12.5%, and 6.3–11.3% for
1-day, 3-day, and 5-day lead times, respectively, showing that these post-processing techniques are
effective to improve the PM
2.5
forecast at locations with observations. Figure
??
shows that all methods
can improve NME and RMSE at the majority of avg-monitors. For example, the CART-LM-KF-AN
method decreases NME and RMSE at about 70% to 80% avg-monitors for all three lead times.
Table 2.
Percentage changes (%) of the performance statistics with respect to original model forecast by
the five methods in the “correction estimation” step.
Lead Time Metrics 7-Day-MA DIST-AN DIST-KF-AN
CART-RF-
KF-AN
CART-LM-
KF-AN
1 day
R
2
−2.7 0.3 3.1 3.5 6.1
NME
−19.3 −18.3 −17.3 −7.4 −14.7
RMSE
−16.0 −16.1 −15.6 −7.8 −14.6
3 day
R
2
−8.3 −13.3 −4.0 −1.5 0.7
NME
−13.6 −7.3 −10.0 −12.6 −14.4
RMSE
−11.6 −7.5 −10.9 −10.1 −12.5
5 day
R
2
−8.9 −17.7 −7.0 −2.0 −0.3
NME
−11.2 −4.5 −8.9 −10.7 −12.2
RMSE
−10.0 −6.3 −11.1 −9.0 −11.3
41
Books
MDPI
Atmosphere 2017, 8, 147
Figure 5.
Percentage (%) of avg-monitors with increased R
2
values (
a
), and decreased NME (
b
) and
RMSE values (c) by the five post-correction methods and for the 1, 3, and 5-day lead times.
For most of these methods, however, R
2
only increases marginally for the 1-day lead time and
even decreases slightly in the 3-day and 5-day lead times. For example, DIST-AN decreases the R
2
by
17.7% for the 3-day lead time. The ineffectiveness in increasing R
2
may reflect that these analog-based
methods, although good at reducing biases, do not improve the ability to capture the variability in
the data, especially for longer lead times. Among the five methods in question, the CART-LM-KF-AN
method has the largest R
2
for all the three lead times, with a 6% increase in R
2
for the 1-day lead and
essentially no change for the 3-day and 5-day leads from the raw model forecast.
The results also show the impact of forecasting lead times on the performance of bias-correction
techniques (
????
, Figure
??
). In general, the enhancement in the forecast performance decreases with
the longer lead time. The decreasing effectiveness of the post-processing procedure with lead times
may partly result from the fact that the increasing uncertainties in the model forecasted meteorology
and pollutant concentrations lead to larger uncertainties in the analog determination for the longer
lead times. Among the five methods in this study, the performance of the CART-LM-KF-AN method is
most insensitive to varied lead times (NME 0.41, 0.43, 0.45, and RMSE 27.5, 27.5, 28.7 ug/m
3
for 1-day,
3-day, and 5-day lead times, respectively), showing that the combination of the CART and LM methods
constitutes a more robust analog determination algorithm for the PM
2.5
model forecast in China. In
contrast, the performance of the DIST-AN method (NME 0.40, 0.46, 0.49, and RMSE 27.0, 29.0, 30.3
ug/m
3
for 1-day, 3-day, and 5-day lead times, respectively) and the 7-Day-MA method (NME 0.39,
0.43, 0.46 and RMSE 27.0, 27.8, 29.1 ug/m
3
for 1-day, 3-day, and 5-day lead time, respectively) degrade
significantly with longer lead times. Although the performance of the CART-LM-KF-AN method is
similar to or a little worse than the 7-Day-MA and DIST-AN for the 1-day lead time, CART-LM-KF-AN
outperforms other methods for a longer lead time, which is a good property for the purpose of dynamic
air quality management.
3.2. Performance of the Final Product
After we estimated the correction for the model forecasts at each of the individual avg-monitors,
we spread the local corrections across the entire Chinese domain by spatially interpolating the local
corrections with ordinary Kriging. Figure
??
shows an example of the “final product” of a 5-day lead
forecast for 30 December 2016, using the CART-LM-KF-AN method.
42
Books
MDPI
Atmosphere 2017, 8, 147
Figure 6.
Bias-corrected CMAQ PM
2.5
forecast over China for 30 December 2016 (5-day lead time) by
the CART-LM-KF-AN method. The dots represent observed PM
2.5
levels.
By comparing the “final product” with observations at 211 avg-monitors (whose data were not
used in the “local correction estimation” step), we can evaluate the performance of the “final product”
at locations without avg-monitors. Table
??
shows that the correction estimated through the spatial
interpolation can also effectively reduce forecast errors, even at locations without PM
2.5
monitors.
Depending on the methods and lead times, the fraction of avg-monitors that finds improvements in
NME and RMSE varies from 50 to 80% (Figure
??
). The improvements, for most methods, are slightly
less but still comparable to those at locations with observations (
????
), indicating that, compared to
the forecast errors at locations with observations, the uncertainties induced by spatial interpolation
are likely insignificant. In other words, the “local correction estimation” (including feature selection
and analog determination) rather than “correction spread” is the “bottle-neck” in the post-correction
framework. Efforts to further improve the performance should be directed to improve the estimation
of local corrections.
Table 3.
Performance statistics for R
2
, NME, and RMSE (in ug/m
3
) at locations without monitors after
correction spread.
Lead Time Metrics Raw 7-Day-MA DIST-AN
DIST-KF-
AN
CART-RF-
KF-AN
CART-LM-
KF-AN
1 day
R
2
0.38 0.33 0.40 0.39 0.44 0.43
NME 0.48 0.47 0.42 0.46 0.42 0.41
RMSE 28.4 27.7 25.6 27.4 24.5 24.3
3 day
R
2
0.34 0.29 0.33 0.34 0.33 0.33
NME 0.49 0.47 0.44 0.47 0.44 0.44
RMSE 27.7 26.8 25.3 26.6 25.8 25.6
5 day
R
2
0.31 0.26 0.29 0.31 0.30 0.30
NME 0.50 0.49 0.46 0.48 0.46 0.46
RMSE 28.3 27.6 26.1 27.1 26.5 26.5
43
Books
MDPI
Atmosphere 2017, 8, 147
Figure 7.
Percentage (%) of avg-monitors with increased R
2
values (
a
) and decreased NME (
b
) and
RMSE values (
c
) by the five post-correction methods and for the 1, 3, and 5-day lead times by spatially
interpolating the estimated biases.
3.3. Discussion
In comparison with previous studies conducted in the U.S., our results generally show less
improvements (in terms of percentage) from the raw model forecasts. For example, Djalalova, Delle,
Monache, and Wilczak [
?
] used KF-AN (similar to DIST-KF-AN in this study) to post-correct hourly
CMAQ PM
2.5
forecasts for the 1-day lead time and reduced the MAE values by 65% from 8.5
μ
g/m
3
to
3
μ
g/m
3
. Kang, et al. [
?
] also applied the KF-AN method to daily PM
2.5
forecasts and reduced RMSE
by 33% from 7.5
μ
g/m
3
to 5
μ
g/m
3
. Previous studies have found that the KF- and AN-based methods
better perform in lower pollution level regions [
?
]. The differences in the performance between this
and previous studies may result from the fact that PM
2.5
levels in China are much higher than in the
U.S. Consistent with previous studies, our analysis also shows that the percentage improvements by
the CART-LM-KF-AN method are generally larger in relatively cleaner regions (e.g., the Pearl River
Delta in South china, Northeast China, and other remote regions) than in heavily polluted regions (e.g.,
the North China Plain and the Yangtze River Delta in East China) (Figure
??
), suggesting that there
might be important factors missing in the trained relationship between model biases and predictor
variables over polluted regions. One such factor is the fast-changing emissions in both magnitude and
distribution in regions such as the North China Plain and the Yangtze River Delta during the modeled
three years [
???
], a result of increasingly more strict emission control enforcements and/or economic
fluctuations. The significant change of emission rates in these regions between the training years
(2014–2015) and the prediction year (2016) could confound the trained bias correction relationships. In
contrast, the actual emission rates were likely to vary insignificantly in the cleaner regions over the
same time period.
44
Books
MDPI
Atmosphere 2017, 8, 147
Figure 8.
The difference in R
2
(
a
,
d
,
g
), NME (
b
,
e
,
h
), and RMSE (
c
,
f
,
i
) values between bias-corrected
forecasts by the CART-LM-KF-AN method and raw CMAQ forecasts (Corrected forecast minus raw
forecasts) at the 545 avg-monitors for the 1, 3, and 5-day lead times. The differences in RMSE is
in μg/m
3
.
4. Conclusions
To improve the PM
2.5
forecast, we proposed a bias-correction framework that utilized the
relationships between biases and select forecasted and observational variables. The framework consists
of four steps: feature selection, analog determination, local correction estimation, and correction
spread. We applied this bias-correction framework to PM
2.5
forecasts in 2014–2016 over China from
the operational AiMa air quality forecasting system using the CMAQ model. Five methods, differing
in how to perform feature selection, analog determination, and local correction estimation, were tested
in this study, and we found all the five methods were able to improve the overall forecast performance
in terms of RMSE and NME, though to a relatively limited degree.
Based on our results, we recommend the CART-LM-KF-AN method. In most cases, the
performance of this method is better or comparable to other methods. Particularly, the method
shows consistent improvement for longer lead times (3–5 day) when other methods degrade in their
performance. This is important for dynamic air quality management, as this type of practice often
requires longer lead time.
In comparison with previous studies that were all conducted in areas outside of China, our results
generally show fewer improvements (in terms of percentage) from the raw model forecast, especially in
regions with much higher pollution levels. A major reason for this is that the fast-changing emissions
45
Books
MDPI
Atmosphere 2017, 8, 147
in high pollution regions of China can confound the relationship between model biases and predictor
variables. On the other side, the correction spread is not found to be a significant source of errors at
locations without monitors. Future efforts should be directed to improve the performance in the local
correction estimation, especially to explore methods that can build relationships that better represent
the missing factors of changing emissions.
Acknowledgments:
The authors thank the Hangzhou AiMa Technologies, Inc. for providing the archived AiMa
5-day air quality forecasting products for years 2014–2016.
Author Contributions:
Baolei Lyu, Yuzhong Zhang, and Yongtao Hu conceived and designed the study;
Baolei Lyu
gathered data and implemented the algorithm; Baolei Lyu and Yuzhong Zhang performed the analysis;
Baolei Lyu, Yuzhong Zhang, and Yongtao Hu wrote the paper.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Zhang, Y.L.; Cao, F. Fine particulate matter (PM
2.5
) in China at a city level. Sci. Rep.
2015
, 5, 14884. [CrossRef]
[PubMed]
2.
Fang, D.; Wang, Q.g.; Li, H.; Yu, Y.; Lu, Y.; Qian, X. Mortality effects assessment of ambient PM
2.5
pollution
in the 74 leading cities of China. Sci. Total Environ. 2016, 569–570, 1545–1552. [CrossRef] [PubMed]
3.
Gao, M.; Saide, P.E.; Xin, J.; Wang, Y.; Liu, Z.; Wang, Y.; Wang, Z.; Pagowski, M.; Guttikunda, S.K.;
Carmichael, G.R. Estimates of health impacts and radiative forcing in winter haze in eastern China through
constraints of surface PM
2.5
predictions. Environ. Sci. Technol. 2017, 51, 2178–2185. [CrossRef] [PubMed]
4.
Hu, Y.; Odman, M.T.; Chang, M.E.; Russell, A.G. Operational forecasting of source impacts for dynamic air
quality management. Atmos. Environ. 2015, 116, 320–322. [CrossRef]
5.
Zhang, Y.; Bocquet, M.; Mallet, V.; Seigneur, C.; Baklanov, A. Real-time air quality forecasting, part I: History,
techniques, and current status. Atmos. Environ. 2012, 60, 632–655. [CrossRef]
6.
Lv, B.; Cobourn, W.G.; Bai, Y. Development of nonlinear empirical models to forecast daily PM
2.5
and ozone
levels in three large Chinese cities. Atmos. Environ. 2016, 147, 209–223. [CrossRef]
7.
Perez, P.; Salini, G. PM
2.5
forecasting in a large city: Comparison of three methods. Atmos. Environ.
2008
, 42,
8219–8224. [CrossRef]
8.
Eder, B.; Kang, D.; Mathur, R.; Yu, S.; Schere, K. An operational evaluation of the Eta–CMAQ air quality
forecast model. Atmos. Environ. 2006, 40, 4894–4905. [CrossRef]
9.
Zhang, Y.; Bocquet, M.; Mallet, V.; Seigneur, C.; Baklanov, A. Real-time air quality forecasting, part II: State of
the science, current research needs, and future prospects. Atmos. Environ. 2012, 60, 656–676. [CrossRef]
10.
Byun, D.; Schere, K.L. Review of the governing equations, computational algorithms, and other components
of the models-3 community multiscale air quality (CMAQ) modeling system. Appl. Mech. Rev.
2006
, 59,
51–77. [CrossRef]
11.
Mckeen, S.; Grell, G.; Peckham, S.; Wilczak, J.; Djalalova, I.; Hsie, E.Y.; Frost, G.; Peischl, J.; Schwarz, J.;
Spackman, R. An evaluation of real-time air quality forecasts and their urban emissions over eastern texas
during the summer of 2006 second texas air quality study field study. J. Geophys. Res. Atmos.
2009
, 114, D7.
[CrossRef]
12.
Vaughan, J.; Lamb, B.; Frei, C.; Wilson, R.; Bowman, C.; Figueroa-Kaminsky, C.; Otterson, S.; Boyer, M.;
Mass, C.; Albright, M.; et al. A numerical daily air quality forecast system for the Pacific Northwest. Bull.
Am. Meteorol. Soc. 2004, 85, 549–561. [CrossRef]
13.
Otte, T.L.; Pouliot, G.; Pleim, J.E.; Young, J.O.; Schere, K.L.; Wong, D.C.; Lee, P.C.S.; Tsidulko, M.;
McQueen, J.T.; Davidson, P.; et al. Linking the Eta model with the community multiscale air quality
(CMAQ) modeling system to build a national air quality forecasting system. Weather Forecast.
2005
, 20,
367–384. [CrossRef]
14.
Zhou, G.Q.; Xu, J.M.; Xie, Y.; Chang, L.Y.; Gao, W.; Gu, Y.X.; Zhou, J. Numerical air quality forecasting over
eastern China: An operational application of WRF-Chem. Atmos. Environ. 2017, 153, 94–108. [CrossRef]
15.
Binkowski, F.S.; Roselle, S.J. Models-3 community multiscale air quality (CMAQ) model aerosol component
1. Model description. J. Geophys. Res. Atmos. 2003, 108, 335–346. [CrossRef]
46
Books
MDPI
Atmosphere 2017, 8, 147
16.
Djalalova, I.; Delle Monache, L.; Wilczak, J. PM
2.5
analog forecast and kalman filter post-processing for the
community multiscale air quality (CMAQ) model. Atmos. Environ. 2015, 108, 76–87. [CrossRef]
17.
Delle Monache, L.; Nipen, T.; Deng, X.X.; Zhou, Y.M.; Stull, R. Ozone ensemble forecasts: 2. A kalman filter
predictor bias correction. J. Geophys. Res. Atmos. 2006, 111, D05308. [CrossRef]
18.
Kang, D.W.; Mathur, R.; Rao, S.T. Implementation of real-time bias-corrected o-3 and PM
2.5
air quality
forecast and their performance evaluations during 2008 over the continental united states. In Air Pollution
Modeling and its Application XX; Steyn, D.G., Rao, S.T., Eds.; Springer: Berlin/Heidelberg, Germany, 2010;
p. 283.
19.
Djalalova, I.; Wilczak, J.; McKeen, S.; Grell, G.; Peckham, S.; Pagowski, M.; DelleMonache, L.; McQueen, J.;
Tang, Y.; Lee, P.; et al. Ensemble and bias-correction techniques for air quality model forecasts of surface o-3
and PM
2.5
during the texaqs-ii experiment of 2006. Atmos. Environ. 2010, 44, 455–467. [CrossRef]
20.
De Ridder, K.; Kumar, U.; Lauwaet, D.; Blyth, L.; Lefebvre, W. Kalman filter-based air quality forecast
adjustment. Atmos. Environ. 2012, 50, 381–384. [CrossRef]
21.
Delle Monache, L.; Djalalova, I.; Wilczak, J. Analog-based postprocessing methods for air quality forecasting.
In Air Pollution Modeling and its Application XXIII; Steyn, D.G., Rao, S.T., Eds.; Springer: Berlin/Heidelberg,
Germany, 2014; pp. 237–239.
22.
Huang, J.P.; McQueen, J.; Wilczak, J.; Djalalova, I.; Stajner, I.; Shafran, P.; Allured, D.; Lee, P.; Pan, L.; Tong, D.;
et al. Improving NOAA NAQFC PM
2.5
predictions with a bias correction approach. Weather Forecast.
2017
,
32, 407–421. [CrossRef]
23.
Delle Monache, L.; Nipen, T.; Liu, Y.; Roux, G.; Stull, R. Kalman filter and analog schemes to postprocess
numerical weather predictions. Mon. Weather Rev. 2011, 139, 3554–3570. [CrossRef]
24.
Dolman, B.K.; Reid, I.M. Bias correction and overall performance of a VHF spaced antenna boundary layer
profiler for operational weather forecasting. J. Atmos. Sol. Terr. Phys. 2014, 118, 16–24. [CrossRef]
25.
Mckeen, S.; Wilczak, J.; Grell, G.; Djalalova, I.; Peckham, S.; Hsie, E.Y.; Gong, W.; Bouchet, V.; Menard, S.;
Moffet, R. Assessment of an ensemble of seven real-time ozone forecasts over eastern north america during
the summer of 2004. J. Geophys. Res. Atmos. 2005, 110, 3003–3013. [CrossRef]
26.
Monache, L.D.; Grell, G.A.; Mckeen, S.; Wilczak, J.; Pagowski, M.O.; Peckham, S.; Stull, R.; Mchenry, J.;
Mcqueen, J. A kalman-filter bias correction of ozone deterministic, ensemble-averaged, and probabilistic
forecasts. Tellus 2006, 60b.
27.
Wilczak, J.; Mckeen, S.; Djalalova, I.; Grell, G.; Peckham, S.; Gong, W.; Bouchet, V.; Moffet, R.; Mchenry, J.;
Mcqueen, J. Bias-corrected ensemble and probabilistic forecasts of surface ozone over eastern North America
during the summer of 2004. J. Geophys. Res. Atmos. 2012, 111, 6443–6445. [CrossRef]
28.
Kang, D.; Mathur, R.; Rao, S.T. Real-time bias-adjusted O
3
and PM
2.5
air quality index forecasts and their
performance evaluations over the continental united states. Atmos. Environ.
2010
, 44, 2203–2212. [CrossRef]
29.
Crooks, J.L.; Özkaynak, H. Simultaneous statistical bias correction of multiple PM
2.5
species from a regional
photochemical grid model. Atmos. Environ. 2014, 95, 126–141. [CrossRef]
30.
Neal, L.S.; Agnew, P.; Moseley, S.; Ordonez, C.; Savage, N.H.; Tilbee, M. Application of a statistical
post-processing technique to a gridded, operational, air quality forecast. Atmos. Environ.
2014
, 98, 385–393.
[CrossRef]
31.
Silibello, C.; Bolignano, A.; Sozzi, R.; Gariazzo, C. Application of a chemical transport model and optimized
data assimilation methods to improve air quality assessment. Air. Qual. Atmos. Health
2014
, 7, 283–296.
[CrossRef]
32.
Skamarock, W.C.; Klemp, J.B.; Dudhia, J.; Gill, D.O.; Barker, D.M.; Wang, W.; Powers, J.G. A Description of
the Advanced Research WRF Version 2; No. NCAR/TN-468+ STR; Mesoscale and Microscale Meteorology
Division, National Center For Atmospheric Research: Boulder, CO, USA, 2005.
33.
Zhou, Y.; Wu, Y.; Yang, L.; Fu, L.X.; He, K.B.; Wang, S.X.; Hao, J.M.; Chen, J.C.; Li, C.Y. The impact of
transportation control measures on emission reductions during the 2008 olympic games in Beijing, China.
Atmos. Environ. 2010, 44, 285–293. [CrossRef]
34.
China National Urban Air Quality Real-Time Publishing Platform. Available online: http://106.37.208.233:
20035 (accessed on 14 January 2017).
35. Malik, M.B. Applied linear regression. Technometrics 2005, 47, 371–372. [CrossRef]
47
Books
MDPI
Atmosphere 2017, 8, 147
36.
Jones, P.W.; Quirk, F.H.; Baveystock, C.M.; Littlejohns, P. A self-complete measure of health status for chronic
airflow limitation. The st. George’s respiratory questionnaire. Am. Rev. Respir. Dis.
1992
, 145, 1321–1327.
[CrossRef] [PubMed]
37.
Chen, Y.W.; Lin, C.J. Combining svms with various feature selection strategies. In Feature Extraction;
Guyon, I., Nikravesh, M., Gunn, S., Zadeh, L.A., Eds.; Springer: Berlin/Heidelberg, Germany, 2006; Volume 3,
pp. 315–324.
38. Team, R.C. R: A Language and Environment for Statistical Computing; The R Team: Vienna, Austria, 2016.
39.
Breiman, L.; Friedman, J.H.; Olshen, R.; Stone, C.J. Classification and regression trees. Biometrics
1984
, 40,
358.
40.
Liu, J.; Sun, D.; He, F.; Zhang, W.; Guan, X. Land use/cover classification with classification and regression
tree applied to MODIS imagery. J. Appl. Sci. 2013, 13, 3070–3773.
41.
Youssef, A.M.; Pourghasemi, H.R.; Pourtaghi, Z.S.; Al-Katheeri, M.M. Landslide susceptibility mapping
using random forest, boosted regression tree, classification and regression tree, and general linear models
and comparison of their performance at Wadi Tayyah Basin, asir region, Saudi Arabia. Landslides
2016
, 13,
839–856. [CrossRef]
42.
Mertens, M.; Nestler, I.; Huwe, B. Gis-based regionalization of soil profiles with classification and regression
trees (CART). J. Plant. Nutr. Soil Sci. 2002, 165, 39–43. [CrossRef]
43.
De’Ath, G.; Fabricius, K.E. Classification and regression trees: A powerful yet simple technique for the
analysis of complex ecological data. Ecology 2000, 81, 3178–3192. [CrossRef]
44.
Jain, A.K.; Duin, R.P.W.; Mao, J. Statistical pattern recognition: A review. IEEE Trans. Pattern Anal.
2000
, 22,
4–37. [CrossRef]
45.
Kang, D.; Mathur, R.; Rao, S.T.; Yu, S. Bias adjustment techniques for improving ozone air quality forecasts.
J. Geophys. Res. Atmos. 2007, 113, 2036–2044. [CrossRef]
46.
Oliver, M.A.; Webster, R. Kriging: A method of interpolation for geographical information systems. Int. J.
Geogr. Inf. Syst. 1990, 4, 313–332. [CrossRef]
47. Kang, D.; Mathur, R.; Rao, S.T. Assessment of bias-adjusted PM
2.5
air quality forecasts over the continental
united states during 2007. Geosci. Model. Dev. 2009, 2, 309–320.
48.
Duncan, B.N.; Lamsal, L.N.; Thompson, A.M.; Yoshida, Y.; Lu, Z.F.; Streets, D.G.; Hurwitz, M.M.;
Pickering, K.E. A space-based, high-resolution view of notable changes in urban NOX pollution around the
world (2005–2014). J. Geophys. Res. Atmos. 2016, 121, 976–996. [CrossRef]
49.
Krotkov, N.A.; McLinden, C.A.; Li, C.; Lamsal, L.N.; Celarier, E.A.; Marchenko, S.V.; Swartz, W.H.;
Bucsela, E.J.; Joiner, J.; Duncan, B.N.; et al. Aura omi observations of regional SO
2
and NO
2
pollution
changes from 2005 to 2015. Atmos. Chem. Phys. 2016, 16, 4605–4629. [CrossRef]
50.
Wu, Y.; Zhang, S.J.; Hao, J.M.; Liu, H.; Wu, X.M.; Hu, J.N.; Walsh, M.P.; Wallington, T.J.; Zhang, K.M.;
Stevanovic, S. On-road vehicle emissions and their control in China: A review and outlook. Sci. Total Environ.
2017, 574, 332–349. [CrossRef] [PubMed]
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
48
Books
MDPI
atmosphere
Article
Air Quality and Control Measures Evaluation during
the 2014 Youth Olympic Games in Nanjing and its
Surrounding Cities
Hui Zhao
1
, Youfei Zheng
1,2,
* and Ting Li
2
1
Key Laboratory for Aerosol-Cloud-Precipitation of China Meteorological Administration,
Nanjing University of Information Science and Technology, Nanjing 210044, China; zhaohui_nuist@163.com
2
Key Laboratory of Atmospheric Environment Monitoring and Pollution Control,
Collaborative Innovation Center of Atmospheric Environment and Equipment Technology,
Nanjing University of Information Science and Technology, Nanjing 210044, China; c2015liting@163.com
* Correspondence: zhengyf@nuist.edu.cn
Academic Editors: Pius Lee, Rick Saylor and Jeff McQueen
Received: 28 April 2017; Accepted: 3 June 2017; Published: 4 June 2017
Abstract:
Air pollution had become a vital concern for the 2014 Youth Olympic Games in Nanjing.
In order to control air pollutant emissions and ensure better air quality during the Games, the Nanjing
municipal government took a series of aggressive control measures to reduce pollutant emissions
in Nanjing and its surrounding cities during the Youth Olympic Games. The Air Quality Index
(AQI) is an index of air quality which is used to inform the public about levels of air pollution
and associated health risks. In this study, we use the AQI and air pollutant concentrations data
to evaluate the effectiveness of the implementation of control measures. The results suggest that
the emission reduction measures significantly improved air quality in Nanjing. In August 2014,
the mean concentrations of PM
2.5
,PM
10
,SO
2
,NO
2
, CO and O
3
were 42.44
μ
g
·
m
−3
, 59.01
μ
g
·
m
−3
,
11.12
μ
g
·
m
−3
, 31.09
μ
g
·
m
−3
, 0.76 mg
·
m
−3
and 38.39
μ
g
·
m
−3
, respectively, and fell by 35.92%, 36.75%,
20.40%, 15.05%, 8.54% and 47.15%, respectively, compared to the prophase mean before the emission
reduction. After the emission reduction, the mean concentrations of PM
2.5
,PM
10
,SO
2
,NO
2
, and O
3
increased by 20.81%, 41.84%, 22.84%, 21.16% and 60.93%, respectively, which is due to the cancellation
of temporary atmospheric pollution control measures. The air pollutants diurnal variation curve
during the emission reduction was lower than the other two periods, except for CO. In addition,
the AQI of Nanjing and its surrounding cities showed a downward trend, compared with July
2014. The most of effective method to control air pollution is to implement the measures of regional
cooperation and joint defense and control, and reduce local emissions during the polluted period,
such as airborne dust, coal-burning, vehicle emissions, mobile sources and industrial production.
Keywords:
air pollutants; PM
2.5
; Youth Olympic Games; meteorological conditions; emission
reduction; air quality
1. Introduction
As the largest developing country in the world, China has suffered from serious air pollution
due to the rapid economic development and industrial reconstruction in the past three decades. It has
become one of the major environmental concerns in some Chinese cities, including Beijing, Shanghai,
Guangzhou and Nanjing [
1
]. Some major air pollutants in the atmosphere, including particulate matter
(PM), sulfur dioxide (SO
2
), carbon monoxide (CO), nitrogen dioxide (NO
2
) and ozone (O
3
), have
attracted increasing attention due to their impacts on air quality, visibility reduction, human health and
global climate [
2
–
6
]. From June 2000, the China National Environmental Monitoring Center started
Atmosphere 2017, 8, 100 49 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2017, 8, 100
reporting the status of ambient air quality by using an air pollution index (API), which is calculated
based on the highest index of 24-h average concentrations of PM
10
,SO
2
and NO
2
. From March 2012,
the Ministry of Environmental Protection (MEP) started using the air quality index (AQI), which is
calculated based on the concentrations of the six major pollutants including PM
2.5
,PM
10
,SO
2
,NO
2
,
CO and O
3
. The AQI ranges from 0 to 500. The greater the value of the AQI, the higher the level of air
pollution. Currently, the AQI is widely used to describe the quality of the air.
Nanjing, the capital city of Jiangsu Province, is located in the western Yangtze River Delta and has
a population of over 8.2 million. As a highly industrialized and urbanized city, Nanjing suffers from
severe air pollution. It hosted the 2nd Summer Youth Olympic Games (YOG) from 17 to 28 August
2014. In view of the significant air pollution in Nanjing, to ensure good air quality during the YOG,
the Nanjing municipal government actively took temporary control measures to reduce pollutant
emissions. These measures mainly included dust control, coal-burning control, vehicle emission
control, industrial production halts, construction site shutdowns and regional joint prevention and
control. In addition, other cities in the surrounding area cooperated with Nanjing to guarantee good
air quality during the same period.
In recent years, China has hosted many major international events. The local government has
implemented numerous stringent local emission standards, which significantly reduced the emissions
and concentrations of air pollutants in the city. During the 2008 Beijing Olympics Games, the mean
concentration of SO
2
,PM
2.5
and NO
2
were reduced by 51.0%, 43.7% and 13% compared to the period
before Olympic Games in Beijing and its surrounding area [
7
]. Wang et al. [
8
] demonstrated that the
concentration of O
3
,SO
2
, CO and NOx decreased 23%, 61%, 25% and 21%, respectively, compared
to previous years. Wang et al. [
9
] found that the average concentrations of PM
2.5
,PM
10
,SO
2
and
NO
2
during the Asia-Pacific Economic Cooperation period decreased by 47%, 36%, 62% and 41%
respectively, whereas concentrations of O
3
increased by 102%. Meanwhile, emission control measures
which were implemented in Shanghai and Guangzhou also successfully improved air quality for the
World Expo and the Asian Games [10,11].
In this study, we used air quality and air pollutants measurement data at nine sites in Nanjing
City from July to September 2014 to quantify the efficiency of the emission control measures in YOG.
The objectives of this study were (i) to show the temporal and spatial characteristics of air quality
and air pollutants during the YOG, and (ii) to give some suggestions for controlling air pollution in
large cities.
2. Data and Methods
In this paper, the AQI and hourly concentrations data for six air pollutants (PM
2.5
,PM
10
,
SO
2
,NO
2
, CO and O
3
) were provided by the China National Environmental Monitoring Center
(CNEMC). The AQI approach is based on the National Ambient Air Quality Standards of China
(NAAQS-2012) [
12
]. There are nine sites located in Nanjing, including Maigao Bridge (MG),
Caochangmen (CC), Shanxi Road (SX), Zhonghuamen (ZH), Ruijin Road (RJ), Xuanwu Lake (XW),
Pukou (PK), Olympic Stadium (OS) and Xianlin University Town (XL). To show the effectiveness of
the implementation of control measures, we acquired hourly monitoring data from July to September
2014. We divided the data into three time periods, before the period of the emission reduction
(
1 to 31 July 2014
), during the period of the emission reduction (
1 to 31 August 2014
) and after the
period of the emission reduction (
1 to 30 September 2014
), to assess the impacts of emission reduction
measures on air quality. In addition, we also recorded the data of the AQI in the surrounding cities,
including Shanghai (SH), Hangzhou (HZ1), Huzhou (HZ2), Jiaxing (JX), Hefei (HF), Maanshan (MAS),
Wuhu (WH), Xuancheng (XC), Chuzhou (CZ1), Bengbu (BB), Suzhou (SZ), Wuxi (WX), Changzhou
(CZ2), Zhenjiang (ZJ), Yangzhou (YZ), Huaian (HA), Taizhou (TZ), Nantong (NT), Xuzhou (XZ),
Yancheng (YC), Suqian (SQ) and Lianyungang (LYG), which are shown in Figure 1.
Meteorological data, including air temperature (T), relative humidity (RH), wind speed (WS)
and atmospheric pressure (AP), were obtained from the China Meteorological Data Sharing Service
50
Books
MDPI
Atmosphere 2017, 8, 100
System Administration (http://data.cma.cn/site/index.html), and solar radiation (SR) was recorded
by Watchdog weather station (Watchdog 2900ET, Spectrum Technologies, Inc., USA) on the campus of
Nanjing University of Information Science & Technology (NUIST) (32
◦
14
N, 118
◦
42
E).
This study analyzed temporal variation and characteristics of air quality and air pollutants in
different periods using the measured time-series. All graphs were created using origin 9.0 software
(Origin Lab, Northampton, MA, USA).
Figure 1. Distribution of air quality monitoring stations in Nanjing and surrounding cities.
3. Results and Discussion
3.1. Variation in Meteorological Conditions
Air temperature, relative humidity, wind speed, solar radiation and atmospheric pressure
were important external factors affecting the concentration of pollutants. In order to evaluate the
effectiveness of the implementation of control measures, we need to eliminate the interference of
meteorological conditions, so we compare meteorological parameters at different periods. Table 1
shows the mean of meteorological parameters before the emission reduction (July, 2014), during the
emission reduction (August, 2014) and during the same period in 2013 (August, 2013). The mean air
temperature during the period of the emission reduction was 25.6
◦
C, which was slightly lower than
the average level in July 2014. Relative humidity in August 2013 was far lower than the average level
in August 2014, while relative humidity in July 2014 was close to the average level in August 2014.
The average wind speed during the emission reduction period was 2.19 m/s, which was also very
close to the period before the emission reduction. Solar radiation during the emission reduction was
very much lower than that in 2013 for the same time, while solar radiation in July 2014 was higher
than the average level in August 2014. Atmospheric pressure was not much different across the three
periods. Overall, compared with the period during the emission reduction, solar radiation in July 2014
was higher, and other meteorological parameters were close to the average level during the emission
reduction. By comparison, we found that the meteorological parameters in July 2014 were close to
the average level during the emission reduction, but were very different from that in August 2013.
51
Books
MDPI
Atmosphere 2017, 8, 100
Thus, the meteorological conditions in July 2014 were more appropriate than those in August 2013 to
evaluate the effect of air quality improvement during the period of the emission reduction.
Table 1. Comparison of meteorological conditions during the different periods.
Air Temperature
(
◦
C)
Relative
Humidity (%)
Wind Speed
(m/s)
Solar Radiation
(w/m
2
)
Atmospheric
Pressure (hPa)
August 2013 30.8 65.9 3.09 205.6 1000.9
July 2014 27.1 83.4 2.26 161.6 1001.5
August 2014 25.6 87.1 2.19 127.7 1003.5
3.2. Comparison of Air Quality and Air Pollutants in Different Periods
During the emission reduction, a number of control measures were adopted to guarantee good
air quality in Nanjing. To understand the air quality and air pollutants variation characteristics during
this period, the AQI and air pollutant concentrations were plotted against time using the average daily
concentration of pollutants data in different periods (Figure 2). Table 2 shows the mean value of the
AQI and the mean concentration of air pollutants during the different time periods in Nanjing.
XUH
Figure 2. Daily variation of the AQI and air pollutant concentrations in different periods.
52
Books
MDPI
Atmosphere 2017, 8, 100
Table 2. Mean value of the AQI and mean air pollutant concentrations during the different periods.
Period
Air
Quality
Index
PM
2.5
(μg·m
−3
)
PM
10
(μg·m
−3
)
SO
2
(μg·m
−3
)
NO
2
(μg·m
−3
)
CO
(mg·m
−3
)
O
3
(μg·m
−3
)
Before the
emission reduction
1–31 July 2014 93.35 66.23 93.29 13.97 36.60 0.831 72.64
During the
emission reduction
1–31 August 2014 59.65 42.44 59.01 11.12 31.09 0.760 38.39
During the Youth
Olympic Games
17–28 August 2014 53.40 37.84 49.60 10.86 32.89 0.778 39.12
After the
emission reduction
1–30 September 2014 76.20 51.27 83.70 13.66 37.67 0.732 61.78
The AQI is an indicator of air quality, which is used to describe the air quality level of a place and
ranges from 0 to 500, where higher values represent worse air quality. The air quality was the best in
August 2014 with the AQI of 59.65, an average decrease of 36.10% compared with July 2014, reflecting
a significant improvement in the air quality during the emission reduction.
During the emission reduction period, the average PM
2.5
mass concentration was 42.44
μ
g
·
m
−3
.
Compared with the period before the emission reduction, it exhibited a decrease of 35.92%. The PM
2.5
levels were much lower than the mean PM
2.5
concentration of 106
μ
g
·
m
−3
in August 2012 in
Nanjing [13].
From Figure 2, we can see that the trend and variability of PM
10
showed a similar trend to PM
2.5
.
PM
10
concentration declined from 93.29
μ
g
·
m
−3
in July 2014 to 59.01
μ
g
·
m
−3
during the emission
reduction period, a 36.75% reduction. It is suggested that the pollution control measures of halting
work at construction sites and shutting down heavy-industry factories during the emission reduction
yielded a significant effect on the concentration of PM
10
. In addition, we calculated the ratios between
PM
2.5
and PM
10
and found that the higher ratios occurred in August 2014 (0.73) and the lower ratios
occurred in August 2013 (0.50), indicating that PM
2.5
was a main source for PM
10
during the emission
reduction period.
SO
2
concentration was reduced from 13.97
μ
g
·
m
−3
in July 2014 to 11.12
μ
g
·
m
−3
during the
emission reduction period, a decrease of 20.40%. When the control measures were terminated, some
factories and enterprises returned to normal production in Nanjing and the surrounding cities after
the Youth Olympic Games, thus the SO
2
level began to rise to 13.66
μ
g
·
m
−3
. The variation of SO
2
concentration was affected mainly by the emission sources and weather conditions. Due to low
emission sources, high temperature and strong convection in the summer, the concentration of SO
2
was
much lower than in other seasons due to the influence of gas-to-particle conversion and wet scavenging.
NO
2
and CO concentrations were reduced by 15.1% and 8.54%, respectively, in comparison to the
levels recorded for July 2014. However, they showed a tendency to increase during the YOG, thus the
concentrations of NO
2
and CO were 32.89
μ
g
·
m
−3
and 0.778 mg
·
m
−3
, respectively, compared with
August 2014, which was closely related to meteorological conditions. NO
2
and CO are important
precursors for O
3
formation, and they were affected by the photochemical reaction which enhanced
with the increase of solar radiation. According to the comparison of meteorological parameters, we
found that solar radiation during the emission reduction (127.7 w
·
m
−2
) was higher than that that
during the YOG (120.31 w
·
m
−2
). This indicated that NO
2
and CO levels in August 2014 were more
likely to convert to O
3
than those of 17–28 August 2014.
Ground-level O
3
is a greenhouse gas, a strong oxidant and a secondary pollutant mainly produced
via photochemical reactions of nitrogen oxides, carbon monoxide and volatile organic compounds [
14
].
Its concentration is affected by solar radiation, air temperature and mixing/transport [
15
].
O
3
concentration in July 2014 was 72.64
μ
g
·
m
−3
, and was 38.39
μ
g
·
m
−3
during the emission reduction
period, showing a 47.15% decrease, which was significantly greater than the those other air pollutants.
As a secondary pollutant, the decrease of O
3
was due to the reduction of precursors and the weather
conditions during the period of the emission reduction.
53
Books
MDPI
Atmosphere 2017, 8, 100
3.3. Diurnal Evolution of Air Pollutant Concentrations
Figure 3 shows the diurnal variation of air pollutant concentrations in different periods. Although
the air pollutant concentrations of each period are different, the basic pattern shows that the air
pollutants diurnal variation curve in August 2014 was lower than those of the other two periods,
except for CO.
Figure 3. Diurnal variation of air pollutant concentrations in different periods.
PM
2.5
concentration was gradually reduced from midnight to 2:00–4:00, followed by a morning
peak around 9:00–10:00, except for July 2014 (before the emission reduction). Concentrations then
decreased until 16:00–18:00, after which they rose until 21:00–22:00. Therefore, two periods displayed
a bimodal pattern with peaks between 9:00–10:00 and 21:00–22:00. The morning peak was attributed
to enhanced anthropogenic activity during rush hour [
16
]. After sunrise, strong turbulence in the
developing convective boundary layer led to lower PM
2.5
concentration. During the night, lower mixed
layer height led to higher PM
2.5
concentration. The diurnal variation characteristic of PM
10
was similar
to that of PM
2.5
concentration, which reflected the positive effects of the emission reduction measures.
SO
2
concentration showed a unimodal pattern; the maximum SO
2
concentrations tended to appear
around 9:00–10:00 during the day, which was similar to that of PM
2.5
. The mixing layer was high
and photochemical conversion was strong at this time, as SO
2
in the upper air was transferred to the
ground which led to the peak [
9
]. The main source of NO
2
in Nanjing is automobile exhaust. Diurnal
variation of NO
2
presented a double-peak curve. The first peak appeared at 08:00 after the morning
peak traffic, then the concentration of NO
2
was reduced by photolysis due to the enhancement of solar
radiation, and the minimum values of NO
2
concentration occurred at 13:00–14:00. With the arrival of
the evening peak traffic, NO
2
concentration gradually increased from 13:00–14:00 to 21:00–22:00, and
another significant peak appeared at 21:00–22:00, similar to that observed in Chengdu City [17].
The highest CO hourly concentration appeared at around 08:00–10:00 in the morning, followed
by a sharply decreasing trend to the lowest value at around 14:00–15:00, except for July 2014 (before
the emission reduction). A slight but continuous increase was observed from 14:00–15:00 to 21:00,
and finally a second weaker peak was seen at night due to the decrease in boundary layer height,
which resulted in the accumulation of pollutants. The diurnal variations of O
3
concentrations in three
periods showed similar characteristics; the lowest ozone concentrations for three periods appeared at
around 07:00–08:00 with values around 20–30
μ
g
·
m
−3
. With increasing temperature and solar radiation
54
Books
MDPI
Atmosphere 2017, 8, 100
in the daytime, O
3
concentration increased from 07:00–08:00 until 15:00, when it reached its peak.
The concentration then rapidly fell until midnight.
3.4. Spatial Variation of the AQI and Air Pollutant Concentrations
In order to analyze the spatial variation of air quality and air pollutants during the period of the
emission reduction, the mean values and standard errors of the AQI and six air pollutant concentrations
in nine different sites of Nanjing were compared with the period before the emission reduction, as
shown in Figure 4. The AQI during the emission reduction was lower than that in July 2014 at all
monitoring stations; the percentage of decrease of ranged from 31.54 to 44.90% on spatial variation,
which indicated the effects on air quality of implementing the reduction measures.
Figure 4.
Spatial variation of the AQI and air pollutants at each site before and during the
emission reduction.
Concentrations of PM
2.5
,PM
10
and O
3
during the emission reduction period decreased
significantly at nine monitoring sites compared with those in July 2014. The percentage of decrease of
PM
2.5
ranged from 31.00 to 44.91%, showing that the strict pollution control measures carried out in
Nanjing could reduce the concentration of PM
2.5
. The percentage of decrease of PM
10
ranged from
30.87 to 41.46%, which was similar to PM
2.5
in the spatial variation. The percentage of decrease of
O
3
ranged from 15.93 to 55.61%, which was the largest of the six air pollutants. The percentage of
decrease of SO
2
ranged from 3.36 to 31.89%, which was lower than that observed in Beijing during
the APEC period [
9
]. Most of the coal-burning factories were shut down during the period of the
55
Books
MDPI
Atmosphere 2017, 8, 100
emission reduction in Nanjing. Thus, the decrease of SO
2
indicated the effects of the reduction of coal
consumption [
18
]. The percentage of decrease of NO
2
ranged from 0.66 to 48.51%, which reflected
the good influence of reducing motor vehicle exhaust and industrial production. The percentage of
decrease of CO ranged from 1.43 to 25.97%, which was the smallest of the six air pollutants; moreover,
the daily average of CO has not exceeded the Chinese national assessment standard in Nanjing in
recent years.
The decrease of the AQI at the ZH site was the highest of all the sites. The decrease of PM
2.5
at
the ZH site was the highest, while the decrease of PM
2.5
at the XW site was lower than the others.
The decrease of PM
10
at the OS site was higher than those at other monitoring sites, reflecting the
effectiveness of control measures for PM
10
in the main stadium of the Nanjing Youth Olympic Games.
The PK site was located in the suburbs of Nanjing, and the decrease of SO
2
and NO
2
at the PK site was
the highest of all the monitoring sites, indicating the effectiveness of controlling motor vehicles and
industrial production. The decrease of O
3
exceeded 40% in the spatial variation, except for the RJ site,
with a decrease of 55.61% at the ZH site, which was the largest of all the monitoring sites.
3.5. Variation of Air Quality in Nanjing and the Surrounding Cities
From 1 to 31 August 2014, Nanjing cooperated with 22 surrounding cities to establish an air
pollution joint prevention group to guarantee the air quality in August in Nanjing. Figure 5 shows
the AQI of Nanjing and 22 nearby cities in July and August 2014. In July 2014, the air quality was the
worst in WX, compared with the other cities; the AQI for WX was the highest (105), followed by CZ2
and YZ. The air quality was the best in LYG, XZ and SH, where the AQI was the lowest (74). In August
2014, the maximum value of the AQI occurred in JX and the minimum value of the AQI appeared
in Nanjing. We found that the difference between the AQI in July 2014 and August 2014 was very
significant (
p < 0.05
). Figure 6 represents the percentage of decrease of the AQI during the emission
reduction period compared with July 2014 in Nanjing and 22 nearby cities. The AQI of all cities showed
a downward trend. The decrease in the AQI in Nanjing was 35.48%, which was the largest of the
23 cities, followed by YZ and HZ2, and was almost the same in ZJ, YC and WX. The decrease in the
AQI in CZ2, NT and TZ were 19.15%, 18.60% and 18.18%, respectively. The decrease in the AQI in SH,
HF, XZ, JX, HA and LYG were less than 10% and greater than 0%, and was the lowest in LYG with
only a 2.70% reduction.
Figure 5.
Variation of the AQI in Nanjing and its surrounding cities during the emission
reduction period.
56
Books
MDPI
Atmosphere 2017, 8, 100
Figure 6.
Decrease percentage of the AQI during the emission reduction period compared with July
2014 in Nanjing and its surrounding cities.
4. Discussion and Conclusions
Air pollutant concentrations are not only related to emissions from air pollution sources, but are
also influenced by meteorological conditions. In order to eliminate the interference of meteorological
conditions, we divided the meteorological data into three time periods: before, during the emission
reduction and during the same period of the previous year (2013). According to the comparison
of meteorological data in three periods, we found that most of the meteorological parameters in
July 2014 were closer to the average level during the period of the emission reduction than that in
August 2013. Therefore, we analyzed the concentrations of air pollutants before, during and after the
emission reduction, which was more appropriate than the comparison with the year 2013 to evaluate
the effectiveness of air pollution reduction actions.
According to our analysis and comparison, the concentrations of six air pollutants during the
emission reduction period were significantly lower than those in the prophase mean before the emission
reduction, especially O
3
. Compared with the average levels in July 2014, the mean concentrations of
PM
2.5
,PM
10
,SO
2
,NO
2
, CO and O
3
decreased by 35.92%, 36.75%, 20.40%, 15.05%, 8.54% and 47.15%,
respectively. All these reductions in air pollutants indicated that the emission control measures were
successful during the YOG, and little related to the local meteorological conditions. After the YOG, air
pollutants increased significantly due to the cancellation of the emission control measures, except for
CO. In addition, we also analyzed the diurnal variations of air pollutant concentrations in different
periods. Overall, the diurnal variation curve of air pollutants during the period of the emission
reduction was lower than that in July 2014, except for CO. In the spatial variation of pollutants, the
AQI during the emission reduction was lower than that in July 2014 at all monitoring stations, the
decrease of which ranged from 31.54 to 44.90% in spatial variation. Meanwhile, the decrease of PM
2.5
ranged from 31.00% to 44.91% at different sites, the decrease of PM
10
ranged from 30.87% to 41.46%,
the decrease of SO
2
ranged from 3.36% to 31.89%, the decrease of NO
2
ranged from 0.66% to 48.51%,
the decrease of CO ranged from 1.43% to 25.97%, and the decrease of O
3
ranged from 15.93% to
55.61%. Our results suggested that the AQI of Nanjing and 22 nearby cities showed a downward trend
when compared with July 2014, which reflected the effects on air quality of implementing the regional
emission reduction measures.
In recent years, China has successfully hosted many international events. In order to improve
air quality during the events, the local government has implemented some short-term periods with
temporary emission control measures to reduce the emissions of pollutants. During the 2008 Olympic
Games, the Beijing municipal government took many air pollution control measures, which included
57
Books
MDPI
Atmosphere 2017, 8, 100
relocating some heavily polluting enterprises, encouraging natural gas instead of coal-fired boilers and
domestic stoves, limiting the use of cars and so on [
19
]. During the 2010 Asian Games, the Guangzhou
government made great efforts to improve the air quality, which mainly included controlling emissions
from industries and transportation restrictions; vehicles could be driven only on alternate days
depending on license plate numbers, and all construction activities were put on hold during the
Games [
20
]. During the APEC period, the Chinese government took more stringent emission reduction
measures in Beijing and the surrounding region, including banning heavily polluting vehicles, closing
heavily polluting factories, slowing down construction activities and so on [
21
]. Some studies found
that the mass concentration of air pollutants during the events decreased significantly, except for
O
3
[
9
,
22
]. The results of our studies showed that the emission control measures were the most effective
on O
3
out of the six air pollutants, indicating that the major O
3
precursors (nitrogen oxides and volatile
organic compounds) were well controlled, and that meteorological conditions also played a positive
role during the YOG.
The Nanjing government carried out temporary strict environmental regulations to ensure good
air quality during the YOG in 2014. Other surrounding cities cooperated with Nanjing to guarantee
good air quality during the Games. Approximately 2630 construction sites were halted. Heavy-industry
factories, including the petrochemical industry, iron and steel industry, building materials industry
and so on were required to reduce manufacturing by 20%. High-emission or yellow-labeled vehicles
were not allowed to drive on the road. Open space barbecue restaurants were closed. Over 900 electric
buses and 500 electric taxis have been put into operation [
23
]. In addition, 22 surrounding cities in
the Yangtze River Delta region were asked to cooperate with Nanjing to close industries with high
pollution emissions during the emission reduction period [24].
According to the analysis of this study, several suggestions have been made to improve the air
quality of Nanjing. On the one hand, as regional emission reduction measures were the main factor in
improving the air quality in Nanjing, regional cooperation and joint defense and control are considered
the future direction in air pollution control. On the other hand, greater efforts should be exerted to
control dust, coal-burning, vehicle emissions and industrial production around Nanjing, which would
play a significant role in improving the air quality of Nanjing.
In order to protect human health and improve air quality, we may consider taking the following
several measures and recommendations for air pollution reduction over long-term period:
1.
To reduce emissions. Use more non-pollution energy, such as solar energy, wind energy and
hydropower. Reform energy structure and use low-polluting energy, such as natural gas, biogas
and alcohol. In addition, before the pollutants enter the atmosphere, use technology such as dust
and smoke abatement to eliminate the partial pollutant in the waste gas, as it can reduce the
quantity of pollutants entering the atmosphere.
2.
To curb emissions and make full use of the atmosphere’s self-purification ability. Due to different
meteorological conditions, the efficiency of the removal of pollutants in the atmosphere is different.
The pollutants concentration may be different in different regions, even when the same amount
of pollutants has been released into the atmosphere. Therefore, respond to different regions at
different times for the effective control of emissions.
3.
Factors such as the location of factory sites, chimney designs, city planning and the location of
industrial districts must be given proper attention so as to not cause redundant iterations of
pollution, which can in turn result in serious local pollution events.
4.
Afforestation enables more plants to absorb pollutants, thus reducing the degree of air pollution.
Acknowledgments: This work was financially supported by the National Natural Science Fund (41475108).
Author Contributions:
Hui Zhao and Youfei Zheng conceived and designed the experiments as well as writing
the paper; Hui Zhao analyzed the data; Ting Li helped perform the statistical analysis.
Conflicts of Interest: The authors declare no conflict of interest.
58
Books
MDPI
Atmosphere 2017, 8, 100
References
1. Chan, C.K.; Yao, X.H. Air pollution in mega cities in China. Atmos. Environ. 2008, 42, 1–42. [CrossRef]
2.
Gao, J.J.; Tian, H.Z.; Cheng, K.; Lu, L.; Zheng, M.; Wang, S.X.; Hao, J.M.; Wang, K.; Hua, S.B.; Zhu, C.Y.;
Wang, Y. The variation of chemical characteristics of PM
2.5
and PM
10
and formation causes during two haze
pollution events in urban Beijing, China. Atmos. Environ. 2015, 107, 1–8. [CrossRef]
3.
Leiva, M.A.; Santibanez, D.A.; Ibarra, S.; Matus, P.; Seguel, R. A five-year study of particulate matter (PM
2.5
)
and cerebrovascular diseases. Environ. Pollut. 2013, 181, 1–6. [CrossRef] [PubMed]
4.
Pascal, M.; Falq, G.; Wagner, V.; Chatignoux, E.; Corso, M.; Blanchard, M.; Host, S.; Pascal, L.; Larrieu, S.
Short-term impacts of particulate matter (PM
10
,PM
10–2.5
,PM
2.5
) on mortality in nine French cities.
Atmos. Environ. 2014, 95, 175–184. [CrossRef]
5.
Shankardass, K.; Jerrett, M.; Dell, S.D.; Foty, R.; Stieb, D. Spatial analysis of exposure to traffic-related air
pollution at birth and childhood atopic asthma in Toronto, Ontario. Health Place
2015
, 34, 287–295. [CrossRef]
[PubMed]
6.
Yang, Y.R.; Liu, X.G.; Qu, Y.; Wang, J.L.; An, J.L.; Zhang, Y.H.; Zhang, F. Formation mechanism of continuous
extreme haze episodes in the megacity Beijing, China, in January 2013. Atmos. Res.
2015
, 155, 192–203.
[CrossRef]
7.
Xin, J.Y.; Wang, Y.S.; Tang, G.Q.; Wang, L.L.; Sun, Y.; Wang, Y.H.; Hu, B.; Song, T.; Ji, D.S.; Wang, W.F.; et al.
Variability and reduction of atmospheric pollutants in Beijing and its surrounding area during the Beijing
2008 Olympic Games. Chin. Sci. Bull. 2010, 55, 1937–1944. [CrossRef]
8.
Wang, W.T.; Primbs, T.; Tao, S.; Simonich, S.M. Atmospheric Particulate Matter Pollution during the 2008
Beijing Olympics. Environ. Sci. Technol. 2009, 15, 5314–5320. [CrossRef]
9.
Wang, Z.S.; Li, Y.T.; Chen, T.; Li, L.J.; Liu, B.X.; Zhang, D.W.; Sun, F.; Wei, Q.; Jiang, L.; Pan, L.B. Changes
in atmospheric composition during the 2014 APEC conference in Beijing. J. Geophys. Res. Atmos.
2015
, 120,
695–707. [CrossRef]
10.
Hao, N.; Valks, P.; Loyola, D.; Cheng, Y.F.; Zimmer, W. Space-based measurements of air quality during the
World Expo 2010 in Shanghai. Environ. Res. Lett. 2011, 6, 67–81. [CrossRef]
11.
Wu, F.C.; Xie, P.H.; Li, A.; Chan, K.L.; Hartl, A.; Wang, Y.; Si, F.Q.; Zeng, Y.; Qin, M.; Xu, J.; et al.
Observations of SO2 and NO2 by mobile DOAS in the Guangzhou eastern area during the Asian Games
2010. Atmos. Meas. Tech. 2013, 6, 2277–2292. [CrossRef]
12.
Hu, J.L.; Ying, Q.; Wang, Y.G.; Zhang, H.L. Characterizing multi-pollutant air pollution in China: Comparison
of three air quality indices. Environ. Int. 2015, 84, 17–25. [CrossRef] [PubMed]
13.
Shen, G.F.; Yuan, S.Y.; Xie, Y.N.; Xia, S.J.; Li, L.; Yao, Y.K.; Qiao, Y.Z.; Zhang, J.; Zhao, Q.Y.; Ding, A.J.; et al.
Ambient levels and temporal variations of PM2.5 and PM10 at a residential site in the mega-city, Nanjing, in
the western Yangtze River Delta, China. J. Environ. Sci. Health. A. 2014, 49, 171–178.
14.
Stella, P.; Personne, E.; Loubet, B.; Lamaud, E.; Ceschia, E.; Beziat, P.; Bonnefond, J.M.; Irvine, M.; Keravee, P.;
Mascher, N.; et al. Predicting and partitioning ozone fluxes to maize crops from sowing to harvest:
The Surfatm-O3 model. Biogeosciences 2011, 8, 2869–2886. [CrossRef]
15.
Leung, L.R.; Gustafson, W.I. Potential regional climate change and implications to U.S. air quality.
Geophys. Res. Lett. 2005, 32, 367–384. [CrossRef]
16.
Zhao, X.J.; Zhang, X.L.; Xu, X.F.; Xu, J.; Meng, W.; Pu, W.W. Seasonal and diurnal variations of ambient PM2.5
concentration in urban and rural environments in Beijing. Atmos. Environ. 2009, 43, 2893–2900. [CrossRef]
17.
Xie, Y.Z.; Pan, Y.P.; Ni, C.J.; Chen, Z.H.; Wei, X. Temporal and spatial variations of atmospheric pollutants in
urban Chengdu during summer [in Chinese]. Acta. Sci. Circumst. 2015, 35, 975–983.
18.
Apergis, N.; Loomis, D.; Payne, J.E. Are fluctuations in coal consumption transitory or permanent? Evidence
from a panel of US states. Atmos. Environ. 2010, 87, 2424–2426. [CrossRef]
19.
Wang, S.X.; Zhao, M.; Xing, J.; Wu, Y.; Zhou, Y.; Lei, Y.; He, K.B.; Fu, L.X.; Hao, J.M. Quantifying the air
pollutants emission reduction during the 2008 Olympic Games in Beijing. Environ. Sci. Technol.
2010
, 44,
2490–2496. [CrossRef] [PubMed]
20.
Liu, H.; Wang, X.M.; Zhang, J.P.; He, K.B.; Wu, Y.; Xu, J.Y. Emission controls and changes in air quality in
Guangzhou during the Asian Games. Atmos. Environ. 2013, 76, 81–93. [CrossRef]
21.
Li, R.P.; Mao, H.J.; Wu, L.; He, J.J.; Ren, P.P.; Li, X.Y. The evaluation of emission control to PM concentration
during Beijing APEC in 2014. Atmos. Pollut. Res. 2016, 7, 363–369. [CrossRef]
59
Books
MDPI
Atmosphere 2017, 8, 100
22.
Wang, Z.S.; Li, Y.T.; Zhang, D.W.; Chen, T.; Sun, F.; Li, L.J.; Li, J.X.; Sun, N.D.; Chen, C.; Wang, B.Y. Analysis
on air quality in Beijing during the 2014 APEC conference [in Chinese]. J. Environ. Sci. 2016, 36, 675–683.
23.
Ding, J.; Vander, A.R.J.; Mijling, B.; Levelt, P.F.; Hao, N. NOx emission estimates during the 2014 Youth
Olympic Games in Nanjing. Atmos. Chem. Phys. 2015, 15, 6337–6372. [CrossRef]
24.
Li, S.W.; Li, H.B.; Luo, J.; Li, H.M.; Qian, X.; Liu, M.M.; Bi, J.; Cui, X.Y.; Ma, L.Q. Influence of pollution
control on lead inhalation bioaccessibility in PM
2.5
: A case study of 2014 Youth Olympic Games in Nanjing.
Environ. Int. 2016, 94, 69–75. [CrossRef] [PubMed]
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
60
Books
MDPI
atmosphere
Article
Characterization of Particulate Matter (PM
2.5
and
PM
10
) Relating to a Coal Power Plant in the Boroughs
of Springdale and Cheswick, PA
Casey D. Bray *, William Battye, Pornpan Uttamang, Priya Pillai and Viney P. Aneja
Department of Marine, Earth, and Atmospheric Sciences, North Carolina State University, Raleigh, NC 27695,
USA; whbattye@ncsu.edu (W.B.); puttama@ncsu.edu (P.U.); prpillai@ncsu.edu (P.P.); vpaneja@ncsu.edu (V.P.A.)
* Corresponding: cdbray@ncsu.edu; Tel.: +1-919-515-3690
Received: 7 August 2017; Accepted: 18 September 2017; Published: 23 September 2017
Abstract:
Ambient concentrations of both fine particulate matter (PM
2.5
) and particulate matter
with an aerodynamic diameter less than 10 micron (PM
10
) were measured from 10 June 2015 to
13 July 2015 at three locations surrounding the Cheswick Power Plant, which is located between
the boroughs of Springdale and Cheswick, Pennsylvania. The average concentrations of PM
10
observed during the periods were 20.5
±
10.2
μ
gm
−3
(Station 1), 16.1
±
4.9
μ
gm
−3
(Station 2)
and 16.5
±
7.1
μ
gm
−3
(Station 3). The average concentrations of PM
2.5
observed at the stations
were 9.1
±
5.1
μ
gm
−3
(Station 1), 0.2
±
0.4
μ
gm
−3
(Station 2) and 11.6
±
4.8
μ
gm
−3
(Station 3).
In addition, concentrations of PM
2.5
measured by four Pennsylvania Department of Environmental
Protection air quality monitors (all within a radius of 40 miles) were also analyzed. The observed
average concentrations at these sites were 12.7
±
6.9
μ
gm
−3
(Beaver Falls), 11.2
±
4.7
μ
gm
−3
(Florence), 12.2
±
5.3
μ
gm
−3
(Greensburg) and 12.2
±
5.5
μ
gm
−3
(Washington). Elemental analysis
for samples (blank – corrected) revealed the presence of metals that are present in coal (i.e., antimony,
arsenic, beryllium, cadmium, chromium, cobalt, lead, manganese, mercury, nickel and selenium).
Keywords: PM
2.5
;PM
10
; coal-fired power plant; particulate matter emissions
1. Introduction
With over 600 active coal-fired power plants in the US, 39% of the country’s total electricity
generation is attributed to coal (Available online: http://www.eia.gov). While these plants may
bring jobs and prosperity to their surrounding regions, they also emit dangerous pollutants into
the atmosphere. According to the US Environmental Protection Agency (EPA) National Emissions
Inventory, coal-fired power plants emit sulfur dioxide, oxides of nitrogen and particulate matter into
the atmosphere. They also emit 84 of the 187 Hazardous Air Pollutants (HAP) regulated by the US
EPA [
1
]. In addition to the stack emissions from the coal-fired power plant, coal handling may also
emit pollutants into the atmosphere and thus degrade the air quality in the vicinity near the power
plant [
2
]. Several studies have been conducted on the relationship between particulate matter and
emissions from coal-fired power plants [
2
–
14
]. Particulate matter measured near a coal-fired power
plant is known to contain a number of harmful chemicals that are present in coal and are also known
to have carcinogenic properties, such as antimony, arsenic, beryllium, cadmium, chromium, cobalt,
lead, manganese, mercury, nickel, selenium and polycyclic organic matter (POM) [
2
–
4
]. All of these
chemicals are known to be hazardous and are thus regulated by the US EPA as HAPs under the Clean
Air Act, as amended in 1990 [
15
]. The emission of these pollutants into the atmosphere is known to be
dangerous to both human health and welfare. Exposure to high concentrations of HAPs can lead to a
number of adverse health effects such as damage to the eyes, skin, lungs, kidneys and the nervous
system, and can even cause cancer, pulmonary disease and cardiovascular disease [
14
,
16
]. Particulate
Atmosphere 2017, 8, 186 61 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2017, 8, 186
matter (PM
2.5
), which is particulate matter with an aerodynamic diameter of 2.5 microns or less, is also
a dangerous atmospheric pollutant due to its small size, which can travel deep into people’s lungs and
lead to a number of severe health effects. Elevated concentrations of PM
2.5
are known to be associated
with cardiovascular issues (heart disease, heart attacks, etc.) as well as respiratory issues, reproductive
issues and even cancer [
16
,
17
]. In addition to harming human health, coal-fired power plants can
also lead to a number of environmental impacts as well, such as acidification of the environment,
bioaccumulation of toxic metals, the contamination of water sources, reduced visibility due to haze as
well as degradation of buildings and monuments.
The Cheswick Power Plant is located in the southwestern part of Pennsylvania along the Allegheny
River right between the boroughs of Springdale (40.5414
◦
N, 79.7821
◦
W) and Cheswick (40.5416
◦
N,
79.8002
◦
W) (Figure 1). While the Cheswick Power Plant has brought economic benefits to the small
boroughs of Springdale and Cheswick, PA, the plant also brought with it a multitude of harmful
effects on both human health and welfare. According to the Allegheny County Health Department’s
Point Source Emissions Inventory Report, the Cheswick Power Station is the largest point source
emitter of both criteria pollutants and hazardous air pollutants (HAP) in Allegheny County, PA [
18
].
In 2006, the plant was listed as the 17th Dirtiest Power Plant for sulfur dioxide [
1
]. In 2010, the plant
once again made headlines, ranking 41st in the Top Power Plant Hydrochloric Acid Emitters and
91st in the Top US Power Plant Lead Emitters [
19
]. Since then, significant efforts have been made to
reduce emissions [
18
]. Overall, emissions of carbon monoxide and sulfur dioxide did decrease in 2011,
while emissions of nitrogen oxides, particulate matter, NO
x
, and volatile organic compounds (VOC)
increased [
18
]. However, the plant remains Allegheny County’s largest point source emitter of both
criteria pollutants and hazardous air pollutants. The objective of this study was to measure the air
quality at three different locations within the two boroughs surrounding the power plant in order to
determine concentrations of both PM
2.5
and PM
10
due to emissions from the power plant. Particulate
matter measurements measured by the PA Department of Environmental Protection (DEP) were also
analyzed for the study period and compared against the measurements taken for this study.
2. Experiments
2.1. NCSU Monitoring Stations
Three experimental sites were set up within Springdale and Cheswick, PA, each within a mile of
the Cheswick power plant (Figure 1). There were two sampling periods in this study. Two of the three
sites took samples from 10 June 2015 to 27 June 2015 and then from 30 June 2015 to
13 July 2015
. Station 1
was located at 244 Center St., Springdale, PA (40.5402
◦
N, 79.7884
◦
W), which is less than half of a mile
from the power plant. Samples were taken at this location during both sampling periods. Station 2 was
located at 200 Hill Avenue, Cheswick, PA (40.542
◦
N, 79.80
◦
W), which is approximately one mile from
the power plant. The sampling period for this study was from
10 June 2015
to 27 June 2015. The third
station was located at 1212 Fairmont St, Cheswick, PA (40.5465
◦
N, 79.8031
◦
W). The sampling period
for this site was from 30 June 2015 to 13 July 2013. All three of these sites are in residential areas, very
close to the power plant. The monitoring sites were chosen considering the prevailing wind direction
(west southwesterly) such that during both sampling periods, one site was upwind of the power
plant (those sites in Cheswick) and one site was downwind of the power plant (the site in Springdale).
However, it must be noted that the wind during the sampling period was highly variable.
Each experimental site was equipped to measure PM
2.5
and PM
10
using Reference Ambient Air
Sampler (RAAS) high volume air samplers by Anderson Instruments as well as meteorological data
measured by Met One Model Automet portable weather stations equipped with an onboard data
logger. The sampling period for each instrument was set to 24 h, and the air volumetric flow on both
types of instruments was set at the standard flow rate of 16.67 Liters/minute. The sampling filters
used to collect both PM
2.5
and PM
10
were 47 mm diameter Teflon filters. For PM
2.5
sampling, there
was an additional smaller filter used (~39 mm diameter) in the WINS impactor, which was wetted
62
Books
MDPI
Atmosphere 2017, 8, 186
with impactor oil, in order to restrict particles larger than 2.5 micron from reaching the 47 mm Teflon
filter. All the filters were pre-tared and numbered.
Figure 1.
Location of monitoring stations in relation to the Cheswick Power Plant and the two boroughs
of Springdale and Cheswick.
High Volume (Hi Vol) 47 mm MTL Teflon filters were used to analyze mass concentration
of PM
10
while Hi Vol 39 mm Teflon filters were used to analyze mass concentration of PM
2.5
.
Filter weighing measurements took place inside a temperature/relative humidity-controlled ISO
Class 6 (<1000 PM
0.5
/cf) clean room employing a draft-shielded microbalance with anti-static wand.
Temperature and relative humidity were controlled to 21
◦
C and 35%, respectively. The filter mass
was determined through five weightings of the filter, with each weight bracketed by a reading of the
internal zero of the balance. A buoyancy correction was applied to the mass and the average of the
zeros bracketing the mass was then subtracted from the result. This was repeated four additional times,
with the average of these five results being reported as the final mass. The gravitational analysis was
performed by the Research Triangle Park’s (RTP) office of Applied Research Associates, Inc. (ARA).
Ten samples were chosen from this study to undergo an inorganic analysis, i.e., five for PM
2.5
and
five for PM
10
. In addition, two sample blanks were analyzed for calibration. Samples were analyzed
from all three monitoring locations. The samples were selected to represent a mix of particulate matter
concentrations collected during times when the plant was running. The inorganic analysis was then
performed by X-Ray Fluorescence (XRF) and then Ion Chromatography (IC) analysis.
2.2. PA PM
2.5
Monitoring Sites
In addition to the measurement sites used in the study, data from four monitoring stations run
by the state of PA (PA Department of Environmental Protection) was also used to compare against
measurements taken in the field campaign. The four monitoring sites used in this study (Figure 2)
were Beaver Falls (40.7478
◦
N, 80.3157
◦
W), Florence (40.4454
◦
N, 80.4212
◦
W), Greensburg (40.3043
◦
N,
79.5060
◦
W) and Washington (40.1706
◦
N, 80.2617
◦
W). These four sites contain continuous Met One
BAM 1020 PM
2.5
monitors [
20
]. It is important to note that while the NCSU monitoring sites were
located fairly close to the Cheswick Power Plant (i.e., less than 5 miles), the PA DEP monitoring sites
were located 20–40 miles from the power plant.
63
Books
MDPI
Atmosphere 2017, 8, 186
Cheswick Powerplant
Figure 2.
Location of the PA DEP monitoring sites in relation to the NCSU monitoring stations.
The PA DEP monitoring sites are represented by the green stars while the NCSU monitoring sites are
represented by the blue dots.
3. Results and Discussion
The observed average 24-hour concentrations of both PM
2.5
(0.02–22.76
μ
gm
−3
) and PM
10
(0.9
−
44.15
μ
gm
−3
) from the NCSU monitoring sites were found to be lower than the US EPA National
Ambient Air Quality Standard of 35 μgm
−3
and 150 μgm
−3
, respectively (Figure 3).
The average concentrations of PM
10
observed during the periods were 20.5
±
10.2
μ
gm
−3
(Station 1), 16.1
±
4.9
μ
gm
−3
(Station 2) and 16.5
±
7.1
μ
gm
−3
(Station 3). The average concentrations
of PM
2.5
observed at the stations were 9.1
±
5.1
μ
gm
−3
(Station 1), 0.2
±
0.4
μ
gm
−3
(Station 2)
and 11.6
±
4.8
μ
gm
−3
(Station 3). Station 1 observed the maximum concentrations of both PM
2.5
(22.76
μ
gm
−3
) and PM
10
(44.15
μ
gm
−3
μ
gm
−3
). While Station 3 also observed higher concentrations
of particulate matter (both PM
2.5
and PM
10
), the results of a t test (at
α
= 0.05) indicated that the
concentrations at Station 1 were indeed higher than those observed at Stations 2 and 3. There are
several plausible causes for this: the location of Station 1 versus Stations 2 and 3 in relation to
the meteorological conditions as well as additional potential sources of particulate matter into the
atmosphere (i.e., traffic emissions, roadway dust, transport). Concentrations of particulate matter
tended to be higher when conditions were warm and sunny. Conversely, the lowest concentrations
of particulate matter were primarily observed during rainy conditions. This was expected due
to the removal of the particulates from the atmosphere through wet deposition. However, there
were a few days where elevated concentrations were observed when the plant was running during
rainy conditions. This may have resulted from the stability of the atmosphere in these conditions,
which allows for the atmospheric accumulation of particulate matter. Nevertheless, comparisons of
particulate matter concentrations with meteorological conditions (Figures 4 and 5) failed to show any
strong correlations.
64
Books
MDPI
Atmosphere 2017, 8, 186
Figure 3.
The daily 24-hour concentration of PM
10
and PM
2.5
at Station 1 (blue), Station 2 (magenta)
and Station 3 (gold) plotted against the US EPA National Ambient Air Quality Standards value for
24-hour average PM
10
concentration (red).
Figure 4.
Comparing concentrations of PM
2.5
with meteorological variables for both Station 1 and
Stations 2 and 3, denoted by Station 2 in this figure.
65
Books
MDPI
Atmosphere 2017, 8, 186
Figure 5.
Comparing concentrations of PM
10
with meteorological variables for both Station 1 and
Stations 2 and 3, denoted by Station 2 in this figure.
When comparing the median wind speed versus the concentration of PM
10
at Stations 1 and 2,
positive correlations of 0.2 and 0.27, respectively, were observed. When comparing the median
relative humidity with the concentration of PM
10
, negative correlations of 0.3 and 0.44 were observed,
respectively, while minor positive correlations of 0.08 and 0.12 were observed for temperature and
concentration at Stations 1 and 2, respectively. When comparing meteorological conditions with
concentrations of PM
2.5
at Stations 1 and 2, a negative correlation of 0.32 between wind speed and
concentration of PM
2.5
was observed at Station 2 while Station 1 observed a positive correlation of 0.12.
The strongest meteorological correlation was a negative correlation of
−
0.56 between relative humidity
and the concentration of PM
2.5
at Stations 1, which is nominally downwind from the plant. In contrast,
the correlation between relative humidity and PM
2.5
was weakly positive, at 0.09, for Station 2. Positive
correlations of 0.54 and 0.46 were observed when comparing temperature versus concentrations of
PM
2.5
at Stations 1 and 2, respectively. However, when considering these correlations, it is important
to acknowledge the short sample period and thus small sample.
The median wind direction at upwind versus downwind stations was compared with the average
24-hour concentrations of particulate matter, where the categorizations of upwind and downwind
were denoted based on the daily median wind direction. The results of a t test provide p = 0.5 and
p = 0.4, for concentrations of PM
10
and PM
2.5
, respectively, which suggests that the relationship is not
statistically significant.
Despite the fact that the median wind direction was not significantly correlated with the
concentrations of particulate matter, the local hourly wind direction likely did impact concentrations
of particulate matter. Furthermore, it is likely that some of the relatively high concentrations of both
particulate matter observed at Station 1 were found to have come from the direction of the coal site on
the power plant’s property (Figures 6 and 7). In addition to observing elevated concentrations coming
directly from the plant, elevated concentrations of both PM
2.5
and PM
10
were also observed coming
from other directions. This can likely be attributed to a number of sources, such as emissions from
vehicles, road dust. In addition, the elevated concentrations can also be attributed to transport due to
local and regional meteorological phenomena. Furthermore, fairly elevated concentrations of PM
10
are
observed at faster wind speeds. While wind speeds would generally reduce ambient concentrations of
particulate matter in the atmosphere, the observed elevated concentrations could simply be due to an
emission source located extremely close to the monitoring site (i.e., a stalled truck on the side of the road).
66
Books
MDPI
Atmosphere 2017, 8, 186
Figure 6.
Wind rose created for PM
2.5
concentrations measured at Station 1. Green colors represent
the highest concentrations of particulate matter while the orange and red colors represent lower
concentrations of particulate matter, as shown in the legend. Units of particulate matter concentration
are
μ
gm
−3
. The upper side of the wind rose represents North. The length of the wind rose vectors
represents average wind speed, where the longer vectors represent a higher average wind speed.
Figure 7.
Wind rose created for PM
10
concentrations measured at Station 1. Green colors represent
the highest concentrations of particulate matter while the orange and red colors represent lower
concentrations of particulate matter, as shown in the legend. Units of particulate matter concentration
are
μ
gm
−3
. The upper side of the wind rose represents North. The length of the wind rose vectors
represents average wind speed, where the longer vectors represent a higher average wind speed.
67
Books
MDPI
Atmosphere 2017, 8, 186
In contrast to this, neither Stations 2 nor 3 (not pictured) showed extremely high concentrations
coming from the power plant. However, this is not entirely surprising for two reasons: the first reason
is that the winds were primarily such that those stations were upwind of the power plant and that
both of these stations were twice as far as Station 1 for the plant, thus allowing for some dispersion of
pollutants emitted from the plant and the coal site to occur.
In addition to this, concentrations of both PM
2.5
and PM
10
were compared at calm wind conditions
(median wind speeds less than 1 kt) and non-calm wind conditions (Table 1). When comparing average
concentrations of PM
2.5
during calm winds (9.29
±
5.59
μ
gm
−3
) with average concentrations during
not calm winds (3.53
±
5.18
μ
gm
−3
), it is clear that the average concentrations during calm winds are
higher than the average concentrations during not calm winds. Similarly, average concentrations of
PM
10
observed during calm winds (19.83
±
10.29
μ
gm
−3
) was higher than the average concentrations
observed when the winds were not calm (16.25
±
4.87
μ
gm
−3
). Two t tests were also conducted in
order to determine whether or not the comparison between concentrations of PM
2.5
and PM
10
during
calm and not calm winds was statistically significant. The results of the t tests provide p = 0.052 and
p = 0.001, for concentrations of PM
10
and PM
2.5
, respectively. This suggests that while the comparison
between concentrations of PM
2.5
during calm and not calm winds is statistically significant, the
comparison for concentrations of PM
10
is not quite statistically significant.
Concentrations of particulate matter were also compared with the plant’s gross load during
the period (power plant gross load obtained via: EPA. Available online: http://www.ampd.epa.
gov/ampd). Figure 8 compares the Cheswick Power Plant’s daily gross load with the daily 24-hour
average concentration of PM
10
and PM
2.5
. When comparing the daily gross load with the particulate
matter concentrations, the correlation coefficients (ranging between
−
0.23 and 0.15) suggest that the
concentrations of particulate matter at each station are not statistically correlated with the plant gross load.
Table 1.
Comparison of concentrations of PM
2.5
and PM
10
(
μ
gm
−3
) during both calm and not calm
wind conditions. Calm wind conditions are considered to be conditions where the median wind speed
is less than 1 kt.
Calm Winds
(μgm
−3
)
Not Calm Winds
(μgm
−3
)
PM
2.5
PM
10
PM
2.5
PM
10
Maximum 22.76 44.15 15.05 29.31
Minimum 0.16 0.91 0.02 8.30
Mean 9.29 19.83 3.53 16.25
Median 8.26 20.55 0.35 15.03
Standard Deviation 5.59 10.29 5.18 4.87
Number of Samples 31 33 15 16
Five PM
10
samples and five PM
2.5
samples were subjected to XRF and IC analysis. The XRF
analysis represents the PM samples as elements, while the IC analysis represents the PM samples as
ions. When comparing the results of the analyses, it is evident that they are consistent. The results of
the XRF analysis (Figure 9) showed that the primary constituents of PM
10
are sulfur
(0.66–1.77 μgm
−3
)
,
silicon (0.14–2.47
μ
gm
−3
), aluminum
(0.03–1.06 μgm
−3
)
and iron
(0.08–0.95 μgm
−3
)
. Similarly, the
primary constituents of PM
2.5
were found to be sulfur (0.41–1.17
μ
gm
−3
), silicon (0.02–0.23
μ
gm
−3
)
and iron (0.03–0.06
μ
gm
−3
). Particulate matter is typically composed of a complex mixture of chemicals
that are strongly dependent on source characteristics. Inorganic analysis of the particulate matter
revealed the presence of antimony, arsenic, beryllium, cadmium, chromium, cobalt, lead, manganese,
mercury, nickel and selenium (Figure 9). All of these metals present in the PM samples are known
to be present in coal [
21
]. While these elements are all found within coal and coal ash, they are also
present in crustal material. Based on the small number of samples analyzed, it was not possible to
discern differences between the upwind and downwind samples. The highest concentration of sulfur
observed in a sample during the period was observed at Station 1. In this case, it is possible that
68
Books
MDPI
Atmosphere 2017, 8, 186
this can be attributed to the coal pit, which is located less than half of a mile away from the station.
However, it is also important to note that there are other coal-fired power plants located within 50 miles
of the Cheswick Power plant, which could also contribute to the elevated sulfur concentrations.
Figure 8.
Comparing concentrations of PM
10
and PM
2.5
at each station with the power plant gross load.
The blue dots represent the concentrations at Station 1, the magenta dots represent the concentrations
at Station 2 and the yellow/gold dots represent the concentrations at Station 3.
Figure 9. The results of the XRF analysis plotted on a log scale.
69
Books
MDPI
Atmosphere 2017, 8, 186
The results of the IC analysis (Figure 10) showed that the dominant species of particulate matter
in this region are primarily sulfate, nitrate and ammonium. Within the PM
2.5
samples analyzed,
concentrations of sulfate ranged from 2.31
μ
gm
−3
to 3.08
μ
gm
−3
, nitrate concentrations ranged
from 0.03
μ
gm
−3
to 0.58
μ
gm
−3
, and concentrations of ammonium ranged from 0.82
μ
gm
−3
to
1.15 μgm
−3
. Within the analyzed PM
10
samples, concentrations of sulfate ranged from 1.05
μ
gm
−3
to
4.73
μ
gm
−3
, concentrations of nitrate ranged from 0.01
μ
gm
−3
to 0.22
μ
gm
−3
and concentrations
of ammonium ranged from 0.38
μ
gm
−3
to 1.69
μ
gm
−3
. As described above, it was not possible to
discern differences between the upwind and downwind samples based on the small sample size. It is
also important to note that sulfate, nitrate and ammonium are also common constituents of particulate
matter and thus cannot be attributed entirely to the power plant.
Figure 10. The results of the IC analysis, plotted on a log scale.
The concentrations of PM
2.5
measured by four PA Department of Environmental Protection air
quality monitors were also analyzed in this study. The observed average 24-hour concentrations at
these sites were 12.7
±
6.9
μ
gm
−3
(Beaver Falls), 11.2
±
4.7
μ
gm
−3
(Florence),
12.2 ± 5.3 μgm
−3
(Greensburg) and
12.2 ± 5.5 μgm
−3
(Washington). Similar to what was observed at the NCSU
monitoring sites, the PA DEP monitors also observed 24-hour average concentrations of fine particulate
(2–28.2
μ
gm
−3
) matter below the US EPA 24-hour NAAQS (35
μ
gm
−3
), with the exception of the
Beaver Falls monitoring stations, which observed a concentration of 42.6
μ
gm
−3
on 4 July 2015
(Figure 11). However, this elevated concentration can likely be explained by the presence of fireworks
and other combustion processes due to the holiday.
When comparing the average PA DEP monitoring station PM
2.5
concentrations [
20
] with the
average NCSU monitor PM
2.5
concentrations (Figure 12), it is evident that the monitors used in this
study observed lower concentrations than the concentrations observed at the PA DEP monitoring
sites. However, the general trend in concentrations for both monitoring networks is the same. There
are several possible explanations for this discrepancy. One potential cause for the differences in
observations could be due to differences in the instrumentation used to measure the particulate matter.
In addition, another potential cause for the observed differences could be due to location. Since the
monitors are farther away, it is possible that the pollutants were transported aloft before mixing down
to the surface level. Furthermore, it is also possible that there are other potential sources of particulate
matter located near the sites.
70
Books
MDPI
Atmosphere 2017, 8, 186
Figure 11.
PM
2.5
measurements taken by the PA DEP monitoring sites during the study period. The red
line represents the US EPA 24-hour NAAQS, the blue line represents the Washington monitoring site,
the magenta represents the Greensburg monitoring site, the gold line represents the Florence monitoring
site and the green line represents the Beaver Falls monitoring site.
Figure 12.
The average concentration measured by the NCSU monitors (this study, green line)
compared with the average concentration measured by the PA DEP monitors (blue line line) and
the US EPA 24-hour NAAQS (red line) for PM
2.5
.
71
Books
MDPI
Atmosphere 2017, 8, 186
4. Conclusions
The ambient 24-hour average concentrations of both PM
2.5
(measured in this study as well as
by the PA DEP) and PM
10
did not reach levels higher than what is permitted by the US EPA 24-hour
NAAQS. The average concentrations of PM
10
observed during the periods were 20.5
±
10.2
μ
gm
−3
for Station 1, 16.1
±
4.9
μ
gm
−3
for Station 2 and 16.5
±
7.1
μ
gm
−3
for Station 3. The average
concentrations of PM
2.5
observed at the stations were 9.1
±
5.1
μ
gm
−3
for Station 1, 0.2
±
0.4
μ
gm
−3
for Station 2 and 11.6
±
4.8
μ
gm
−3
for Station 3. The highest average concentration was observed at
Station 1 for PM
10
and at Station 3 for PM
2.5
. However, Station 1 observed the highest daily average
concentration for both PM
2.5
and PM
10
. Not only was Station 1 the closest to the power plant, but the
wind rose analysis showed that some of the elevated concentrations potentially came directly from the
power plant’s coal pit, in addition to other PM sources. However, other elevated concentrations were
observed coming from other directions, suggesting that there are several sources for particulate matter
located in the region.
The IC analysis showed that the dominant species of particulate matter were primarily sulfate,
nitrate and ammonium. The results of the XRF analysis showed that the primary constituents of PM
10
and PM
2.5
were sulfur, silicon, aluminum and iron. While these are all constituents that are observed
from coal combustion emissions, these constituents are also prominent in most particulate matter
speciation and therefore cannot be contributed directly to emissions from the Cheswick Power Plant.
The low particulate matter emissions and results of the speciation analyses could be attributed to a
number of factors. Because the study period was so short, it is likely that the local scale meteorological
conditions led to bias in the results. In addition, it is also possible that there were errors in the
instrumentation that led to such low concentrations, particularly for PM
2.5
concentrations observed at
Station 2. Furthermore, the low concentrations of particulate matter observed at the sites could also
be attributed to the emission control technology that is added to the Cheswick power plant, which
includes wet lime flue gas desulfurization, low NO
x
burner technology with separated over fire air
selective catalytic reduction and an electrostatic precipitator (Available online: http://www.ampd.epa.
gov/ampd), which would explain the reduced concentrations of pollutants observed in this study.
Based on the results of this study, is not possible to determine a concrete conclusion on the role
of the Cheswick Power Plant in concentrations of particulate matter in the Cheswick and Springdale
boroughs. These inconclusive results can be attributed to a number of factors, including unfavorable
meteorological conditions, potential issues with the measurement equipment as well as an extremely
short sampling period. It must be emphasized that this study was conducted over a limited time
period. Therefore, it is recommended that further work be done on this matter, with longer sampling
periods occurring in each season in order to capture a seasonal profile of concentrations of particulate
matter in this region.
Acknowledgments:
The authors acknowledge the support from the Sierra Club. We want to thank Marty Blake,
Karen Petito, and Natalie Bick for their help with the monitoring stations. We also want to thank Karsten Baumann
for help with the sample mass analysis, John Walker for help with the inorganic analysis of PM samples, and
Dennis Mikel for his help with the X-Ray Fluorescence analysis of the PM samples. We would also like to thank
Sean Nolan and the PA Department of Environmental Protection, Bureau of Air Quality, for their assistance in
obtaining the data for the PA DEP monitoring sites. In addition, we would also like to thank Zachary Fabish for
his assistance with this project.
Author Contributions:
Viney Aneja was the PI for this project. Priya Pillai led the measurement process.
Pornpan Uttamang helped take measurements for the project and helped in research for the manuscript.
William Battye helped in the data analysis and in the revision and improvement of the manuscript. Casey Bray
assisted in the measurement process, analyzed the data and wrote the manuscript.
Conflicts of Interest: The authors report no conflict of interest.
References
1.
Dirty Kilowatts. America’s most Polluting Power Plants 2006. Available online: http://www.dirtykilowatts.
org/dirty_kilowatts.pdf (accessed on 15 September 2015).
72
Books
MDPI
Atmosphere 2017, 8, 186
2.
Aneja, V.P.; Isherwood, A.; Morgan, P. Characterization of Particulate Matter (PM
10
) Related to Surface Coal
Mining Operations in Appalachia. Atmos. Environ. 2012, 54, 496–501. [CrossRef]
3.
Nielsen, M.T.; Livbjerg, H.; Fogh, C.L.; Jensen, J.N.; Simonsen, P.; Lund, C.; Poulsen, K.; Sander, B. Formation
and Emission of Fine Particles from Two Coal-Fired Power Plants. Combust. Sci. Technol.
2002
, 174, 79–113.
[CrossRef]
4.
Tian, H.; Wang, Y.; Xue, Z.; Qu, Y.; Chai, F.; Hao, J. Atmospheric emissions estimation of Hg, As, and Se from
coal-fired power plants in China. Sci. Total Environ. 2007, 409, 3078–3081. [CrossRef] [PubMed]
5.
Contini, D.; Cesari, D.; Conte, M.; Donateo, A. Application of PMF and CMB Receptor Models for the
Evaluation of the Contribution of a Large Coal-Fired Power Plant to PM
10
Concentrations. Sci. Total Environ.
2016, 560, 131–140. [CrossRef] [PubMed]
6.
Chow, J.C.; Watson, J.G. Review of PM
2.5
and PM
10
Apportionment for Fossil Fuel Combustion and Other
Sources by the Chemical Mass Balance Receptor Model. Energy Fuels 2002, 16, 222–260. [CrossRef]
7.
Cohen, D.D.; Crawford, J.; Stelcer, E.; Atanacio, A.J. Application of Positive Matrix Factorization, Multi-Linear
Engine and Back Trajectory Techniques to the Quantification of Coal–Fired Power Station Pollution in
Metropolitan Sydney. Atmos. Environ. 2012, 61, 204–211. [CrossRef]
8.
Gladney, E.S.; Small, J.A.; Gordon, G.E.; Zoller, W.H. Composition and Size Distribution of in-Stack
Particulate Material at a Coal-Fired Power Plant. Atmos. Environ. 1976, 10, 1071–1077. [CrossRef]
9.
Lee, S.W. Source Profiles of Particulate Matter Emissions from a Pilot-Scale Boiler Burning North American
Coal Blends. J. Air Waste Manag. Assoc. 2001, 51, 1568–1578. [CrossRef]
10.
Liu, X.; Xu, M.; Yao, H.; Yu, D.; Zhang, Z.; Lü, D. Characteristics and Composition of Particulate Matter from
Coal-Fired Power Plants. Sci. China Ser. E 2009, 52, 1521–1526. [CrossRef]
11.
Wu, H.; Pedersen, A.J.; Glarborg, P.; Frandsen, F.; Dam-Johansen, K.; Sander, B. Formation of Fine Particles in
Co-Combustion of Coal and Solid Recovered Fuel in a Pulverized Coal-Fired Power Station. Proc. Combust. Inst.
2011, 33, 2845–2852. [CrossRef]
12.
Young, G.S.; Fox, M.A.; Trush, M.; Kanarek, N.; Glass, T.A.; Curriero, F.C. Differential Exposure to Hazardous
Air Pollution in the United States: A Multilevel Analysis of Urbanization and Neighborhood Socioeconomic
Deprivation. Int. J. Environ. Res. Public Health 2012, 9, 2204–2225. [CrossRef] [PubMed]
13.
Zhou, W.; Cohan, D.; Pinder, R.; Neuman, J.; Holloway, J.; Peischl, J.; Ryerson, T.; Nowak, J.; Flocke, F.;
Zheng, W. Observation and Modeling of the Evolution of Texas Power Plant Plumes. Atmos. Chem. Phys.
2012, 12, 455–468. [CrossRef]
14.
Anderson, H.R.; Atkinson, R.W.; Bremner, S.A.; Marston, L. Particulate air pollution and hospital admissions
for cardiorespiratory diseases: Are the elderly at greater risk? Eur. Pespir. J. 2003, 21, 39–46. [CrossRef]
15.
Congress.gov. Clean Air Act Amendments of 1990. Available online: https://www.congress.gov/bill/101st-
congress/senate-bill/1630/amendments (accessed on 24 July 2017).
16.
MacIntosh, D.; Spengler, J. Emissions of Hazardous Air Pollutants from Coal Fired Power Plants.
2011. Available online: http://www.lung.org/assets/documents/healthy-air/coal-fired-plant-hazards.pdf
(accessed on 15 September 2015).
17.
Pope, C.A., III; Dockery, D.W. Health Effects of Fine Particulate Air Pollution: Lines that Connect. J. Air
Waste Manag. Assoc. 2006, 56, 709–742. [CrossRef]
18.
Kelly, M.; Fischman, G. Point Source Emission Inventory Report. 2011. Available online: http://www.achd.
net/air/pubs/pdf/2011_emissions_inventory_report.pdf (accessed on 15 September 2015).
19.
Levin, I. America’s Top Power Plant Toxic Polluters 2011. Available online: http://www.environmentalintegrity.
org/documents/report-topuspowerplanttoxicairpolluters.pdf (accessed on 15 September 2015).
20. Nolan, S.; PA DEP. PADEP PM
2.5
Air Quality Monitoring Data. Personal Communication, 2017.
21.
Finkelman, R.B. Modes of occurrence of environmentally-sensitive trace elements in coal. In Environmental
Aspects of Trace Elements in Coal; Anonymous; Springer: Dordrecht, The Netherlands, 1995; pp. 24–50.
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
73
Books
MDPI
atmosphere
Article
Comparing CMAQ Forecasts with a Neural Network
Forecast Model for PM
2.5
in New York
Samuel D. Lightstone *, Fred Moshary and Barry Gross
Optical Remote Sensing Lab, City College of New York, New York, NY 10031, USA;
moshary@ccny.cuny.edu (F.M.); gross@ccny.cuny.edu (B.G.)
* Correspondence: slights01@citymail.cuny.edu
Received: 30 June 2017; Accepted: 26 August 2017; Published: 29 August 2017
Abstract:
Human health is strongly affected by the concentration of fine particulate matter (PM
2.5
).
The need to forecast unhealthy conditions has driven the development of Chemical Transport
Models such as Community Multi-Scale Air Quality (CMAQ). These models attempt to simulate
the complex dynamics of chemical transport by combined meteorology, emission inventories (EI’s),
and gas/particle chemistry and dynamics. Ultimately, the goal is to establish useful forecasts that
could provide vulnerable members of the population with warnings. In the simplest utilization,
any forecast should focus on next day pollution levels, and should be provided by the end of
the business day (5 p.m. local). This paper explores the potential of different approaches in
providing these forecasts. First, we assess the potential of CMAQ forecasts at the single grid cell
level (12 km), and show that significant variability not encountered in the field measurements occurs.
This observation motivates the exploration of other data driven approaches, in particular, a neural
network (NN) approach. This approach makes use of meteorology and PM
2.5
observations as model
predictors. We find that this approach generally results in a more accurate prediction of future
pollution levels at the 12 km spatial resolution scale of CMAQ. Furthermore, we find that the NN is
able to adjust to the sharp transitions encountered in pollution transported events, such as smoke
plumes from forest fires, more accurately than CMAQ.
Keywords:
air quality model; Air Quality System (AQS); Community Multi-Scale Air Quality
(CMAQ) model; fine particulate matter (PM
2.5
); Aerosol Optical Depth (AOD)
1. Introduction
Fine particulate matter air pollution (PM
2.5
) is an important issue of public health, particularly
for the elderly and young children. The study by Pope et al. suggests that exposure to high levels of
PM
2.5
is an important risk factor for cardiopulmonary and lung cancer mortality [
1
,
2
]. Furthermore,
increased risk of asthma, heart attack and heart failure have been linked to exposure to high PM
2.5
concentrations [3].
PM
2.5
levels are dynamic and can fluctuate dramatically over different time scales. In addition
to local emission sources, pollution events can be the result of aerosol plume transport and intrusion
into the lower troposphere. When there is a potential high pollution event, the local air quality
agencies must alert the public, and advise the population on proper safety measures, as well as direct
the reduction of emission producing activities. Therefore, accurately measuring and predicting fine
particulate levels is crucial for public safety.
The U.S. Environmental Protection Agency (EPA) established the National Ambient Air Quality
Standards (NAAQS), which regulate levels of pollutants such as fine particulate matter. The New York
State Department of Environment Conservation (NYSDEC) operates ground stations for monitoring
PM
2.5
and speciation throughout NY State [
4
]. However, surface sampling is expensive and existing
Atmosphere 2017, 8, 161 74 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2017, 8, 161
networks are limited and sparse. This results in data gaps that can affect the ability to forecast PM
2.5
over a 24-h period. The EPA developed the Models-3 Community Multi-scale Air Quality system
(CMAQ), to provide 24–48 h air quality forecasts. CMAQ provides an investigative tool to explore
proper emission control strategies. CMAQ has been the standard for modeling air pollution for nearly
two decades because of its ability to independently model different pollutants while describing the
atmosphere using “first-principles” [5].
In their studies, McKeen et al. and Yu et al. evaluate the accuracy of CMAQ forecasts [
6
,
7
]. To do
so, they use the CMAQ 1200 UTC (Version 4.4) forecast model. They observe the midnight-to-midnight
local time forecast and compare the hourly and daily average forecasts to the ground monitoring
stations. McKeen et al. [
6
] observed minimal diurnal variations of PM
2.5
at urban and suburban
monitor locations, with a consistent decrease of PM values between 0100 and 0600 local time. However,
the CMAQ model showed significant diurnal variations, leading McKeen et al. to conclude that aerosol
loss during the late night and early morning hours has little effect on PM
2.5
concentrations, while the
CMAQ model does not account for this. Therefore, in addition to testing the hourly CMAQ forecast for
a 24-h period, we focus on the daytime window for two reasons: (1) to assess the accuracy of CMAQ
when aerosols do not play a reduced roll in forecasting; (2) the forecast should predict the air quality
during the time of maximum human exposure.
While these studies make a distinction between rural and urban locations, they take the average
results for all rural and urban locations respectively; thereby, their assessment of the CMAQ model
was as at a regional scale, rather than a localized one. In addition to regional emissions, these studies
also considered extreme pollution events such as the wildfires in western Canada and Alaska, which
occurred during the observation period for the studies by Yu et al. and McKeen et al. The results of
this assessment concluded that due to insufficient representation of transport pollution associated with
the burning of biomass, CMAQ significantly under predicted the PM
2.5
values for these events.
In the study by Huang et al. [
8
], the bias corrected CMAQ forecast was assessed for both the
0600 and 1200 UTC release times. The study revealed a general improvement of forecasting skill
for the CMAQ model. However, it was observed that the bias correction was limited in predicting
extreme events, such as wildfires, and new predictors must be included in the bias correction to
predict these events. In this study, CMAQ was assessed as a regional forecasting tool, taking 551 sites,
and evaluating the average results in six sub-regions.
In our present assessment of the current operational CMAQ forecast model (Version 4.6), we differ
from the regional studies above in the following ways: Firstly, in addition to the 1200 UTC forecast,
we evaluated the 0600 UTC forecast for the same period to determine if release time affects the CMAQ
forecast. Second, we focused on specific locations, both rural and urban, to assess the potential of
CMAQ as a localized forecasting tool. In addition, we revisited the forecast potential of CMAQ for high
pollution events, to determine if these events are generally caused by transport, or by local emissions.
Finally, we tailor the forecast comparisons to focus on the potential of providing next day forecasts
using data prior to 5 p.m. of the previous day, since this is an operational requirement for the state
environmental agencies.
In focusing on both rural and urban areas in New York State, previous studies have shown
anomalies in PM
2.5
from CMAQ forecasts. For example, in [
9
], using CMAQ (Version 4.5) with various
planetary boundary layer (PBL) parameterizations, PM
2.5
forecasts during the summer pre-dawn and
post-sunset periods were often highly overestimated in New York City (NYC). Further analysis of
these cases demonstrated that the most significant error was the retrieval of the PBL height, which
was often compressed by the CMAQ model, and did not properly take into account the Urban Heat
Island mechanisms that expand the PBL layer [
10
]. This study showed the importance of PBL height
dynamics and meteorological factors that motivated the choice of meteorological forecast inputs used
during the NN development.
The objective of this paper is to determine the best method to forecast PM
2.5
by direct comparison
with CMAQ output products. In particular, using the CMAQ forecast model, as a baseline, we explore
75
Books
MDPI
Atmosphere 2017, 8, 161
the performance of a NN based data driven approach with suitable meteorological and prior PM
2.5
input factors.
Paper Structure
Our present paper is organized in the following manner: In Section 2, we analyze CMAQ as the
baseline forecaster. We briefly describe the CMAQ model and the forecast schedules that are publically
available, as well as the relevant ground stations we use for comparison. We then describe and perform
a number of statistical tests using both the direct, as well as the bias compensated, CMAQ outputs.
In this section, we show the large dispersion in using the direct results without bias correction.
In Section 3, we present our NN data driven strategy. This includes a description of all the relevant
input factors used, including a combination of present and predicted meteorology, as well as diurnal
trends of prior PM
2.5
levels. We present our first statistical results for the comparisons between CMAQ
and the NN for a variety of experiments in order to highlight the conditions in which the NN results
are generally an improvement. Then we explore the forecast performance for high pollution multiday
transport events, which result in the highest surface PM
2.5
levels during the observed time period.
In this comparison, analyzed by combining a sequence of next-day forecasts together, we find that the
neural network seems to follow the trends in PM
2.5
more accurately than the CMAQ model.
In Section 4, we summarize our results and describe potential improvements.
2. CMAQ Local and Regional Assessment
2.1. Datasets
2.1.1. Models
The CMAQ V4.6 (CB05 gas-phase chemistry) with 12 km horizontal resolution was used for
this paper. The CMAQ product for meteorology predictions used is the North American Model
Non-hydrostatic Multi-scale Model (NAM-NMMB). This version was made available starting February
2016. The CMAQ data used for this paper is from 1 February 2016 until 31 October 2016. The station
names and locations are listed in Table A1. The data can be accessed from [
11
], and the model
description can be found in [12,13].
The CMAQ model used has a few different configurations: release times of 0600 UTC and
1200 UTC
, and each release time has a standard forecast as well as a bias corrected forecast. The analog
ensemble method is used for bias corrections. The idea is to look at similar weather patterns for
the forecast period, and statistically correct the numerical PM
2.5
forecast based on historical errors.
The analog ensemble method is described in detail in Huang et al. [
8
]. For each release time, CMAQ
provides a 48-h forecast. The release time of 0600 UTC and 1200 UTC (2 a.m. and 8 a.m. EDT) does not
give the public enough time to react to the forecast on the same day as the release. Consequently, for
the 0600 UTC release time, the forecast hours 22–45 were used, and for the release time of 1200 UTC
the forecast hours of 16–39 were used. This allowed us to construct a complete 24-h diurnal period for
the forecast time window, which facilitated comparison with the field station data.
2.1.2. Ground-Based Observations
PM
2.5
ground data is collected from the EPA’s AirNow, which collects NYSDEC monitoring
station measurements in real time. The station data used for the forecast experiments in this article
are from the New York State stations listed in Table A2, from 1 January 2011 until 31 December 2016.
To assess the accuracy of CMAQ model forecasts, matching the model to the ground monitoring station
is necessary. To do this, we use the ground NYSDEC stations that lay within the CMAQ grid cell only.
Ground stations that are not found in a CMAQ grid cell were not used for comparison; therefore,
no spatial interpolation was done on the model results while mapping the model or meteorological
data to the AirNow ground stations. This matching method is widely used for comparing the CMAQ
model to ground monitoring stations [
6
,
7
,
14
]. The locational data-points are depicted in Figure A1,
76
Books
MDPI
Atmosphere 2017, 8, 161
the CMAQ grid cell information can be found in Table A1, and the NYSDEC station information can
be found in Table A2.
2.2. Methods
Assessing Accuracy of CMAQ Forecasting Models
The forecasting skill of the different models were evaluated by computing the R
2
and the
root mean square (RMSE) values from a regression analysis comparing the model to the AirNow
observations. High R
2
values and low RMSE values indicated a good match between the prediction
and the observations. Finally, to directly assess potential biases in the regression assessment, residual
plots are provided to show significant concentration bias.
2.3. Results
2.3.1. Effects of Bias and Release Time
Figure 1 shows the regression plots for the hourly CMAQ model output compared to the ground
station data for the City College of New York (CCNY Station) to illustrate the general behavior of the CMAQ
model, and how the forecast is affected by different forecast release times, and by the bias corrections
applied. The results of the R
2
analysis for all ground stations can be found in the supplementary materials.
All forecasts from the CMAQ model over CCNY have a positive correlation to the ground data.
The effect on the forecast for different release times, if any, is minimal.
As seen in Figure 1a,c, the standard model generally overestimates the ground. While the bias
correction improves the over-prediction, the results are more dispersed. This can be verified from the
fact that the bias correction decreases the root mean square error (RMSE), but it also decreases the R
2
value for both release times.
In Figure 1 we assess the overall skill for a 24-h CMAQ forecast. In Figure 2, we determine if the
CMAQ model could be improved by simply moving the forecast release time to a later point in the day,
thereby including the most up-to-date inputs in the model. To do this, we make a direct comparison
between CMAQ forecasts with different release times. In Figure 2, the R
2
value is computed for each
hour of the day. The release time of 0600 UTC, with forecast hours of 22–45, is compared to the 1200 UTC
release time, with forecast hours 16–39, to determine if the lower number of forecast hours yields more
accurate predictions. It is clear from Figure 2 that the later release time does not lead to a significant
improvement in the accuracy of the forecast, and this is true for both urban and non-urban test sites.
(a)
(b)
0 5 10 15 20 25 30 35 4
0
Airnow
0
5
10
15
20
25
30
35
40
QMAQ: 6Z
R
egress
i
on
A
na
l
ys
i
s
R
2
=0.28
RMSE =7.78
CCNY
QMAQ: 6Z vs.Airnow
BestFit: y=1.29x+2.99
0 5 10 15 20 25 30 35 4
0
Airnow
0
5
10
15
20
25
30
35
40
QMAQ: 6Z Bias Co
r
r
ected
R
egress
i
on
A
na
l
ys
i
s
R
2
=0.20
RMSE =4.55
CCNY
QMAQ: 6Z Bias Corrected vs.Airnow
BestFit: y=0.62x+6.14
Figure 1. Cont.
77
Books
MDPI
Atmosphere 2017, 8, 161
(c) (d)
0 5 10 15 20 25 30 35 4
0
Airnow
0
5
10
15
20
25
30
35
40
QMAQ: 12Z
R
egress
i
on
A
na
l
ys
i
s
R
2
=0.27
RMSE =7.69
CCNY
QMAQ: 12Z vs.Airnow
BestFit: y=1.25x+3.07
0 5 10 15 20 25 30 35 4
0
Airnow
0
5
10
15
20
25
30
35
40
QMAQ: 12Z Bias Co
r
r
ected
R
egress
i
on
A
na
l
ys
i
s
R
2
=0.26
RMSE =5.28
CCNY
QMAQ: 12Z Bias Corrected vs.Airnow
BestFit: y=0.83x+5.02
Figure 1.
Community Multi-Scale Air Quality (CMAQ) regression analysis. (
a
) Standard, 06Z release
time; (
b
) Bias Corrected, 06Z release time; (
c
) Standard, 12Z release time; (
d
) Bias Corrected, 12Z
release time.
(a) (b)
(c) (d)
05101520
Hour of the Day
E
aste
rn Tim
e
0.15
0.2
0.25
0.3
0.35
0.4
0.45
R
2
CMAQ Analysis
R
2
as a function of Forecast Hour
CCNY
QMAQ: 6Z
QMAQ: 12Z
0 5 10 15 20
Hour of the Day
E
aste
rn Tim
e
0.15
0.2
0.25
0.3
0.35
0.4
0.45
R
2
CMAQ Analysis
R
2
as a function of Forecast Hour
ROCHESTER 2
QMAQ: 6Z
QMAQ: 12Z
0 5 10 15 20
Hour of the Day
E
aste
rn Tim
e
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
R
2
CMAQ Analysis
R
2
as a function of Forecast Hour
ALBANY COUNTY HEALTH DEPT
QMAQ: 6Z
QMAQ: 12Z
0 5 10 15 20
Hour of the Day
E
aste
rn Tim
e
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
R
2
CMAQ Analysis
R
2
as a function of Forecast Hour
BROOKSIDE TERRACE
QMAQ: 6Z
QMAQ: 12Z
Figure 2.
Comparing the effect of different release times for CMAQ by plotting the R
2
value
as a function of time of day. (
a
) City College of New York (CCNY); (
b
) Rochester; (
c
) Albany;
(d) Brookside Terrace.
78
Books
MDPI
Atmosphere 2017, 8, 161
It can be seen from this analysis that the CMAQ model performs best for midday hours, which
is reasonable, since this is the period when convective mixing is most dominant. As discussed in
reference [
9
], PBL modeling is very complex during the predawn/post-sunset period and errors in the
PBL height clearly are a significant concern for further model development.
2.3.2. Differences between Urban and Non-Urban Locations
To get a better understanding of the spatial performance of the model, a multi-year time-series
of daily averaged PM
2.5
observations from ground monitoring is used to compare the relationship
between PM
2.5
values in New York City to the rest of New York State. Figure 3a is the regression
analysis for this time period, and shows how the PM
2.5
values for NYC are strongly correlated to
non-NYC areas, R
2
~0.6. This indicates that while PM
2.5
values in NYC are generally higher than the
rest of the state, the PM
2.5
level in NYC are still correlated to the levels in the rest of the state.
(a) (b)
0 5 10 15 20 25 30 35 4
0
Airnow NYS
0
5
10
15
20
25
30
35
40
Ai
r
now NYC
R
egress
i
on
A
na
l
ys
i
s
R
2
=0.60
RMSE =3.23
Airnow: NYC vs NYS
Airnow NYC vs.Airnow NYS
BestFit: y=1.00x+1.42
0 5 10 15 20 25 30 35 4
0
CMAQ NYS
0
5
10
15
20
25
30
35
40
CMAQ NYC
Regression Analysis
R
2
=0.22
RMSE =3.59
CMAQ: NYC vs NYS
CMAQ NYC vs.CMAQ NYS
BestFit: y=0.68x+3.25
Figure 3.
Regression analysis comparing PM
2.5
levels between NYC and the rest of NYS (non-NYC
sites). (
a
) Multi-year day-averaged PM
2.5
analysis from NYSDEC ground observations; (
b
) CMAQ
model comparison between NYC and NYS.
The same analysis comparing NYC to the rest of NYS was done with CMAQ forecast values as
seen in Figure 3b. In this case, the correlation between NYC and NYS is not so strong, R
2
~0.2. From
this analysis alone, we can only speculate the reason for a low correlation between CMAQ forecasts for
NYC and the rest of NYS is due to strong spatial differences in the National Emission Inventory (NEI)
entries. However, the strong correlation in ground observations between NYC and NYS shows that
while urban source emission may be a significant cause for somewhat higher levels of PM
2.5
, there is
still a strong correlation between NYC and NYS, and an accurate forecasting model must take this
into account.
The limitations of CMAQ forecasting on a local pixel level indicate that other approaches should
be explored. In particular, we explore the potential of data-driven models for localized forecasting in
the next section.
79
Books
MDPI
Atmosphere 2017, 8, 161
3. Data Driven (Neural Network) Development
3.1. Datasets
3.1.1. Ground-Based Observations
PM
2.5
data collected from NYSDEC ground-monitoring stations is used for inputs in the neural
network. These are the same ground stations listed above, in Section 2.1.2.
3.1.2. Models
The meteorological data was collected from the National Centers for Environmental Prediction
(NCEP) North American Regional Reanalysis (NARR). NARR has high-resolution reanalysis of the
North American region, 0.3 degrees (32 km) at the lowest latitude, including assimilated precipitation.
The NARR makes available 8-times-daily and monthly means respectively. The data collected for this
paper is the 8-times-daily means for the duration 1 January 2011 until 31 December 2016. Figure A1
shows the proximity of the meteorological data and the CMAQ model outputs to the ground stations.
The NN network was created and tested using historical data. In this paper, meteorology “forecast”
data refers to NARR data that was observed the day of the PM
2.5
forecast. “Observed” or “measured”
meteorology refers to NARR data that was observed before the forecast release time.
3.2. Methods
3.2.1. Development of the Neural Network
As stated above, the accurate prediction of PM
2.5
values is crucial for air quality agencies, so that
they could alert the public of the severity and duration of a high pollution event. Therefore, it is
imperative that the forecast predictions are released to the public the day before the event. For this
paper, we chose 5 p.m. as a target for the forecast release time. Therefore, we ensure that all the
methods tested, utilize factors that are available to the state agency prior to 2100 UTC (5 p.m. EDT).
Input Selection Scenarios
The NN input includes the following NARR meteorological data: surface air temperature, surface
pressure, planetary boundary layer height (PBLH), relative humidity, and horizontal wind (10 m).
To account for the seasonal variations, the month is also used as an input in the neural network.
The PM input variables for the NN are the PM
2.5
measurements averaged over a three-hour frequency
to match the meteorological dataset. The NN output is the next day PM
2.5
values.
In order to optimize the performance of the neural network, preliminary tests were done to
determine the optimum utilization of the meteorological input variables. These test were done to
determine if the “forecast” or the “observed” meteorology, or a combination of the two, should be
used as input variables.
The forecast time window is midnight-to-midnight EDT for the forecast day, while the time
window with the observed data is midnight to 5 p.m. EDT the day the forecast is released.
For the PBLH, the forecast value is always used as the input. One NN design employed only the
forecast meteorological values as inputs. The second design utilized a combination of the forecast
and the observed data, by subtracting the eight observation datasets from the eight forecast datasets.
This first NN architecture uses the meteorological values as predictors, while the second design uses
metrological trends as predictors. We note that this comparison does not affect the number of inputs
used, allowing for a direct comparison of information content.
In scenario 1, where only the MET forecasts are used, we use the following inputs, where i
represents the indices for time windows for the observation day, and j represents the indices for time
windows for the forecast day (from the NARR forecasts), the NN inputs design is:
80
Books
MDPI
Atmosphere 2017, 8, 161
PM
2.5
(
i
)
i = 1:5 time window
(
i
)
=
(
i − 1
)
×
3:i × 3 (Field measurements)
MET
f orecast
(
j
)
j = 1:8 time windo w
(
j
)
=
(
j − 1
)
×
3:j × 3 (NARR Forecasts)
PBLH
(
j
)
j = 1:8 time windo w
(
j
)
=
(
j − 1
)
×
3:j × 3 (NARR Forecasts)
In scenario 2, where the differential between the observation day and forecast day of the MET
variables are used, the architecture for the NN inputs is:
PM
2.5
(
i
)
i = 1:5 time window
(
i
)
=
(
i − 1
)
×
3:i × 3 (Field measurements)
MET
f orecast
(
j
)
−
MET
observed
(
i
)
j = 1:8
i
= 1:8
time window
(
j
)
=
(
j − 1
)
×
3:j × 3
time window
(
i
)
=
(
i − 1
)
×
3:i × 3
(NARR Forecasts)
(NARR Observations)
PBLH
(
j
)
j = 1:8 time windo w
(
j
)
=
(
j − 1
)
×
3:j × 3 (NARR Forecasts)
To show the robustness of the NN, the data used for training the neural networks came from
2011–2015 alone, while the network was tested with data from 2016. In both scenarios, the targets for
the NN were taken to be the complete set of PM
2.5
over all time windows of the forecast day:
Targets: PM
2.5
(
j
)
j = 1:8 time windo w
(
j
)
=
(
j − 1
)
×
3:j × 3 (Field measurements)
Neural Network Training Approach
In developing a NN PM
2.5
forecast for all of New York State (NYS), we needed to take into account
the very different emission sources, and to a lesser extent the meteorological conditions, between New
York City (NYC) and the other sites in NYS. We found that the best solution is to design two different
neural networks. The first is trained only over NYC sites, while the second is trained for the rest of
NYS. It is important to note that we do not try to build a unique NN for every station, since this is
not a useful approach for local agencies. PM and Meteorological data from 2011–2015, were used
for training.
For NYC, since the stations are very close to each other, the NN was trained with spatial
mean values of the ground PM monitors and NARR meteorological datasets. For NYS, all the PM
and meteorological data from each site outside of NYC were used. Some site-specific information
was implicitly included by using the surface pressure as inputs, which provides some indicator of
surface elevation.
The neural network was developed using the MATLAB Neural Network Toolbox [
15
]. The
Levenberg-Marquardt network was deployed using 10 hidden nodes. The break down for the NN
input data is: 70% training, 15% validation, and 15% testing. Because the sample set of training,
validation, and testing is divided randomly over the entire dataset, accuracy of the NN was determined
by testing each network over 2016 data only, a time window that was not included in training. Once
the NN function was created, the 2016 meteorological and PM data was passed through the network,
and the outputs were stored with the date-time and station location as indices.
Neural Network Scenario Results
Figure 4a shows the performance of the NN using the forecast metrological data as inputs, while
Figure 4b shows the performance of the NN using the difference between the forecast and the current
days measurements. The NN utilizing the difference configuration is clearly better, with a higher R
2
value, 0.44 compared to 0.36, and a lower root-mean-square value, 3.09 compared to 4.59. In addition,
there are substantially less anomalous high PM
2.5
forecasts. Since this improvement was seen in all test
cases, we only used scenario 2, (differential meteorology) NN configuration. From these results, we see
that meteorological trends are better indicators of PM
2.5
than meteorology alone. This appears to us
to be a reasonable result since the meteorology trend better isolates particular mesoscale conditions,
which is known to be a significant factor in boundary layer dynamics.
81
Books
MDPI
Atmosphere 2017, 8, 161
(a) (b)
0 5 10 15 20 25 30 35 4
0
Airnow
0
5
10
15
20
25
30
35
40
NN forecast only
R
egress
i
on
A
na
l
ys
i
s
R
2
=0.36
RMSE =4.59
CCNY
NN forecast only vs.Airnow
BestFit: y=0.93x+1.97
0 5 10 15 20 25 30 35 40
Airnow
0
5
10
15
20
25
30
35
40
Neural Network
R
egress
i
on
A
na
l
ys
i
s
R
2
=0.44
RMSE =3.09
CCNY
Neural Network vs.Airnow
BestFit: y=0.74x+3.17
Figure 4.
Results from the regression analysis to maximize Neural Network performance for the
different scenarios. (
a
) NN designed with the forecast meteorological data (Scenario 1); (
b
)NN
designed by taking the difference between the forecast and the current days measurements (Scenario 2).
3.3. Results
3.3.1. Neural Network and CMAQ Comparison
The R
2
value for CMAQ and the NN, both compared to AirNow observations, is computed for
each forecast model and for each location. As a representative example of the overall performance,
the R
2
value for NYC, represented by CCNY, is compared to NYS, represented by Brookside Terrace,
a non-NYC, non-urban station, and these results are displayed in Figure 5. The individual results for
each location can be found in the supplementary materials.
(a) (b)
CMA
Q
6Z 6Z Bias Corrected CMA
Q
12Z 12Z Bias Corrected NN
0
0.1
0.2
0.3
0.4
0.5
R
2
Analysis
Full 24hrs
BROOKSIDE TERRACE
Hourly
3hr Average
NYS Average
Neural Network
CMA
Q
6Z 6Z Bias Corrected CMA
Q
12Z 12Z Bias Corrected NN
0
0.1
0.2
0.3
0.4
0.5
R
2
Analysis
Full 24hrs
CCNY
Hourly
3hr Average
NYS Average
Neural Network
Figure 5.
Regression analysis is computed for the comparison between AirNow observations and the
various prediction models, for the complete CMAQ database time period (February 2016, through
October 2016). The R
2
value for each model is plotted in the figure above to compare CMAQ to the NN.
The CMAQ model includes the different release times as well as bias compensated vs. uncompensated
runs. In addition, different time and spatial averaging of CMAQ is considered at each location.
(a) Brookside Terrace, representative of non-NYC; (b) CCNY, representative of NYC.
82
Books
MDPI
Atmosphere 2017, 8, 161
From Figure 5 above, it can be seen that the most accurate forecast model is the neural network
for both NYS and NYC over any of the CMAQ forecasts studied. Regarding CMAQ, we note better
performance for NYC than for non-urban areas. This is in contrast to the neural network, where there
is very little variation in the results for locations that are urban versus non-urban, indicating that
locational inputs in the model, such as the surface pressure, improves forecasting skill.
In addition, for all cases, it can be seen that taking the time average improves the CMAQ results.
Furthermore, the spatial averaging over NYS (with 1-h time sampling) shows more improvement in
most NYC cases and some non-NYC cases as well. These results indicate the possibility that the best
use for CMAQ forecasting is on a regional level. This is supported from the 12 km grid cell resolution
for CMAQ, a cell size typical for regional analysis.
We note again that the different release times for CMAQ has almost no effect on the forecast
accuracy. In Figure 6, we compared the diurnal performance of the NN to the CMAQ model. The most
apparent result is the dramatic improvement of the NN during the night and morning hours, where
the CMAQ model has the most difficulty. This is clearly due to the machine learning approach where
the time differences, the inputs, and forecast periods have a dramatic effect on output performance.
This also explains the general downward trend, where performance tails off in the late afternoon
and becomes closer to the CMAQ performance. This can be expected, because larger time delays
should lead to more dispersion between the outputs and input PM levels.
0 5 10 15 20
Hour of the Day
Eastern Time
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
R
2
Forecast Analysis: CMAQ vs NN
R
2
as a function of Forecast Hour
CCNY
Neural Network
QMAQ 3-hour: 12Z
Figure 6.
Comparing the effect of different release times for the NN in comparison to CMAQ by
plotting the R
2
value as a function of time of day.
Figure 7 below shows the residual results for CMAQ in comparison to the neural network.
For CMAQ, as noted above, there seems to exist a non-random bias pattern, where CMAQ generally
over predicts for low and high PM values, and under predicts for medium values. This pattern seems
to indicate that the CMAQ model may not capture all of the underlying variability factors. On the
other hand, for the neural network, the behavior of the residuals is clearly stochastic in nature.
We find that an optimized NN approach generally results in a more accurate prediction of future
pollution levels, as compared to CMAQ, for a single grid cell (resolution 12 km).
83
Books
MDPI
Atmosphere 2017, 8, 161
(a) (b)
Figure 7.
Residual analysis. The standard deviation of PM
2.5
from the AirNow ground monitoring
sites was calculated to be 4
μ
g/m
3
, therefore, +/
−
4
μ
g/m
3
was used for the error bounds (
a
) CMAQ;
(b) Neural Network.
3.3.2. Heavy Pollution Transport Events
Because the neural network is data-driven, the network performs better when the most up-to-date
inputs are used. This explains the degradation of performance with time, as seen in Figure 6.
In the current design of the neural network, we only used five PM
2.5
inputs, instead of maximum
possible in a 24-h period, eight. In the training of the NN, there were very few extreme event cases,
PM
2.5
>25μg/m
3
. The lack of suitable training statistics for these events causes the NN approach to
have difficultly in adjusting to the sharp contrast with the onset of the event.
Therefore, a second neural network was trained with the same design as the neural network
illustrated above; however, this neural network produces a 24-h forecast at 5 p.m. for the time period,
5 p.m.–5 p.m. (instead of a next day 24-h midnight-to-midnight forecast). This neural network uses all
eight PM measurements, because there is no lag time between the release time and the first forecast
hour. This neural network, referred to as NN Continuous, was not used in the statistical analysis for
the different forecast models (because the 24-h forecast period is different than the forecast analysis
above), but is being explored in the extreme event cases. The reason for developing this continuous
neural network is to determine if the continuous nature of the network produces better results in
extreme pollution events.
To explore the behavior of the different models under high pollution transport conditions,
the forecasts coinciding with the wildfires of Fort McMurray in Alberta, Canada were analyzed.
The wildfire started on 1 May 2016, and was declared under control on 5 July 2016. Although the
wildfire lasted for over two months, evidence of increased PM
2.5
surface levels in NYC resulting
from the wildfire were detected on 9 May, and on 25 May. On these dates, instances of aloft plume
intrusions and the mixing down into the planetary-boundary layer were observed by a ceilometer and
a Raman-Mie Lidar [
16
]. In Figure 8, we plot the CMAQ and NN model forecasts, focusing on the
transport intrusions into NYC on 25 May.
The first thing to notice in Figure 8a, is the oscillations in the CMAQ model, and to notice how
these oscillations smooth out in Figure 8b,c, where the three-hour time average and the New York State
84
Books
MDPI
Atmosphere 2017, 8, 161
spatial average are tested respectively. It is logical that for heavy transport cases, domain averaging
helps decrease oscillations; however, we still see significant underestimation of the event.
This is the first case where we analyze the behavior of the continuous neural network. Looking at
Figure 8c,d, it is clear that the continuous neural network is able to respond to the trend of the high
pollution event faster, and more accurately, then the standard neural network.
(a) (b)
(c)
(d)
Figure 8.
NYC surface PM
2.5
levels affected from the wildfires of Alberta Canada 2016. The plots
focus on the aloft plumes mixing down into the PBL on May 25. The plots show different models
vs. AirNow observation (
a
) CMAQ Biased hourly, NN continuous; (
b
) CMAQ Biased 3-h average,
NN continuous; (
c
) CMAQ Biased state average, NN continuous; (
d
) CMAQ Biased state average,
standard Neural Network.
4. Conclusions
In this paper, we first made a baseline assessment of the V4.6 CMAQ forecasts, and found
significant dispersion as well as a tendency for the model to overestimate the ground truth field
measurements. Even in the bias corrected case, the residuals error in the model was found to have
significant bias patterns, indicating that there are predictors not included in the model that could
significantly improve the results.
These results motivated the development of data driven approaches such as a NN. In developing a
data driven NN next day forecast model, we found a general improvement of performance when using
prior PM
2.5
inputs together with the difference between present and next day meteorological parameter
85
Books
MDPI
Atmosphere 2017, 8, 161
forecasts. This “differential NN” approach performed significantly better than if we used only the
future forecast variables, indicating that meteorological pattern trends are important indicators.
Using this NN architecture, we then made extensive regression based comparisons between
CMAQ next day forecast models and regionally trained NN next day forecasts for the NYS and NYC
regions. In general, we found that the NN results are a significant improvement over the CMAQ
forecasts in all cases. These comparisons were made to be consistent with state agencies where forecasts
should be available by 5 p.m. In addition, we also made a diurnal comparison, which illustrated
that; the NN approach had superior forecasting skills during the early part of the day but degraded
smoothly as the forecast time increased. By mid-day, the differences between the two approaches was
much closer.
To improve the CMAQ forecasts, we found limited improvement when spatial averaging is
extended beyond the single pixel 12 km resolution to all of New York State. Even in this case, the NN
results were generally more accurate.
Finally, we focused on forecast performance for transported high pollution events such as
Canadian wildfires. In these cases, we found that the CMAQ forecasts had large temporal fluctuations,
which could hide most of the event. In this case, significant improvement was obtained when using
state averaged bias corrected outputs; however, in general, the smoothed results underestimate the
local PM
2.5
measurements.
In this application, we found the neural network approach provides a reasonably smooth forecast,
although the transition from a clean state to a polluted state is very poor. Nevertheless, the standard
NN performed better than CMAQ in this scenario. Further improved results for the NN were obtained
in the transition period when the forecast time of the NN was reduced (NN continuous), making the
transition from training to testing continuous.
Future Work
While the continuous NN does adjust quickly to the sharp contrast in transport events, this design
limits the scope of the forecast period. Clearly, local data alone is not ideal for this application.
Non-local data that can identify high pollution events and assesses their potential mixing with our
region is needed. As a preliminary analysis, we explored the use of a combination of HYSPLIT Air
Parcel Trajectories with GOES satellite Aerosol Optical Depth (AOD) retrievals to improve the NN.
In particular, we analyzed the use of these tools to quantify the relative AOD levels for all air parcels
that reach our target area. We found that by properly counting the trajectories weighted by the AOD,
a good correlation was seen between the relative AOD and the PM
2.5
levels. Therefore, we believe
that using the relative AOD metric as an additional input factor can make improvements in the
NN approach. When GOES-R AOD retrievals, with high data latency and multispectral inversion
capabilities [
17
,
18
], become available, we plan to incorporate these AOD metrics as predictors in
the NN.
Supplementary Materials:
The following are available online at www.mdpi.com/2073-4433/8/9/161/s1,
Figure S1: Regression Analysis.
Acknowledgments:
The authors gratefully acknowledge the NOAA Air Resources Laboratory (ARL) for the
provision of the HYSPLIT transport and dispersion model used in this publication. We further gratefully
acknowledge Jeff McQueen for providing the V4.6 CMAQ forecasts. Finally, S. Lightstone would like to
acknowledge partial support of this project through a NYSERDA EMEP Fellowship award.
Author Contributions:
Samuel Lightstone was responsible for the NN experiments and the CMAQ assessment;
Lightstone also wrote the paper. Barry Gross was responsible for design of the different experiments as well as
conceiving the use of the GOES AOD and HYSPLIT trajectories to estimate transported AOD as a predictor of
PM
2.5
. Fred Moshary provided significant critical assessment and suggestions of all results.
Conflicts of Interest: The authors declare no conflict of interest.
86
Books
MDPI
Atmosphere 2017, 8, 161
Appendix A. Datasets
Table A1. CMAQ Grid Cell Information.
Name Abbreviation Latitude Longitude Land Type
Amherst AMHT 42.99 −78.77 Suburban
CCNY CCNY 40.82
−73.95 Urban
Holtsville HOLT 40.83
−73.06 Suburban
IS 52 IS52 40.82
−73.90 Suburban
Loudonville LOUD 42.68
−73.76 Urban
Queens College 2 QC2 40.74
−73.82 Suburban
Rochester Pri 2 RCH2 43.15
−77.55 Urban
Rockland County RCKL 41.18
−74.03 Rural
S. Wagner HS WGHS 40.60
−74.13 Urban
White Plains WHPL 41.05
−73.76 Suburban
Table A2. NYSDEC Station Information.
NYSDEC ID Station Name Latitude Longitude Land Type
360010005
Albany County Health
Dept
42.6423 −73.7546 Urban
360050112 IS 74 40.8155
−73.8855 Suburban
360291014 Brookside Terrace 42.9211
−78.7653 Suburban
360551007 Rochester 2 43.1462
−77.5482 Urban
360610135 CCNY 40.8198
−73.9483 Urban
360810120 Maspeth Library 40.7270
−73.8931 Suburban
360850055 Freshkills West 40.5802
−74.1983 Suburban
360870005 Rockland County 41.1821
−74.0282 Rural
361030009 Holtsville 40.8280
−73.0575 Suburban
361192004 White Plains 41.0519
−73.7637 Suburban
Location Match
NYSDEC, MET, CMAQ
8
0
°
W
7
8
°
W
7
6
°
W
7
4
°
W
7
2
°
W
7
0
°
W
4
0
°
N
4
2
°
N
4
4
°
N
4
6
°
N
New York
NYSDEC Station
Met Data
CMAQ
Figure A1.
This map shows the proximity of the ground NYSDEC stations to the NARR meteorological
data, and the CMAQ forecast data.
87
Books
MDPI
Atmosphere 2017, 8, 161
References
1.
Pope, C.A., III; Dockery, D.W. Health effects of fine particulate air pollution: Lines that connect. J. Air Waste
Manag. Assoc. 2006, 56, 709–742. [CrossRef] [PubMed]
2.
Pope, C.A., III; Burnett, R.T.; Thun, M.J.; Calle, E.E.; Krewski, D.; Ito, K.; Thurston, G.D. Lung cancer,
cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. JAMA
2002
, 287,
1132–1141. [CrossRef] [PubMed]
3.
Weber, S.A.; Insaf, T.Z.; Hall, E.S.; Talbot, T.O.; Huff, A.K. Assessing the impact of fine particulate matter
(PM
2.5
) on respiratory-cardiovascular chronic diseases in the New York City Metropolitan area using
Hierarchical Bayesian Model estimates. Environ. Res. 2016, 151, 399–409. [CrossRef] [PubMed]
4.
Rattigan, O.V.; Felton, H.D.; Bae, M.; Schwab, J.J.; Demerjian, K.L. Multi-year hourly PM
2.5
carbon
measurements in New York: Diurnal, day of week and seasonal patterns. Atmos. Environ.
2010
, 44,
2043–2053. [CrossRef]
5.
Byun, D.; Schere, K.L. Review of the governing equations, computational algorithms, and other components
of the Models-3 Community Multiscale Air Quality (CMAQ) modeling system. Appl. Mech. Rev.
2006
, 59,
51–77. [CrossRef]
6.
McKeen, S.; Chung, S.H.; Wilczak, J.; Grell, G.; Djalalova, I.; Peckham, S.; Gong, W.; Bouchet, V.; Moffet, R.;
Tang, Y.; et al. Evaluation of several PM
2.5
forecast models using data collected during the ICARTT/NEAQS
2004 field study. J. Geophys. Res. Atmos. 2007, 112. [CrossRef]
7.
Yu, S.; Mathur, R.; Schere, K.; Kang, D.; Pleim, J.; Young, J.; Tong, D.; Pouliot, G.; McKeen, S.A.; Rao, S.T.
Evaluation of real-time PM
2.5
forecasts and process analysis for PM
2.5
formation over the eastern United
States using the Eta-CMAQ forecast model during the 2004 ICARTT study. J. Geophys. Res. Atmos.
2008
, 113.
[CrossRef]
8.
Huang, J.; McQueen, J.; Wilczak, J.; Djalalova, I.; Stajner, I.; Shafran, P.; Allured, D.; Lee, P.; Pan, L.; Tong, D.;
et al. Improving NOAA NAQFC PM
2.5
Predictions with a Bias Correction Approach. Weather Forecast.
2017
,
32, 407–421. [CrossRef]
9.
Doraiswamy, P.; Hogrefe, C.; Hao, W.; Civerolo, K.; Ku, J.Y.; Sistla, G. A retrospective comparison of
model-based forecasted PM
2.5
concentrations with measurements. J. Air Waste Manag. Assoc.
2010
, 60,
1293–1308. [CrossRef] [PubMed]
10.
Gan, C.-M.; Wu, Y.; Madhavan, B.L.; Gross, B.; Moshary, F. Application of active optical sensors to probe the
vertical structure of the urban boundary layer and assess anomalies in air quality model PM
2.5
forecasts.
Atmos. Environ. 2011, 45, 6613–6621. [CrossRef]
11.
Files in /mmb/aq/sv/grib. Available online: http://www.emc.ncep.noaa.gov/mmb/aq/sv/grib/
(accessed on 1 December 2016).
12.
NCEP Operational Air Quality Forecast Change Log. Available online: http://www.emc.ncep.noaa.gov/
mmb/aq/AQChangelog.html (accessed on 1 May 2017).
13.
Lee, P.; McQueen, J.; Stajner, I.; Huang, J.; Pan, L.; Tong, D.; Kim, H.; Tang, Y.; Kondragunta, S.; Ruminski, M.;
et al. NAQFC developmental forecast guidance for fine particulate matter (PM
2.5
). Weather Forecast.
2017
, 32,
343–360. [CrossRef]
14.
McKeen, S.; Grell, G.; Peckham, S.; Wilczak, J.; Djalalova, I.; Hsie, E.; Frost, G.; Peischl, J.; Schwarz, J.;
Spackman, R.; et al. An evaluation of real-time air quality forecasts and their urban emissions over eastern
Texas during the summer of 2006 Second Texas Air Quality Study field study. J. Geophys. Res. Atmos.
2009
,
114. [CrossRef]
15.
Demuth, H.; Beale, M. MATLAB and Neural Network Toolbox Release 2016a; The MathWorks, Inc.: Natick, MA,
USA, 1998.
16.
Wu, Y.; Pena, W.; Diaz, A.; Gross, B.; Moshary, F. Wildfire Smoke Transport and Impact on Air Quality
Observed by a Multi-Wavelength Elastic-Raman Lidar and Ceilometer in New York City. In Proceedings of
the 28th International Laser Radar Conference, Bucharest, Romania, 25–30 June 2017.
88
Books
MDPI
Atmosphere 2017, 8, 161
17.
Anderson, W.; Krimchansky, A.; Birmingham, M.; Lombardi, M. The Geostationary Operational Satellite
R Series SpaceWire Based Data System. In Proceedings of the 2016 International SpaceWire Conference
(SpaceWire), Yokohama, Japan, 25–27 October 2016.
18.
Lebair, W.; Rollins, C.; Kline, J.; Todirita, M.; Kronenwetter, J. Post launch calibration and testing of the
Advanced Baseline Imager on the GOES-R satellite. In Proceedings of the SPIE Asia-Pacific Remote Sensing,
New Delhi, India, 3–7 April 2016.
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
89
Books
MDPI
atmosphere
Article
Multi-Year (2013–2016) PM
2.5
Wildfire Pollution
Exposure over North America as Determined from
Operational Air Quality Forecasts
Rodrigo Munoz-Alpizar
1,
*, Radenko Pavlovic
1
, Michael D. Moran
2
, Jack Chen
2
,
Sylvie Gravel
2
, Sarah B. Henderson
3
, Sylvain Ménard
1
, Jacinthe Racine
1
, Annie Duhamel
1
,
Samuel Gilbert
1
, Paul-André Beaulieu
1
, Hugo Landry
1
, Didier Davignon
1
, Sophie Cousineau
1
and Véronique Bouchet
1
1
Canadian Meteorological Centre Operations, Environment and Climate Change Canada,
2121 Rte Transcanadienne, Montreal, QC H9P 1J3, Canada; radenko.pavlovic@canada.ca (R.P.);
sylvain.menard@canada.ca (S.M.); jacinthe.racine@canada.ca (J.R.); annie.duhamel@canada.ca (A.D.);
samuel.gilbert@canada.ca (S.G.); paul-andre.beaulieu@canada.ca (P.-A.B.); hugo.landry@canada.ca (H.L.);
didier.davignon@canada.ca (D.D.); sophie.cousineau@cafnada.ca (S.C.); veronique.bouchet@canada.ca (V.B.)
2
Air Quality Research Division, Environment and Climate Change Canada, Toronto, ON M3H 5T4, Canada;
mike.moran@canada.ca (M.D.M.); jack.chen@canada.ca (J.C.); sylvie.gravel@canada.ca (S.G.)
3
Environmental Health Services, British Columbia Centre for Disease Control, Vancouver, BC V5Z 4R4,
Canada; sarah.henderson@bccdc.ca
* Correspondence: rodrigo.munoz-alpizar@canada.ca; Tel.: +1-514-421-7266
Received: 30 June 2017; Accepted: 14 September 2017; Published: 19 September 2017
Abstract:
FireWork is an on-line, one-way coupled meteorology–chemistry model based on
near-real-time wildfire emissions. It was developed by Environment and Climate Change Canada to
deliver operational real-time forecasts of biomass-burning pollutants, in particular fine particulate
matter (PM
2.5
), over North America. Such forecasts provide guidance for early air quality alerts that
could reduce air pollution exposure and protect human health. A multi-year (2013–2016) analysis of
FireWork forecasts over a five-month period (May to September) was conducted. This work used an
archive of FireWork outputs to quantify wildfire contributions to total PM
2.5
surface concentrations
across North America. Different concentration thresholds (0.2 to 28
μ
g/m
3
) and averaging periods
(24 h to five months) were considered. Analysis suggested that, on average over the fire season, 76%
of Canadians and 69% of Americans were affected by seasonal wildfire-related PM
2.5
concentrations
above 0.2
μ
g/m
3
. These effects were particularly pronounced in July and August. Futhermore,
the analysis showed that fire emissions contributed more than 1
μ
g/m
3
of daily average PM
2.5
concentrations on more than 30% of days in the western USA and northwestern Canada during the
fire season.
Keywords:
air quality modeling; wildfire smoke; fine particulate matter; wildfire pollution exposure
1. Introduction
Wildfires are large, uncontrolled vegetation fires that result from natural processes or
anthropogenic activities. In North America they are a major natural hazard, with high interannual
variability in both the number of fires and the total burned area. Every year, wildfires consume
millions of hectares of forest in North America, resulting in several community evacuations due
to the direct threat of fire or the indirect threat of heavy smoke [
1
]. According to the 2016 report
of the Canadian Interagency Forest Fire Centre (CIFFC), during the last decade an average of 7000
wildfires occurred each year in Canada and burned an average of 2.6 million hectares per year [
2
].
Annual costs of wildfire suppression in Canada have ranged from about $0.5 billion to $1 billion in the
Atmosphere 2017, 8, 179 90 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2017, 8, 179
last decade [
3
]. As an extreme example, one wildfire in western Canada from 1 May to 4 July 2016
burned an area of 590,000 ha, roughly the size of the Canadian province of Prince Edward Island,
and forced the evacuation of nearly 90,000 people from the city of Fort McMurray in northeastern
Alberta. The damages caused by this fire were estimated to be on the order of $9.5 billion [4].
In addition to economic impacts, wildfires can adversely affect both air quality (AQ) and human
health. The AQ impacts depend on the amount and chemical composition of the emissions from
these fires, the smoke plume dynamics, and the meteorological conditions that drive the transport
and diffusion of wildfire smoke. Biomass burning from wildfires can release significant amounts of
pollutants into the atmosphere, including particulate matter (PM), ammonia (NH
3
), and ozone (O
3
)
precursors such as nitrogen oxides (NOx), volatile organic compounds (VOCs), and carbon monoxide
(CO) [
5
]. Although many of these species are harmful to human health, the population health impacts
of wildfire smoke have been attributed mainly to short-term concentrations of PM less than 2.5
μ
min
aerodynamic diameter (PM
2.5
). Two recent systematic reviews found that short-term smoke exposure
is strongly associated with small increases in daily mortality from all causes, and with acute respiratory
outcomes ranging in severity from increased reporting of symptoms through to increased risk of
hospital admissions. Associations were weaker for acute cardiovascular morbidity and birth outcomes,
but suggestive of effects in both cases [
6
,
7
]. In addition, new evidence about wildfire smoke is
emerging rapidly given the severity of fires across North America over the past decade. Furthermore,
wildfire smoke is playing an increasingly important role in long-term air pollution as fires get larger
and other sources such as motor vehicles and industry come under increasingly strict regulation [
8
].
Long-term exposure to PM
2.5
is associated with the development of a wide and growing range of
chronic diseases [9].
During wildfire events, PM
2.5
concentration at the ground level may be significantly increased,
such that it exceeds the levels established by regulatory agencies to protect the environment and
human health. In Canada, for example, the established Canadian Ambient Air Quality Standards
(CAAQS) for PM
2.5
concentration are an annual mean of 10
μ
g/m
3
and a daily mean of 28
μ
g/m
3
[
10
].
The CAAQS metric for annual concentration of PM
2.5
is based on the three-year average of the annual
average concentrations, and on the three-year average of the annual 98th percentile of the daily 24-h
average concentrations. Furthermore, the AQ impacts of wildfire emissions are not limited to the
local or regional scales. Under some meteorological conditions, wildfire smoke plumes can disperse
widely and travel thousands of kilometers, affecting people living far away from the fire location [
11
].
Observational evidence indicates that the long-range transport of wildfire smoke can episodically
increase PM and O
3
ground-level concentrations at regional and continental scales. For example,
smoke from Canadian wildfires was associated with high concentrations of PM in areas great distances
from the fire source, such as Baltimore and Washington, D.C. in the eastern USA and as far away as
Europe [
11
–
14
]. Moreover, Canadian wildfires have also been linked to increased O
3
concentrations
in Houston, TX and the northeastern USA, as well as Europe [
15
–
19
]. On the other hand, long-range
transport of Siberian wildfire smoke has also contributed to exceedances in O
3
and PM
2.5
on the west
coast of Canada [20,21].
The assessment of human exposure to smoke from wildfires is challenging because such smoke
episodes are typically sporadic and short-lived, with highly variable concentrations in both space and
time [
22
,
23
]. Furthermore, the spatial variability of population exposure to wildfire smoke cannot be
correctly represented based solely on regulatory monitoring data because these data provide limited
spatial coverage. For example, impacts are often observed in populated non-urban areas where
regulatory monitoring networks are sparse or not available. On the other hand, remote sensing data
from satellites can be used over very large areas, covering locations where the monitoring networks
are missing. However, these measurements provide information about the total atmospheric column
of air pollutants rather than the ground-level concentrations. They are also generally not available at
night and can be masked by the occurrence of clouds. Additionally, satellite overpasses may occur
only once a day or every few days, resulting in large amounts of missing information. Therefore,
91
Books
MDPI
Atmosphere 2017, 8, 179
deterministic AQ forecast models have become a useful tool to fill the temporal and spatial gaps in
available measurements and provide guidance about AQ over the coming hours and days [24–31].
AQ forecasting systems consisting of 3D numerical weather prediction (NWP) models with
on-line or off-line chemical transport models (CTMs) have become a valuable tool in the past 15 years.
They can provide guidance in the production of AQ forecasts and assist public health authorities in
understanding pollutant exposures and developing public actions to protect populations against those
exposures. The accuracy of pollutant exposure estimates using modeled AQ data depends on the
ability of the forecast systems to reproduce observed concentrations of air pollutants. The differences
among the current AQ forecast systems that consider anthropogenic emissions of pollutants have
been reviewed recently [
24
,
25
], but to date only a few AQ forecast systems have been developed that
combine information from wildfires and meteorology to retrospectively or prospectively estimate the
emissions, transport, and diffusion of wildfire smoke [26,27].
In order to provide guidance to regional AQ forecasters, first responders, and public health
decision-makers about the dispersion of smoke from large wildfires, Environment and Climate Change
Canada (ECCC) has developed FireWork [
27
], an on-line, one-way coupled meteorology–chemistry
model based on near-real-time wildfire emissions. FireWork was built on the existing ECCC operational
AQ forecast system, and was first deployed in 2013 during the Canadian wildfire season to deliver
real-time forecasts of wildfire smoke plumes over North America. Previous studies have shown the
ability of FireWork to forecast PM
2.5
in terms of statistical scores and spatial distributions, as well as
public health impacts [
27
–
29
]. Observed trends and seasonal variability of PM
2.5
are well captured by
the model. However, FireWork’s ability to simulate the emission and dispersion of wildfire smoke
is currently limited by factors such as the accuracy of wildfire emission factors, the treatment of fire
behavior, and the suitability of plume-rise algorithms.
Here we conduct a multi-year (2013–2016) analysis of FireWork forecasted PM
2.5
concentrations
from biomass burning over North America to provide estimates of the population exposure to PM
2.5
from wildfires for several concentration thresholds. The number of days that exceed these thresholds
as well as the magnitude of the area in exceedance was estimated. The goal of this work is to help
public health professionals, policymakers, and the general public better understand the human health
impacts of wildfire-related PM
2.5
pollution.
2. Methodology
2.1. North American Wildfire AQ Forecasting System
The FireWork system was first run in an experimental mode beginning in 2013 at ECCC’s
Canadian Centre for Meteorological and Environmental Prediction (CCMEP). The system became
operational in April 2016. The FireWork system is identical to the ECCC operational Regional
Air Quality Deterministic Prediction System (RAQDPS) [
30
–
32
], except for the inclusion of
satellite-derived, near-real-time biomass burning emissions from natural, prescribed, and agricultural
fires [
27
]. The on-line RAQDPS modeling system relies on the GEM-MACH (Global Environmental
Multi-scale-Modelling Air quality and Chemistry) model, an on-line, one-way coupled CTM
(i.e., meteorology affects chemistry, but chemistry does not affect meteorology), embedded within the
Global Environmental Multi-scale (GEM) model. Both the RAQDPS and FireWork systems input the
same hourly anthropogenic gridded emissions fields based on processing the 2010 Canadian national
Air Pollutant Emission Inventory (APEI), the 2011 U.S. National Emissions Inventory (NEI), and the
1999 Mexican emissions inventory, as well as biogenic and sea-salt emissions from natural sources [
31
].
Each of the three national anthropogenic inventories accounts for emissions of at least seven criteria
air pollutants: SO
2
,NO
x
, VOC, CO, NH
3
,PM
2.5
, and PM
10
.
The calculation of the near-real-time biomass burning emissions required by FireWork starts
with the Canadian Forest Service’s operational Canadian Wildland Fire Information System (CWFIS),
which provides fire activity and fire danger conditions across Canada and the continental United
92
Books
MDPI
Atmosphere 2017, 8, 179
States during the active wildfire season [
33
]. The primary data used by the CWFIS to capture
fire activity come from satellite-based detection systems: NASA’s Moderate Resolution Imaging
Spectroradiometer (MODIS) instrument, NOAA’s Advanced Very High Resolution Radiometer
(NOAA/AVHRR), and Visible Infrared Imaging Radiometer Suite (VIIRS) imagery through NASA and
the U.S. Forest Service Remote Sensing Applications Center [
34
]. During the fire season, fire activity
is updated six times daily in the CWFIS, corresponding to the frequency of available satellite-based
retrievals. Relevant fire information estimated by CWFIS for each fire hotspot includes fuel type,
surface fuel consumption, crown fuel consumption, total fuel consumption, and forest floor fuel
consumption. Estimates of daily biomass-burning emissions for individual hotspots are then obtained
using fuel consumption values from the CWFIS and emission factors from the Fire Emission Production
Simulator (FEPS), a component of the BlueSky Modeling Framework [
26
]. More detailed information
about the FireWork modelling system framework and its data flow is provided in other recent
publications [27,28].
In the current operational setup, the seasonal FireWork system runs twice per day at 00 UTC and
12 UTC during the North American fire season from 1 April to 31 October. FireWork simulation results
provide numerical AQ forecast guidance over North America with a 48-h lead time. In 2013, 2014,
and 2015, when FireWork was run as an experimental model version at CCMEP, the period from April
to October was only partially covered (see Table 1). In April 2016, however, when the FireWork System
became operational [35], FireWork forecasts were extended to cover the full wildfire season.
The seasonal peak for wildfire events in Canada occurs in the months of June, July, and August [
2
],
and initially our analysis focused on this three-month period. However, due to the extreme wildfires
that occurred in northern Alberta in May 2016, we decided to extend our analysis to a five-month
period from May to September. In order to backfill this five-month period for the years 2013–2015
(Table 1), FireWork was rerun using the operational forecasting approach [
27
] with the same FireWork
version that had been used each year. This required three older versions of FireWork to be run because
new, updated versions of the RAQDPS had been introduced before each fire season [31,35].
Table 1.
2013–2016 experimental and operational FireWork start/end forecast periods together with
additional periods for which FireWork was rerun retrospectively.
Year Experimental/Operational FireWork Added Periods
Start End
2013 June 1 August 31 May 1–31; Sep. 1–31
2014 June 9 October 1 May 1–June 8
2015 May 21 October 31 May 1–20
2016 April 1 October 31 -
Seasonal fire emissions for May 1 through September 30 were estimated for North America
using the FireWork emissions system [
27
]. The total fire emissions of key trace gases and particulate
species for each year from 2013–2016 show a significant variation in the seasonal totals (Table S1;
see Supplementary Materials). Maximum emissions were observed for the 2014 season, and there
was substantial spatial variability in the regional estimates (data not shown). The mean seasonal
FireWork primary PM
2.5
emissions for North America for 2013–2016 of 1.4 Tg/season are lower than
but comparable to previous estimates of North American annual PM
2.5
emissions from wildfires
(1.9 and 2.2 Tg/y) [
36
]. For context, total U.S. anthropogenic PM
2.5
emissions in 2010 were estimated
to be 4.1 Tg/y [37], so fire emissions are an important source of PM
2.5
.
The FireWork domain covers most of Canada and the USA (including Alaska), as well as northern
Mexico, with a 10 km
×
10 km grid (Figure 1). A new operational version of FireWork with a
new domain and new 10 km
×
10 km grid (Figure 1) was introduced during the 2016 wildfire
season [
35
]. Results presented in Section 3 are calculated on the original grid used by FireWork prior
to 7 September 2016. FireWork results for the period after the grid change (approximately three weeks)
93
Books
MDPI
Atmosphere 2017, 8, 179
were interpolated to the original grid to allow for consistent analysis. The area covered by the original
FireWork domain was 33,085,648 km
2
.
Figure 1.
FireWork domain boundaries before (green) and after (red) 7 September 2016.
The 10 km × 10 km grid is not shown.
2.2. Wildfire Emissions’ Contribution to PM
2.5
Pollution
We used FireWork forecasts to analyze the contribution of fire-originated fine particulate matter
(fire-PM
2.5
)toPM
2.5
pollution over North America. In order to estimate the direct contribution
of fire-PM
2.5
to the total PM
2.5
concentration forecasted by FireWork, the RAQDPS forecast PM
2.5
concentration field valid at the same hour was subtracted from the FireWork field. This simple strategy
removes the contribution of the anthropogenic sources and other natural sources considered by both
the RAQDPS and FireWork, and makes it possible to isolate wildfire smoke plume locations and follow
their evolution over time [
27
]. The analysis of forecasted wildfire smoke presented in this paper is
based on the set of hourly PM
2.5
concentration fields generated by this subtraction. Note that fire-PM
2.5
includes contributions from both primary PM
2.5
emissions and secondary aerosol formation from
primary gas-phase emissions.
An essential part of characterizing the impacts of exposure to wildfire pollution is to understand
both long-term (monthly to yearly) and short-term (hourly to daily) exposures. We used multi-year
FireWork simulations (2013–2016) to characterize both long- and short-term wildfire pollution exposure
over North America by calculating averages based on multi-year, seasonal, monthly, daily, and hourly
concentrations and assessing areas affected by different concentration thresholds. Furthermore,
we compared these averages with the PM
2.5
CAAQS of 10
μ
g/m
3
(annual standard) and 28
μ
g/m
3
(daily standard) [
10
], and with lower thresholds of 0.2, 1, and 5
μ
g/m
3
. The 0.2
μ
g/m
3
threshold was
based on the U.S. Environmental Protection Agency (EPA) Significant Impact Level (SIL) guidance
document, which defines 0.2
μ
g/m
3
as the threshold below which any annual PM
2.5
change is
considered negligible [
38
]. Also, from our own experience with FireWork, 0.2
μ
g/m
3
is the lowest
value not susceptible to numerical noise that can be considered when analyzing the contributions of
fire-PM
2.5
to total forecasted PM
2.5
concentrations. The 1 and 5
μ
g/m
3
thresholds were considered to
transition between the minimal 0.2 μg/m
3
threshold and the 10 μg/m
3
threshold.
94
Books
MDPI
Atmosphere 2017, 8, 179
2.3. Population Exposure Estimation
Statistics on population exposure to wildfire smoke for Canada and the USA can be calculated by
combining FireWork fields of direct contributions of fire-PM
2.5
to total surface PM
2.5
concentration
with population data. We used population data from the 2016 Canadian census [
39
,
40
] and from the
2010 U.S. census [
41
]. For the 2016 Canadian census, we used population reported at the dissemination
area (DA) level, where each DA typically has a population of 200 to 1000 people. For the 2010 U.S.
census, we used population reported at the block-group level (Figure 2). Although U.S. population
projections are available for 2016, they are only available at the coarser census-tract level rather than
the more finely resolved block-group level (see Figure S1 in the Supplementary Materials).
Figure 2.
Population count per FireWork grid cell (10 km
×
10 km) based on the 2016 Canadian
census and the 2010 U.S. census. The red box over the four southern Great Lakes marks the location of
the inset.
A number of steps were required to estimate the population affected at different wildfire PM
2.5
concentration thresholds. The first step was to determine the population for each 10 km
×
10 km
grid cell on the FireWork domain. To do so, 2016 Canadian population data reported by DA [
39
]
were incorporated into a shapefile containing DA polygons [
40
]. The same step was performed at
the sub-county level for 2010 U.S. population data [
41
]. The two population shapefiles were then
interpolated separately to the 10 km by 10 km FireWork grid using a normalized conservative approach
that preserved population within polygons. This interpolation approach divided the population value
of each polygon between the grid cells wholly or partly contained within the polygon based on
fractional area and assuming uniform population density within a DA or sub-county (Figure 2).
The total populations contained in the older FireWork domain (Figure 1) for these two censuses were
35,148,512 in Canada and 305,744,285 in the USA.
The next step was to identify the aggregate population for the set of FireWork grid cells above a
PM
2.5
concentration threshold. The Canadian and U.S. populations were processed separately. For grid
cells along the Canada-USA border, it was necessary to determine in which country each cell was
mainly located. This was done using a mask indicating the country associated with each grid cell.
The population values for the set of FireWork grid cells associated with each PM
2.5
threshold were then
summed together. The final step was to assess population exposure to different PM
2.5
concentrations.
95
Books
MDPI
Atmosphere 2017, 8, 179
To do this, five PM
2.5
thresholds (0.2, 1, 5, 10, and 28
μ
g/m
3
) were considered and applied to monthly
and seasonal (May to September) contribution of fire-PM
2.5
to total surface PM
2.5
concentrations.
2.4. Exposure Frequency Estimation
The temporal frequency of PM
2.5
pollution exposure is another critical factor in determining
health impacts from fire-PM
2.5
. One way to characterize this factor is to examine the number of
occurrences of hourly or daily fire-PM
2.5
concentrations above specific thresholds for individual grid
cells. This is done by counting the number of hours or days in a wildfire season with wildfire-related
PM
2.5
concentrations above four different levels: 1, 5, 10 and 28
μ
g/m
3
. In this case the lowest threshold
of 0.2 μg/m
3
was not considered because it is less meaningful over shorter averaging periods.
3. Results
3.1. Area Affected by Wildfire Smoke
In 2013–2016, almost all areas of Canada and the USA included in the FireWork domain were
affected by wildfire smoke based on a seasonal fire-PM
2.5
> 0.2
μ
g/m
3
exceedance at least once
per grid cell (Figures 3–6). Western North America was more affected by wildfire smoke than
eastern North America in all four years. From a continental perspective, 2014 had the most intense
wildfire season, based on the seasonal values of the area affected by wildfire pollution for PM
2.5
concentration thresholds from 1
μ
g/m
3
to 28
μ
g/m
3
(Table 2). This year was followed by 2015
and 2013, while 2016 was the year least affected by wildfire smoke. Based on the average seasonal
concentrations, the percentage areas of the FireWork domain (including land and water areas) above
the 0.2
μ
g/m
3
threshold were 52%, 49%, 44%, and 22% for the years 2013–2016, respectively (Table 2).
Above the 1
μ
g/m
3
threshold the corresponding percentage areas were 14%, 17%, 13%, and 6%,
and above the 5 μg/m
3
threshold the values were 0.8%, 1.9%, 1.1%, and 0.4%.
Table 2.
FireWork domain area affected (km
2
and percentages) by wildfire pollution above five PM
2.5
concentration thresholds based on average monthly and seasonal fire-PM
2.5
contributions to total
average monthly and seasonal surface PM
2.5
concentrations. For reference, the total area of North
America is 24.71 million km
2
and for the FireWork domain is 33.09 million km
2
. The numbers in
parentheses correspond to the percentage of the area affected.
μg/m
3
>0.2 >1 >5 >10 >28
2013
May 754,020 (2.3%) 30,662 (0.1%) 1377 (0.0%) 196 (0.0%) 0 (0.0%)
June 12,478,900 (37.7%) 2,532,748 (7.7%) 184,058 (0.6%) 48,991 (0.1%) 7378 (0.0%)
July 17,235,170 (52.1%) 7,279,618 (22.0%) 861,986 (2.6%) 194,966 (0.6%) 13,326 (0.0%)
August 20,155,062 (60.9%) 9,873,087 (29.8%)
1,783,453
(5.4%) 604,476 (1.8%) 71,844 (0.2%)
September 9,351,929 (28.3%) 2,072,752 (6.3%) 84,693 (0.3%) 39,022 (0.1%) 10,611 (0.0%)
2014
May 1,746,428 (5.3%) 61,310 (0.2%) 6507 (0.0%) 2137 (0.0%) 200 (0.0%)
June 3,411,109 (10.3%) 872,761 (2.6%) 57,926 (0.2%) 13,461 (0.0%) 1686 (0.0%)
July 16,196,537 (49.0%) 7,260,984 (21.9%)
1,562,298
(4.7%) 541,468 (1.6%) 146,616 (0.4%)
August 19,536,859 (59.0%) 8,436,851 (25.5%)
2,560,671
(7.7%) 934,005 (2.8%) 157,462 (0.5%)
September 13,305,055 (40.2%) 4,531,984 (13.7%) 444,316 (1.3%) 137,840 (0.4%) 24,680 (0.1%)
2015
May 2,837,325 (8.6%) 252,707 (0.8%) 12,768 (0.0%) 2950 (0.0%) 197 (0.0%)
June 9,375,291 (28.3%) 2,604,739 (7.9%) 291,593 (0.9%) 42,184 (0.1%) 2552 (0.0%)
July 14,245,102 (43.1%) 5,173,825 (15.6%)
1,033,576
(3.1%) 306,687 (0.9%) 30,776 (0.1%)
August 15,581,769 (47.1%) 5,918,077 (17.9%)
1,236,416
(3.7%) 636,671 (1.9%) 143,120 (0.4%)
September 11,999,087 (36.3%) 1,629,073 (4.9%) 102,071 (0.3%) 45,570 (0.1%) 14,173 (0.0%)
2016
May 3,829,030 (11.6%) 972,851 (2.9%) 167,935 (0.5%) 58,168 (0.2%) 11,955 (0.0%)
June 2,532,347 (7.7%) 334,211 (1.0%) 35,392 (0.1%) 13,575 (0.0%) 3599 (0.0%)
July 7,866,765 (23.8%) 2,709,907 (8.2%) 223,084 (0.7%) 57,209 (0.2%) 8215 (0.0%)
August 7,807,827 (23.6%) 2,839,017 (8.6%) 477,835 (1.4%) 146,340 (0.4%) 31,658 (0.1%)
September 4,914,590 (14.9%) 1,715,135 (5.2%) 204,326 (0.6%) 71,075 (0.2%) 15,620 (0.0%)
96
Books
MDPI
Atmosphere 2017, 8, 179
Table 2. Cont.
μg/m
3
>0.2 >1 >5 >10 >28
Seasonal Exceedances (Area)
2013 17,151,407 (51.8%) 4,534,747 (13.7%) 269,933 (0.8%) 62,226 (0.2%) 9267 (0.0%)
2014 16,335,869 (49.4%) 5,616,429 (17.0%) 625,489 (1.9%) 216,897 (0.7%) 28,417 (0.1%)
2015 14,507,744 (43.8%) 4,203,396 (12.7%) 373,565 (1.1%) 90,825 (0.3%) 16,248 (0.0%)
2016 7,229,006 (21.8%) 1,931,583 (5.8%) 121,897 (0.4%) 40,841 (0.1%) 9388 (0.0%)
In terms of total area burned, in 2013 Canada had its seventh most intense wildfire season of the
past 34 years [
2
], whereas the USA was 37% below its 2006–2016 average [
42
]. The wildfire season in
2013 effectively started in June, with maximum intensity reached in July and August. During these
two months, the area for which the monthly average of fire-PM
2.5
exceeded the 0.2
μ
g/m
3
threshold
covered most of North America (Figure 3). In July, an area of 194,966 km
2
(the size of South Dakota)
had average monthly fire-PM
2.5
above 10
μ
g/m
3
and 13,326 km
2
were above 28
μ
g/m
3
(Table 2).
In August, using the same thresholds, these areas were 604,476 km
2
(larger than California; close to
the size of Manitoba and Texas) and 71,844 km
2
, respectively. August 2013 was the most active month
of the year, with intense activity in northwestern and western Canada and the northwestern USA.
In 2014, Canada had its fifth most intense wildfire season of the past 34 years in terms of total area
burned, while in the USA the value was 48% below its 2006–2016 average [
2
,
42
]. The extreme wildfire
event in the Northwest Territories near the city of Yellowknife started in June and peaked in July
(Figure 4). In July, intense wildfires began burning in British Columbia, Alberta, Washington, Oregon,
California, and Idaho, bringing the total area with average monthly fire-PM
2.5
>10
μ
g/m
3
to 541,468
km
2
,and>28
μ
g/m
3
to 146,616 km
2
. These fires persisted into August, making August 2014 the most
extreme month of the 2013–2016 period in terms of area affected by wildfire smoke. In this month,
average monthly fire-PM
2.5
values above the 5, 10, and 28
μ
g/m
3
thresholds covered 2,560,671 km
2
(larger than Alaska or Nunavut), 934,005 km
2
(the size of British Columbia), and 157,462 km
2
(the size
of Georgia), respectively (Table 2).
The 2015 season was the sixth most intense of the past 34 years for Canada in terms of area
burned [
2
]. In the USA, it was the peak year of the 2006–2016 period, with more than 10 million acres
(40,500 km
2
) burned, 45% above the period average [
42
]. The 2015 fire season was marked by two
intense wildfire periods: the first from 15 June to 15 July and the second from 1 to 15 August [
27
]
(Figure 5). In the first period, most of the wildfires occurred in northwestern Canada (Alberta and
Saskatchewan), whereas in the second period most of the wildfires occurred in the western USA
(Washington, Oregon, Idaho, and California). In July and August areas of 306,687 km
2
and 636,671 km
2
had average monthly fire-PM
2.5
>10
μ
g/m
3
, respectively, and areas of 30,776 km
2
and 143,120 km
2
were >28 μg/m
3
(Table 2).
The 2016 fire season included unprecedented impacts in Canada on both people and the national
economy. The entire city of Fort McMurray, Alberta, with a population of nearly 90,000, was evacuated
in May when it was overrun by a large, fast-moving wildfire (Figure 6). Estimated insured fire damages
to Fort McMurray were 9.6 billion dollars, the costliest insured natural disaster in Canadian history [
4
].
Despite this disaster, 2016 was the least intense wildfire season among the four years analyzed in
terms of area burned [
2
] and area affected by wildfire smoke across Canada (Table 2). In the USA,
the 2016 area burned was 21% below its 2006–2016 average [
42
]. On the other hand, the early start to
the wildfire season in Canada made the month of May 2016 the most affected by wildfire pollution
among the four Mays assessed. For May 2016 the area with average monthly fire-PM
2.5
>10
μ
g/m
3
and > 28 μg/m
3
was 58,168 km
2
and 11,955 km
2
respectively (Table 2).
97
Books
MDPI
Atmosphere 2017, 8, 179
(A) (B)
(C) (D)
(E) (F)
Figure 3.
2013 average monthly fire-PM
2.5
contribution to total forecasted surface PM
2.5
concentrations
(
μ
g/m
3
) for (
A
) May, (
B
) June, (
C
) July, (
D
) August, (
E
) September, and (
F
) seasonal (May–September
average). Note that the color scale is non-linear and white areas indicate values < 0.2 μg/m
3
.
It is also of interest to examine the importance of fire-PM
2.5
relative to other sources of PM
2.5
.
Figure S2 shows the seasonal fire-PM
2.5
contribution to total PM
2.5
as a percentage for each of the
four years. For 2013 to 2015, the seasonal contribution of fire-PM
2.5
to total PM
2.5
was 50% or more
over much of northwestern North America and parts of the U.S. mountain west. In 2014 the seasonal
wildfire contribution was 90% or greater for a large part of the Northwest Territories and parts of
the interior of British Columbia. These results are not surprising considering that these areas have
relatively few inhabitants and low anthropogenic emissions.
It is difficult to compare the monthly model forecasted fire-PM
2.5
directly with PM
2.5
measurements, as measurements are influenced by all PM
2.5
sources, not just biomass burning
98
Books
MDPI
Atmosphere 2017, 8, 179
emissions. As an indirect comparison, however, we note that archived near-real-time measurement
from the AirNow data feed (www.airnow.gov) include at least one U.S. or Canadian AQ station
located close to wildfires having monthly PM
2.5
concentrations above 30, 50, and even 150
μ
g/m
3
for each of the four years analyzed here. The most extreme month was August 2015, when 13 PM
2.5
measurement stations reported mean monthly PM
2.5
concentrations of 30
μ
g/m
3
or above, including
six stations in Idaho, two stations each in Oregon, Washington State, and British Columbia, and one
station in Montana (see Table S2). FireWork also forecasted mean monthly fire-PM
2.5
concentrations
above 30
μ
g/m
3
for two regions of California in August 2015; although no stations with available
measurements were located in these regions, one nearby station in Shasta county in northern California
had a mean monthly PM
2.5
concentration of 24.5 μg/m
3
and 100% data completeness.
(A) (B)
(C) (D)
(E) (F)
Figure 4. Same as Figure 3 but for 2014.
99
Books
MDPI
Atmosphere 2017, 8, 179
(A) (B)
(C) (D)
(E) (F)
Figure 5. Same as Figure 3 but for 2015.
(A) (B)
Figure 6. Cont.
100
Books
MDPI
Atmosphere 2017, 8, 179
(C) (D)
(E) (F)
Figure 6. Same as Figure 3 but for 2016.
3.2. Population Exposure to Wildfire Pollution
In terms of population exposure to wildfire-related PM
2.5
pollution, for 17 of the 20 months
considered, more than 1 million Canadians (3% of the population) were estimated to have been
affected by average monthly fire-PM
2.5
> 0.2
μ
g/m
3
(Figure 7). During the same period, more than
14 million Canadians (39% of the population) were affected by average monthly fire-PM
2.5
> 0.2
μ
g/m
3
in 10 of the 20 months, and over 32 million Canadians (90% of the population) were affected in 7 months.
The months affecting the most people were July and August (Figure 8 and Table S3), with August 2015
being the worst for average monthly fire-PM
2.5
>10
μ
g/m
3
. During August 2015 the proportion of the
Canadian population affected by fire-PM
2.5
above thresholds of 0.2, 1, 5, 10, and 28
μ
g/m
3
were 97%,
21%, 11%, 8%, and 1.2%, respectively. The three periods in which the most Canadians were exposed to
>28
μ
g/m
3
were: (a) August 2015 (417,171 people or 1.2% of the population) due to extreme wildfires
in British Columbia and the northwestern USA; (b) May 2016 (69,909 people or 0.2% of the population)
due to extreme wildfires in northern Alberta near Fort McMurray; and (c) August 2014 (27,160 people
or 0.1% of the population) due to extreme wildfires in northwestern Canada. In all other months,
less than 0.1% of the Canadian population was affected at this very high threshold (Table S3).
101
Books
MDPI
Atmosphere 2017, 8, 179
(A)
(B)
0
5
10
15
20
Total Number of Months
Population (millions)
Distribution of monthly Canadian population affected by
wildfire pollution
>0.2
>1
>5
>10
>28
0
5
10
15
20
Total Number of Months
Population (millions)
Distribution of monthly U.S. population affected by wildfire
pollution
>0.2
>1
>5
>10
>28
Figure 7.
Cumulative distribution of number of months with the portions of the Canadian (
A
) and
U.S. (
B
) population affected by monthly fire-PM
2.5
above five PM
2.5
concentration thresholds for four
five-month wildfire seasons (2013–2016).
Based on the average seasonal statistics (Table S3), over 90% of the Canadian population was
affected by seasonal fire-PM
2.5
> 0.2
μ
g/m
3
in 2013, 2014 and 2015. The 2016 wildfire season was much
milder, with 19% of Canadian population affected by wildfire smoke above this threshold (Figure 8).
The corresponding proportions for the 1
μ
g/m
3
threshold ranged from 0.4% (2016) to 26% (2014).
The population affected by concentrations above the 10
μ
g/m
3
threshold reached its maximum in
2015, with over 100,000 people (Table S3).
In the USA, more than 200 million people (65% of the population) were exposed to average
monthly fire-PM
2.5
> 0.2
μ
g/m
3
during nine of the 20 months considered (Figure 7). As well, a much
greater proportion of the U.S. population was affected by wildfire pollution in 2013, 2014, and 2015
than in 2016, similar to Canada (Figure 8), and for both the USA and Canada, the proportion of the
population exposed to wildfire pollution in September was larger than in June.
The total percentages of the U.S. population affected by seasonal fire-PM
2.5
> 0.2
μ
g/m
3
ranged
from 21% (2016) to 90% (2015) (Figure 8 and Table S4), and the corresponding range above 1
μ
g/m
3
was 5% (2015) to 15% (2014). Based on the four-year average seasonal statistics for population exposure
to fire-PM
2.5
> 0.2
μ
g/m
3
, a smaller percentage of the U.S. population (69%) than the Canadian
population (76%) was exposed to pollution above this threshold (Tables S3 and S4). For concentrations
>1
μ
g/m
3
the corresponding percentages were 10% in the USA and 12% in Canada. However,
for average seasonal fire-PM
2.5
>28
μ
g/m
3
, a higher percentage of Americans than Canadian were
exposed. The affected U.S. population ranged from 32,549 in 2014 to 56,442 in 2015 (Table S4), giving a
four-year average exposure of 0.015% for the U.S. population compared with 0.004% for the Canadian
population (Table S3).
102
Books
MDPI
Atmosphere 2017, 8, 179
(A) (B)
(C) (D)
(E) (F)
(G) (H)
(I) (J)
Figure 8.
Percentage of population in Canada (left panels:
A
,
C
,
E
,
G
, and
I
) and the USA (right panels:
B
,
D
,
F
,
H
, and
J
), affected by wildfire pollution above five PM
2.5
concentration thresholds based
on the average monthly and seasonal fire-PM
2.5
contribution to total average monthly and seasonal
surface PM
2.5
concentrations for four wildfire seasons (2013–2016). The percentage of the affected
population for Canada and the USA was calculated using the 2016 Canadian and the 2010 U.S. censuses,
respectively. See also Tables S3 and S4.
3.3. Frequency of Wildfire-Related Pollution Events
In 2013–2016 most of North America (except Alaska) was affected by wildfire smoke on at least
one day (Figure 9). The land-use map for Alaska used by National Resources Canada to determine total
fuel consumption was only updated in 2015. Given the direct impact of wildfire emissions estimates
103
Books
MDPI
Atmosphere 2017, 8, 179
from input land-use in FireWork, there may be an underestimation of wildfire emissions for regions of
Alaska prior to 2015. The highest daily frequency of wildfire smoke occurred in 2014, where for most
of western Canada and the northwestern USA more than 30% of the days from May to September had
surface fire-PM
2.5
greater than 1
μ
g/m
3
for at least one hour. In parts of Alberta, Saskatchewan and
the Northwest Territories, the daily frequency was above 40%, and in part of the southern Northwest
Territories it was over 60%. For hourly frequency, in 2013 the majority of Canada and the USA had
more than 10% of hourly forecasted fire-PM
2.5
above 1
μ
g/m
3
. Hourly frequency was slightly lower in
2014 and 2015, but in 2016, only a portion of western North America had hourly frequencies over 10%
(Figure 9).
The daily frequency of forecasted fire-PM
2.5
concentrations greater than 5
μ
g/m
3
was above 10%
for the majority of western Canada in 2013, 2014 and 2015 (Figure 10). The same was true for northern
Quebec in 2013. In the case of the USA, regions over 10% were found only in the west over the same
period. In 2016, frequencies above 10% were less common when compared with other years and
were limited to areas close to the wildfires in northwestern Canada and the northwestern USA and
California. Frequencies above 30% were found in 2014 in northwestern Canada, while this percentage
was limited to areas very close to the wildfires in the other years. The results of the hourly frequency
analysis are similar to those of the daily frequency analysis, with values of 10–20% covering large areas
of northwestern Canada and the western USA, especially for 2014. Hourly frequencies over 20% were
not present over Canada in 2016 and were limited to areas close to wildfires in 2013, 2014, and 2015
(Figure 10).
The spatial patterns of daily and hourly frequencies for the 10
μ
g/m
3
concentration threshold
were very similar to the patterns observed for the 5
μ
g/m
3
threshold (Figure 11). However, 2014 was
the only year with daily and hourly frequencies above 30%, in northwestern Canada. For a threshold
of 28
μ
g/m
3
(Figure 12), the frequency of days and hours with hourly forecasts above this threshold
was generally below 10%.
We can also look at the number of days with elevated fire-PM
2.5
from a population exposure
perspective. Tables S5 and S6 provide this information for Canada and the USA, respectively. In 2014
wildfire season, more than 14% of Canadians were exposed to a daily fire-PM
2.5
>5
μ
g/m
3
on at least
30 days (i.e., 20% or more days) and 12% of Canadians were exposed to a daily
fire-PM
2.5
>10μg/m
3
on at least 15 days (Table S5). For the USA, the corresponding values in 2014 were 2% and 3%,
but interestingly they were higher (3% and 4%) in 2016 (Table S6), even though in other respects 2016
had a less active fire season (e.g., Table 2).
MAY–SEPTEMBER
% of days % of hours
2013
Figure 9. Cont.
104
Books
MDPI
Atmosphere 2017, 8, 179
2014 2015 2016
Figure 9.
Percentage frequency of the number of days (
left
) and the number of hours (
right
) with
forecasted 24-h moving average PM
2.5
concentration above 1
μ
g/m
3
from fire-PM
2.5
contribution for
the period May–September for years 2013, 2014, 2015, and 2016. White areas indicate locations that
experienced no days or hours above the threshold during the period.
MAY–SEPTEMBER
% of days % of hours
2013
Figure 10. Cont.
105
Books
MDPI
Atmosphere 2017, 8, 179
2014 2015 2016
Figure 10. Same as Figure 9 but for 5 μg/m
3
.
MAY–SEPTEMBER
% of days % of hours
2013
Figure 11. Cont.
106
Books
MDPI
Atmosphere 2017, 8, 179
2014 2015 2016
Figure 11. Same as Figure 9 but for 10 μg/m
3
.
MAY–SEPTEMBER
% of days % of hours
2013
Figure 12. Cont.
107
Books
MDPI
Atmosphere 2017, 8, 179
2014 2015 2016
Figure 12. Same as Figure 9 but for 28 μg/m
3
.
4. Discussion
The spatial distributions of monthly and seasonal wildfire plumes illustrate the large spatial
and temporal variability of wildfire occurrence in North America (Figures 3–6). Our results also
highlight how common wildfires are each summer and the large spatial extent of their influence
due to long-range transport of wildfire emissions. Analyses of daily elemental and organic carbon
measurements from the U.S. IMPROVE speciated PM
2.5
measurement network support the analysis of
the FireWork wildfire smoke forecasts presented here, indicating that smoke from wildfires contributes
substantially to PM
2.5
levels in the western USA [43–46].
A related finding is that wildfires can sometimes impact the same location many times during a
single season (Figures 9–12). The frequent presence of fire-PM
2.5
shown here for 2013–2016, especially
in western North America, has implications for regional attainment of PM
2.5
regulatory objectives.
Both Canadian and U.S. standards for PM
2.5
allow the exclusion of days with concentrations above
the national standard due to wildfire smoke. However, to invoke such an exclusion, it is necessary
to demonstrate that an unmanaged emission source such as wildfires is the cause of an elevated
concentration. FireWork forecasts could provide useful evidence for this purpose.
108
Books
MDPI
Atmosphere 2017, 8, 179
It is also relevant to compare the distribution of population in North America (Figure 2) with the
distribution of wildfire smoke (Figures 3–6). At the continental scale the majority of the North American
population lives in the eastern half of the continent whereas the majority of large wildfires occur in
the western half. This anticorrelation reduces the degree of population exposure to wildfire smoke.
Nevertheless, 32% of the Canadian population lives west of Ontario and 41% of the U.S. population
lives west of the Mississippi River, closer to western wildfires. As an example of this difference between
wildfire location and population location, consider that for the 2013–2016 period (a) August 2014
was identified as the month with the greatest areal extent of average monthly
fire-PM
2.5
>28μg/m
3
,
whereas (b) August 2015 was identified as the month with the largest U.S. population exposure to
average monthly fire-PM
2.5
>28μg/m
3
.
Short-term exposure to PM
2.5
from wildfire smoke puts the population at an increased risk of
experiencing a wide range of acute health outcomes, particulary those with chronic conditions such
as asthma [
47
] or heart disease [
48
]. Long-term exposure to PM
2.5
from all sources, including the
contribution from wildfire smoke, increases the risk of developing chronic conditions, such as asthma
or heart disease [
9
]. Global estimates suggest that approximately 340,000 deaths per year can be
attributed to smoke from landscape fires, of which only 18% are due to short-term effects while 82%
are due to the long-term effects [
49
]. Our results confirm that fire-PM
2.5
puts the Canadian and U.S.
populations at risk from both short- and long-term exposure. Although our long-term averages covered
the five-month fire season rather than the entire year, we found that up to 26.4% of the Canadian
population and 14.7% of the U.S. populations were affected by increases of > 1
μ
g/m
3
in the extreme
fire seasons. This likely indicates that the annual fire-PM
2.5
averages were > 0.2
μ
g/m
3
for these
populations, which is defined by the U.S. EPA as a non-negligible impact [
38
]. Given that wildfires are
becoming more frequent and intense across North America [
2
,
42
], tools such as FireWork can help to
characterize their contribution to the long-term exposure most responsible for the burden of disease
attributable to air pollution.
A similar wildfire pollution exposure study was recently published for the USA [
23
], in which
fire-PM
2.5
contributions were estimated for an earlier five-year period (2008–2012) using paired
retrospective simulations performed with another AQ modeling system. As in this study, one simulation
considered wildfire emissions and one did not, and then a post-simulation subtraction of predicted
paired surface PM
2.5
fields yielded the fire-PM
2.5
contribution estimate. One difference between the two
studies was the fire seasons sampled: the average annual U.S. area burned during their study period
(2008–2012) was 11% higher than that during our study period (2013–2016) [
42
]. Other differences were
the concentrations thresholds (in
μ
g/m
3
) that they considered (0.15, 0.75, 1.5) compared with those that we
considered (0.2, 1, 5, 10, 28), and the annual concentration values aggregated from
12 km × 12 km
grid cells
to the county level that they considered vs. the monthly and five-month values for
10 km × 10 km
grid cells
that we considered (i.e., higher temporal and spatial resolution). Nevertheless, a limited comparison of the
two studies is possible. Based on the 2010 U.S. census (also used in this study), Rappold et al. [
23
] estimated
that 10% of the U.S. population lived in areas where the contribution of fire-PM
2.5
was >1.5
μ
g/m
3
.
We estimated that between 5.3% (in 2015) and 14.7% (in 2014) of the U.S. population lived in areas where
the five-month fire-PM
2.5
contribution was >1 μg/m
3
. Rappold et al. [23] also estimated that 10.3 million
individuals in the U.S. lived in areas having 10 or more days (between 2008 and 2012) with fire-PM
2.5
contribution > 35
μ
g/m
3
, and we estimated that 10.6 million individuals in the USA lived in areas having 10
or more days (for all five-month seasons between 2013 and 2016) with fire-PM
2.5
>28
μ
g/m
3
. Considering
the differences between these two studies, this basic comparison suggests that the results are comparable
and consistent.
This study was an “analysis of opportunity” based on the availability of four years of daily North
American wildfire smoke forecasts, and further improvements are likely possible. For example, wildfires
in Siberia are known to affect Alaska and western Canada, but FireWork does not currently consider
wildfire emissions external to North America. In our operational on-line FireWork performance evaluation,
a negative bias in PM
2.5
concentration forecasts is observed for long-range wildfire pollution advection. As a
109
Books
MDPI
Atmosphere 2017, 8, 179
consequence, the results presented here are likely to be conservative. Work is ongoing to examine potential
improvements to the PM
2.5
emission factors and the plume-rise parameterization used by FireWork, both of
which influence PM
2.5
concentration predictions. Given the horizontal grid spacing of 10 km used by
FireWork, it is expected too that forecasts of near-source concentration will be underestimates due to the
limitation of assuming uniform emissions within a grid cell. ECCC also has an AQ objective analysis
system that combines AQ model predictions with AQ observations [
50
]. This system is now used with
FireWork, opening the possibility of analyzing optimally-combined fields of model predictions and PM
2.5
measurements instead of relying solely on model forecast fields. Finally, our results might have differed
slightly had we run the same FireWork version for all four years instead of using archived FireWork
versions from each year (which ensured that our analyses reflected the information available to ECCC
forecasters and stakeholders at the time). However, none of these changes are likely to alter our overall
conclusions about the impact of wildfires in North America.
5. Conclusions
The FireWork AQ forecast system with near-real-time wildfire emissions has been run daily for
a North American domain by ECCC from 2013 to 2016 during the May–September wildfire season.
A multi-year analysis for this period showed the importance of accounting for contributions from wildfire
PM
2.5
emissions to total PM
2.5
surface concentrations (denoted here as fire-PM
2.5
) during the wildfire
season. For both Canada and the USA, the months of July and August usually showed the maximum
fire-PM
2.5
, although intense wildfires can also occur in September in the western USA, likely due to a
longer summer season [51].
Monthly and seasonal analyses of the mean forecasted fire-PM
2.5
suggested that, on average, over 76%
of Canadians and 69% of Americans were at least minimally affected by wildfire smoke during the
four-year study period. Comparison of average monthly fire-PM
2.5
showed large year-to-year variations in
both timing and spatial locations of wildfires between 2013 and 2016. Wildfire impacts are often driven
by a few major wildfire events that can lead to poor air quality for several consecutive weeks near the
emission sources and beyond. In August 2015 approximately 3 million Canadians and 3 million Americans
were exposed to mean monthly fire-PM
2.5
>10μg/m
3
.
Calculations of the number of days and hours with forecasted fire-PM
2.5
above various concentration
thresholds ranging from 1
μ
g/m
3
to 28
μ
g/m
3
for 2013–2016 showed that most wildfire events over
North America occurred in the western part of the USA and in western, northern, and central Canada.
During months of extreme wildfire activity, some areas in northwestern Canada and the western USA had
up to 20% of days where the fire-PM
2.5
was > 28
μ
g/m
3
. The eastern USA and eastern Canada had fewer
days with threshold exceedances, but most of North America was affected by fire-PM
2.5
>1
μ
g/m
3
on at
least one day per year.
FireWork is a valuable prognostic tool used as guidance by AQ meteorologists to issue forecasts on
a daily basis, allowing advance warnings to populations at risk to reduce their exposure. In addition,
this study has shown that FireWork is also useful for retrospective analysis of past wildfire events.
The statistical analyses of these forecasts over multiple years can be used by public health researchers,
AQ regulators and policymakers, and others interested in wildfire impacts to understand and
characterize exposure to wildfire smoke and its interannual and geographic variability.
Supplementary Materials: The following are available online at www.mdpi.com/2073-4433/8/9/179/s1.
Acknowledgments:
The ECCC FireWork team is very grateful to the Natural Resources Canada team behind
the Canadian Wildland Information System (CWFIS), who provided crucial assistance in the development of
the FireWork system. We are particularly grateful to Kerry Anderson and Peter Englefield, who provided useful
analyses and guidance on the CWFIS system. We also thank the U.S. Forest Service BlueSky team, particularly
Susan O’Neill, for analyses and advice on wildfire emission estimates for the USA. We appreciate the assistance
of Alexandru Lupu and Cassandra Bolduc for preparing tables presented in this article. Finally, we would like
to acknowledge three anonymous reviewers for their valuable and constructive comments, which considerably
improved this paper.
110
Books
MDPI
Atmosphere 2017, 8, 179
Author Contributions:
Rodrigo Munoz-Alpizar and Radenko Pavlovic conceived the study; Rodrigo Munoz-Alpizar,
Radenko Pavlovic and Sylvain Ménard backfilled missing FireWork forecasts over 2013–2015 five-month periods;
Hugo Landry, Samuel Gilbert and Paul-André Beaulieu developed analysis tools; Jacinthe Racine and Annie Duhamel
contributed to the results presentation; Rodrigo Munoz-Alpizar, Radenko Pavlovic, Michael D. Moran, Sarah B.
Henderson, Jack Chen, Sylvie Gravel and Sylvain Ménard wrote the paper; and Didier Davignon, Sophie Cousineau,
and Véronique Bouchet contributed with organizational support and scientific suggestions.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Beverly, J.L.; Bothwell, P. Wildfire evacuations in Canada 1980–2007. Nat. Hazards
2011
, 59, 571–596.
[CrossRef]
2.
Canadian Interagency Forest Fire Centre. Canada Report 2016. Available online: http://www.ciffc.ca/
images/stories/pdf/2016_canada_report_2017_05_15_v02.pdf (accessed on 15 May 2017).
3.
Natural Resources Canada. Cost of Fire Protection. Available online: http://www.nrcan.gc.ca/forests/
climate-change/forest-change/17783 (accessed on 15 May 2017).
4.
Flannigan, M.; Tymstra, C. The Fort McMurray Wildfire: By the Numbers. Canadian Wildland Fire &
Smoke Newsletter. 2016; pp. 1–2. Available online: https://sites.ualberta.ca/~wcwfs/CWFSN/newsletters/
CWFSN_Fall_2016.pdf (accessed on 15 September 2017).
5.
Urbanski, P.U.; Hao, W.M.; Baker, S. Chemical composition of wildland fire emissions. Dev. Environ. Sci.
2009, 8, 79–107.
6.
Reid, C.E.; Brauer, M.; Johnston, F.H.; Jerrett, M.; Balmes, J.R.; Elliott, C.T. Critical Review of Health Impacts
of Wildfire Smoke Exposure. Environ. Health Perspect. 2016, 124, 1334–1342. [CrossRef] [PubMed]
7.
Liu, J.C.; Pereira, G.; Uhl, S.A.; Bravo, M.A.; Bell, M.L. A systematic review of the physical health impacts
from non-occupational exposure to wildfire smoke. Environ. Res. 2015, 136, 120–132. [CrossRef] [PubMed]
8.
Liu, J.C.; Mickley, L.J.; Sulprizio, M.P.; Dominici, F.; Yue, X.; Ebisu, K.; Anderson, G.B.; Khan, R.F.; Bravo, M.A.;
Bell, M.L. Particulate air pollution from wildfires in the Western US under climate change. Clim. Chang.
2016
,
138, 655–666. [CrossRef] [PubMed]
9.
Thurston, G.D.; Kipen, H.; Annesi-Maesano, I.; Balmes, J.; Brook, R.D.; Cromar, K.; de Matteis, S.;
Forastiere, F.; Forsberg, B.; Frampton, M.W.; et al. A joint ERS/ATS policy statement: What constitutes an
adverse health effect of air pollution? An analytical framework. Eur. Respir. J.
2017
, 49, 1600419. [CrossRef]
[PubMed]
10.
Canadian Council of Ministers of the Environment (CCME). Guidance Document on Achievement
Determination Canadian Ambient Air Quality Standards for Fine Particulate Matter and Ozone. 2012; p. 43.
Available online: http://www.ccme.ca/files/Resources/air/aqms/pn_1483_gdad_eng.pdf (accessed on
15 May 2017).
11.
Sapkota, A.; Symons, J.; Kleissl, J.; Wang, L.; Parlange, M.; Ondov, J.; Breysse, P.N.; Diette, G.B.;
Eggleston, P.A.; Buckley, T.J. Impact of the 2002 Canadian forest fires on particulate matter air quality
in Baltimore city. Environ. Sci. Technol. 2005, 39, 24–32. [CrossRef] [PubMed]
12.
Forster, C.; Wandinger, U.; Wotawa, G.; James, P.; Mattis, I.; Althausen, D.; Simmonds, P.; O’Doherty, S.;
Jennings, I.M.; Kleefeld, C.; et al. Transport of boreal forest fire emissions from Canada to Europe.
J. Geophys. Res. 2001, 106, 22887–22906. [CrossRef]
13.
Adam, M.; Pahlow, M.; Kovalev, V.A.; Ondov, J.M.; Parlange, M.B.; Nair, N. Aerosol optical characterization
by nephelometer and lidar: The Baltimore Supersite experiment during the Canadian forest fire smoke
intrusion. J. Geophys. Res. 2004, 109, D16. [CrossRef]
14.
Colarco, P.R.; Schoeberl, M.R.; Doddridge, B.G.; Marufu, L.T.; Torres, O.; Welton, E.J. Transport of smoke
from Canadian forest fires to the surface near Washington, DC: Injection height, entrainment, and optical
properties. J. Geophys. Res. 2004, 109, 1–12. [CrossRef]
15.
Spichtinger, N.; Wenig, M.; James, P.; Wagner, T.; Platt, U.; Stohl, A. Satellite detection of a continental-scale
plume of nitrogen oxides from boreal forest fires. Geophys. Res. Lett. 2001, 28, 4579–4582. [CrossRef]
16.
Morris, G.A.; Hershey, S.; Thompson, A.M.; Stohl, A.; Colarco, P.R.; McMillan, W.W.; Warner, J.; Johnson, B.J.;
Witte, J.C.; Kucsera, T.L.; et al. Alaskan and Canadian forest fires exacerbate ozone pollution in Houston,
Texas, on 19 and 20 July. J. Geophys. Res. 2004, 111, 1–10. [CrossRef]
111
Books
MDPI
Atmosphere 2017, 8, 179
17.
DeBell, L.J.; Talbot, R.W.; Dibb, J.E.; Munger, J.W.; Fischer, E.V.; Frolking, S.E. A major regional air pollution
event in the northeastern United States caused by extensive forest fires in Quebec, Canada. J. Geophys. Res.
2004, 109, 1–16. [CrossRef]
18.
Fiore, A.M.; Pierce, R.B.; Dickerson, R.R.; Lin, M.; Bradley, R. Detecting and attributing episodic high
background ozone events. J. Air Waste Manag. Assoc. 2014, 64, 22–28.
19.
Dreessen, J.; Sullivan, J.; Delgado, R. Observations and impacts of transported Canadian wildfire smoke
on ozone and aerosol air quality in the Maryland region on June 9–12. J. Air Waste Manag. Assoc.
2016
, 66,
842–862. [CrossRef] [PubMed]
20.
Cottle, P.; Strawbridge, K.; McKendry, I. Long-range transport of Siberian wildfire smoke to British Columbia:
Lidar observations and air quality impacts. Atmos. Environ. 2014, 90, 71–77. [CrossRef]
21.
Teakles, A.D.; So, R.; Ainslie, B.; Nissen, R.; Schiller, C.; Vingarzan, R.; McKendry, I.; Macdonald, A.M.;
Jaffe, D.A.; Bertram, A.K.; et al. Impacts of the July 2012 Siberian fire plume on air quality in the Pacific
Northwest. Atmos. Chem. Phys. 2017, 17, 2593–2611. [CrossRef]
22.
Centre for Disease Control. Evidence Review: Exposure Measures for Wildfire Smoke Surveillance. 31 March
2014. Available online: http://www.bccdc.ca/resource-gallery/Documents/Guidelines%20and%20Forms/
Guidelines%20and%20Manuals/Health-Environment/WFSG_EvidenceReview_Smokesurveillance_
FINAL_v2_edstrs.pdf (accessed on 15 May 2017).
23.
Rappold, A.G.; Reyes, J.; Pouliot, G.; Cascio, W.E.; Diaz-Sanchez, D. Community Vulnerability to Health
Impacts of Wildland Fire Smoke Exposure. Environ. Sci. Technol. 2017, 51, 6674–6682. [CrossRef] [PubMed]
24.
Zhang, Y.; Bocquet, M.; Mallet, V.; Seigneur, C.; Baklanov, A. Real-time air quality forecasting, part I: History,
techniques, and current status. Atmos. Environ. 2012, 60, 632–655. [CrossRef]
25.
Zhang, Y.; Bocquet, M.; Mallet, V.; Seigneur, C.; Baklanov, A. Real-time air quality forecasting, part II: State of
the science, current research needs, and future prospects. Atmos. Environ. 2012, 60, 656–676. [CrossRef]
26.
Larkin, N.K.; O’Neill, S.M.; Solomon, R.; Raffuse, S.; Strand, T.; Sullivan, D.C.; Krull, C.; Rorig, M.; Peterson, J.;
Ferguson, S.A. The BlueSky smoke modeling framework. Int. J. Wildland Fire 2010, 18, 906–920. [CrossRef]
27.
Pavlovic, R.; Chen, J.; Anderson, K.; Moran, M.D.; Beaulieu, P.-A.; Davignon, D.; Cousineau, S. The FireWork
air quality forecast system with near-real-time biomass burning emissions: Recent developments and
evaluation of performance for the 2015 North American wildfire season. J. Air Waste Manage. Assoc.
2016
, 66,
819–841. [CrossRef] [PubMed]
28.
Pavlovic, R.; Chen, J.; Davignon, D.; Moran, M.D.; Beaulieu, P.A.; Landry, H.; Sassi, M.; Gilbert, S.;
Munoz-Alpizar, R.; Anderson, K.; et al. FireWork—A Canadian Operational Air Quality Forecast Model
With Near-Real-Time Biomass Burning Emissions. Canadian Wildland Fire & Smoke Newsletter. 2016;
pp. 18–29. Available online: https://sites.ualberta.ca/~wcwfs/CWFSN/newsletters/CWFSN_Fall_2016.pdf
(accessed on 15 September 2017).
29.
Yuchi, W.; Yao, J.; McLean, K.E.; Stull, R.; Pavlovic, R.; Davignon, D.; Moran, M.D.; Henderson, S.B. Blending
forest fire smoke forecasts with observed data can improve their utility for public health applications.
Atmos. Environ. 2016, 145, 308–317. [CrossRef]
30.
Moran, M.D.; Ménard, S.; Pavlovic, R.; Anselmo, D.; Antonopoulos, S.; Makar, P.A.; Gong, W.; Stroud, C.;
Zhang, J.; Zheng, Q.; et al. Recent advances in Canada’s National Operational AQ Forecasting System.
In Proceedings of the 32nd NATO/SPS ITM on Air Pollution Modelling and Its Application, Utrecht,
The Netherlands, 7–11 May 2012; pp. 215–220.
31.
Moran, M.; Zheng, Q.; Zhang, J.; Pavlovic, R. RAQDPS Version 013: Upgrades to the CMC
Operational Regional Air Quality Deterministic Prediction System Released in June 2015. 2015; p. 57.
Available online: http://collaboration.cmc.ec.gc.ca/cmc/cmoi/product_guide/docs/lib/op_systems/
doc_opchanges/Technical_Note_GEM-MACH10_v1.5.3+SET2.1.1_Emissions_9Nov2015.pdf (accessed on
15 May 2017).
32.
Im, U.; Bianconi, R.; Solazzo, E.; Kioutsioukis, I.; Badia, A.; Balzarini, A.; Baro, R.; Bellasio, R.; Brunner, D.;
Chemel, C.; et al. Evaluation of operational online-coupled regional air quality models over Europe and
North America in the context of AQMEII phase 2. Part I: Ozone. Atmos. Environ.
2015
, 115, 404–420.
[CrossRef]
33.
Lee, B.S.; Alexander, M.E.; Hawkes, B.C.; Lynham, T.J.; Stocks, B.J.; Englefield, P. Information systems in
support of wildland fire management decision making in Canada. Comput. Electron. Agric.
2002
, 37, 185–198.
[CrossRef]
112
Books
MDPI
Atmosphere 2017, 8, 179
34.
Anderson, K.R.; Englefield, P.; Little, J.M.; Reuter, G. An approach to operational forest fire growth predictions
for Canada. Int. J. Wildland Fire 2009, 18, 893–905. [CrossRef]
35.
Moran, M.; Pavlovic, R.; Chen, J. FireWork 2016: Release of the Initial Operational Version of the CMC
Regional Air Quality Deterministic Prediction System with Near-Real-Time Satellite-Derived Wildfire
Emissions. 2016; p. 4. Available online: http://collaboration.cmc.ec.gc.ca/cmc/CMOI/product_guide/
docs/tech_notes/technote_raqdps015fw_20160428_e.pdf (accessed on 15 May 2017).
36.
Kaiser, J.W.; Heil, A.; Andreae, M.O.; Benedetti, A.; Chubarova, N.; Jones, L.; Morcrette, J.-J.; Razinger, M.;
Schultz, M.G.; Suttie, M.; et al. Biomass burning emissions estimated with a global fire assimilation system
based on observed fire radiative power. Biogeosciences 2012, 9, 527–554. [CrossRef]
37.
Xing, J.; Pleim, J.; Mathur, R.; Pouliot, G.; Hogrefe Gan, C.-M.; Wei, C. Historical gaseous and primary aerosol
emissions in the United States from 1990–2010. Atmos. Chem. Phys. 2013, 13, 7531–7549. [CrossRef]
38.
EPA. Draft Guidance for Comment: Significant Impact Levels for Ozone and Fine Particle in the Prevention
of Significant Deterioration Permitting Program. 2016; p. 13. Available online: https://www.epa.
gov/sites/production/files/2016-08/documents/pm2_5_sils_and_ozone_draft_guidance.pdf (accessed on
15 May 2017).
39.
Statistics Canada. Dissemination Areas, Census 2016. Available online: http://www12.statcan.gc.ca/census-
recensement/2016/dp-pd/hlt-fst/pd-pl/Comprehensive.cfm (accessed on 15 May 2017).
40.
Statistics Canada. Boundary Files, Census 2016. Available online: http://www12.statcan.gc.ca/census-
recensement/2011/geo/bound-limit/bound-limit-eng.cfm (accessed on 15 May 2017).
41.
United States Census Bureau. Census 2010. Available online: https://www.census.gov/geo/maps-data/
data/tiger-data.html (accessed on 15 May 2017).
42.
National Intergancy Fire Center. Available online: https://www.nifc.gov/fireInfo/fireInfo_stats_totalFires.
html (accessed on 15 May 2017).
43.
Jaffe, D.; Hafner, W.; Chand, D.; Westerling, A.; Spracklen, D. Interannual variations in PM
2.5
due to wildfires
in the Western United States. Environ. Sci. Technol. 2008, 42, 2812–2818. [CrossRef] [PubMed]
44.
Mao, Y.H.; Li, Q.B.; Zhang, L.; Chen, Y.; Randerson, J.T.; Chen, D.; Liou, K.N. Biomass burning contribution
to black carbon in the Western United States Mountain Ranges. Atmos. Chem. Phys.
2011
, 11, 11253–11266.
[CrossRef]
45.
Spracklen, D.V.; Logan, J.A.; Mickley, L.J.; Park, R.J.; Yevich, R.; Westerling, A.L.; Jaffe, D.A. Wildfires drive
interannual variability of organic carbon aerosol in the western U.S. in summer. Geophys. Res. Lett.
2007
, 34,
L16816. [CrossRef]
46. Zeng, T.; Wang, Y. Nationwide summer peaks of OC/EC ratios in the contiguous United States. Atmos. Environ. 2011,
45, 578–586. [CrossRef]
47.
Henderson, S.B.; Johnston, F.H. Measures of forest fire smoke exposure and their associations with respiratory
health outcomes. Curr. Opin. Allergy Clin. Immunol. 2012, 12, 221–227. [CrossRef] [PubMed]
48.
Dennekamp, M.; Straney, L.D.; Erbas, B.; Abramson, M.J.; Keywood, M.; Smith, K.; Sim, M.R.; Glass, D.C.;
Monaco, A.D.; Haikerwal, A.; et al. Forest Fire Smoke Exposures and Out-of-Hospital Cardiac Arrests in
Melbourne, Australia: A Case-Crossover Study. Environ. Health Perspect.
2015
, 123, 959–964. [CrossRef]
[PubMed]
49.
Johnston, F.H.; Henderson, S.B.; Chen, Y.; Randerson, J.T.; Marlier, M.; DeFries, R.S.; Kinney, P.;
Bowman, D.M.; Brauer, M. Estimated Global Mortality Attributable to Smoke from Landscape Fires.
Environ. Health Perspect. 2012, 120, 695–701. [CrossRef] [PubMed]
50.
Robichaud, A.; Ménard, R.; Zaïtseva, Y.; Anselmo, D. Multi-pollutant surface objective analyses and mapping
of air quality health index over North America. Air Qual. Atmos. Health
2016
, 9, 743–759. [CrossRef] [PubMed]
51.
Jolly, W.M.; Cochrane, M.A.; Freeborn, P.H.; Holden, Z.A.; Brown, T.J.; Williamson, G.J.; Bowman, D.M.
Climate-induced variations in global wildfire danger from 1979 to 2013. Nat. Commun.
2015
, 6, 7537.
[CrossRef] [PubMed]
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
113
Books
MDPI
atmosphere
Review
Interpreting Mobile and Handheld Air Sensor
Readings in Relation to Air Quality Standards and
Health Effect Reference Values: Tackling
the Challenges
George M. Woodall
1,
*, Mark D. Hoover
2
, Ronald Williams
1
, Kristen Benedict
1
,
Martin Harper
2,†
, Jhy-Charm Soo
2
, Annie M. Jarabek
1
, Michael J. Stewart
1
, James S. Brown
1
,
Janis E. Hulla
3
, Motria Caudill
4
, Andrea L. Clements
1
, Amanda Kaufman
1
, Alison J. Parker
5
,
Martha Keating
1
, David Balshaw
6
, Kevin Garrahan
7
, Laureen Burton
7
, Sheila Batka
8
,
Vijay S. Limaye
8
, Pertti J. Hakkinen
9
and Bob Thompson
1
1
Environmental Protection Agency, Research Triangle Park, NC 27711, USA; williams.ronald@epa.gov (R.W.);
benedict.kristen@epa.gov (K.B.); jarabek.annie@epa.gov (A.M.J.); stewart.michael@epa.gov (M.J.S.);
brown.james@epa.gov (J.S.B.); clements.andrea@epa.gov (A.L.C.); kaufman.amanda@epa.gov (A.K.);
keating.martha@epa.gov (M.K.); thompson.bob@epa.gov (B.T.)
2
National Institute for Occupational Safety and Health, Morgantown, WV 26505, USA;
mhoover1@cdc.gov (M.D.H.); mharper@zefon.com (M.H.); jsoo@cdc.gov (J.-C.S.)
3
Army Corps of Engineers, Sacramento, CA 95814, USA; janis.e.hulla@usace.army.mil
4
Agency for Toxic Substances and Disease Registry, Atlanta, GA 30329, USA; caudill.motria@epa.gov
5
ORISE Fellow hosted by U.S. Environmental Protection Agency, Washington, DC 20004, USA;
parker.alison@epa.gov
6
National Institute for Environmental Health Sciences, Research Triangle Park, NC 27709, USA;
balshaw@niehs.nih.gov
7
Environmental Protection Agency, Washington, DC 20004, USA; garrahan.kevin@epa.gov (K.G.);
burton.laureen@epa.gov (L.B.)
8
Environmental Protection Agency, Chicago, IL 60605, USA; batka.sheila@epa.gov (S.B.);
limaye.vijay@epa.gov (V.S.L.)
9
National Library of Medicine, Bethesda, MD 20892, USA; hakkinenp@mail.nlm.nih.gov
* Correspondence: woodall.george@epa.gov; Tel.: +1-919-541-3896
† Retired. Current affiliation with Zefon International, Inc., Ocala, FL 34474, USA.
Received: 17 August 2017; Accepted: 14 September 2017; Published: 21 September 2017
Abstract:
The US Environmental Protection Agency (EPA) and other federal agencies face a number of
challenges in interpreting and reconciling short-duration (seconds to minutes) readings from mobile
and handheld air sensors with the longer duration averages (hours to days) associated with the
National Ambient Air Quality Standards (NAAQS) for the criteria pollutants-particulate matter (PM),
ozone, carbon monoxide, lead, nitrogen oxides, and sulfur oxides. Similar issues are equally relevant
to the hazardous air pollutants (HAPs) where chemical-specific health effect reference values are the
best indicators of exposure limits; values which are often based on a lifetime of continuous exposure.
A multi-agency, staff-level Air Sensors Health Group (ASHG) was convened in 2013. ASHG represents
a multi-institutional collaboration of Federal agencies devoted to discovery and discussion of sensor
technologies, interpretation of sensor data, defining the state of sensor-related science across each
institution, and provides consultation on how sensors might effectively be used to meet a wide
range of research and decision support needs. ASHG focuses on several fronts: improving the
understanding of what hand-held sensor technologies may be able to deliver; communicating what
hand-held sensor readings can provide to a number of audiences; the challenges of how to integrate
data generated by multiple entities using new and unproven technologies; and defining best practices
in communicating health-related messages to various audiences. This review summarizes the
challenges, successes, and promising tools of those initial ASHG efforts and Federal agency progress
Atmosphere 2017, 8, 182 114 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2017, 8, 182
on crafting similar products for use with other NAAQS pollutants and the HAPs. NOTE: The opinions
expressed are those of the authors and do not necessary represent the opinions of their Federal
Agencies or the US Government. Mention of product names does not constitute endorsement.
Keywords: air pollutants; ambient air; indoor air; citizen science; toxic chemicals
1. Introduction
The Air Sensors 2013: Data Quality and Applications workshop, held in Research Triangle Park, N.C.,
highlighted the substantial advances in the development of portable air sensors capable of providing
real-time measurements of ambient air pollution [
1
]. One anticipated benefit for the use of air sensors
is the potential to expand upon the already well-established network of air quality monitors for key
air pollutants. Sensor manufacturers introduced a number of new, relatively low-cost, portable air
sensors capable of continuously measuring ambient levels of ozone (O
3
) and nitrogen dioxide (NO
2
).
In addition, considerable progress was described regarding the development of air sensors capable of
measuring ambient concentrations of particulate matter (PM), and total volatile organic compounds
(VOCs). Since that 2013 workshop, the field of portable air quality sensing technology has continued
to evolve at a rapid pace, with commercially available products now available for O
3
,NO
2
, PM, VOCs,
as well as for other pollutants, both alone and in complex mixtures. While the potential for these
technologies continues to be great, significant challenges in their application remain. Most notably,
there is a wide range of data quality differences between different sensor manufactures and models [
2
],
and there is a significant question about how air quality data collected on timescales as short as 1-min
should be interpreted in comparison to the available health reference values for air pollutants which are
typically based on exposure durations of several hours to many years [
3
]. For example, commercially
available portable air sensors for O
3
can provide the user with minute-by-minute O
3
concentrations;
however, interpreting these very short-term measurements with respect to potential adverse health
effects is difficult. The health-based National Ambient Air Quality Standard (NAAQS) for O
3
is
based on an 8-h average concentration which is backed by thousands of studies from numerous
independent researchers and is established through a rigorous process to be scientifically defensible
and legally enforceable; no such standards have been developed for these shorter duration exposures.
Similar challenges exist when applying Occupational Safety and Health Administration (OSHA)
standards. The most comparable OSHA exposure standards include the Permissible Exposure Limit
(PEL), which is an 8-h time-weighted average, and the Short Term Exposure Limit (STEL), which is a
15-min time-weighted average.
Detection and monitoring of most non-NAAQS environmental chemicals (including the hazardous
air pollutants), toxins and pathogens still largely involves identifying each individual agent,
which often requires sending samples to a remote analytical laboratory for analyses. Delivery of
laboratory results may take days, weeks or even months. Although the deployment of portable
direct-reading instruments, such as photon ionization detectors (PIDs) for total VOCs, can provide
some real-time information, these screening instruments lack the sensitivity or selectivity delivered
by analytical laboratories and thus, are unable to fully inform a user with a critical need for
high-specificity (e.g., rapid response decision-makers). New technologies such as miniaturized light
emitting diodes, ultra-violet light detectors and functionalized graphene resistors have enabled the
development of chemical detectors capable of delivering laboratory-quality analyses in near real-time.
Coupling these types of sensors with global positioning and cell phone technologies may enable
the detection, quantification, and visual monitoring of environmental contamination in real-time
from remote locations. These promising technologies are expensive, however, and will likely remain
out of reach for all but the most dedicated citizen scientist. As described in the following sections,
understanding and advancing these types of applications has been a focus of the ASHG: beginning
115
Books
MDPI
Atmosphere 2017, 8, 182
with defining the relevance of air sensors; facing the challenges of communicating across diverse key
audiences; dealing with data validity issues; and moving toward a future with more reliable and useful
sensor readings.
Formation of the Air Sensors Health Group (ASHG)
Recognizing the potential widespread use of portable air sensors, the likely data interpretation
challenges these sensors would present to state and local governments, and the opportunity for
collaboration across Federal agencies, the ASHG was formed in 2013. The ASHG is a multi-institutional
collaboration of Federal agencies devoted to keeping abreast of new sensor technologies and to assist in
the proper interpretation of sensor data as potential indicators of air quality. ASHG consists of experts
in a number of areas, including toxicology, public health, engineering, monitoring and sampling,
ambient air, indoor air, and occupational health, to name a few.
The ASHG monitors the state of the science at each institution and aims to find common ground
on how sensors might effectively be used to meet a wide range of research needs in the occupational,
indoor air, and ambient air settings. The ASHG also aims to be a resource for state, regional, and tribal
organizations, as well as for citizen scientists and community members considering the use of portable
air sensors for air pollution research and decision-making. To date, the ASHG has focused on multiple
fronts: assisting EPA Program Offices in developing tools and message statements regarding the
potential for adverse health effects from the short-duration air sensor readings for PM and O
3
;
providing analysis comparing short-term readings to longer-duration averages from existing official
monitoring stations used in determining compliance to the NAAQS; and developing prototype visual
tools to assist in communicating appropriate interpretive messages. These ASHG contributions have
been incorporated into projects and programmatic products which are discussed in later sections
of this paper. This review summarizes the challenges, and successes of those initial ASHG efforts,
and Federal agency progress on crafting similar products for use with other pollutants.
2. Challenges
2.1. Relevance of Sensor Measurements
A critical concern for interpreting readings from air quality sensors is to have a realistic
understanding of what the sensors are actually measuring; how that relates to the expectations
of users (including citizen scientists, researchers, regulatory agencies, etc.) for sensitivity, specificity,
and robustness of the intended application; and the extent to which those measurements of exposure
can appropriately be used to communicate potential hazard and manage risks to public health.
The magnitude of this concern varies greatly across the types of pollutants purportedly being measured
and is related to a number of factors, such as the chemical and physical nature of the pollutant,
the reactivity of the pollutant, concentration and compositional changes over time and location,
and the influences with various atmospheric conditions. EPA guidance on achieving high-quality
data through systematic planning using the data quality objectives process can be found online [
4
].
Standards for data quality can differ slightly depending on the context; additional standards are
discussed as those contexts are examined in the ensuing sections of this review. The ASHG recognized
early in their discussions that this ideal for collecting high-quality data may not be readily obtainable
for all potential users of air pollutant sensors.
Low-cost air sensors on the market to date are predominantly of three types; optical,
electrochemical, or metal oxide. Generally, these sensor types are not as specific and do not incorporate
the front end conditioning or selection that is present in Federal Reference Method (FRM) or Federal
Equivalent Method (FEM) measurement devices. Therefore, low-cost sensor measurements may suffer
from the influence of co-responsive pollutants, environmental conditions, and even sensor component
production variations. The need to understand what the sensor is measuring is of prime importance.
In order to better understand the measurement and potential confounding influences, an informed
116
Books
MDPI
Atmosphere 2017, 8, 182
user must understand the physical and chemical nature of a pollutant of interest and how it responds
to the environment as well as the sensor measurement technology itself. Field testing may help
a user understand the influence of confounding factors and could identify co-responsive pollutants.
The interpretation and use of low-cost air sensor data will be enhanced by simultaneous collection
of co-responsive pollutant concentrations and environmental data including temperature, relative
humidity, wind, and weather (i.e., precipitation, fog), as well as observations about local pollution
sources. In this context, chemical sensors must be validated for chemical specificity and sensitivity
under the environmental conditions expected in the field prior to any consideration for demonstrating
regulatory compliance.
2.1.1. Interpreting Sensor Readings
Managing expectations about what sensor measurement data can and cannot be used for is
paramount in communicating with both citizen scientists and the general public at large, who for a
variety of reasons (e.g., personal health, general air quality interest) may, in the future, routinely consult
a sensor. In anticipation of this interest on the part of the public, the U.S. EPA began a pilot effort in
June 2016, to begin addressing the interpretation of short-term sensor readings in the context of air
quality [
5
]. There are several challenges to interpreting these data. Among these challenges are sensor
performance and the short-term, sometimes instantaneous output from a sensor. Short-term sensor
data come from instruments of unknown performance quality and, importantly, these short-term
concentrations cannot be compared to the NAAQS to draw conclusions about what these nearly
instantaneous exposures may mean in terms of health impacts. The NAAQS are based on longer
exposure durations (e.g., 8-h or 24-h averages) consistent with the health evidence from the reviews
of these standards. This health evidence does not support linking 1-min (or shorter) ozone or PM
2.5
concentrations to adverse health effects, thus a 1-min sensor reading is not directly comparable to the
NAAQS, or to the related Air Quality Index (AQI) categories.
To help the public understand the implications of these readings, scientists at EPA have piloted
a color-coded Sensor Scale that might be used in conjunction with the AQI to help the public better
understand what their short-term sensor data mean in the context of local and regional air quality and
consequently make behavioral decisions about outdoor activities. Statistical approaches were used
to understand the relationship between short-term ozone and fine particulate matter measurements
with longer term averages. The Sensor Scales and explanatory background materials are housed in
the Air Sensor Toolbox on the EPA website [
6
] and have also been described elsewhere. To test the
effectiveness of these messages and their visual presentation, EPA conducted focus groups in 2015 and
2017, before and after the deployment of the pilot respectively. Analysis of the outcomes of these focus
groups is underway. After considering the input from the focus groups, EPA will refine the messages
as appropriate and consider outreach to air quality sensor developers on these focus group findings.
Additional work is also on-going to develop similar message schemes for selected HAPs.
In addition to the NAAQS, EPA’s Integrated Risk Information System (IRIS) provides inhalation
reference concentration (RfC) values for HAPs and other key pollutants important to many EPA
Programs with a focus on chronic exposure durations (from years to a lifetime). The Agency for Toxic
Substances and Disease Registry (ATSDR), a federal public health agency of the U.S. Department of
Health and Human Services, develops similar substance-specific minimal risk level (MRL) values for
acute (1–14 days), intermediate (15–364 days), and chronic (365 days and longer) exposure durations [
7
].
ATSDR’s work is focused on Superfund sites, but environmental health specialists apply the MRLs in
a wide range of investigations. Many state agencies also develop inhalation health effect reference
values in support of their programmatic needs. More generically, the term reference value is used in
this text to include all of the various values referred to as standards, guideline values, toxicity values,
health benchmarks, etc., and includes values developed for use in emergency response, occupational
exposure monitoring, and those protective of the general public. Additional reports are available for a
117
Books
MDPI
Atmosphere 2017, 8, 182
more complete comparison of the available systems of health effect reference values [
3
,
8
], and between
values for specific pollutants [3,9–13].
2.1.2. Occupational Versus Environmental Exposures
One issue likely to be resolved on a case-by-case basis is the overlap between the environment and
the workplace. An employee may be exposed to a chemical in the workplace at a higher concentration
than in an outdoor environment. In addition, because the worker is assumed to be fit and healthy
and to have periods of recovery from exposure, the levels of concentration deemed tolerable are also
higher. A citizen wearing a sensor outdoors, may well carry it into their workplace, and could then
experience problems reconciling the levels of pollutants considered acceptable in these different milieus.
For example, a sensor designed to help citizens avoid pollution from motor vehicles might measure
carbon monoxide. EPA sets a primary NAAQS of 9 ppm averaged over an 8-h period and 35 ppm
averaged over a 1-h period, not to be exceeded more than once per year. In the workplace, the situation
is not quite as straightforward, but all the limit values which might be applied are higher than those
under the NAAQS. The OSHA Permissible Exposure Limit is 50 ppm, while the National Institute for
Occupational Safety and Health (NIOSH) Recommended Exposure Limit is 35 ppm, and the American
Conference of Governmental Hygienists (ACGIH) Threshold Limit Value is 25 ppm, all averaged over
an 8-h period, as with the 9 ppm NAAQS standard. If the monitor alarm is set for the 8-h average
NAAQS value, or possibly even at the one-hour average value, the alarm might easily be triggered in
the workplace, which might be, for example, a bus garage or foundry. The ensuing discussion with
the workplace safety manager or the employer regarding the acceptability of the exposure situation
might be difficult for both parties in the absence of well-thought out responses. Nevertheless, in a
holistic vision of the exposome, which should consider all exposures over all life-stages, it may not be
appropriate to regard ambient, indoor and workplace exposures as somehow “different”, to be always
measured and assessed separately [
14
], and so it is to be hoped that it will not be necessary to turn off
ambient continuous air monitors “at the factory gates”.
2.1.3. Global/International Perspectives
Low cost air quality sensors are of interest worldwide. Multiple studies have distributed sensors in
an effort to apply them to local environmental research [
15
–
18
]. One of the most notable efforts involving
multiple countries and metropolitan areas was the Citi-Sense project [
19
,
20
]. This diverse research
project involved technology developers, citizen scientists, academics, and professional research
organizations using low cost sensor technologies with a purposeful intent [
21
,
22
]. The pan-European
initiative, EuNetAir, has goals of developing harmony in sensor selection and deployment strategies
as well as coordinating sensor evaluation protocols to ensure the timely integration of air quality
sensors into monitoring networks [
23
–
25
]. The Clean Air Asia consortium, a partnership of multiple
Asian-based cities, sensor enthusiasts, academics, and air quality professionals are attempting to
improve air quality and improve the overall living conditions in some of the most polluted cities in the
world [26].
Even with the apparent world-wide enthusiastic use of low cost sensors to inform public
awareness of environmental conditions, there is also a call to ensure the data being collected from
these devices across the globe are accurate enough to be used in a purposeful manner [
27
]. This call
for an adequate understanding of sensor performance is not only a reasonable approach but one that
must be pursued with the same enthusiasm as those wishing to disseminate low cost sensors to a
global population.
A multitude of pseudo air quality index messaging applications are available on the internet from
sources around the world [
28
]. Some of these applications are using low cost sensors to collect air
quality data, potentially not accounting for the accuracy of the measurement or the environmental
setting in which the measurements are taking place (e.g., indoor, outdoor, near source categories).
Furthermore, use of subjective data messaging on health impacts of such sensor measurements without
118
Books
MDPI
Atmosphere 2017, 8, 182
a scientific basis has the potential of confusing the public-at-large and their understanding of sound
air quality awareness indices.
The World Health Organization (WHO) describes air pollution as a “major environmental risk
to health”. The WHO Air Quality Guidelines (AQG) provide an assessment of air pollution health
effects and recommend reference values for health-harmful pollution levels of ozone, PM
10
,PM
2.5
,SO
2
,
and NO
2
[
29
]. ATSDR has considered AQGs in the screening process of health evaluations for multiple
site investigations in recent years [
30
–
33
]. The European Commission (EC) has developed air quality
standards for the six US criteria pollutants, as well as benzene, polycyclic aromatic hydrocarbons
(PAHs), and three metals: arsenic, cadmium, and nickel.
2.2. Communicating Across Audiences
Portable sensors present a great deal of promise for identifying personal exposure to toxicants
present in the environment. However, communication between government or public health agencies,
sensor manufacturers, researchers, employers and employees, and citizen scientists is particularly
important given the varying degrees of accuracy across sensors, the complexities of environmental
exposure, and the difficulties in interpreting potential health risk based on readings from a portable
sensor. Thus, the subsections below summarize the various efforts that agencies represented on the
ASHG membership have put toward communicating with these groups.
2.2.1. Citizen Scientists and Communities
Over the last decade, members of the public have become increasingly engaged in taking
measurements of their environments and otherwise contributing to scientific research. Enabled by
the rapid pace of growth of sensor technology, many citizen scientists collect data on air quality
in their local environments, both individually and as a part of organized projects. The growth of
these technologies and a surge in enthusiasm for these approaches has pushed the boundaries of
traditional institution-driven research. Often called “citizen science”, these efforts are also referred to
as civic or community science, community-based monitoring, crowdmapping, participatory science,
open science, or crowdsourcing. In citizen science, the public participates voluntarily in the scientific
process, addressing problems in ways that may include formulating research questions, conducting
scientific experiments, collecting and analyzing data, interpreting results, making new discoveries,
developing technologies and applications, and solving complex problems [
34
]. Of particular relevance
to air quality research and air sensors, community science or community citizen science is “collaboratively
led scientific investigation and exploration to address community-defined questions, allowing for
engagement in the entirety of the scientific process. Unique in comparison to citizen science, community
science may or may not include partnerships with professional scientists, emphasizes the community’s
ownership of research and access to resulting data, and orients toward community goals and working
together in scalable networks to encourage collaborative learning and civic engagement” [35].
The growth of citizen science offers significant opportunities and challenges for federal agencies.
Citizen science increases public understanding and community and civic engagement with science and
environmental issues, especially locally. Citizen science can connect agencies to the public, provide
opportunities for working together towards common goals, support innovation, and make science more
accessible and available. It provides data that would otherwise be inaccessible, helps generate a more
comprehensive understanding of variation over space and time, and can increase our understanding
of social science and human behavior. At the same time, low cost sensor technologies and citizen
science introduce challenges to federal agencies, such as communicating risk to project participants,
increased public pressure for actions like regulations and enforcement, and communication of data
quality. In December 2016, the National Advisory Council for Environmental Policy and Technology
provided EPA with advice and recommendations for how to maximize the benefits of citizen science
and respond to the corresponding challenges. Recommendations include building technical capacity,
119
Books
MDPI
Atmosphere 2017, 8, 182
providing guidance and communicating data quality needs for different data uses, and integrating
citizen science into the full range of EPA’s work [34].
EPA is collaborating with citizen scientists to share tools and technology, conduct air monitoring
studies, and interpret sensor data. The Air Sensor Toolbox website [
6
] was created in 2014 to
provide resources and tools to citizens interested in learning more about conducting successful air
monitoring projects. Topics include how to use sensors, interpreting sensor data, information about
EPA air monitoring projects, funding resources, and local air monitoring examples across the United
States. The Air Sensor Guidebook [
36
] provides a comprehensive overview of air pollutants, sensors,
study design, data collection, and data interpretation. Other resources include sensor evaluation
reports highlighting performance of various sensors on the market, standard operating procedures for
sensors, fact sheets, blogs, and training videos from the EPA-sponsored Community Air Monitoring
Training in 2015.
Another way EPA engages communities is through collaborations. One such collaboration was
conducted in the Ironbound community in Newark, New Jersey through a Regional Applied Research
Effort (RARE) grant [
37
]. EPA scientists trained citizen scientists from the Ironbound Community
Corporation (ICC) (The Ironbound section of Newark is a multicultural, multiracial mosaic whose
population of 50,000 reflects the diversity and the challenges in urban America. ICC impacts the lives
of nearly 1000 people daily and thousands annually. The majority of ICC’s 3000 annual clients are
from very low to low income households with low literacy and English proficiency and multiple
family stressors.) to operate sensor pods developed by the Office of Research and Development (ORD).
The collaboration allowed for joint decisions regarding study design, sensor siting, and collection,
validation and interpretation of sensor readings. Following the Ironbound project, a similar RARE
grant collaborative project was performed in Ponce, Puerto Rico. EPA scientists used lessons learned
from the first project to attempt to improve study design topics such as roles and responsibilities and
data validation and interpretation. Tasks such as managing, validating, and interpreting large datasets
can be an obstacle for community groups involved in air monitoring projects.
ATSDR and NIOSH have partnered with University of Cincinnati, Georgia Tech, and University of
Texas Arlington to pilot the use of real-time sensors as sentinels that trigger sample collection via more
conventional devices. For example, a low-budget hydrogen sulfide detector can be programmed such
that, once a certain threshold is exceeded, a VOC canister will be filled and then sent for laboratory
analysis. ATSDR is developing these projects either to characterize peak exposures or as the first stage
in assessing the need for a full-blown traditional air monitoring study.
Sensor collocation is an important step to perform before embarking on any monitoring project,
regardless of who is conducting the monitoring. This process involves siting sensors in line with
regulatory monitors (the gold standard) for a period of time in order to compare the two datasets and
establish a regression equation to normalize sensor data and make it more accurate. EPA scientists
recently created tools designed to assist citizens in this process, including an easy-to-use Excel
™
macro
that allows one to compare two datasets, such as sensor and reference data, and with one click provides
a regression equation, comparative data graphs, and descriptive statistics. A training document that
explains collocation and why it is important, and how to perform a proper collocation, accompanies
the macro. EPA is partnering with the Clean Air Carolina community group and the Eastern Band
of Cherokee Indians to conduct their own collocation projects, pilot test these resources and provide
input on how to improve them for use by the general public.
2.2.2. State, Local and Tribal Agencies
State, tribal and local health and regulatory agencies were identified as a primary target audience
early in discussions within the ASHG. The more proximately located agencies and offices are more
likely to be called by the public and press, and may be more resource-challenged to interpret data
generated using sensors. The EPA Regional Offices were identified to be the most likely first contact
point for EPA. Similarly, ATSDR has Regional Offices that are another resource for health agencies.
120
Books
MDPI
Atmosphere 2017, 8, 182
Health agencies at all levels of government are called upon by the public and press to interpret air
monitoring data. City and county health departments, depending on their size and community needs,
may have environmental health specialists on staff who are familiar with air monitoring methods.
Frequently, local agencies will defer to their respective state health department, many of which have
specialists that are supported by the ATSDR Cooperative Agreement Program. State or local agencies
may in turn also contact ATSDR or its sister agency, the National Center for Environmental Health
(NCEH), which is part of the Centers for Disease Control and Prevention (CDC). As noted earlier
in Section 2.1.1, there are several reference value systems which have been applied in interpreting
exposures to non-criteria air pollutants, including the RfC and MRL values. Health agencies also
frequently make site-specific exposure dose calculations and may evaluate health risks for shorter
averaging periods by adapting the dose-response information in toxicology studies on which the
reference values are based.
In order to address these needs, EPA Regional Offices have initiated outreach to state, tribal,
and local environmental agencies to facilitate information sharing amongst stakeholders and to develop
an inventory of sensor-based community air quality investigations. EPA Regional Office staff have
highlighted EPA’s Air Sensor Toolbox for Citizen Scientists [
38
] in this outreach, facilitating exchanges
on sensor operation and maintenance, funding opportunities, and tools for data interpretation.
For example, the Minnesota Pollution Control Agency provided input to the EPA Regional Office
in Chicago (Region 5) on its own experience managing an air sensor loan program, and shared
documentation it had developed to inform community members about proper sensor operation.
Monitoring for the NAAQS pollutants is delegated to state agencies with oversight from EPA
Regional Offices and EPA’s Office of Air Quality Planning and Standards (OAQPS); there is a structured
regulatory program to determine compliance status with each NAAQS. Emissions of HAP are regulated
at the source of the emissions, and monitoring is mostly initiated at the local level (i.e., state, tribal and
local agencies), with federal guidance and some federal funding. There are very specific regulatory
requirements in monitoring for the NAAQS pollutants but the States and Tribes have more leeway in
monitoring for other purposes (e.g., for HAPs). EPA’s Superfund Program uses various technologies
for emergency response, and new sensors may be useful to inform decisions in these scenarios.
Federally recognized Tribes interface with EPA and ATSDR on a government-to-government basis;
many receive EPA grants or ATSDR technical assistance to do monitoring and other environmental
health projects.
2.2.3. Sensor Manufacturers
Low cost air quality sensors have exploded onto the commercial market in the last decade.
Development and manufacturing of these devices is not isolated to a single sector or industry.
Some of the earliest developers were design teams associated with academic industrial/design arts
programs, telecommunication research groups, and non-profit citizen scientists [
39
]. In particular,
these developers took advantage of inexpensive sensor components and their individual areas
of expertise to craft air quality sensors to meet a wide variety of needs [
40
]. More traditional
instrumentation manufacturers have recently started to invest research capital into the low cost
sensor area [
41
]. Their presence is the result of the growing citizen science movement and an obvious
market niche.
The existence of a multitude of sensor manufacturers has resulted in both obvious benefits
and deficits. On the plus side, there are numerous low cost sensors in a price range typically
well under $1000 available for the interested user for both particle as well as gas phase pollutant
monitoring. Many offer user friendly features such as immediate data visualization through smart
phone applications with wireless data transmission. Some concerns associated with low cost sensor
production is that often manufacturers may lack the technical know-how or capabilities to adequately
test or calibrate their devices prior to releasing them to the market [
27
]. Other concerns include
reliability issues between replicate copies of the same sensors attributed to either a poor overall
121
Books
MDPI
Atmosphere 2017, 8, 182
manufacturing process associated with the fully assembled sensor or issues with the individual
components themselves [42,43].
Recent sensor evaluation efforts by recognized institutions, including the U.S. EPA and
others [6,44]
would appear to be having a positive impact upon sensor manufacturing. In particular,
work conducted in 2012 resulted in some of the first reported performance tests of sensors from
a wide variety of manufacturing sources [
2
]. The evaluations resulted in an almost immediate
update of the technology by many of the manufacturers to overcome issues first revealed in the
tests (e.g., battery failures, low detection sensitivities, poor telecommunications protocols). Some of
those early manufacturers would appear to have left the market while other new ones have joined.
These observations suggest that it is likely that more mature instrument manufacturers will continue
to invest in this area and capture more of the market share as start-up groups continue to develop
unique devices to meet a particular niche.
2.3. Calibration and Validation
As mentioned previously, the calibration procedures conducted by most low-cost sensor
manufacturers are often severely inadequate or non-existent and fail to establish the sensor’s
performance characteristics, especially when used in the ambient environment. End users and research
organizations have often taken the lead on establishing the performance of these emerging technologies
and sharing their results. Noted groups establishing sensor performance include the European
Union’s Joint Research Center [
45
], the U.S. EPA [
46
], and the South Coast Air Quality Management
District’s AQ-SPEC laboratory [
47
]. Results show that while nephelometric (light scattering) devices
might perform well in direct chamber-based evaluations, they often reveal significant departures
from a true reference grade instrument response under real-world (ambient) conditions [
1
]. In like
manner, while excellent chamber-based response relationships have been observed for select gas phase
pollutants [
2
], ambient test results have been less promising or inconsistent [
42
]. There are of course,
exceptions to these observations and certain devices and sensors appear to show more promise relative
to their performance features [48].
Gas detection in the workplace has a long history of being driven by flammability concerns,
beginning with the miners’ safety lamp, but direct-reading real-time instruments have also been
used to determine toxic gases for almost 100 years; colorimetric detector tubes patented for detecting
carbon monoxide in the 1930’s are examples. More recent developments of infra-red analyzers
and gas chromatography detectors (with or without an associated gas chromatograph) have been
portable, but not personal, although current research into miniaturization may alter that situation.
Electrochemical cells have been developed for specific acute hazards, such as carbon monoxide,
hydrogen sulfide and chlorine. These are used in process safety applications, but have also been
adapted for use in personal dosimeters. The quality of data used for safety monitoring and personal
exposure assessment in the workplace is of paramount importance, particularly in relation to
acutely hazardous substances. The European Union countries have developed guidance for the
performance, testing, selection, installation, use and maintenance of electrical apparatus used for
the direct detection and direct concentration of toxic gases and vapors. Other Standards-setting
organizations have published similar products, and a joint Working group of the International
Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC)
are working on umbrella standards to replace these (these Standards are detailed in Appendix A).
In the USA, NIOSH has developed guidance for ensuring data quality for gas and vapor detection,
including during emergency response [
49
,
50
]. The American Industrial Hygiene Association, through
their Gas and Vapor Detection Systems Committee (now Real-Time Detection Systems Committee) has
developed guidance for manufactures to report specifications for electronic real-time gas and vapor
detection equipment [51].
Recognizing the need to calibrate instrument response under real-world conditions, variable
performance based on changing conditions (e.g., aerosol size-distribution), and deteriorating sensor
122
Books
MDPI
Atmosphere 2017, 8, 182
response as due to age, new research efforts have been focused on the idea of auto-calibration
methods [
52
]. Such methods may be able to maintain the accuracy of a node-based distributed
sensor system for longer term measurements.
Extending beyond the need for validation in controlled laboratory settings to “real world”
conditions and environmental epidemiological and citizen science applications, the National Institute
for Environmental Health Sciences (NIEHS) released a phased field validation program in 2013 [
53
].
Through this effort, teams of engineers, exposure scientists and environmental epidemiologists are
working together to test sensor performance at the pilot scale, including iterative improvements
in sensor system design and performance and then scaling up the application to a full-sized
epidemiological study to demonstrate the added value of temporally and spatially resolved air
pollution exposure assessment relative to existing ‘citizen’ measures.
3. Related Projects and Programs
Although the ASHG has not been a prime mover in development of key products or programs,
it has served as an advisory group and has contributed to a number of projects and programs, as noted
in the sections below.
3.1. Village Green
The U.S. EPA’s Village Green project has provided a test-bed for long-term evaluation and
application of emerging air quality sensors in a variety of community settings [
54
]. While the sensor
technologies for particulate matter and ozone are not considered truly low cost (~$6000/each), they do
represent mid-tier technologies [
55
] which are showing good to excellent capabilities for certain
attributes. In particular, they are capable of providing extended periods (months) of ambient air
quality monitoring using sustainable energy (solar power) with little or no technical support and
often with a high degree of agreement with local reference monitoring [
56
]. Another unique feature of
the Village Green is that it was designed to stream data continuously to the public via its web-based
data portal. It was in fact, the first real-time public reporting of air quality data for environmental
awareness purposes provided to the general public by the US EPA. Since its conception in 2014,
Village Green stations have now been deployed in a total of eight (8) U.S. metropolitan areas and
involve a variety of air quality sensor and emerging technologies (variable sustainable power supplies,
data microprocessing features, etc.).
At the time of its development as a technology test bed and community air quality awareness tool,
its capabilities to meet purposeful air quality data analyses were not established. EPA’s investigations
into short time interval sensor data messaging provided an opportunity to use the Village Green for
such an analysis due to its extensive database containing 1-min measurements of both ozone and fine
particulate matter. Analysis of that database resulted in a pilot Sensor Scale associated with potential
short time interval air quality measurements [
5
]. EPA is launching a pilot project to test a new tool
for making instantaneous outdoor air quality data useful for the public. The new “Sensor Scale”
is designed to be used with air quality sensors that provide data in short time increments–often as
little as one minute [
5
]. EPA developed the scale to help people understand the 1-min data the stations
provide and how to use those data as an additional tool for planning outdoor activities.
3.2. The E-Enterprise Advanced Monitoring Team (EEAMT)
Under the direction of the E-Enterprise Leadership Council, a joint EPA/State Advanced
Monitoring team was formed in April 2015 to address the challenges and opportunities presented
by rapidly changing air and water technologies. Data interpretation is one of five priority projects
identified in the path forward for EPA, States, and Tribes [
57
]. In order to provide context and
interpretation of advanced monitoring data in formats relevant and understandable to users, the team
was charged with advancing (1) statistical analyses to understand the relationship between continuous
data and data collected over longer-term averaging times or via discrete (e.g., bi-weekly) sampling;
123
Books
MDPI
Atmosphere 2017, 8, 182
(2) development of visualization tools (e.g., interactive maps) and websites with appropriate messaging;
and (3) development of outreach and communication materials.
Conducting statistical analyses to understand the relationship between short term measurements
and longer term standards or discrete measurements is a blanket need across media and pollutants.
While messaging already exists to alert the public about air or water quality conditions experienced
over specific time periods (e.g., 1-h, 8-h, or 24-h), the same messaging should not be used to translate
short term (e.g., 1-min) measurements. Two articles discussing an approach to relate short term ozone
and particulate matter (PM
2.5
) measurements with longer term averages have been published [
58
,
59
].
Other needs identified by the team include data analysis and messaging for SO
2
,NO
2
, CO, PM
10
,
benzene, total volatile organic compounds (VOCs), and other specific VOC compounds.
In order to create visualization tools that make continuous and discrete state/federal data more
accessible and understandable, EPA and States have determined it is important to present real-time
(e.g., hourly concentration values) along with appropriate caveats. Examples of caveats include
marking data as “raw”, “provisional”, or “final” and distinguishing between different forms of data
(e.g., regulatory vs. peer reviewed). A need exists to display continuous data along with discrete
measurements while giving the data context including geographical and meteorological information.
Mockups were drafted as part of a thought process to address these issues and allow for EPA and States
to provide guidance or expertise to big data developers desiring to stream, collect, or use regulatory
and sensor data. Once short term messaging has been developed, it is important to develop appropriate
outreach and communication materials that EPA and States can use for consistent messaging including
frequently asked questions (FAQs) and a standardized, centralized repository of metrics that break
down toxicity/health information by pollutant for both short term and long term effects (e.g., gaseous,
metals, PM, pH, toxins, dissolved oxygen, etc.). Outreach material should describe the limitations of
sensors and sensor networks including information on what we do not know and cannot measure
with confidence.
3.3. Homeland Security Applications
Air sensors have many important applications relating to homeland security, but two roles are
especially important: (1) serving as sentinels to detect the release of dangerous substances into the
air, and (2) providing measurements of the amount of dangerous substances present in the air so that
health risks can then be assessed and appropriate decisions made to protect public health.
Serving as sentinels to provide early warning of the release of dangerous substances is a very
critical function for real-time air sensors. In this capacity, air sensors are designed to monitor
background levels of various substances (dangerous chemicals or/or their indicators) and detect any
significant increases in their presence. Air sensor networks have already been deployed extensively
as sentinels in several major cities, chemical plants, military installations, and during major public
events. These networks have been deployed on roof tops, in underground transportation systems,
and in various types of mobile ground vehicles and aircraft.
In addition to their role as sentinels to detect the release of dangerous chemicals, air sensors may
be used to measure concentrations of harmful chemicals which can potentially provide an estimate of
health risks to persons breathing the air. Estimating health risks from acute or short-term exposure
durations is a challenging task, as has already been noted in this paper. Most of the tools used to assess
health risks are more centered on long-term, chronic, or lifetime exposure to chemicals. For example,
as EPA’s premier program for assessing health risks from chemicals, the Integrated Risk Information
System (IRIS) has focused on dose-response data for chronic exposures. Health effect reference values
derived to assess long-term health risks have very little, if any, value for assessing health risks from
acute, short-term, or intermittent exposure scenarios, and few have been developed to assess those
health risks to the general public.
A number of chemical release accidents, including the December 1984 accidental release of methyl
isocyanate from a Union Carbide chemical plant in Bhopal, India, led to the formation of the Acute
124
Books
MDPI
Atmosphere 2017, 8, 182
Exposure Guideline Level (AEGL) program in 1995 [
60
]. Standing Operating Procedures for deriving
AEGL values were developed by the National Research Council (NRC) [
61
] for use in both planning
for and during a catastrophic chemical release event, using values covering inhalation exposures from
10-min to 8-h.
Shortly after the events of 11 September 2001, the need for additional sets of values for use
for remediation in the aftermath of such events led to the development of another set of health
metrics, called Provisional Advisory Levels (PALs). PALs were derived by building upon the AEGL
methodology developed by the NRC [
61
], to create values for use at 1-day, 30-days, 90-days, and
2-years durations [
62
]. Under the PALs program, EPA has drafted over 3000 numeric values for over
one hundred priority chemical agents, two routes of exposure (ingestion and inhalation), three levels of
harmful effects, and four relatively short exposure durations. These values, in addition to those already
published by many other organizations [
3
], can provide a context on which to compare sensor-derived
exposure levels.
3.4. Indoor Air–Non-Industrial
Indoor air quality (IAQ), which is the air quality within buildings (i.e., homes, schools, offices and
other non-industrial buildings) and other enclosed spaces, can affect the health, comfort and ability to
perform for occupants, from infants to senior citizens. IAQ involves many factors including: outdoor
air contributions; chemical, micro-biological, and particulate contaminants; and, characteristics of the
indoor climate such as temperature, humidity and airflow. Americans spend about 90 percent of their
time indoors [
63
–
66
], where pollutant levels, like some VOCs, may be two to five times higher—and
occasionally 100 times higher than outdoors [
67
–
69
]. EPA traditionally has focused its IAQ efforts to
identify sources and developing guidance to reduce human exposure to unhealthy indoor air and
to provide low-cost mitigation strategies consistent with public health practices. In addition, many
international professional, trade, and standards organizations have dedicated committees addressing a
broad range of indoor air quality issues. With the increased availability of low-cost sensor technology,
however, the resources we use to assess indoor air have been expanding.
One challenge for sensor work within indoor environments is the fact that the several pollutants
important in the indoor environment cannot be detected accurately by current low-cost sensor
technology. Indoor air sensors have the same challenges that ambient air sensors encounter when
attempting to evaluate short-term sensor readings. Most of the indoor air pollutants also share the
challenge associated with HAPs, the lack of enforceable or agreed upon health based standards to with
which to compare sensor readings. As with HAPs, the available health effect reference values may be
inappropriate for direct use with indoor air sensor readings. There are several other challenges unique
to using sensors indoors—for example, unlike the ability of ambient air sensors to be collocated with
regulatory monitors to help assess accuracy, there is no similar availability for indoor air.
While guidance development around IAQ best practices will continue within EPA, the potential
for increased use of sensor technology for indoor air quality continues to evolve. Sensors that
work with occupants’ health-assessment technologies and those that can actuate building systems or
individual appliances automatically may help enhance IAQ efforts in the built environment. There is
a great potential that, as low-cost sensor technology improves, it may be a viable complement to
comprehensive building system approaches currently used by the IAQ community and would
help inform and improve IAQ stakeholders’ ability (both professional and consumer level) to
use IAQ management tools, interpret air quality data for their space, and assess the benefits of
IAQ-related actions.
It is important that we not only continue the engineering “fixes” that are constantly improving
source control, as well as ventilation and filtration/air cleaning systems, but also work to improve
sensor technology as a tool to help optimize building performance and increase the accessibility
of these tools as a complement to integrated strategies for improving IAQ for all parts of society.
Within the context of IAQ, this may involve creating future systems that can be automated, improving
125
Books
MDPI
Atmosphere 2017, 8, 182
the integration of these low-cost tools with building and health-assessment systems, establishing
guidance that creates IAQ best practices that integrates these tools into existing public health
practices and provides appropriate messaging for the public to better understand their health in
their indoor environments.
4. Maximizing the Usefulness of Sensor Readings
As discussions within the ASHG have progressed over the years, approaches have been considered
on how sensors may improve our understanding of real-world dosimetry, and how we might better
organize the process of developing and using sensors through life-cycle analysis. Both of these
approaches are more immediately relevant to research-grade sensor technologies; however, as more
improvements in sensitivity and specificity are made to the low-cost sensors available to citizen
scientists, these approaches will apply to a more universal set of sensor users.
4.1. Transitioning Sensors into Dosimeters and Future of Exposure Science
Radiation biology provides a paradigm for transitioning sensor-derived data into estimates of
internal dose. Biological effect data from early animal studies modeled radiation exposures to define
internal dose and algorithms were developed so that the readout from portable detector could be
translated into the dose delivered to internal target organs.
It is the internal dose (the concentration of the agent at the cell/organ/tissue being affected) and
not the environmental exposure level that is more closely related to toxicological effects. Health-based
standard setting and development of health effect reference values are moving to apply emerging
advances in exposure and toxicological sciences aimed at a more seamless integration of exposure
and internal dose metrics, together with the incorporation, coherent alignment, and coordination
of novel data streams [
70
–
72
]. Elucidation of the linkages between exposures and adverse effects
in humans and the ecosystem will result in a better understanding on which to develop effective
management strategies.
Constructing a robust context for this integration calls for coordinated research with human-health
and ecologic-health scientists to identify, collect, and evaluate data that capture internal and external
markers of exposure in a format that improves the analysis and modeling of exposure–response
relationships and links to emerging methods for hazard identification such as high-throughput test
systems (HTS). Figure 1 depicts selected scientific and technologic advances for measuring and
monitoring considered in relation to a conceptual integration of exposure and dose as espoused by the
National Research Council [71].
Mattingly et al. [
73
] developed an exposure ontology (ExO) designed to address the lack
of exposure information required to elucidate environmental contributions to diseases, translate
molecular insights from new technologies such as HTS, and aid assessment of human and ecological
risks. The ExO formalized definition such as “exposure receptor” and centralized the role of exposure
science with the intent to extend its ability to integrate and analyze exposure information within the
broader context of environmental health. The exposure receptor can be an organ, tissue or cell, and the
exposure stressor can be a biological, physical, or psychosocial agent, so that exposure assessment may
include estimating the magnitude, frequency, and duration of an exposure, along with characteristics of
the specific receptor. This concept of the exposome is consistent with the life-cycle approach articulated
above regarding sensors. Such integration will expand the impact of exposure data and inform existing
environmental health data by providing associated real-world exposure context.
On the response side of the exposure-dose-response linkage, adverse outcome pathways (AOP)
are emerging as an important construct for the integration of toxicological effects across various levels
of biological organization (e.g., genomic, cellular, target tissue, individual, population) based on HTS
and identification of molecular initiating events [
74
,
75
]. This construct is also completely consistent
with the ExO, and more recently Teeguarden et al. [
76
] have proposed formal linkage of exposure
science with the AOP by means of the aggregate exposure pathway (AEP).
126
Books
MDPI
Atmosphere 2017, 8, 182
Therefore, emerging technologies now provide a method whereby measures of environmental
exposure can be translated to an internal dose which can then be used in applications such as
biomonitoring of populations or in cell systems. These approaches essentially provide a scalable
platform with which to depict both exposure and internal dose. Alignment of exposures across
experimental toxicity-testing systems can be achieved by understanding, measuring, and applying
this information on the processes that control the time course of concentrations and delivery of
chemicals and particles to target cells in various test systems at different scales (e.g., scale of ecological
epidemiological studies, target tissue dose for
in vivo
animal studies, delivered concentration in
HTS) that may be used as the basis for developing health-based values to which sensor data may
be compared. Such exposure and internal dosimetry considerations can provide a context for the
interpretation of emerging sensor data and may also inform future considerations of the form of
various health-based reference values.
Figure 1. Conceptual integration of exposure and dose (from National Research Council).
4.1.1. Dosimeters in the Workplace
The Army projects aimed at developing a naphthalene dosimeter are an example of how
biomarkers of exposure are being integrated with chemical sensor-derived data [
77
]. The first in the
series of naphthalene dosimeter projects was funded through the Army’s SBIR program. It developed
an instrument capable of measuring naphthalene from air every three minutes [
78
]. A second project,
underway at NIOSH, is an independent evaluation of the performance of the instrument. A third
project lead by the U.S. Army Research Institute of Environmental Medicine is deploying the prototype
instrument on military fuel-handlers to measure the concentration of naphthalene in their breathing
zone. Concurrent to collecting the personal real-time exposure data, biomarkers of naphthalene
exposure are being collected from the exhaled breath, urine and skin of the individual wearing the
air sensor. Following the radiation dosimetry paradigm, the concurrently collected exposure and
biomarker data will be used in a fourth project that will inform development of a model to estimate
internal dose [79].
127
Books
MDPI
Atmosphere 2017, 8, 182
4.1.2. Dosimetry for NAAQS Air Pollutants
Portable sensors available for some NAAQS pollutants (namely, PM, O
3
, and NO
2
) have been
evaluated by EPA. There are not specific metabolites that can be used to quantify the general
relationship between exposure and dose. However, unlike naphthalene which enters the body
by both dermal absorption and inhalation, at least these three NAAQS pollutants enter the body
predominately by inhalation. In general, greater than 80% of inhaled ozone is absorbed in the
respiratory tract (see Table 5-1 of U.S. EPA, 2013 [
80
]). Thus, dose can be approximated as the product
of O
3
concentration, minute ventilation, and duration of exposure. Conceivably, an estimate of
minute ventilation could be derived using an accelerometer in combination with data on the mass
and gender of an individual. A more exact linkage to predicted decrements in lung function due to
O
3
exposure, at least in healthy individuals, could be calculated by linking dosimeter data with an
existing model [
81
]. This type of approach could be applied to both occupational and ambient ozone
exposures or adapted for other compounds where health endpoints are closely related to inhaled dose.
Development of a dosimeter for PM is complicated due to the dependence of particle deposition
on inhaled particle size, route of breathing, tidal volume, breathing rate, and lung size [
82
]. A couple of
PM samplers that mimic the deposition in adults during nasal breathing and light exercise based on the
International Commission on Radiological Protection (ICRP) model [
83
] have been developed. First,
Koehler et al. [
84
] employed the use of a foam plug in which particle deposition efficiencies are similar
to the ICRP [
83
] predicted total respiratory tract deposition (average of adult males and females) for
particles between 0.05 and 2
μ
m. Total deposited dose in the respiratory tract can be determined either
gravimetrically or by digesting the foam and extracting metals or organs for quantification by other
means. Second, TSI has developed a Nanoparticle Surface Area Monitor (Model 3550) which can
be used to determine particle surface area depositing in the tracheobronchial and alveolar regions
of the lung for particles between 0.01 and 1
μ
m. The Model 3550 is designed to match predicted
deposition in an adult male. Its operation is based on the diffusion charging of particles followed by
detection of the aerosol using an electrometer. Particle doses can be assessed by the second, computed
as a time-weighted average, or cumulative total deposed particle surface area. Available portable
PM sensors typically use light scattering to estimate PM mass or number concentration in air for
micron-sized or larger particle fractions such as PM
2.5
. Unless an underlying PM size distribution is
known or assumed, estimates of dose cannot be derived based on data from these portable PM sensors.
The utility of a dosimeter for ambient NO
2
is questionable. For NO
2
, one of the critical endpoints
is asthma exacerbation via an increase in airways responsiveness, as discussed in Section 5.2.2 of the
Integrated Science Assessment (ISA) for NO
2
[
85
]. Since the increase in airway responsiveness does not
appear to be associated with NO
2
dose for NO
2
concentrations between 100 and 600 ppb [
86
], it may
be sufficient to monitor NO
2
concentration. Additionally, ambient NO
2
concentrations are generally
only elevated near roadways, as discussed in Section 2.5.3 of the 2016 ISA [
85
]. An elevated NO
2
sensor reading could serve as a warning for a person with asthma to take actions to reduce exposure,
e.g., by using recirculation of air in an automobile or avoiding outdoor activities in close proximity to
major roadways or during periods of increased traffic.
4.1.3. Dosimeters for Hazardous Air Pollutants (HAPs)
There are 187 air pollutants designated as HAPs, including several VOC, that are emitted by point
sources and under the purview of Section 112 of the Clean Air Act [
87
]. These pollutants are known or
suspected to cause cancer or other serious health effects, such as reproductive effects or birth defects, or
adverse environmental effects and are often also encountered in the workplace. As sensor technologies
emerge to address these pollutants, our recommendations for life-cycle characterization and dosimetry
as discussed above will be essential to create proper context for interpretation of sample measurements
with comparisons to appropriate health effect reference values, when such values are available [3].
128
Books
MDPI
Atmosphere 2017, 8, 182
4.2. Adopting a Life-Cycle Approach
An overarching life-cycle framework and decision-making process that the ASHG has encouraged
as an ideal for air quality sensor applications is illustrated in Figure 2. The life-cycle concept was
originally developed for use in emergency response situations [
88
], adapted and applied for radioactive
air sampling and instrumentation [
89
], adopted as a systematic way to organize the framework of the
White House’s signature initiative on Nanotechnology for Sensors and Sensors for Nanotechnology [
90
],
and most recently expanded to meet all manner of the emerging sensor needs for safety, health,
well-being and productivity [
91
]. The lifecycle begins with a clear and complete identification of the
purpose of the measurement, including what needs to be measured, under which conditions it needs to
be measured, and how well it needs to be measured. The lifecycle guides the research and development,
prototype testing, qualification type testing, production control testing, and training needs for the
sensor system. The lifecycle further defines procedures for acceptance testing, initial calibration,
functional checks, conduct and evaluation of operational experience, maintenance and recalibration,
and periodic performance testing to confirm continued successful use of the sensor system. Effective
following of the life-cycle process ensures that the sensor methods and instrumentation will work
as intended under realistic conditions. Documentation and continuous improvement are essential at
each step.
Figure 2. The life-cycle approach for sensor methods and instrumentation [88–91].
Use of the lifecycle supports an approach to sensor methods and instrumentation that is consistent
with the roles served by resources such as My Air, My Health: An HHS/EPA Challenge [
92
] to develop
and validate new methods, and the online AirNow [
93
] maps and forecast data that are collected and
shared using federal reference or equivalent monitoring techniques or techniques approved by the
state, local or tribal monitoring agencies.
4.2.1. A Working Definition of Air Quality Sensors and Health Informatics
A proposed working definition of Air Quality Sensors and Health Informatics might be “the
science and practice of determining which information is relevant to meeting air sensor and health
objectives; developing and implementing effective mechanisms for collecting, validating, storing,
sharing, analyzing, modeling, and applying the information; and then confirming that appropriate
decisions were made and that desired mission outcomes were achieved” [
94
]. The additional steps
129
Books
MDPI
Atmosphere 2017, 8, 182
in the informatics lifecycle include “conveying experience to the broader community, contributing to
generalized knowledge, and updating standards and training” [
95
]. Successful informatics endeavors
will apply all of the steps in the process. ASHG members have also assisted with the development of
an attempt to define “data readiness levels”, which could help with relevance and reliability issues for
the collection, sharing, and application of air sensor data [96].
4.2.2. Roles and Responsibilities of Sensor and Data Customers, Creators, Curators, and Analysts
In the context of our working definition of informatics for air sensors and health, the roles and
responsibilities of the myriad individuals who are engaged in the development and application
of air quality sensors can be viewed as fitting into four categories: sensor and data customers
(who specify the sensors and data needs for their intended purposes), sensor and data creators
(who will develop relevant and reliable sensors and data to meet the customer needs), sensor and data
curators (who will maintain and ensure the quality of the sensors and sensor data), and sensor data
analysts (who will develop and apply models for data analysis and interpretation that are consistent
with the quality and quantity of the data and that those data meet the customers’ needs). In some
instances, the same individuals may perform all roles, and in the larger global reality the individuals
and their roles may extend over significant distances, organizations, and time periods. As shown in
Figure 3, effective communication across the many customer, creator, curator, and analyst interfaces
is essential, and that communication across each of the six interfaces must work effectively in both
directions [
95
]. This vision follows the views of Hendren et al. [
97
] on a collaborative approach to
assessing, evaluating, and advancing the state of the field for data curation in the emerging field of
nanomaterials and nanotechnology.
Figure 3.
Communication interfaces, roles, and responsibilities for air quality sensor and health data
customers, creators, curators, and analysts [95].
5. Forecast of Advancing Technologies
Major areas of opportunity to advance the state of the art and application of sensor technologies
include: strengthening and sustaining infrastructures for investments and collaborations, improving
the sharing of information, integrating data and a coherent interpretation across various venues to
construct a cumulative accounting of exposures, and advancing breakthroughs in miniaturization
of sensor systems. At this writing, there are efforts across multiple organizations (public and
private; large and small; formal and informal) with additional involvement by independent
130
Books
MDPI
Atmosphere 2017, 8, 182
inventors. Entrepreneurial organizations such as Aclima have become involved in collecting data
from mobile sensors attached to Google street-view cars [
98
] and are leading the way to making those
data accessible.
Equally important are opportunities to increase the linkage of measurements of exposure via
sensor readings to additional parameters useful to estimate dosimetry: biometric data (e.g., pulse rate,
breath rate, and the like, which are now available from devices such as FitBit); geospatial measurements
of location and daily movement patterns; and of the physical environment (temperature, humidity,
etc.). These capabilities are most actively being developed within the occupational exposure arena but
migration of these technologies to the low-cost sensor market is anticipated as value-added features.
Properly informing and providing useful guidance to ensure these technologies are appropriately
engaged will be another challenge for groups like the ASHG.
5.1. Infrastructure Needs
5.1.1. Forecast and Statement of Needed Investment
When real-time sensing is combined with concurrent GPS coordinates, the derived data are
multidimensional. Today, sensor-derived data include streaming of latitude, longitude, altitude in
addition to the chemical data. However, the hardware and software to fully capture the potential
usefulness is lacking. For example, visualization of multidimensional data might be imaged as a video
of a multi-colored topographical map that flexes as an individual moves through different levels of
the positional and chemical data. Depth of color might represent various concentrations of chemicals
detected by the sensors. If the full potential is to be captured, investment is essential. Infrastructure is
needed to not only capture multidimensional data streams but also analyze it as it is being produced.
This is especially relevant to homeland security and for rapid response following a catastrophic event.
The application of sensor technology to the field of epidemiology can also be cited as an investment
opportunity. New evidence of cause-and-effect will become evident by developing the ability to
retrieve archived multidimensional sensor-derived data and overlaying it with geographical disease
prevalence information. Challenges of scalability across dimensions of time and geography must be
addressed to properly interface with available health-based standards and reference values.
As described in Section 3.1, there is great interest, and investment, in transforming toxicity testing
with the application of high throughput analyses using cell cultures [
70
]. Generating evidence to
demonstrate the validity of these testing methods is an ongoing challenge. In this context, we have an
opportunity to transform systematic hazard identification and risk assessment processes by combining
real-time exposure monitoring with real-time image analysis. Again, investment in the data capture,
analytics and archiving capabilities is essential if this opportunity is to be realized.
5.1.2. Scientific Literature Collection and Coordination
Another essential investment is needed to network the fields of Health Science, Electrical
Engineering and Computer Science. A first step might be to encourage sensor-related journals to
petition the National Library of Medicine for incorporation into the database. The National Library
of Medicine’s PubMed database [
99
] provides the ability to search for peer-reviewed publications.
This includes research about the use of sensors to assess indoor and outdoor air. As an example,
PubMed includes air sensor-related publications from a journal called “Environmental Monitoring and
Assessment” such as this 2017 one about “Public engagement on urban air pollution: an exploratory
study of two interventions” [
100
]. However, not all scientific journals are indexed in PubMed and this
is especially true for journals specializing in emerging areas and technologies Editors and publishers
can submit their journals for inclusion in PubMed; however, not all journals will be accepted based
on NLM’s selection criteria [
101
]. The relatively new PubMed Commons allows for the sharing of
opinions and information about PubMed citations [
102
]. This could include comments about the
131
Books
MDPI
Atmosphere 2017, 8, 182
techniques used in a publication or alerting readers to consider looking at publications with more
current approaches and findings.
5.2. Miniaturization
Miniaturization has been essential to recent successes in moving laboratory-scale technologies to
the field. For example, to meet the demands for personal exposure monitoring of the size characteristics
and spatial distribution of ultrafine particles, Fierz et al. [
103
] have developed a compact, real-time
instrument called the diffusion size classifier (DiSC) which provides particle size information that is in
good agreement with the much larger and more expensive laboratory-based aerosol spectrometers.
Another personal monitoring device recently released into the market is the portable aerosol mobility
spectrometer (PAMS) which simultaneously measures the number-weighted size distribution of
submicrometer aerosol, including the nanoparticle fraction [
104
]. In developing the PAMS, NIOSH
investigators used miniaturization to overcome the prohibitive size, weight, and cost limitations of the
previous technology, and they also eliminated burdensome regulatory and administrative limitations
for record-keeping and transportation of the instrument by replacing the traditionally used radioactive
source for particle charge conditioning with a nonradioactive bipolar diffusion charger. Compared
to traditionally used instruments, these innovations provided reductions in size (by a factor of 20),
weight (by a factor of 10–15), and cost (by a factor of 4), along with improvements in analytical
performance. However, further improvements in instrument size, weight, cost, and performance are
still needed to make these technologies more widely affordable, deployable, and able to operate in the
wide range of particle number concentrations that can be encountered in both workplaces and ambient
environments. Needed improvements include smaller, more reliable power sources, including the
possibility of body-heat or motion-driven power sources, and more compact data collection, processing,
and memory capabilities. As described in the perspective article by Fadel et al. [
105
], advances in
nanotechnologies could enable and accelerate the development of inexpensive, portable devices for
the broad detection, identification, and quantification of biological and chemical substances, including
sensors for air quality and health applications.
6. Summary and Conclusions
The goals of the ASHG are to provide an open dialogue from the multiple disciplines and
agencies represented on a number of related issues: (1) improving the understanding of what sensor
technologies may be able to deliver for these agencies to meet their missions; (2) communication of
what small-scale sensor readings can provide to a number of audiences; and most centrally, (3) best
practices in communicating health-related messages to numerous audiences.
In this paper, we have attempted to cross-reference other related projects and provide additional
resources to an interested reader to pursue additional information from credible sources. Additional
aims were to provide an update on the advances made to date under the auspices of these various
programs, to forecast potential applications of rapidly emerging sensor technologies and to foster a
collaborative response to challenges involved in their application to research and support of decisions
related to air quality management.
Acknowledgments:
Funding for publication costs were provided by the U.S. Environmental Protection Agency,
Washington, DC USA, This project was supported in part by an appointment to the Internship/Research
Participation Program at the U.S. Environmental Protection Agency, administered by the Oak Ridge Institute
for Science and Education through an interagency agreement between the U.S. Department of Energy and EPA,
with additional support provided by the National Institutes of Health, National Library of Medicine.
Author Contributions:
George Woodall, Mark Hoover, Ronald Williams and Janis Hulla were the primary
architects of the outline for this review paper; all authors contributed in writing the paper.
Conflicts of Interest: The authors declare no conflict of interest.
132
Books
MDPI
Atmosphere 2017, 8, 182
Appendix A
Standards for equipment used to detect and determine toxic gases and vapors in a workplace or
similar situation.
A.1. Published
Europe:
1.
EN 45544:2000 Workplace atmospheres. Electrical apparatus used for the direct detection and
direct concentration measurement of toxic gases and vapours. Part 1: General requirements and
test methods; Part 2: Performance requirements for apparatus used for measuring concentrations
in the region of limit values; Part 3: Performance requirements for apparatus used for
measuring concentrations well above limit values; Part 4: Guide for selection, installation,
use and maintenance.
USA:
1. Underwriters Laboratory: UL 2075 Gas and Vapor Detectors and Sensors
2.
American National Standards Institute/International Safety Association:
ANSI/ISA-92.00.01-2010 Performance Requirements for Toxic Gas Detectors; ANSI/ISA
92.00.02-2013 Installation, Operation, and Maintenance of Toxic Gas-Detection Instruments
3.
American Society for Testing and Materials: ASTM E2885-13 Standard Specification for Handheld
Point Chemical Vapor Detectors (HPCVD) for Homeland Security Application
International Electrotechnical Commission:
1.
IEC 60079-29-1:2007 Explosive atmospheres—Part 29-1: Gas detectors—Performance
requirements of detectors for flammable gases
2.
IEC 60079-29-2:2007 Explosive atmospheres—Part 29-2: Gas detectors—Selection, installation,
use and maintenance of detectors for flammable gases and oxygen
Other:
1.
Australian/New Zealand Standard: AS/NZS 4641:2007 Electrical apparatus for the detection of
oxygen and other gases and vapours at toxic levels—General requirements and test methods.
2. ISO database: https://www.iso.org/committee/52702/x/catalogue/p/0/u/1/w/0/d/0
3.
CEN Database: https://standards.cen.eu/dyn/www/f?p=204:7:0::::FSP_ORG_ID:6245&cs=
178094E67E1897102F190938A48C7A285
A.2. Standards Proceeding through Process
ISO/IEC (IEC 62990-1) Workplace Atmospheres—Part 1: Gas detectors—Performance requirements of
detectors for toxic gases
ISO/IEC (IEC 62990-2) Work-place Atmospheres—Part 2: Gas detectors—Selection, installation,
use and maintenance of detectors for toxic gases and vapours and oxygen sensors.
References
1.
Williams, R. Findings from the 2013 EPA Sensors Workshop. Available online: https://www.epa.gov/air-
research/findings-2013-epa-air-sensors-workshop (accessed on 5 May 2017).
2.
Williams, R.; Long, R.; Beaver, M.; Kaufman, A.; Zeiger, F.; Heimbinder, M.; Hang, I.; Yap, R.; Acharya, B.;
Ginwald, B.; et al. Sensor Evaluation Report; U.S. Environmental Protection Agency: Washington, DC,
USA, 2014.
133
Books
MDPI
Atmosphere 2017, 8, 182
3.
U.S. Environmental Protection Agency. Graphical Arrays of Chemical-Specific Health Effect Reference Values
for Inhalation Exposures; U.S. Environmental Protection Agency: Research Triangle Park, NC, USA, 2009;
EPA/600/R-09/061. Available online: http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=211003
(accessed on 5 May 2017).
4.
U.S. Environmental Protection Agency. Guidance on Systematic Planning Using the Data Quality Objectives
Process. Available online: https://www.epa.gov/fedfac/guidance-systematic-planning-using-data-quality-
objectives-process (accessed on 5 May 2017).
5.
U.S. Environmental Protection Agency. Air Sensor Toolbox: What Do My Sensor Readings Mean?
Sensor Scale Pilot Project. Available online: https://www.epa.gov/air-sensor-toolbox/what-do-my-sensor-
readings-mean-sensor-scale-pilot-project (accessed on 2 May 2017).
6.
U.S. Environmental Protection Agency. Air Sensor Toolbox for Citizen Scientists, Researchers and Developers.
Available online: https://www.epa.gov/air-sensor-toolbox (accessed on 2 May 2017).
7.
ATSDR. Agency for Toxic Substances and Disease Registry Minimal Risk Levels (MRLs) 2016.
Available online: https://www.atsdr.cdc.gov/mrls/index.asp (accessed on 12 June 2017).
8.
Woodall, G.; Lipscomb, J.; Taylor, M. Review of health-based reference values for inhalation exposures. 2017;
in preparation.
9.
U.S. Environmental Protection Agency. Inhalation Health Effect Reference Values for Toluene (CASRN 108-88-3);
U.S. Environmental Protection Agency: Research Triangle Park, NC, USA, 2012. Available online:
http://oaspub.epa.gov/eims/eims-comm.getfile?p_download_id=512650 (accessed on 2 May 2017).
10.
U.S. Environmental Protection Agency. Chemical-Specific Reference Values for Benzene (CASRN 71-43-2); U.S.
Environmental Protection Agency: Research Triangle Park, NC, USA, 2012; EPA/600/R-12/047F1. Available
online: http://oaspub.epa.gov/eims/eimscomm.getfile?p_download_id=512648 (accessed on 2 May 2017).
11.
U.S. Environmental Protection Agency. Inhalation Health Effect Reference Values for Ethylbenzene
(CASRN 100-41-4); U.S. Environmental Protection Agency: Research Triangle Park, NC, USA, 2012;
EPA/600/R-12/047F2. Available online: http://oaspub.epa.gov/eims/eimscomm.getfile?p_download_
id=512649 (accessed on 2 May 2017).
12.
U.S. Environmental Protection Agency. Inhalation Health Effect Reference Values for Xylene—All
Isomers (CASRNs Mixed Isomers—1330-20-7; m-xylene—95-47-6; o-xylene—108-38-3; p-xylene—106-42-3);
U.S. Environmental Protection Agency: Research Triangle Park, NC, USA, 2012; EPA/600/R-12/047F4.
Available online: http://oaspub.epa.gov/eims/eimscomm.getfile?p_download_id=512651 (accessed on
2 May 2017).
13.
U.S. Environmental Protection Agency. Inhalation Health Effect Reference Values for Manganese (CASRN
7439-96-5—Manganese) and Compounds (CASRN 1344-43-0; 1317-35-7; and 1129-60-5); U.S. Environmental
Protection Agency: Research Triangle Park, NC, USA, 2012; EPA/600/R-12/047F5. Available online:
http://oaspub.epa.gov/eims/eimscomm.getfile?p_download_id=512652 (accessed on 2 May 2017).
14.
Harper, M.; Weis, C.; Pleil, J.D.; Blount, B.C.; Miller, A.; Hoover, M.D.; Jahn, S. Commentary on the
contributions and future role of occupational exposure science in a vision and strategy for the discipline of
exposure science. J. Expo. Sci. Environ. Epidemiol. 2015, 25, 381–387. [CrossRef] [PubMed]
15.
Fishbain, B.; Lerner, U.; Castell, N.; Cole-Hunter, T.; Popoola, O.; Broday, D.M.; Iñiguez, T.M.;
Nieuwenhuijsen, M.; Jovasevic-Stojanovic, M.; Topalovic, D.; et al. An evaluation tool kit of air quality
micro-sensing units. Sci. Total Environ. 2017, 575, 639–648. [CrossRef] [PubMed]
16.
Bart, M.; Williams, D.E.; Ainslie, B.; McKendry, I.; Salmond, J.; Grange, S.K.; Alavi-Shoshtari, M.; Steyn, D.;
Henshaw, G.S. High density ozone monitoring using gas sensitive semi-conductor sensors in the lower
Fraser Valley, British Columbia. Environ. Sci. Technol. 2014, 48, 3970–3977. [CrossRef] [PubMed]
17.
Gao, M.L.; Cao, J.J.; Seto, E. A distributed network of low-cost continuous reading sensors to measure
spatiotemporal variations of PM2.5 in Xi’an, China. Environ. Pollut. 2015, 199, 56–65. [CrossRef] [PubMed]
18.
Mead, M.I.; Popoola, O.A.M.; Stewart, G.B.; Landshoff, P.; Calleja, M.; Hayes, M.; Baldovi, J.J.; McLeod, M.W.;
Hodgson, T.F.; Dicks, J.; et al. The use of electrochemical sensors for monitoring urban air quality in low-cost,
high-density networks. Atmos. Environ. 2013, 70, 186–203. [CrossRef]
19.
Jovasevic-Stojanovic, M. CITI-SENSE Development of Sensor-Based Citizens’ Observatory Community for
Improving Quality of Life in Cities. Available online: http://www.citi-sense.eu/Portals/106/Documents/
Dissemination%20material/CITI-SENSE_Newsletter%20No4.pdf (accessed on 19 September 2017).
134
Books
MDPI
Atmosphere 2017, 8, 182
20.
CITI-SENSE. Deliverable D9.16 Project Overview for the Lay People. 2016. Available online: http://co.citi-
sense.eu/TheProject/Publications/Deliverables.aspx (accessed on 2 May 2017).
21.
CITI-SENSE. Development of Sensor-Based Citizens’ Observatory Community for Improving Quality of
Life in Cities. Available online: http://www.citi-sense.eu/ (accessed on 2 May 2017).
22.
Aspuru, I.; García, I.; Herranz, K.; Santander, A. Citi-Sense: Methods and tools for empowering citizens to
observe acoustic comfort in outdoor public spaces. Noise Mapp. 2016, 3, 37–48. [CrossRef]
23.
European Commission. Air Quality Standards. 8/6/16. Available online: http://ec.europa.eu/environment/
air/quality/standards.htm (accessed on 2 May 2017).
24.
Borrego, C.; Costa, A.M.; Ginja, J.; Amorim, M.; Coutinho, M.; Karatzas, K.; Sioumis, T.; Katsifarakis, N.;
Konstantinidis, K.; De Vito, S.; et al. Assessment of air quality microsensors versus reference methods:
The EuNetAir joint exercise. Atmos. Environ. 2016, 147, 246–263. [CrossRef]
25.
Penza, M.; Consortium, E. COST Action TD1105: Overview of sensor-systems for air-quality monitoring.
Procedia Eng. 2014, 87, 1370–1377. [CrossRef]
26.
Clean Air Asia. Clean Air Asia India. Available online: http://cleanairasia.org/india/ (accessed on
2 May 2017).
27.
Lewis, A.; Edwards, P. Validate personal air-pollution sensors. Nature
2016
, 535, 29–31. [CrossRef] [PubMed]
28.
BreezoMeter. BreezoMeter: Hyperlocal Air Quality Data, Available Globally. Available online: https:
//breezometer.com/ (accessed on 2 May 2017).
29.
WHO. Ambient (Outdoor) Air Quality and Health. Fact Sheet; WHO: Geneva, Switzerland, 2016. Available
online: http://www.who.int/mediacentre/factsheets/fs313/en/ (accessed on 2 May 2017).
30.
ATSDR. Health Consultation: Evaluation of Particulate Matter, Bennett Landfill Fire, Chester, South Carolina;
ATSDR: Atlanta, GA, USA, 2015.
31.
ATSDR. Health Consultation: Assessing the Public Health Implications of the Criteria (NAAQS) Air Pollutants and
Hydrogen Sulfide as Part of the Midlothian Area Air Quality Petition Response, Midlothian, Ellis County, Texas;
ATSDR: Atlanta, GA, USA, 2016. Available online: https://www.atsdr.cdc.gov/sites/midlothian/health_
consultations.html (accessed on 2 May 2017).
32.
ATSDR. Health consultation: Brooklyn Township PM2.5, Brooklyn Township, Susquehanna County, Pennsylvania;
ATSDR: Atlanta, GA, USA, 2016; Cost Recovery Number: 3A4K00. Available online: https://www.atsdr.cdc.
gov/HAC/pha/BrooklynTownship/BrooklynTwnsp_pm2-5_HC_Final_04-22-2016_508.pdf (accessed on
2 May 2017).
33.
ATSDR. Health Consultation: Review of Analysis of Particulate Matter and Metal Exposures in Air, KCBX
(AKA, Chicago Petroleum Coke sites), Cook County, IL, USA; ATSDR: Atlanta, GA, USA, 2016.
34.
NACEPT. NACEPT 2016 Report: Environmental Protection belongs to the Public, a Vision for Citizen
Science at EPA; U.S. Environmental Protection Agency: Washington, DC, USA, 2016; EPA 219-R-16-001.
Available online: https://www.epa.gov/faca/nacept-2016-report-environmental-protection-belongs-
public-vision-citizen-science-epa (accessed on 2 May 2017).
35.
Dosemagen, S.; Gehrke, G. Civic Technology and Community Science: A New Model for Public Participation
in Environmental Decisions in Confronting the Challenges of Public Participation: Issues in Environmental,
Planning, and Health Decision-Making. In Proceedings of the Iowa State University Summer Symposia on
Science Communication, Ames, IA, USA, 30 May–1 June 2013.
36.
Williams, R.; Kilaru, V.; Snyder, E.; Kaufman, A.; Dye, T.; Rutter, A.; Russell, A.; Hafner, H. Air Sensor
Guidebook; U.S. Enviromental Protection Agency: Research Triangle Park, NC, USA, 2016; EPA/600/R-14/159.
Available online: https://cfpub.epa.gov/si/si_public_record_report.cfm?direntryid=277996 (accessed on
2 May 2017).
37.
U.S. Environmental Protection Agency. Citizen Science in Newark, New Jersey. 2015. Available online:
https://www.epa.gov/sciencematters/citizen-science-newark-new-jersey (accessed on 12 July 2017).
38.
U.S. Environmental Protection Agency. Air sensor toolbox: Resources and funding. 2016. Available online:
https://www.epa.gov/air-sensor-toolbox/air-sensor-toolbox-resources-and-funding#TER (accessed on
5 June 2017).
39.
Heimbinder, M. Apps & Sensors for Air Pollution. Taking Space: The HabitatMap & AirCasting Blog 2012
4/9/2012. Available online: http://www.takingspace.org/epa-apps-sensors-for-air-pollution-workshop/
(accessed on 2 May 2017).
135
Books
MDPI
Atmosphere 2017, 8, 182
40.
Snyder, E.G.; Watkins, T.H.; Solomon, P.A.; Thoma, E.D.; Williams, R.W.; Hagler, G.S.; Shelow, D.;
Hindin, D.A.; Kilaru, V.J.; Preuss, P.W. The changing paradigm of air pollution monitoring.
Environ. Sci. Technol. 2013, 47, 11369–11377. [CrossRef] [PubMed]
41.
MacDonnell, M.; Raymond, M.; Wyker, D.; Finster, M.; Chang, Y.; Raymond, T.; Temple, B.; Scofield, M.;
Vallano, D. Mobile Sensors and Applications for Air Pollutants; U.S. Environmental Protection Agency:
Washington, DC, USA, 2013.
42.
Jiao, W.; Hagler, G.; Williams, R.; Sharpe, R.; Brown, R.; Garver, D.; Judge, R.; Caudill, M.; Rickard, J.;
Davis, M.; et al. Community Air Sensor Network (CAIRSENSE) project: Evaluation of low-cost sensor
performance in a suburban environment in the southeastern United States. Atmos. Meas. Tech.
2016
, 9,
5281–5292. [CrossRef]
43.
Smith, K.; Edwards, P.M.; Evans, M.J.J.; Lee, J.D.; Shaw, M.D.; Squires, F.; Wilde, S.; Lewis, A.C. Clustering
approaches that improve the reproducibility of low-cost air pollution sensors. Faraday Discuss.
2017
, 200,
621–637. [CrossRef] [PubMed]
44.
Kaufman, A.; Brown, A.; Barzyk, T.; Williams, R. The Citizen Science Toolbox: A One-Stop Resources for
Air Sensor Technology. In EM: Air and Waste Management Association’s Magazine for Environmental Managers;
Air & Waste Management Association: Pittsburgh, PA, USA, 2014.
45.
Spinelle, L.; Gerboles, M.; Aleixandre, M.; Bonavitacola, F. Evaluation of metal oxides sensors for the
monitoring of O3 in ambient air at Ppb level. Chem. Eng. Trans. 2016, 54, 319–324.
46.
Williams, R. Evaluation of Elm and Speck Sensors; U.S. Environmental Protection Agency: Washington, DC,
USA, 2015.
47.
Air Quality Sensor Performance Evaluation Center (AQ-SPEC). South Coast Air Quality Management
District. Available online: http://www.aqmd.gov/aq-spec/ (accessed on 2 May 2017).
48.
Lin, C.; Gillespie, J.; Schuder, M.D.; Duberstein, W.; Beverland, I.J.; Heal, M.R. Evaluation and calibration of
Aeroqual series 500 portable gas sensors for accurate measurement of ambient ozone and nitrogen dioxide.
Atmos. Environ. 2015, 100, 111–116. [CrossRef]
49.
NIOSH. Components for Evaluation of Direct-Reading Monitors for Gases and Vapors; NIOSH: Cincinnati, OH,
USA, 2012; Publication No. 2012-12-162. Available online: https://www.cdc.gov/niosh/docket/archive/
pdfs/niosh-220/2012-162.pdf (accessed on 2 May 2017).
50.
NIOSH. Addendum to Components for Evaluation of Direct-Reading Monitors for Gases and Vapors: Hazard
Detection in First Responder Environments; NIOSH: Cincinnati, OH, USA, 2012; Publication No. 2012-163.
Available online: https://www.cdc.gov/niosh/docket/archive/pdfs/niosh-220/2012-163.pdf (accessed on
2 May 2017).
51.
AIHA. Reporting Specifications for Electronic Real Time Gas and Vapor Detection Equipment Fact Sheet.
Available online: https://www.aiha.org/governmentaffairs/PositionStatements/Reporting%20Specifica-
tions%20for%20Real%20Time%20Detection_Final.pdf (accessed on 2 May 2017).
52.
Moltchanov, S.; Levy, I.; Etzion, Y.; Lerner, U.; Broday, D.M.; Fishbain, B. On the feasibility of measuring
urban air pollution by wireless distributed sensor networks. Sci. Total Environ.
2015
, 502, 537–547. [CrossRef]
[PubMed]
53.
National Institute of Environmental Health Science. Validation and Demonstration of Devices for
Environmental Exposure Assessment (R21/R33). 2013. Available online: https://grants.nih.gov/grants/
guide/rfa-files/RFA-ES-13-013.html (accessed on 2 May 2017).
54.
U.S. Environmental Protection Agency. Village Green Project. 4/12/17. Available online: https://www.epa.
gov/air-research/village-green-project (accessed on 2 May 2017).
55.
U.S. Environmental Protection Agency. (2013) DRAFT Roadmap for Next Generation Air Monitoring.
Available online: https://www.epa.gov/sites/production/files/2014-09/documents/roadmap-20130308.
pdf (accessed on 12 July 2017).
56.
Wan, J.; Hagler, G.; Williams, R.; Sharpe, B.; Weinstock, L.; Rice, J. Field assessment of the Village Green
Project: An autonomous community air quality monitoring system. Environ. Sci. Technol. 2015. [CrossRef]
57. Hindin, D.; Grumbles, B.; Wyeth, G.; Beneditc, K.; Watkins, T.; Aburn, G., Jr.; Ulrich, M.; Lang, S.; Poole, K.;
Dapolito Dunn, A. Advanced monitoring technology: Opportunities and challenges a path forward for EPA,
states, and tribes. EM Environ. Manag. 2016, 11, 16–21.
58.
Keating, M.; Benedict, K.; Evans, R.; Jenkins, S.; Mannshardt, E.; Stone, S.L. Interpreting and communicating
short-term air sensor data. EM Environ. Manag. 2016, 11, 22–25.
136
Books
MDPI
Atmosphere 2017, 8, 182
59.
Mannshardt, E.; Benedict, K.; Jenkins, S.; Keating, M.; Mintz, D.; Stone, S. Analysis of short-term ozone and
PM2.5 measurements: Characteristics and relationships for air sensor messaging. J. Air Waste Manag. Assoc.
2016, 67, 462–474. [CrossRef] [PubMed]
60.
U.S. Environmental Protection Agency. History of Acute Exposure Guideline Levels (AEGLs).
Available online: https://www.epa.gov/aegl/history-acute-exposure-guideline-levels-aegls (accessed on
12 July 2017).
61.
NRC. Standing Operating Procedures for Developing Acute Exposure Guideline Levels (Aegls) for Hazardous
Chemicals; National Academy Press: Washington, DC, USA, 2001.
62.
Young, R.A.; Bast, C.B.; Wood, C.S.; Adeshina, F. Overview of the Standing Operating Procedure (SOP) for
the development of Provisional Advisory Levels (PALs). Inhal. Toxicol.
2009
, 21, 1–11. [CrossRef] [PubMed]
63.
Klepeis, N.E.; Nelson, W.C.; Ott, W.R.; Robinson, J.P.; Tsang, A.M.; Switzer, P.; Behar, J.V.; Hern, S.C.;
Engelmann, W.H. The National Human Activity Pattern Survey (NHAPS): A resource for assessing exposure
to environmental pollutants. J. Expo. Anal. Environ. Epidemiol. 2001, 11, 231–252. [CrossRef] [PubMed]
64.
Ott, W. Human Activity Patterns: A Review of the Literature for Estimating Time Spent Indoors, Outdoors, and
In-Transit; U.S. Environmental Protection Agency, Office of Research and Development: Las Vegas, NV,
USA, 1989.
65.
Robinson, J.P.; Thomas, J. Time Spent in Activities, Locations, and Microenvironments: A California National
Comparison; U.S. Environmental Protection Agency: Las Vegas, NV, USA, 1992.
66.
U.S. Environmental Protection Agency. Child-Specific Exposure Factors Handbook (Final Report) 2008;
U.S. Environmental Protection Agency: Washington, DC, USA, 2008; EPA/600/R-06/096F. Available online:
http://cfpub.epa.gov/ncea/cfm/recordisplay.cfm?deid=199243 (accessed on 12 July 2017).
67.
Wallace, L.A. Project Summary: The Total Exposure Assessment Methodology (TEAM) Study; U.S. Environmental
Protection Agency: Washington, DC, USA, 1987.
68.
SAB. Integrated Human Exposure Committee Commentary on Indoor Air Strategy; U.S. Environmental Protection
Agency: Washington, DC, USA, 1998.
69.
Nehr, S.; Hosen, E.; Tanabe, S.I. Emerging developments in the standardized chemical characterization of
indoor air quality. Environ. Int. 2017, 98, 233–237. [CrossRef] [PubMed]
70.
NRC. Toxicity Testing in the 21st Century: A Vision and a Strategy; National Academies Press:
Washington, DC, USA, 2007.
71.
NRC. Exposure Science in the 21st Century: A Vision and a Strategy; National Academies Press: Washington,
DC, USA, 2012.
72.
National Academies of Sciences, Engineering, and Medicine. Using 21st Century Science to Improve Risk-Related
Evaluations; National Academies Press: Washington, DC, USA, 2017.
73.
Mattingly, C.J.; Mckone, T.E.; Callahan, M.A.; Blake, J.A.; Hubal, E.A. Providing the missing link:
The exposure science ontology ExO. Environ. Sci. Technol. 2012, 46, 3046–3053. [CrossRef] [PubMed]
74.
Villeneuve, D.L.; Crump, D.; Garcia-Reyero, N.; Hecker, M.; Hutchinson, T.H.; Lalone, C.A.; Landesmann, B.;
Lettieri, T.; Munn, S.; Nepelska, M.; et al. Adverse outcome pathway (AOP) development I: Strategies and
principles. Toxicol. Sci. 2014, 142, 312–320. [CrossRef] [PubMed]
75.
Villeneuve, D.L.; Crump, D.; Garcia-Reyero, N.; Hecker, M.; Hutchinson, T.H.; Lalone, C.A.; Landesmann, B.;
Lettieri, T.; Munn, S.; Nepelska, M.; et al. Adverse outcome pathway development II: Best practices.
Toxicol. Sci. 2014, 142, 321–330. [CrossRef] [PubMed]
76.
Teeguarden, J.G.; Tan, Y.M.; Edwards, S.W.; Leonard, J.A.; Anderson, K.A.; Corley, R.A.; Kile, M.L.;
Simonich, S.M.; Stone, D.; Tanguay, R.L.; et al. Completing the link between exposure science and toxicology
for improved environmental health decision making: The aggregate exposure pathway framework.
Environ. Sci. Technol. 2016, 50, 4579–4586. [CrossRef] [PubMed]
77.
Hulla, J.; Snawder, J.E.; Proctor, S.P.; Chapman, G.D. DOD impact assessment and management of
naphthalene-related risks. Toxicologist 2010, 114, 400.
78.
Hug, W.F.; Bhartia, R.; Reid, R.D.; Reid, M.R.; Oswal, P.; Lane, A.L.; Sijapati, K.; Sullivan, K.; Hulla, J.E.;
Snawder, J.; et al. Advanced environmental, chemical, and biological sensing technologies IX. In Wearable
Real-Time Direct-Reading Naphthalene and VOC Personal Exposure Monitor; Vo-Dinh, T., Lieberman, R.A.,
Gauglitz, G., Eds.; SPIE: Bellingham, WA, USA, 2012; Volume 8366.
79.
Hulla, J.; Proctor, S.; Snawder, J.E. The naphthalene dosimeter—Vanguard technology for improved health
protection. Toxicologist 2015, 144, 110.
137
Books
MDPI
Atmosphere 2017, 8, 182
80.
U.S. Environmental Protection Agency. Integrated Science Assessment for Ozone and Related Photochemical
Oxidants; U.S. Environmental Protection Agency, Office of Research and Development, National Center for
Environmental Assessment-RTP Division: Research Triangle Park, NC, USA, 2013.
81.
Mcdonnell, W.F.; Stewart, P.W.; Smith, M.V.; Kim, C.S.; Schelegle, E.S. Prediction of lung function response
for populations exposed to a wide range of ozone conditions. Inhal. Toxicol.
2012
, 24, 619–633. [CrossRef]
[PubMed]
82.
Brown, J.S.; Gordon, T.; Price, O.; Asgharian, B. Thoracic and respirable particle definitions for human health
risk assessment. Part. Fibre Toxicol. 2013, 10, 12. [CrossRef] [PubMed]
83.
ICRP. Human Respiratory Tract Model for Radiological Protection: A Report of a Task Group of the International
Commission on Radiological Protection; Pergamon Press: New York, NY, USA, 1994; Volume 24, pp. 1–482.
84.
Koehler, K.A.; Clark, P.; Volckens, J. Development of a sampler for total aerosol deposition in the human
respiratory tract. Ann. Occup. Hyg. 2009, 53, 731–738. [CrossRef] [PubMed]
85.
U.S. Environmental Protection Agency. Integrated Science Assessment for Oxides of Nitrogen-Health Criteria
(Final Report); U.S. Environmental Protection Agency, Office of Research and Development, National Center
for Environmental Assessment: Research Triangle Park, NC, USA, 2016.
86.
Brown, J.S. Nitrogen dioxide exposure and airway responsiveness in individuals with asthma. Inhal. Toxicol.
2015, 27, 1–14. [CrossRef] [PubMed]
87.
U.S. Congress. Clean Air Act as Amended in 1990, Section 112 (b) (1) Hazardous Air Pollutants. 1990.
Available online: https://www.epa.gov/history/epa-history-clean-air-act-amendments-1990 (accessed on
12 July 2017).
88.
Hoover, M.D.; Cox, M. Public protection from nuclear, chemical, and biological terrorism: Health Physics
Society 2004 summer school. In A Life-Cycle Approach for Development and Use of Emergency Response and Health
Protection Instrumentation; Brodsky, A., Johnson, R.H., Groans, R.E., Eds.; Medical Physics: Madison, WI,
USA, 2004; pp. 317–324.
89.
Hoover, M.D.; Cox, M. Radioactive air sampling methods. In A Life-Cycle Approach to Development and
Application of Air Sampling Methods and Instrumentation; Maiello, M.L., Hoover, M.D., Eds.; CRC Press/Taylor
& Francis: Boca Raton, FL, USA, 2011; pp. 43–52.
90.
NNI. Nanotechnology for Sensors and Sensors for Nanotechnology: Improving and Protecting Health, Safety, and
Environment; NNI: Washington, DC, USA, 2012.
91.
Hoover, M.D.; Debord, D.G. Turning numbers into knowledge: Sensors for safety, health, well-being, and
productivity. Synergist 2015, 26, 22–26.
92.
InnoCentive. My Air, My Health: An HHS/EPA Challenge. Available online: https://www.innocentive.
com/ar/challenge/9932947 (accessed on 12 July 2017).
93.
U.S. Environmental Protection Agency. AirNow. Available online: https://www.airnow.gov/ (accessed on
12 July 2017).
94.
De la Iglesia, D.; Harper, S.; Hoover, M.D.; Klaessig, F.; Lippelli, P.; Maddux, B.; Morse, J.; Nel, A.; Rajan, K.;
Reznik-Zellen, R.; et al. Nanoinformatics 2020 Roadmap. National Nanomanufacturing Network 2011.
Available online: http://eprints.internano.org/607/ (accessed on 20 June 2017).
95.
Hoover, M.D.; Myers, D.S.; Cash, L.J.; Guilmette, R.A.; Kreyling, W.G.; Oberdörster, G.; Smith, R.; Cassata, J.R.;
Boecker, B.B.; Grissom, M.P. Application of an informatics-based decision-making framework and process to
the assessment of radiation safety in nanotechnology. Health Phys.
2015
, 108, 179–194. [CrossRef] [PubMed]
96.
NNI. Nanotechnology Signature Initiative: Nanotechnology Knowledge Infrastructure (NKI) Data Readiness Levels
Discussion Draft; NNI: Washington, DC, USA, 2013. Available online: https://www.nano.gov/node/1015
(accessed on 20 June 2017).
97.
Hendren, C.O.; Powers, C.M.; Hoover, M.D.; Harper, S.L. The Nanomaterial Data Curation Initiative:
A collaborative approach to assessing, evaluating, and advancing the state of the field. Beilstein J. Nanotechnol.
2015, 6, 1752–1762. [CrossRef] [PubMed]
98.
Google. Mapping the Invisible: Street View Cars Add Air Pollution Sensors. Available online:
https://environment.google/projects/airview/ (accessed on 12 July 2017).
99.
U.S. National Library of Medicine. PubMed. Available online: https://www.ncbi.nlm.nih.gov/pubmed
(accessed on 17 August 2017).
100.
Oltra, C.; Sala, R.; Boso, A.; Asensio, S.L. Public engagement on urban air pollution: An exploratory study of
two interventions. Environ. Monit. Assess. 2017, 189, 296. [CrossRef] [PubMed]
138
Books
MDPI
Atmosphere 2017, 8, 182
101.
U.S. National Library of Medicine. Fact Sheet: MEDLINE
®
Journal Selection. 30 June 2017. Available online:
https://www.nlm.nih.gov/pubs/factsheets/jsel.html (accessed on 17 August 2017).
102.
U.S. National Library of Medicine. PubMed Commons. Available online: https://www.ncbi.nlm.nih.gov/
pubmedcommons/ (accessed on 17 August 2017).
103.
Fierz, M.; Houle, C.; Steigmeier, P.; Burtscher, H. Design, Calibration, and field performance of a miniature
diffusion size classifier. Aerosol Sci. Technol. 2011, 45, 1–10. [CrossRef]
104.
Kulkarni, P.; Qi, C.; Fukushima, N. Miniature differential mobility analyzer for compact field-portable
spectrometers. Aerosol Sci. Technol. 2016, 50, 1167–1179. [CrossRef] [PubMed]
105.
Fadel, T.R.; Farrell, D.F.; Friedersdorf, L.E.; Griep, M.H.; Hoover, M.D.; Meador, M.A.; Meyyappan, M.
Toward the responsible development and commercialization of sensor nanotechnologies. ACS Sens.
2016
, 1,
207–216. [CrossRef] [PubMed]
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
139
Books
MDPI
atmosphere
Article
Mobile DOAS Observations of Tropospheric NO
2
Using an UltraLight Trike and Flux Calculation
Daniel-Eduard Constantin
1,
*, Alexis Merlaud
2
, Mirela Voiculescu
1
, Carmelia Dragomir
1
,
Lucian Georgescu
1
, Francois Hendrick
2
, Gaia Pinardi
2
and Michel Van Roozendael
2
1
European Center of Excellence for the Environment, Faculty of Sciences and Environment,
“Dunarea de Jos” University of Galati, Str. Domneasca 111, Galati 800008, Romania;
mirela.voiculescu@ugal.ro (M.V.); carmelia.dragomir@ugal.ro (C.D.); lucian.georgescu@ugal.ro (L.G.)
2
Royal Belgian Institute for Space Aeronomy, Ringlaan-3-Avenue Circulaire, B-1180 Brussels, Belgium;
alexis.merlaud@aeronomie.be (A.M.); francois.hendrick@aeronomie.be (F.H.);
gaia.pinardi@aeronomie.be (G.P.); michel.vanroozendael@aeronomie.be (M.V.R.)
* Correspondence: daniel.constantin@ugal.ro
Academic Editors: Pius Lee, Rick Saylor and Jeff McQueen
Received: 24 February 2017; Accepted: 19 April 2017; Published: 22 April 2017
Abstract: In this study, we report on airborne Differential Optical Absorption Spectroscopy (DOAS)
observations of tropospheric NO
2
using an Ultralight Trike (ULT) and associated flux calculations.
The instrument onboard the ULT was developed for measuring the tropospheric NO
2
Vertical Column
Density (VCD) and it was operated for several days between 2011 and 2014, in the South-East of
Romania. Collocated measurements were performed using a car-DOAS instrument. Most of the
airborne and mobile ground-based measurements were performed close to an industrial platform
located nearby Galati city (45.43
◦
N, 28.03
◦
E). We found a correlation of R = 0.71 between tropospheric
NO
2
VCDs deduced from airborne DOAS observations and mobile ground-based DOAS observations.
We also present a comparison between stratospheric NO
2
Slant Column Density (SCD) derived from
the Dutch OMI NO
2
(DOMINO) satellite data product and stratospheric SCDs obtained from ground
and airborne measurements. The airborne DOAS observations performed on 13 August 2014 were
used to quantify the NO
2
flux originating from an industrial platform located nearby Galati city.
Measurements during a flight above the industrial plume showed a maximum tropospheric NO
2
VCD of (1.41
±
0.27)
×
10
16
molecules/cm
2
and an associated NO
2
flux of (3.45
±
0.89)
×
10
−3
kg/s.
Keywords: mobile DOAS; airborne observations; nitrogen dioxide; emission flux
1. Introduction
Nitrogen dioxide (NO
2
) is a chemical gaseous compound with an important role in the Earth’s
atmosphere. NO
2
is a key trace element in the chemistry of ozone, since it is involved in the catalytic
destruction of ozone in the stratosphere [
1
], while in the troposphere its photolysis leads directly to
the formation of ozone (O
3
) in the presence of VOCs (volatile organic compounds). NO
2
is released in
the atmosphere from natural sources (soil, lightning, solar cosmic rays) and anthropogenic emissions
(fossil fuels and biomass burning, industrial activities). Long-term exposure to NO
2
may affect the
respiratory system and lead to coronary diseases. NO
2
can lead to acidification of the aquatic ecosystem
following the oxidation to HNO
3
.
The Differential Optical Absorption Spectroscopy (DOAS) technique [
2
] has been used for NO
2
atmospheric measurements since the early 1970s [
3
,
4
]. Nowadays, besides ground-based zenith sky
measurements, DOAS techniques have developed into Multi-Axis Differential Optical Absorption
Spectroscopy (MAX-DOAS) observations [
5
]. The mobile DOAS technique was recently used on
Atmosphere 2017, 8, 78 140 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2017, 8,78
several platforms such as: cars [
6
,
7
], airplanes [
8
–
10
], Unmanned Aerial Vehicles (UAVs) [
11
]or
satellites [12–16].
Airborne observations have a number of important advantages for atmospheric research such as:
the flexibility during the flights and the possibility to access remote areas such as oceans, deserts, rural
or areas without roads.
The ULMs (Ultralight Motorized) are airborne platforms with an important scientific potential for
atmospheric research. So far, ULMs have been used to study the ultraviolet actinic radiation flux [
17
],
formaldehyde distribution [18], aerosol profiles [19], SO
2
,NO
2
and ozone distribution [20–23].
This work highlights the capability of a low-cost system (ULT-DOAS) used for measurements
of tropospheric NO
2
VCD and associated flux calculations. This study presents airborne DOAS
observations of tropospheric NO
2
using an Ultralight Trike (ULT) and associated flux calculations.
The work presented here was motivated by the need to further assess the intrapixel variability of
NO
2
detected by UV-VIS DOAS instruments onboard satellites. This work comes in the context of
a validation programme of the future Atmospheric Sentinels, starting with the Sentinel-5 Precursor to
be launched in summer 2017. A similar system based on DOAS onboard the ULM was used during
the Airborne Romanian Measurements of Aerosols and Trace gases (AROMAT) campaign, held in
Romania in August 2015 [
24
]. The AROMAT campaign was conducted under the aegis of the European
Space Agency (ESA) in the framework of a series of ESA field campaigns.
2. Methodology
2.1. Experimental and Instrumental Descriptions
The mobile DOAS observations were performed onboard of an Ultralight Trike (ULT),
in the South-East of Romania (Figure 1) during several days between 2011 and 2014. All measurements
were performed under clear-sky conditions (see Table 1). The Mobile DOAS system used for
measurements will be described in the following as the ULT-DOAS system. The measurements took
place in an area around Galati (located at 45.43
◦
N, 28.03
◦
E), Braila (45.26
◦
N, 27.95
◦
E), and close to
the industrial areas of Slobozia (44.56
◦
N, 27.35
◦
E). Note that an operational steel mill factory is located
in the vicinity of Galati, while Slobozia was chosen due to the presence of a fertilizer factory. Due to
security concerns, direct flights above the cities or industrial platforms were not performed. Most of
the airborne DOAS observations were performed in nadir geometry. Details about the ULT-DOAS
measurements are presented in Table 1.
Figure 1.
The main locations where airborne and/or ground-based mobile Differential Optical
Absorption Spectroscopy (DOAS) observations were performed.
141
Books
MDPI
Atmosphere 2017, 8,78
Table 1. Coordinates and temporal coverage of the mobile DOAS measurements.
Day Time Interval UTC *
Route of ULT-DOAS **
Measurements
NO
2
Source Target
1 September 2011 8.31–9.45 Galati–Braila Braila urban area
25 August 2012 7.53–8.89 Galati–Braila Braila urban area
21 July 2014 9.51–10.96 Galati–Slobozia Slobozia industrial area
13 August 2014 7.32–8.19 Galati Galati industrial area
* Coordinated Universal Time. ** ULT-DOAS = Ultralight Trike-Differential Optical Absorption Spectroscopy.
Airborne-DOAS measurements were accompanied by car-DOAS measurements and static twilight
observations on 21 July 2014. Static twilight DOAS measurements, used for determination of the NO
2
content from the reference spectra, were performed in a rural area close to Galati city. The car-DOAS
measurements were performed right before or after the experimental flights.
The aircraft used for all the flight experiments presented in this paper is a double-seated,
open-cockpit ultralight aircraft, trike type (model Fanagoria 21, produced by Plovdiv Air Bulgaria).
The flexible wing (Atlant-21, produced by Plovdiv Air Bulgaria) has an area of 16 m
2
. The ULT is
powered by a Subaru EA81 engine with 75 HP. The cruise speed is 75 km/h and the maximum speed
is 100 km/h relative to the ground. The aircraft has a maximum total weight at take-off of 450 kg.
The ULT-DOAS instrument consists of a compact Czerny-Turner spectrometer (AvaSpec-ULS2048XL-
USB2, of 175
×
110
×
44 mm dimensions and 855 g weight) placed in the Ultralight Trike. Figure 2
presents the instrumental DOAS set-up. The spectral range of the spectrometer is 280–550 nm with
0.7 nm resolution (FWHM—Full Width at Half Maximum) with a focal length of 75 mm. The entrance
slit is 50
μ
m and the grating is 1200 L/mm, blazed at 250 nm. The spectrometer is connected to the
telescope through a 400
μ
m chrome-plated brass optical fiber. The telescope achieves a 2.3
◦
field-of-view
with fused silica collimating lenses. Each spectrum is recorded by a laptop and georeferenced by a GPS
receiver. The spectrometer and the GPS receiver are powered by the laptop USB ports. The entire
set-up is powered by 12 V of the ULT through an inverter. Each measurement is a 10-second average
of 10 scans accumulations at an integration time between 50 and 150 ms.
This work is mostly based on nadir-DOAS observations but we also present zenith-sky
observations onboard ULT for stratospheric NO
2
measurements. The same DOAS system was used in
the case of the zenith sky car-DOAS observations.
Figure 2. Schematic of the ULT-DOAS measurement principle.
142
Books
MDPI
Atmosphere 2017, 8,78
2.2. Determination of the NO
2
Tropospheric Vertical Column and Flux Calculation
2.2.1. Retrieval of NO
2
Slant Column
The analysis of the measured spectra was performed using the QDOAS software [
25
]. For the NO
2
fit, the spectral window of 425–490 nm was used. The NO
2
spectral analysis included five absorption
cross sections: the NO
2
cross sections at 298 K and 220K [
26
], the O
3
cross section at 223 K [
27
],
the O
4
cross section [
28
] and a Ring spectrum [
29
]. A fifth-degree polynomial to account for scattering
processes and broad-band absorption in the atmosphere was used in the DOAS analysis. The direct
result of the spectral analysis is a differential slant column density (DSCD), which is the integrated
trace gas concentration along the light path through the atmosphere. The DSCD is the difference
between the slant column densities in the measured spectra (SCD
meas
) and the Fraunhofer reference
spectrum (SCD
ref
). The NO
2
amount in the Fraunhofer reference spectrum is unknown and its retrieval
is important for the determination of the SCD
meas
(Equation (1)).
SCD
meas
= DSCD + SCD
ref
(1)
Figure 3 presents a typical DOAS fit using a spectrum recorded during the experiment close to
the Galati steel factory, on 13 August 2014.
Figure 3.
Example of a DOAS fit realized with the QDOAS software; the analyzed spectrum was
recorded close to Galati, on 13 August 2014. Black line corresponds to molecular cross sections scaled
to the detected absorptions in the measured spectrum (red line).
The Slant Column Density (SCD) is converted to a Vertical Column Density (VCD) by means of
an Air Mass Factor (AMF), which is defined as the ratio between SCD and VCD (Equation (2))
AMF
=
SCD
VCD
=
τ
SCD
τ
VCD
(2)
where
τ
SCD
and
τ
VCD
are the optical thickness for the slant column (SCD) and vertical column
(VCD), respectively.
Since the measured spectra contain information about both stratospheric and tropospheric NO
2
content, the SCD
meas
can be written as:
SCD
meas
= AMF
tropo
· VCD
tropo
+ AMF
strato
· VCD
strato
− SCD
ref
(3)
143
Books
MDPI
Atmosphere 2017, 8,78
The above equations can be further simplified assuming that SCD
ref
is dominated by stratospheric
NO
2
. Using this assumption, the stratospheric contributions can be canceled by the NO
2
amount in
the reference spectrum (Equation (4)) [8].
VCD
tropo
= DSCD/AMF
tropo
(4)
The assumption presented above is valid if the reference spectrum (needed for the spectral
evaluation) is recorded at noon, in an area with a very low NO
2
content and if SCD
strato
does not
vary in time. The Fraunhofer reference spectrum could also be a zenith-sky spectrum recorded at high
altitude over the boundary layer [
30
]. However, in this work the tropospheric NO
2
VCD is based on
calculations using Equation (3).
2.2.2. Deduction of the SCD
ref
and VCD
strato
To avoid systematic errors due to the use of multiple reference spectra, only one spectrum will be
used for the spectral analysis of all DOAS observations presented in this paper.
The NO
2
amount in the reference spectrum was calculated using a photo-chemically modified
Langley plot [
6
,
31
]. The SCD
ref
corresponds to a zenith spectrum recorded at noon, in a clean rural
area close to Galati city. The spectrum was recorded on 13 August 2014 (9.70 UTC and solar zenith
angle (SZA) = 31.55
◦
).
The photo-chemically modified Langley plot was applied for the twilight sunrise observations
performed on 21 July 2014. By applying the Langley plot method, we calculated the SCD
ref
as
4.1 × 10
15
molecules/cm
2
of NO
2
.
The stratospheric contribution used for the retrieval of the VCD
tropo
is derived from the
assimilated vertical stratospheric columns simulated by Dutch OMI NO
2
(DOMINO). Table 2 shows
the satellite overpass data sets that were used for the retrieval algorithm presented in this work.
Table 2. OMI satellite overpasses data sets.
Day Orbit Nr.
Overpass Time
UTC
Stratospheric VCD
[
×10
15
molecules/cm
2
]
1 September 2011 37,923 11:04:28 3.76
25 August 2012 43,151 11:11:07 3.75
21 July 2014 53,272 11:17:36 4.14
13 August 2014 53,607 11:23:47 3.74
VCD = Vertical Column Density.
Figure 4A presents the SCD determined at twilight sunrise on 21 July 2014 compared with
the SCD
strato
derived from the DOMINO Level 2 product [
32
] scaled with a chemically modified
AMF calculated by PSCBOX [
33
,
34
]. More details about the retrieval of the SCD
strato
using twilight
observations and model simulations are presented in [
6
]. A good agreement between the two types of
SCD
strato
determinations is obtained, which gives confidence in the stratospheric SCD measured by
our static DOAS observations.
Figure 4B shows the SCDstrato derived from DOMINO compared with the SCDs determined by
car-DOAS zenith-sky observations and ULT-DOAS measurements performed in the zenith geometry,
for the same day of 21 July 2014. From this plot, one can see that the car-DOAS measurements are
dominated by tropospheric NO
2
while the zenith-sky ULT-DOAS observation presents a low amount
of NO
2
close to the stratospheric NO
2
SCD derived from OMI. This is due to the fact that zenith-sky
ULT-DOAS observations are performed above the NO
2
plume or above the planetary boundary layer.
144
Books
MDPI
Atmosphere 2017, 8,78
Figure 4.
Comparisons between measured SCD using various methods of determination and
stratospheric SCD derived from OMI (21 July 2014); (
A
) Comparison between SCD determined from
ground-based (GB) observations and stratospheric SCD derived from DOMINO; (
B
) Comparison
between SCD determined from CAR-DOAS (black dots) and ULT-DOAS (red dots) and stratospheric
SCD derived from DOMINO.
2.2.3. Radiative Transfer Calculation
In order to determine the VCD, the SCD retrieved with the DOAS method has to be converted
using an appropriate AMF. The geometric approximation for the airborne DOAS observations assumes
a simple reflection of the sunlight on the earth’s surface. In this case (neglecting the earth’s sphericity)
the nadir AMF can be described as a function of the solar zenith angle (SZA) as:
AMF
geo
= 1 + 1/ cos(SZA) (5)
Since the ULT-DOAS measurements were performed in the open atmosphere using scattered
sunlight radiation, the radiative transfer during the observations needs to be modeled to interpret
the retrieved data. In this work, the AMF was calculated using the radiative transfer model (RTM)
UVspec/DISORT [35], which is a fully spherical model. This RTM has been validated using six other
different codes [34].
The general assumptions made for the radiative transfer calculation using the RTM
UVspec/DISORT are introduced in Table 3.
Table 3. Input parameters used for the radiative transfer model (RTM) calculations.
Trace Gas Nitrogen Dioxide
wavelength 440 nm
flight altitude 1000 m 1500 m
albedo 0.1
visibility 1 km 5 km 20 km
line of sight 0
◦
Results of AMF simulations for the nadir flight performed on 13 August 2014, using the input
parameters presented in Table 3, are displayed in Figure 5. The geometric AMF for the nadir view is
also shown. The visibility parameter accounts for the effect of aerosols.
145
Books
MDPI
Atmosphere 2017, 8,78
Figure 5.
AMF simulations obtained from RTM calculations using UVspec/DISORT for various input
parameters, for 13 August 2014. AMF = Air Mass Factor.
3. Results and Discussions
The airborne DOAS observations were designed to determine the distribution of tropospheric
NO
2
from the South-East of Romania from urban, industrial and rural areas and associated flux.
The first flights were performed on 1 September 2011 and 25 August 2012, and aimed at measuring
the NO
2
around the industrial area of Galati city and from Braila city. The flight performed on
21 July 2014 aimed to measure the NO
2
emitted by a fertilizer factory located nearby Slobozia
city; unfortunately, during the DOAS flight the fertilizer factory was not operational. Figure 6
presents the horizontal distribution of the tropospheric NO
2
determined in nadir geometry for
1 September 2011, while Figure 7 depicts the results during a similar flight, but on 25 August
2012. During this experiment, the plume from the industrial platform was not fully crossed by the
optical instrument onboard the ULT. The wind was northerly resulting in local increases of the NO
2
amount detected by the spectrometer. The maximum tropospheric NO
2
VCD detected during this
experiment was (1.1
±
0.24)
×
10
16
molecules/cm
2
while the minimum tropospheric NO
2
VCD was
(2.1 ± 0.81) × 10
15
molecules/cm
2
.
Figure 6.
Map of tropospheric NO
2
VCD determined on 1 September 2011 using ULT-DOAS observations.
146
Books
MDPI
Atmosphere 2017, 8,78
Figure 7.
Map of tropospheric NO
2
VCD determined on 25 August 2012 using ULT-DOAS observations.
The trajectory of the flight on 13 August 2014 gave us the opportunity of calculating the NO
2
flux
emissions around the industrial area of Galati city. This was not possible for the other flights because
encircling the NO
2
source was not authorized.
On 1 September 2011, the NO
2
amount was low relative to the other day. The flight comprised
almost two complete circles around Braila; however, the NO
2
displayed no clear variation.
The horizontal distribution of NO
2
was quite homogenous over Braila city on this day.
A double experiment was performed on 13 August 2014, using both a ULT-DOAS and a car-based
DOAS system. The mobile ground-based DOAS observations were performed using the same
equipment during 9.75–10 UTC, while the ULT-DOAS observations were performed during
7.30–8.15 UTC.
Figure 8 shows the tropospheric NO
2
VCD derived along the trajectory of the ULT-DOAS
measurements. The right plot shows a photograph of the plume crossed by the ULT flights. During the
same day, approximately 1 h after the acquisition of the ULT-DOAS measurements, a zenith-sky
car-DOAS system was used to sample the NO
2
plume at the same location as the ULT-DOAS
observations. We assume that the quantity of the NO
2
emitted by the steel factory was almost constant
during the airborne and car-DOAS system.
Figure 8.
The tropospheric NO
2
VCD along the flight trajectory using the ULT-DOAS system on
13 August 2014 (left); Photography of the NO
2
plume determined on the same day (right).
147
Books
MDPI
Atmosphere 2017, 8,78
Figure 9 presents the NO
2
VCD derived from the nadir airborne DOAS observations performed
over the industrial area of Galati city compared with zenith-sky ground-based mobile DOAS
measurements performed over the same area in the same day (13 August 2014). In these figures,
we show the original SCDs (A) and the retrieved tropospheric NO
2
VCD (B). Figure 9A shows that
the DSCDs determined from the ULT-DOAS system are ~30% higher than the DSCDs determined
using car-DOAS observations. This difference is attributed to the different observation geometries.
After appropriate AMF calculation (see Section 2.2.2), both observations show close results.
The ULT flight above the industrial plume led to the detection of a maximum tropospheric NO
2
VCD of (1.41
±
0.27)
×
10
16
molecules/cm
2
while the car-DOAS observation shows a maximum
tropospheric NO
2
VCD of (1.36
±
0.21)
×
10
16
molecules/cm
2
. Figure 10 displays the correlation
between the tropospheric NO
2
VCD retrieved by the ULT-DOAS and the car-DOAS instrument, where
closest spatially coincident data were selected. A Pearson correlation coefficient R = 0.71 was obtained
between ground mobile DOAS observations and airborne DOAS measurements.
Figure 9.
Comparisons between ULT-DOAS and car-DOAS observations performed on 13 August 2014.
(
A
) The results of the preliminary DOAS analysis (DSCDs) and (
B
) after determination of the vertical
columns (VCDs).
Figure 10.
Correlation between tropospheric VCDs measured by the ULT-DOAS and the car-DOAS
instrument on 13 August 2014.
148
Books
MDPI
Atmosphere 2017, 8,78
NO
2
Flux Calculation
The NO
2
flux above the industrial platform located nearby Galati city was calculated by
performing upwind and downwind measurements around the point source using ULT-DOAS
observations on 13 August 2014. The calculation of the emission flux is based on the following
parameters: the NO
2
VCD determined from the transect over the plume, the wind speed and the
wind factor correction, taking into account the angle between the flight direction and wind direction
(Equation (6)), [20,36,37]:
Flux
NO
2
=
∑
i
VCD
NO
2
(s
i
) · v · sin (α
i
) · Δs
i
(6)
where VCD
NO2
is the NO
2
tropospheric vertical column, v is the wind speed,
α
the angle between
wind direction and driving route, i is the observations index, and
Δ
s
i
is the distance between
two successive spectra.
The wind data used for the NO
2
flux calculation rely on measurements of the automatic weather
station (Davis Vantage Pro2) located in the campus university of Galati city (45.44
◦
N, 28.05
◦
E),
while vertical wind profiles come from the Hybrid Single Particle Lagrangian Integrated Trajectory
Model (HYSPLIT) [
38
] model using archived dataset GDAS 0.5
◦
×
0.5
◦
. The weather station is located
at 30 m height and acquires data every 30 min. Since no atmospheric sounding was possible during the
experiments, the wind measured on the ground was scaled to the output of HYSPLIT model simulation
at 1000 m altitude. During the period of the ULT-DOAS measurements, the mean NO
2
emission flux
was determined to be (3.45
±
0.89)
×
10
−3
kg/s. A local environmental report indicates ~600 tons/year
NO
x
emissions emitted by the steel factory [
39
]. Using a Leighton ratio (L = [NO]/[NO
2
]) of 0.3,
we calculated that the steel factory has emitted a mean of ~10
×
10
−3
kg/s of NO
2
. The difference
between the two types of estimation may be attributed to the fact that the NO
x
emissions from the steel
factory are dependent on the quantity of the steel produced, which can vary from one day to another.
Also, the derived NO
2
fluxes must be dealt with some care because of the probably incomplete NO to
NO
2
conversion [40].
4. Conclusions
Ultralight-trike DOAS observations were performed in the South-East of Romania during four
days between 2011 and 2014. The first two flights were focused over Braila city, the third aimed at
measurements of the NO
2
plume emitted by a fertilizer factory near Slobozia city. Unfortunately,
during the DOAS flight the fertilizer factory was not operational. The last flight, performed on
13 August 2014, was focused over the industrial area of Galati city. Nadir observations were performed
around the industrial platform of Galati city aiming at measuring the tropospheric NO
2
VCD around
the source and at evaluating the associated NO
2
flux. To retrieve the tropospheric NO
2
VCD from
ULT-DOAS observations, complementary ground- and space-based measurements were used.
We showed that the tropospheric NO
2
VCD deduced from the ULT-DOAS observations are
consistent with measurements performed from the ground using a zenith-sky car-DOAS system.
Although two hours separated the two experiments, a correlation coefficient of R = 0.71 was found
between the two results, a tropospheric NO
2
VCD of (1.41
±
0.27)
×
10
16
molecules/cm
2
and an
estimated associated flux of (3.45
±
0.89)
×
10
−3
kg/s was measured close to the industrial area of
Galati city on 13 August 2014, the only day that it was possible to determine the NO
2
flux.
Also, we showed that the stratospheric SCD derived from ground-based and airborne
measurements correlates well with stratospheric NO
2
derived from observations by the OMI
satellite sensor.
Based on this study, we conclude that the ULT is an efficient tool which allows determining
with a high resolution the NO
2
distribution around urban or industrial sources. Also, the ULT-DOAS
system is a very efficient tool for measuring fluxes due to its flexibility during the flights and the
149
Books
MDPI
Atmosphere 2017, 8,78
possibility to access remote areas. The ULT-DOAS system might also constitute a promising tool for
satellite validation and calibration under clear-sky conditions, especially for upcoming high-resolution
sensors such as the TROPOMI/Sentinel-5 Precursor instrument to be launched in summer 2017.
Acknowledgments:
The work of D.E. Constantin was supported by Project PN-II-RU-TE-2014-4-2584, a grant of
the Romanian National Authority for Scientific Research and Innovation, CNCS-UEFISCDI. The DOMINO data
product was taken from the ESA TEMIS archive (www.temis.nl) maintained at KNMI, The Netherlands.
Author Contributions:
Daniel-Eduard Constantin and Alexis Merlaud conceived and designed the study;
Daniel-Eduard Constantin analyzed the data; and Mirela Voiculescu, Carmelia Dragomir, Lucian Georgescu,
Gaia Pinardi, Francois Hendrick and Michel Van Roozendael improved the paper.
Conflicts of Interest: The authors declare no conflict of interest.
References
1.
Solomon, S.; Portmann, R.W.; Sanders, R.W.; Daniel, J.S.; Madsen, W.; Bartram, B.; Dutton, E.G. On the role
of nitrogen dioxide in the absorption of solar radiation. J. Geophys. Res. 1999, 104, 12047–12058. [CrossRef]
2. Platt, U. Differential optical absorption spectroscopy (DOAS). Chem. Anal. Ser. 1994, 127, 27–83.
3.
Brewer, A.W.; McElroy, C.T.; Kerr, J.B. Nitrogen dioxide concentration in the atmosphere. Nature
1973
, 246,
129–133. [CrossRef]
4.
Noxon, J.F. Nitrogen dioxide in the stratosphere and troposphere measured by ground-based absorption
spectroscopy. Science 1975, 189, 547–549. [CrossRef] [PubMed]
5.
Hendrick, F.; Müller, J.-F.; Clémer, K.; Wang, P.; De Mazière, M.; Fayt, C.; Gielen, C.; Hermans, C.; Ma, J.Z.;
Pinardi, G.; et al. Four years of ground-based MAX-DOAS observations of HONO and NO
2
in the Beijing area.
Atmos. Chem. Phys. 2014, 14, 765–781. [CrossRef]
6.
Constantin, D.-E.; Merlaud, A.; van Roozendael, M.; Voiculescu, M.; Fayt, C.; Hendrick, F.; Pinardi, G.;
Georgescu, L. Measurements of Tropospheric NO
2
in Romania Using a Zenith-Sky Mobile DOAS System
and Comparisons with Satellite Observations. Sensors 2013, 13, 3922–3940. [CrossRef]
7.
Dragomir, C.; Constantin, D.-E.; Voiculescu, M.; Georgescu, L.; Merlaud, A.; Roozendael, M.V. Modeling
results of atmospheric dispersion of NO
2
in an urban area using METI–LIS and comparison with coincident
mobile DOAS measurements. Atmos. Pollut. Res. 2015, 6, 503–510. [CrossRef]
8.
Merlaud, A.; van Roozendael, M.; van Gent, J.; Fayt, C.; Maes, J.; Toledo-Fuentes, X.; Ronveaux, O.;
de Mazière, M. DOAS measurements of NO
2
from an ultralight aircraft during the Earth Challenge
expedition. Atmos. Meas. Tech. 2012, 5, 2057–2068. [CrossRef]
9.
Meier, A.C.; Schönhardt, A.; Bösch, T.; Richter, A.; Seyler, A.; Ruhtz, T.; Constantin, D.-E.; Shaiganfar, R.;
Wagner, T.; Merlaud, A.; et al. High-resolution airborne imaging DOAS-measurements of NO
2
above
Bucharest during AROMAT. Atmos. Meas. Tech. Discuss. 2016. [CrossRef]
10.
Tack, F.; Merlaud, A.; Iordache, M.-D.; Danckaert, T.; Yu, H.; Fayt, C.; Meuleman, K.; Deutsch, F.; Fierens, F.;
van Roozendael, M. High resolution mapping of the NO
2
spatial distribution over Belgian urban areas based
on airborne APEX remote sensing. Atmos. Meas. Tech. Discuss. 2017. [CrossRef]
11.
Merlaud, A.; Constantin, D.; Fayt, C.; Maes, J.; Mingireanu, F.; Mocanu, I.; Georgescu, L.; Roozendael, M.
Small whiskbroom imager for atmospheric composition monitoring (SWING) from an unmanned areal
vehicle (UAV). In Proceedings of the 21st ESA Symposium on European Rocket and Balloon Programmes
and related Research, Thun, Switzerland, 9–13 June 2014; pp. 1–7.
12.
Bovensmann, H.; Burrows, J.P.; Buchwitz, M.; Frerick, J.; Nöel, S.; Rozanov, V.V.; Chance, K.V.; Goede, A.H.P.
SCIAMACHY—Mission objectives and measurement modes. J. Atmos. Sci. 1999, 56, 127–150. [CrossRef]
13.
Levelt, P.; van den Oord, G.; Dobber, M.; Malkki, A.; Visser, H.; de Vries, J.; Stammes, P.; Lundell, J.; Saari, H.
The ozone monitoring instrument. IEEE T. Geosci. Remote. 2006, 44, 1093–1101. [CrossRef]
14.
Munro, R.; Eisinger, M.; Anderson, C.; Callies, J.; Corpaccioli, E.; Lang, R.; Lefebvre, A.; Livschitz, Y.;
Albinana, A.P. GOME-2 on MetOp. In Proceedings of the 2006 EUMETSAT Meteorological Satellite
Conference, Helsinki, Finland, 12–16 June 2006; p. 48.
15.
van der A, R.J.; Eskes, H.J.; Boersma, K.F.; van Noije, T.P.C.; van Roozendael, M.; de Smedt, I.;
Peters, D.H.M.U.; Meijer, E.W. Trends, seasonal variability and dominant NO
x
source derived from a ten year
record of NO
2
measured from space. J. Geophys. Res. 2008, 113, D04302. [CrossRef]
150
Books
MDPI
Atmosphere 2017, 8,78
16.
Varotsos, C.; Christodoulakis, J.; Tzanis, C.; Cracknell, A.P. Signature of tropospheric ozone and nitrogen
dioxide from space: A case study for Athens, Greece. Atmos. Environ. 2014, 89, 721–730. [CrossRef]
17.
Junkermann, W. An ultralight aircraft as platform for research in the lower troposphere: System performance
and first results from radiation transfer studies in stratiform aerosol layers and broken cloud conditions.
J. Atmos. Ocean. Technol. 2001, 18, 934. [CrossRef]
18.
Junkermann, W. On the distribution of formaldehyde in the western Po-Valley, Italy, during FORMAT
2002/2003. Atmos. Chem. Phys. 2009, 9, 9187–9196. [CrossRef]
19.
Chazette, P.; Sanak, J.; Dulac, F. New Approach for Aerosol Profiling with a Lidar Onboard an Ultralight
Aircraft: Application to the African Monsoon Multidisciplinary Analysis. Environ. Sci. Technol.
2007
, 41,
8335–8341. [CrossRef] [PubMed]
20.
Wang, P.; Richter, A.; Bruns, M.; Burrows, J.P.; Scheele, R.; Junkermann, W.; Heue, K.-P.; Wagner, T.; Platt, U.;
Pundt, I. Airborne multi-axis DOAS measurements of tropospheric SO
2
plumes in the Po-valley, Italy.
Atmos. Chem. Phys. 2006, 6, 329–338. [CrossRef]
21.
Grutter, M.; Basaldud, R.; Rivera, C.; Harig, R.; Junkerman, W.; Caetano, E.; Delgado-Granados, H.
SO
2
emissions from Popocatépetl volcano: Emission rates and plume imaging using optical remote sensing
techniques. Atmos. Chem. Phys. 2008, 8, 6655–6663. [CrossRef]
22.
General, S.; Pöhler, D.; Sihler, H.; Bobrowski, N.; Frieß, U.; Zielcke, J.; Horbanski, M.; Shepson, P.B.;
Stirm, B.H.; Simpson, W.R.; et al. The Heidelberg Airborne Imaging DOAS Instrument (HAIDI)—A novel
imaging DOAS device for 2-D and 3-D imaging of trace gases and aerosols. Atmos. Meas. Tech.
2014
, 7,
3459–3485. [CrossRef]
23.
Liu, L.; Flatøy, F.; Ordóñez, C.; Braathen, G.O.; Hak, C.; Junkermann, W.; Andreani-Aksoyoglu, S.;
Mellqvist, J.; Galle, B.; Prévôt, A.S.H.; et al. Photochemical modelling in the Po basin with focus on
formaldehyde and ozone. Atmos. Chem. Phys. 2007, 7, 121–137. [CrossRef]
24.
Airborne Romanian Measurements of Aerosols and Trace Gases (AROMAT). Available online: http://uv-
vis.aeronomie.be/aromat/index.php (accessed on 15 January 2017).
25.
Danckaert, T.; Fayt, C.; van Roozendael, M.; de Smedt, I.; Letocart, V.; Merlaud, A.; Pinardi, G.
QDOAS Software user manual Version 2.111-April 2016, UV-Visible DOAS Research Group of the Royal
Belgian Institute for Space Aeronomy Web Site. Available online: http://uv-vis.aeronomie.be/software/
QDOAS/QDOAS_manual.png (accessed on 15 December 2016).
26.
Vandaele, A.; Hermans, C.; Simon, P.; Carleer, M.; Colin, R.; Fally, S.; Mérienne, F.; Jenouvrier, A.; Coquart, B.
Measurements of the NO
2
absorption cross-section from 42000 cm
−1
to 10000 cm
−1
(238–1000 nm) at 220 K
and 294 K (220 K). J. Quant. Spectrosc. Radiat. Transf. 1998, 59, 171–184. [CrossRef]
27.
Bogumil, K.; Orphal, J.; Homann, T.; Voigt, S.; Spietz, P.; Fleischmann, O.C.; Vogel, A.; Hartmann, M.;
Kromminga, H.; Bovensmann, H.; et al. Measurements of molecular absorption spectra with the
SCIAMACHY pre-flight model: Instrument characterization and reference data for atmospheric
remote-sensing in the 230–2380 nm region. J. Photochem. Photobiol. A Chem. 2003, 157, 167–184. [CrossRef]
28.
Thalman, R.; Volkamer, R. Temperature dependent absorption cross-sections of O
2
-O
2
collision pairs between
340 and 630 nm and at atmospherically relevant pressure. Phys. Chem. Chem. Phys.
2013
, 15, 15371–15381.
[CrossRef] [PubMed]
29.
Chance, K.V.; Spurr, R.J.D. Ring effect studies: Rayleigh scattering, including molecular parameters for
rotational Raman scattering, and the Fraunhofer spectrum. Appl. Opt.
1997
, 36, 5224–5230. [CrossRef]
[PubMed]
30.
Baidar, S.; Oetjen, H.; Coburn, S.; Dix, B.; Ortega, I.; Sinreich, R.; Volkamer, R. The CU Airborne MAX-DOAS
instrument: Vertical profiling of aerosol extinction and trace gases. Atmos. Meas. Tech.
2013
, 6, 719–739.
[CrossRef]
31.
Vaughan, G.; Quinn, P.T.; Green, A.C.; Bean, J.; Roscoe, H.K.; van Roozendael, M.; Goutail, F. SAOZ
measurements of stratospheric NO
2
at Aberystwyth. J. Environ. Monit.
2006
, 8, 353–361. [CrossRef] [PubMed]
32.
Dirksen, R.J.; Boersma, K.F.; Eskes, H.J.; Ionov, D.V.; Bucsela, E.J.; Levelt, P.F.; Kelder, H.M. Evaluation of
stratospheric NO
2
retrieved from the Ozone Monitoring Instrument: Intercomparison, diurnal cycle and
trending. J. Geophys. Res. 2011, 116, D08305. [CrossRef]
151
Books
MDPI
Atmosphere 2017, 8,78
33.
Hendrick, F.; Barret, B.; van Roozendael, M.; Boesch, H.; Butz, A.; De Mazière, M.; Goutail, F.;
Hermans, C.; Lambert, J.-C.; Pfeilsticker, K.; et al. Retrieval of nitrogen dioxide stratospheric profiles
from ground-based zenith-sky UV-visible observations: Validation of the technique through correlative
comparisons. Atmos. Chem. Phys. 2004, 4, 2091–2106. [CrossRef]
34.
Hendrick, F.; van Roozendael, M.; Kylling, A.; Petritoli, A.; Rozanov, A.; Sanghavi, S.; Schofield, R.;
von Friedeburg, C.; Wagner, T.; Wittrock, F.; et al. Intercomparison exercise between different radiative
transfer models used for the interpretation of ground-based zenith-sky and multi-axis DOAS observations.
Atmos. Chem. Phys. 2006, 6, 93–108. [CrossRef]
35.
Mayer, B.; Kylling, A. Technical note: The libRadtran software package for radiative transfer
calculations—Description and examples of use. Atmos. Chem. Phys. 2005, 5, 1855–1877. [CrossRef]
36.
Rivera, C.; Sosa, G.; Wöhrnschimmel, H.; de Foy, B.; Johansson, M.; Galle, B. Tula industrial complex (Mexico)
emissions of SO
2
and NO
2
during the MCMA 2006 field campaign using a mobile mini-DOAS system.
Atmos. Chem. Phys. 2009, 9, 6351–6361. [CrossRef]
37.
Johansson, M.; Rivera, C.; de Foy, B.; Lei, W.; Song, J.; Zhang, Y.; Galle, B.; Molina, L. Mobile mini-DOAS
measurement of the outflow of NO
2
and HCHO from Mexico City. Atmos. Chem. Phys.
2009
, 9, 5647–5653.
[CrossRef]
38.
Stein, A.F.; Draxler, R.R.; Rolph, G.D.; Stunder, B.J.B.; Cohen, M.D.; Ngan, F. NOAA‘s HYSPLIT atmospheric
transport and dispersion modeling system. Bull. Am. Meteorol. Soc. 2015, 96, 2059–2077. [CrossRef]
39. Studiu Privind Calitatea Aerului in Municipiul Galati; Municipiul Galati: Galati, Romania, 2016.
40.
Frins, E.; Bobrowski, N.; Osorio, M.; Casaballe, N.; Belsterli, G.; Wagner, T.; Platt, U. Scanning and mobile
multi-axis DOAS measurements of SO
2
and NO
2
emissions from an electric power plant in Montevideo,
Uruguay. Atmos. Environ. 2014, 98, 347–356. [CrossRef]
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
152
Books
MDPI
atmosphere
Article
Evaluation of Analysis by Cross-Validation.
Part I: Using Verification Metrics
Richard Ménard *and Martin Deshaies-Jacques
Air Quality Research Division, Environment and Climate Change Canada, 2121 Transcanada Highway,
Dorval, QC H9P 1J3, Canada; martin.deshaies-jacques@canada.ca
* Correspondence: richard.menard@canada.ca; Tel.: +1-514-421-4613
Received: 5 September 2017; Accepted: 24 February 2018; Published: 27 February 2018
Abstract:
We examine how passive and active observations are useful to evaluate an air quality analysis.
By leaving out observations from the analysis, we form passive observations, and the observations
used in the analysis are called active observations. We evaluated the surface air quality analysis of O
3
and PM
2.5
against passive and active observations using standard model verification metrics such
as bias, fractional bias, fraction of correct within a factor of 2, correlation and variance. The results
show that verification of analyses against active observations always give an overestimation of the
correlation and an underestimation of the variance. Evaluation against passive or any independent
observations display a minimum of variance and maximum of correlation as we vary the observation
weight, thus providing a mean to obtain the optimal observation weight. For the time and dates
considered, the correlation between (independent) observations and the model is 0.55 for O
3
and 0.3
for PM
2.5
and for the analysis, with optimal observation weight, increases to 0.74 for O
3
and 0.54 for
PM
2.5
. We show that bias can be a misleading measure of evaluation and recommend the use of a
fractional bias such as the modified normalized mean bias (MNMB). An evaluation of the model bias
and variance as a function of model values also show a clear linear dependence with the model values
for both O
3
and PM
2.5
.
Keywords: chemical data assimilation; air quality model diagnostics; cross-validation
1. Introduction
Since 2003, Environment and Climate Change Canada (ECCC) has been producing hourly surface
analyses of pollutants covering North America [
1
,
2
] which became operational products in February
2013 [
3
]. The analyses are produced using an optimum interpolation scheme that combines the operational
air quality forecast model GEM-MACH output [
4
] (CHRONOS model output was used prior to 2010 [
5
])
with real-time hourly observations of O
3
,PM
2.5
,PM
10
,NO
2
, and SO
2
from the AirNow gateway with
additional observations from Canada. As those surface analyses are not used to initialize an air quality
model, it raises the issue on how to evaluate them. We conduct routine evaluations using the same set of
observations as those used to produce the analysis. Once in a while, when there is a change in the system,
a more thorough evaluation is conducted where we leave out a certain fraction of the observations and use
them as independent observations, a process known as cross-validation. Observations used in producing
the analysis are called active observations while those not used for evaluation are passive observations.
Cross-validation is used to validate any model that depends on data. In air quality applications it has
been used, for example, for mapping and exposure models [
6
–
8
]. The purpose of this two-part paper
is to examine the relative merit of using active or passive observations (or independent observations in
general) viewed from different evaluation metrics, but also to develop, in the second part, a mathematical
framework to estimate the analysis error, and in doing so, to improve the analysis.
The evaluation of an analysis is important, even in the case where it is used to initialize an air
quality forecast model, since the evaluation of the resulting air quality forecast may not be a good
Atmosphere 2018, 9, 86 153 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2018, 9,86
measure of the quality of the analysis. In air quality forecasting, the forecast error growth is small,
depicts little sensitivity to initial conditions and is in fact more sensitive to numerous modeling errors
such as: photochemistry, clouds, meteorology, boundary conditions and emissions just to name a
few [
9
–
13
]. Furthermore, chemical species that are observed are incomplete compared to species
needed to initialize an air quality model, incomplete in terms of the number of species observed
as well as in their kind [
9
,
11
,
13
,
14
]. Only a fraction of the observed species (either of secondary
or primary pollutants) are usable for data assimilation; important chemical mechanisms are left
completely unobserved and for aerosols, information on size distribution is quite limited and almost
nonexistent when it comes to speciation [
9
,
13
]. In addition, the observational coverage is limited
to the surface or to total column measurements which, up until now, were available at one or two
local times per day. There are thus many assumptions to be made from an analysis to a proper 3D
initial chemical condition and surface emission correction and its subsequent impact on the air quality
forecast. These considerations warrant an independent evaluation of the quality of the analysis on its
own [15].
Evaluating an analysis with observations is quite different from evaluating a model with observations,
since analyses are created from observations. From a statistical point of view, the observation and
analysis cannot be considered independent. However, let us assume that observation errors are spatially
uncorrelated. Then, since the passive and active observation sites are never collocated, then the errors
from passive observations are uncorrelated with errors of active observations (i.e., observations that are
used for the analysis). Furthermore, since the modelling errors are usually assumed to be uncorrelated
with observation errors, then it is also uncorrelated with the analysis errors. Cross-validation thus offers a
means to evaluate analyses with statistically independent (passive) observations [16].
In part one of this paper, we evaluate the relative merit of passive and active observations in the
evaluation of analyses using standard metrics used for model evaluation. We show how and when the
use of active observations can be misleading and that passive observations can provide a means to
identify optimal analyses. Our examples show that optimal analyses, at the independent observation
sites, have much smaller biases than the model biases and increase the correlation coefficient by nearly
a factor of 2.
The paper is thus organized as follows. First we present the analysis scheme we will be using,
as well as the cross-validation design, the evaluation metrics and the configuration of the experiments.
Then in Section 3, we assess the quality of the analyses in both active and passive observation spaces
using standard air quality evaluation metrics, identify some pitfalls of some metrics and advocate
using active observations. Conclusions are presented in Section 4.
2. Experimental Design
2.1. Design of the Objective Analysis Solver
In optimum interpolation there is no use of an explicit interpolation observation operator.
The correlation between a pair of locations, either from two observation sites or from an observation
site to a model grid point, is computed as a function of distance using a prescribed correlation function.
The observation operator is in effect a delta function applied over a continuous spatial domain [17].
In this study we interpolate the gridded analysis field to observation locations, using bilinear
interpolation, to compute residuals such as observation-minus-analysis. Thus there can be a discrepancy
between the observation operator used to generate the analysis, i.e., delta functions, and the observation
operator used to interpolate the analysis field at the observation location, i.e., bilinear interpolation.
To eliminate this discrepancy in observation operators we have revised the optimum interpolation
scheme to use explicitly the same bilinear interpolation in handling the error covariance. We will give
details below.
As in the operational optimum interpolation, the inversion of the innovation covariance matrix for
the analysis solver is done using Choleski decomposition on the full matrix. The number of observations
154
Books
MDPI
Atmosphere 2018, 9,86
to be processed per analysis being of the order of a thousand or less, there was no need for computational
simplification for large number of observations by using either data selection [
18
] or compact support
correlation functions [
17
,
19
]. Thus, the analysis scheme used in this study computes explicitly the gain
matrix
K as,
K
=
BH
T
(H
BH
T
+
R
)
−1
(1)
where
H
is a bilinear interpolation operator,
B
is the prescribed background error covariance and
R
is the prescribed observation error covariance. The tilde (
∼
·
) emphasizes that these are prescribed,
potentially suboptimal, quantities.
The computational demand of the Kalman gain was kept low by computing the background error
correlation function only at model grid points needed for the bilinear interpolation. For example,
to calculate the correlation between a pair of observations requires the computation of correlation
between four points surrounding observation 1 (needed for the bilinear interpolation) and the other
four points surrounding observation 2, thus forming a 4
×
4 correlation matrix
C
between the target
model grid points. Then we calculate
HCH
T
which gives the correlation between two observation sites.
This procedure is generalized for the
N
observations needed for the analysis. Equation (1) also involves
the computation of
BH
T
that we compute as a set of
N
representers (i.e., columns of
BH
T
), each being
a 2D field that maps the background error covariance in model space with a single observation
location, using again the bilinear interpolation approach to get a single interpolated representor for
each observation location. By doing so we keep the consistency between the observation operators
used for interpolation of a field and the observation operator used to manipulate matrices.
2.2. Cross-Validation
Cross-validation is a technique to evaluate an analysis (or in general any model that depends
on observations) by partitioning the original observation data set into a training set, used to create
the analysis, and an independent (or passive) set, used to evaluate the analysis. The most common
cross-validation designs are: the k-fold cross-validation, where the original observation data set is
partitioned into k equal size subsamples and the leave-one-out cross-validation, where N subsamples
are created, each with one different observation set aside for the evaluation while the other
N −
1
observations are used in producing the analysis. The cross-validation is then repeated with all the
different sets until all observations have been used for evaluation. Clearly, there are k analyses
computed in the k-fold cross-validation and N in the leave-one-out cross-validation, which is being
computationally demanding when N is large. The main disadvantage of the k-fold cross-validation is
that the analyses being evaluated uses a smaller number of observations (actually
(k −
1
)N/k
) than
the original observation data set, whereas the leave-one-out cross-validation evaluates analysis that
uses nearly the same number of observations (actually
N −
1) as the original observation data set.
This actually matters with the k-fold cross-validation if we need an estimate of the analysis error
variance (or any other second moments) as the analysis error variance depends on the number of
observations used.
Let
O
j
be a vector that contains the jth set of observations used for evaluation, and let
A
(j)
be
a vector of analysis value interpolated at the verification observation locations of
O
j
and where the
analysis used all observations except those in
O
j
(the index in parenthesis, i.e.,
(j)
, indicates all sets
except the set j). It is customary in cross-validation literature (e.g., [
20
]) to construct a mean square
error cost function, often denoted by CV,
CV
=
∑
j
(O
j
− A
(j)
)
T
(O
j
− A
(j)
) (2)
that represents a misfit quadratic error of the model
A
- in our case the analysis. Different model
A
can be compared and selected from which the CV value is smallest. Likewise, a tunable parameter
in
A
can be obtained by minimizing the cost function CV with respect to that parameter. As we shall
155
Books
MDPI
Atmosphere 2018, 9,86
discuss later in this paper, in Section 4 and onwards, the bias of
(O
j
− A
(j)
)
needs to be removed from
the cost function in order to estimate the input error covariance parameters.
In applications and thus in all experiments that follow, the analyses and verification against
passive observations are made only with a set of observations that have passed a quality control.
The quality control is nearly identical to the quality control used for the operational implementation
of the analysis of surface pollutants at ECCC (see supplementary material in Robichaud et al. [
3
]).
It consists in discarding observations that report a negative value, or whose value exceeds a certain
unrealistic threshold set to 300 ppbv for ozone (300
μ
g/m
3
for PM
2.5
). Observations are also discarded
based on innovations (or observed-minus-background values) when, for ozone, they exceed 50 ppbv
(100
μ
g/m
3
for PM
2.5
) in absolute value. The quality-controlled observations are then separated into
three sets of observations of equal numbers, i.e., a 3-fold cross-validation procedure, as illustrated in
Figure 1.
Figure 1.
Spatial distribution of the three subsets of PM
2.5
observations used for cross-validation.
The selection algorithm is based on regular picking of station by ID number.
The selection into three sets is made by station ID number, selecting on a regular basis each fourth
station, starting with station 1 for the first set, station 2 for the second set and station 3 for the third
set, and resulting in locally spatially random distribution of each sets of stations. The cross-validation
is then made by leaving one set out of the three sets, and using the remaining two sets to produce
the analysis.
2.3. Verification Metrics
We will evaluate the analyses against passive and active observations with the following standard
evaluation metrics used for air quality models [
21
–
24
]; the bias, the modified normalized mean
bias (MNMB), the fraction of correct within a factor of 2 (FC2), the variance (
var(O − A)
) and the
correlation coefficient (
cor(O
,
A)
), where the statistics is computed over time t for each station, and then
the resulting metric is averaged over all the verifying station i,
bias
=
1
N
i
∑
i
1
N
k
∑
k
(O
i
(t
k
) − A
i
(t
k
))
(3)
MNMB
=
1
N
i
∑
i
2
N
k
∑
k
O
i
(t
k
) − A
i
(t
k
)
O
i
(t
k
)+A
i
(t
k
)
(4)
156
Books
MDPI
Atmosphere 2018, 9,86
FC2 =
1
N
i
∑
i
1
N
k
count
0.5 ≤
A
i
(t
k
)
O
i
(t
k
)
≤
2
(5)
var
(O − A)=
1
N
i
∑
i
1
N
k
− 1
∑
k
[(O
i
(t
k
) − A
i
(t
k
)) − (O
i
− A
i
)]
2
(6)
cor
(O, A)=
1
N
i
∑
i
⎧
⎪
⎪
⎨
⎪
⎪
⎩
1
N
k
− 1
∑
k
(O
i
(t
k
) − O
i
)( A
i
(t
k
) − A
i
)
∑
k
(O
i
(t
k
) − O
i
)
2
∑
k
(A
i
(t
k
) − A
i
)
2
⎫
⎪
⎪
⎬
⎪
⎪
⎭
(7)
where
O
i
(t
k
)
is the observed value at time
t
k
at the station i,
A
i
(t
k
)
is the analysis at time
t
k
interpolated
at the location of the station i,
N
k
is the total number of time sample per station,
N
s
is the total number
of stations (in the sample or over the domain), and the overbar
()
denotes the time average. The bias
and the MNMB are metrics of the first moment that have distinctive properties. The bias gives a
representative measure of the systematic discrepancy between analyzed and observed values over
the whole set of observations used for verification. However, since atmospheric constituents exhibit
a range of values that can vary in time and space, and different constituents have different range of
values and may as well be expressed with different units, a relative error measure such as the MNMB is
often preferred [
24
]. The MNMB is a dimensionless quantity that falls in the range
[−
2,
+
2
]
. The factor
of 2 is introduced so to give a % error interpretation to the MNMB. This metric has the additional
advantage of treating over- and under-estimation in a symmetric way [
24
]. However, the MNMB is
relatively insensitive to relatively large discrepancies between analysis (or model) values and observed
values, that is when its values are close to +2 (200%) or −2(−200%) [23].
The fraction of correction within a factor of 2 (FC2) is a measure of reliability. It is based on counts
and has the distinctive advantage that it is insensitive to outliers. It is worth mentioning that it accounts
both high values outliers and also low values outliers that is a unique property of this metric [
22
].
The FC2 metric is also symmetric with respect to permutation of A and O, it is also dimensionless
and its values must lie between 0 and 1. Our experience with this metric indicates that it is relatively
insentive for relatively good agreement between analysis and observed values.
The variance,
var(O − A )
, and the correlation coefficient
cor(O
,
A)
are metrics that depend on the
spread of the discrepancy between analysis and observed values. The variance is not a dimensionless
metric. It gives a representative measure of the spread of the discrepancy between analyses and
observations and is not sensitive to systematic errors. As we will show in Section 4 and also shown in
Marseille et al. [
16
],
var(O − A)
with passive observations has the distinct advantage of providing a
measure of the true analysis error variance (i.e., the error with respect to the truth) and
var(O − A)
can
be considered as a cross-validation cost function CV, Equation (2), with debiased (
O − A
) increments.
As for any second moment metric,
var(O − A)
is sensitive to outliers; they must be removed, and this
is done by gross check of the
(O − B)
, as explained in the previous subsection Section 2.2. Finally,
the correlation coefficient is a dimensionless quantity that lies in the range [
−
1, +1]. It is also invariant
to shifts in the mean (i.e., not sensitive to systematic errors), and multiplicative rescaling of either
analysis or observations. The correlation is also relatively insensitive to improvement when the
correlation is close to 1 or −1.
2.4. Description of the Ensemble of Analyses and Their Verification Statistics
A series of hourly analyses of O
3
and PM
2.5
at 21 UTC for a period of 60 days (14 June to
12 August 2014) were performed with given input error statistics using the operational model
GEM-MACH and the real-time AirNow observations as described in the introduction and with quality
controlled observations (see Section 2.2 above). In all experiments, the observation and background
error variances,
σ
2
o
and
σ
2
b
, used in the analysis are uniform. The prescribed observation error and
background error covariances are given as
R
= σ
2
o
I
,
B
= σ
2
b
C
, where the correlation model
C
is a
homogeneous isotropic second-order autoregressive model with a correlation length obtained by
157
Books
MDPI
Atmosphere 2018, 9,86
maximum likelihood, as in Ménard et al. [
17
]. Note that aside from quality control, that ends up
rejecting some observations, the analysis uses the observation values and model realizations as is,
with no bias correction.
We repeat the series of 60 day analyses for different observation and background error variances
chosen in such a way that their sum
σ
2
o
+ σ
2
b
is equal to
var(O − B)
but with different ratios of error
variances
γ = σ
2
o
/σ
2
b
. We perform the series of analyses over a wide range of
γ
ratios in the interval
[
10
−2
,10
2
]
, thus creating on one end analyses with very large observation weights, i.e.,
γ
1, such that
the analysis interpolated at the active observation sites tend to match the observed value, and on the
other end, with
γ
1, creating analyses with very small observation weight producing analyses that
are very close to the background (model) state.
The condition
σ
2
o
+ σ
2
b
= var(O − B)
, called the innovation variance consistency, is an important
constraint that is useful for the estimation of the true error statistics [
25
]. Indeed, the stronger condition
for the full covariance matrices, the innovation covariance consistency criterion, takes the form:
< (O − B)
(
O − B)
T
>=
R
+ H
BH
T
, where
<>
represents the mean over an ensemble of realizations,
H
is the
interpolation from model grid to the observation location (or observation operator). It is one of the
two necessary and sufficient condition to obtain the true error covariance statistics (in observation
space) [25,26].
As explained in Section 2.3 above, the verification metrics are first calculated over 60 days for a
given hour and for each station. Then the metric is averaged over all the verifying stations, resulting in
one value of the metric for each hour of the day. Here, however, we computed the metrics for 21 UTC
only. If
N
s
is the total number of stations, the statistics over one of the 3-fold subset then involves
an average of the metric over
N
s
/
3 passive stations. Doing this for all three subsets, and taking the
average of the subsets’ results, is equivalent to taking the average of the metric over all stations. In the
results that will be presented in the following sections, we always present the average metric over
the three passive subsets so that, in the end, the sample size of the passive observation experiments
and of the active observation experiments are equal and thus can be presented side by side on the
same graphic.
3. Verification against Passive and Active Observations
In this series of experiments, analyses of O
3
and PM
2.5
were produced using a fixed homogeneous
isotropic correlation function, where the correlation length was obtained by maximum likelihood using a
second-order auto-regressive model and error variances computed using a local Hollingsworth-Lönnberg
fit [
17
]. A correlation length of 124 km was obtained for O
3
and of 196 km for PM
2.5
. Our correlation length
is defined from the curvature at the origin as in Daley [
27
] and is different from the length-scale parameter
of the correlation model (see Ménard et al. [
17
] for a discussion of these issues). We did a series of
60-days analyses for different values of
σ
2
o
and
σ
2
b
but such that their sum respects the innovation variance
consistency,
σ
2
o
+ σ
2
b
= var(O − B)
, an important condition for an optimal analysis [
25
], as explained in
Section 2.4. This is the experimental procedure that has been used to generate the Figures 2–7. The results
are shown for a wide range of variance ratios
γ = σ
2
o
/σ
2
b
from 10
−2
to 10
2
in Figures 2–5 and 7 in particular.
Note that
γ
1 corresponds to a very large observation weight while
γ
1 correspond to very small
observation weight.
The
var(O − A)
using passive observations (red curve with circles) and active observations (black
curve with squares) is presented in Figure 2 for O
3
(left panel) and PM
2.5
(right panel). The solid
blue line represents
var(O − B)
, the variance of observation-minus-model, i.e., prior to an analysis.
As mentioned in Section 2.4, in the cross-validation experiments we averaged the verification metric
over the 3-fold subsets so that, in effect, the total number of observations that end up being used for
verification is
N
s
, the total number of stations. We thus argue that the verification sampling error for
the cross-validation experiments (red curve) is the same as for the active observations using the full
analysis (i.e., analysis using the total number of stations; black curve). In addition, note that from
Figure 1 the station sampling strategy gives rise to spatially random selection of stations, so that
158
Books
MDPI
Atmosphere 2018, 9,86
the individual metric on each set should be comparable. Furthermore, there is roughly 1300 quality
controlled O
3
observations over the domain and 750 PM
2.5
quality controlled observations, each with
60 time samples or less. To give some qualitative idea of the sampling error, the different metric
values for the individual 3-fold sets are presented in the Supplementary Material Figures S1 and S2,
where we can see that for
var(O − A)
and
cor(O
,
A)
the metric values for the individual sets are nearly
indistinguishable from the means of the 3-subset.
(a) (b)
Figure 2.
Variance of observation-minus-analysis residuals of O
3
and PM
2.5
for both active and
cross-validation passive observations as a function of
γ = σ
2
o
/σ
2
b
.(
a
) is for O
3
with ordinates in ppbv
2
units, and (
b
) is for PM
2.5
with ordinates in
(μg/m
3
)
2
. Red curve results from the evaluation at the
passive observation sites (average of the 3-fold subsets). Black curve results from evaluation at the
active observation sites with analyses using all observations. Green curve results from the evaluation
at the active observation sites in the cross-validation experiment (i.e., using 2/3 of the observations;
average of the three subsets). Blue curve is the variance of observation-minus-model.
The difference between the verification against passive observations in cross-validation analyses
(red curve) and the verification against active observations using full analyses (black curve) can be
attributed to two effects: (1) the analysis used in the cross-validation uses 2/3rd of the total number
of observations and thus the analysis error has larger variance than analyses using all observations,
(2) since the analysis error variance has typically a local minimum at the individual active observation
sites and increases away from it (see for example Figure 4a in [
28
] i.e., Part II of this paper), an evaluation
of the analysis error at passive sites (i.e., away from the active sites) has larger error variance than
those evaluated at the active sites [
16
]. We may call this the distance effect of passive observation sites.
In order to separate these two effects, we also display the 3-fold average of the metric verifying against
active observation for the cross-validation analyses as a green curve with squares. Thus in summary
we display a;
• red curve: using analysis with 2N
s
/3 observations with an evaluation at passive sites
• green curve: using analysis with 2N
s
/3 observations with an evaluation at active sites
• black curve: using analysis with N
s
observations with an evaluation at active sites.
The difference between the red and green curves show the influence of distance between passive
and active observation sites, whereas the difference between the green and black curves show the
influence of having different number of observations in creating the analysis for verification.
Let us first examine the results of verifying against active observations. As the observation
weights get smaller (i.e.,
γ
1), the analysis draws closer to the background, so that
var(O − A)
increases toward
var(O − B)
. On the other end, when
γ
1, the
var(O − A)
continuously decreases
159
Books
MDPI
Atmosphere 2018, 9,86
as
γ
diminishes to ultimately reach zero. This is in effect an expected result from the inner working
of an analysis scheme that the analysis error variance goes when the observation error variance goes
to zero. This effect does not depend on the observed values or the model values. For this reason,
the
var(O − A)
using active observations cannot provide a true measure of the quality of an analysis.
Now let us examine the results of verifying against passive observations with cross-validation
analyses. As the observation weights get smaller (i.e.,
γ
1), as for active observations the analysis
draws closer to the background, so that
var(O − A)
increases toward
var(O − B)
. On the other
end when
γ
1, the
var(O − A)
using passive observation increases as
γ
diminishes, whereas the
var(O − A)
evaluated at the active observation sites (green and black curves) decreases, indicating
that the analysis tries to overfit active observations which results in a spatially noisy analysis between
the active observation sites. Somewhere in between these two extreme values of
γ
lies a minimum
of
var(O − A)
where there is neither an overfitting nor an underfitting to the active observations.
This “optimal” ratio, that is found by inspection, actually corresponds the optimal analysis. It is also
where the analysis error variance with respect to the truth is minimum, but to show this last statement
requires an extensive analysis of the problem that we will discuss in part two of this study.
We also computed the verification of the subset of active observations used in the cross-validation
experiments with green curves. The difference between the black and green curves indicate the
effect of having more observations in the analysis. One would expect that having a larger number
of observations in the full analysis active
var(O − A)
compared to the active
var(O − A)
for the
cross-validation analyses would result in slightly smaller
var(O − A)
. This is indeed observed between
the black and green curves when the observation weight is small (i.e.,
γ
1). However, surprisingly,
when the observation weight is large,
γ
1, we observe the opposite. This intriguing behavior may
indicate an inconsistency between the assumption of uniform error variances for
σ
2
o
and
σ
2
b
(assumed
in the input error statistics) and the real spatial distribution of error variances. This discrepancy being
simply amplified when the observation weight is large and when there are less observations to produce
the analysis.
The difference between
var(O − A)
at passive sites and active sites (with the same number of
observations to construct the analyses) is substantial. For O
3
and for an optimal ratio, the
var(O − A)
at
passive sites is 51.02 ppbv
2
(red curve) while at active sites is 22.77 ppbv
2
(green curve). For PM2.5 and
for an optimal ratio, the
var(O − A)
at passive sites is 38.09
(μg/m
3
)
2
(red curve) while at active sites
is 15.41
(μg/m
3
)
2
(green curve). For both species, the error variance at active sites gives a significant
overestimation of the error variance by more than a factor of 2.
In Figure 3, we present the correlation metric between the observations and the analysis using,
as in Figure 2, the verification against passive observations in cross-validation analyses (red curve),
the verification against active observations using full analyses (black curve) and the verification
against active observations in the cross-validation analyses (green curve). The blue curve depicts the
correlation between the model and the observations, that is the prior correlation.
The evaluation against passive observations with cross-validation analyses (red curve) shows a
maximum at the same values of
γ = σ
2
o
/σ
2
b
than for the
var(O − A)
. We argue that the same arguments
of underfitting and overfitting are responsible for this maximum. The correlation between the active
observations and the analysis (black and green curves) increases as the observation weight increases
(
γ
decreases), theoretically reaching a value 1 for
σ
2
o
=
0, which is again unrealistic and simply shows
the impact of ill-prescribed error statistics in an analysis scheme. The gain in correlation between
independent observations and analysis is significant. For O
3
, it increases from a value of 0.55 with
respect to the model to a value of 0.74 with respect to an optimal analysis (when
γ = σ
2
o
/σ
2
b
is optimal).
For PM
2.5
, the correlation against the model has a value of 0.3 which basically has no skill, to a value of
0.54 for optimal analysis, which represent a modest but useable skill. The correlation evaluated at the
active sites for an optimal ratio, is 0.85 for O
3
(green curve) and 0.74 for PM
2.5
(green curve), being a
substantial overestimation with respect to values obtained at passive sites.
160
Books
MDPI
Atmosphere 2018, 9,86
(a) (b)
Figure 3.
Correlation between observations and analysis for (
a
)O
3
and (
b
)PM
2.5
for both active and
cross-validation passive observations as a function of
γ = σ
2
o
/σ
2
b
. The red, black and green curves are
as in Figure 2.
(a) (b)
Figure 4.
FC2 for (
a
)O
3
and (
b
)PM
2.5
for both active and cross-validation passive observations as a
function of γ = σ
2
o
/σ
2
b
. The red, black and green curves are as in Figure 2.
Another metric that we have considered is the FC2, Equation (5) [
3
]. The evaluation of this metric
against passive and active observations is presented in Figure 4 for O
3
(left panel) and PM
2.5
(right
panel). Note that the scale in the ordinate is quite different between the left and right panels. Although
the results bear similarity with the correlation between O and A presented in Figure 3, the maximum
with passive observations is reached at larger
γ
values than those obtained for
var(O − A)
or
cor(O
,
A)
,
which are identical. Individual fold results are presented in the supplementary materials Figure S3.
The interpretation of this metric is, however, not clear. Although the ratio
z = A/O
is a dimensionless
quantity the spread of z is generally not independent of the variance of A or O and there are cases
where it is. So to count the number of occurrence of
z
between the dimensionless values 0.5 and 2 is
confusing. As a simplified illustration, suppose that A is normally distributed as
N(
0,
σ
2
a
)
and similarly
with
O ∼ N(
0,
σ
2
o
)
. The ratio of these two random variables is then a Cauchy distribution whose
probability density function (pdf) is
σ
o
σ
a
/[π(σ
2
o
z
2
+ σ
2
a
)]
. The mean, variance and higher moments of
Cauchy probability distributions are not defined since the integral of the pdf is not bounded; only the
mode is defined. Cauchy distributions also have a spread parameter, which in this case is equal to
161
Books
MDPI
Atmosphere 2018, 9,86
σ
a
/σ
o
. If the variance of A and O are equal, then the number count between the dimensionless bounds
0.5 and 2 depends only on the shape of the probability distribution function, not on the variance.
If the variance of A and O are different, then it also depends on the ratio of variances. Furthermore,
in principle this metric also depends on the bias (which is not the case here for these analyses). It may
be a difficult metric to interpret but if used as a quality control, the FC2 have the unique ability of
rejecting too low as well as too high values of z.
In Figure 5 we present the bias between observations and analyses, and where the verification is
made against passive and active observations as done with the other metrics. Bias is not a dimensionless
quantity; note that the range and scale presented for O
3
and PM
2.5
in Figure 5 are different. The blue
curve is the mean
(O − B)
and thus indicates that for O
3
in average over all observation stations (for
the time and dates considered) the model overpredicts, and that for PM
2.5
the model underpredicts.
(a) (b)
Figure 5.
Bias between observation and analysis for (
a
)O
3
and (
b
)PM
2.5
for both active and cross-
validation passive observations as a function of
γ = σ
2
o
/σ
2
b
. The red, black and green curves are as in
Figure 2.
Contrary to all metric results seen so far, the 3-fold variability of the bias is substantial: it is of the
order of
±
0.5 ppbv (in average) for O
3
at passive sites and of the order of
±
0.1
μg/m
3
(in average) for
the PM
2.5
at passive sites (results shown in the supplementary material Figure S4). The distinction
between the red, black and green curves may not be statistically significant for both O
3
and PM
2.5
.
However, the difference between the analysis bias and model bias is large and statistically significant
(see supplementary material). For O
3
, the model bias is eliminated at the passive observation sites
(red curve) as long as the observation weight
γ ≤
1. The situation is not so clear for PM
2.5
. In fact,
when the observation weight is small, we get the intruiging result that the bias of the analysis is larger
bias than the model. How can that be when the observation weight is small (i.e.,
γ >
1); should the
analysis not be close to the model values? This apparent contradiction reveals a more complex issue
underlying the bias metric.
To explore the possible causes, we have calculated the bias per bin of model values, displayed in
Figure 6. In order to have a decent sample size per bin, we collect all the
(O − A)
and
(O − B)
over time
and observation sites, create bins of model values and calculate the statistic per bin (and not per station
as before). The result shows that the model bias is nearly linearly dependent on the model values
(black boxes in the bias panel). Both O
3
and PM
2.5
show an underprediction for low model values
and an overprediction for large model values. The origin of this bias is not known but one would
argue that it is not directly related to chemistry as such since both constituents, O
3
and PM
2.5
, present
the same feature. Possible explanations could be related to the model boundary layer, the emissions
being too low for low polluted areas and too large for polluted areas, insufficient transport away from
162
Books
MDPI
Atmosphere 2018, 9,86
polluted areas to unpolluted areas, species destruction/scavenging could be too low in low polluted
areas and too high in polluted areas. The lower panels of Figure 6a,b show the count of stations per
model bin size. We observe that the majority of stations have O
3
model values in the range of 40 to
55 ppbv, where the bias is negative. Over all the stations, this gives rise to a negative mean
(O − B)
,
and this is how we make the claim that the model overpredicts. However, for PM
2.5
the situation is
different: the majority of stations lie in the low model value range, and there are gradually less stations
for increasingly larger model values. Although the
(O − B)
have large negative values in the high
model value bin while small model value bins have positives
(O − B)
’s, the effect over all stations is to
yield a modestly positive mean
(O − B)
and thus the model underestimates the PM
2.5
. The results
of the analysis evaluated at the passive observation sites are presented with the yellow and grey
histogram boxes. In yellow, near optimal analyses with optimal observation weight, as determined by
the minimum of var(O − A) are used, and in grey non-optimal analyses with γ = 10.
(a)
(b)
Figure 6.
Biases per bin of model values. Figure (
a
), presents the statistics for O
3
and in (
b
), for PM
2.5
.
In the upper portion, (
a
,
b
) are the residual statistics per bin; in black, the
(O − B)
,ingrey,the
(O − A)
at passive observation sites (mean of the 3-fold subsets) for a non-optimal analysis with
γ =
10,
and in yellow, the
(O − A)
at passive observation sites (mean of the 3-fold subsets) using the optimal
observation weight. In the lower portion, (a,b) are the station number count per model values.
163
Books
MDPI
Atmosphere 2018, 9,86
We observe that the effect of the optimal analysis is nearly insentive to model bin values, where near
zero biases are obtained in most of the range except for very small and very large model values. The fact
that we are not able to capture the full benefit of analysis on all model values may be an artefact of
the assumption that we are using uniform observation and background error variances whereas the
model values varies considerably. In grey, we used the non-optimal analyses with a small observation
weight were we set
γ =
10. In the non-optimal case, the state-dependent bias is still present but
appears to be nearly perfectly anti-symmetric, positive in the low model value bins and nearly the
exact opposite in high model value bins. Since for O
3
the majority of observations lie in the range
40 to 55 ppbv,
(O − A)
for the optimal analyses at passive observation sites is nearly zero. However,
for the non-optimal analysis with
γ =
10, the
(O − A)
at passive sites is negative, i.e., the analysis is
overpredicting, as shown in Figure 5.
(a) (b)
Figure 7.
Modified normalized mean bias (MNMB) between observation and analysis for (
a
)O
3
and
(
b
)PM
2.5
for both active and cross-validation passive observations as a function of
γ = σ
2
o
/σ
2
b
. The red,
black and green curves are as in Figure 2.
For PM
2.5
, the weighted sum of the
(O − A)
bins is such that over all stations the bias for an
optimal analysis is nearly zero. In the case of the non-optimal analysis with
γ =
10, the weighted sum
of the nearly anti-symmetric
(O − A)
bias per bin gives more weight to the positive bias at smaller
model values, so that overall there is a positive (O − A), as in Figure 5.
To circumvent the state-dependency of the
(O − A)
biases it is useful to consider instead a
fractional bias metric, such as the modified normalized mean bias, MNMB Equation (4). The MNMB
metric is a dimensionless measure and as defined with a factor of 2, Equation (4), represents a % error.
The MNMB metric is a relative measure with respect to the mean observed-analysis value and is thus
less sensitive to spatially varying distribution of the concentrations, revealing instead the intrinsic
difference between the fields. The MNMB for O
3
and PM
2.5
for passive and active observations are
displayed in Figure 7 using the same color as in Figure 2. We note immediately that the MNMB analysis
bias does not exceed the MNMB model bias as we observed for the bias metric of PM
2.5
(Figure 5 right
panel). The MNMB bias also varies smoothly as a function of
γ
(at variance with the bias metric for
PM
2.5
—Figure 5).
Furthermore, examining the 3-fold variability of the cross-validation analysis MNMB at the
passive sites and the variability of the MNMB at the active sites (see Figure S5 in supplementary
materials), we infer that for PM
2.5
, where we can actually deduce that the difference between the
cross-validation and the validation against active observations is statistically significant when
γ <
1.
There is also another important point to make; although analyses are designed to reduce the error
164
Books
MDPI
Atmosphere 2018, 9,86
variance, it so happens that for a near optimal analysis the fractional bias MNMB is very small, around
1% for O
3
and about 1–2% for PM
2.5
. We argue that it results from an optimal use of observations.
There is also some information to gain from the variance of observed-minus-analysis per bin
size, as illustrated in Figure 8, using the same color histograms as in Figure 6. We note that for O
3
,
the model error variance against observations increases gradually with larger model values. However,
the fraction of analysis variance vs. model variance is roughly uniform across all bins. This can be
explained by the fact that the observation and background error variances are uniform, and thus the
reduction of variance across all bins is uniform as well. However, the situation is different for PM
2.5
.
We note a relatively poor performance of the model at low model values, with standard deviation of
7
μ
g/m
3
. For slightly larger model values (3–6
μ
g/m
3
), the error variance is smaller to 5.5
μ
g/m
3
and then increases almost linearly with model values. The fraction of analysis variance vs. model
variance decreases steadily with larger model values. These results thus indicate that the assumption
that observation and background error variances are uniform and independent of the model value
may have to be revisited.
(a)
(b)
Figure 8.
Same as Figure 6 except that we display the variance of analysis-minus-passive observations
per bin of model values.
165
Books
MDPI
Atmosphere 2018, 9,86
4. Conclusions
We have developed an approach by which analyses can be evaluated and optimized without
using a model forecast but rather by partitioning the original observation data set into a training set,
to create the analysis, and an independent (or passive) set, used to evaluate the analysis. This kind of
evaluation by partitioning is called cross-validation.
The need for such a technique came about from our desire to evaluate our operational surface air
quality analyses that are created off-line with no assimilation cycling. Evaluating a surface air quality
analysis based on its chemical forecast does in fact require additional information or assumptions, such as
vertical correlation, aerosol speciation and bin distribution (while surface measurement is primarily about
mass) or unobserved chemical variables correlations, and so on. So that the quality of the chemical forecast
is not solely dependent on the quality of the analysis and, if there are compensating errors, can actually be
a misleading assessment of the quality of the analysis.
We have applied this cross-validation procedure to the operational analyses of surface O
3
and
PM
2.5
over North America for a period of 60 days and present an evaluation using different metrics;
bias, modified normalized mean bias, variance of observation-minus-analysis residuals, correlation
between observation and analysis, and fraction of correction within a factor of 2.
Our results show that, in terms of variance and correlation, the verification of analyses against
active observations always yield an overestimation of the accuracy of the analysis. This overestimation
also increases as the observation weight increases. On the other hand for biases, the distinction between
the verification against active observations and passive observations is unclear and drowned in the
sample variability. However, using a fractional bias metric, in particular the MNMB, shows that the
verification against passive observations can be close to one percent for an optimal analysis while the
verification against active observations is much larger.
Results also show the importance of having an optimal analysis for verification. The variance of
the analysis with respect to independent observations is minimum and the correlation between the
analysis and independent observations is maximum for an optimal analysis. By being a compromise
between an overfit to the active observations (which produce noisy analysis field) and an underfit,
the optimal analysis offers the best use of observations throughout. At optimality, the analysis
fractional bias (MNMB) at the passive observation sites has only one or two percent error whereas the
fractional bias of the model is 6.5% for O
3
and 21% for PM
2.5
. The correlation between the analysis
and independent observations is also significantly improved with an optimal analysis: the correlation
between the model and independent observations is 0.55 for O
3
and increases to 0.74 with the analysis,
while for PM
2.5
the correlation between the model and independent observations is only 0.3 (which is
basically no skill) but rises to 0.54 for the analysis.
We also argue that the fraction of correct within a factor of 2, is a metric whose interpretation is
unclear as it mixes information about bias, variance and probability distribution in a non-uniform way
and does not seem to add anything new to other metrics. The bias is also very sensitive to sample
variability and can lead to wrong conclusions. For example, we have seen that the mean analysis
bias can be larger than the mean model bias, whether verifying against active or passive observations.
However, since an analysis is always closer to the truth than its prior (i.e., the model), it results in
an apparent contradiction. This implies that the bias metric cannot be used to faithfully compare
model states accurately. Such wrongful conclusions do not arise, however, with the MNMB. We thus
recommend avoiding using bias as a measure of truthfulness, and use instead a fractional bias measure
such as the MNMB.
We also found that errors in the GEM-MACH model grow almost linearly with the model value.
This is particularly evident for the bias where the model underestimates at small model values and
overestimates at large model values. Furthermore, this occurs in equal ways for O
3
and PM
2.5
,
thus indicating that the source of this bias is not related to chemistry. The fact that, over the entire
domain, the model overestimates O
3
, and underestimates PM
2.5
is simply a result of the concentrations.
We have not conducted a systematic study of model error for other times of the day and other periods
166
Books
MDPI
Atmosphere 2018, 9,86
of the year, but it would be very interesting to look at this, to see whether or not changes of biases are
due primarily to changes in the distribution of values rather than a fundamental change in the bias per
model value bin.
Finally, we have also examined the variance against independent observations per model value
bin, and concluded that the error variance is not quite uniform with model values but increases slowly
with model values for O
3
and in a more pronounced way for PM
2.5
.
In part two, we will focus on the estimation of the analysis error variance and develop a
mathematical formalism that permits the comparison of different diagnostics of variance under
different assumptions, optimizes the analysis parameters and gains confidence on the estimate of
analysis error as we obtain coherent estimated values across different diagnostics.
Supplementary Materials:
The following are available online at www.mdpi.com/xxx/s1, Figure S1: Verification
of variance for O
3
and PM
2.5
for the individual sets. Figure S2: Same as Figure S1 but for the correlation between
observations and analysis. Figure S3: Same as Figure S1 but for the fraction of correct within a factor of 2. Figure S4:
Same as Figure S1 but for bias. Figure S5: Same as Figure S1 but for modified normalized mean bias.
Acknowledgments:
We are grateful to the US/EPA for the use of the AIRNow database for surface pollutants
and to all provincial governments and territories of Canada for kindly transmitting their data to the Canadian
Meteorological Centre to produce the surface analysis of atmospheric pollutants. We are also thankful for the
proof read by Kerill Semeniuk, and for three anonymous reviewers for their comments and help in improving
the manuscript.
Author Contributions:
This research was conducted as a joint effort by both authors. R.M. contributed to the
theoretical development and wrote the paper, and M.D.-J. conducted all experiments design and execution,
proof reading and introduced a new diagnostic of optimal analysis error that was furher extended to passive
observations space.
Conflicts of Interest:
The authors declare no conflict of interest. The founding sponsors, which is the government
of Canada, had no role in the design of the study; in the collection, analyses, or interpretation of data; in the
writing of the manuscript, and in the decision to publish the results.
References
1.
Ménard, R.; Robichaud, A. The chemistry-forecast system at the Meteorological Service of Canada.
In Proceedings of the ECMWF Seminar Proceedings on Global Earth-System Monitoring, Reading, UK,
5–9 September 2005; pp. 297–308.
2.
Robichaud, A.; Ménard, R. Multi-year objective analysis of warm season ground-level ozone and PM
2.5
over
North-America using real-time observations and Canadian operational air quality models. Atmos. Chem. Phys.
2014, 14, 1769–1800. [CrossRef]
3.
Robichaud, A.; Ménard, R.; Zaïtseva, Y.; Anselmo, D. Multi-pollutant surface objective analyses and mapping
of air quality health index over North America. Air Qual. Atmos. Health
2016
, 9, 743–759. [CrossRef] [PubMed]
4.
Moran, M.D.; Ménard, S.; Pavlovic, R.; Anselmo, D.; Antonopoulus, S.; Robichaud, A.; Gravel, S.; Makar, P.A.;
Gong, W.; Stroud, C.; et al. Recent Advances in Canada’s National Operational Air Quality Forecasting System,
32nd ed.; Springer: Dordrecht, The Netherlands, 2014.
5.
Pudykiewicz, J.A.; Kallaur, A.; Smolarkiewcz, P.K. Semi-lagrangian modelling of tropospheric ozone. Tellus B
1997, 49, 231–248. [CrossRef]
6. Cressie, N.; Wikle, C.K. Statistics for Spatio-Temporal Data; Wiley: Hoboken, NJ, USA, 2011.
7.
Schneider, P.; Castell, N.; Vogt, M.; Dauge, F.R.; Lahoz, W.A.; Bartonova, A. Mapping urban air quality in
near real-time using observations from low-cost sensors and model information. Environ. Int.
2017
, 106,
234–247. [CrossRef] [PubMed]
8.
Lindström, J.; Szpiro, A.A.; Oron, P.D.; Richards, M.; Larson, T.V.; Sheppard, L. A flexible spatio-temporal
model for air pollution and spatio-temporal covariates. Environ. Ecol. Stat.
2014
, 21, 411–433. [CrossRef]
[PubMed]
9.
Carmichael, G.R.; Sandu, A.; Chai, T.; Daescu, D.N.; Constantinescu, E.M.; Tang, Y. Predicting air quality:
Improvements through advanced methods to integrate models and measurements. J. Comput. Phys.
2008
, 227,
3540–3571. [CrossRef]
167
Books
MDPI
Atmosphere 2018, 9,86
10.
Dabberdt, W.F.; Carroll, M.A.; Baumgardner, D.; Carmichael, G.; Cohen, R.; Dye, T.; Ellis, J.; Grell, G.;
Grimmond, S.; Hanna, S.; et al. Meteorological research needs for improved air quality forecasting: Report of
the 11th prospectus development team of the US weather research program. Bull. Am. Meteorol. Soc.
2004
, 85,
563–586. [CrossRef]
11.
Sportisse, B. A review of current issues in air pollution modeling and simulation. Comput. Geosci.
2007
, 11,
159–181. [CrossRef]
12.
Elbern, H.; Strunk, A.; Nieradzik, L. Inverse modelling and combined state-source estimation for chemical
weather. In Data Assimilation; Lahoz, W., Khattatov, B., Ménard, R., Eds.; Springer: Berlin/Heidelberg,
Germany, 2010; pp. 491–513.
13.
Bocquet, M.; Elbern, H.; Eskes, H.; Hirtl, M.; Žabkar, R.; Carmicheal, G.R.; Flemming, J.; Inness, A.;
Pagaoski, M.; Pérez Camaño, J.L.; et al. Data assimilation in atmospheric chemistry models; current status
and future prospects for coupled chemistry meteorology models. Atmos. Chem. Phys.
2015
, 15, 5325–5358.
[CrossRef]
14.
Chai, T.; Carmichael, G.R.; Sandu, A.; Tang, Y.H.; Daescu, D.N. Chemical data assimilation of transport and
chemical evolution over the pacific (TRACE-P) aircraft measurements. J. Geophys. Res.
2006
, 111, D02301.
[CrossRef]
15. Sandu, A.; Chai, T. Chemical data assimilation—An overview. Atmosphere 2011, 2, 426–463. [CrossRef]
16.
Marseille, G.J.; Barkmeijer, J.; De Haan, S.; Verkle, W. Assessment and tuning of data assimilation systems
using passive observations. Q. J. R. Meteorol. Soc. 2016, 142, 3001–3014. [CrossRef]
17.
Ménard, R.; Deshaies-Jacques, M.; Gasset, N. A comparison of correlation-length estimation methods for
the objective analysis of surface pollutants at Environment and Climate Change Canada. J. Air Waste
Manag. Assoc. 2016, 66, 874–895. [CrossRef] [PubMed]
18.
Cohn, S.E.; Da Silva, A.; Guo, J.; Sienkiewicz, M.; Lamich, D. Assessing the effects of data selection with the
DAO physical-space statistical analysis system. Mon. Weather Rev. 1998, 126, 2913–2926. [CrossRef]
19.
Houtekamer, P.L.; Mitchell, H.L. A sequential ensemble Kalman filter for atmospheric data assimilation.
Mon. Weather Rev. 2001, 129, 123–137. [CrossRef]
20. Efron, B.; Tibshirani, R.J. An Introduction to Boostrap; Chapman & Hall: New York, NY, USA, 1993.
21.
Seigneur, C.; Pun, B.; Pai, P.; Louis, J.F.; Solomon, P.; Emery, C.; Morris, R.; Zahniser, M.; Worsnop, D.;
Koutrakis, P.; et al. Guidance for the performance evaluation of three-dimensional air quality modeling
systems for particulate matter and visibility. J. Air Waste Manag. Assoc.
2000
, 50, 588–599. [CrossRef] [PubMed]
22.
Chang, J.C.; Hanna, S.R. Air quality model performance evaluation. Meteorol. Atmos. Phys.
2004
, 87, 167–196.
[CrossRef]
23.
Savage, N.H.; Agnew, P.; Davis, L.S.; Ordóñez, C.; Thorpe, R.; Johnson, C.E.; O’Connor, F.M.; Dalvi, M.
Air quality modelling using the Met Office Unified Model (AQUM OS24-26): Model description and initial
evaluation. Geosci. Model Dev. 2013, 6, 353–372. [CrossRef]
24.
Katragkou, E.; Zanis, P.; Tsikerdekis, A.; Kapsomenakis, J.; Melas, D.; Eskes, H.; Flemming, J.; Huijnen, V.;
Inness, A.; Schultz, M.G.; et al. Evaluation of near surface ozone over Europe from the MACC reanalysis.
Geosci. Model Dev. 2015, 8, 2299–2314. [CrossRef]
25.
Ménard, R. Error covariance estimation methods based on analysis residuals: Theoretical foundation and
convergence properties derived from simplified observation networks. Q. J. R. Meteorol. Soc.
2016
, 142,
257–273. [CrossRef]
26.
Desroziers, G.; Berre, L.; Chapnik, B.; Poli, P. Diagnosis of observation, background, and analysis-error
statistics in observation space. Q. J. R. Meteorol. Soc. 2005, 131, 3385–3396. [CrossRef]
27. Daley, R. Atmospheric Data Analysis; Cambridge University Press: New York, NY, USA, 1991; p. 457.
28.
Ménard, R.; Deshaies-Jacques, M. Evaluation of analysis by cross-validation, Part II: Diagnostic and
optimization of analysis error covariance. Atmosphere 2018, 9, 70. [CrossRef]
©
2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
168
Books
MDPI
atmosphere
Article
Evaluation of Analysis by Cross-Validation, Part II:
Diagnostic and Optimization of Analysis
Error Covariance
Richard Ménard *and Martin Deshaies-Jacques
Air Quality Research Division, Environment and Climate Change Canada, 2121 Transcanada Highway, Dorval,
QC H9P 1J3, Canada; martin.deshaies-jacques@canada.ca
* Correspondence: richard.menard@canada.ca; Tel.: +1-514-421-4613
Received: 7 November 2017; Accepted: 13 February 2018; Published: 15 February 2018
Abstract:
We present a general theory of estimation of analysis error covariances based on
cross-validation as well as a geometric interpretation of the method. In particular, we use the variance
of passive observation-minus-analysis residuals and show that the true analysis error variance can
be estimated, without relying on the optimality assumption. This approach is used to obtain near
optimal analyses that are then used to evaluate the air quality analysis error using several different
methods at active and passive observation sites. We compare the estimates according to the method
of Hollingsworth-Lönnberg, Desroziers et al., a new diagnostic we developed, and the perceived
analysis error computed from the analysis scheme, to conclude that, as long as the analysis is near
optimal, all estimates agree within a certain error margin.
Keywords:
data assimilation; statistical diagnostics of analysis residuals; estimation of analysis error;
air quality model diagnostics; Desroziers et al. method; cross-validation
1. Introduction
At Environment and Climate Change Canada (ECCC) we have been producing hourly surface
pollutants analyses covering North America [
1
–
3
] using an optimum interpolation scheme which
combines the operational air quality forecast model GEM-MACH output [
4
] with real-time hourly
observations of O
3
,PM
2.5
,PM
10
,NO
2
, and SO
2
from the AirNow gateway with additional observations
from Canada. These analyses are not used to initialize the air quality model and we wish to evaluate
them by cross-validation, that is by leaving out a subset of observations from the analysis to use them
for verification. Observations used to produce the analysis are called active observations while those
used for verification are called passive observations.
In a first-part paper of this study, i.e., Ménard and Deshaies-Jacques [
5
], we have examined
different verification metrics using either active or passive observations. As we changed the ratio of
observation error to background error variances
γ = σ
2
o
/σ
2
b
, while keeping the sum
σ
2
o
+ σ
2
b
equal to
var(O − B)
, we found a minimum in
var(O − A)
in the passive observation space. In this second-part
paper, we formalize this result, develop the principles of estimation of the analysis error covariance by
cross-validation, and apply it to estimate and optimize the analysis error covariance of ECCC’s surface
analyses of O
3
and PM
2.5
.
When we refer to analysis error, or analysis error covariance, it is important to distinguish the
perceived analysis error with the true analysis error [
6
]. The perceived analysis error is the analysis
error that results from the analysis algorithm itself, whereas the true analysis error is the difference
between the analysis and the true state. Analysis schemes are usually derived from an optimization of
some sort. In a variational analysis scheme for example, the analysis is obtained by minimizing a cost
function with some given or prescribed observation error and background error covariances,
R
and
B
Atmosphere 2018, 9, 70 169 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2018, 9,70
respectively. In a linear unbiased analysis scheme, the gain matrix
K
is obtained by minimum variance
estimation, yielding an expression of the form,
K
=
BH
(H
BH +
R
)
−1
, where
H
is the observation
operator. The perceived analysis error covariance is then derived as
A
=(I −
KH
)
B
. In order to
derive an expression for the perceived analysis error covariance we in fact assume that given error
covariances,
R
and
B
, are error covariances with respect to the true state, i.e., the true error covariances.
We also assume that the observation operator is not an approximation with some error, but is the
true error-free observation operator. Of course, in real applications neither of
R
and
B
are covariance
measures with respect to the true state, but only a more or less accurate estimate of those. Daley [
6
]
have argued that in principle for an arbitrary gain matrix
K
, the true analysis error covariance
A
can
be computed as
A =(I −
KH
)B(I −
KH
)
T
+
KR
K
T
provided that we know the true observation and
background error covariances,
R
and
B
. This expression is a quadratic matrix equation, and has the
property that the true analysis error variance,
tr(A)
is minimum when
K
= BH(HBH + R)
−1
= K
.
In that sense, the analysis is truly optimal. The optimal gain matrix
K
is called the Kalman gain. It thus
illustrates that although an analysis is obtained through some minimization principle, the resulting
analysis error is not necessarily the true analysis error.
One of the main sources of information to obtain the true
R
and
B
is from the
var(O − B)
statistic.
However, it has always been argued that this is not possible without making some assumptions [
7
–
9
],
the most useful one being that background errors are spatially correlated while the observation errors
are spatially uncorrelated, or at least on a much shorter length-scale. Even under those assumptions,
different estimation methods such as the Hollingsworth-Lönnberg method [
10
], the maximum
likelihood gives different error variances and different correlation lengths [
11
]. Other methods use
var(O − B)
for rescaling but assume that the observation error is known. The assumption that the
observation error is known is also debated as they contain representativeness errors [12] that include
observation operator errors. How to obtain an optimal analysis is thus unclear.
The evaluation of the true or perceived analysis error covariance using its own active observations
is also a misleading problem unless the analysis is already optimal. Hollingsworth and Lönnberg [
13
]
addressed this issue for the first time where they noted that in the case of an optimal gain (i.e., optimal
analysis), the statistics of observation-minus-analysis residuals
O −
ˆ
A
are related to the analysis error
by
E[(O −
ˆ
A
)(O −
ˆ
A
)
T
]=R − H
ˆ
AH
T
, where
ˆ
A
is the optimal analysis error covariance and
H
and
R
are the observation operator and observation error covariance respectively. The caret (ˆ) over
A
indicates that the analysis uses an optimal gain. In the context of spatially uncorrelated observation
errors, the off-diagonal elements of
E[(O −
ˆ
A
)(O −
ˆ
A
)
T
]
would then give the analysis error covariance
in observation space. Hollingsworth and Lönnberg [
13
] argued that for most practical purposes,
the negative intercept of
E[(O −
ˆ
A
)(O −
ˆ
A
)
T
]
at zero distance and the prescribed observation weight
should be nearly equal, and thus could be used as an assessment of optimality of an analysis. However,
in case where such agreement does not exist, an estimate of the actual analysis error is not possible.
Another method, proposed by Desroziers et al. [
14
], argued that the diagnostic
E[(O −
ˆ
A
)(
ˆ
A
− B)
T
]
should be equal to the analysis error covariance in observation space but, again, only if the gain is
optimal and the innovation covariance consistency is respected [15].
The impasse of the estimation of the true analysis error seems to be tied with using active
observations, i.e., using the same observations as those used to create the analysis. A robust approach
that does not require an optimal analysis is to use observations whose errors are uncorrelated with
the analysis error. For example, if we assume that observation errors are temporarily (serially)
uncorrelated, an estimation of the analysis error can be made with the help of a forecast model
initialized by the analysis by verifying the forecast against these observations. This is the essential
assumption used traditionally in meteorological data assimilation to assess indirectly the analysis
error by comparing the resulting forecast with observations valid at the forecast time. As forecast
error grows with time, the observation-minus-forecast can be used to assess whether an analysis is
better than another. In a somewhat different method but making the same assumption, Daley [
6
]
170
Books
MDPI
Atmosphere 2018, 9,70
used the temporal (serial) correlation of the innovations to diagnose the optimality of the gain matrix.
This property was first established in the context of Kalman filter estimation theory by Kailath [
16
].
However, both the traditional meteorological forecast approach and the Daley method [
6
] are subject
to limitations: they assume that the model forecast has no bias and the analysis corrections are made
correctly on all the variables needed to initialize the model. In practice, improper initialization of
unobserved meteorological variables gives rise to spin-up problems or imbalances. Furthermore, with
the traditional meteorological approach, compensation due to model error can occur, so that an optimal
analysis does not necessarily yield an optimal forecast [7].
An alternative approach introduced by Marseille et al. [
17
], which we will follow here, is to use
independent observation or passive observations to assess the analysis error. The essential assumption
of this method is that the observations have spatially uncorrelated errors, so that the observations used
for verification, i.e., the passive observations, have uncorrelated errors with the analysis. The advantage
of this approach is that it does not involve any model to propagate the analysis information to a later
time. Marseille et al. [
17
] then showed that by multiplying the Kalman gain with an appropriate scalar
value, one can reduce the analysis error. In this paper, we go further by using principles of error
covariance estimation to obtain a near optimal Kalman gain. In addition we impose the innovation
covariance consistency [
15
] and show that all diagnostics of analysis error variance nearly agree with
one another. These include the Hollingsworth and Lönnberg [
13
], the Desroziers et al. [
14
] and new
diagnostics that we will introduce.
The paper is organized as follows. First we present in Section 2 the theory and diagnostics of
analysis error covariance in both passive and active observation spaces, as well as a geometrical
representation. This leads us to a method to minimize the true analysis error variance. In Section 3,
we present the experimental setup on how we obtain near optimal analyses and presents the results
of several diagnostics in active and passive observation spaces, and compare with the analysis error
variance obtained from the optimum interpolation scheme itself. In Section 4, we discuss the statistical
assumptions being used, how and if they can be extended and how this formalism can be used in other
applications such as the estimation of correlated observation errors with satellite observations. Finally,
we draw some conclusions in Section 5.
2. Theoretical Framework
This section is composed of mainly three parts. In Sections 2.1 and 2.2, we first describe the
diagnostics to obtain the true error covariances using passive observations whether the analysis is
optimal or not. We then give a geometric interpretation in Section 2.3 that indicate the way to obtain
the optimal analysis, that is, one that minimizes the true analysis error. Lastly, in Sections 2.4 and 2.5,
we formulate the diagnostics of analysis error for optimal analyses.
2.1. Diagnostic of Analysis Error Covariance in Passive Observation Space
Let us decompose the observation space in two disjoint sets; the active observation set or training
set
{
y
}
used to create the analysis, and the independent or passive observation set
{
y
c
}
used to
evaluate the analysis. An analysis built from prescribed background and observation error covariances,
B and
R respectively, is given by
x
a
= x
f
+
BH
T
(H
BH
T
+
R
)
−1
d = x
f
+
Kd (1)
where
K
is the gain matrix built from the prescribed error covariances,
H
is the observation operator
for the active observation set,
d
is the active innovation vector,
d = y − Hx
f
= ε
o
− Hε
f
,
x
f
is the
171
Books
MDPI
Atmosphere 2018, 9,70
background state or model forecast,
ε
o
is the active observation error and
ε
f
the background error.
The observation-minus-analysis residual (O − A) for the active set is given by [14,15,18].
(O − A)=y − Hx
a
= ε
o
− Hε
a
= d − H
BH
T
(H
BH
T
+
R
)
−1
d
=
R
(H
BH
T
+
R
)
−1
d
(2)
where
ε
a
is the analysis error. The analysis interpolated at the passive observation sites
can be denoted by
H
c
x
a
, where
H
c
is the observation operator at passive observation sites.
The observation-minus-analysis residual at the passive observation sites (O − A)
c
is then given by
(O − A)
c
= y
c
− H
c
x
a
= ε
o
c
− H
c
ε
a
= d
c
− H
c
BH
T
(H
BH
T
+
R
)
−1
d
, (3)
where
d
c
= y
c
− H
c
x
f
= ε
o
c
− H
c
ε
f
is the innovation at the passive observation sites. Note that the
formalism introduced here is general, and can be used with any set of independent observations
such as different instruments or observation networks as long as the
H
c
operator is properly defined.
Consequently, for generality, we distinguish the passive observation errors or independent observation
error, ε
o
c
, from the active observation error ε
o
.
There are two important statistical assumptions from which we derive cross-validation diagnostics.
Assuming that the observation errors are spatially uncorrelated, it follows that
E
[ε
o
(ε
o
c
)
T
]=0 (4)
where
E[]
is the mathematical expectation that represents the mean over an ensemble of realizations.
It has been argued by Marseille et al. [
17
] that representativeness error can violate this assumption
for a close pair of active-passive observations, but we will neglect this effect. Also, assuming that
observation errors are uncorrelated with background error, we have
E
[ε
o
(Hε
f
)
T
]=0,E[ε
o
c
(H
c
ε
f
)
T
]=0 (5)
We come now to the most important property: since the analysis is a linear combination of the
active observations and the background state, the analysis error is then uncorrelated with the passive
observation errors,
E
[(
c
ε
a
)(ε
o
c
)
T
]=0 (6)
and thus we get the following cross-validation diagnostic in passive observation space,
E
[(O − A)
c
(O − A)
T
c
]=R
c
+ H
c
AH
T
c
(7)
similarly to Marseille et al. [
17
]. A very important point to note is that
A
is the true analysis error
covariance—it does not assume that the gain is optimal. The matrices in Equation (7) are of the
dimension of the passive observation space and
R
c
is the observation error covariance matrix for the
passive or independent observations.
2.2. A Complete Set of Diagnostics of Error Covariances in Passive Observation Space
It is also possible to define a set of diagnostics that would determine, in principle, the true error
covariances, R, B and A. From Equations (4) and (5) we get another cross-validation diagnostic,
E
[(O − B)
c
(O − B)
T
]=H
c
BH
T
(8)
172
Books
MDPI
Atmosphere 2018, 9,70
This diagnostic is related to the Hollingsworth-Lönnberg [
10
] estimation of the spatially correlated
part of the innovation, and in practice can be dominated by sampling error. Note that it is not a square
matrix and an estimation of parameters of B may not be trivial.
We can also obtain the innovation covariance matrix in passive observation space (identical in
fact to the one in active observations space) as
E
[(O − B)
c
(O − B)
T
c
]=R
c
+ H
c
BH
T
c
(9)
The system of Equations (7)–(9) gives a complete set of equations to determine the true
R
,
B
and
A
at the passive observation sites provided that by interpolation/extrapolation we can obtain
H
c
BH
T
c
from H
c
BH
T
.
Nevertheless, and for sake of completeness, we also investigated the meaning of the statistics
E[(O − A)
c
(O − B)
T
]
and came to the conclusion that it can interpreted as a misfit to the Desroziers et al.
estimate of
B
[
14
] with the true value
B
. We recall that the first iterate of the Desroziers et al. estimate
for
B
in active observation space is given by
HB
D
H
T
= E[(Hx
a
− Hx
f
)(y − Hx
f
)
T
]
[
15
], where we
used the superscript D indicate the Desroziers et al. first iterate estimate. We can actually generalize
this diagnostic to be a cross-covariance between state space and active observation space as
B
D
H
T
= E[(x
a
− x
f
)(y − Hx
f
)
T
] (10)
By applying
H
c
to Equation (10) we can then introduce a generalized Desroziers et al. estimate of the
background error covariance B between passive and active observation spaces as,
E
[( A − B)
c
(O − B)
T
]=H
c
B
D
H
T
(11)
Since
(O − A)
c
=(O − B)
c
− (A − B)
c
, we get with Equations (8) and (10),
E
[(O − A)
c
(O − B)
T
]=H
c
(B − B
D
)H
T
(12)
That is the difference between the true
B
and the Desroziers et al. first estimate
B
D
in the
cross passive-active observation spaces. Similarly to Equation (8) this diagnostics requires spatial
interpolation of error covariances from active sites to passive sites of basically noisy statistics.
The estimation of
B
from this diagnostics is further complicated by the fact that what is being
interpolated, that is B − B
D
, may not even be positive definite, but the augmented matrix,
cov
(O − B)
(
O − A)
c
=
R
+ HBH
T
H(B − B
D
)H
T
c
H
c
(B − B
D
)H
T
R
c
+ H
c
AH
T
c
(13)
is positive definite. Except for the diagonal of Equation (13), we have not attempted in this study to
conduct this complete estimation of
R
,
B
and
A
, but rather focused on getting a reliable estimate of the
analysis error covariance A.
2.3. Geometrical Interpretation
A geometrical illustration of some of the relationships obtained above can be made by using a
Hilbert space representation of random variables in observation space. A 2D representation for the
analysis of a scalar quantity was used in Desroziers et al. [
14
] to illustrate their a posteriori diagnostics.
We will generalize this approach to include passive observations by considering a 3D representation.
As in Desroziers et al. [
14
] let’s consider the analysis of a scalar quantity. Several variables are to be
considered in this observation space:
y
o
the active observation (or measurement) of the scalar quantity,
y
b
the background (or prior) value equivalent in observation space (i.e.,
y
b
= Hx
b
),
y
a
the analysis
in observation space (i.e.,
y
a
= Hx
a
), and for verification
y
c
an independent observation (or passive
173
Books
MDPI
Atmosphere 2018, 9,70
observation) that is not used to compute the analysis. Each of these quantities is a random variable
as they each contain random errors, and any linear combination of random variables in observation
space also belong to observation space. For example,
y
o
− y
b
is the innovation (commonly denoted
by O-B) and that belongs to observation space,
y
a
− y
b
is the analysis increment in observation space
(commonly denoted by A-B), and
y
o
− y
a
is the analysis residual in observation space (commonly
denoted by O-A). We can also define an inner product of any random variables in observation space.
y
1
, y
2
,as
y
1
, y
2
:= E
[
(
y
1
− E(y
1
))(y
2
− E(y
2
))
]
(14)
The squared norm then represents the variance,
y
2
:=
y, y
=
σ
2
y
(15)
so the inner product has the following geometric interpretation
y
1
, y
2
=
y
1
y
2
cos θ (16)
where
cos θ
is the correlation coefficient. Uncorrelated random variables are thus statistically
orthogonal. With this inner product, the observation space forms a Hilbert space of random variables.
Figure 1 illustrates the statistical relationship in observation space between: the active observation
y
o
(illustrated as O in the figure), the prior or background
y
b
(i.e., B), the analysis
y
a
(i.e., A), and the
independent observation
y
c
(i.e., O
c
). The origin T corresponds to the truth of the scalar quantity,
and also corresponds to the zero of the central moment of each random variables, e.g.,
y − E[y]
, since
each variables are assumed to be unbiased. We also assume that the background, active and passive
observations errors are uncorrelated to one another, so the three axes;
ε
o
for the active observation
error,
ε
b
for the background error, and
ε
o
c
for the passive observation error are orthogonal. The plane
defined by
ε
o
and
ε
b
axes is the space where the analysis takes place, and is called the analysis
plane. However, since we define the analysis to be linear and unbiased, only linear combinations
of the form
y
a
= ky
o
+(
1
− k)y
b
where k is a constant are allowed. The analysis A then lies on
the line (B, O). The thick lines in Figure 1 represent the norm of the associated error. For example,
the thick line along the
ε
o
axis depict the (active) observation standard deviation
σ
o
, and similarly for
the other axes and other random variables. Since the active observation error is uncorrelated with
the background error, the triangle
Δ
OTB is a right triangle, and by Pythagoras theorem we have,
(y
o
− y
b
)
2
:=
(O − B), (O − B)
= σ
2
o
+ σ
2
b
. This is the usual statement that the innovation variance
is the sum of background and observation error variances. The analysis is optimum when the analysis
error
ε
a
2
= σ
2
a
is minimum, in which case the line (T, A) is perpendicular to line (O, B).
Now let’s consider the passive observation O
c
. The passive observation error is perpendicular to
the analysis plane, thus the triangle ΔO
c
TA is a right triangle,
(y
c
− y
a
)
2
:=
(O − A)
c
, (O − A)
c
=
σ
2
c
+ σ
2
a
(17)
where
σ
2
c
is the passive observation error variance. The most important fact to stress here is that the
orthogonality expressed in Equation (17) is true whether or not the analysis is optimal. Furthermore,
as the distance
(y
c
− y
a
)
2
varies with the position of
A
along the line (O, B), the distance
(y
c
− y
a
)
2
reaches a minimum value when
σ
2
a
is minimum that is when the analysis is optimal. We thus also
argue from this representation that there is always a minimum, and the minimum is unique. Finally,
we note that
Δ
BTO
c
is also a right triangle so that
(y
c
− y
b
)
2
:=
(O − B)
c
, (O − B)
c
=
σ
2
c
+ σ
2
b
, which
is the scalar version of Equation (9).
174
Books
MDPI
Atmosphere 2018, 9,70
Figure 1.
Hilbert space representation of a scalar analysis and cross-validation problem. The arrows
indicate the directions of variability of the random variables, and the plane defined by the background
and observation errors
ε
b
,
ε
o
defines the analysis plane. The thick lines represent the norm associated
with the different random variables. T indicate the truth, O the active observation, B the background,
A the analysis and O
c
the passive observation.
To extend this formalism to random vectors requires to define a proper matricial inner product
as briefly described in Section 1.2 of Caines [
19
]. It is important here to distinguish the stochastic
metric space from the observation vector space. In the stochastic metric space a “scalar” needs not to
be a number, but only a non-random quantity invariant with respect to
E[]
. Hence, we can define a
Hilbert space with the following (stochastic scalar) matricial inner product
y, w
=
E[(y − E(y))(w −
E(w)
T
)] = cov(y
,
w)
. The matrix nature of
cov(y
,
w)
pertains to the observation vector space, but it
remains a scalar with respect to the stochastic Hilbert space herein defined. In order to obtain a true
scalar (
∈ R
), one would need to define a metric on the observation space matrices as well, such as the
trace. Also to be able to compare active and passive observations, projectors need to be introduced on
the observation space, implying yet another metric structure on the observation space. We do not carry
out this formalism here as it would represent a rather lengthy development that would distract us
from the main purpose of this paper; this will be considered in a future manuscript. Finally, we remark
that Hilbert space representation of random variables in infinite dimensional space (i.e., continuous
space) can also be defined, see Appendix 1 of Cohn [20].
2.4. Error Covariance Diagnostics in Active Observation Space for Optimal Analysis
An analysis is optimal if the analysis error
E[(ε
a
)
T
ε
a
]
is minimum. This implies that the gain
matrix using the prescribed error covariances,
K
in Equation (1), must be identical to the gain using
the true error covariances, i.e.,
K
= BH
T
(HBH
T
+ R)
−1
[
14
,
15
]. It is important to mention that
necessary and sufficient conditions to obtain the true error covariances
HBH
T
and
R
in observation
space, are: 1—the Kalman gain condition,
H
K = HK
true
and 2—the innovation covariance consistency,
E[(O − B)(O − B)
T
]=H
BH
T
+
R
. For a proof see the Theorem on error covariance estimates in
Ménard [15].
From the optimality of the analysis (or Kalman gain) alone, we derive that
E[(
ˆ
A
− T)(O − B)
T
]=
0
or
E[(
ˆ
A
− T)(O −
ˆ
A
)
T
]=0
. Indeed, from Equation (2), we get
(O −
ˆ
A
)=R(HBH + R)
−1
d
,
and for the analysis error in observation space we get,
(
ˆ
A
− T)=R(HBH
T
+ R)
−1
Hε
f
+
HBH
T
(HBH
T
+ R)
−1
ε
o
, from which we derive the expectations above. Using the geometrical
representation in Section 2.3 the distance between A and T is minimum, when
ΔTAO
and
ΔTAB
(Figure 1) are right triangles. We should also note that for the scalar problem, the Kalman gain depends
175
Books
MDPI
Atmosphere 2018, 9,70
only on the ratio of observation to background error variances and thus the scalar Kalman gain is
optimal if the ratio of the prescribed error variances is equal to the ratio of the true error variances.
If in addition to the optimality of the analysis or Kalman gain we add the innovation covariance
consistency then we get three different statistical diagnostics of the (optimal) analysis error covariance.
Hollingsworth and Lönnberg [
13
] was the first to introduce a statistical diagnostic of analysis error in
the active observation space, as:
E
[(O −
ˆ
A
)(O −
ˆ
A
)
T
]=R − H
ˆ
A
HL
H
T
(18)
Here we use a subscript, HL to indicate that this is the Hollingsworth-Lönnberg estimate. Equation
(18) can obtained from the covariance of
(O −
ˆ
A
)
and that, for an optimal gain matrix,
H
ˆ
AH
T
=
R − R(HBH + R)
−1
R
which derives from the usual formula,
ˆ
A
= B − BH
T
(HBH + R)
−1
HB
.
Geometrically it derives from the fact that the triangle
ΔT
ˆ
AO
is a right triangle, and from the innovation
covariance consistency that implies that the triangle
ΔOTB
is a right triangle. Using data to construct
E[(O −
ˆ
A
)(O −
ˆ
A
)
T
]
, an estimated analysis error covariance obtained from Equation (18) is symmetric
but could be non-positive definite as it is obtained by subtracting two positive definite matrices.
The effect of misspecification in the prescribed error covariances resulting from a lack of innovation
covariance consistency will be discussed in the result Section 3 and in Appendix B.
Inspired from the geometrical interpretation that
ΔTAB
is also be a right triangle we derived the
following diagnostic,
E
[(
ˆ
A
− B)(
ˆ
A
− B)
T
]=HBH
T
− H
ˆ
A
MD J
H
T
= H(B −
ˆ
A
MD J
)H
T
(19)
where MDJ stands for Ménard-Deshaies-Jacques. This relationship is obtained by using the expression
(
ˆ
A
− B)=HBH
T
(HBH + R)
−1
d
, the innovation covariance consistency and the formula for the
optimal analysis error covariance
ˆ
A
= B − BH
T
(HBH + R)
−1
HB
. As for the HL diagnostics,
the estimated error covariance obtained from this diagnostic is symmetric by construction but may not
be positive definite. Another way of looking at Equation (19) is that it expresses, in observation space,
the error reduction due to the use of observations.
Another diagnostic of analysis error covariance was proposed by Desroziers et al. [
14
].
By combining (O −
ˆ
A
) and (
ˆ
A
− B) we get
E
[(O −
ˆ
A
)(
ˆ
A
− B)
T
]=H
ˆ
A
D
H
T
(20)
where the subscript D denotes the Desroziers et al. estimate. By construction, the estimated analysis
error covariance is not necessarily symmetric. A geometrical derivation is provided in Appendix A.
We also provide in Appendix B a sensitivity analysis on the departure from innovation covariance
consistency for each diagnostics in both active and passive observation spaces.
2.5. Error Covariance Diagnostics in Passive Observation Space for Optimal Analysis
We can also derive optimal analysis diagnostics in the passive observation space. Considering the
3D geometric interpretation, and in particular the tetrahedron
(O
c
,
T
,
A
,
B)
, we notice that since
ΔTAB
is a right triangle, so is
ΔO
c
AB
, which is a projection of the triangle
ΔTAB
on the plane passing through
O
c
, O and B. We thus have
E[(
ˆ
A
− B)
c
(
ˆ
A
− B)
T
c
]+E[(O −
ˆ
A
)
c
(O −
ˆ
A
)
T
c
]=E[(O − B)
c
(O − B)
T
c
]
.
Combining this result with Equation (7) and using,
E[(O − B)
c
(O − B)
T
c
]=H
c
BH
T
c
+ R
c
, we then get
E[(
ˆ
A
− B)
c
(
ˆ
A
− B)
T
c
]=H
c
BH
T
c
− H
c
ˆ
A
MD J
H
T
c
(21)
Note that our analysis diagnostic is the only diagnostic that is valid in both active and passive
observation spaces, i.e., Equation (21) is similar to Equation (19).
176
Books
MDPI
Atmosphere 2018, 9,70
The other, less direct, diagnostic for optimal analysis is simply based on Equation (7) that is
argmin
γ, L
c
{E[(O − A)
c
(O − A)
T
c
]} = R
c
+ H
c
ˆ
A
(γ, L
c
)H
T
c
(22)
These five diagnostics will be used later in the results section.
3. Results with Near Optimal Analyses
3.1. Experimental Setup
We will just give here a short summary of the experimental setup we are using in this study. More
details can be found in the Part I paper (Ménard and Deshaies-Jacques [
5
]). A series of hourly analyses
of O
3
and PM
2.5
at 21 UTC for a period of 60 days (14 June to 12 August 2014) were performed using
an optimum interpolation scheme combining the operational air quality model GEM-MACH forecast
and the real-time AirNow observations (see Section 2 of [
5
] for further details). The analyses are
made off-line so they are not used to initialize the model. As input error covariances, we use uniform
observation and background error variances, with
R
= σ
2
o
I
and
B
= σ
2
b
C
, where
C
is a homogeneous
isotropic error correlation based on a second-order autoregressive model. The correlation length is
estimated by using a maximum likelihood method using at first, error variances obtained from a local
Hollingworth-Lönnberg fit [
11
] (and only for the purpose of obtaining a first estimate of the correlation
length). We then conduct a series of analyses by changing error variance ratio
γ = σ
2
o
/σ
2
b
while at
the same time respecting the innovation variance consistency condition,
σ
2
o
+ σ
2
b
= var(O − B)
. This
corresponds basically in searching for the minimum of the tr (trace) of Equation (7) while the trace of
the innovation covariance consistency, tr[R + HBH
T
]=tr{E[(O − B)(O − B)
T
]}, is respected.
First the observations are separated into 3 sets of observations of equal number and distributed
randomly in space. By leaving out one set of observations for verification and constructing analyses
with the remaining 2 other sets, we construct a cross-validation setup from which we can evaluate
the diagnostic
var(O − A)
c
= tr{E[(O − A)
c
(O − A)
T
c
}
in passive observation space. This constitutes
our first guess experiment that we will refer to as
iter 0
. No observation or model bias correction was
applied, nor were the mean of the innovation at the stations were removed prior to performing the
analysis. The variance statistics are first computed at the station using the 60-day members, and then
averaged over the domain to give the equivalent of
tr{E[(O − A)
c
(O − A)
T
c
}
. We repeat the procedure
by exhausting all permutations possible, in this case 3. The mean value statistic for the three verifying
subsets are then averaged. More details can be found in Section 2 and beginning of Section 3 of Part I [
5
].
The red curve on Figure 2 illustrates how this diagnostic varies with and exhibits a minimum.
This minimum can easily be understood by referring to Figure 1: as the analysis point
A
changes
position along the line
(O
,
B)
the distance
O
c
− A
reaches a minimum, and this is what we observe
in Figure 2.
In our next step,
iter 1
, we first re-estimate the correlation length by applying a maximum
likelihood method, as in Ménard [
15
], using the
iter 0
error variances that are consistent with the
optimal ratio
ˆ
γ
obtained in
iter 0
and the innovation variance consistency. Then, with this new
correlation length, we estimate a new optimal ratio
ˆ
γ
(
iter 1
), which turn out to be very close to the
value obtained in
iter 0
. We recall that we use uniform error variances, both to keep things simple but
also because the optimal ratio is obtained by minimizing a domain-averaged variance
var(O
c
− A)
.
A summary of the error covariance parameters obtained for iter 0 and iter 1 are presented in Table 1.
177
Books
MDPI
Atmosphere 2018, 9,70
Figure 2.
Red line, variance of observation-minus-analysis of O
3
in passive observation space. Blue
line, variance of observation-minus-model.
Table 1. Input error statistics for the first experiment and optimized variance ratio experiment.
Experiment L
c
(km)
(
O − B
)
2
ˆγ
=
ˆ
σ
2
o
/
ˆ
σ
2
b
ˆ
σ
2
o
ˆ
σ
2
b
χ
2
/N
s
O
3
iter 0 124 101.25 0.22 18.3 83 2.23
O
3
iter 1 45 101.25 0.25 20.2 81 1.36
PM
2.5
iter 0 196 93.93 0.17 13.6 80.3 2.04
PM
2.5
iter 1 86 93.93 0.22 16.9 77 1.25
We have also added in Table 1, the
χ
2
/N
s
diagnostic values [
15
] which is the 60-day mean
value of
χ
k
2
/N
s
(k)
where
N
s
(k)
is the total number of observations available at time
t
k
and
χ
k
2
=
d
T
k
H
BH
+
R
−1
d
k
. The
χ
2
/N
s
diagnostics [
15
] is closely related to the
J
min
diagnostic used in
variational methods, and should have a value of 1 in the case where the innovations are consistent
with the prescribed error statistics. When there is innovation covariance consistency then
χ
2
/N
s
=
1
but the reverse is not true. We observe that there was a significant improvement from
iter 0
to
iter 1
in
terms of χ
2
/N
s
but is still not equal to one. We thus refer the analysis of iter 1 as near optimal.
The repeated application of this sequence of estimation methods, i.e., find the correlation length
by maximum likelihood and optimize the variance ratio, converges really fast and in practice there is
no need to go beyond
iter 1
. Figure 3 displays iterates 0 to 4 with our estimation procedure for O
3
.
With one iteration update we nearly converge. A similar procedure was used in Ménard [
15
], where
the variance and correlation length (estimated by maximum likelihood) were estimated in sequence,
which taught us that a slow and fictitious drift in estimated variances and correlation length can occur
when the correlation model is not the true correlation. So in regard of similar considerations that may
occur here, we do not extend our iteration procedure beyond the first iterate.
3.2. Statistical Diagnostics of Analysis Error Variance
For each of these experiments, statistics related diagnostics for analysis error variance, discussed
in Section 2, are computed and the results are presented in Table 2 for the verification made against
active observations, and in Table 3 to the verification made against the passive observations. If the
analysis was truly optimal the different diagnostics would all agree.
178
Books
MDPI
Atmosphere 2018, 9,70
Figure 3.
Optimal estimates of
σ
2
o
,
σ
2
b
and maximum likelihood estimate of correlation length
L
c
for the
first four iterates. Blue, is the optimal background error variance, green, the optimal observation error
variance and in red the correlation length (in km, with labels on the right side of the figure).
Table 2. Analysis statistics against active observations.
Experiment
Active
var
(
ˆ
A
− B)
Active
tr
(H
ˆ
A
MDJ
H
T
)/N
s
Active
tr
(H
ˆ
A
D
H
T
)/N
s
Active
var
(O −
ˆ
A
)
Active
tr
(H
ˆ
A
HL
H
T
)/N
s
O
3
iter 0 60.29 22.69 9.61 24.33 −6.03
O
3
iter 1 67.66 13.32 13.68 11.26 8.94
PM
2.5
iter 0 62.29 17.98 7.71 16.78 −3.18
PM
2.5
iter 1 66.3 10.68 9.51 9.57 7.33
In the second, third and last column of Table 2 are tabulated estimates of the analysis error variance
at the active location sites, i.e.,
tr(HAH
T
)/N
s
, obtained by three different methods. The second column
is an estimate given with our method
σ
2
b
− var(
ˆ
A
− B)=tr(H
ˆ
A
MD J
H
T
)/N
s
. The third column is the
Desroziers et al. estimate of analysis error [
14
], Equation (20), and the last column is the estimate using
the method proposed by Hollingsworth and Lönnberg [
13
], Equation (18). We note that the analysis
error variance estimate provided by the first two methods is fairly consistent for an updated correlation
length estimate, i.e.,
iter 1
(but not
iter 0
). We also note that
χ
2
/N
s
is closer to one for
iter 1
. These two
facts indicate that the updated correlation length (
iter 1
) with uniform error variances is closer to the
innovation covariance consistency. The Hollingsworth and Lönnberg [
13
] method however, is very
sensitive and negatively biased in the lack of innovation covariance consistency.
Estimate of the analysis error variance at the passive observation locations, i.e.,
tr(H
c
AH
T
c
)/N
s
,
provided by two different methods are given by Equation (21) in column 3 and by Equation (22) in
column 5 of Table 3. As for the estimate at the active locations (Table 2), there is a general agreement
on the analysis error estimates with the updated correlation length (
iter 1
), although this distinction is
not that clear for PM
2.5
.
Table 3. Analysis statistics against passive observations.
Experiment
Passive
var
[(
ˆ
A
− B)
c
]
Passive
tr
(H
c
ˆ
A
MDJ
H
T
c
)/N
s
Passive
var
[(O −
ˆ
A
)
c
]
Passive
var
[(O −
ˆ
A
)
c
] − σ
2
oc
O
3
iter 0 56.95 26.03 51.02 32.72
O
3
iter 1 52.04 28.95 48.95 28.75
PM
2.5
iter 0 62.29 22.65 38.09 24.49
PM
2.5
iter 1 66.3 24.62 38.28 21.38
We note also that the analysis error variance at the active sites is smaller than the analysis error
variance at the passive observation sites. This involves in particular the fact that since the passive
179
Books
MDPI
Atmosphere 2018, 9,70
observation are away from the active observation sites, the reduction of variance at the passive
observation sites is smaller than at the active observation sites.
3.3. Comparison with the Perceived Analysis Error Variance
We computed the analysis error covariance
A
resulting from the analysis scheme, the so-called
perceived analysis covariance [6], using the expression,
A
=
B
−
BH
T
(H
BH +
R
)
−1
H
B =
B
− GG
T
(23)
Contrary to the statistical diagnostics above, the perceived analysis error variance is obtained
at each model grid point. We then compared the perceived analysis error variance at the active
observation sites with the estimated active analysis error variance obtained from previous diagnostics.
Since we showed that in both active and passive observations sites the different diagnostics agrees
when the analysis is optimal, this would indicate that if the perceived analysis error variance agrees
with the analysis error variance diagnostics, that the whole map of analysis error variance is credible.
In order to calculate the perceived analysis error covariance Equation (23) we first perform a
Choleski decomposition of
H
BH
T
+
R
= LL
T
, where
L
is a lower triangular matrix. Then with a
forward substitution we obtain
L
−1
, from which we compute
G =
BH
T
L
−T
. The perceived analysis
error variance for the ozone optimal analysis (i.e., O
3
iter 1
) is displayed in Figure 4 (A similar figure
but for PM
2.5
is given in supplementary material). We note that although the input statistics used for
the analysis are uniform (i.e., uniform background and observation error variances, and homogeneous
correlation model), the computed analysis error variance at the active observation location displays
large variations, which is attributed to the non-uniform spatial distribution of the active observations.
In Figure 5 we display a histogram of those variances for the ozone optimal analysis O
3
iter 1
(panel
b
) and for the first experiment O
3
iter 0
(panel
a
) without optimization (A similar figure is but
for PM
2.5
is given in supplementary material).
Note that median or mean values of variances are significantly different between the optimal and
non-optimal analysis cases. Although the observation and background errors are uniform in both
optimal and non-optimal analyses, in the optimal analysis case the perceived analysis uncertainty at
the observation location is distributed more equally across all values, indicating that there is a better
propagation of information, measured by analysis error variance, across the different observation sites.
The exception being the isolated observation sites, for which we observe a maxima on the high end of
the histograms. At those sites the analysis error variance is simply obtained by the scalar equation
1
/σ
2
a
=
1
/σ
2
o
+
1
/σ
2
b
. For O
3
iter 1
the scalar analysis error variance gives 16.2, and for O
3
iter 0
we
get 15.0, thus explaining the secondary maxima on the high end of the histogram.
Figure 4.
Analysis error variance for ozone optimal analysis case O
3
iter 1
.(
a
) is the analysis error
on the model grid and (
b
) at the active observation sites. Note that the color bar of the left and right
panels are different. The maximum of the color bar for the left panel correspond to σ
2
o
+ σ
2
b
.
180
Books
MDPI
Atmosphere 2018, 9,70
Figure 5.
Distribution (histogram) of the ozone analysis error variance at the active observation
locations. (
a
) First analysis experiment O
3
iter 0
(no optimization) on the left panel; (
b
) Optimal
analysis case O
3
iter 1. Note that the scales are different between the left and right panels.
The mean perceived analysis error variance for all experiments is presented in Table 4. Comparing
these values with the estimated values of analysis error variance based on diagnostics in Table 2 we
note that for both optimal experiments, O
3
iter 1
and PM
2.5
iter 1
, the perceived analysis error variance
roughly agrees with all analysis error variances estimated with diagnostics (Table 2).
Table 4. Perceived analysis error variance. Mean over active observation sites.
Experiment
Perceived tr
(H
ˆ
A
P
H
T
)/N
s
O
3
iter 0 5.77
O
3
iter 1 11.60
PM
2.5
iter 0 4.37
PM
2.5
iter 1 8.21
However, for the non-optimal analyses, O
3
iter 0
and PM
2.5
iter 0
, there is a general disagreement
between all estimated values. Looking more closely, however, we note that the agreement in the optimal
case is not perfect. The perceived analysis error variance is about 20% lower than the best estimates
tr(H
ˆ
A
MD J
H
T
)/N
s
and
tr(H
ˆ
A
D
H
T
)/N
s
. The optimal
χ
2
/N
s
values in the “optimal” cases are slightly
above one, thus indicating that the innovation covariance consistency is not exact and some further tuning
of the error statistics could be done. More on that matter will be presented in Section 4.5.
4. Discussion on the Statistical Assumptions and Practical Applications
4.1. Representativeness Error with In situ Observations
The statistical diagnostics presented in Section 2 derive from the assumption that the observation
errors are horizontally uncorrelated and uncorrelated with the background error. Although this
assumption is never entirely observed in reality, there are ways to work around it. In the case of
in situ observations, and assuming that any systematic error have been removed, random errors
are still present, due to the difference between the observation and the model’s equivalent of the
observation—called representativeness error (see Janjic et al. [
12
] for a review). Representativeness
error is due to unresolved scales and processes in the model and interpolation or forward observation
model errors. These errors are typically roughly at the scale of the model grid [
21
,
22
], so typically a
few tens of kilometers for air quality models. This should not be confused with the representativeness
of an observation, where, for example, remote stations are representative of large area (e.g., several
181
Books
MDPI
Atmosphere 2018, 9,70
hundreds of kilometers), whereas urban and suburban stations are at the scale of human activity in
the cities, traffic and industries, etc. and are, depending on the chemical specie, of a few kilometers
and less.
Representativeness error of in situ measurements can be discarded altogether by simply filtering
any pair of observations that are in the range of a few model grid sizes, both in assimilation and
estimation of error statistics [
1
] or in pairs of passive-active observations for cross-validation [
17
].
Once this filtering is done, the assumption on observation errors being spatially uncorrelated and
uncorrelated with the background error then applies.
4.2. Correlated Observation-Background Errors
In any case, it is interesting to show how the different diagnostics, introduced in Section 2,
depends on the statistical assumptions of the observation error. One way to get an understanding of
the effect of these assumptions is to look at it from a geometrical point of view, using the representation
introduced in Section 2.2. Note that the same results can be obtained analytically, but the geometrical
interpretation gives a simple and appealing way of looking at the problem.
Let us consider the effect on the analysis of observation error correlated with background error.
The case where the observation error is uncorrelated with background error is represented in Figure 6
on the panel a and when they are correlated on the panel b.
Figure 6.
Geometrical representation of the analysis. (
a
) for observation errors uncorrelated with the
background error. (
b
) with correlated errors. T indicate the truth, O the observation, B the background
and
ˆ
A the optimal analysis.
The observation and background error variances are kept unchanged, with the same
(O
,
T)
length
and
(B
,
T)
length in both panels. In the case of correlated errors the angle
∠BTO
is no longer a right
angle. Yet, it is still possible to obtain an optimal analysis,
ˆ
A
, as a linear combination of the observation
and the background, on the line
(O
,
B)
, for which the distance
ˆ
A
to T (i.e., the analysis error variance)
is minimum. In this case,
(
ˆ
A
,
T)⊥(O
,
B)
. Note that for strongly correlated errors and when
σ
2
b
> σ
2
o
,
although
ˆ
A is still on the line (O, B), it may actually lie outside the segment [O, B]. Yet, the principles
and theory still hold in that case.
When the observation error is uncorrelated with the background error,
(
O, T
)
⊥
(
B, T
)
, the triangles
ΔOT
ˆ
A
and
ΔTB
ˆ
A
are similar and it follows that
(O −
ˆ
A
)(
ˆ
A
− B)
=
ˆ
A
− T
2
, which is the
Desroziers et al. [
14
] diagnostic for analysis error variance. However, when the observation error
is correlated with the background error (right panel of Figure 6), the triangles
ΔOT
ˆ
A
and
ΔTB
ˆ
A
are no longer similar triangles and the Desroziers et al. [
14
] diagnostics for analysis error does
not hold (see derivation in Appendix A). However, the HL, Equation (18), and MDJ diagnostic,
Equation (19), depend only on having right triangles
ΔOT
ˆ
A
and
ΔTB
ˆ
A
, and not on the orthogonality
of
(B, T
)
with
(O, T
)
. Therefore, the HL and MDJ diagnostics are valid with or without correlated
observation-background errors.
182
Books
MDPI
Atmosphere 2018, 9,70
4.3. Estimation of Satellite Observation Errors with In situ Observation Cross-Validation
One of the important problems in satellite assimilation is the estimation of the satellite observation
error, which could be addressed with a simple modification of our cross-validation procedure. Let
us assume that we have in situ observations that we assume to have uncorrelated errors between
themselves (or use a filter with a minimum distance as discussed in Section 4.1), with the background
errors and the satellite observation errors. Yet, the satellite observation errors could be correlated
with the background error. Satellite observations could come from a multi-channel instrument with
channel-correlated observation errors, as found with many instruments, and yet our validation
procedure can still be used. Let us consider that the analyses comprise of satellite and in situ
observations but, for the purpose of cross-validation, we use only 2/3rd of the in situ observations in
the analysis, and keep the remaining 1/3rd as passive to carry out the cross-validation procedure.
The first thing to note is that the passive in situ observations have uncorrelated errors with the
analysis error (the analysis is composed of satellite observations and 2/3rd of the in situ observations).
We then use Equation (7) where the interpolation of the analysis is made only at the in situ active
observations. Minimizing
E[(O − A)
T
c
(O − A)]
(i.e., the trace of the left hand side of Equation (7))
results in finding the optimal in situ observation weight. Then, computing the analysis error covariance
in the satellite observation space from the analysis scheme (either from a Hessian of a variational
cost function, or with an explicit gain as in Equation (23)), i.e., H
sat
ˆ
AH
T
sat
, we use the HL formulation
Equation (18) to obtain the satellite observation error covariance,
H
sat
ˆ
AH
T
sat
+ E[(O − A)
sat
(O − A)
T
sat
]=R
sat
(24)
The Equation (24) has the important properties that the estimated observation error covariance is
symmetric and positive definite by construction. Then, a new analysis could be carried out to obtain a
more realistic H
sat
ˆ
AH
T
sat
, with a resulting updated R
sat
, and so forth until convergence.
4.4. Remark on Cross-Validation of Satellite Retrievals
As a last remark, it appears that cross-validation of satellite retrieval observations using a k-fold
approach where the observations are used as passive observations to validate the analysis can be
a difficult problem. Retrievals from passive remote sensing at nadir generally involve a prior or
climatology or a model assumption over different regions, and is thus likely to have spatially correlated
errors and errors correlated with the background error. It does not mean, however, that nothing can
be done in that case. For example, for certain sensors, such as infrared sensors, it is possible to
disentangle the prior from the retrieval, so that by an appropriate transformation of the measurements,
observations can be practically decorrelated from the background [
23
,
24
]. However to the authors’
knowledge, such an approach have never been undertaken for visible measurements such as for NO
2
or AOD’s.
4.5. Lack of Innovation Covariance Consistency and Its Relevance to the Statistical Diagnostics
The error covariance diagnostics for optimal analysis, presented in Sections 2.4 and 2.5, depends
on the innovation covariance consistency,
E[(O − B)(O − B)
T
]=H
BH
T
+
R
, and our results presented
in Section 3 have shown that the different estimates for the optimal analysis error variance are close,
but do not strictly agree to each other. This disagreement is related to the lack of innovation consistency
as follows.
Let us introduce a departure matrix Δ from innovation covariance consistency as,
E
[dd
T
](H
BH
T
+
R
)
−1
= I + Δ (25)
183
Books
MDPI
Atmosphere 2018, 9,70
The trace of Equation (25), which is related to χ
2
, is given by
E
[χ
2
]=tr{E[dd
T
](H
BH
T
+
R
)
−1
} = N
s
+ tr(Δ) (26)
We recall that in the experiment
iter 1
we got
χ
2
/N
s
values of 1.36 for O
3
and 1.25 for PM
2.5
(see
Table 1), indicating that the innovation covariance consistency is deficient, although less serious than
with the experiment iter 0 where values of 2 and higher have been obtained.
If we take into account the fact that there can be a difference between
E
dd
T
and
(H
BH
T
+
R
)
and
we rederive the (active) analysis error covariance for HL, MDJ and D schemes, we get (see Appendix B)
tr{H
ˆ
A
HL
H
T
} = tr{H
ˆ
A
true
H
T
} + tr{
RΔ
}−tr{H
BH
T
Δ} + tr{error
MD J
} (27)
tr
{H
ˆ
A
MD J
H
T
} = tr{H
ˆ
A
true
H
T
}−tr{error
MD J
} (28)
tr
{H
ˆ
A
D
H
T
} = tr{H
ˆ
A
true
H
T
}−tr{error
D
} (29)
where
tr{error
MD J
} = tr {(H
BH
T
)(H
BH
T
+
R
)
−1
Δ(H
BH
T
)}
,
tr{error
D
} =
tr{
R
(H
BH
T
+
R
)
−1
Δ(H
BH
T
)}. We note that although the error terms are complex expressions, they
all depend linearly on
Δ
. Thus, the disagreement between the HL, MDJ and D analysis error variance
estimates is due to lack of innovation covariance consistency.
5. Conclusions
We showed that analysis error variance can be estimated and optimized, without using a model
forecast, by partitioning the original observation data set into a training set, to create the analysis,
and an independent (or passive) set, used to evaluate the analysis. This kind of evaluation by
partitioning is called cross-validation. The method derives from assuming that the observations
have spatially uncorrelated errors or, minimally, that the independent (or passive) observations have
uncorrelated errors with the active observation, and are uncorrelated the background error. This leads
to the important property that passive observations are uncorrelated with the analysis error and can
then be used to evaluate the analysis [17].
We have developed a theoretical framework and a geometric interpretation that has allowed
us to derive a number of statistical estimation formulas of analysis error covariance that can be
used in both passive and active observation spaces. It is shown that by minimizing the variance of
observation-minus-analysis residuals in passive observation space we actually identify the optimal
analysis. This has been done with respect to a single parameter, namely the ratio of observation to
background error variances, to obtain a near optimal Kalman gain. The optimization is also done under
the constraint of the innovation covariance consistency [
14
,
15
]. This optimization could have been
done with more than one error covariance parameter but this has not been attempted here. The theory
does suggest, however, that the minimum is unique.
Once an optimal analysis is identified we conduct an evaluation of the analysis error covariance
using several different formulas; Desroziers et al. [
14
], Hollingsworth Lönnberg [
13
], and one that we
develop in this paper which works in either active or passive observation spaces. As a way to validate
the analysis error variance computed by the analysis scheme itself, the so-called perceived analysis
error variance [6], we compare it with the values obtained from the different statistical diagnostics of
analysis error variance.
This methodology arises from a need to assess and improve ECCC’s surface air quality analyses
using our operational air quality model GEM-MACH and real-time surface observations of O
3
and
PM
2.5
. Our method applied the theory in a simplified way. First by considering the averaged
observation and background error variances and finding an optimal ratio
γ = σ
2
o
/σ
2
b
using as a
constraint the trace of the innovation covariance consistency [
15
]. Second, using a single parameter
correlation model, its correlation length, we used the maximum likelihood estimation [
11
] to obtain near
184
Books
MDPI
Atmosphere 2018, 9,70
optimal analyses. Also we did not attempt to account for representativeness error in the observations
by, for example, filtering observations that are close. Despite all these limitations, our results show
that with near optimal analyses, all estimates of analysis error variance roughly agree with each other,
while disagreeing strongly when the input error statistics are not optimal. This check on estimating
the analysis error variance gives us confidence that the method we propose is reliable, and provides
us an objective method to evaluate different analysis components configurations, such as the type of
background error correlation model, the spatial distribution of error variances and possibly the use of
thinning observations to circumvent effects of representativeness errors.
The methodology introduced here for estimating analysis error variances is general and not
restricted to the case of surface pollutant analysis. It would be desirable to investigate other areas
of applications, such as surface analysis in meteorology and oceanography. The method could,
in principle, provide guidance for any assimilation system. By considering the observation space
subdomain [
25
], proper scaling, local averaging [
26
], or other methods discussed in Janjic et al. [
12
]
it may also be possible to extend this methodology to spatially varying error statistics. Based on our
verification results in Part I [
5
], we found that there is a dependence between model values and error
variances, which we will investigate further in view of our next operational implementation of the
Canadian surface air quality analysis and assimilation.
One strong limitation of the optimum interpolation scheme we are using (i.e., homogeneous
isotropic error correlation and uniform error variances), which is also the case for most 3D-Var
implementations, is the lack of innovation covariance consistency. Ensemble Kalman filters seem,
however, much better in that regard although they have their own issues with localization and inflation.
Experiments with chemical data assimilation using an ensemble Kalman filter does gives
χ
2
/N
s
values
very close to unity after simple adjustments for observation and model error variances [
27
]. We thus
argue that ensemble methods, such as the ensemble Kalman filter, would produce analysis error
variance estimates that are much more consistent between the different diagnostics.
Estimates of analysis uncertainties can also be obtained by resampling techniques, such as the
jackknife method and bootstrapping [
28
]. In bootstrapping with replacement, the distribution of the
analysis error is obtained by creating new analyses by replacing and duplicating observations from
an existing set of observations [
28
]. This technique relies on the assumption that each member of
the dataset is independent and identically distributed. For surface ozone analyses where there is
persistence to next day and the statistics is spatially inhomogeneous, the assumption of statistical
independence may not be adequate. The comparison of these resampling estimates of analysis
uncertainties could be compared with our analysis error variance estimates to help us identify
limitations and areas of improvement.
Supplementary Materials:
The following are available online at http://www.mdpi.com/2073-4433/9/2/70/s1,
Figure S1: Analysis error variance for ozone optimal analysis case PM
2.5
iter 1
. Left panel is the analysis error on
the model grid and on the right panel at the active observation sites. Note that the color bar of the left and right
panels are different. The maximum of the color bar for the left panel correspond to
σ
2
o
+ σ
2
b
, Figure S2: Distribution
(histogram) of the ozone analysis error variance at the active observation locations. First analysis experiment
PM
2.5
iter 0 (no optimization) on the left panel, and optimal analysis case PM
2.5
iter 1 on the right panel.
Acknowledgments:
We are grateful to the US/EPA for the use of the AIRNow database for surface pollutants
and to all provincial governments and territories of Canada for kindly transmitting their data to the Canadian
Meteorological Centre to produce the surface analysis of atmospheric pollutants. We are also thankful for the
proof read by Kerill Semeniuk, and for two anonymous reviewers for their comments and help in improving
the manuscript.
Author Contributions:
This research was conducted as a joint effort by both authors. R.M. contributed to the
theoretical development and wrote the paper, and M.D.-J. conducted all experiments design and execution,
proof reading and introduced a new diagnostic of optimal analysis error that was further extended to passive
observations space.
Conflicts of Interest:
The authors declare no conflict of interest. The founding sponsors, which is the government
of Canada, had no role in the design of the study; in the collection, analyses, or interpretation of data; in the
writing of the manuscript, and in the decision to publish the results.
185
Books
MDPI
Atmosphere 2018, 9,70
Appendix A. A Geometrical Derivation of the Desroziers et al. Diagnostic
Let
u
and
v
be two random variables of a real Hilbert space as defined in Section 2.3.
Two properties of Hilbert spaces are: the polarization identity
u, v
=
1
4
u + v
2
−u − v
2
!
(A1)
and the parallelogram identity [19]
u + v
2
+ u − v
2
= 2
u
2
+ v
2
(A2)
Combining these two equations, we get:
u, v
=
1
2
u + v
2
−u
2
−v
2
!
(A3)
Let u
= O − A and v = A − B, then u + v = O − B, and so we have
(
O − A), (A − B)
=
1
2
O − B
2
−O − A
2
−A − B
2
!
(A4)
If we assume that the analysis is optimal, so that
∠OAT is a right triangle (see Figure 1), then
ˆ
A
− T
2
+ O −
ˆ
A
2
= O − T
2
(A5)
and similarly that
∠TAB is a right triangle,
ˆ
A
− T
2
+
ˆ
A
− B
2
= B − T
2
(A6)
and substitute these expression into Equation (A4) we get
(O −
ˆ
A
), (
ˆ
A
− B)
=
1
2
O − B
2
−O − T
2
−B − T
2
− 2
ˆ
A
− T
2
!
(A7)
If in addition we assume that we have uncorrelated observation-background errors, that is
O − B
2
= O − T
2
+ B − T
2
(A8)
we then get
(O −
ˆ
A
), (
ˆ
A
− B)
=
ˆ
A
− T
2
(A9)
Note that the difference with the result obtained in Appendix A (property 6) of Ménard [
15
] where
it is shown that the necessary and sufficient condition for the perceived analysis error covariance in
active observation space to meet the Desroziers et al. diagnostics for analysis error, is to have innovation
covariance consistency and uncorrelated observation-background errors. The result obtained here is
different; it concerns the optimal analysis error covariance.
Appendix B. Diagnostics of Analysis error Covariance and the Innovation
Covariance Consistency
Let us introduce a departure matrix Δ from innovation covariance consistency as,
E
[dd
T
](H
BH
T
+
R
)
−1
= I + Δ (A10)
186
Books
MDPI
Atmosphere 2018, 9,70
The HL diagnostic, Equation (18), can be expanded as
E
[(O −
ˆ
A
)(O −
ˆ
A
)
T
]=E[dd
T
]+H
BH
T
(H
BH
T
+
R
)
−1
E[dd
T
](H
BH
T
+
R
)
−1
−E[dd
T
](H
BH
T
+
R
)
−1
H
BH
T
− H
BH
T
(H
BH
T
+
R
)
−1
E[dd
T
]
=
R
− H(
B
−
BH
(H
BH
T
+
R
)
−1
H
B)H
T
+ error
HL
=
R
− H
ˆ
AH
T
+ error
HL
(A11)
where the error
HL
is given as
error
HL
= Δ
R − (H
BH
T
)Δ
T
+ H
BH
T
(H
BH +
R
)
−1
Δ(H
BH
T
) (A12)
which includes three error terms. A scalar version of Equation (A11) is given as
error
HL
=
σ
2
b
Δ
1 + γ
− σ
2
b
Δ + σ
2
o
Δ (A13)
The error analysis for the MDJ diagnostic, Equation (19) is
E
[(
ˆ
A
− B)(
ˆ
A
− B)] = H
KE[dd
T
]
KH
T
= H
BH
T
(H
BH
T
+
R
)
−1
H
BH
T
+ error
MD J
(A14)
where the error
MD J
is given as
error
MD J
= H
BH
T
(H
BH
T
+
R
)
−1
Δ(H
BH
T
) (A15)
There is only one term, and its scalar version is given as,
error
MD J
=
σ
2
b
Δ
1 + γ
(A16)
The error analysis for the Desroziers et al. diagnostic, Equation (20) is
E[(O −
ˆ
A
)(
ˆ
A
− B)] = E[dd
T
](H
BH
T
+
R
)
−1
H
BH
T
− H
BH
T
(H
BH
T
+
R
)E[dd
T
](H
BH
T
+
R
)
−1
H
BH
T
= H
BH
T
− H
BH
T
(H
BH
T
+
R
)
−1
H
BH
T
+ error
D
(A17)
where the error
D
is given as
error
D
= {I − H
BH
T
(H
BH
T
+
R
)
−1
}Δ(H
BH
T
)=
R
(H
BH
T
+
R
)
−1
Δ(H
BH
T
) (A18)
Again only one error term that is similar to the MDJ diagnostic, and its scalar version is given as,
error
D
=
σ
2
o
Δ
1 + γ
(A19)
Error analysis for the diagnostics using passive observations can also be derived. For the passive
MDJ diagnostics, Equation (21), we have similarly to (A14),
E
[(
ˆ
A
− B)
c
(
ˆ
A
− B)
c
]=H
c
KE
[dd
T
]
KH
T
c
= H
c
BH
T
(H
BH
T
+
R
)
−1
H
BH
T
c
+ error
MD J_passive
(A20)
with a single error term given as,
error
MD J_passive
= H
c
BH
T
(H
BH
T
+
R
)
−1
Δ(H
BH
T
c
) (A21)
187
Books
MDPI
Atmosphere 2018, 9,70
To express a scalar version of this equation we need to account for the background error correlation
ρ
between the active observation location and the passive observation location, and thus expressed as,
error
MD J_passive
=
ρ
2
σ
2
b
Δ
1 + γ
(A22)
Finally, the fundamental diagnostic of cross-validation Equation (7) does not depend explicitly on
the innovation covariance consistency. However, attaining its true minimum by tuning only
γ
and
L
c
as would suggest Equation (22), does introduce some innovation in-consistency, which all other
optimal diagnostics Equations (18)–(21) has to account for.
References
1.
Ménard, R.; Robichaud, A. The chemistry-forecast system at the Meteorological Service of Canada.
In Proceedings of the ECMWF Seminar Proceedings on Global Earth-System Monitoring, Reading, UK,
5–9 September 2005; pp. 297–308.
2.
Robichaud, A.; Ménard, R. Multi-year objective analysis of warm season ground-level ozone and PM
2.5
over
North-America using real-time observations and Canadian operational air quality models. Atmos. Chem. Phys.
2014, 14, 1769–1800. [CrossRef]
3.
Robichaud, A.; Ménard, R.; Zaïtseva, Y.; Anselmo, D. Multi-pollutant surface objective analyses and mapping
of air quality health index over North America. Air Qual. Atmos. Health
2016
, 9, 743–759. [CrossRef] [PubMed]
4.
Moran, M.D.; Ménard, S.; Pavlovic, R.; Anselmo, D.; Antonopoulus, S.; Robichaud, A.; Gravel, S.; Makar, P.A.;
Gong, W.; Stroud, C.; et al. Recent advances in Canada’s national operational air quality forecasting system.
In Proceedings of the 32nd NATO-SPS ITM, Utrecht, The Netherlands, 7–11 May 2012.
5.
Ménard, R.; Deshaies-Jacques, M. Evaluation of analysis by cross-validation. Part I: Using verification
metrics. Atmosphere 2018, in press.
6.
Daley, R. The lagged-innovation covariance: A performance diagnostic for atmospheric data assimilation.
Mon. Weather Rev. 1992, 120, 178–196. [CrossRef]
7. Daley, R. Atmospheric Data Analysis; Cambridge University Press: New York, NY, USA, 1991; p. 457.
8.
Talagrand, O. A posteriori evaluation and verification of analysis and assimilation algorithms. In Proceedings
of Workshop on Diagnosis of Data Assimilation Systems, November 1998; European Centre for Medium-Range
Weather Forecasts: Reading, UK, 1999; pp. 17–28.
9.
Todling, R. Notes and Correspondence: A complementary note to “A lag-1 smoother approach to
system-error estimaton”: The intrinsic limitations of residuals diagnostics. Q. J. R. Meteorol. Soc.
2015
,
141, 2917–2922. [CrossRef]
10.
Hollingsworth, A.; Lönnberg, P. The statistical structure of short-range forecast errors as determined from
radiosonde data. Part I: The wind field. Tellus 1986, 38A, 111–136. [CrossRef]
11.
Ménard, R.; Deshaies-Jacques, M.; Gasset, N. A comparison of correlation-length estimation methods for
the objective analysis of surface pollutants at Environment and Climate Change Canada. J. Air Waste
Manag. Assoc. 2016, 66, 874–895. [CrossRef] [PubMed]
12.
Janjic, T.; Bormann, N.; Bocquet, M.; Carton, J.A.; Cohn, S.E.; Dance, S.L.; Losa, S.N.; Nichols, N.K.;
Potthast, R.; Waller, J.A.; et al. On the representation error in data assimilation. Q. J. R. Meteorol. Soc.
2017. [CrossRef]
13.
Hollingsworth, A.; Lönnberg, P. The verification of objective analyses: Diagnostics of analysis system
performance. Meteorol. Atmos. Phys. 1989, 40, 3–27. [CrossRef]
14.
Desroziers, G.; Berre, L.; Chapnik, B.; Poli, P. Diagnosis of observation-, background-, and analysis-error
statistics in observation space. Q. J. R. Meteorol. Soc. 2005, 131, 3385–3396. [CrossRef]
15.
Ménard, R. Error covariance estimation methods based on analysis residuals: Theoretical foundation and
convergence properties derived from simplified observation networks. Q. J. R. Meteorol. Soc.
2016
, 142,
257–273. [CrossRef]
16.
Kailath, T. An innovation approach to least-squares estimation. Part I: Linear filtering in additive white
noise. IEEE Trans. Autom. Control 1968, 13, 646–655.
188
Books
MDPI
Atmosphere 2018, 9,70
17.
Marseille, G.-J.; Barkmeijer, J.; de Haan, S.; Verkley, W. Assessment and tuning of data assimilation systems
using passive observations. Q. J. R. Meteorol. Soc. 2016, 142, 3001–3014. [CrossRef]
18.
Waller, J.A.; Dance, S.L.; Nichols, N.K. Theoretical insight into diagnosing observation error correlations
using observation-minus-background and observation-minus-analysis statistics. Q. J. R. Meteorol. Soc.
2016
,
142, 418–431. [CrossRef]
19. Caines, P.E. Linear Stochastic Systems; John Wiley and Sons: New York, NY, USA, 1988; p. 874.
20.
Cohn, S.E. The principle of energetic consistency in data assimilation. In Data Assimilation; Lahoz, W.,
Boris, K., Richard, M., Eds.; Springer: Berlin/Heidelberg, Germany, 2010.
21.
Mitchell, H.L.; Daley, R. Discretization error and signal/error correlation in atmospheric data assimilation:
(I). All scales resolved. Tellus 1997, 49A, 32–53. [CrossRef]
22.
Mitchell, H.L.; Daley, R. Discretization error and signal/error correlation in atmospheric data assimilation:
(II). The effect of unresolved scales. Tellus 1997, 49A, 54–73. [CrossRef]
23.
Joiner, J.; da Silva, A. Efficient methods to assimilate remotely sensed data based on information content.
Q. J. R. Meteorol. Soc. 1998, 124, 1669–1694. [CrossRef]
24.
Migliorini, S. On the quivalence between radiance and retrieval assimilation. Mon. Weather Rev.
2012
, 140,
258–265. [CrossRef]
25.
Chapnik, B.; Desroziers, G.; Rabier, F.; Talagrand, O. Properties and first application of an error-statistics
tunning method in variational assimilation. Q. J. R. Meteorol. Soc. 2005, 130, 2253–2275. [CrossRef]
26.
Ménard, R.; Deshiaes-Jacques, M. Error covariance estimation methods based on analysis residuals and its
application to air quality surface observation networks. In Air Pollution and Its Application XXV; Mensink, C.,
Kallos, G., Eds.; Springer International AG: Cham, Switzerland, 2017.
27.
Skachko, S.; Errera, Q.; Ménard, R.; Christophe, Y.; Chabrillat, S. Comparison of the ensemlbe Kalman filter
and 4D-Var assimilation methods using a stratospheric tracer transport model. Geosci. Model Dev.
2014
, 7,
1451–1465. [CrossRef]
28. Efron, B. An Introduction to Boostrap; Chapman & Hall: New York, NY, USA, 1993; p. 436.
©
2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
189
Books
MDPI
atmosphere
Article
Overview of the Model and Observation Evaluation
Toolkit (MONET) Version 1.0 for Evaluating
Atmospheric Transport Models
Barry Baker
1,2,
* and Li Pan
1,2
1
NOAA Air Resources Laboratory, College Park, MD 20740, USA; li.pan@noaa.gov
2
Cooperative Institute for Climate and Satellites, University of Maryland at College Park,
College Park, MD 20740, USA
* Correspondence: Barry.Baker@noaa.gov; Tel.: +1-301-683-1373
Received: 16 June 2017; Accepted: 21 October 2017; Published: 31 October 2017
Abstract:
This paper describes the development and initial applications of the Model and Observation
Evaluation Tool (MONET) v1.0. MONET was developed to evaluate the Community Multiscale Air
Quality Model (CMAQ) for the NOAA National Air Quality Forecast Capability (NAQFC) modeling
system. MONET is designed to be a modularized Python package for (1) pairing model output to
observational data in space and time; (2) leveraging the pandas Python package for easy searching
and grouping; and (3) analyzing and visualizing data. This process introduces a convenient method
for evaluating model output. MONET processes data that is easily searchable and that can be grouped
using meta-data found within the observational datasets. Common statistical metrics (e.g., bias,
correlation, and skill scores), plotting routines such as scatter plots, timeseries, spatial plots, and more
are included in the package. MONET is well modularized and can add further observational datasets
and different models.
Keywords: CMAQ; evaluation; air quality; software; visualization; statistics
1. Introduction
Ozone (O
3
) and particulate matter smaller than 2.5
μ
m in diameter (PM
2.5
) are among a handful of
criteria air pollutants—pollutants the Clean Air Act requires to be monitored and regulated—that are
primarily responsible for adverse impacts on human health [
1
]. Breathing these pollutants is recognized
as major causes of acute and chronic respiratory and cardiovascular diseases, and premature mortalities
associated with air pollution [2].
In relation to health hazards caused by air pollutants, air quality also has a direct impact on the
economy. Trasande et al. [
3
] found the cost of medical care for preterm births attributable to PM
2.5
exposure to be between $2.43 and $9.66 billion. Ghude et al. [
4
] demonstrates the economic cost of
poor air quality in India to crop yields is estimated to be around $1.26 billion annually. Tong et al. [5]
reveals a similar influence to crop yields in the United States (U.S.).
Evaluating model simulations is critical for the development and implementation for
forecasting [
6
]. Evaluation of meteorological and air quality parameters are essential in validating and
improving model simulations.
Recently, air quality simulations increased from running for days or weeks to months or years,
modeling domains increased resolution, and the use of ensembles greatly increased the amount of
model output analyzed. The Community Multiscale Air Quality (CMAQ) modeling simulations,
used for air quality and air composition modeling, currently run spanning modeling domains of
regional to hemispheric scale and modeling times of days to years resulting in terabytes of model output
available for analysis. Conventional methods of data analysis, such as spreadsheets, are not suited for a
Atmosphere 2017, 8, 210 190 www.mdpi.com/journal/atmosphere
Books
MDPI
Atmosphere 2017, 8, 210
task this large. There are several model evaluation tools available for evaluating meteorological model
simulations but few for evaluating air quality models [
7
]. Although there are many software packages,
such as AMET, or visualization tools, such as Verdi, pycmbs, or Panoply, MONET is built using a
single open-source language that retrieves, downloads, and analyses both model and observations
on a regional scale. MONET is available on all major computing platforms, i.e., Mac, Linux, and PC,
and provides the power of a dedicated programming language such as IDL, MATLAB, or ferret
if needed.
The National Oceanic and Atmospheric Administration (NOAA) Air Resources Laboratory
(ARL) developed the Model and Observation Evaluation Tool (MONET) to aid in the assessment
of the National Air Quality Forecasting Capability (NAQFC) [
8
,
9
]. MONET reads, interpolates,
and organizes model results to observation sites in both space and time resulting in a fast and
flexible method to evaluate air quality modeling simulations. Although MONET was originally
created to only evaluate CMAQ simulations, it can easily be expanded to include different model
outputs (e.g., the Weather Research and Forecasting model, Next Generation Global Prediction System,
and Comprehensive Air Quality Model with Extensions), along with adding additional observational
sources (e.g., different ground based networks, satellite observations, and other in-situ observations).
This paper describes the structure and functionality of the MONET Python package (version 2.7).
A broad description of MONET will be provided, followed by detailed description of how MONET
works. Examples of the analysis products available from MONET will be presented and described.
Finally, a discussion and future directions of MONET will be provided.
2. Tool Description
MONET pairs observations and gridded model prediction in space and time. This evaluates
the model’s performance for a set of predicted or diagnosed fields such as aggregating nitrous oxide
(NO) and nitrogen dioxide (NO
2
) into nitrogen oxides (NO
x
). MONET is built in Python v2.7+ and is
intended to follow object oriented concepts. Each model and observation has a specific object intended
to be used for its specific cases. Stemming from such object structures, a verification object that inherits
observation and model objects can be created allowing for targeted verification between sets of models
and observations. MONET runs interactively or with a simple adaptable Python script.
Most of the dependencies can be obtained using commonly available Python packages, such
as the Anaconda Python Package or the Enthought Canopy distribution. MONET uses the pandas
Python package [
10
,
11
] to enable fast and efficient data manipulation using meta-data available in
the observational datasets. Pairing of model and observations is done using the pyresample package,
(available online: https://pyresample.readthedocs.io/en/latest/), which uses pykdTree to interpolate.
Several interpolation methods are available; nearest neighbor, Gaussian, elliptical weighted averaging,
or a user defined method, such as inverse distance weighting. Analysis is done using scipy, numpy,
and scikit-learn Python functions, while plotting is achieved with Basemap (for spatial plotting),
matplotlib, and seaborn [12,13]. Figure 1 presents a flowchart of the MONET software.
MONET runs in four phases; (1) creation of a model object that determines the model
grid, variable names, and run duration; (2) creation of an observational object that parses
observational data; (3) combines and interpolates model results to observations; and (4) plotting
and statistical comparisons.
191
Books
MDPI
Atmosphere 2017, 8, 210
Figure 1. Flow chart of Model and ObservatioN Evaluation Tool (MONET).
2.1. Creation of Model and Observation Objects
MONET is designed to be modular in that each observational network or model
used has its own set of specialized functions to handle the differences in each dataset.
Currently, MONET can handle the hourly Environmental Protection Agency AirData (EPA AQS;
available online:
https://aqsdr1.epa.gov/aqsweb/aqstmp/airdata/download_files.html#Raw
),
including the Chemical Speciation Network (CSN) and AirNow (available online: https://www.
airnow.gov) datasets, along with the Interagency Monitoring of Protected Visual Environments
(IMPROVE; available online:
http://vista.cira.colstate.edu/improve
) and Aerosol Robotic Network
(AERONET; available online: http://aeronet.gsfc.nasa.gov) datasets. Datasets can be readily
downloaded through an FTP or http by using an array of datetime objects given to the observation
object. Other networks, such as IMPROVE, require that the data be downloaded manually.
Each observation network is designed to monitor specific sets of air quality parameters,
from individual pollutants to derived physical parameters (e.g., AOD). MONET makes it easy
to compare individual pollutants or to aggregate model results to pair with the monitor dataset.
Examples of this are NO
x
(NO+NO
2
) in AQS or AIRNOW or particle sulfate, an aggregate of the
Aitken, accumulation, and coarse mode contribution to the sulfate particle concentration within CMAQ,
from AQS and IMPROVE. MONET includes a set of standard aggregations for different pollutants but
is versatile enough to allow user defined aggregations.
The observation objects require that a time range is given as an array of datetime objects to the
object instance attribute. This is used to tailor data extraction from a pre-downloaded file covering the
prescribed time range or otherwise expand the data retrieval. Then, observation objects read the raw
datafile and process it into a pandas DataFrame, thus assigning it to an object instance with common
columns of observational data (Obs), date (datetime), local date (datetime_local), latitude (Latitude),
and longitude (Longitude). Depending on the network, more meta-data may be available, meaning
that an observational object in MONET can be used as a standalone method to analyze observational
data without the need for pairing. If adding a new network, only the latitude, longitude, observations,
and date or timestamp is needed for the interpolation.
For model objects, the only information needed is the output file(s). In the case of overlapping
simulations, the model results with the latest creation time stamp will be used. For instance, if two 48 h
simulations with 24 h of overlap are given to MONET, then the first 24 h from the first simulation and
the 48 h of the second day are used. MONET assigns the model object file to a class instance attribute
and creates corresponding helper functions, such as that defining map projections for spatial plotting
192
Books
MDPI
Atmosphere 2017, 8, 210
and retrieving and aggregating data. In the future, the model object will be done with the xarray
Python package to keep a consistent view with observations. The xarray package is an N-Dimensional
implementation of the pandas library that is highly suitable for scientific data that allows for larger
then memory computations and efficient data extraction and resampling.
2.2. Pairing Observations and Model Results
MONET has a built-in driver for each model and observation pair, meaning that the model and
observation are separately imported objects. This creates a simple and efficient method for adding new
model results—once a model and observational dataset has its own Python object it can easily follow
the same pairing algorithms. The driver creates the model object, reads the time range, and creates the
model instance. The model time range is then provided to the observational object, where the data is
downloaded and processed into a pandas DataFrame or loaded from a preprocessed file.
At this point, each species available in observations and model prediction are paired. MONET
has the ability to calculate or aggregate data, such as the calculation of PM
2.5
(particulate matter
with diameter less than 2.5
μ
m) mass concentration, from several modeled chemical species. Herein,
a helper function retrieves the data in the model object to derive the needed data array. MONET then
interpolates the model results for each timestamp that is available in the observational network to the
observational site and merges them into a common pandas DataFrame containing pairings of model
results and observations for all specified observation locations.
The Pyresample library (available online: https://pyresample.readthedocs.io/en/latest/)
interpolates the model to observations. Pyresample uses (k-dimensional) KDTrees to transform
data defined in one geometrical grid to another. Resampling uses the nearest neighbor, Gaussian,
or a customized method, such as inverse distance weighting. MONET allows the user to define the
radius of influence and number of neighbors to be used for data re-gridding.
After re-gridding, depending on the dataset, MONET resamples in time and then assigns these
processed data to a separate pandas DataFrame for deriving certain species, such as the 8 h max ozone
or the daily PM
2.5
concentration. Timing for opening model results, processing of observation data,
and pairing takes less than 3 min with AIRNOW for a 48 h simulation. At this point, MONET proceeds
to analyze the dataset with the observation and model result pairs.
3. Example of Tool Applications
MONET is a new software package developed for verifying the NAQFC and more generally the
CMAQ model. It has been newly developed as described above and is available on GitHub. Detailed
examples and installation instructions are available at https://github.com/noaa-oar-arl/MONET.
Once MONET makes the verification object, all of the non-spatial plots, such as timeseries, scatter
plots, and histograms, are executed through as a single line-command with flags specifying the user’s
selection of geographical extent and variables to be plotted. For instance, a user can specify that
the geographical extent of their plot be within the U.S. EPA defined Conterminous United States
(CONUS) region. MONET performs verification within the entire domain, a region, a state, a county,
a metropolitan area, or a specific site. MONET also compares multiple simulations on a single figure
simply by creating two verification objects and passing the figure handle. Figures 2–6 showcases some
of the plotting routines found within MONET.
As an example of how easily MONET does a quick pairing and analysis, the following example
shows how to use MONET to pair AirNow data and a CMAQ simulation. Begin by entering an
interactive python session.
ipython –pylab (1)
Import the MONET object. This is the python object interface to the model and observation objects.
193
Books
MDPI
Atmosphere 2017, 8, 210
import monet (2)
Set the concentration file or files, in this case concfiles, and the gridcro2d file, gridcro.
concfiles = ‘/path/to/concentration/files/ *.ncf’ (3)
gridcro = ‘/path/to/gridcro2dfile.ncf’ (4)
Use the MONET object to pair the AirNow data and CMAQ simulation by creating an instance of
the verified airnow object, which imports both the CMAQ and AIRNOW objects.
m = monet.vairnow (concobj = concfiles, gridobj = gridcro) (5)
MONET retrieves the observations if not in the current directory and the model is interpolated in
space to the observational data points. The paired data can be accessed through the pandas DataFrame
contained within the monet object, m.
m.df.head() (6)
Then simple commands can be used to create different plots. For example, MONET creates a
timeseries plot averaged over all of the observations found within the modeling domain with a simple
one line execution as follows:
m.compare_param (param = ‘OZONE’, timeseries = True) (7)
Figure 2a–d shows a time series of the average concentration over the northeastern United States
during the summer of 2016. The average value is shown in solid lines, observations in black, and model
results in blue and purple. Shaded areas display one standard deviation from the mean. The footer
illustates forecasting performance statistical measures specified by the user. The footer also shows the
start date, end date, number of sites, and the number of measurements. Figure 2b presents an example
of time series root mean square error plot comparing how two simulations performed for the EPA AQS
sites in the U.S. EPA CONUS domain. Likewise, Figure 2c displays a mean bias time series of the two
simulations. Figure 2d provides an example of speciated PM
2.5
sulfate using the IMPROVE network
over the Pacific region. With the IMPROVE network daily average measurement every three days,
MONET interpolates the hourly CMAQ results to each individual site and then averaged to create a
daily concentration as if were an IMPROVE measurement. Then, MONET merges the daily averaged
CMAQ results into the IMPROVE DataFrame using a mysql like merge found within Pandas.
(a)
Figure 2. Cont.
194
Books
MDPI
Atmosphere 2017, 8, 210
(b)
(c)
(d)
Figure 2.
Examples of time-series (
a
) Hourly averaged values are shown in solid lines for EPA AQS
observations (black), and two different simulations, National Air Quality Forecast Capability (NAQFC)
(blue) and NAQFC-Beta (purple). Shaded regions represent one standard deviation for each hourly
averaged time series; (
b
) Hourly averaged values of the root mean square error (RMSE) from EPA
AQS observations are shown for two different simulations of PM
2.5
, NAQFC (blue) and NAQFC-Beta
(purple); (
c
) Hourly averaged values of the RMSE from EPA AQS observations are shown for two
different simulations of PM
2.5
, NAQFC (blue) and NAQFC-Beta (purple); (
d
) Daily averaged values are
shown in solid lines for IMPROVE observations (black), and two different simulations, NAQFC (blue)
and NAQFC-Beta (purple) in the Pacific region. Shaded regions represent one standard deviation for
each daily averaged time series.
195
Books
MDPI
Atmosphere 2017, 8, 210
Figure 3a presents an example of a kernel density estimation (KDE) comparison. The black line
shows an observational KDE and the blue represents the CMAQ model results. This type of plot
(Figure 3a) is useful for inspection for distributional discrepancy between the model and observations.
Using conditional distributions, such as conditioning on time of day or other meteorological factors
like boundary layer height or wind speed, one can provide a more in depth analysis using MONET.
Figure 3b displays the difference in the KDE between model and observation results. Difference KDE
are important to show model biases in the most probable value.
(a)
(b)
Figure 3.
Examples of kernel density estimations (KDEs) (
a
) Daytime kernel density estimations
for O
3
concentrations over the continental United States (U.S.) for both observations (black) and a
Community Multiscale Air Quality Model (CMAQ) simulation (blue); (
b
) A kernel density estimation
of the difference between the simulated NO
2
concentrations and observations.
Figure 4a shows an example of a spatial plot of model results against overlaid observations.
The colorbar is customizable by passing colormaps that can be user defined or predefined in matplotlib.
The colorbar levels can be tailored individually to any single observation. By default, the colorbar
is scaled automatically based on the model data and a discrete colorbar with 15 value bins using
the ‘viridis’ colormap. The ‘viridis’ colormap is perceptually uniform and sensitive for colorblind
individuals. Keyword arguments for the matplotlib.pyplots.imshow can be passed to the MONET
function to give additional control over the spatial plot. Spatial plots are useful to give quick glimpses
196
Books
MDPI
Atmosphere 2017, 8, 210
of spatial patterns found within the model. Figure 4b shows an example of a spatial difference plot
which is useful for providing spatial assessments for absolute discrepancy. The same type of plot can
be created for different metrics such as for RMSE or correlation. In this particular example, it is easy to
pertain that over large cities such as Los Angeles or Las Vegas there is a significant under prediction
of ozone.
(a)
(b)
Figure 4.
Examples of spatial plots. (
a
) Example of a spatial contour map with a discrete colorbar and
observations overlaid; (
b
) Displayed is the normalized spatial O
3
bias (model–observation) at all of the
monitors within the domain. The marker size is dependent on the absolute value of the normalized bias.
Figure 5a presents a scatter plot, while Figure 5b shows a difference scatter plot. Observations
are displayed on the x-axis and model results on the y-axis. A one to one line is automatically created
along with a line of best fit. The scipy linregress function creates the line of best fit. Currently,
there is not a more complicated fitting algorithm available in MONET. The full power of scipy and
scikit-learn is available however if a more in-depth analysis is needed and hopefully will be included
in a future version of MONET. This type of plot is useful for determining the correlation of simulation
to observational results.
197
Books
MDPI
Atmosphere 2017, 8, 210
(a)
(b)
Figure 5.
Examples of scatter plots in MONET (
a
) Scatter plot example of model versus ozone; (
b
)
Scatter plot example of the difference in model and observation versus observations.
MONET uses the correlation, centered root mean square error, and the standard deviation
measures to graphically display on its Taylor diagrams [
14
]. Taylor diagrams are an efficient way to
graphically summarize model performance with observations. Currently, it is up to the user to ensure
that data is close to a Gaussian distribution.
The relative merits of various model simulations can be inferred from Figure 6. Simulations
that best compare to observations lie closest to the x-axis and have a low RMSE, while simulations
198
Books
MDPI
Atmosphere 2017, 8, 210
that lie above the x-axis have a greater RMSE and lower correlation. The concentric arcs around
the observation point show RMSE (in this graph at approximately 5.9 on the x-axis). Dispersion or
variation within the simulations are graphically displayed by the distance from the concentric arcs.
Simulations closest to the dotted arc have a standard deviation similar to observations.
Figure 6.
Taylor diagram example available in MONET. Taylor diagrams provide a concise visualization
of multiple statistics on a single graph. In this example, the closer the simulation to the x-axis the better
the correlation. The closer to the dotted line the better the model captured the variability of the model.
The RMSE values follow the contours emanating from the x-axis.
In this example, SIM5 performed the best as it has the highest RMSE, correlation, and a respectable
standard deviation as compared to observations. SIM9 has the lowest standard deviation and SIM2
and SIM8 have the lowest RMSE.
Different visualizations are being developed for future use. Specifically, cumulative distributions,
spectral density analysis, and principle component analysis will be added. The addition of different
observational data is also a high priority for use with MONET, including data from radiosondes,
aircraft, ceilometers, and satellite data. MONET is a simple, easy to use object oriented approach for
analyzing model results initially from NAQFC and more generally the CMAQ model.
4. Software and Data Availability
MONET is available on GitHub (https://github.com/noaa-oar-arl/MONET) as a free download.
The verification package is written in Python v2.7+. The authors recommend the Anaconda Python
package, available for free at https://www.continuum.io/downloads, to install supporting packages
described in the readme file included at GitHub. As mentioned previously, MONET uses the Pandas
python package to manipulate and store intermediate files. For one day of paired data, the file size
is approximately 100 mb for all of the variables and meta-variables in AirNow within the NAQFC
199
Books
MDPI
Atmosphere 2017, 8, 210
domain. MONET uses freely available observational data from the U.S. EPA, U.S. NOAA, U.S. NASA,
and U.S. NPS.
5. Conclusions
MONET is a comprehensive software package used to analyze observational data alone or pair
observations with gridded model data for chemical transport models. Currently, the observational
datasets included are focused within North America, except for AERONET, which is global. MONET
performs statistical calculations and creates visualizations to enable researchers to evaluate and
improve the scientific understanding of model outputs and observational data. MONET is built
entirely on open-source software and while the only model available for use is CMAQ used in
the NAQFC, the software is easily able to include different models. To include a new model into
MONET, a new object needs to be written that can: (1) read the output and (2) map of the output
variables to observation variables needed to pair data; (3) provide an output latitude and longitude
for interpolation. Work is ongoing to add the Comprehensive Air Quality Model with Extensions
(CAMx) [
15
] model output to the MONET software package through the use of PseudoNetCDF
(https://github.com/barronh/pseudonetcdf).
The MONET software package will continue to grow and improve through internal development
at NOAA ARL, and external development through collaborating partners and contributions made
through the community. The authors encourage collaborations with intergovernmental agencies,
private industry, and academia to improve and further develop MONET. Major software developments
are planned for the next release of MONET. One improvement is the inclusion of different models,
such as the Weather Research and Forecasting (WRF), CAMx, and the Next Generation Global
Prediction System (NGGPS). The next major development planned for MONET is to add different
observational sources such as radiosonde, profiler, and radar observations from the Meteorological
Assimilation Data Ingest System (MADIS; available online: http://madis.noaa.gov) and AWIPS II
(available online: http://www.unidata.ucar.edu/software/awips2/). Another improvement planned
is the use of pyresample to re-grid satellite swath data to model grids, as well as calculate comparable
variables (e.g., column aggregated aerosol optical depth values from model output).
Analysis enhancements to the MONET software will include implementing non-parametric
analysis techniques, such as cumulative distribution frequency plots, and Q-Q plots; adding temporal
decomposition capabilities such as Kolmogorov-Zurbenko filtering (Rao and Zurbenko [
16
]; Wise and
Comrie [
17
]) and spectral density plots; improving the overall speed through the use of parallel
processing in pyresample for spatial re-gridding, xarray for temporal averaging, and dask for larger
than memory computations. Many improvements to the MONET software are in progress and will be
included in the next release.
Supplementary Materials:
Initial application of the MONET software is available at
https://github.com/noaa-oar-arl/MONET
. Please look to the README.md file and the Wiki
(
https://github.com/noaa-oar-arl/MONET/wiki
) for further tutorials. The software can be install with
conda package manager: conda install –c bbakernoaa MONET. Included is a sample script along with a fortran
style namelist that can be used with the MONET package.
Acknowledgments:
The authors would like to thank the reviewers and editors for their dedication and valuable
input in improving this manuscript.
Author Contributions:
Barry Baker developed the infrastructure and implementation of MONET while Li Pan
assisted in testing and evaluation of the NAQFC simulations.
Conflicts of Interest: The authors declare no conflict of interest.
200
Books
MDPI
Atmosphere 2017, 8, 210
References
1.
George, B.J.; Schultz, B.D.; Palma, T.; Vette, A.F.; Whitaker, D.A.; Williams, R.W. An evaluation of
EPA’s National-Scale Air Toxics Assessment (NATA): Comparison with benzene measurements in Detroit,
Michigan. Atmos. Environ. 2011, 45, 3301–3308. [CrossRef]
2.
Fann, N.; Lamson, A.D.; Anenberg, S.C.; Wesson, K.; Risley, D.; Hubbell, B.J. Estimating the National
Public Health Burden Associated with Exposure to Ambient PM
2.5
and Ozone. Risk Anal.
2011
, 32, 81–95.
[CrossRef] [PubMed]
3.
Trasande, L.; Malecha, P.; Attina, T.M. Particulate Matter Exposure and Preterm Birth: Estimates of U.S.
Attributable Burden and Economic Costs. Environ. Health Perspect.
2016
, 124, 1913–1918. [CrossRef]
[PubMed]
4.
Ghude, S.D.; Jena, C.; Chate, D.M.; Beig, G.; Pfister, G.G.; Kumar, R.; Ramanathan, V. Reductions in India’s
crop yield due to ozone. Geophys. Res. Lett. 2014, 41, 5685–5691. [CrossRef]
5.
Tong, D.; Mathur, R.; Schere, K.; Kang, D.; Yu, S. The use of air quality forecasts to assess impacts of air
pollution on crops: Methodology and case study. Atmos. Environ. 2007, 41, 8772–8784. [CrossRef]
6.
Dennis, R.; Fox, T.; Fuentes, M.; Gilliland, A.; Hanna, S.; Hogrefe, C.; Irwin, J.; Rao, S.T.; Scheffe, R.;
Schere, K.; et al. A framework for evaluating regional-scale numerical photochemical modeling systems.
Environ. Fluid Mech. 2010, 10, 471–489. [CrossRef] [PubMed]
7.
Appel, K.W.; Gilliam, R.C.; Davis, N.; Zubrow, A.; Howard, S.C. Overview of the Atmospheric Model
Evaluation Tool (AMET) V1.1 for Evaluating Meteorological and Air Quality Models. Environ. Model. Softw.
2011, 26, 434–443. [CrossRef]
8.
Lee, P.; McQueen, J.; Stajner, I.; Huang, J.; Pan, L.; Tong, D.; Kim, H.; Tang, Y.; Kondragunta, S.;
Ruminski, M.; et al. NAQFC developmental forecast guidance for fine particulate matter (PM
2.5
).
Weather Forecast. 2017. [CrossRef]
9.
Stajner, I.; Lee, P.; McQueen, J.; Draxler, R.; Dickerson, P.; Upadhayay, S. Update on NOAA’s Operational Air
Quality Predictions. In Air Pollution Modeling and Its Application XXIV; Steyn, D.G., Chaumerliac, N., Eds.;
Springer: Cham, Switzerland, 2016; pp. 593–597.
10.
McKinney, W. Data Structures for Statistical Computing in Python. In Proceedings of the 9th Python in
Science Conference, Austin, TX, USA, 28 June–3 July 2010; pp. 51–56.
11.
McKinney, W. Python for data analysis: Data wrangling with Pandas, NumPy, and IPython; O’Reilly Media, Inc.:
Sebastopol, CA, USA, 2012.
12.
Van der Walt, S.; Colbert, S.C.; Varoquaux, G. The NumPy Array: A Structure for Efficient Numerical
Computation. Comput. Sci. Eng. 2011, 13, 22–30. [CrossRef]
13. Hunter, J.D. Matplotlib: A 2D graphics environment. Comput. Sci. Eng. 2007, 9, 90–95. [CrossRef]
14.
Taylor, K.E. Summarizing multiple aspects of model performance in a single diagram. J. Geophys. Res. Atmos.
2001, 106, 7183–7192. [CrossRef]
15.
ENVIRON. User’s Guide, Comprehensive Air Quality Model with Extensions (CAMx), Version 5.40; ENVIRON:
Arlington, TX, USA, 2011.
16.
Rao, S.T.; Zurbenko, I.G. Detecting and Tracking Changes in Ozone Air Quality. Air Waste
1994
, 44, 1089–1092.
[CrossRef] [PubMed]
17.
Wise, E.K.; Comrie, A.C. Extending the Kolmogorov–Zurbenko Filter: Application to Ozone, Particulate
Matter, and Meteorological Trends. J. Air Waste Manag. Assoc. 2005, 55, 1208–1216. [CrossRef] [PubMed]
©
2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
201
Books
MDPI
MDPI
St.Alban‐Anlage66
4052Basel,Switzerland
Tel.+41616837734
Fax+41613028918
http://www.mdpi.com
AtmosphereEditorialOffice
E‐mail:atmosphere@mdpi.com
http://www.mdpi.com/journal/atmosphere
Books
MDPI
MDPI
St. Alban-Anlage 66
4052 Basel
Switzerland
Tel: +41 61 683 77 34
Fax: +41 61 302 89 18
www.mdpi.com
ISBN 978-3-03842-840-4
Books
MDPI
