Subscriber access provided by UNIV OF CALIFORNIA SAN DIEGO LIBRARIES
Article
Daily estimation of ground-level PM2.5 concentrations over Beijing using 3 km resolution MODIS AOD Yuanyu Xie, Yuxuan Wang, Kai Zhang, Wenhao Dong, Baolei Lv, and Yuqi Bai Environ. Sci. Technol., Just Accepted Manuscript • DOI: 10.1021/acs.est.5b01413 • Publication Date (Web): 27 Aug 2015 Downloaded from http://pubs.acs.org on September 3, 2015
Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Environmental Science & Technology is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.
Page 1 of 35
1 2 3
4
5
Environmental Science & Technology
Daily estimation of ground-level PM2.5 concentrations over Beijing using 3 km resolution MODIS AOD Yuanyu Xie1, Yuxuan Wang*1,2,3, Kai Zhang4, Wenhao Dong1, Baolei Lv1, Yuqi Bai1 1
Ministry of Education Key Laboratory for Earth System Modeling, Center for Earth
6
7
System Science, Tsinghua University, Beijing, China. 2
Department of Marine Science, Texas A&M University at Galveston, Galveston,
8
9
TX, 77553, USA. 3
Department of Atmospheric Science, Texas A&M University, College Station,
10
11 12
TX, 77853, USA. 4
Department of Epidemiology, Human Genetics and Environmental Sciences, University of Texas School of Public Health, Houston, Texas, 77030, USA;
13
14
*Corresponding Author: yxw@tsinghua.edu.cn
1 ACS Paragon Plus Environment
Environmental Science & Technology
15
Page 2 of 35
TOC art
16
17
ABSTRACT
18
Estimating exposures to PM2.5 within urban areas requires surface PM2.5
19
concentrations at high temporal and spatial resolutions. We developed a mixed effects
20
model to derive daily estimations of surface PM2.5 levels in Beijing, using the 3 km
21
resolution satellite aerosol optical depth (AOD) calibrated daily by the newly
22
available high-density surface measurements. The mixed effects model accounts for
23
daily variations of AOD-PM2.5 relationships and shows good performance in model
24
predictions (R2 of 0.81~0.83) and cross validations (R2 of 0.75~0.79). Satellite derived
25
population-weighted mean PM2.5 for Beijing was 51.2 μg/m3 over the study period
26
(Mar 2013 to Apr 2014), 46% higher than China’s annual-mean PM2.5 standard of 35
27
μg/m3. We estimated that more than 19.2 million people (98% of Beijing’s population)
28
are exposed to harmful level of long-term PM2.5 pollution. During 25% of the days
29
with model data, the population-weighted mean PM2.5 exceeded China’s daily PM2.5
30
standard of 75 μg/m3. Predicted high-resolution daily PM2.5 maps are useful to identify
31
pollution “hot spots” and estimate short- and long-term exposure. We further 2 ACS Paragon Plus Environment
Page 3 of 35
Environmental Science & Technology
32
demonstrated that a good calibration of the satellite data requires a relatively large
33
number of ground-level PM2.5 monitoring sites and more are still needed in Beijing.
34
Introduction
35
Particulate matter (PM) air pollution in China is a major public health problem.
36
According to the Global Burden of Disease report, ambient PM air pollution is
37
responsible for over 1.2 million premature deaths annually in China, and 50% of its
38
1.3 billion population is currently exposed to ambient PM2.5 (particulate matter with
39
aerodynamic diameters of less than 2.5 μm) in exceedance of an annual average of 35
40
μg/m3 (China’s National Ambient Air Quality Standard, or NAAQS, available at
41
http://kjs.mep.gov.cn/).1 In recent years, extremely high PM2.5 concentrations have
42
been reported over many urban areas in China, notably in Beijing. 2-5 Some of the
43
worst pollution episodes saw PM2.5 concentrations exceeding 500 μg/m3 and lasting
44
multiple days.6-7 Accurate assessment of population exposure to such elevated levels
45
of PM2.5 over densely populated urban areas requires surface PM2.5 concentrations at
46
high temporal and spatial resolutions, but ground-level monitors have been sparse in
47
China which hinders our ability to accurately assess the health impact of PM2.5
48
pollution.
49
Satellite retrievals of AOD provide larger spatial coverage and have been widely
50
employed as a proxy to infer surface PM2.5 concentrations.8-13 Early studies obtained
51
the AOD-PM2.5 relationships using simple linear regression models or chemical 3 ACS Paragon Plus Environment
Environmental Science & Technology
Page 4 of 35
52
transport models (CTM) at a global or continental scale, including the United States
53
(US) and Europe.10-11,14-17 Those studies reported R2, a measure of goodness of fit of
54
linear regression, generally between 0.3 and 0.6 and the AOD-PM2.5 relationship was
55
typically assumed to be constant in time and location.18 Advanced statistical models
56
(e.g. generalized linear model, land use regression model, geographically weighted
57
regression, and generalized additive model) have been applied in order to account for
58
variability in the AOD-PM2.5 relationships associated with meteorology or land
59
use.13,19-22 These advanced models yielded higher prediction performance (R2 at
60
0.6~0.8) for large regions such as the northeastern and southeastern US and China,
61
but the increased complexity of these models requires additional data as inputs. Lee et
62
al23 first introduced an AOD daily calibration approach using a mixed effects model
63
to account for day-to-day and location-specific variability in the AOD-PM2.5
64
relationships over New England. They demonstrated that daily adjustment of the
65
AOD-PM2.5 relationships leads to a significant improvement in model performance
66
with R2 reaching 0.92, as compared to the linear regression (R2~0.51) applied to the
67
same region. The mixed effects model provides a valuable tool for estimating PM2.5
68
concentrations in urban areas where ground monitoring sites are not of sufficient
69
density.
70
Dense surface networks and high-resolution satellite AOD data are essential for
71
improving the accuracy of exposure assessment models. Since Mar 2013, 35 PM2.5
72
monitoring sites have been deployed in Beijing Municipality by Beijing 4 ACS Paragon Plus Environment
Page 5 of 35
Environmental Science & Technology
73
Environmental Protection Bureau, providing hourly PM2.5 concentrations at the urban
74
and suburban areas of Beijing. In early 2014, the Moderate Resolution Imaging
75
Spectroradiometer (MODIS) team released the new collection 6 (C006) AOD
76
products at a 3 km resolution.24 Prior standard AOD products from MODIS have a
77
spatial resolution of 10 km or coarser. Most of the previous studies applying the
78
MODIS AOD to derive surface PM2.5 concentrations over China focused on the
79
regional scale with resolutions ranging from 10 km to 100 km. 12,13,22 Utilizing the
80
newly available surface monitoring data in Beijing, this study is the first attempt to
81
develop a mixed effects model to estimate ground-level PM2.5 concentrations using
82
the 3 km resolution AOD data on a daily basis for a heavily polluted urban
83
environment in China. The severe pollution events offer a valuable opportunity to
84
assess the performance of the daily AOD calibration approach through this mixed
85
effects model. The paper is organized as follows: surface and satellite data as well as
86
model development are introduced in the “Materials and Methods” section; the model
87
performance, PM2.5 prediction results and uncertainties are provided in the “Results
88
and Discussion” section.
89 90
Materials and methods
91
Ground-level PM2.5 data. Hourly PM2.5 concentrations observed at 35 monitoring
92
stations in Beijing (Figure 1b) were obtained from Beijing Municipal Environmental
93
Monitoring Center (http://zx.bjmemc.com.cn/). Surface PM2.5 mass concentrations are 5 ACS Paragon Plus Environment
Environmental Science & Technology
Page 6 of 35
94
measured by the Tapered Element Oscillating Microbalance (TEOM) method, and the
95
measurements have undergone the calibration processes and quality controls
96
according to the environmental protection standard of China (HJ 618-2011;
97
MEPCN).25 Beijing Municipality consists of 16 districts with six urban districts
98
located in the center of the city and ten suburban districts surrounding the urban
99
regions (Figure 1c). Among the 35 PM2.5 monitors, 17 are located in the urban
100
districts and 18 in suburban districts. We averaged the surface concentrations
101
measured between 13:00 pm and 14:00 pm local time at each site to derive the daily
102
PM2.5 concentrations that match with the over-passing time of the Aqua satellite. The
103
study period is from March 26th 2013 to April 23th 2014, spanning a total of 394 days.
104
MODIS 3 km AOD products and calibration. MODIS 3 km AOD product.
105
MODIS on board the NASA Aqua satellite has been in operation since 2002,
106
providing retrieval products of aerosol and cloud properties with nearly daily global
107
coverage.26 The dark target algorithms make use of surface reflectance at three
108
wavelength channels (0.47 μm, 0.66 μm, and 2.12 μm) for AOD retrievals over land.
109
Standard MODIS Level 2 (L2) AOD products are distributed at a 10 km resolution.
110
To fulfill the need for higher resolution pollution detection, the most recently released
111
MODIS Collection 6 product (MYD04_3K) provides aerosol products at a 3 km
112
resolution in addition to the L2 10 km product. The retrieval algorithm of the higher
113
resolution product is similar to that of the 10 km standard product, but averages 6 × 6
114
pixels in a single retrieval box rather than the 20 × 20 pixels after cloud screening and 6 ACS Paragon Plus Environment
Page 7 of 35
Environmental Science & Technology
115
other surface mask processes. Pixels outside the reflectivity range of the brightest 50%
116
and darkest 20% at 0.66 μm are discarded to reduce uncertainty. Validation against
117
surface sun photometer shows that two thirds of the 3 km retrievals fall within the
118
expected error on a regional comparison but with a high bias of ~0.06 especially over
119
urban surface.27
120
MODIS AOD validation. The ground-based AOD measurements from three AErosol
121
RObotic NETwork (AERONET) sites (http://aeronet.gsfc.nasa.gov/) in Beijing were
122
used to validate the satellite-derived AOD. To be comparable with the spectrum
123
setting of MODIS, AERONET AOD at 550 nm were calculated by interpolating AOD
124
at 440 nm and at 675 nm using the reported angstrom exponent at the respective
125
wavelengths. The 3 km MODIS AOD products from Aqua show high temporal
126
correlations with the AERONET AOD (Pearson correlation coefficient r of 0.93, 0.93
127
and 0.94 at the three sites respectively, Text S1, Supporting Information (SI)). The
128
mean AOD difference between MODIS and AERONET at the three sites is 0.29. The
129
higher AOD from MODIS could be partly attributed to the reflectance bias over the
130
brighter surface of urban areas, which was also observed over other urban regions.
131
27
132
systematic bias when deriving the PM2.5 to AOD ratio and is not expected to influence
133
surface PM2.5 estimation in this study because we calibrated MODIS AOD with
134
ground-level PM2.5 measurements on a daily basis.
24,
Since this bias is persistent for the full range of AOD values, it can be treated as a
7 ACS Paragon Plus Environment
Environmental Science & Technology
Page 8 of 35
135
Model development and validation. We selected the site-collocated satellite AOD
136
values for each surface site where it falls within a 3 km grid. If there are more than
137
one site within a single 3 km grid, the PM2.5 values of those sites are averaged. With
138
this process, site #12 and #14 are averaged and there remain 34 pairs of AOD and
139
PM2.5 data for model development.
140 141
A linear regression model was first applied to the collocated AOD and PM2.5 datasets. The linear regression model for all the 34 sites follows the form of:
142
PM 2.5 AOD
143
Where and are the fixed intercept and slope, respectively. The intercept and
144
slope do not have spatial and temporal variations since the linear regression model
145
assumes that the AOD-PM2.5 relationship is constant for all the sites throughout the
146
study period. The log transformed AOD and PM2.5 were also tested, and no significant
147
improvement was found in the regression performance. Therefore, we only present
148
the regression results with the original AOD and PM2.5 datasets.
(1)
149
In reality, the AOD-PM2.5 relationship may exhibit spatial and temporal variations
150
due to changing meteorology and other factors. To represent the varying AOD-PM2.5
151
relationship, we developed a mixed effects model following the approach proposed by
152
Lee et al.23 which takes into consideration daily variations of the AOD-PM2.5
153
relationship and derives site-specific parameters for spatial adjustment. The mixed
154
effects model estimates surface PM2.5 concentrations for each site i of day j
155
( PM 2.5,ij ) from collocated MODIS AOD ( AODij ) in the form of 8 ACS Paragon Plus Environment
Page 9 of 35
Environmental Science & Technology
156
PM 2.5,ij ( u j ) ( v j ) AODij si ij
157
(u j v j ) ~ N [(00), ]
158
Where and are the fixed intercept and slope independent of time and
159
location, u j and v j are the random intercept and slope for all the sites at each day.
160
When the subscript i is not present for a parameter in Equation (2), it indicates that
161
parameter does not change by site. The random terms ( u j and v j ) reflect the
162
day-to-day variations of the AOD-PM2.5 relationship influenced by meteorology,
163
satellite retrieval conditions, etc. The si ~N (0, s2 ) is a site term which accounts for
164
the spatial difference of the AOD-PM2.5 relationship due to differences in site specific
165
characteristics (i.e. surface reflectivity, topography, PM2.5 emissions, and pollution
166
transported to the observation sites). The comparison between the mixed effects
167
models with and without the site effect si provides a measure of the sensitivity in
168
model-derived day-to-day variations of the AOD-PM2.5 relationship to site locations.
169
ij represents the error term, and
170
day-specific random effects. We excluded days with less than two pairs of
171
AOD-PM2.5 data and performed model predictions on the remaining available days.
172
Model performances were evaluated by comparing the predictions to ground
173
measurements using R2, mean prediction error (MPE), and root mean square error
174
(RMSE). The correlation coefficient, r, reported in this study is the Pearson
175
correlation coefficient unless stated otherwise.
(2)
is the variance-covariance matrix for the
9 ACS Paragon Plus Environment
Environmental Science & Technology
Page 10 of 35
176
A cross-validation (CV) method was implemented to test the performance of the
177
linear regression model and mixed effects model. We isolated one site at a time,
178
performed model fitting with the remaining 33 sites and validated model
179
performances on the isolated site. The process was repeated for each of the 34 sites.
180
The CV statistics are represented with R2, MPE and RMSE. We also varied the
181
number of isolated sites during cross validation to 5, 10, and 20 as a way of testing the
182
sensitivity of the mixed effects model to the density of surface monitors needed.
183 184
185
Descriptive statistics. Table 1 presents the descriptive statistics of measured PM2.5
186
concentrations at the 35 surface sites and site-collocated MODIS AOD during the
187
study period. Daily time series of the site-average PM2.5 concentrations and
188
site-collocated MODIS AOD (Figure 1a) indicate an overall good correlation between
189
them (r = 0.6). Average PM2.5 from the ground-based monitors is 81.04 μg/m3 during
190
the study period, more than a factor of 2 higher than China’s NAAQS of 35 μg/m3 for
191
annual-mean PM2.5, and the standard deviation (SD) is 73.23 μg/m3. The average
192
AOD is 0.68 with a SD of 0.49. The maximum site-average daily mean PM2.5 is
193
385.80 μg/m3 and the maximum single-site daily concentration reaches 500 μg/m3.
194
Many sites have high AOD values larger than 2.0. The data are further divided into
195
the warm season (Apr 15th to Oct 14th) and the cold season (Oct 15th to Apr 14th). The
196
average PM2.5 concentrations does not differ much between the two seasons but the
Results and discussion
10 ACS Paragon Plus Environment
Page 11 of 35
Environmental Science & Technology
197
AOD values is much lower in the cold season than in the warm season. Besides the
198
impact of lower planetary boundary layer (PBL) depths in the cold season as well as
199
other meteorological conditions (e.g. lower relative humidity), a lower percent of
200
available satellite AOD retrievals over snow also explains the lower mean and SD of
201
AOD in the cold season. The detailed statistics for each site and description for
202
seasonal comparisons are shown in Text S2 (SI).
203
Spatially surface PM2.5 from the ground-level monitors presents a decreasing
204
gradient from south to north due to topography and land use difference in Beijing
205
(Figure 1b). The south to north PM2.5 gradient generally demonstrates the emission
206
difference between suburban and urban regions. In addition, the southern sites are
207
more influenced by pollution transported from the cities south of Beijing, while the
208
north of Beijing is surrounded by mountains. Satellite AOD averaged for the entire
209
study period exhibits a similarly strong spatial gradient (Figure 1c), with highest
210
values over the southeast urban districts. The correlation between the period-mean
211
PM2.5 and site-collocated AOD is 0.64 across the 35 sites, indicating a relatively tight
212
spatial consistency between the two datasets.
213
Linear regression model. The linear regression model gives a regression slope of
214
58.67 and intercept of 10.08 between measured PM2.5 and site-collocated MODIS
215
AOD, with an overall R2 of 0.47 (p