Long-term trends and spatial patterns of PM2.5 ...

Science of the Total Environment 631–632 (2018) 1504–1514

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

Long-term trends and spatial patterns of PM2.5-induced premature mortality in South and Southeast Asia from 1999 to 2014 Yusheng Shi a,b,c,⁎, Tsuneo Matsunaga b,c, Yasushi Yamaguchi d, Aimei Zhao a, Zhengqiang Li a, Xingfa Gu a a

State Environmental Protection Key Laboratory of Satellite Remote Sensing, Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100101, China Center for Global Environmental Research, National Institute for Environmental Studies, Tsukuba 305-8506, Japan Satellite Observation Center, National Institute for Environmental Studies, Tsukuba 305-8506, Japan d Graduate School of Environmental Studies, Nagoya University, Nagoya 464-8601, Japan b c

H I G H L I G H T S

G R A P H I C A L

A B S T R A C T

• Multi-year premature death due to PM2.5 in South-Southeast Asia reached 1,447,000. • Premature deaths attributed to PM2.5 have grown 38% from 1999 to 2014. • Stroke and ischemic heart disease were the two principal PM2.5-related diseases. • India and Bangladesh were the major contributors of deaths in SouthSoutheast Asia. • South Asia estimated more premature deaths than Southeast Asia during 1999–2014.

a r t i c l e

i n f o

Article history: Received 14 January 2018 Received in revised form 1 March 2018 Accepted 14 March 2018 Available online xxxx Editor: Jianmin Chen Keywords: PM2.5 pollution Premature mortality Spatial variations Long-term trends South and Southeast Asia

a b s t r a c t Fine particulate matter (PM2.5) poses a potential threat to human health, including premature mortality under long-term exposure. Based on a long-term series of high-resolution (0.01° × 0.01°) satellite-retrieved PM2.5 concentrations, this study estimated the premature mortality attributable to PM2.5 in South and Southeast Asia (SSEA) from 1999 to 2014. Then, the long-term trends and spatial characteristics of PM2.5-induced premature deaths (1999–2014) were analyzed using trend analyses and standard deviation ellipses. Results showed the estimated number of PM2.5-induced average annual premature deaths in SSEA was 1,447,000. The numbers increased from 1,179,400 in 1999 to 1,724,900 in 2014, with a growth rate of 38% and net increase of 545,500. Stroke and ischemic heart disease were the two principal contributors, accounting for 39% and 35% of the total, respectively. High values were concentrated in North India, Bangladesh, East Pakistan, and some metropolitan areas of Southeast Asia. An estimated 991,600 deaths in India was quantified (i.e., ~69% of the total premature deaths in SSEA). The long-term trends (1999–2014) of PM2.5-related premature mortality exhibited consistent incremental tendencies in all countries except Sri Lanka. The findings of this study suggest that strict controls of PM2.5 concentrations in SSEA are urgently required. © 2018 Elsevier B.V. All rights reserved.

1. Introduction

⁎ Corresponding author. E-mail address: [email protected] (Y. Shi).

https://doi.org/10.1016/j.scitotenv.2018.03.146 0048-9697/© 2018 Elsevier B.V. All rights reserved.

Long-term exposure to high concentrations of fine particulate matter (aerodynamic diameter: ≤2.5 μm; PM2.5) poses considerable potential threat, including premature mortality (Lelieveld et al., 2013;

Y. Shi et al. / Science of the Total Environment 631–632 (2018) 1504–1514

Burnett et al., 2014; Apte et al., 2015; Zhang et al., 2017). In the Global Burden of Diseases (GBD) study in 2015 (Cohen et al., 2017), exposure to PM2.5 was cited as causing 4.2 million deaths worldwide in 2015, representing 7.6% of total death globally. Approximately 59% of these deaths occurred in South Asia (1.36 million) and East Asia (1.14 million). Moreover, the number of global deaths attributable to ambient PM2.5 is believed to have increased from 3.5 million in 1990 to 4.2 million in 2015 (Cohen et al., 2017). Over recent decades, South and Southeast Asia (SSEA) (e.g., India, Bhutan, Nepal, Bangladesh, Myanmar, Laos and Thailand) has witnessed increasing concentrations of PM2.5 due to combustion emissions from multiple sources, e.g., coal-fired power plants (Koplitz et al., 2017), household use of solid fuel (Bonjour et al., 2013), agricultural and other open burning (Shi and Yamaguchi, 2014; Shi et al., 2015) and industrial- and transportation-related sources (Zhang et al., 2017). According to the “State of Global Air/2017” report by the Health Effects Institute and the Institute for Health Metrics and Evaluation (Health Effects Institute, 2017), the population-weighted annual average PM2.5 concentrations in 2015 were rated extremely high in Bangladesh (89 μg/m3), Nepal (75 μg/m3) and India (74 μg/m3). All of these countries had higher values than China (58 μg/m3), with substantial variation in PM2.5 concentrations among provinces (19–79 μg/m3). Besides, Bangladesh and India have experienced the steepest increases of PM2.5 concentrations since 2010 among the 10 most populous countries (Health Effects Institute, 2017). Consequently, India and Bangladesh have experienced some of the greatest increases (i.e., on the order 50–60%) in PM2.5-attributed premature mortality. Therefore, the SSEA region has contributed considerably to both the total and increase in global premature mortality. A number of methods and algorithms have been developed to estimate ground-level PM2.5 concentrations based on satellite-derived aerosol optical depth (AOD) (Li et al., 2016; Guo et al., 2017; Lv et al., 2017; He and Huang, 2018). These approaches have provided promising alternatives for estimating ambient PM2.5 exposure (van Donkelaar et al., 2010; Ma et al., 2014; Zhang and Li, 2015) and the associated health impacts with improved spatiotemporal coverage (Burnett et al., 2014; Apte et al., 2015; Lelieveld et al., 2015; Zhang et al., 2017). For example, the GBD study reported on PM2.5-induced premature mortality, based on exposure methodology, by taking advantage of a global AOD-derived PM2.5 dataset (Brauer et al., 2016). Several studies have tried to evaluate the health impacts caused by PM2.5 pollution in recent years. One study estimated that 3.2 million PM2.5-related premature mortalities occurred worldwide in 2010 (Lelieveld et al., 2015). A more recent study calculated the number of premature deaths due to PM2.5 pollution globally in 2007 as 3.45 million (Zhang et al., 2017). However, both of these PM2.5-based health assessments reported results for only one year and failed to consider long-term spatiotemporal variations and trends. To date, most multiyear cohort studies on PM2.5-related health assessments have been undertaken in China (Xie et al., 2016; Hu et al., 2017; Liu et al., 2017). However, only a few studies have investigated the long-term trends and spatial variations of PM2.5-induced premature mortality in SSEA, where many countries have witnessed the greatest increases in PM2.5 concentrations in association with surging and dense populations (Shi et al., 2018). Primarily, this has been because of the unavailability of the long-term series of high-resolution PM2.5 data. In addition, the application of a fixed uniform baseline mortality for each disease across continental regions might overlook differences among countries over time (Chowdhury and Dey, 2016). As an improvement on earlier research, this study estimated the annual premature mortality attributable to PM2.5 across SSEA from 1999 to 2014 based on a newly developed high-resolution (0.01° × 0.01° grid) long-term satellite-retrieved ground-level PM2.5 dataset, highresolution population density data, and national-level time-varying baseline mortality data. The long-term (1999–2014) trends and spatial variations of PM2.5-related premature deaths from five diseases in SSEA and each individual country were analyzed. Knowledge of these long-

1505

term trends and spatial variations of premature mortality is essential for understanding the effects of PM2.5 exposure on the health of the populations of different countries. 2. Methods and data source 2.1. Methods 2.1.1. Premature mortality assessment Five PM2.5-related diseases were chosen for this study: chronic obstructive pulmonary disease (COPD), ischemic heart disease (IHD), stroke, lung cancer (LC) for adults, and acute lower respiratory illness (ALRI) for infants, all of which have direct causal links to ambient PM2.5 exposure (Brauer et al., 2016; Cohen et al., 2017). Premature mortality (ΔMi,j) due to a particular disease j for country i attributable to PM2.5 exposure was estimated using the traditional epidemiological relationship (Anenberg et al., 2012; Silva et al., 2013; Chowdhury and Dey, 2016; Ghude et al., 2016): N

N

N

∑i; j¼1 ΔM i; j ¼ ∑i; j;k¼1 Y i; j;k

∑i¼1 RRi; j −1 N ∑i¼1 RRi; j

N

∑i¼1 P i ;

ð1Þ

where Yi,j,k is the baseline mortality for disease j in grid pixel i within a country k, RRi,j represents the relative risk of disease j in grid cell i for a specific PM2.5 concentration, Pi represents the exposed adult population (i.e., above 25 years of age, consistent with the GBD study) and infant population (i.e., below 5 years of age) in each grid. The population structures and ages of the studied countries (1999–2014) were obtained from the Population Pyramids of the World (http:// populationpyramid.net). Then, the annual exposed adult and infant population (Pi) of each grid cell was quantified as a proportion of the total population. The baseline mortality for each disease varied temporally and among countries. We obtained the baseline mortality attributable to each of the five diseases for each country (1999–2014) from the online GBD database (https://vizhub.healthdata.org/gbd-compare/) compiled in Supporting Information (SI) Table S1. In addition, the extrapolation of RRi,j to the observed PM2.5,i concentration for each disease was estimated based on existing epidemiological studies of the integrated exposure–response (IER) functions for ambient PM2.5 exposure (Burnett et al., 2014; Apte et al., 2015; Cohen et al., 2017): ( RRi; j ¼

h i δ 1 þ α j 1− exp −γ j ðX−X 0 Þi j ; 1;

if X NX 0 else

) ;

ð2Þ

where X denotes the annual mean PM2.5 concentration and X0 is the threshold concentration below which no additional risk is assumed. For each disease category, X0 represents the theoretical-minimum-risk concentration (range: 2.4–5.9 μg/m3) (Cohen et al., 2017). In addition, we adopted the bounds representing the 95% confidence interval (CI), derived by Burnett et al. (2014) from 1000 sets of coefficients and exposure–response functions based on Monte Carlo simulations (Liu et al., 2016; Maji et al., 2018). The parameters αj, γj, and δj are specific to each disease and were determined using a stochastic fitting process (Apte et al., 2015). We used the mean values of RR for each disease (age-specific values of RR were used for stroke and IHD) provided as a look-up table by Apte et al. (2015) in SI Table S2 to estimate the number of premature mortalities. 2.1.2. Trend analysis Trend analysis is used commonly in temporal dynamics analyses to explore interannual variation characteristics. In this study, a long sequence of premature mortality change trends was analyzed quantitatively based on the trend analysis method. Tendencies in variations of

1506


premature mortality can be determined as follows:

3. Results and discussion

n n n n ∑i¼1 ði PMi Þ− ∑i¼1 i ∑i¼1 PMi ; Trend ¼ 2 n n n ∑i¼1 i2 − ∑i¼1 i

3.1. Spatial patterns of the estimated premature mortality ð3Þ

where PM is the grid unit premature mortality, n is the time span, and i is the time unit.

2.1.3. Rate of change The rate of change (ROC) was used to compare the variance in premature mortality of each disease between 1999 and 2014 throughout SSEA. The ROC was defined as follows (Shi et al., 2012):

ROC ¼

B−A 100%; A

ð4Þ

where A and B represent the premature mortality in 1999 and 2014, respectively.

2.1.4. Standard deviation ellipse analysis The standard deviation ellipse (SDE), or directional distribution, represents elements in the main distribution area. The calculated major and minor axes of the ellipse indicate the direction and range of the data distribution (Wang et al., 2015; Wachowicz and Liu, 2016). Thus, the calculated multiyear SDE can reflect clear trends in the directionality of the elements over time. The SDE calculation is described separately in SI section S1. The median center identifies the location that minimizes the overall Euclidean distance to the features in a dataset; therefore, it can express the center of the entire dataset (Peng et al., 2016). Herein, by comparing the interannual changes in the SDE and median center of the time series, it was possible to determine the overall characteristics of the geospatial distributions and their spatial dynamic process over time (Shi et al., 2018). In this study, the spatial characteristics of premature mortality and the corresponding PM2.5 concentrations and populations in 1999 and 2014 were quantified to trace their changes over the 16-year study period.

2.2. Data The analysis of premature mortality in SSEA attributable to PM2.5 exposure during 1999–2014 was based on two data sources: annual PM2.5 concentrations and population density. For the former, we applied the 0.01° × 0.01° high-spatial-resolution ground-level PM2.5 concentrations derived from satellite-retrieved datasets that were inverted by van Donkelaar et al. (2016). These global long-term series of ground-level PM2.5 concentrations were estimated using the GEOS-Chem chemical transport model based on AOD retrievals from different datasets. The validation against the measured ground-level point data across SSEA showed that the satellite-derived annual PM2.5 concentrations presented consistent agreement with site observation data (van Donkelaar et al., 2016; Shi et al., 2018). The latter data source included Gridded Population of the World version 4 (GPWv4) maps of national demographic data for 2000, 2005, 2010, and 2015 (resolution: 30 arcseconds worldwide), which were developed by the Center for International Earth Science Information Network at Columbia University (USA). The populations adopted for other years were extrapolated linearly using the data of the abovementioned four years as a base. We subsequently resampled them to match the resolution of the PM2.5 concentration dataset (0.01° × 0.01°) from 1999 to 2014.

During 1999–2014, the estimated total average annual premature mortality due to PM2.5 exposure in SSEA reached 1,447,000 (95% CI: 935,300–2,541,100) (Table 1, SI Table S3). The five diseases contributed differently to the total annual average. The estimates of annual premature mortality from COPD, IHD, stroke, LC, and ALRI attributable to ambient PM2.5 numbered 187,700 (95% CI: 101,400–271,900), 509,400 (95% CI: 374,600–668,000), 556,400 (95% CI: 442,800–673,200), 30,700 (95% CI: 12,500–41,100), and 162,900 (95% CI: 87,300–259,600), respectively. In terms of their relative contributions, COPD, IHD, stroke, LC, and ALRI accounted for 13%, 35%, 39%, 2%, and 11%, respectively, of total premature deaths in SSEA. The two largest contributors (stroke and IHD) accounted for 74% of total premature deaths in SSEA, whereas the contribution from LC was minimal. Fig. 1 presents the spatial distribution of estimated annual PM2.5related premature mortality linked to COPD, IHD, stroke, LC, and ALRI in SSEA during 1999–2014. In general, premature deaths showed obvious spatial patterns and strong variations across the entire study area from 1999 to 2014 (Fig. 2). High levels of premature mortality were widespread in North India (Figs. 1 and 2). This area showed the highest numbers of estimated premature deaths because of high PM2.5 concentrations and high population densities (SI Fig. S1). Other isolated regions with high values of premature mortality were evident across India, corresponding to the dense population centers of urban areas throughout the study period. Bangladesh showed high levels of premature mortality throughout the entire country because of high PM2.5 concentrations and population density. The highest levels of premature mortality in Pakistan were found mostly in the east. Figs. 1 and 2 show strong variations and a clear decreasing tendency of premature mortality from South Asia (SA) to Southeast Asia (SEA) during 1999–2014 and in each specific year. In SEA, estimates of premature deaths in most countries were low, except in Naypyidaw (Myanmar), Bangkok (Thailand), and Hanoi and Ho Chi Minh City (Vietnam). The high population densities and high PM2.5 concentrations in these urban areas were the dominant reasons for the comparatively higher estimates in relation to neighboring rural areas (SI Fig. S1). Specifically, premature deaths from stroke and IHD (Fig. 3), the two greatest contributors, throughout SSEA during 1999–2014 presented spatial patterns and variations similar to total deaths (Figs. 1 and 2), i.e., highest values in North India, Bangladesh, and East Pakistan. Premature deaths attributable to COPD, LC, and ALRI were low throughout most of SSEA, except for some high values estimated in large cities (Fig. 3). Overall, different countries contributed differently to the total number of premature deaths in SSEA because of variations in population density and PM2.5 concentration (Fig. 4, Table 1, SI Table S3). The three countries with the highest numbers of premature deaths were India (991,600 (95% CI: 378,100–1,872,800)), Bangladesh (163,000 (95% CI: 99,500–281,800)), and Pakistan (134,600 (95% CI: 76,400–269,200)), which accounted for 69%, 11%, and 9% of the total values in SSEA during 1999–2014. Together, these three countries accounted for up to 89% of premature mortality in SSEA, indicating that air pollution in SA generally poses a greater risk to human life than that in SEA, where relatively smaller populations and moderate PM2.5 concentrations are found. Vietnam, Myanmar, Thailand, and Nepal accounted for 48,900 (95% CI: 23,700–73,800), 45,700 (95% CI: 22,100–72,100), 29,500 (95% CI: 10,100–41,600), and 16,200 (95% CI: 7400–28,300) premature deaths, respectively, followed by Cambodia, Laos, Sri Lanka, and Bhutan with 9000 (95% CI: 3400–20,000), 5500 (95% CI: 3100–12,500), 2800 (95% CI: 1700–4900), and 200 (95% CI: 100–400), respectively. It is not surprising that India accounted for more premature mortalities than all other SSEA countries combined, which was because it has had persistently high and increasing PM2.5 concentrations over most of its land area during the past 16 years (Shi


1507

Table 1 Average annual premature mortalities (in thousands) attributable to each of the five diseases in each country (1999–2014), and their corresponding changes (%) between 1999 and 2014. IHD

India Bangladesh Pakistan Vietnam Myanmar Thailand Nepal Cambodia Laos Sri Lanka Bhutan Total

COPD

Stroke

LC

ALRI

Total

Mean

Change (%)

Mean

Change (%)

Mean

Change (%)

Mean

Change (%)

Mean

Change (%)

Mean

Change (%)

335.9 58.6 48.1 22.7 19.5 12.3 4.5 4.1 2.0 1.6 0.1 509.4

40 25 47 32 37 28 42 55 49 −19 83 38

162.4 11.2 7.6 1.0 1.7 1.5 1.7 0.2 0.1 0.1 0.0 187.7

47 41 50 29 50 34 54 66 62 −49 95 47

371.1 66.6 54.4 21.2 18.6 12.7 5.0 3.5 2.0 1.0 0.1 556.4

42 28 47 30 47 31 45 66 56 −49 92 40

15.0 3.7 4.2 3.1 1.6 2.7 0.2 0.2 0.1 0.1 0.0 30.7

48 42 50 28 51 34 55 67 63 −54 96 44

107.1 23.0 20.3 0.9 4.3 0.3 4.8 1.0 1.2 0.0 0.0 162.9

13 26 27 −12 21 −9 3 42 36 −97 31 17

991.6 163.0 134.6 48.9 45.7 29.5 16.2 9.0 5.5 2.8 0.2 1447.0

39 28 44 30 40 30 33 58 49 −32 83 38

et al., 2018) together with a dense and surging population, especially in North India. The largest annual PM2.5 concentration among all SSEA countries was recorded in Bangladesh (van Donkelaar et al., 2016), which also has a highly dense population, resulting in large numbers of premature deaths annually. Although Bhutan experienced rapid growth in the number of premature deaths from 1999 to 2014, its absolute number of premature deaths due to PM2.5 concentrations was the smallest proportion among the SSEA countries. Among the five diseases studied, stroke and IHD were the two primary contributors to total premature deaths, not only in the entire SSEA region but also in each country (Fig. 4, Table 1, SI Table S3). For example, annually, stroke and IHD in India caused 371,100 (95% CI: 283,400–496,700) and 335,900 (95% CI: 214,300–450,100) premature deaths (1999–2014), respectively, which accounted for 37% and 34% of the total number, respectively. Although the contributions of the five diseases in each country ranked similarly, their corresponding proportions varied greatly. This could be attributed to the changing baseline mortality among the various countries during 1999–2014. 3.2. Long-term trends of premature mortality Changes in PM2.5 concentration, population, and the baseline mortality for each disease in the different countries of the SSEA mean the

annual premature mortality fluctuated annually. Therefore, the longterm changes in annual premature deaths (Fig. 5) and their percent contributions (SI Fig. S2) were quantified over the entire of SSEA (1999–2014). Overall, PM2.5-induced premature mortality in SSEA showed an increasing trend from 1999 to 2014 (Fig. 5). Our calculations suggested that premature deaths from COPD, IHD, stroke, LC, and ALRI attributable to ambient PM2.5 concentrations numbered 1,179,400 in 1999 and 1,724,900 in 2014, equivalent to a growth rate of 38% and net increase of 545,500. However, the total number of PM2.5-induced premature deaths in 2005 was lower than 2004, this was attributable to the relative lower PM2.5 concentrations of SSEA in 2005 (34.8 μg/ m3) compared with that in 2004 (40.0 μg/m3). Premature deaths from all five studied diseases increased gradually from 1999 to 2014; however, the largest increments occurred in stroke and IHD, followed by COPD, ALRI, and LC. Because of the variations in the trends of the five diseases and their absolute amounts, their relative contributions to total premature mortality also varied during the 16-year period. The changes in the proportions of the contributions of the diseases to total premature deaths showed that IHD and LC varied little annually (SI Fig. S2). Both COPD and stroke increased their contributions from 1999 to 2014 with minor increases in their proportions, corresponding to a decrease in the percentage contribution of ALRI to total premature mortalities. With increasing economic development in the SSEA

Fig. 1. Spatial distribution of the total average annual premature mortality (COPD, IHD, stroke, LC, and ALRI) in South and Southeast Asia from 1999 to 2014.

1508


Fig. 2. Annual total premature mortality (COPD, IHD, stroke, LC, and ALRI) in South and Southeast Asia from 1999 to 2014.

countries, affordable healthcare has helped prevent premature mortality from ALRI, resulting in a decline in the baseline mortality rate for infants (Chowdhury and Dey, 2016). Therefore, the increasing number of premature deaths attributed to ALRI was relatively small compared with the other diseases, despite the surging population and increasing PM2.5 concentrations. Interannual variations in premature deaths from the five diseases in each country also presented consistent increasing tendencies from 1999 to 2014, except in Sri Lanka (SI Fig. S3). Premature deaths from ALRI in India, Bangladesh, and Pakistan showed strong interannual variations with a flat trend of change. Overall, most SSEA countries have areas that have experienced increasing PM2.5 concentrations and populations (Shi et al., 2018). India accounted for 69% of total premature deaths; therefore, it showed increasing trends consistent with the entire SSEA region. Conversely, premature mortality in Sri Lanka has decreased because of its decline in PM2.5 concentration and its largely unchanged population. Other reason can be attributable to the consistent decline of baseline mortality (COPD, IHD, stroke, ALRI) during 1999–2014 (SI Table S1). In fact, most diseases in most countries of SSEA experienced decreased baseline mortality from 1999 to 2014, which affected the increasing trends of the estimated premature death. The spatial patterns of the interannual trends of premature deaths from 1999 to 2014 were quantified using linear regression, and significant positive or negative trends were revealed at the 95% confidence level after testing (p b 0.05) (Fig. 6). Overall, premature deaths showed a significant

incremental tendency across most of the SSEA region. However, the trend of change showed strong spatial variations among countries. The areas with the most significant and remarkable increases in premature deaths were found in North India, where the annual growth rate was N0.02 deaths/km2/year. India had a larger spatial extent with a higher increase in premature deaths than any other country in SSEA. A significant increase in premature deaths was found in East Pakistan and Bangladesh, where a rate of increase N0.02 deaths/km2/year was recorded. The spatial distribution of the long-term trends in premature deaths during 1999–2014 revealed that SA countries experienced increases that were more remarkable than SEA countries (Fig. 6), similar to the spatial characteristics of temporal trends in PM2.5 concentrations (Shi et al., 2018). In SEA countries, premature deaths in Myanmar increased to a limited extent around the largest city of Yangon at Yangon Region of Southern Myanmar at an annual rate of 0.005–0.02 deaths/km2/year. During 1999–2014, Myanmar experienced a significant increasing tendency of PM2.5 concentration over most areas (Shi et al., 2018). However, the PM2.5 concentrations remained consistently at a moderate level (b25 μg/m3); therefore, the related premature deaths remained very low, which resulted in indistinct trends of change in most areas. Thailand witnessed nonsignificant trends in premature deaths in most areas during 1999–2014. However, an exception was found in Bangkok, where premature deaths increased at an annual rate of 0.02–0.1 deaths/km2/ year. Phnom Penh in Cambodia and Ho Chi Minh City and Hanoi in Vietnam also experienced increasing tendencies of premature deaths.


1509

Fig. 3. Averaged annual premature death from COPD, IHD, stroke, LC and ALRI in South-Southeast Asia from 1999 to 2014.

In SSEA, we found significant negative trends in parts of Nepal, coastal southern India, Naypyidaw (Myanmar), and North Vietnam (Phu Tho Province) with rates of decrease ranging from −0.01 to −1. deaths/

km2/year. The reason for the negative trends in these areas was the steady decline in population from 1999 to 2014 based on the gridded demographic data from GPWv4.

Fig. 4. Multi-year averaged premature mortality of each disease in each country during 1999–2014.

Fig. 5. Interannual changes in total premature mortality (i.e., attributable to COPD, IHD, stroke, LC, and ALRI) in South and Southeast Asia from 1999 to 2014.

1510


Fig. 6. Spatial distributions of interannual trends of total premature mortality (i.e., attributable to COPD, IHD, stroke, LC, and ALRI) due to PM2.5 exposure in South and Southeast Asia from 1999 to 2014. Negative or positive trends were all significant at the 95% confidence level after testing (p b 0.05). Non-significant trends in the study area are shown in white.

The long-term trends of premature deaths from stroke and IHD presented spatial patterns similar to the interannual trends of total premature mortality across the entire SSEA region during 1999–2014 (SI Fig. S4). The consistent significant increases in East Pakistan, North India, and Bangladesh and the significant decreases in Naypyidaw (Myanmar) and Phu Tho Province (Vietnam) corresponded well with the changing trends of total premature deaths (1999–2014), indicating that the trends of change of total premature deaths in SSEA were attributable primarily to the changes in stroke and IHD. However, COPD also contributed to the increase of premature deaths in North India, with an annual rate of increase of b0.01 deaths/km2/year. Throughout the study period, LC had the smallest net increase of premature deaths; therefore, the spatial distribution of the trend of change in each pixel showed almost no variation. ALRI also experienced a small increase of premature deaths during 1999–2014. Therefore, many areas presented no obvious trend of change, or the trend was not significant. 3.3. Net changes in premature mortality during 1999–2014 Overall, the absolute number of deaths attributable to PM2.5 increased from 1,179,400 in 1999 to 1,724,900 in 2014 due to both an increase in air pollution and population growth (Fig. 7). The estimated premature mortality is a function of both the population and air quality in each grid cell, resulting in high levels of premature mortality in heavily polluted and densely populated areas. Regions with high numbers of premature deaths were associated with the major cities across India, Naypyidaw (Myanmar), Bangkok (Thailand), and Hanoi (Vietnam). By 2014, there was a significant increase in both the magnitude and the areal extent of high premature mortality. Specifically, in North India and the major cities of India, the rate of change between 1999 and 2014 was generally N0.2 death/km2. Areas with distinct increases in premature mortality corresponded well with areas with already high numbers of premature deaths, whereas areas with low numbers of premature deaths experienced little change between 1999 and 2014. The major contributors of stroke and IHD also presented spatial patterns in the changes of premature death (SI Fig. S5) similar to the total changes (Fig. 7). Notably, SA countries showed higher numbers of premature deaths than SEA, not only in

1999 and 2014, but also in their changes between 1999 and 2014. Because of low population densities and PM2.5 concentrations, SEA countries witnessed relatively few premature deaths and small changes. In some countries (e.g., Myanmar, Thailand, Cambodia, and Laos), the population density was extremely low and decreased between 1999 and 2014. Therefore, the number of premature deaths changed only slightly, despite some small increases in PM2.5 concentration. The numbers of premature deaths in SEA were high only in major cities, such as Hanoi and Ho Chi Minh City. Because of increases in the populations of these cities, premature deaths showed distinct increases even though PM2.5 concentrations changed little. However, Naypyidaw in Myanmar and Phu Tho Province in North Vietnam are exceptions because their populations experienced obvious net decreases between 1999 and 2014. Overall, the number of premature deaths did not always increase in all areas (grid cells) within an individual country (Fig. 7). It generally occurred where both PM2.5 concentration and population density were high (e.g., East Pakistan, North India, and Bangladesh) or in areas with very high population densities (e.g., Bangkok and Hanoi). Furthermore, these regions were usually the areas with the highest increments in absolute premature deaths between 1999 and 2014. Conversely, the number of premature deaths in areas with reduced PM2.5 concentrations or low population densities in SEA countries changed little between 1999 and 2014. The overall changes in the spatial pattern of premature mortality across SSEA between 1999 and 2014 were evaluated using SDE analysis, which revealed elements distributed within the main region (Fig. 8, SI Table S4). The results showed that premature deaths in SSEA were concentrated primarily in India. The SDE showed that the main distribution of premature deaths was aligned in the northwest–southeast direction and that it decreased gradually in this direction. The major and minor axes of the ellipses declined from 2792 to 2781 km and from 1630 to 1586 km, respectively. The reduction of the two axes of the ellipses illustrated the spatial aggregation tendency and the spatial changes in premature mortality from 1999 to 2014. The major and minor axes of the PM2.5 ellipse decreased in length from 3714 to 3643 km and from 1441 to 1408 km, respectively, between 1999 and 2014. Meanwhile, the major and minor axes of the population ellipse shortened from 3064 to 3034 km and from 1780 to 1751 km, respectively. The different


1511

Fig. 7. Spatial distributions of PM2.5 concentration, population density, and total premature mortality due to PM2.5 in South and Southeast Asia in 1999 and 2014 and their changes between 1999 and 2014.

Fig. 8. Spatial changes in the median center and standard deviation ellipse of PM2.5 concentrations, population, and premature mortalities in South and Southeast Asia in 1999 and 2014.

1512


estimated the premature mortality in SSEA from COPD, IHD, stroke, LC, and ALRI, linked to PM2.5 concentrations, to be 1.36 million (100 × 100 km spatial resolution) in 2010. Our estimate of 1.52 million premature deaths in SSEA attributable to PM2.5 pollution in 2010 showed good agreement. However, some differences were found in certain countries in specific years. This study calculated the PM2.5-induced premature death linked to COPD, IHD, stroke, and LC in India in 2007 as 878,000, which was similar to the figure of 811,000 reported by Chowdhury and Dey (2016) but higher than the estimate of 584,000 by Zhang et al. (2017). However, there was an overall contribution of 74% from stroke and IHD to the total number of premature deaths in SSEA, which was very close to their corresponding proportions at the global level (i.e., ~70%) (Apte et al., 2015).

lengths of the major and minor axes of PM2.5, population, and premature deaths demonstrated different distributional patterns of these three elements across SSEA. The ratio between the major and minor axes of the ellipses calculated by premature deaths presented an increasing tendency between 1999 and 2014, indicating that the directional trend became increasingly evident, which displayed patterns consistent with PM2.5 concentration and population. By 2014, the ellipse of premature deaths moved northwest relative to 1999. This tendency was attributed to the rapid increase in premature death in SA compared with SEA between 1999 and 2014 (Fig. 7). The median centers were detected to the northwest of the arithmetic center of the SDE, which suggested that higher numbers of premature deaths were concentrated in the northwest (i.e., SA countries) of SSEA rather than the southeast (i.e., SEA countries). This was supported by the consistent spatial patterns of PM2.5 concentration and population, which revealed the same distributions with decreasing tendencies from the northwest to the southeast. From 1999 to 2014, the center of premature deaths across SSEA moved from 23.2°N, 82.8°E to 23.4°N, 82.6°E (SI Table S4). The shift of the center from the southeast to the northwest indicated faster increases in premature deaths in East Pakistan, North India, and Bangladesh, which had increasing influences on the total premature deaths of SSEA. The center of population also moved northwest from 1999 to 2014, which indicated greater population increase in SA. However, the center of PM2.5 concentration showed the opposite pattern. Although the growing influence of PM2.5 concentration in SEA countries resulted in the southeastward movement of the center, PM2.5 concentration had little effect on the patterns and changes of premature deaths because the PM2.5 concentration in SEA remained low relative to SA. Therefore, the trend of change of premature mortality from 1999 to 2014 comprised the integrated effects of PM2.5 concentration and population. In SSEA, population had the greatest effect on the spatial distribution of premature deaths because of the similarity in their patterns over time. During 1999–2014, premature deaths attributed to PM2.5 have grown 38% in SSEA (Table 1). The three largest contributing countries (India, Bangladesh, and Pakistan) experienced rapid changes with incremental rates ranging from 28% to 44%. Overall, the estimated numbers of premature deaths increased for all countries except Sri Lanka, which showed a declining trend and the lowest growth rate of −32%. Among all SSEA countries, the largest rate of growth in premature mortality was found in Bhutan (83%), followed by Cambodia (58%) and Laos (49%). However, although these countries recorded faster rates of increase than India, Bangladesh, and Pakistan, their annual average numbers of premature deaths remained low. Therefore, the absolute increases in premature death remained small. Because of the increases in PM2.5 concentration and population, India had a 39% net increase in premature deaths (386,600 persons) during 1999–2014, which contributed considerably to the total change in premature deaths across SSEA. Among the five studied diseases, the largest two contributors, stroke and IHD, both experienced notable increases in premature death from 448,800 to 672,000 (net increase: 223,200) and from 415,000 to 608,500 (net increase: 193,500), respectively, with ROCs of up to 40% and 38%, respectively, between 1999 and 2014. However, larger ROCs were found for COPD and LC (47% and 44%, respectively), which also presented net increases of 88,100 and 13,600 deaths, respectively. ALRI showed the smallest ROC (17%), which resulted in an increase of 27,200 deaths during the study period.

Estimates of premature death due to PM2.5 pollution can be subject to large uncertainties (Roman et al., 2008; Brauer et al., 2012). First, the baseline mortality across SSEA for each country was used as an input (i.e., not pixel-based data). Although the baseline mortality varied among countries over time, using the uniform baseline mortality within each country augmented the uncertainties. For example, using a single value of baseline mortality for India prevented the description of spatial variations among pixels. By introducing a spatially varying baseline mortality (i.e., pixel-based, not country-based data), the uncertainties of premature death estimations can be alleviated considerably. Second, the IER relationships for stroke and IHD dominated total mortality and were found to be supralinear, increasing most rapidly at low concentrations (Cohen et al., 2017; GBD 2016 Risk Factors Collaborators, 2017). The relationship was regressed linearly using the global datasets with supralinear relationships (Apte et al., 2015). Additional long-term cohort studies are needed to more accurately constrain the shape of the PM2.5 concentration–response relationships in SSEA, especially under the cleanest and most polluted conditions (Apte et al., 2015; Cohen et al., 2017). Third, the populations in intervening years between 1999 and 2014 were resampled for the years 2000, 2005, 2010, and 2015. Therefore, the linearly extrapolated populations might not truly reflect the actual populations. Finally, the satellite-retrieved PM2.5 concentration estimation in SSEA achieved a high correlation (0.82) with the measured point data (Shi et al., 2018). However, the inverted PM2.5 values in SSEA were generally lower than the ground-based measurements (Shi et al., 2018) because of an AOD bias in the retrievals over South and East Asia (Kahn et al., 2009) and the Indian subcontinent (Dey and Di Girolamo, 2010) and because of missing satellite observations (van Donkelaar et al., 2015). From above analysis, we concluded that the amount of PM2.5 concentrations, population and baseline mortality were the sources that caused the uncertainties in the estimation. According to van Donkelaar et al. (2016), the typical uncertainty of PM2.5 concentration was on the order of 10–20% in South and Southeast Asia. The population was within an uncertainty range of approximately ±50% around the mean value (Doxsey-Whitfield et al., 2015). Given the large presumed uncertainties of statistics for baseline mortality, the probability of the value used was assumed to have a normal distribution with a coefficient of variation of 30%. We subsequently ran Monte Carlo simulations to quantitatively estimate the range of premature deaths with a 95% CI (SI Table S3) of each kind of disease for each country in SSEA.

3.4. Comparison with other studies

4. Conclusions

The estimated long-term (1999–2014) and high-spatial-resolution (0.01° × 0.01°) data of premature deaths due to PM2.5 exposure were in good agreement with other studies. For example, our estimate of 1.67 million premature deaths in 2013 in SSEA, attributed to PM2.5 concentrations, agreed well with the GBD 2013 calculation of 1.80 million (Brauer et al., 2016; World Bank, 2016). Lelieveld et al. (2015)

This study quantified the 0.01° × 0.01° high-resolution premature mortality (from COPD, IHD, stroke, LC and ALRI) induced by PM2.5 exposure in SSEA during 1999–2014. In addition, the long-term trends and spatial patterns of PM2.5-induced premature mortality in SSEA during 1999–2014 were analyzed. The total annual premature deaths attributable to PM2.5 reached 1,447,000 (95% CI: 935,300–2,541,100) in SSEA.

3.5. Uncertainties


The estimated annual numbers of premature deaths were 187,700 (95% CI: 101,400–271,900) for COPD, 509,400 (95% CI: 374,600–668,000) for IHD, 556,400 (95% CI: 442,800–673,200) for stroke, 30,700 (95% CI: 12,500–41,100) for LC, and 162,900 (95% CI: 87,300–259,600) for ALRI, respectively. Among the five diseases, stroke and IHD were the two largest contributors, accounting for 74% of the total premature deaths in SSEA. Among all countries, India contributed the largest premature deaths across the study area with 991,600 (95% CI: 378,100–1,872,800) and 69%. The total number of premature deaths attributable to PM2.5 concentrations in SSEA showed an increasing trend from 1999 to 2014, with a growth rate of 38% and a net increase of 545,500 deaths. Stroke was the largest contributor to the total rise in premature deaths from 1999 to 2010. India was the primary contributor to the increase in total premature deaths in SSEA. The spatial pattern showed that high numbers of premature deaths were concentrated in North India, Bangladesh, East Pakistan, and some metropolises in Southeast Asian countries. These areas, with high PM2.5 concentrations and population densities, also showed changes in premature deaths. The long-term trends in premature deaths presented consistent incremental tendencies for COPD, IHD, stroke, LC, and ALRI in all countries except Sri Lanka during 1999–2014. Although some uncertainties exist, this study contributes valuable insights into the strong spatial variations and long-term trends in PM2.5-related premature mortality in SSEA. The following work will focus on the attributions of the increasing trends in PM2.5-induced premature mortality to individual driving factor. The contributions of the increase in population both in number and spatial distribution and the changes in baseline mortality of certain diseases and PM2.5 concentrations will be further investigated and quantified. Supplementary data to this article can be found online at https://doi. org/10.1016/j.scitotenv.2018.03.146. Conflict of interest The authors declare no conflict of interest. Acknowledgements This research was supported by the NIES GOSAT-2 Project, Japan, CAS Pioneer Hundred Talents Program, China (Y8YR2200QM, Y7S00100CX, Y7S0010030) and National Natural Science Foundation of China (41701498). The authors thank Dr. Aaron van Donkelaar of the Atmospheric Physics Institute at Dalhousie University in Canada who offered the PM2.5 data used in this research. References GBD 2016 Risk Factors Collaborators, 2017. Global, regional, and national comparative risk assessment of 84 behavioural, environmental and occupational, and metabolic risks or clusters of risks, 1990–2016: a systematic analysis for the Global Burden of Disease Study 2016. Lancet 390, 1345–1422. Anenberg, S.C., Schwartz, J., Shindell, D., Amann, M., Faluvegi, G., Klimont, Z., JanssensMaenhout, G., Pozzoli, L., Van Dingenen, R., Vignati, E., Emberson, L., Muller, N.Z., West, J.J., Williams, M., Demkine, V., Hicks, W.K., Kuylenstierna, J., Raes, F., Ramanathan, V., 2012. Global air quality and health co-benefits of mitigating nearterm climate change through methane and black carbon emission controls. Environ. Health Perspect. 120, 831–839. Apte, J.S., Marshall, J.D., Cohen, A.J., Brauer, M., 2015. Addressing global mortality from ambient PM2.5. Environ. Sci. Technol. 49, 8057–8066. Bonjour, S., Adair-Rohani, H., Wolf, J., Bruce, N.G., Mehta, S., Pruss-Ustun, A., Lahiff, M., Rehfuess, E.A., Mishra, V., Smith, K.R., 2013. Solid fuel use for household cooking: country and regional estimates for 1980-2010. Environ. Health Perspect. 121, 784–790. Brauer, M., Amann, M., Burnett, R.T., Cohen, A., Dentener, F., Ezzati, M., Henderson, S.B., Krzyzanowski, M., Martin, R.V., Van Dingenen, R., van Donkelaar, A., Thurston, G.D., 2012. Exposure assessment for estimation of the global burden of disease attributable to outdoor air pollution. Environ. Sci. Technol. 46, 652–660. Brauer, M., Freedman, G., Frostad, J., van Donkelaar, A., Martin, R.V., Dentener, F., van Dingenen, R., Estep, K., Amini, H., Apte, J.S., Balakrishnan, K., Barregard, L., Broday, D., Feigin, V., Ghosh, S., Hopke, P.K., Knibbs, L.D., Kokubo, Y., Liu, Y., Ma, S., Morawska, L., Sangrador, J.L., Shaddick, G., Anderson, H.R., Vos, T., Forouzanfar, M.H., Burnett, R.T., Cohen, A., 2016. Ambient air pollution exposure

1513

estimation for the Global Burden of Disease 2013. Environ. Sci. Technol. 50, 79–88. Burnett, R.T., Pope, C.A.I.I.I., Ezzati, M., Olives, C., Lim, S.S., Mehta, S., Shin, H.H., Singh, G., Hubbell, B., Brauer, M., Anderson, H.R., Smith, K.R., Balmes, J.R., Bruce, N.G., Kan, H., Laden, F., Prüss-Ustün, A., Turner, M.C., Gapstur, S.M., Diver, W.R., Cohen, A., 2014. An integrated risk function for estimating the global burden of disease attributable to ambient fine particulate matter exposure. Environ. Health Perspect. 122, 397–403. Chowdhury, S., Dey, S., 2016. Cause-specific premature death from ambient PM2.5 exposure in India: estimate adjusted for baseline mortality. Environ. Int. 91, 283–290. Cohen, A.J., Brauer, M., Burnett, R., Anderson, H.R., Frostad, J., Estep, K., Balakrishnan, K., Brunekreef, B., Dandona, L., Dandona, R., Feigin, V., Freedman, G., Hubbell, B., Jobling, A., Kan, H., Knibbs, L., Liu, Y., Martin, R., Morawska, L., Pope III, C.A., Shin, H., Straif, K., Shaddick, G., Thomas, M., van Dingenen, R., van Donkelaar, A., Vos, T., Murray, C.J.L., Forouzanfar, M.H., 2017. Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: an analysis of data from the Global Burden of Diseases Study 2015. Lancet 389, 1907–1918. Dey, S., Di Girolamo, L., 2010. A climatology of aerosol optical and microphysical properties over the Indian subcontinent from 9 years (2000–2008) of Multiangle Imaging Spectroradiometer (MISR) data. J. Geophys. Res. 115, D15204. van Donkelaar, A., Martin, R.V., Brauer, M., Kahn, R., Levy, R., Verduzco, C., Villeneuve, P.J., 2010. Global estimates of ambient fine particulate matter concentrations from satellite-based aerosol optical depth: development and application. Environ. Health Perspect. 118, 847–855. van Donkelaar, A., Martin, R.V., Brauer, M., Boys, B.L., 2015. Use of satellite observations for long-term exposure assessment of global concentrations of fine particulate matter. Environ. Health Perspect. 123, 135–143. van Donkelaar, A., Martin, R.V., Brauer, M., Hsu, N.C., Kahn, R.A., Levy, R.C., Lyapustin, A., Sayer, A.M., Winker, D.M., 2016. Global estimates of fine particulate matter using a combined geophysical-statistical method with information from satellites, models, and monitors. Environ. Sci. Technol. 50, 3762–3772. Doxsey-Whitfield, E.K., MacManus, K., Adamo, S.B., Pistolesi, L., Squires, J., Borkosvska, O., Baptista, S.R., 2015. Taking advantage of the improved availability of census data: a first look at the gridded population of the world, version 4 (GPWv4). Pap. Appl. Geogr. 1, 226–234. Ghude, S.D., Chate, D.M., Jena, C., Beig, G., Kumar, R., Barth, M.C., Pfister, G.G., Fadnavis, S., Pithani, P., 2016. Premature mortality in India due to PM2.5 and ozone exposure. Geophys. Res. Lett. 43, 4650–4658. Guo, Y., Tang, Q., Gong, D., Zhang, Z., 2017. Estimating ground-level PM2.5 concentrations in Beijing using a satellite-based geographically and temporally weighted regression model. Remote Sens. Environ. 198, 140–149. He, Q., Huang, B., 2018. Satellite-based mapping of daily high-resolution ground PM2.5 in China via space-time regression modeling. Remote Sens. Environ. 206, 72–83. Health Effects Institute, 2017. State of Global Air 2017. Special Report. Health Effects Institute, Boston, MA https://www.stateofglobalair.org/sites/default/files/SoGA2017_report.pdf. Hu, J., Huang, L., Chen, M., Liao, H., Zhang, H., Wang, S., Zhang, Q., Ying, Q., 2017. Premature mortality attributable to particulate matter in China: source contributions and responses to reductions. Environ. Sci. Technol. 51, 9950–9959. Kahn, R.A., Nelson, D.L., Garay, M.J., Levy, R.C., Bull, M.A., Diner, D.J., Martonchik, J.V., Paradise, S.R., Hansen, E.G., Remer, L.A., 2009. MISR aerosol product attributes and statistical comparisons with MODIS. IEEE Trans. Geosci. Remote Sen. 47, 4095–4114. Koplitz, S.N., Jacob, D.J., Sulprizio, M.P., Myllyvirta, L., Reid, C., 2017. Burden of disease from rising coal-fired power plant emissions in Southeast Asia. Environ. Sci. Technol. 51, 1467–1476. Lelieveld, J., Barlas, C., Giannadaki, D., Pozzer, A., 2013. Model calculated global, regional and megacity premature mortality due to air pollution. Atmos. Chem. Phys. 13, 7023–7037. Lelieveld, J., Evans, J.S., Fnais, M., Giannadak, D., Pozzer, A., 2015. The contribution of outdoor air pollution sources to premature mortality on a global scale. Nature 525, 367–371. Li, Z., Zhang, Y., Shao, J., Li, B., Hong, J., Liu, D., Li, D., Wei, P., Li, W., Li, L., Zhang, F., Guo, J., Deng, Q., Wang, B., Cui, C., Zhang, W., Wang, Z., Lv, Y., Xu, H., Chen, X., Li, L., Qie, L., 2016. Remote sensing of atmospheric particulate mass of dry PM2.5 near the ground: method validation using ground-based measurements. Remote Sens. Environ. 173, 59–68. Liu, J., Han, Y., Tang, X., Zhu, J., Zhu, T., 2016. Estimating adult mortality attributable to PM2.5 exposure in China with assimilated PM2.5 concentrations based on a ground monitoring network. Sci. Total Environ. 568, 1253–1262. Liu, M., Huang, Y., Ma, Z., Jin, Z., Liu, X., Wang, H., Liu, Y., Wang, J., Jantunen, M., BiDr, J., KinneyDr, P.L., 2017. Spatial and temporal trends in the mortality burden of air pollution in China: 2004–2012. Environ. Int. 98, 75–81. Lv, B., Hu, Y., Chang, H.H., Russell, A.G., Cai, J., Xu, B., Bai, Y., 2017. Daily estimation of ground-level PM2.5 concentrations at 4 km resolution over Beijing-Tianjin-Hebei by fusing MODIS AOD and ground observations. Sci. Total Environ. 615, 177–186. Ma, Z., Hu, X., Huang, L., Bi, J., Liu, Y., 2014. Estimating ground-level PM2.5 in China using satellite remote sensing. Environ. Sci. Technol. 48, 7436–7444. Maji, K.J., Dikshit, A.K., Arora, M., Deshpande, A., 2018. Estimating premature mortality attributable to PM2.5 exposure and benefit of air pollution control policies in China for 2020. Sci. Total Environ. 612, 683–693. Peng, J., Chen, S., Lv, H., Liu, Y., Wu, J., 2016. Spatiotemporal patterns of remotely sensed PM2.5 concentration in China from 1999 to 2011. Remote Sens. Environ. 174, 109–121. Roman, H.A., Walker, K.D., Walsh, T.L., Conner, L., Richmond, H.M., Hubbell, B.J., Kinney, P.L., 2008. Expert judgment assessment of the mortality impact of changes in ambient fine particulate matter in the U.S. Environ. Sci. Technol. 42, 2268–2274.

1514


Shi, Y., Yamaguchi, Y., 2014. A high-resolution and multi-year emissions inventory for biomass burning in Southeast Asia during 2001–2010. Atmos. Environ. 98, 8–16. Shi, Y., Xiao, J., Shen, Y., Yamaguchi, Y., 2012. Quantifying the spatial differences of landscape change in the Hai River Basin, China, in the 1990s. Int. J. Remote Sens. 33, 4482–4501. Shi, Y., Matsunaga, T., Yamaguchi, Y., 2015. High-resolution mapping of biomass burning emissions in three tropical regions. Environ. Sci. Technol. 49, 10806–10814. Shi, Y., Matsunaga, T., Yamaguchi, Y., Li, Z., Gu, X., Chen, X., 2018. Long-term trends and spatial patterns of satellite-retrieved PM2.5 concentrations in South and Southeast Asia from 1999 to 2014. Sci. Total Environ. 615, 177–186. Silva, R.A., West, J.J., Zhang, Y., Anenberg, S.C., Lamarque, J.F., Shindell, D.T., Collins, W.J., Dalsoren, S., Faluvegi, G., Folberth, G., Horowitz, L.W., Nagashima, T., Naik, V., Rumbold, S., Skeie, R., Sudo, K., Takemura, T., Bergmann, D., Cameron-Smith, P., Cionni, I., Doherty, R.M., Eyring, V., Josse, B., MacKenzie, I.A., Plummer, D., Righi, M., Stevenson, D.S., Strode, S., Szopa, S., Zeng, G., 2013. Global premature mortality due to anthropogenic outdoor air pollution and the contribution of past climate change. Environ. Res. Lett. 8, 034005. Wachowicz, M., Liu, T., 2016. Finding spatial outliers in collective mobility patterns coupled with social ties. Int. J. Geogr. Inf. Sci. 30, 1806–1831.

Wang, B., Shi, W., Miao, Z., 2015. Confidence analysis of standard deviational ellipse and its extension into higher dimensional Euclidean space. PLoS One 10 (3), e0118537. World Bank, 2016. The Cost of Air Pollution: Strengthening the Economic Case for Action. World Bank Group, Washington, D.C. http://documents.worldbank.org/curated/en/ 781521473177013155/The-cost-of-air-pollution-strengthening-the-economic-casefor-action. Xie, R., Sabel, C.E., Lu, X., Zhu, W., Kan, H., Nielsen, C.P., Wang, H., 2016. Long-term trend and spatial pattern of PM2.5 induced premature mortality in China. Environ. Int. 97, 180–186. Zhang, Y., Li, Z., 2015. Remote sensing of atmospheric fine particulate matter (PM2.5) mass concentration near the ground from satellite observation. Remote Sens. Environ. 160, 252–262. Zhang, Q., Jiang, X., Tong, D., Davis, S.J., Zhao, H., Geng, G., Feng, T., Zheng, B., Lu, Z., Streets, D.G., Ni, R., Brauer, M., van Donkelaar, A., Martin, R.V., Huo, H., Liu, Z., Pan, D., Kan, H., Yan, Y., Lin, J., He, K., Guan, D., 2017. Transboundary health impacts of transported global air pollution and international trade. Nature 543, 705–709.