Decomposing the Long-term Variation in Population Exposure to Outdoor PM_2.5 in the Greater Bay Area of China Using Satellite Observations

Abstract

The Greater Bay Area (GBA) of China is experiencing a high level of exposure to outdoor PM_2.5 pollution. The variations in the exposure level are determined by spatiotemporal variations in the PM_2.5 concentration and population. To better guide public policies that aim to reduce the population exposure level, it is essential to explicitly decompose and assess the impacts of different factors. This study took advantage of high-resolution satellite observations to characterize the long-term variations in population exposure to outdoor PM_2.5 for cities in the GBA region during the three most-recent Five-Year Plan (FYP) periods (2001–2015). A new decomposition method was then used to assess the impact of PM_2.5 variations and demographic changes on the exposure variation. Within the decomposition framework, an index of pollution-population-coincidence–induced PM_2.5 exposure (PPCE) was introduced to characterize the interaction of PM_2.5 and the population distribution. The results showed that the 15-year average PPCE levels in all cities were positive (e.g., 6 µg/m³ in Guangzhou), suggesting that unfavorable city planning had led to people dwelling in polluted areas. An analyses of the spatial differences in PM_2.5 changes showed that urban areas experienced a greater decrease in PM_2.5 concentration than did rural areas in most cities during the 11th (2006–2010) and 12th (2011–2015) FYP periods. These spatial differences in PM_2.5 changes reduced the PPCE levels of these cities and thus reduced the exposure levels (by as much as -0.58 µg/m³/year). The population migration resulting from rapid urbanization, however, increased the PPCE and exposure levels (by as much as 0.18 µg/m³/year) in most cities during the three FYP periods considered. Dongguan was a special case in that the demographic change reduced the exposure level because of its rapid development of residential areas in cleaner regions adjacent to Shenzhen. The exposure levels in all cities remained high because of the high mean PM_2.5 concentrations and their positive PPCE. To better protect public health, control efforts should target densely populated areas and city planning should locate more people in cleaner areas.

1. Introduction

Long-term exposure to outdoor particulate matter with an aerodynamic diameter of <2.5 μm (PM_2.5) is associated with a range of adverse health effects [1,2,3,4,5]. The level of population exposure to outdoor PM_2.5 in China is much higher than that in the U.S. or Europe [6]. As one of the major city clusters in China, the Pearl River Delta (PRD) region is exposed to a high level of PM_2.5 pollution [7]. The population-weighted mean PM_2.5 concentrations have been extensively used by researchers as an indicator of the population exposure level [8,9,10,11,12,13,14,15]. It is estimated by averaging the ambient PM_2.5 concentration associated with each individual within the study region. Burnett et al. [16] assessed the global population exposure to outdoor PM_2.5 in 2015 and concluded that the population-weighted mean PM_2.5 concentrations varied from the lowest levels in Canada/U.S. (7.9 μg/m³) to the highest levels in China (57.5 μg/m³) and India (74.0 μg/m³). Using satellite-derived PM_2.5 data, Lin et al. [17] estimated the population-weighted mean PM_2.5 concentration to be 43.9 μg/m³ in the PRD region from 2000 to 2014. In addition, their results showed that 84.0% of the population lived in areas with PM_2.5 concentrations that exceeded the World Health Organization (WHO) Interim Target 1 (IT-1) or the Chinese National Ambient Air Quality Standard (NAAQS), both of which are 35 µg/m³. In February 2019, the “Outline Development Plan for Guangdong-Hong Kong-Macao Greater Bay Area” (the Greater Bay Area (GBA) Plan) was promulgated, which emphasized green development in this region. The regulation of pollutant emissions was tightened to further alleviate air pollution.

The variations in the mean PM_2.5 concentration level directly affect a population’s exposure level in any given city. During the past decade, the governments of Guangdong Province and Hong Kong promulgated a series of control measures to reduce the PM_2.5 concentration [18]. Using satellite observations, Ma et al. [19] showed that the mean PM_2.5 concentration in the GBA region remained relatively constant from 2004 to 2007 and decreased from 2008 to 2013. Another study reported that all cities in the GBA region experienced an increasing mean PM_2.5 concentration during the 10th Five-Year Plan (FYP) period (2001–2005) and a decreasing mean PM_2.5 concentration during the 11th (2006–2010) and 12th (2011–2015) FYP periods [20].

A decrease in the mean PM_2.5 concentration is not necessarily associated with a decreasing population exposure level for the city if the PM_2.5 concentrations decrease in vast rural areas but increase in densely populated urban areas. The spatial differences in PM_2.5 changes therefore play an important role in exposure management [21]. Ideally, control efforts to reduce the PM_2.5 concentration should target densely populated urban areas. Lin et al. [22] assessed the differences in PM_2.5 variations between urban and rural areas for Chinese provinces, and they showed that the urban areas in the GBA region experienced a greater decrease in the PM_2.5 concentration than did rural areas during the 11th and 12th FYP periods. This type of spatial difference in PM_2.5 change assists a region to reduce its population exposure level.

Rapid urbanization typically leads to population migration into large urban areas. As reported by the Chinese statistical yearbook, the urban population in Guangdong Province increased from 47.5 million (55% of total population) in 2000 to 74.5 million (69% of total population) in 2015 (http://www.stats.gov.cn/tjsj/ndsj/). As urban areas are typically highly polluted, the migration of people to the urban areas in the GBA region during the past decade tended to increase its overall population exposure level. Ideally, demographic changes can reduce the exposure level if city planners relocate the population away from polluted areas. Further scientific and quantitative investigation of the impact of demographic change is needed.

The variations in the population exposure level are therefore determined by changes in the mean PM_2.5 concentration, spatial differences in PM_2.5 change and demographic changes. The level of population exposure to PM_2.5 for a city decreases if: (1) the mean PM_2.5 concentration declines, (2) urban areas experience a greater reduction of PM_2.5 concentration than rural areas, or (3) city planners relocate people away from polluted areas. To better guide public policies that aim to reduce exposure levels, it is essential to explicitly decompose and assess the impact of different factors.

Traditional studies have used ground-based networks to monitor the PM_2.5 concentration. However, such monitoring in China is much more limited than in developed countries [23]. Satellite remote sensing is an important step toward filling this data gap [24,25,26]. This study took advantage of high-resolution satellite observations to characterize long-term variations in human exposure to PM_2.5 for cities in the GBA region during the three most-recent FYP periods (2001–2015). A new decomposition method was then used to assess the impact of PM_2.5 variations and demographic changes on the exposure variation. Finally, suggestions for future environmental policies and urbanization planning were developed.

2. Data

2.1. Population Density

The GBA region of China contains nine major cities in Guangdong Province [Guangzhou (GZ), Shenzhen (SZ), Zhuhai (ZH), Dongguan (DG), Foshan (FS), Zhongshan (ZS), Zhaoqing (ZQ), Jiangmen (JM) and Huizhou (HZ)]. Guangzhou is the capital city of Guangdong Province. Shenzhen and Zhuhai are two special economic zones of China. The GBA region also contains two special administrative zones of China [i.e., Hong Kong (HK) and Macau (MC)]. All eleven administrative regions as cities were considered.

The yearly population density data was obtained at a spatial resolution of 1 km for the GBA region for 2001 and 2015 from the LandScan population database (http://web.ornl.gov/sci/landscan/). LandScan population data was constructed with the use of spatial data and imagery analyses to disaggregate census counts within administrative boundaries. A linear interpolation method was then applied to derive population densities for other years of the study period. The year- and city-specific gross population data were obtained from 2001 to 2015 from local statistical yearbooks (http://www.gdstats.gov.cn/). The LandScan-derived population data were adjusted by year- and city-specific factors to match the gross populations from the statistical yearbooks. Figure 1 shows the spatial distribution of the 15-year mean of population density in the GBA region. The population densities exceeded 10³ persons/km² in urban areas of the GBA region.

2.2. Satellite-Derived PM_2.5

A high-resolution satellite-based PM_2.5 dataset was taken advantage of to characterize the long-term variations in the PM_2.5 concentration in the GBA region (http://envf.ust.hk/dataview/aod2pm/current). The algorithm was first used to build aerosol optical depth (AOD) data at a spatial resolution of 0.01° × 0.01° using spectral data from the two Moderate Resolution Imaging Spectroradiometer (MODIS) instruments aboard the Terra and Aqua satellites [27]. The verification of the satellite-retrieved AOD against the ground-based sunphotometer observations showed retrieval errors within 15–20%. An observational data-driven AOD-PM_2.5 algorithm, which used ground-observed visibility and relative humidity data as inputs, was then applied to derive the ground-level PM_2.5 concentrations [28,29]. The uncertainty in the PM_2.5 concentrations resulted from the input data such as the AOD, meteorological values and model parameters. The evaluation of the satellite-derived PM_2.5 data against observations from the ground monitoring network within the GBA region during the entire study period identified the correlation coefficient, root mean square error, mean absolute bias and mean absolute percentage bias as 0.86 (N = 363), 4.7 µg/m³, 3.6 ± 3.1 µg/m³, and 9.3 ± 8.1%, respectively [20]. This accuracy is comparable to those with a correlation coefficient of 0.76–0.82 (N = 974–1145) in the U.S. and 0.74–0.79 (N = 45–68) in China for other existing long-term PM_2.5 data sets [11,30,31,32]. This study focused on residential areas with population densities of ≥10 persons/km². Figure 2 shows the spatial distribution of the 15-year mean of the PM_2.5 concentration in the residential areas of the GBA region. The PM_2.5 concentrations exceeded 60 µg/m³ in the central GBA region.

3. Methodology

The exposure decomposition framework is summarized as follows (more information can be found in the Appendix). In a given region, the spatial distributions of PM_2.5 concentration and population density are denoted as $c_{i,j}$ and ${\rho }_{i,j}$, respectively, where i = {1, 2, 3,…, X} and j = {1, 2, 3,…, Y}. The population-weighted mean PM_2.5 concentration (c_ρ) can be estimated by Equation (1):

$$c_{\rho }=\frac{{{\displaystyle \sum }}_{i=1}^X{{\displaystyle \sum }}_{j=1}^Y{c}_{i,j}·{\rho }_{i,j}}{{{\displaystyle \sum }}_{i=1}^X{{\displaystyle \sum }}_{j=1}^Y{\rho }_{i,j}}$$

(1)

The spatial averages of the PM_2.5 concentration and population density in a region are represented by c₀ and ρ₀, respectively. The c₀ and ρ₀ values are estimated by averaging the PM_2.5 concentration and population density from all grids within the study region. Further, $c{{}^{\prime }}_{i,j}=c_{i,j}-c_0$ and $p{{}^{\prime }}_{i,j}=\frac{{\rho }_{i,j}-{\rho }_0}{{\rho }_0}$ was then defined as the deviation of the PM_2.5 concentration from c₀ and the relative deviation of population density from ρ₀, respectively. The $c{{}^{\prime }}_{i,j}$ and $p{{}^{\prime }}_{i,j}$ values therefore represent the differences in the PM_2.5 concentration and population density, respectively, from their regional average levels. The term c_ρ can be estimated by Equation (2):

$$c_{\rho }=c_0+\frac{1}{A}·{\displaystyle \sum }_{i=1}^X{\displaystyle \sum }_{j=1}^Y{c^{\prime }}_{i,j}·{p^{\prime }}_{i,j}$$

(2)

where A = X·Y is the total grid number, representing the area of the study region. The authors defined ${\mathit{C}}^{\prime }$ and ${\mathit{P}}^{\prime }$ as the matrices of ${c^{\prime }}_{i,j}$ and ${p^{\prime }}_{i,j}$, respectively. The term c_ρ can then be expressed as Equation (3):

$$c_{\rho }=c_0+\frac{1}{A}·\left(\mathit{C}{}^{\prime }·\mathit{P}{}^{\prime }\right)$$

(3)

An index of pollution-population-coincidence–induced PM_2.5 exposure (PPCE) of $c_1=\frac{1}{A}·\left(\mathit{C}{}^{\prime }·\mathit{P}{}^{\prime }\right)$ was introduced to represent the PM_2.5 exposure resulting from the spatial coincidence between the PM_2.5 concentration and population density. This PPCE is associated with the dot product of ${\mathit{C}}^{\prime }$ and ${\mathit{P}}^{\prime }$.

An increase in population density in polluted areas leads to a positive value of c₁, which increases the c_ρ level. Ideally, more people would reside in cleaner areas, which would be associated with a negative value of c₁. The trend in c_ρ can be expressed as Equation (4):

$$\frac{d{c}_{\rho }}{dt}=\frac{d{c}_0}{dt}+\frac{1}{A}·\left({\mathit{P}}^{\prime }·\frac{d{\mathit{C}}^{\prime }}{dt}\right)+\frac{1}{A}·\left({\mathit{C}}^{\prime }·\frac{d{\mathit{P}}^{\prime }}{dt}\right)$$

(4)

Equation (4) contains three parts: $\frac{d{c}_{{\rho }_1}}{dt}=\frac{d{c}_0}{dt}$; $\frac{d{c}_{{\rho }_2}}{dt}=\frac{1}{A}·\left({\mathit{P}}^{\prime }·\frac{d{\mathit{C}}^{\prime }}{dt}\right)$; and $\frac{d{c}_{{\rho }_3}}{dt}=\frac{1}{A}·\left({\mathit{C}}^{\prime }·\frac{d{\mathit{P}}^{\prime }}{dt}\right)$. The combination of the second and third parts represents the trend in PPCE ($d{c}_1/dt$). The c_ρ value decreases under three conditions: $\frac{d{c}_0}{dt}<0$; ${\mathit{P}}^{\prime }·\frac{d{\mathit{C}}^{\prime }}{dt}<0$; or ${\mathit{C}}^{\prime }·\frac{d{\mathit{P}}^{\prime }}{dt}<0$. The first condition occurs if the mean PM_2.5 concentration decreased; the second condition occurs if densely populated urban areas experience a greater reduction of PM_2.5 concentration than rural areas; the third condition occurs if city planners relocate people from polluted areas to clean areas. These conditions are associated with the respective effects of the change in the mean PM_2.5 concentration, the spatial difference in PM_2.5 change and the demographic change.

4. Results

4.1. 15-Year Mean Population Exposure

Based on the 15-year averages of the PM_2.5 concentration and population density, Figure 3a,b show the deviation of the PM_2.5 concentration (${\mathit{C}}^{\prime }$) and the relative deviation of population density (${\mathit{P}}^{\prime }$), respectively, for different cities in the GBA region. The c’_i,j or p’_i,j values were positive (in red) in areas where the PM_2.5 concentration or population density was higher than the city’s averages. It is noted that the population density in urban areas could be several times larger than the city average.

Figure 4 shows the 15-year average of PPCE for different cities in the GBA region. Ideally, the PPCE values should be negative. However, all cities had a positive PPCE value because of a positive spatial coincidence between the PM_2.5 concentration and population density. Hong Kong and Zhuhai had the lowest PPCE (c₁ = 1.3 µg/m³). In contrast, the highest PPCE was found in Guangzhou (c₁ = 5.9 µg/m³).

The blue and red bars in Figure 5 show the 15-year averages of the mean PM_2.5 concentration (c₀) and the population-weighted mean PM_2.5 concentration (c_ρ), respectively, for different cities in the GBA region. The c₀ values were the lowest in Hong Kong (c₀ = 33.2 µg/m³) and Macau (c₀ = 33.9 µg/m³). In contrast, Foshan experienced the highest concentration of c₀ (57.1 µg/m³). The high mean PM_2.5 concentrations and the positive PPCE values resulted in high levels of population exposure. The highest c_ρ values were found in the central GBA region: Foshan (c_ρ = 59.8 µg/m³) and Guangzhou (c_ρ = 51.9 µg/m³). These values were much higher than the WHO air-quality standards for annual PM_2.5 concentration, including those for IT-1 (35 µg/m³), IT-2 (25 µg/m³), IT-3 (15 µg/m³) and Air Quality Guideline (AQG, 10 µg/m³). Although the lowest c_ρ value (34.5 µg/m³ in Hong Kong) did not exceed the WHO IT-1, it was still much higher than the WHO IT-2, IT-3 and AQG.

4.2. Decomposition of Exposure Variation

4.2.1. Effect of Changes in Mean PM_2.5 Concentration

Figure 6 shows the trends in the population-weighted mean PM_2.5 concentration resulting from changes in the mean PM_2.5 concentration ($\frac{d{c}_{{\rho }_1}}{dt}=\frac{d{c}_0}{dt}$) for different cities in the GBA region during the three most-recent FYP periods. During the 10th FYP period, all cities experienced an increase in c₀. The most substantial increase in c_ρ resulting from increasing c₀ was observed in Foshan, where the concentration increased by 2.77 µg/m³/year. During the 11th and 12th FYP periods, all cities experienced a decrease in c₀. The most substantial reduction of c_ρ resulting from decreasing c₀ was observed in Foshan, where the exposure level decreased by −2.25 µg/m³/year during the 11th FYP period and in Zhongshan, where the exposure level decreased by −3.71 µg/m³/year during the 12th FYP period.

4.2.2. The Effect of Spatial Differences in PM_2.5 Change

Figure 7a shows the trends in the deviation of the PM_2.5 concentration ($d{\mathit{C}}^{\prime }/dt$) for different cities during the three most-recent FYP periods. The areas with increasing ${c^{\prime }}_{i,j}$ values (in red) experienced less of a reduction of the PM_2.5 concentration than those with decreasing ${c^{\prime }}_{i,j}$ values (in blue). During the 10th FYP period, densely populated urban areas (${p^{\prime }}_{i,j}>0$) experienced greater increases in the PM_2.5 concentration than rural areas (${p^{\prime }}_{i,j}<0$) in most cities (8 out of 11). As a result, the spatial difference in the PM_2.5 change increased c_ρ in these cities. The left panel of Figure 7b shows the trends in the population-weighted mean PM_2.5 concentration resulting from the spatial difference in the PM_2.5 change [$\frac{d{c}_{{\rho }_2}}{dt}=\frac{1}{A}·\left({\mathit{P}}^{\prime }·\frac{d{\mathit{C}}^{\prime }}{dt}\right)$] for different cities during the 10th FYP period. The most substantial increase in c_ρ resulting from the spatial difference in the PM_2.5 change was observed in Guangzhou, where the exposure level increased by 1.03 µg/m³/year.

During the 11th and 12th FYP periods, densely populated urban areas (${p^{\prime }}_{i,j}>0$) experienced a greater decline in the PM_2.5 concentration than rural areas (${p^{\prime }}_{i,j}<0$) in most cities (7 and 10 cities, respectively). As a result, the spatial difference in the PM_2.5 change reduced c_ρ in these cities. The middle and right panels of Figure 7b show the trends in the population-weighted mean PM_2.5 concentration resulting from the spatial difference in PM_2.5 change for different cities during the 11th and 12th FYP periods, respectively. During the 11th FYP period, the effect reduced c_ρ most substantially in Guangzhou, where the exposure level decreased by −0.31 µg/m³/year. During the 12th FYP period, the effect reduced c_ρ most substantially in Macau, where the exposure level decreased by −0.58 µg/m³/year and in Shenzhen, where the exposure level decreased by −0.42 µg/m³/year.

4.2.3. The Effect of Demographic Change

Figure 8a plots trends in the relative deviation of population density ($d{\mathit{P}}^{\prime }/dt$) for different cities in the GBA region during the three most-recent FYP periods. The areas with increasing ${p^{\prime }}_{i,j}$ values (in red) experienced a greater percentage increase in the population density than those with decreasing ${p^{\prime }}_{i,j}$ values (in blue). It was found that urban areas in most cities experienced an increase in ${p^{\prime }}_{i,j}$ value. As these urban areas were typically more polluted than the rural areas, the increasing ${p^{\prime }}_{i,j}$ value in urban areas (typically with ${c^{\prime }}_{i,j}>0$) increased c_ρ in these cities. Figure 8b plots trends in the population-weighted mean PM_2.5 concentration resulting from the demographic changes [$\frac{d{c}_{{\rho }_3}}{dt}=\frac{1}{A}·\left({\mathit{C}}^{\prime }·\frac{d{\mathit{P}}^{\prime }}{dt}\right)$] for different cities in the GBA region during the three most-recent FYP periods. The effect increased c_ρ by as much as 0.18 µg/m³/year.

Dongguan was a special case, in which the demographic change decreased c_ρ [i.e., $\frac{1}{A}·\left({\mathit{C}}^{\prime }·\frac{d{\mathit{P}}^{\prime }}{dt}\right)<0$] by as much as -0.10 µg/m³/year during the three FYP periods studied. Figure 9a shows the 15-year trend in the relative deviation of the population density ($d{\mathit{P}}^{\prime }/dt$) from 2001 to 2015 in Dongguan. It was found that the southeastern region experienced a greater increase of the population density than the northwestern region. To evaluate this demographic change, gross population data was collected for different towns in Dongguan from the statistical yearbook (http://tjj.dg.gov.cn). The ratios of the gross population for 2001 and 2015 in different towns are shown in Figure 10. The results showed that towns southeast of the city adjacent to Shenzhen experienced a greater percentage increase in population than those in the northwest. This population change characteristic is consistent with that using the gridded data.

Based on the 15-year average of PM_2.5 concentration, Figure 9b shows the deviation of the PM_2.5 concentration (${\mathit{C}}^{\prime }$) in Dongguan. Higher PM_2.5 concentrations were observed in the northwestern region, whereas lower PM_2.5 concentrations were observed in the southeastern region. These results indicate that Dongguan rapidly developed residential areas in the cleaner regions that were adjacent to Shenzhen during the past decade. This demographic change pattern helped Dongguan reduce its population PM_2.5 exposure level.

4.2.4. Long-Term Variation in PPCE

The variation in PPCE was determined by the combined effect of spatial differences in PM_2.5 changes and demographic changes. Figure 11 shows the inter-annual variations in PPCE for different cities in the GBA region from 2001 to 2015. Most cities experienced an increase in PPCE during the 10th FYP period and a decrease in PPCE during the 12th FYP period. It is encouraging to find that the PPCE levels decreased to approximately half the previous maximum in most cities except in a few non-central cities such as Zhaoqing, Jiangmen and Huizhou. In 2015, the PPCE levels were still positive in all cities. The maximum PPCE levels in 2015 were found in Zhaoqing (3.7 µg/m³), Guangzhou (3.4 µg/m³) and Huizhou (3.2 µg/m³).

4.2.5. The Long-Term Variations in Population Exposure

The variations in the population-weighted mean PM_2.5 concentration (c_ρ) were determined by the combined effect of changes in the mean PM_2.5 concentration, spatial differences in PM_2.5 change and demographic changes. Figure 12 shows inter-annual variations in c_ρ for different cities in the GBA region from 2001 to 2015. In general, the c_ρ values started to decrease during the 11th and 12th FYP periods in all cities. However, in 2015, the c_ρ values remained high in all cities. Although the lowest c_ρ value (c_ρ = 25.9 µg/m³ in Hong Kong) was lower than that of WHO IT-1, it still exceeded those of WHO IT-2, IT-3 and AQG.

Figure 13 shows the trends in the population-weighted mean PM_2.5 concentration ($d{c}_{\rho }/dt$) resulting from the combined effect for different cities in the GBA region during the three most-recent FYP periods. During the 10th FYP period, c_ρ increased in all cities. The most substantial increases in c_ρ were observed in Foshan, where the exposure level increased by 3.27 µg/m³/year and in Guangzhou, where the exposure level increased by 3.28 µg/m³/year. During the 11th and 12th FYP periods, c_ρ declined in all cities. The reduction in c_ρ was the most substantial in Foshan, where the exposure level decreased by −2.53 µg/m³/year during the 11th FYP period, and in Zhongshan, where the exposure level decreased by −3.78 µg/m³/year during the 12th FYP period.

5. Discussion

Lin et al. [17] compared the spatial mean and population-weighted mean PM_2.5 concentrations for the cities in the GBA region from 2000 to 2014 and found that the population-weighted mean PM_2.5 concentrations were systematically higher than the spatial averages. The systematic differences were related to the positive spatial correlation between the PM_2.5 concentration and population density. For the entire GBA region, the population-weighted mean PM_2.5 concentration was higher than the spatial average by 5.9 µg/m³. This study further investigated the characteristics of this difference using the pollution-population-coincidence induced PM_2.5 exposure (PPCE). The PPCE is an important part of the population exposure level not only because its current level is significantly high, but also because its optimal level should be negative.

The mean PM_2.5 concentration levels in all cities of the GBA region of China decreased during the 11th and 12th FYP periods, due to the implementation of improved control measures. In addition to the beneficial effect of the decreasing mean PM_2.5 concentrations, it was found that spatial differences in the PM_2.5 changes also reduced the population exposure levels in most cities during the 11th and 12th FYP periods. In contrast, demographic changes increased the exposure levels in most cities because of their rapid urbanization. Dongguan was a special case that experienced the best city planning in terms of reducing population exposure. The rapid development of residential areas in cleaner regions adjacent to Shenzhen reduced its population exposure level during the past decade.

Although the population exposure levels greatly decreased for cities in the GBA region during the past decade, their levels remained high because of the high mean PM_2.5 concentrations and a positive PPCE. To better protect public health, more control efforts are required to reduce both the mean PM_2.5 concentration and PPCE. Currently, the reduction of the mean PM_2.5 concentration contributes significantly to the reduction of exposure levels in the GBA region. However, further reduction in the mean PM_2.5 concentration becomes more difficult as it drops to a lower level. The positive PPCE levels in all cities suggest that unfavorable city planning has located residential dwellings in polluted areas. To reduce PPCE, it is suggested that control efforts further target densely populated areas (e.g., controls of mobile and residential sources) and that city planners relocate some sources of pollution (e.g., emissions from industrial sources or power generation) away from these densely populated urban areas and locate more residential dwellings in cleaner areas.

It is difficult to find a truly objective criterion to classify a geographical area as urban or rural. All classification methods require a choice of threshold, which is subjective to a certain extent. The Organization for Economic Co-operation and Development (OECD) uses population size cutoffs (50,000 or 100,000 people) and population density cutoffs (1000 or 1500 persons/km² depending on the country) to define urban cores [33]. Following the OECD approach, Lin et al. [22] used the population density threshold of 1500 persons/km² for the classification of urban and rural areas, and investigated the differences in PM_2.5 variations between urban and rural areas for provinces in China. Their results showed that the urban areas in the GBA region experienced a greater increase in the PM_2.5 concentration than did rural areas during the 10th FYP period and a greater decrease in PM_2.5 concentration than did rural areas during the 11th and 12th FYP periods. These results are consistent with this study showing that the spatial difference in the PM_2.5 change increased population exposure levels in most cities during the 10th FYP period and reduced population exposure levels in most cities during the 11th and 12th FYP periods.

The uncertainties of exposure assessment might result from both the PM_2.5 concentration and population density. The evaluation of the satellite-derived PM_2.5 concentration against ground observations in the GBA region obtained a correlation coefficient of 0.86 and a mean percentage bias of <20% [20]. To reduce the bias of the LandScan-derived population density data, the gridded population data was adjusted using gross population data collected from the statistical yearbooks. The assessment in this study focused on residential areas with population densities of ≥10 persons/km². These residential areas contained 99.9% of the total population within the study region.

This study was focused on the decomposition of long-term variations in the population exposure to outdoor PM_2.5, which were measured using a satellite remote sensing technique. The exposure to indoor PM_2.5 would also negatively impact human health [34]. The indoor and outdoor exposure levels differ because of factors such as the spatial variability of the PM_2.5 concentration and outdoor-to-indoor infiltration. The measurement of indoor pollution exposure requires advanced and portable sensor devices [35,36]. Future studies could assess the indoor exposure and its health impact using the sensor measurements.

The climate in the GBA region is affected by seasonal monsoons. The prevailing winds are from the north in winter and from the south in summer. The monsoons affect the monthly variation in the PM_2.5 concentration in this region. Based on the satellite-derived monthly PM_2.5 data, Figure 14 shows the monthly variation of the population-weighted mean PM_2.5 concentration for different cities during the study period. All cities experienced lower population exposure levels in summer and higher population exposure levels in winter. The highest population exposure levels were typically found in January and December, with a level as high as >60 µg/m³ for cities in central GBA region (e.g., Foshan and Guangzhou). The population exposure levels actually doubled from summer to winter in several cities such as Zhongshan, Zhuhai and Jiangmen. Stronger regional transport of pollutants from inland, fewer mixing layers and less precipitation leads to higher levels of PM_2.5 exposure in winter than in summer [17].

The highly developed economy and industry in the GBA region offer a large number of employment opportunities. Therefore, many residents in the GBA region come from other provinces and work in this region. During the Chinese New Year, which is usually a one-week holiday in February, most of these people move back to their hometowns. The movement of these people changes their short-term exposure levels. The changes are determined by the differences in the PM_2.5 concentration level between the GBA region and their hometowns. Future studies could collect detailed information on the movement of populations over specific short-term periods and assess its effect on population exposure level.

The rapid economic development and urbanization in the GBA region has also caused complex environmental and health challenges from other air pollutants, such as sulfur dioxide (SO₂), nitrate oxides (NO_x) and ozone [37,38,39]. To better guide local environmental policies, it would be of great value to assess the population exposure to other air pollutants and the factors that influence such exposure. Further research could delineate exposure decompositions for other air pollutants in the study region.

6. Conclusions

This study used a decomposition method to assess the impact of PM_2.5 variations and demographic changes on the exposure variations for cities in the GBA region of China during the three most-recent FYP periods. This study took advantage of high-resolution satellite observations to characterize the long-term variation in human exposure to PM_2.5. The results showed that the 15-year average PPCE levels were positive in all cities, suggesting that unfavorable city planning had placed people in the polluted areas. The analyses of the spatial differences in PM_2.5 changes showed that urban areas experienced a greater decrease in the PM_2.5 concentration than rural areas did in most cities during the 11th (2006–2010) and 12th (2011–2015) FYP periods. These spatial differences in PM_2.5 changes reduced their PPCE levels and thus reduced the exposure levels. The migration of population resulting from rapid urbanization, however, increased the PPCE and exposure levels in most cities during the three FYP periods considered. Dongguan was the only city in which the demographic change reduced the exposure level because of its rapid development of residential areas in cleaner regions adjacent to Shenzhen. The exposure levels in all cities remained high because of the high mean PM_2.5 concentrations and their positive PPCE. To better protect public health, control efforts should target densely populated areas and city planning should place more residential dwellings in cleaner areas.