Spatiotemporal Changes in PM2.5 and Their Relationships with Land-Use and People in Hangzhou

Increases in the extent and level of air pollution in Chinese cities have become a major concern of the public and burden on the government. While ample literature has focused on the status, changes and causes of air pollution (particularly on PM2.5 and PM10), significantly less is known on their effects on people. In this study we used Hangzhou, China, as our testbed to assess the direct impact of PM2.5 on youth populations that are more vulnerable to pollution. We used the ground monitoring data of air quality and Aerosol optical thickness (AOT) product from the Moderate Resolution Imaging Spectroradiometer (MODIS) for the spatiotemporal changes of PM2.5 by season in 2015. We further explored these distributions with land cover, population density and schools (kindergarten, primary school and middle school) to explore the potential impacts in seeking potential mitigation solutions. We found that the seasonal variation of PM2.5 concentration was winter > spring > autumn > summer. In Hangzhou, the percentage of land area exposed to PM2.5 > 50 µg m−3 accounted for 59.86% in winter, 56.62% in spring, 40.44% in autumn and 0% in summer, whereas these figures for PM2.5 of <35 µg m−3 were 70.01%, 5.28%, 5.17%, 4.16% in summer, winter, autumn and spring, respectively. As for land cover, forest experienced PM2.5 of 35–50 µg m−3 (i.e., lower than those of other cover types), likely due to the potential filtering and absorption function of the forests. More importantly, a quantitative index based on population-weighted exposure level (pwel) indicated that only 9.06% of the population lived in areas that met the national air quality standards. Only 1.66% (14,055) of infants and juveniles lived in areas with PM2.5 of <35 µg m−3. Considering the legacy effects of PM2.5 over the long-term, we highly recommend improving the monitoring systems for both air quality and people (i.e., their health conditions), with special attention paid to infants and juveniles.


Introduction
In haze days, PM 2.5 (the particulate matter with aerodynamic diameter ≤2.5 µm) and its particle concentration account for 56.7-75.4% of the total suspended particles and >80%-90% of PM 10 (the particles measuring ≤10 µm in aerodynamic diameter) [1]. The high rate of exposure of the youth Our primary study objectives are to: (1) understand the spatiotemporal distribution of PM2.5 concentrations in Hangzhou using data from 2015; (2) examine the empirical relationships between the spatiotemporal changes of PM2.5 and land cover; (3) analyze the populations exposed to different levels of PM2.5 concentrations; and (4) analyze the distribution of students and schools (e.g., kindergarten, primary school and middle school students) living in areas of different PM2.5 concentrations. We aim at reducing the potential health threats of long-term exposure of infants and juveniles to PM2.5.

Study Area
Hangzhou city, the capital city of the Zhejiang Province, is located in southeast China. This study covers the main urban area of Hangzhou, including eight districts: Shangcheng, Xiacheng, Jianggan, Gongshu, Xihu, Binjiang, Yuhang and Xiaoshan (Figure 1a). The study area was 3376 km 2 and the population density was 2111.96 per km 2 in 2015 (http://www.hzfc.gov.cn/web). It experiences a humid subtropical climate with four distinct seasons and is characterized by long, hot, humid summers and chilly, cloudy winters. The average annual precipitation is 1,438 mm; rainfall is abundant during summer and relatively low during winter [35].

Data Sources
The MODIS has 36 spectral channels ranging from visible to infrared, providing an effective means of detecting global aerosol properties. In this study, the MOD04-3K AOT product at 3 km

Data Sources
The MODIS has 36 spectral channels ranging from visible to infrared, providing an effective means of detecting global aerosol properties. In this study, the MOD04-3K AOT product at 3 km resolution for 2015 was acquired from the Level 1 and Atmospheric Archive and Distribution System (LAADS) (https://earthdata.nasa.gov/about/daacs/daac-laads). Geometric correction was applied to AOT images. In addition, meteorological data (e.g., wind, relative humidity from NCEP (http://dss.ucar.edu/)) were also used for calibrating AOT. Real-time hourly monitoring data of PM 2.5 density from 10 ground stations in Hangzhou were collected from the National Environmental Monitoring Centre from January 2015 to December 2015 and were converted to daily averages to match with MODIS AOT. The land cover map was developed based on high-resolution (<1 m) aerial photos in 2015 (Figure 1b). The spatial distribution of kindergartens, primary schools, middle schools and their numbers of enrollment were collected from the Hangzhou Education Bureau (Figure 1c). The district-level population data was collected from the statistical yearbook of Hangzhou (http://tjj.hangzhou.gov.cn/tjnj/nj2017/index.htm). To create spatially continuous population distributions, we assume that the population density is highly correlated with building density. The district-level population was reallocated spatially on a standard grid map (100 m × 100 m); the proportion of build-up area in a grid was used as a weighting factor to calculate the population of a grid by ensuring that the total population of each district remains the same (Figure 1d). Figure 2 shows the process of modeling the relationship between AOT and PM 2.5 . The three consequent steps are: AOT retrieval and calibration, match of ground monitoring data with AOT and regression modelling. The key for calibrating AOT data is correcting aerosol altitude and water vapor density. The density of aerosol particles decreases with increasing altitude because of the gravity impact. The relationship between AOT and the aerosol extinction coefficient was expressed as:

Spatial Modeling of PM 2.5 Distribution
where τ a (λ) stands for the AOT value; k a,0 (λ) is the near-ground horizontal extinction coefficient, which is affected by the atmospheric water vapor content; H A stands for aerosol scaling height. Aerosol scaling height is a key parameter that can be approximated by mixed-layer height. The mixed-layer height is closely related to the aerosol stability and can be calculated following the protocols of the State Bureau of Technical Supervision and the State Environmental Protection Administration [50] (Equation (2)).
where L stands for the mixed-layer height (m); u 10 is the wind speed at the altitude of 10 m (m s −1 ) and its maximum value is 6 m s −1 ; Ω stands for the rotational angular velocity of the earth and is assigned a value of 7.29 × 10 −5 rad s −1 ; ϕ stands for geodetic latitude; S is related to the aerosol stability referring to Pasquill stability classes (see details in Reference [50]) and its corresponding values in Hanzhou can be found in Table 1. After the corrections, the aerosol extinction coefficient can be obtained. Water vapor correction is further applied to retrieve the "dry" aerosol extinction coefficient as: where E dry stands for the "dry" aerosol extinction coefficient; RH represents the relative humidity (%).
To establish a relationship between the ground measurements and the AOT, further statistics were applied to ensure the spatiotemporal consistency of the ground measurements with the remote sensing images. The precision of temporal match should be within ±1 h between the monitoring data and the satellite passing time. The mean value in 3 × 3 pixel cells of AOT is used for matching with the value of the monitoring location. Linear regression is applied to explore the correlation between the "dry" aerosol extinction coefficient and PM 2.5 density (Equation (4)). Independent regression models were established by season because the climatic differences may result in different aerosol distributions.
To establish a relationship between the ground measurements and the AOT, further statistics were applied to ensure the spatiotemporal consistency of the ground measurements with the remote sensing images. The precision of temporal match should be within ±1 h between the monitoring data and the satellite passing time. The mean value in 3 × 3 pixel cells of AOT is used for matching with the value of the monitoring location. Linear regression is applied to explore the correlation between the "dry" aerosol extinction coefficient and PM2.5 density (Equation (4)). Independent regression models were established by season because the climatic differences may result in different aerosol distributions.

Spatial Correlation between PM2.5 Distribution and Land Use Types
Recent studies have shown that the land cover (i.e., traffic roads) could be spatially correlated with the density of PM2.5 [17,[51][52][53]. Incorporating this information would help to increase the model accuracy. In this study, we quantified the empirical relationship of PM2.5 density with different land cover types by season. Specifically, the spatial PM2.5 concentration was divided into three levels: <35 µg m −3 (non-polluted), 35-50 µg m −3 (intermediate) and >50 µg m −3 (heavy) and their proportions for each land cover type were calculated. Such results can be useful for understanding the landscape contribution and further improving PM2.5 predictions by including land use regression models.

Potential Impact of PM2.5 Distribution in Hangzhou
In order to estimate the impacts of PM2.5, several demographic data were used for calculating the proportion of the population affected by different levels of PM2.5. A quantitative index-population-weighted exposure level (pwel)-was calculated to identify the areas with potential high risk of population exposure to atmospheric particulates:

Spatial Correlation between PM 2.5 Distribution and Land Use Types
Recent studies have shown that the land cover (i.e., traffic roads) could be spatially correlated with the density of PM 2.5 [17,[51][52][53]. Incorporating this information would help to increase the model accuracy. In this study, we quantified the empirical relationship of PM 2.5 density with different land cover types by season. Specifically, the spatial PM 2.5 concentration was divided into three levels: <35 µg m −3 (non-polluted), 35-50 µg m −3 (intermediate) and >50 µg m −3 (heavy) and their proportions for each land cover type were calculated. Such results can be useful for understanding the landscape contribution and further improving PM 2.5 predictions by including land use regression models.

Potential Impact of PM 2.5 Distribution in Hangzhou
In order to estimate the impacts of PM 2.5 , several demographic data were used for calculating the proportion of the population affected by different levels of PM 2.5 . A quantitative index-population-weighted exposure level (pwel)-was calculated to identify the areas with potential high risk of population exposure to atmospheric particulates: where P i stands for the population in grid i and P total stands for the total population in the research area; C i is the simulated PM 2.5 density in grid i. Additionally, the number of kindergartens, primary schools and middle schools located in different PM 2.5 concentration zones were calculated to show the affected "key" population (i.e., infants and juveniles) in Hangzhou.

Relationship between AOT and the PM 2.5 Concentration
After AOT inversion and calibration, the linear regression models were successfully established between AOT and the PM 2.5 concentration for the four seasons ( Figure 3). Model correlations varied by season, with the correlation coefficient of determination (R 2 ) varying between 0.347 and 0.740. The accuracy order of the model was determined as summer > spring > autumn > winter. This seasonal difference was affected by the height of the atmospheric mixing layer (i.e., low in autumn and winter when the diffusion of air particle pollutants was low). During autumn and winter, cold waves were frequent and the resulting weather conditions led to increasing atmospheric pollutants and greater spatiotemporal variability. The model fitting accuracy decreased. Nevertheless, the appeared acceptable for all four seasons [54].
where stands for the population in grid i and stands for the total population in the research area; is the simulated PM2.5 density in grid i. Additionally, the number of kindergartens, primary schools and middle schools located in different PM2.5 concentration zones were calculated to show the affected "key" population (i.e., infants and juveniles) in Hangzhou.

Relationship between AOT and the PM2.5 Concentration
After AOT inversion and calibration, the linear regression models were successfully established between AOT and the PM2.5 concentration for the four seasons ( Figure 3). Model correlations varied by season, with the correlation coefficient of determination (R 2 ) varying between 0.347 and 0.740. The accuracy order of the model was determined as summer > spring > autumn > winter. This seasonal difference was affected by the height of the atmospheric mixing layer (i.e., low in autumn and winter when the diffusion of air particle pollutants was low). During autumn and winter, cold waves were frequent and the resulting weather conditions led to increasing atmospheric pollutants and greater spatiotemporal variability. The model fitting accuracy decreased. Nevertheless, the appeared acceptable for all four seasons [54].

The Spatiotemporal Distribution in PM 2.5
The annual average of PM 2.5 concentrations was 43 µg m −3 (std = 5.28), indicating that Hangzhou's regional air quality was better than half of China's cities (53.0 µg m −3 ) in 2015 [13,27]. Following China's peak PM value in 2013, the PM 2.5 concentration showed a significant decreasing trend. In Hangzhou, the PM 2.5 concentration was 52.2 µg m −3 in 2013 [42]. However, this level is far from China's National Air Quality Standard for Grade II limit of 35 µg m −3 [22]. Among the 366 cities in China, over 80% did not reach the standard of Grade II [55]. As for the spatial distribution of the annual average, PM 2.5 was mainly concentrated in Gongchu, Shangcheng, Xiacheng and parts of Xihu, Yuhang and Xiaoshan. The mean PM 2.  (Figure 4a-d). The lowest values were found in the northwest mountainous areas (Figure 4). Among the seasons, the concentration was winter > spring > autumn > summer. The seasonal characteristics of PM 2.5 concentration were consistent with the ground observations in Hangzhou. During winter, air pollution remained as a serious issue that severely affected people. The administrative department continued to struggle to find efficient ways to reduce the pollution level [56][57][58][59]. During winter, air pollution remained as a serious issue that severely affected people. The administrative department continued to struggle to find efficient ways to reduce the pollution level [56][57][58][59].  (Table 2). Additionally, there appeared multiple "hot spots" in all four seasons ( Figure 4).   (Table 2). Additionally, there appeared multiple "hot spots" in all four seasons ( Figure 4). The histogram statistics of PM 2.5 for the four seasons were also calculated. The distributions of PM 2.5 in spring and autumn were relatively narrow, presenting a typical single peak distribution (Figure 5a,c). However, winter and summer showed dispersed values for PM 2.5 ; these values were especially complex for winter, where there were multiple peaks (Figure 5b,d). This complexity was likely due to winter's mixed atmospheric layer height being low and not conductive to the diffusion of atmospheric particle pollutants. In addition, cold waves frequently changed the weather conditions, which subsequently led to an increase in the spatiotemporal variability of atmospheric pollution and resulted in the regularity of distribution being less significant than that of spring and autumn [15,16]. Even in northern China, the distribution of PM 2.5 was complex and contained multiple peaks during the winter [57,60]. The histogram statistics of PM2.5 for the four seasons were also calculated. The distributions of PM2.5 in spring and autumn were relatively narrow, presenting a typical single peak distribution (Figure 5a,c). However, winter and summer showed dispersed values for PM2.5; these values were especially complex for winter, where there were multiple peaks (Figure 5b,d). This complexity was likely due to winter's mixed atmospheric layer height being low and not conductive to the diffusion of atmospheric particle pollutants. In addition, cold waves frequently changed the weather conditions, which subsequently led to an increase in the spatiotemporal variability of atmospheric pollution and resulted in the regularity of distribution being less significant than that of spring and autumn [15,16]. Even in northern China, the distribution of PM2.5 was complex and contained multiple peaks during the winter [57,60].

Correlation Analysis between Land Use and the Spatial Distribution of PM 2.5 Concentration
Of the seven land cover types in the study area, the landscape was composed of 26.86% (905.84 km 2 ) built-up area, 12.87% (434.22 km 2 ) water, 4.94% (166.23 km 2 ) grassland, 26.31% (887.42 km 2 ) forest, 11.95% (403.07 km 2 ), cultivated land, 5.38% (181.53 km 2 ), roads, 11.67% (393.91 km 2 ) and orchard land (Figure 1b). We delineated portions of the seven land cover types by three classes of PM 2.5 concentrations (Table 3). In spring, 4.16% of the land surface in Hangzhou experienced PM 2.5 of <35 µg m −3 . Meanwhile, 39.23% and 56.62% of the land surface was experiencing air conditions of PM 2.5 between 35-50 µg m −3 and PM 2.5 of >50 µg m −3 , respectively. In summer, the air quality was better, as 70.01% of the land surface was exposed to PM 2.5 of <35 µg m −3 and no area accounted for PM 2.5 of >50 µg m −3 . In autumn, only 5.17% of the land surface experienced PM 2.5 of <35 µg m −3 and 40.44% of the land surface was exposed to PM 2.5 of >50 µg m −3 . The air pollution in winter was more severe, with 59.86% of the land surface under PM 2.5 of >50 µg m −3 . Overall, the PM 2.5 concentrations in the winter and spring seasons were higher than those of the other seasons and showed multiple peaks ( Figure 5). Hangzhou's meteorological conditions in these seasons were not conducive to the emission of air pollutants [37,44,49].
In regard to the land cover types, the forests in Hangzhou were distributed mainly around the area where the PM 2.5 concentration was 35-50 µg m −3 . Within PM 2.5 of <35 µg m −3 , forests occupied the highest proportion of the land surface (Table 3). This may due to the filtering/abortion function as particulate air pollutants move through the forest landscape [17,61,62]. Janhäll [27] has advocated that increasing the vegetation barriers should help absorb and filter the particulate air pollution. For PM 2.5 of >50 µg m −3 , built-up area showed the largest area proportion in spring, autumn and winter (Table 3). This result again highlighted the trend of 'more human activities, more air pollution sources' [26]. Based on the annual mean PM 2.5 concentration and its spatial distribution in Hangzhou, the population-weighted exposure level (pwel) showed the risk level of populations exposed to different concentrations of PM 2.5 . We found 249.18 Pop km −2 (±746.53) of the population live in PM 2.5 of <35 µg m −3 , covering 266.29 km 2 ; for PM 2.5 of 35-50 µg m −3 , the population density was 1521.60 Pop km −2 (±3584.08) in 1483.99 km 2 ; for PM 2.5 of >50 µg m −3 , the population density was 1582.66 Pop km −2 (±3124.79) in 1188.18 km 2 (Figure 1d). Clearly, most people reside in high PM 2.5 concentration areas. On the other hand, gaseous and particulate pollutants were also exposed due to human activity. Pollution from human activities has severely contributed to the health impacts on people over a long period of time [63]. Considering infants and juveniles attending school, younger individuals are less resistant to disease and daily exposure to high PM 2.5 concentrations can cause both current and future health problems.
In the study area, the number of kindergartens was 623, with 239,459 infants. The number of primary schools was 265, facilitating 389,260 students. The number of middle schools was 123, with 217,959 students. By the different PM 2.5 concentration levels, 294 kindergarten students were under  (Table 4). These results indicated that at each of the aforementioned educational levels, only 1.66% (14,055) of infants and juveniles lived in an environment that met China's National Air Quality Standard for Grade II. This number fell far below the national mean level [13,26,64]. In addition, 41 (Table 4). Although we only generated statistics for the number of infants and juveniles, the families and schools near the residential areas experienced a similar atmospheric environment. Although children's disease attributed to PM 2.5 exposure has not been well studied, other studies have showed that China's leading mortality causes (e.g., stroke, IHD, LC and COPD) could be attributed to PM 2.5 exposure to some extent [35,64]. Considering the legacy effects on human health from long-term PM 2.5 exposure, it is necessary to track the health status of infants and juveniles from birth until they have entered into adulthood. By doing so, we might reduce the harms of PM 2.5 on people.

Conclusions
We used a combination of the dispersed monitoring ground data, land cover data and MODIS remote-sensing AOT to model the distribution of PM 2.5 concentrations and to analyze its effects on residents, with a particular focus on infants and juveniles attending schools in Hangzhou in 2015. First, the seasonal variation in PM 2.5 concentration was winter > spring > autumn > summer. For the eight main urban districts, the highest PM 2.5 concentrations in spring, summer and autumn were located in the Xiacheng district and the lowest value was located in the Yuhang district. However, in winter, the highest value was found in the Gongshu district and the lowest value in the Yuhang district. In addition, the lowest value for all four seasons appeared in the Yuhang district, where there is abundant vegetation and a low population density. Secondly, for the different land cover types, we found that in winter and spring, 59.86% and 56.62% of the land area was exposed to PM 2.5 concentrations of >50 µg m −3 , while the built-up area occupied 20.65% in winter and 19.72% in spring. In autumn, 54.38% of the land area was exposed PM 2.5 35-50 µg m −3 and forest occupied the largest proportion (15.49%). In the summer, the air particulate content was the lowest, with 70.01% of the land surface area exposed to PM 2.5 of <35 µg m −3 and the forests accounted for 23.39%. Finally, based on the spatial distribution of different classes of PM 2.5 concentrations, only 9.06% of the population lived in an environment that met the national air quality standards. For infants and juveniles, only 1.66% (14,055) lived in areas of PM 2.5 of <35 µg m −3 ; 56.49% of infants and juveniles (478,257) lived in an intermediately polluted environment (PM 2.5 of 35-50 µg m −3 ) and 41.97% (355,333) lived in a heavily polluted environment (PM 2.5 > 50 µg m −3 ) in Hanzhou. We estimated site-specific annual PM 2.5 concentrations. Most infants and juveniles currently live in an atmospherically polluted environment not only in Hangzhou but also in most cities in China. We believe that air quality modelling and cost-benefit analyses of emission reduction scenarios and corresponding health benefits play key roles in meeting the site-specific annual PM 2.5 concentration goals. Actions must be taken and attention must be paid in order to safeguard the future of the country.