2.1.

Study Area

China is located in the east of Asia and the west of the Pacific; its climate is significantly affected by both continent and ocean. As a result of its complex terrain, both temperature and precipitation exhibit a complex spatial pattern. Similarly, land-use type in China has a large variability. For instance, sandy deserts and Gobi are mainly located in northwestern China, arable land is the dominant land type in the eastern plain, grassland scatters over the northern part of Inner Mongolia, and forest land is mainly in the northeastern and southwestern China. In addition, the economic development in China is also regionally imbalanced. Eastern areas are more economically developed than western China. Consequently, population density is higher in Yangtze River Delta, Pearl River Delta, and the Bohai Rim Economic Circle, and lower in western China.

With the rapid economic development, air pollution has become one of the top environmental concerns in China.²¹ The growing demand for energy and the increasing number of motor vehicles and fast industrialization have led to a serious deterioration in air quality and consequent serious negative effects on human health and ecosystems. In some parts of China, due to the overlaying of different kinds of pollutants, air pollution is more serious over cities and industrial zones. Environmental protection in China is facing huge challenges and becoming more urgent.

2.2.

Data Collection and Processing

In this research, NASA MODIS C5.1 daily Aerosol Level-2 Product MOD04, over the time period from 2010 to 2014, was used to estimate ${\mathrm{PM}}_{10}$ concentration. The data are produced at a spatial resolution of $10\times 10{\mathrm{km}}^2$.²² We first extracted the valid AOD [aerosol optical thickness at 0.55 micron for both ocean (best) and land (corrected) with best quality data ($\text{quality}\text{flag}=3$)] from MOD 04 daily Aerosol Product; the valid range of this data is from $-0.05$ to 5.0. The AOD was derived from the Dark Target algorithm.²³ Due to the limitation of the Dark Target algorithm,²⁴ some data over bright areas and cloudy areas were missing. To deal with this problem, we calculated the missing AOD by integral averaging the value of the day before and the day after. Then we obtained the monthly average AOD and annual average AOD based on the daily Aerosol Product.

After that, we collected ${\mathrm{PM}}_{10}$ annual average concentration measured over 228 Chinese cities from 2010 to 2014. The annual average ${\mathrm{PM}}_{10}$ concentration in each city is the mean value of all the monitoring stations in both urban and suburban areas in this city. The average ${\mathrm{PM}}_{10}$ concentration measured in these 228 Chinese cities can be classified into two groups, one group was used for establishing the exponential model, and the other group was used for estimating validations of satellite-derived ${\mathrm{PM}}_{10}$ concentration. In this paper, 176 were used to establish the empirical model, and the others were used to evaluate the validation of the model. Monthly average temperature and actual vapor pressure data from 2010 to 2014 were acquired from 194 international exchange meteorological stations in China. And monthly average aerosol scale height was obtained from 98 solar radiation observation stations in China over the time period from 2010 to 2014.

2.3.

Methodology

2.3.1.

Relationship between aerosol optical depth and ground ${\mathsf{PM}}_{\mathsf{10}}$ mass concentration

Estimating ground aerosol mass concentrations from aerosol optical depth

Since the ground ${\mathrm{PM}}_{10}$ concentration is defined as the surface concentration of the particles, while the AOD retrieved from satellite observations corresponds to total column concentration of particles under ambient RH, the direct correlation between satellite-based AOD and the ground concentration of ${\mathrm{PM}}_{10}$ is relatively low and is influenced by humidity. Due to the hydroscopic growth of aerosols, to accurately estimate the ground ${\mathrm{PM}}_{10}$ concentration from satellite-retrieved AOD, RH has to be taken into account. According to the empirical relationship derived by White and Roberts²⁵ and Li et al.,⁷ the aerosol extinction coefficient (${\beta }_0$) is defined as follows:

Eq. (1)

$${\beta }_0=\rho \times f(\mathrm{RH})=\rho \times \frac{1}{1-\mathrm{RH}},$$

where ${\beta }_0$ is the aerosol extinction coefficient, $\rho$ is the mass concentration and RH is the relative humidity.

In addition, the aerosol extinction coefficient is also a function of height. The variation of the aerosol extinction coefficient (${\beta }_z$) with the height can be described as an exponential function:¹⁰

Eq. (2)

$${\beta }_z={\beta }_0\times e^{-\frac{z}{\mathrm{ASH}}},$$

where ASH is the aerosol scaling height and $z$ is the height.

Since the AOD is an integral of the aerosol extinction coefficient in the total column,

Eq. (3)

$$\mathrm{AOD}=\underset{0}{\overset{\infty }{\int }}{\beta }_z\mathrm{d}z=\underset{0}{\overset{\infty }{\int }}{\beta }_0\times e^{-\frac{z}{\mathrm{ASH}}}\mathrm{d}z=\mathrm{ASH}\times {\beta }_0.$$

Therefore, the ground aerosol mass concentration (AMC) can finally be written as

Eq. (4)

$$\mathrm{AMC}=\frac{{\beta }_0}{f(\mathrm{RH})}=\frac{\mathrm{AOD}}{\mathrm{ASH}}\times (1-\mathrm{RH}).$$

The spatial distribution of the monthly mean ASH was obtained from 98 solar radiation observation stations in China over the time period from 2010 to 2014 by using Kriging interpolation. RH was calculated with the modified Magnus equation.²⁶

Eq. (5)

$$\mathrm{RH}=\frac{e_a}{e_0(T)}\times 100,$$

Eq. (6)

$$e_0(T)=610.78\times e^{\frac{17.269(T-273.16)}{T-35.86}},$$

where $e_a$ is the actual vapor pressure and $e_0(T)$ is the saturation vapor pressure at air temperature $T$. It is obvious that knowledge of the spatial distribution of both $T$ and $e_a$ are required to generate the RH spatial distribution over China. To this end, $e_a$ was obtained through interpolation of the monthly mean vapor pressure data in China by using Kriging interpolation. From Eq. (6), it is seen that to obtain the spatial distribution of the saturated vapor pressure, air temperature has to be known. First of all, monthly mean air temperature data from 183 meteorological stations in China are converted to air temperature at sea level according to the following definition:

Eq. (7)

$$T_2=T_1+\frac{0.65h}{100},$$

where $T_1$ is the measured air temperature at meteorological stations, $T_2$ is the temperature at virtual sea level, and $h$ is the altitude, which is calculated with a digital elevations model. Since the variation of air temperature at the same level is considered as continuous, the air temperature at sea level was calculated through Kriging interpolation. Second, the actual air temperatures were obtained by converting the interpolated air temperature at virtual sea level back to actual elevation with Eq. (7). Finally, by combining Eqs. (4), (5), and (6), the monthly mean surface aerosol mass concentration was estimated.

Relationship between ground ${\mathsf{PM}}_{\mathsf{10}}$ concentration and ground AMC

${\mathrm{PM}}_{10}$ is microscopic solid or liquid matter suspended in the Earth’s atmosphere, with an aerodynamic diameter $<10\mu \mathrm{m}$. AMC includes aerosols with various sizes, while ${\mathrm{PM}}_{10}$ accounts for only the aerosols with sizes $<10\mu \mathrm{m}$. To obtain the ${\mathrm{PM}}_{10}$ concentration from AMC, we first selected the ${\mathrm{PM}}_{10}$ annual average concentration observed over 176 Chinese cities from 2010 to 2014. Note that the ${\mathrm{PM}}_{10}$ annual concentration for each city is the mean value of all the monitoring stations in both urban and suburban areas. Then we compared the annual average AMC with the annual average surface monitors located in the grid of the remote sensing data and analyzed the regression relation between them. It is found that there is an exponential relationship between annual average ground AMC from 2010 to 2014 and corresponding ground ${\mathrm{PM}}_{10}$ mass concentration over these cities. The derived relationship is given and shown in Fig. 1.

Fig. 1

The relationship between aerosol mass concentration and ground ${\mathrm{PM}}_{10}$ mass concentration.

3.1.

Spatial Distribution of ${\mathsf{PM}}_{\mathsf{10}}$ in China

The annual average ${\mathrm{PM}}_{10}$ spatial distribution from 2010 to 2014 derived from the relationship between ground ${\mathrm{PM}}_{10}$ concentration and satellite-derived AOD is shown in Fig. 2. The spatial resolution of the map is $10\times 10{\mathrm{km}}^2$. It is seen that the highest annual average ${\mathrm{PM}}_{10}$ concentration is mainly concentrated in Northern China, the Sichuan Basin, and Taklimakan Desert regions, where annual average ${\mathrm{PM}}_{10}$ concentration is $>100\mu \mathrm{g}/{\mathrm{m}}^3$. The northwestern, the middle and lower reaches of the Yangtze River, and Inner Mongolia, Shaanxi, Shanxi, and some other provinces in northern China are as high as $80\mu \mathrm{g}/{\mathrm{m}}^3$. In the northeastern area, Liaoning and Jilin, the annual average ${\mathrm{PM}}_{10}$ concentration is $\sim 70\mu \mathrm{g}/{\mathrm{m}}^3$. In southwestern and southern China, ${\mathrm{PM}}_{10}$ concentration is $\sim 60\mu \mathrm{g}/{\mathrm{m}}^3$. Over the Tibetan Plateau, Fujian and Heilongjiang province, ${\mathrm{PM}}_{10}$ concentration has the lowest value. In general, concentration is the highest in the northern zone, followed by the central region, and the concentration is lowest in southern China and the Tibetan Plateau.

Fig. 2

Annual average ${\mathrm{PM}}_{10}$ in China from 2010 to 2014.

3.2.

Spatial Distribution of Population in China

Figure 3 shows the annual average population density over China from 2010 to 2014 in a resolution of 10 km. In general, the permanent population of China is mainly concentrated in the eastern coastal area. Population density in northern China, southern China, and the middle and lower reaches of Yangtze River is much larger than that in western China. Population density in the plain and basin areas is overall higher, while it is much lower in mountainous and plateau regions. Areas along rivers and the coast are more densely populated.

Fig. 3

Annual average population density in China from 2010 to 2014.

3.3.

Population-Weighted ${\mathsf{PM}}_{\mathsf{10}}$ Exposure Level

To accurately assess the impact of ${\mathrm{PM}}_{10}$ exposure on human health, the distribution of ${\mathrm{PM}}_{10}$ has to be combined with the distribution of the population. Therefore, according to the population-weighted ${\mathrm{PM}}_{10}$ exposure model, given in Eq. (9), population-weighted ${\mathrm{PM}}_{10}$ exposure levels are calculated and are shown in Fig. 4. It is clearly seen that, in both northern China and the Sichuan Basin regions, which are highly populated and industrialized and have a higher annual average ${\mathrm{PM}}_{10}$ concentration from 2010 to 2014, the population-weighted ${\mathrm{PM}}_{10}$ exposure level is even higher; it can reach up to $100\mu \mathrm{g}/{\mathrm{m}}^3$. In the middle and lower reaches of the Yangtze River, from Wuhan to Nanjing, the population-weighted ${\mathrm{PM}}_{10}$ exposure level is as high as $\sim 85\mu \mathrm{g}/{\mathrm{m}}^3$. In northeastern, southern, and southwestern China, population-weighted ${\mathrm{PM}}_{10}$ exposure level is $\sim 65\mu \mathrm{g}/{\mathrm{m}}^3$, much lower than that in the northern China and the Yangtze River region. However, most western and northwestern regions have a very low population-weighted exposure level as a result of both less population and underdeveloped industry.

Fig. 4

Annual average population-weighted ${\mathrm{PM}}_{10}$ exposure level from 2010 to 2014.

3.4.

${\mathsf{PM}}_{\mathsf{10}}$ Human Health Losses in China

Finally, according to human health effects of the ${\mathrm{PM}}_{10}$ model, as shown in Eqs. (12) and (13), the human health losses of each health endpoint resulting from ${\mathrm{PM}}_{10}$ exposure are evaluated. It is found that, from 2010 to 2014, exposure to ${\mathrm{PM}}_{10}$ air pollution has caused a negative effect on health for $\sim 6.9$ million people in China every year. Among them, there are 0.9 million cases of death and 6.0 million cases of acute health diseases. More specifically, $\sim 3.5$ million people suffer from acute respiratory illness, and 2.5 million people suffer from acute cardiovascular diseases due to the exposure to ${\mathrm{PM}}_{10}$.

4.1.

Validation of Satellite-Derived PM₁₀

The uncertainties of the satellite-derived ${\mathrm{PM}}_{10}$ lead to the uncertainties of the human health impact resulting from ${\mathrm{PM}}_{10}$ exposure. To estimate the validation of the satellite-derived ${\mathrm{PM}}_{10}$ concentration, the annual average ${\mathrm{PM}}_{10}$ concentration measured in 228 Chinese cities was first analyzed using cluster analysis. Then these ground measured data were divided into four categories. They represent four ${\mathrm{PM}}_{10}$ pollution levels, and the ${\mathrm{PM}}_{10}$ concentration is significantly different in each category. To estimate the overall validation of the satellite-derived ${\mathrm{PM}}_{10}$, we first selected 52 of them as the samples according to the method of stratified sampling.⁴¹ Note that 52 is the minimum number of samples when these cities were divided into four categories. Then we compared the annual average of ground-based ${\mathrm{PM}}_{10}$ concentration with the annual average of satellite-derived ${\mathrm{PM}}_{10}$ concentration of these 52 samples from 2010 to 2014, as shown in Fig. 5. A linear relationship exists between the satellite-derived ${\mathrm{PM}}_{10}$ and ground-based ${\mathrm{PM}}_{10}$; the correlation coefficient is as high as 0.83, root mean square error is 19.27, relative standard deviation is 12.66%, and mean absolute percentage error (MAPE) is only 7.70%. High correlation and low MAPE indicates the applicability and reliability of ${\mathrm{PM}}_{10}$ concentration derived from MODIS data. The mean bias of the concentration of ${\mathrm{PM}}_{10}$ is $<10\mu \mathrm{g}/{\mathrm{m}}^3$ based on the accuracy of MODIS AOD retrievals over land. The corresponding transferred bias for the relative risks from exposure to ${\mathrm{PM}}_{10}$ is $<0.01$ according to the ${\mathrm{PM}}_{10}$ human health impact model and the bias of negative cases from ${\mathrm{PM}}_{10}$ exposure is $<0.2$ million.

Fig. 5

Comparison of annual average of ground-based ${\mathrm{PM}}_{10}$ concentration and satellite-derived ${\mathrm{PM}}_{10}$ concentration.

4.2.

Advantages of Satellite-Derived PM₁₀ in Estimating Human Health Impact

The correlation between short-term satellite data and ground PM is relatively low in some meteorological conditions, especially when troposphere air changes. Tian and Chen⁴² found that the instantaneous satellite data were poorly related to the ground-based ${\mathrm{PM}}_{2.5}$ concentration due to changes in meteorological conditions. Hutchison⁴³ indicated that a stronger correlation can be obtained by averaging longer timescales’ satellite observation data and ground-based ${\mathrm{PM}}_{2.5}$ data. In this study, we obtained annual average AOD from MODIS Aerosol Product data for the years from 2010 to 2014 and annual average ground ${\mathrm{PM}}_{10}$ observations to avoid the problem of low correlation between AOD and ground ${\mathrm{PM}}_{10}$ concentration caused by instantaneous atmospheric vertical instability and different atmosphere conditions. Furthermore, the impact of ${\mathrm{PM}}_{10}$ exposure on human health is a long-term process, and research on human health impact based on long-term satellite-derived ${\mathrm{PM}}_{10}$ data can improve the accuracy of the results.

Correlating the human health impact with long-term ${\mathrm{PM}}_{10}$ concentration derived from satellite observations has many advantages. Currently, most works on the assessment of air pollution to human health in a certain region are usually based on the average ground-based observations data.^44,45 It is known that the air quality monitoring stations are mainly located in the areas where ${\mathrm{PM}}_{10}$ concentration is higher, such as urban areas. Consequently, ${\mathrm{PM}}_{10}$ concentration and subsequent human health risks and losses are overestimated. Although some studies^46,47 used the interpolated ${\mathrm{PM}}_{10}$ concentration from ground-based observations to access the human health risks and losses, these methods are constrained by physiochemical models and may not generate accurate results in complex terrain areas.¹⁸ Surface models²⁰ based on GIS and ground-based observations could provide more accurate results of ${\mathrm{PM}}_{10}$ concentration; however, they are usually not suitable for calculating long-term diffusion of air pollution in large areas. In our study, ${\mathrm{PM}}_{10}$ concentration derived from satellite observations can be considered more realistic than the above methods, since no direct interpolation on ${\mathrm{PM}}_{10}$ concentration is involved.

The distribution of population is another factor that has to be taken into account to accurately assess the impact of air pollution on human health. However, the statistical population data from census are given for each administrative district and do not provide sufficient spatial resolution²⁴ for the purpose of this study. To this end, based on the correlation between the total population and land-use type, we generated a population distribution map with a resolution of 10 km. Such a distribution map demonstrates the spatial variations of population over China. In addition, unlike traditional human health assessment methods, which overlay the in situ ${\mathrm{PM}}_{10}$ concentrations data over statistical population data given by the administrative district, we obtained the spatial distribution of human health risks in China by analyzing the spatial distribution of both ${\mathrm{PM}}_{10}$ concentration and population. To show the advantages of the approach used in our study, we calculated the human health risks and losses based on both the satellite observations data and ground-based observations data. Comparison of the results of the two approaches is given in Figs. 6 and 7.

Fig. 6

Comparison of relative risk based on ground-based observations data with relative risk based on satellite observations data at a provincial scale.

Fig. 7

Comparison of human health losses based on ground-based observations data with losses based on satellite observations data.

It is shown in Fig. 6 that relative risks based on satellite observations are generally lower than that based on ground-based observations in most provinces, except for Shanghai, which includes more highly populated areas, and Hainan, Guangdong, Guangxi, which are located in the subtropical monsoon climate zone. Humid weather and complex atmosphere conditions lead to a lower accuracy in estimating ${\mathrm{PM}}_{10}$ concentration in these regions. For all other provinces, relative risk by ${\mathrm{PM}}_{10}$ exposure based on satellite observations is lower than that based on ground observations data. As for the annual average human health losses caused by ${\mathrm{PM}}_{10}$ exposure from 2010 to 2014, as shown in Fig. 7, the total number of cases of detrimental health impact from ${\mathrm{PM}}_{10}$ pollution estimated with ground-based ${\mathrm{PM}}_{10}$ observations is 10.21 million, which is much larger than the estimation with the method proposed in our study.