Study On Spatiotemporal Distributions Of PM_2.5 In China Using Satellite Remote Sensing

Particulate matter with aerodynamic diameter less than 2.5 μm (PM2.5, or fine particulate matter) is the major pollutant of haze. Numerous epidemiologic studies have shown that PM2.5 is strongly associated with adverse health effects. With the rapid development of social economy, haze pollution has gradually become a severe environmental problem in China. However, Chinese people and government have not realized the PM2.5 issues until recent years. It is not until end of 2012 that China have established regulatory PM2.5 monitoring network. The lack of spatially and temporally continuous ground PM2.5 measurements before 2013 makes it difficult to support the environmental management of PM2.5 and substantially restrains the epidemiological and health effect studies of PM2.5 in China. Against this background, based on satellite remotely sensed aerosol optical depth (AOD) data, this study developed a high-accuracy PM2.5-AOD statistical model and estimated the spatiotemporal PM2.5 concentrations in China, so as to provide exposure data and scientific basis for environmental management and health effect studies.To improve the model accuracy, this study used the latest version (Collection 6, or C6, which was released in 2014) AOD data of the Moderate Resolution Imaging Spectroradiometer (MODIS), which is aboard the U.S. Aqua satellite. To optimize the AOD data coverage, this study developed a custom inverse variance weighting (IVW) approach to combine the MODIS C6 Dark Target (DT) and Deep Blue (DB) AOD data. First, daily regression analyses were performed between collocated DT and DB AOD to fill the missing DB AOD in those pixels where only DT AOD data are available and vice versa. Second, the variance of the differences between gap-filled DT (or DB) AOD and Aerosol Robotic Network (AERONET)AOD values for each season was calculated. Finally, the reciprocals of the variances were used as weights to average the gap-filled DT and DB AOD data. Compared with the AERONET AOD, the R2 of the IVW combined AOD is 0.80, while the R2 of MODIS’s standard combined AOD is 0.81. And IVW combined AOD data have less bias than MODIS’s standard combined AOD. Both of them perform similarly, however, IVW AOD data have 80.6% greater coverage.Based on the PM2.5 ground measurements, IVW combined AOD data, meteorological data, and land use information of 2013, this study fitted a two-stage statistical model which can account for the spatial and temporal variability of PM2.5-AOD relationship. The first-stage linear mixed effect (LME) models were fitted for each province to calibrate the temporal variability of relationship between PM2.5 and AOD. Certain buffer zone was created for each province to ensure enough data records to produce a robust model-fitting dataset. The second-stage generalized additive model was fitted with the smooth terms of coordinates and land use parameters to represent the spatial contrast. The overall model fitting and cross validation R2 are 0.82 and 0.79, respectively. This model greatly improved the model accuracy in previous studies in China and approached the results in regional-scale studies conducted in the U.S.Based on the PM2.5-AOD model of 2013, this study estimated the historical PM2.5 concentration from 2004-2012. Due to the lack of ground measurements before 2013, PM2.5 concentrations of Jan-Jun,2014 were also estimated and evaluated using the ground measurements. The results show that the PM2.5 prediction accuracy increases as the time scale increases. The prediction accuracy at daily level was poor compared to the historical observations (R2=0.41). Nonetheless, the historical PM2.5 estimations at monthly and seasonal level exhibit satisfactory performance (R2=0.74 and 0.80, respectively). Sensitivity analysis shows that that the monthly mean AOD-derived PM2.5 with available days of at least 6 and 11 can represent a true monthly and seasonal mean values, respectively.Finally, this study clarified the spatiotemporal patterns of PM2.5 in China in detail. The spatial patterns of PM2.5 pollution are closely associated with terrain characteristics in China. Highly polluted areas mainly appears in areas with low and flat terrains, e.g., North China Plain, the Middle-lower Yangtze Plain, Sichuan Basin, and the Guanzhong Plain. High density of population, socio-economic activites, and pollutant emissions, together with the terrain and weather conditions which make the pollutants hard to diffuse, are the main reasons for the formation of these highly polluted areas. Population exposure shows that most of the population is concentrated in high PM2.5 areas and over 96% of the population lives in the areas with annual PM2.5 concentrations exceed the National Ambient Air Quality Standard (GB3095-2012) Level 2 standard. Seasonal variations show that winter is the most polluted season and summer is the cleanest season. Time series analysis shows that PM2.5 significantly increased from 2004 to 2007, but slightly decreased after 2008. The inflection point may be due to the synergistic effects of the strict energy conservation and emission reduction policy during the period of "11th five-year plan". Further studies are needed to explain the specific synergistic effects.The results of this study indicates that satellite remote sensing provides a feasible way to fill the temporal and spatial gaps left by ground PM25 monitoring network in China. The spatial pattern of long-term average PM2.5 can provide basic information for environmental risk zoning in China. The monthly mean PM2.5 time series can be used for assessing the performance of historical environmental policies. Besides, PM2.5 data produced in this study can also serve as exposure estimates for environmental epidemiologic studies and health impact assessments of PM2.5 in China, which can greatly advance the research of PM2.5 health effects in China.