Two decades of PM2.5 pollution has seriously hindered China's sustainable development. However, relevant research of PM2.5 has been hindered because of the lack of long-term historical monitoring data. Therefore, ground observations of PM2.5 concentration from 2013 to 2016 in four typical regions of China and the MODIS aerosol optical thickness data, boundary layer height, temperature, and other meteorological data from 2000 to 2016 were used as the basic data. A combined simulation model was constructed by combining the two algorithms of backward artificial neural network and support vector regression and obtains the PM2.5 concentration history for the past 20 years using geospatial analysis technology. The results demonstrate that the combination model is better than the single model, with lower error and higher generalization ability. The spatial-temporal analysis results show that the concentration of PM2.5 continued to increase in the Beijing-Tianjin-Hebei region and in the three northeastern provinces of China, the PM2.5 concentration decreased slowly in the Pearl River Delta, the pollution range of PM2.5 in three of the research areas showed an expanding trend, and the PM2.5 concentration and pollution range remained stable in the Yangtze River Delta. In 2012, the concentration of PM2.5 in the four study areas decreased and the pollution range narrowed, but the PM2.5 concentration rose slightly after that decline and the high pollution range narrowed during 2013-2016, which with the country to take PM2.5 regional defense and other governance measures.
Keywords: PM2.5; backward artificial neural network; four typical regions; geospatial analysis; support vector regression.