PM_2.5 Concentration Inversion And Spatiotemporal Distribution Characteristics In Eastern China

With the rapid development of China's economy and the acceleration of urbanization,air quality problems have become more serious.PM_2.5 refers to particles suspended in the air with a diameter of less than 2.5?m.As a major source of air pollution,PM_2.5 has a great impact on human health and normal life.Therefore,monitoring of PM_2.5 concentration has also become a hot issue.At present,most of China's monitoring of PM_2.5 is carried out in the form of stations.Each city is distributed with a national environmental air quality monitoring network city automatic monitoring station?commonly known as"national environmental air quality automatic monitoring station",referred to as"national control point"?.However,the number of national control points is limited,and most of them are located in the core areas of cities,which cannot be covered by remote areas.An interpolation method needs to be applied to acquire the area distribution of the concentration of PM_2.5,but this method has a large error,so a model for inversion of PM_2.5based on remote sensing is widely established.The use of AOD to build an inversion model is currently an effective way to obtain regional PM_2.5 concentrations,but there is no unified model standard to constrain the relationship between AOD and PM_2.5;and there is no PM_2.5 pollution driving factor analysis result integrating the knowledge of remote sensing and economics.In view of the above problems,this paper proposes a deep learning inversion method combining spatial division and Himawari data.In order to verify the effectiveness of the method,the research was carried out on Eastern China,which has the most economic and cultural development in China from 2016 to 2018,and analyzed the temporal and spatial distribution characteristics of PM_2.5 based on the inversion results.Finally,the fusion STIRPAT and spatial measurement models are used to analyze the factors affecting PM_2.5 pollution in2016.The main research contents are as follows:?1?Construction of inversion dataset based on spatial division.Introduce K-means clustering algorithm to cluster the national control points according to PM_2.5 time series characteristics,as an objective basis for zoning.Based on the time and space resolution of Himawari AOD,other data will be converted.Then,according to the position of PM_2.5national control points,the elements are partitioned in time and space.Eliminate missing values and filter out valid data to complete data preprocessing and construct partitioned inversion data sets.?2?Construction of PM_2.5 inversion model based on deep neural network.Taking the PM_2.5value of the national control point as the true value data,the Himawari-AOD data and various meteorological auxiliary elements are introduced,and the PM_2.5 inversion model is constructed by zones.Comparing the accuracy of the deep neural network model with the traditional linear and nonlinear regression models,the superiority of the deep neural network model in PM_2.5inversion is confirmed.The accuracy of the division model is compared with the global model to confirm the improvement effect of spatial division on PM_2.5 inversion accuracy.?3?Analysis of temporal and spatial distribution characteristics and influencing factors of PM_2.5 concentration.In addition to the analysis of the seasonal and annual spatial distribution characteristics of PM_2.5 based on regional inversion results,this paper further combines multi-source socio-economic factors?including population,economic,industrial,transportation,and greening data?.The improved STIRPAT and spatial measurement model are used to analyze the socio-economic causes of PM_2.5 pollution in different spatial divisions.The experimental results show that the multi-source element joint model based on deep neural network is the optimal PM_2.5 inversion model in Eastern China?R²>0.6,MAD<14,RMSE<20?,and compared with the division model?R²=0.77,0.78,0.72,0.61?and the global model?R²=0.73,0.78,0.72,0.23?accuracy,confirming that the K-means-based spatial division can improve the accuracy of the PM_2.5 inversion model;using the optimal model to analyze the spatial and temporal distribution characteristics,it was found that the PM_2.5 concentration has the characteristics of"year-by-year decrease"in the interannual period,"high in winter and low in summer,spring falls and autumn rises"in the season,and"high in the north and low in the south"in space;the driving model results show that there are differences in the socio-economic driving factors of PM_2.5 pollution in different zones,and the government should formulate a targeted governance plan.Therefore,the proposed PM_2.5 concentration inversion method combining"spatial division","high spatial-temporal resolution"and"deep learning"is significantly better than the accuracy level achieved by traditional methods,effectively fillling the gaps in PM_2.5 ground-based monitoring in time and space,and providing reliable basic data for PM_2.5 application research.In the construction of the driving model,the use of high-precision remote sensing inversion results,combined with economics,econometrics and other disciplines,to obtain more accurate and more reliable analysis results,provide a scientific and reliable basis for government decision-making.