Spatio-temporal Pattern Analysis Of Meteorologicaldata-based PM2.5 Concentrations In China:1980-2016

Since the reform and opening up,China's energy consumption has grown at an alarming rate with its rapid and massive economic development.The current ecological environment problems are extremely serious in China,particularly the air pollution.Previous studies have shown that air pollution can endanger public safety and human health.As a major pollutant causing haze and affecting human health,the PM2.5 pollution of China has caused widespread social concern.However,it was not until 2012 that China started to establish its ground PM2.5 monitoring network.The lack of long-term PM2.5monitoring data has impeded the study of environmental epidemiology and health effects related to PM2.5 in China,and made it difficult to control and manage PM2.5 pollution with scientific support.Due to the lack of relevant satellite data,it is impossible to obtain PM2.5 before2000 through the AOD-based PM2.5 inversion method.In this work,the PM2.5estimation methods were built based on the relationship between PM2.5 concentrations and meteorological factors(e.g.,visibility).The models this study considered included parametric regression models and machine learning models.Explanatory variables selected by stepwise regression contained visibility,temperature,relative humidity,site pressure,wind speed and maximum gust wind speed.Buffer zone with a radius of 40km was created for each meteorological site to match PM2.5 and meteorological data.The buffer zone of 207 meteorological sites could match PM2.5 monitoring sites.The 10-fold cross validation results indicate that machine learning models showed a certain degree of overfitting,the validation accuracy of Support Vector Regression,Random Forest and Artificial Neural Network is higher than their training accuracy.The parametric regression models do not show over-fitting,and the log-regression model has the best fitting effect.Its training R2 and RMSE are 0.72 and 20.14?g/8)~3,and validation R2 and RMSE are 0.69 and 21.04?g/8)~3.This study has tried to solve the over-fitting problem of machine learning models by increasing the training samples,and sites belonging to the same province were used to build models together.The results show that the increase of training samples alleviates the over-fitting to some extent,but the training time and fitting effect are still not as good as the log-regression model.Therefore,log-regression model was selected to estimate the PM2.5 concentration of 207 meteorological stations in China from 1980 to 2016.The results of spatiotemporal pattern analysis and correlations between PM2.5concentration and socio-economic factors are listed as follows.(1)During the period of1980-2016,the average PM2.5 concentration of the stations near E110-120°and N40°was relatively high.The average and standard deviation of the 207 sites were 42.45 and18.89?g/8)~3,and the PM2.5 concentration of northern China was obviously higher than that in the southern China.The north was the most polluted zone and Beijing-Tianjin-Hebei was the urban agglomeration with the highest PM2.5 concentration.(2)The center of the PM2.5 moved southeast from 1980 to 2005,and after that,it turned to move northwest.The two semi-axes of the PM2.5 standard deviation ellipse showed a tendency to shrink during the period of 1980-2016.The rotation of the ellipse could be divided into four phases:clockwise-counterclockwise-counterclockwise–clockwise.(3)The PM2.5 concentration showed an upward trend in most sites(146/207)during 1980-2016.China's PM2.5 concentration increased in a rate of 0.277?g/(8)~3?year),and southern PM2.5 increased more rapidly than that in northern China.The center was the region with the fastest increase in PM2.5 concentration,and the PM2.5 of Yangtze River Delta rose the fastest among all urban agglomerations.(4)All the socio-economic factors(i.e.,population density,non-agricultural population,regional gross domestic product,annual electricity consumption,gross industrial output value,the primary industry output value,the tertiary industry output value and the urban built-up area)were closely related to PM2.5 concentration,but their specific relationships were different among various regions.