Spatio-Temporal Prediction And Pollution Prevention Of PM_2.5 In The Yangtze River Delta

With the rapid growth of the economy,a series of environmental problems such as energy depletion,soil loss,and air pollution have followed,and increasingly serious air pollution issues have brought great challenges to environmental governance.In recent years,China’s economic growth has ranked first in the world,and the problem of air pollution has become increasingly severe.Taking the Yangtze River Delta（YRD）region as typical,PM_2.5coordinated regional composite pollution with other pollutants has become a problem of current air pollution prevention.Facing with the status quo of complex air pollution,grasping the characteristics of spatial-temporal evolution of PM_2.5and its composite pollution on the basis of accurate prediction of PM_2.5concentration,exploring the coordinated prevention and control strategies of composite pollution,are of great significance to the effective implementation of our country’s air pollution joint prevention joint control strategy.Therefore,under the current situation of uneven distribution of air quality stations and widespread absence of PM_2.5historical monitoring data,this paper in-depth discussed the filling effect of various imputation methods in the case of real missing of PM_2.5monitoring data,and used the optimal model to fill the missing of PM_2.5historical monitoring data in the YRD from 2015 to2020.I built the spatial and temporal prediction model of PM_2.5in the YRD from the two scales of site and grid,analyzed the pollution trends of PM_2.5and O₃based on the forecast data set,and studied the regional division method for the coordinated prevention of PM_2.5and O₃composite pollution.The main conclusions of this paper are as follows:（1）In the task of missing value imputation,CSDI model has better interpolation accuracy than methods of ARIMA,KNN,and multiple imputation,and has good interpolation effect on both the continuous missing task and the discontinuous missing task.The ARIMA model has superior performance in the discontinuous missing task,and the interpolation effect on the continuous missing task is very poor,while KNN and multiple imputation methods show instability in the interpolation effect.Based on the K-Shape clustering partitioning results,the filling study of PM_2.5historical monitoring data of the YRD shows that the model accuracy is the highest in the third region which in the eastern part of the YRD,while the lowest in the second region in the northern part of the YRD.This phenomenon is attributed to the accuracy of the clustering region and the PM_2.5variation characteristics of the region itself.（2）The spatio-temporal random forest model constructed by the factors as AOD,temperature,precipitation,relative humidity,wind speed,NDVI,DEM and spatio-temporal information items can effectively predict PM_2.5concentration,the prediction accuracy of monthly mean concentration shows R²=0.764,MAE=7.238μg/m³,RMSE=11.135μg/m³.In the spatio-temporal random forest model,AOD is the most important,with a contribution of 27.48%,while DEM contribute less than 2%.The prediction ability of the deep learning model is significantly higher than that of the machine learning model for the prediction task of PM_2.5concentration in the future one hour at multiple sites.Among them,the prediction accuracy of C-TCN model was the highest,R²reached 0.904,and the model error was significantly decreased compared with the baseline model（LSTM and GRU）,RMSE decreased by 6.484～8.56μg/m³and MAE decreased by 5.894～6.572μg/m³.（3）From 2015 to 2020,PM_2.5concentration in the YRD decreased year by year,while O₃concentration increased steadily,the combined pollution of PM_2.5and O₃is serious in the northern and central part of the YRD,the eastern coastal area is dominated by O₃pollution,and the overall air quality in the southwestern mountainous area is good.Under the different influences of atmospheric temperature and relative humidity,PM_2.5concentration shows a U-shaped trend in monthly variation,while O₃concentration shows an inverted U-shaped trend.PM_2.5concentration in Hefei,Hangzhou,Changzhou,Wuxi decreased by more than 5μg/m³annually,O₃concentration increased significantly in the whole Anhui Province and the western part of Jiangsu Province,with an average annual increase of 5～10μg/m³.The YRD region experienced a transition from PM_2.5pollution to O₃pollution from 2015 to 2020,with the O₃pollution area accounting for 40.3%in 2020.In order to effectively control air pollution,relevant departments should pay attention to the transformation of industrial structure and energy structure,and adopt the strategy of reducing VOCs first and then NOx on limiting the emission of precursors.