Establishment Of PM_2.5 Concentration Prediction Model And Temporal-spatial Analysis In Beijing-Tianjin-Hebei Region

Due to the rapid development of urbanization and industrialization,China's consumption of resources is increasing day by day.Air pollution issues increasingly affect people's lives and health and national development strategies.Traditional monitoring of atmospheric pollutants is generally monitored through stations.China's monitoring of atmospheric pollution started relatively late.Inadequate distribution of monitoring stations and inadequate coverage have limited air pollution monitoring,and the maintenance of ground stations requires a lot of manpower.Material resources.Remote sensing technology is applied more and more based on its advantages of wide coverage area and real-time monitoring.To air quality monitoring.This paper uses the data of remote sensing data,meteorological data,and PM2.5 site monitoring in the Beijing-Tianj in-Hebei region from 2014 to 2016 as the main data source.Based on machine learning algorithms,the PM2.5 concentration prediction model,spatial-temporal distribution characteristics and major atmospheric pollutants Relevance has been discussed and studied in depth.The main research results are as follows:(1)A PM2.5 concentration prediction model was established using BP neural network and DBN algorithm based on MAIAC(Multi-angle implementation of atmospheric correction)AOD,9 types of meteorological parameters such as temperature,humidity and atmospheric pressure,and PM2.5 site monitoring data.In addition,the 1km-resolution PM2.5 concentration distribution was inverted,and the spatial resolution and effective value coverage were greatly improved compared to the MODIS 3km-resolution aerosol product.(2)In the prediction of PM2.5 concentration,the prediction accuracy of DBN model and BP neural network model is compared and analyzed.The results show that:DBN model R2=0.71,root mean square error RMSE=17.63?g/m3,average error is 15.0?g/m3;The BP network prediction model is 0.68,23.51?gg/m3,and 20.03?gg/m3,respectively,which shows that the DBN model is superior to the BP neural network model.(3)The monitoring data of PM2.5 stations were analyzed.The time distribution features are as follows:PM2.5 concentration values is relatively high in January and gradually increased to the highest value in February and March.It began to gradually decrease in April and reached the minimum value of pollution in June.There will be a certain degree of rise in July.And the rise dropped again in August and September.In October,November and December,it gradually increased until January and February next year.The analysis of the spatial distribution of PM2.5 concentration inversion of the prediction model shows that the concentrations in the mountainous areas in the west and north are generally lower,and the concentrations in the central,southern and eastern urban areas are higher,and the areas affected by PM2.5 in Beijing,Langfang,Tianjin,and Baoding are relatively Serious,PM2.5 concentrations were low in Qinhuangdao,Tangshan,and Zhangjiakou.(4)Based on the correlation analysis of major atmospheric pollutants and PM2.5,due to factors such as fuel combustion and automobile exhaust emissions,PM2.5 concentration in the Beijing-Tianjin-Hebei region was significantly affected by CO,NO2,and SO2,and far greater than the effect of O3.In summary,based on the remote sensing image AOD data,meteorological data and PM2.5 site data,the PM2.5 concentration prediction model established has high precision and can be used to better invert the regional PM2.5 concentration spatial distribution.The spatial and temporal distribution characteristics of PM2.5 were analyzed.According to the correlation of PM2.5 concentration with pollutants such as SO2,NO2,CO,and O3,it was concluded that CO contributed the most to PM2.5 concentration.