The Research Of GWR Method To Simulate PM2.5 Concentration Based On PCA

At present the models of predicting PM2.5 concentration distribution mainly include spatial interpolation?MODIS remote sensing imagery reflection?Land Use Regression(LUR)and so on at home and abroad.Spatial interpolation has lots of advantages such as simple principle?convenient operation and wide range of application.However,there is a big discount when facing with such situation when air quality monitoring sites are distributing sparsely.LUR has its own feature such as having lower requirements for data and taking many factors into account,but it can only reflect the impact the variables of geographical elements on PM2.5 concentration in global or average sense,and it could not show the variability characteristics of local spatial.Therefore,this paper integrates with Principal Component Analysis(PCA),uses Geographically Weighted Regression(GWR)to explore non-stationarity of PM2.5 concentration,futher studies the spatial distribution characteristics of the effect each variable on PM2.5 concentration.and finds out the local reason why PM2.5 generated.Not only can this method solve multicollinearity problem between variables of geographic factors,but also it can explore and analyze spatial variability characteristics of regression correlation.In order to verify the effectiveness of using the GWR method to simulate PM2.5 concentration based on PCA,the paper made comparative analysis experiments using LUR and CoKriging(CK)respectively The results indicate that compared to LUR and CK,the model precision of GWR methods based on PCA increased by 18.7 percent and 25.77 percent respectively,and the goodness of fit improved by 0.4 percent and 10.6 percent respectively.