基于FA-SVM优化LUR模型的汾渭平原PM2.5时空格局模拟OA

To accurately capture the complex relationship between PM2.5 and predictive factors,and to obtain spatially continuous PM2.5 pollution distribution with higher resolution and predic-tion accuracy,a regional PM2.5 pollution early warning mechanism is constructed.In this study,the firefly algorithm-support vector machine(FA-SVM)is used to optimize the land use regres-sion(LUR)model,estimating the PM2.5 mass concentrations in the Fenwei Plain in 2019 at a spa-tial resolution of 1 km.The results indicate that,compared to conventional LUR and SVM mod-els,FA-SVM demonstrates superior predictive performance.The ten-fold cross-validation coeffi-cient of determination for FA-SVM is as high as 0.90,with a root mean square error and mean absolute error of 12.29 μg/m3 and 8.99 μg/m3,respectively.In contrast,the validation coefficient of determination for LUR and SVM are 0.75 and 0.85,respectively,with root mean square error values of 19.57 μg/m3 and 14.37 μg/m3,and mean absolute error values of 14.84 μg/m3 and 9.62 μg/m3,respectively.PM2.5 pollution in the Fenwei Plain in 2019 exhibits significant spatio-temporal heterogeneity.Temporally,PM2.5 pollution is most severe in winter,gradually decrea-sing in spring,autumn,and summer.Spatially,areas with relatively higher economic levels show higher PM2.5 mass concentration,forming high-value aggregation zones,while the Qinling Mountains region represents low-value aggregation zones.Overall,PM2.5 exhibits a spatial pattern of higher concentrations in the central region and lower concentrations in the surrounding areas.