Spatial modeling of PM2.5 concentrations using an optimized land use regression method in Jiangsu, China

3rd International Conference on Energy Engineering and Environmental Protection

能源学;生态环境科学

Wang, Xintong^1 ; Qian, Yu^1

State Key Laboratory of Pollution Control and Resource Reuse, School of the Environment, Nanjing University, Xianlin Campus, 163 Xianlin Avenue, Nanjing

210023, China^1

关键词: Geographical locations;    Geographically weighted regression;    Land use regression;    Monitoring stations;    Multivariate linear regressions;    PM2.5 concentration;    Population exposure;    Spatial distribution map;   

Others  :  https://iopscience.iop.org/article/10.1088/1755-1315/227/5/052045/pdf
DOI  :  10.1088/1755-1315/227/5/052045

学科分类：环境科学（综合）

来源: IOP

【 摘 要 】

Land use regression method (LUR) has been recognized as a promising way to predict surface air pollutants concentration and spatial distribution of them globally. While linited studies have been conducted in China, especially in rather large areas. We therefore promoted an optimized LUR method which refined the predictor variables and improved the regression modeling method for PM2.5 spatial distribution in Jiangsu, China. Firstly, the integrated predictor variables which combined the total emission, distance to the monitoring station and wind direction are used to explore a more appropriate expression of variables e.g. pollution from point sources and line sources (traffic). The results showed that a model generated by integrated variables (R2 = 0.52) outperformed that generated by traditional variables (R2 = 0.34). Secondly, we compared the method of geographically weighted regression (GWR) which reflect the influence of geographical location on the effect of regression process with traditional multivariate linear regression (MLR) method. The model comparison results suggested that the overall prediction accuracy is significantly improved by 19% when using GWR model (R2 = 0.62). Furthermore, the spatial distribution map of predicted PM2.5 concentrations from GWR model was definitely finer than that from MLR model. It can be concluded that more appropriate expression of variables and the GWR modeling could definitely improve LUR modeling prediction accuracy. This study would not only demonstrate the applicability of the optimized LUR models in large geographical areas, but also support for fine population exposure studies in the future.