基于土地利用回归模型的江苏省空气污染物空间分布精细化模拟

Finely simulation of spatial distribution of atmospheric pollutants in Jiangsu based onlanduse regression models

目的建立土地利用回归(LUR)模型模拟江苏省空气污染物浓度的空间分布。方法基于国家监测站点,提取土地利用、道路交通、气象条件、地理条件、人口密度和工业污染源等数据,通过后向算法建立江苏省2016年PM2.5、PM10、NO2和CO的LUR模型,结合留一交叉验证和十次10折交叉验证对模型进行检验。建立1 km×1 km网格点,利用建立的LUR模型预测上述4种污染物的空间分布情况。结果共获取73个潜在预测变量,PM10、PM2.5、NO2、CO的LUR模型调整后的R2分别为0.911、0.596、0.590、0.552,交叉验证的R2接近模型R2,模型具有良好的预测能力和稳健性。模型各个变量的方差膨胀因子(VIF)均小于10,不存在共线性问题。结论基于国家监测站点的四种空气污染物(PM2.5、PM10、NO2、CO)的LUR模型较好地解释了省级较大空间尺度的空气污染物浓度空间分异性,可为流行病学研究提供支持。

Objectives To establish land use regression(LUR)models to simulate the spatial distribution of atmospheric pollutants concentrations in Jiangsu province.Methods The data of land use,road traffic,meteorological conditions,geographical conditions,population density and industrial pollution sources were extracted.The LUR model of PM2.5,PM10,NO2 and CO in Jiangsu province was established by backward algorithm.The model was tested by combining one-out cross-validation and ten 10-fold cross-validation.Grid points(1 km×1 km)were established,and the spatial distributions of the above four pollutants were predicted by the established LUR models.Results A total of 73 potential predictors were obtained.The adjusted R2 of the LUR model were 0.911 for PM10,0.596 for PM2.5,0.590 for NO2,0.552 for CO respectively,and the values of R2 of cross-validation were close to those of model.Models have good predictive power and robustness.The variance expansion factor(VIF)of each variable of the model was less than 10,which proved that there is no collinearity.Conclusions The LUR model of four atmospheric pollutants(PM2.5,PM10,NO2,CO)based on national monitoring sites better explains the spatial heterogeneity of atmospheric pollutants concentrations at provincial scales,which can provide support for epidemiological research.