基于iLME+Geoi-RF模型的四川省PM_(2.5)浓度估算  

Estimation of PM_(2.5) Concentration in Sichuan Province Based on Improved Linear Mixed Effect Model and Geo-intelligent Random Forest

作　　者：吴宇宏 杜宁[1] 王莉[1] 蔡宏 周彬 吴磊 敖逍 WU Yu-hong;DU Ning;WANG Li;CAI Hong;ZHOU Bin;WU Lei;AO Xiao(Mining College,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学矿业学院,贵阳550025

出　　处：《环境科学》2021年第12期5602-5615,共14页Environmental Science

基　　金：国家自然科学基金项目(41901225)。

摘　　要：高分辨率PM_(2.5)空间分布数据对动态监测和控制PM_(2.5)污染具有重要意义.选取Himawari-8气溶胶光学厚度(AOD)、ERA5气象再分析资料、DEM、土地利用数据、夜光遥感数据、增强型植被指数和人口数据等作为估算变量,使用改进的重采样法进行数据匹配,并提出改进的线性混合模型(iLME)结合地理智能随机森林(Geoi-RF)构建组合模型估算PM_(2.5)浓度.结果表明:①在选取的估算变量中,气溶胶光学厚度、气压、温度、相对湿度和边界层高度是影响2016年四川省PM_(2.5)浓度的重要因素,其相关系数分别为0.65、0.58、0.55、0.54和0.35.②iLME+Geoi-RF模型精度相较其他模型有较大提升,模型拟合R2、RMSR和MAE分别为0.98、3.25μg·m^(-3)和1.98μg·m^(-3),交叉验证R2、RMSR和MAE分别为0.89、7.95μg·m^(-3)和4.81μg·m^(-3).该模型可获取更高精度的四川省PM_(2.5)时空分布特征,为区域空气质量评估、人体暴露风险评价和环境污染治理提供更加合理地科学参考.③2016年四川省PM_(2.5)浓度存在显著的季节性差异,各季节PM_(2.5)浓度大小关系为:冬季>秋季>春季>夏季.2016年四川省月均PM_(2.5)浓度总体上呈先降后升的"Ⅴ"型趋势,最小值在6月,最大值在12月,8月和11月有微小起伏.在空间分布上四川省PM_(2.5)浓度总体上呈东高西低和局部污染程度较高的特点,高值区主要分布在城市快速发展和人口密集的东部地区,低值区主要分布在经济发展落后和人口稀疏的西部地区.④虽然不同模型估算出的PM_(2.5)浓度整体分布基本一致,但iLME+Geoi-RF模型能更准确有效地估算本研究区污染的空间分布.High-resolution PM_(2.5) spatial distribution data is of great significance for the dynamic monitoring and control of PM_(2.5) pollution.Himawari-8 AOD data,ERA5 meteorological reanalysis data,DEM,land-use data,and luminous remote-sensing data were selected as estimating variables,using an improved resampling method for data matching and an improved linear mixed model(iLME)combined with a Geo-intelligent random forest model to build the combined model for estimating PM_(2.5) concentration.The results showed that:①Among the estimated variables selected,AOD,SP,TEMP,RH,and BLH were important factors affecting the PM_(2.5) concentration of Sichuan Province in 2016,and their correlation coefficients were 0.65,0.58,0.55,0.54,and 0.35,respectively.②The prediction accuracy of the iLME+Geoi-RF model was greatly improved compared to that of other models.The model-fitted R2,RMSR,and MAE were 0.94,5.72μg·m^(-3),and 3.92μg·m^(-3),and the cross-validated R2,RMSR,and MAE were 0.82,10.20μg·m^(-3),and 6.44μg·m^(-3),respectively.The model can obtain more accurate spatial and temporal distribution characteristics of PM_(2.5) in Sichuan Province and provide a more reasonable scientific reference for regional air quality assessment,human exposure risk assessment,and environmental pollution control.③There was a significant seasonal difference in PM_(2.5) concentration in Sichuan Province,with the highest concentration of PM_(2.5) in winter,followed by spring and autumn,with the concentration of PM_(2.5) in summer being the lowest.In 2016,the monthly average PM_(2.5) concentration in Sichuan Province showed a V shape that first decreased and then increased,with the minimum value in June,the maximum value in December,and slight fluctuations in August and November.In terms of spatial distribution,the PM_(2.5) concentration in the eastern area of Sichuan Province was generally higher than that in the west,and the local pollution level was relatively high.The high-valued areas were mainly distributed in the eastern region,whe

关 键 词：PM_(2.5) Himawari-8 AOD 重采样 共线性诊断 iLME+Geoi-RF模型 时空变化 

分 类 号：X513[环境科学与工程—环境工程]