四川盆地PM2.5与PM10高分辨率时空分布及关联分析

为深入了解四川盆地PM_2.5与PM₁₀污染情况,通过机器学习的方法,基于卫星遥感气溶胶产品（MAIAC）与国家环境空气质量监测网数据以及气象、地理、社会经济变量等,构建2个随机森林机器学习模型（R²均为0.86）,反演四川盆地2013~2017年间1km网格逐日PM_2.5与PM₁₀浓度时空分布,并分析两者的时空关联性.结果表明：2013~2017年四川盆地地面PM_2.5与PM₁₀平均浓度分别为47.8,75.2μg/m³.PM_2.5与PM₁₀浓度空间上均整体呈现"倒月牙"状分布,西部与南部区域浓度值较高.5a间,区域颗粒物浓度逐年递减,总降幅均达到27%,季节上则均具有"冬高夏低"的特点;PM_2.5与PM₁₀浓度空间相关性显著（相关系数0.96）,呈现"内强外弱"的格局,春夏季相关系数（0.91、0.90）低于秋冬季（0.96、0.96）.盆地西南部PM_2.5与PM₁₀比值较高,比值高低的季节性排序为冬季 > 秋季 > 夏季 > 春季.

High resolution spatiotemporal distributionand correlation analysis of PM2.5 and PM10 concentrations in the Sichuan Basin

In order to advance the understandings of the regional air pollution in the Sichuan Basin, two machine learning based models, random forests (RF), were developed to estimate the daily PM_2.5 and PM₁₀ concentrations on the 1km grid from 2013 to 2017. The datasets used for the model training included the satellite-retrieved aerosol optical depth (AOD) from the Multi-Angle Implementation of Atmospheric Correction (MAIAC) product, the ground-based observations from the state-managed air quality monitoring network, as well as the meteorological, geographical, and socioeconomic variables. The RF models showed superior performance in predicting PM_2.5 and PM₁₀ (R²=0.86 for both). The multiyear regional average PM_2.5 and PM₁₀ concentrations were 47.8 and 75.2μg/m³, respectively. The PM_2.5 and PM₁₀ levels were predicted to be higher in the western and southern basin than the other areas, exhibiting a shape of crescent. During these five years, the PM_2.5 and PM₁₀ concentrations both decreased by 27%. The particulate matter concentrations exhibited obvious seasonality with the highest in winter and the lowest in summer. The spatial distributions of PM_2.5 and PM₁₀ showed high correlation (r=0.96), with stronger correlation in the lowland areas and slightly weaker correlation in the surrounding mountainous area. The correlations in spring (r=0.91) and summer (r=0.90) were relatively lower than those in fall (r=0.96) and winter (r=0.96). Ratio of PM_2.5 to PM₁₀ was higher in the southwestern basin and showed adescending order of winter > fall > summer > spring.