基于多特征参数的OMI遥感产品气溶胶分类研究——以广东省为例

The Multi-feature Parameter Classification of Aerosol Based on OMI Remote Sensing Data:A Case Study in Guangdong Province

为了解决气溶胶分类精度低和特征参数冗杂的问题,基于OMI(Ozone Monitoring Instrument)遥感产品的气溶胶特征参数,利用随机森林算法,将广东省2014年的气溶胶类型划分为沙尘型气溶胶(Desert Dust,DST)、生物质燃烧型含碳气溶胶(Carbonaceous Aerosols Associated with Biomass Burning,CRB)和硫酸盐型城镇-工业气溶胶(Sulfate-based Urban-industrial Aerosols,SLF)3种类型.并统计分析随机森林以及特征参数的重要性,将分类结果的空间分布与OMI气溶胶类型产品的空间分布进行对比.结果表明:(1)随机森林算法仅需少量训练样本点即可达到97%以上的总体分类精度.(2)通过计算不同气溶胶特征参数在随机森林分类过程中的重要性高低,得到重要性排名前六的特征参数依次为α指数、UVAI、RI388、RI354、SSA500、AAOD500,表明在分类过程中,气溶胶粒径分布和吸收能力起到了最关键的作用.(3)3种气溶胶类型的空间分布显示,SLF型气溶胶为广东省最主要的气溶胶类型;DST型和CRB型气溶胶在珠三角地区占比最高,在粤东、粤北地区的占比最低.

In order to solve the problems of low precision and redundant feature parameters in the process of aerosol classification,several significant aerosol feature parameters were extracted from the OMI(Ozone Monitoring Instrument)remote sensing products and Random Forest(RF)algorithm was used for aerosol type classification and verification.Aerosols of Guangdong Province in 2014 were divided into three types:desert dust(DST),carbonaceous aerosols associated with biomass burning(CRB)and sulfate-based urban-industrial aerosols(SLF).The Random Forest classification results and the importance of feature parameters were analyzed.The spatial distribution of cla-ssification results were compared to that of OMI aerosol type products.The following results are obtained.First,with the RF algorithm,a total precision of over 97%can be reached with a few training samples.Second,calculating the importance of different aerosol feature parameters in RF shows that the most important feature parameters are angstrom exponent,UVAI,RI388,RI354,SSA500 and AAOD500 in turn,indicating that size distribution and absorption ability of aerosols play the key roles.Third,the spatial distribution of three aerosol types shows that sulfate-based urban-industrial aerosols are dominant in the Pearl River Delta.Proportion of biomass burning and desert dust aerosols are highest in the Pearl River Delta and lowest in the eastern and northern Guangdong.