卫星遥感监测近地表细颗粒物多元回归方法研究

对地基监测PM2.5和气象数据、MODISAOD卫星数据与NCEP FNL数据进行了处理分析,在与一元简单线性模型(模型1)进行对比的基础上,建立了适应于北京及其附近地区遥感监测近地面颗粒物(PM2.5)浓度的多元线性(模型2)和非线性(模型3)回归模型,并对模型进行了评价验证和遥感监测初步应用.结果表明:模型1,2,3分别能够解释PM2.5 32.5%,56.1%,62.7%的变异.反演的PM2.5浓度与站点监测值相关性分别为0.5488(R2=0.3012), 0.7449(R2=0.5549), 0.7431(R2=0.5523).对于站点监测PM2.5浓度63.1652μg/m3的均值,反演均方根误差RMSE分别为43.5562, 35.3321, 36.8450μg/m3.模型2和3中气象因子分别能够解释PM2.5 23.6%和12.6%的变异,说明了气象因子影响北京地区春季PM2.5-AOD关系的显著性.3种模型整体上都不同程度地存在着低值高估和高值低估的现象.

A study of multiple regression method for estimating concentration of fine particulate matter using satellite remote sensing

In comparison with the simple linear model (Model 1), we developed two multiple regression models- linearmodel (Model 2) and nonlinear model (Model 3)-to estimate the ground PM2.5concentration using satellite observationsover Beijing and its surrounding area based on the analysis of the PM2.5data, the meteorological data, the MODIS AOD dataand the NCEP FNL data. Results showed that Model 1, Model 2 and Model 3 could explain 32.5%, 56.1%, 62.7% of the variability in ground-level PM2.5 concentration respectively. Correlation coefficients (R) of the three model estimated values of PM2.5 mass concentration with the actual observations were 0.5488, 0.7449, 0.7431 respectively. With an average PM2.5 concentration of 63.1652 μg/m3, their RMSEs were 43.5562, 35.3321,36.8450μg/m3 respectively. Meteorological factors in Model 2 and Model 3 could separatelyexplain 23.6%, 12.6% of the variability in ground-level PM2.5 concentration, which indicatedtheir significant influenceson the PM2.5-AOD relationship. In addition, there were low-value overestimation and high-value underestimation phenomenon in the three models.