利用支持向量机遥感估算京津冀细颗粒物浓度

Estimation of Surface PM_(2.5) Concentration in Beijing-Tianjin-Hebei Region

针对传统地面监测手段难以获取全面的细颗粒物(fine particulate matter,PM_(2.5))浓度信息,以2015年3月份京津冀地区PM_(2.5)污染状况为研究对象,将卫星遥感产品气溶胶光学厚度(aerosol optical depth,AOD)作为单变量输入构建支持向量机回归模型,得到的预测值与真实值的R^2为0.525,相对误差为44.6%。鉴于相对湿度(relative humidity,RH)和边界层高度(planetary boundary layer height,PBLH)是PM_(2.5)形成机制的重要影响因素,遂将RH、PBLH与AOD一起作为输入特征构建支持向量机回归模型预测PM2.5,得到的结果 R^2为0.729,相对误差为33.3%。研究结果表明:基于AOD、RH、PBLH为输入特征构建的支持向量机回归模型能够较好地从空间层次预测PM_(2.5)质量浓度。

Considering the difficulty of traditional monitoring methods in acquiring of the information of PM（2.5）,this study researched the PM（2.5） pollution of Beijing-Tianjin-Hebei region in March 2015 and built support vector regression（SVR）model which takes aerosol optical depth（AOD）as an input feature,aproduct of satellite remote sensing.The results gave R2 between estimated and observed PM（2.5） concentration 0.525,and the relative error 44.6%.Moreover,thinking of the relative humidity（RH）and planetary boundary layer height（PBLH）are important influence factors of PM（2.5）,a new support vector regression model by taking in more effective factors of RH,PBLH and AOD as input features was built,with R2 0.729,and the relative error 33.3%.The results showed that the SVR model based on AOD,RH and PBLH can predict PM（2.5） better.