一种基于机器学习和葵花8号数据的PM_(2.5)浓度监测方法

A Method for PM_(2.5)Concentration Monitoring Based on Machine Learning and Himawari-8 Satellite Data

基于卫星遥感数据和气象数据构建模型是实现PM_(2.5)浓度监测的重要方法。利用2017-2020年河南区域的葵花8号AOD相关产品、地面气象观测数据及PM_(2.5)浓度逐时地面观测数据,建立样本数据库,探究了PM_(2.5)浓度与遥感反演参数、气象因子的相关关系,提出一种基于空间距离的机器学习方法,构建了卫星遥感与气象数据相结合的PM_(2.5)监测模型,并利用PM_(2.5)观测数据对模型进行散点拟合和空间分布的验证。结果表明:PM_(2.5)浓度与能见度(V)、气温(T)、气溶胶光学厚度(AOD)有极为显著的相关关系;从PM_(2.5)浓度反演值与观测值的散点回归分析来看,PM_(2.5)浓度反演值与观测值相关系数R^(2)=0.73;从PM_(2.5)浓度反演值与观测值的空间分布图来看,PM_(2.5)浓度反演值实现了空间格点的小时监测,弥补了非晴空及夜间监测数据的缺失,在空间分布上与观测值一致,具有很高的精确性和可靠性。基于空间距离的机器学习方法构建PM_(2.5)浓度监测模型能很好地应用于PM_(2.5)浓度监测,为PM_(2.5)浓度监测提供了新的思路和方法。

Modeling based on satellite remote sensing data and meteorological data is an important method for PM_(2.5) concentration monitoring.Based on the AOD related products of Himawari-8,meteorological observation data,hourly observation data of PM_(2.5) concentration in Henan from 2017 to 2020,we established a sample database in this research,explored the correlation between the PM_(2.5) concentration and the remote sensing retrieval parameters and meteorological factors,proposed a machine learning method based on spatial distance,constructed a PM_(2.5) monitoring model combining satellite remote sensing and meteorological data,and used PM_(2.5) observation data in verifying the scattered point fitting and spatial distribution of the model.The results show that PM_(2.5) has a very significant correlation with visibility(V),air temperature(T)and aerosol optical thickness(AOD).From the scatter regression analysis of the PM_(2.5) concentration retrieval and observation,the correlation between PM_(2.5) retrieval and observation is very good(R^(2)=0.73).From the spatial distributions of PM_(2.5) retrieval and observation,the retrieved PM_(2.5) has realized the hourly monitoring of spatial grid points,making up for the lack of non-clear sky and night monitoring data,which is consistent with the observed values in spatial distribution,so its accuracy and reliability are very high.Therefore,the PM_(2.5) monitoring model constructed by the machine learning method based on spatial distance could be well applied to PM_(2.5) monitoring,which is a new idea and method for PM_(2.5) monitoring.