随机森林方法在PM2.5遥感反演中的研究及应用

Research and Application of Random Forest Method in PM2.5 Remote Sensing Inversion

伴随着城市建设的不断加快,城市空气质量问题也日益凸显。以重庆市为研究区域,将区域内的PM2.5浓度作为研究对象,利用MODIS传感器的L2级产品生产出分辨率为1 km的AOD数据,结合MERRA-2气象数据与地面监测站数据进行遥感反演,基于随机森林的方法构建了模型,将训练集与验证集数据应用到模型中,验证了该模型在研究区域PM2.5反演的效果,并通过样本均衡化的方法提高了模型的反演精确度。利用模型归纳总结了2022年重庆市PM2.5的分布集聚特征。结果表明:①样本均衡化的处理方式,提升了模型遇到高污染天气的鲁棒性,使得模型的拟合相关系数由0.45提高至0.96。模型与地面检测站数据的拟合相关系数为0.97与0.96,平均绝对误差为4.54 ug/m^(3)与6.05 ug/m^(3),平均相对误差为20.32%与26.76%,可以满足区域大气污染监测的需求与任务。②2022年,重庆市地区的PM2.5大气污染时间上呈现季节性变化,冬季最高,春季、秋季次之,夏季最低。空间上呈现“西高东低”的特征。

With the continuous acceleration of urban construction,the problem of urban air quality has become increasingly prominent.Taking Chongqing as the research area and the concentration of PM2.5 in the area as the research object,AOD data with a resolution of 1km are produced by using L2 products of MODIS sensors,and remote sensing inversion is carried out by combining MERRA-2 meteorological data and ground monitoring station data.A model is constructed based on the method of random forest,and the data of training set and verification set are applied to the model to verify the effect of the model in PM2.5 inversion in the research area,and the inversion accuracy of the model is improved by sample equalization.The distribution and agglomeration characteristics of PM2.5 in Chongqing in 2022 are summarized by using the model.The results show that:①The processing method of sample equalization improves the robustness of the model when it encounters highly polluted weather,and makes the fitting correlation coefficient of the model increase from 0.45 to 0.96.The fitting correlation coefficients between the model and the data of ground monitoring stations are 0.97 and 0.96,the average absolute errors are 4.54 ug/m and 6.05 ug/m,and the average relative errors are 20.32%and 26.76%,which can meet the requirements and tasks of regional air pollution monitoring.②In 2022,the air pollution time of PM2.5 in Chongqing showed seasonal changes,with the highest in winter,followed by spring and autumn,and the lowest in summer.It is characterized by“high in the west and low in the east”in space.