Continuous daily maps of fine particulate matter (PM2.5) air quality in East Asia by application of a random forest algorithm to GOCI geostationary satellite data

We use 2011-2019 aerosol optical depth (AOD) observations from the Geostationary Ocean Color Imager (GOCI) instrument over East Asia to infer 24-h daily surface fine particulate matter (PM2.5) concentrations at continuous 6km x 6km resolution over South Korea, eastern China, and Japan. We use PM2.5 observations from national networks to train and cross-validate a random forest (RF) algorithm that predicts PM2.5 from the gap-filled GOCI AOD, meteorological variables, and other predictor variables. The predicted 24-h PM2.5 for sites entirely withheld from training in a ten-fold crossvalidation procedure correlates highly with observed concentrations (R2 = 0.89) with single-value precision of 26-32% depending on country. Prediction of annual mean values has R2 = 0.96 and single-value precision of 12%. More information is available in the associated publication. Here we supply a NetCDF containing the inferred daily PM2.5 fields from 2011-19 for use in further research. If you use this data, please cite the associated publication, and feel free to reach out via email to discuss this work.本研究利用2011-2019年东亚区域静止海洋水色成像仪（Geostationary Ocean Color Imager, GOCI）获取的气溶胶光学厚度（aerosol optical depth, AOD）观测数据，在韩国、中国东部及日本区域以连续6km×6km的空间分辨率反演逐日24小时平均地表细颗粒物（fine particulate matter, PM2.5）浓度。本研究利用国家监测网络获取的PM2.5观测数据，训练并交叉验证随机森林（random forest, RF）算法：该算法基于填补了数据间隙的GOCI AOD、气象变量及其他预测因子，实现PM2.5浓度的估算。在十折交叉验证流程中，针对完全未参与训练的监测站点，其预测得到的24小时平均PM2.5浓度与实测值相关性极强（决定系数R²=0.89），单值精度依国家不同介于26%~32%之间。年平均PM2.5浓度的预测结果决定系数R²可达0.96，单值精度为12%。更多细节可参见相关研究论文。本数据集附带2011-2019年反演得到的逐日PM2.5浓度场NetCDF文件，可供后续研究使用。若您使用本数据集，请引用相关研究论文，如有任何疑问或交流需求，欢迎通过邮件联系我们。