Hourly Ground-Level PM2.5 Estimation Using Geostationary Satellite and Reanalysis Data via Deep Learning

This study proposes an improved approach for monitoring the spatial concentrations of hourly particulate matter less than 2.5 ?m in diameter (PM2.5) via a deep neural network (DNN) using geostationary ocean color imager (GOCI) images and unified model (UM) reanalysis data over Korean Peninsula. The DNN performance was optimized to determine appropriate training structures, incorporating hyperparameter tuning, regularization, early stopping, input output variable normalization prevent dataset overfitting. Near-surface atmospheric information from UM also used as spatially generalize model. retrieved PM2.5 compared with estimates random forest, multiple linear regression, Community Multiscale Air Quality demonstrated highest accuracy that conventional methods hold-out validation (root mean square error (RMSE) = 7.042 ?g/m3, bias (MBE) ?0.340 coefficient determination (R2) 0.698) cross-validation (RMSE 9.166 MBE 0.293 R2 0.49). Although low due underestimated high concentration patterns, RMSE reliable values (<10 ?g/m3 1 respectively) cross-validation.