Estimation of PM2.5 Concentration Using Deep Bayesian Model Considering Spatial Multiscale

Directly establishing the relationship between satellite data and PM2.5 concentration through deep learning methods for estimation is an important means estimating regional concentration. However, due to lack of consideration uncertainty in methods, based on have certain overfitting problems process estimation. In response this problem, paper designs a Bayesian model that takes into account multiple scales. The uses neural network describe key parameters priori, provide regularization effects network, perform posterior inference parameters, take characteristics uncertainty, which used alleviate problem improve generalization ability model. addition, different-scale Moderate-Resolution Imaging Spectroradiometer (MODIS) ERA5 reanalysis were as input strengthen model’s perception features atmosphere, well further enhance accuracy ability. Experiments with Anhui Province research area showed R2 method independent test set was 0.78, higher than DNN, random forest, BNN models do not consider impact surrounding environment; moreover, RMSE 19.45 μg·m−3, also lower three compared models. experiment different seasons 2019, other models, significantly reduced; however, could still reach 0.66 or more. Thus, has better