A Spatial Neighborhood Deep Neural Network Model for PM2.5 Estimation Across China

Fine particulate matter, specifically PM2.5, has raised increasing public and governmental concerns over the past decade for its threats to the environment and public health. For large-scale PM2.5 estimation, spatial neighborhood information is frequently ignored when modeling the spatiotemporal heterogeneity of the PM2.5-aerosol optical depth (AOD) relationship. In this regard, applying convolutional neural networks (CNNs) to extract the spatial neighborhood characteristic has great potential; therefore, this article establishes a spatial neighborhood deep neural network (SNDNN) model to predict PM2.5 concentrations across China. In addition to the backward propagation neural network (BPNN) model for extracting spatiotemporal features, the model integrates a CNN model to achieve spatial neighborhood data mining within a <tex-math notation="LaTeX">3 × 3 </tex-math> km2 window. The cross-validation (CV) results show that the daily model achieves high accuracy and stability from 2016 to 2020. The coefficient of determination ( <tex-math notation="LaTeX">R² </tex-math>) value reached 0.92 in 2018, with a root mean square error (RMSE) of <tex-math notation="LaTeX">9.39 μ g/m³ </tex-math> and a mean absolute error (MAE) of <tex-math notation="LaTeX">6.09 μ g/m³ </tex-math>. Further, a monthly SNDNN model is established to predict seasonal and annual PM2.5 concentrations with greater accuracy and wider spatial coverage. The study results demonstrate the superiority of introducing spatial neighborhood information to PM2.5 estimation and indicate that the SNDNN can provide a practical reference for spatial neighborhood feature extraction.