Estimating Ground‐Level PM_2.5 by Fusing Satellite and Station Observations: A Geo‐Intelligent Deep Learning Approach

Abstract Fusing satellite observations and station measurements to estimate ground‐level PM 2.5 is promising for monitoring PM 2.5 pollution. A geo‐intelligent approach, which incorporates geographical correlation into an intelligent deep learning architecture, is developed to estimate PM 2.5 . Specifically, it considers geographical distance and spatiotemporally correlated PM 2.5 in a deep belief network (denoted as Geoi‐DBN). Geoi‐DBN can capture the essential features associated with PM 2.5 from latent factors. It was trained and tested with data from China in 2015. The results show that Geoi‐DBN performs significantly better than the traditional neural network. The out‐of‐sample cross‐validation R 2 increases from 0.42 to 0.88, and RMSE decreases from 29.96 to 13.03 μg/m 3 . On the basis of the derived PM 2.5 distribution, it is predicted that over 80% of the Chinese population live in areas with an annual mean PM 2.5 of greater than 35 μg/m 3 . This study provides a new perspective for air pollution monitoring in large geographic regions.