PM2.5 and PM10 Concentration Estimation Based on the Top-of-Atmosphere Reflectance.

Estimating ground-level PM2.5/10 based on satellite aerosol optical depth (AOD) products is a research hotspot at home and abroad. It has large area and high-density coverage characteristics, making up for the lack of ground monitoring stations. The AOD products are usually retrieved from top-of-atmosphere (TOA) reflectance via an atmospheric radiative transfer model. However, the strict surface assumptions in the AOD retrieval process make it impossible to retrieve AOD effectively in specific regions or periods. Therefore, this paper proposes a method based on machine learning to estimate ground-level PM2.5/10 concentration using TOA reflectance, observation angles and meteorological data, called TOA-PM2.5/10 model, and compares it with the AOD-PM2.5/10 model, whose inputs are AOD data and meteorological data. The comparative results show that the R2, RMSE, and MAE of PM2.5/10 concentration estimated using the TOA-PM2.5/10 model can reach 0.888, 6.158, 3.580 for PM2.5 and 0.889, 13.887, 8.141 for PM10 respectively, which is superior to that of the AOD-PM2.5/10 model.