Artificial Neural Network (ANN) for Multi-source PM2.5 Estimation Using Surface, MODIS, and Meteorological Data

In recent years, various kinds of satellite-derived aerosol products have been used to air quality monitoring. However, satellites are not sensitive to the near surface aerosol which impacts human health but the entire aerosol column. In this paper, we establish an artificial neural network (ANN) instead of multiple regression technique to lessen the surface PM2.5 estimation uncertainty from remote sensing. The PM2.5 estimating model was built using the ground monitoring data, MOD04_L2 product and ECMWF meteorological data over northern China from March to May 2008. The comparison with ground measurements on 10 monitoring sites shows that the ANN-derived daily mean PM2.5 has better performance than that obtained by the multiple regression method, with the respectively correlation coefficient of 0.80 and 0.58. Although the monitoring sites and time period in this experiment is limited, this research reveals the potential of estimating PM2.5 with ANN method.