Abstract
Abstract. Aerosol refers to suspensions of small solid and liquid particles in the atmosphere. Although the content of aerosol in the atmosphere is small, it plays a crucial role in atmospheric and the climatic processes, making it essential to monitor. In areas with poor aerosol characteristics, satellite-based aerosol optical depth (AOD) values often differ from ground-based AOD values measured by instruments like AERONET. The use of 3km DT, 10km DT and 10km DTB algorithms in Beijing area has led to significant overestimation of AOD values, highlighting the need for improvement. This paper proposes the use of machine learning techniques, specifically support vector regression (SVR) and artificial neural network (ANN), to correct the deviation of AOD data. Our approach leverages ground-based monitoring data, meteorological reanalysis data and satellite products to train the models. Our results show that the ANN model outperforms the SVR model achieving R2, RMSE and Slope values of 0.88, 0.12 and 0.97, respectively, when applied to nearly two decades of data from 2001 to 2019. This study significantly improves the accuracy of MODIS AOD values, reducing overestimation and bringing them closer to ground-based AOD values measured by AERONET. Our findings have important applications in climate research and environmental monitoring.