The Retrieval of Land Aerosol Optical Depth Based on Machine Learning Approach Over South Asia

Aerosol optical depth is a crucial parameter that characterizes the impact of aerosols on light attenuation. This parameter serves as the fundamental prerequisite for quantitative remote sensing studies and holds significant importance in climate change research, atmospheric pollution monitoring, and environmental governance. In this article, four machine learning models, namely, LSTM, RF, XGBoost, and LightGBM, were used to achieve aerosol optical depth retrieval using MODIS data as input features. Spatiotemporal cross-validation was used for model training and evaluation, and the results showed that the LSTM model had the best performance among the four models. In independent sites validation, both the LSTM and LightGBM models exhibited superior performance compared to the MODIS Deep Blue algorithm, indicating that the machine learning methods will contribute to enhancing the capability of satellite remote sensing for aerosol optical depth retrieval.