Aerosol Optical Depth Forecast over Global Dust Belt Based on LSTM, CNN-LSTM, CONV-LSTM and FFT Algorithms

Aerosols are sources of the uncertainty in the global atmosphere and climate. They have many critical health, economic and social impacts. In this paper, we assess prediction of temporal monthly of Aerosol Optical Depth (AOD) over four dust sources within the global dust belt using three different algorithms. The three models are long-short term memory (LSTM), Convolutional neural networks-long-short term memory (CNN-LSTM) and Convolutional long-short term memory (ConvLSTM). Classical Fast Fourier Transform (FFT) algorithm for time series predication is compared to the three neural networks models. Grid search is used to find the optimal internal weights for the proposed neural network algorithms. The four dust sources are Eastern Libyan Desert, Saudi Arabia Peninsula, Indian subcontinent and China. Monthly temporal (2005-2021) AOD product from Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2) reanalysis is selected for training and validation periods. The presented models for AOD predication show efficient performance and cheap solution from computational point of view. However, ConvLSTM algorthims shows the least RMSE within ± 10%.