Reconstructing aerosol optical depth using spatiotemporal Long Short-Term Memory convolutional autoencoder

Abstract

Aerosol Optical Depth (AOD) is a crucial atmospheric parameter in comprehending climate change, air quality, and its impacts on human health. Satellites offer exceptional spatiotemporal AOD data continuity. However, data quality is influenced by various atmospheric, landscape, and instrumental factors, resulting in data gaps. This study presents a new solution to this challenge by providing a long-term, gapless satellite-derived AOD dataset for Texas from 2010 to 2022, utilizing Moderate Resolution Imaging Spectroradiometer (MODIS) Multi-angle Implementation of Atmospheric Correction (MAIAC) products. Missing AOD data were reconstructed using a spatiotemporal Long Short-Term Memory (LSTM) convolutional autoencoder. Evaluation against an independent test dataset demonstrated the model’s effectiveness, with an average Root Mean Square Error (RMSE) of 0.017 and an R2 value of 0.941. Validation against the ground-based AERONET dataset indicated satisfactory agreement, with RMSE values ranging from 0.052 to 0.067. The reconstructed AOD data are available at daily, monthly, quarterly, and yearly scales, providing a valuable resource to advance understanding of the Earth’s atmosphere and support decision-making concerning air quality and public health.