Spatiotemporal Imputation of MAIAC AOD Using Deep Learning with Downscaling

Area of Science:

• Atmospheric Science
• Environmental Science
• Data Science

Background:

• Aerosols significantly impact human health and climate.
• The Multiangle Implementation of Atmospheric Correction (MAIAC) provides high-resolution Aerosol Optical Depth (AOD) data.
• Missing MAIAC AOD data due to clouds and high surface reflectance limits its utility for exposure and health studies.

Purpose of the Study:

• To develop and validate a deep learning imputation approach for filling gaps in MAIAC AOD data.
• To improve the spatiotemporal characterization of aerosols for health and climate research.
• To enhance air quality modeling applications by providing complete AOD datasets.

Main Methods:

• Utilized a deep learning autoencoder with residual connections and bagging for imputation.
• Incorporated downscaled Modern-Era Retrospective analysis for Research and Applications Version 2 (MERRA-2) GMI Replay Simulation (M2GMI) data as a gap-filling feature.
• Applied the method to impute weekly MAIAC AOD over California from 2000 to 2016.

Main Results:

• The deep residual network achieved high accuracy (mean R² = 0.94, RMSE = 0.007) in independent testing.
• Outperformed traditional methods like regular neural networks and generalized additive models.
• Imputed AOD showed strong correlation with AERONET AOD measurements (R = 0.83, R² = 0.69).

Conclusions:

• Deep learning provides a reliable method for imputing missing AOD data.
• The approach significantly enhances the usability of MAIAC AOD for air quality and health research.
• This method has important downstream applications for air quality modeling and exposure assessment.