Document Type : Original Article

Authors

• Ali Sadeghnezhad
• Ali Shamsoddini

Department of Remote Sensing and GIS, Faculty of Humanities, Tarbiat Modares University, Tehran, Iran

Abstract

Objective: “Atmospheric Optical Depth (AOD) product effectiveness is limited for local-scale applications, particularly in urban air pollution monitoring, due to its coarse spatial resolution. This study aims to improve the accuracy of AOD estimation by leveraging Landsat remote sensing data and integrating advanced machine learning and deep learning techniques.”
Methods: To evaluate the accuracy of AOD estimation, six algorithms were tested: Multiple Linear Regression, Support Vector Regression (SVR), Random Forest Regression, Extra Trees Regression, Deep Multilayer Perceptron, and Deep Forest Regression. A decision-level fusion approach based on averaging was employed to integrate the results. The study utilized albedo data, 1 km resolution AOD products from the MODIS sensor, and top-of-atmosphere reflectance data from the OLI sensor (with 7 bands). Additionally, 180 AOD measurements from the AERONET (Aerosol Robotic Network) were used for training and validation. The study period covered the years 2014 to 2022.
Results: All algorithms demonstrated relatively strong performance, with R² values ranging from 0.73 to 0.86 and Root Mean Square Error (RMSE) values between 0.185 and 0.414. The Deep Multilayer Perceptron algorithm achieved the best performance, although the differences among the algorithms were minor. The decision-level fusion approach improved estimation accuracy, achieving an R² of 0.86 and reducing RMSE to 0.202.
Conclusions: The proposed method, which combines machine learning and deep learning techniques with Landsat-8 imagery, shows significant potential for generating high-resolution AOD datasets. This approach can enhance the accuracy of AOD estimation, making it more suitable for local-scale applications such as urban air quality monitoring.

Keywords

• Downscaling
• Machine learning
• Deep learning
• Aerosol Optical Depth
• MODIS
• Landsat-8

Main Subjects

• Remote sensing and Gis