Modeling PM2.5 concentration in tehran using satellite-based Aerosol optical depth (AOD) and machine learning: Assessing input contributions and prediction accuracy

The adverse effects of PM_2.5 on human health and the environment necessitate precise and continuous monitoring of this pollutant. Satellite remote sensing technology provides an effective and cost-efficient alternative to ground-based measurements. However, accurately estimating ground-based PM_2.5 concentrations using Aerosol Optical Depth (AOD) is challenging due to the influence of various parameters and atmospheric conditions on the AOD-PM_2.5 relationship. In this study, Aerosol Optical Depth (AOD) data were retrieved from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor at a spatial resolution of 1 km, covering a ten-year period from 2012 to 2023. The objective was to estimate PM₂.₅ concentrations for six ground-based monitoring stations in Tehran, Iran. These estimated concentrations were compared with daily measurements from the Tehran Air Quality Control Company air pollution monitoring stations for the same period. To determine the most significant conditioning factors in the modeling process and their impacts, the genetic algorithm optimization method and the Recursive Feature Elimination (RFE) technique were employed. The results indicated that, in addition to the AOD parameter, meteorological parameters such as wind speed, wind direction, temperature, precipitation, normalized difference vegetation index (NDVI), and visibility (VIS) could enhance model accuracy. Predictions were made using three machine learning algorithms: Random Forest (RF), Support Vector Machine (SVM), and Gaussian Process Regression (GPR). The findings revealed that the RF method was the most accurate, achieving accuracy in the range of 94–98 % for predicting PM_2.5 concentrations for all the studied stations. This study's results can significantly aid policymakers and researchers in utilizing satellite data for air pollution monitoring and management.