Assessing particulate matter (PM2.5) concentrations and variability across Maharashtra using satellite data and machine learning techniques

• Published in: Discover sustainability Vol. 6; no. 1; pp. 238 - 20
• Main Authors: Kunjir, Ganesh Machhindra, Tikle, Suvarna, Das, Sandipan, Karim, Masud, Roy, Sujit Kumar, Chatterjee, Uday
• Format: Journal Article
• Language: English
• Published: Cham Springer International Publishing 04.04.2025; Springer Nature B.V; Springer
• Subjects: Accuracy; Aerosols; Air pollution; Earth and Environmental Science; Environment; Fine mode aerosol optical depth (FAOD); Fine particulate matter (PM2.5); Ground stations; Machine learning; Neural networks; Outdoor air quality; Public health; Remote sensing; Satellite remote sensing; Sustainable Development; Western Ghats; Arabian Sea; Maharashtra India; Madhya Pradesh India; India; Satellite remote sensing; Fine particulate matter (PM; ); Fine mode aerosol optical depth (FAOD); Machine learning
• ISSN: 2662-9984; 2662-9984
• DOI: 10.1007/s43621-025-01082-3

Airborne fine particulate matter (PM 2.5 ) is recognized globally as one of the most hazardous air pollutants due to its profound impact on human health, contributing to respiratory and cardiovascular diseases, and increasing the risk of premature mortality. The World Health Organization (WHO) attributes millions of deaths annually to PM 2.5 exposure, making it a critical subject of study for both environmental and public health research. In this context, the present study aims to predict PM 2.5 concentrations across Maharashtra, India, for the year 2023, employing machine learning models to improve spatial and temporal air quality assessments. The analysis utilizes daily station-specific datasets, incorporating PM 2.5 concentrations, Fine Aerosol Optical Depth (FAOD), wind components (u and v), relative humidity (RH), and air temperature (TEMP) to improve prediction accuracy. Four regression models were applied: Random Forest (RF), Multiple Linear Regression (MLR), Linear Regression (LR), and Lasso Regression, using a combination of Fine Aerosol Optical Depth (FAOD) with meteorological data from Google Earth Engine and ground-based observations from Central Pollution Control Board (CPCB) monitoring stations. The study emphasizes the importance of utilizing FAOD as a more refined metric for fine-mode aerosol concentration in PM 2.5 modeling, compared to conventional AOD. The RF model achieved the highest accuracy (R 2  = 0.87, RMSE = 12.57 µg/m 3 , MAE = 6.96 µg/m 3 ), outperforming MLR, LR, and Lasso Regression, which showed significantly lower R 2 values. This highlights the RF model’s effectiveness in capturing the non-linear relationships between PM 2.5 and its environmental factors. This study identified key PM 2.5 hotspots in Maharashtra, particularly in densely urbanized areas like Mumbai, Thane, and Pune, with annual PM 2.5 concentrations reaching 46.34 µg/m 3 , far exceeding the Indian National Ambient Air Quality Standards (NAAQS) of 40 µg/m 3 . Seasonal analysis revealed significant variability, with the highest PM 2.5 concentrations observed during the winter months, while levels significantly decreased during the monsoon due to higher rainfall and increased atmospheric moisture. The study identifies key PM 2.5 hotspots in urban areas, offering crucial insights for policymakers and urban planners to implement targeted air quality interventions. These findings support improved public health and sustainable environmental management in Maharashtra.