Comparative Analysis of Factors Influencing PM2.5 Using Sentinel-5P and CPCB Data by Machine Learning Techniques: Case Study of Gurugram City (2019–2023)

Abstract

PM_2.5 particulates are major contributing factors that pose a serious threat to public health, particularly in urban cities like Gurugram, India. This study investigates the spatiotemporal variations of PM_2.5 concentration in Gurugram from 2019 to 2023 by integrating satellite and surface-level data. Meteorological and ground-level air quality data were collected through the Central Pollution Control Board, and spatial patterns of pollution were collected using satellite data from Google Earth Engine. Supervised Machine learning algorithms were then used to predict PM_2.5 concentrations and identify the key parameters influencing pollution. Among the evaluated models, the Random Forest algorithm demonstrated superior performance, achieving a coefficient of determination (R²) of 0.912, a mean absolute error of 2.946, a root mean square error of 5.013, and a mean square error of 25.13 Analysis has revealed that ground-based predictors exhibited stronger linear association with PM_2.5, whereas satellite-derived predictors captured broader regional trends. Strict tests for accuracy and precision in satellite-retrieved data were performed using comparative studies with ground measurement datasets. Temporal analysis indicated strong seasonal variation with the elevated PM_2.5 recorded during the winter months, whereas spatial analysis using satellite data revealed that the densely populated areas and transportation-dominated zones have high levels of pollutants. The findings demonstrate that the combination of satellite-based atmospheric indicators with ground measurements increases the spatiotemporal characterisations of air quality in cities.