Developing a machine learning model using satellite data to predict the Air Quality Index (AQI) over Korba Coalfield, Chhattisgarh (India)

Abstract

This study investigates air quality dynamics in the Korba district, Asia's largest coal mining hub in India, employing advanced machine learning approaches to analyse seasonal Air Quality Index (AQI) variations and their influencing factors. Through comprehensive evaluation of six machine learning algorithms, Random Forest emerged as the superior model with the highest predictive accuracy (R2 = 0.93), followed by XGBoost (0.92), LightGBM (0.91), CatBoost (0.89), Gradient Boosting (0.82), and K-Neighbors Regressor (0.79). The research revealed distinct seasonal patterns in air quality, with correlation analysis highlighting complex interactions between environmental parameters and pollutants. Strong positive correlations were observed between NO2 and SO2 (0.97) and CO and SO2 (0.94), while significant negative correlations emerged between LST and NDVI (−0.73) and between NO2 and O3 (−0.77). Seasonal analysis demonstrated that winter periods experience elevated pollution levels due to atmospheric inversion conditions, while monsoon seasons show improved air quality due to precipitation-induced pollutant scavenging. Spatial distribution analysis identified critical variations in pollutant concentrations across the region, mainly influenced by elevation, vegetation density, industrial activity, and meteorological conditions. The study emphasises the significance of seasonal variations in pollution patterns and highlights the effectiveness of machine learning approaches in predicting air quality for environmental management. These findings provide valuable insights for developing targeted intervention strategies, including the strategic placement of green belts and seasonal adjustment of industrial operations, to mitigate air quality deterioration in this critically important industrial region. © 2024 Turkish National Committee for Air Pollution Research and Control