A Novel Framework for Airshed Delineation and PM2.5 Estimation across India Using Machine Learning and Spatial Clustering

Air pollution continues to pose a major challenge in India, with PM2.5 being a key contributor to serious health risks. Its spatial distribution is influenced by climatic, topographic, and anthropogenic factors, which are often poorly represented in analyses limited to administrative boundaries. This study developed a novel framework for spatial airshed delineation to support effective air quality management. By integration of the PM2.5 concentration, meteorological data, and land characteristics with clustering algorithms, the study proposes seven major airsheds and five transitional regions across India to minimize inconsistencies between neighboring zones. The clustering patterns were found to be consistent across multiple years, allowing for standardized long-term management strategies. These airsheds also facilitated the identification of dominant pollution sources within each region. The national-level machine learning model using the random forest algorithm was developed using MERRA-2 reanalysis data and ground-based observations to estimate PM2.5 concentrations. Incorporating airshed-based clustering into model development significantly improved predictive performance, increasing the R2 from 0.71 to 0.80 and reducing the RMSE from 27.58 μg/m3 to 23.25 μg/m3. Overall, this study provides a robust, data-driven framework for airshed delineation and region-specific PM2.5 modeling, supporting more accurate, actionable, and localized air quality management strategies.