Abstract Air pollution in India is complex due to the multitude of sources and varying topography, rendering the interplay between meteorology and emission sources significant. To address this challenge, this work presents an integrated methodology for PM 2.5 source apportionment in Bhopal, central India, combining dispersion‐normalized positive matrix factorization (DN‐PMF) with a machine‐learning interpretability approach using Random Forest and SHAP (RF‐SHAP). DN‐PMF improves conventional source identification by incorporating air dilution effects, yielding refined source contributions for nine factors in Bhopal. Seasonal factor contributions peaked during periods with a lower boundary layer height, such as secondary sulfate during the winter season (21.3 μg m −3 , 31.7%). The COVID‐19 lockdowns, a quasi‐natural emissions reduction experiment, led to a decrease in aerosol contributions from industrial, residential and traffic‐related sources. However, during this period, crop residue burning was exposed as a major anthropogenic contributor, which together with unfavorable meteorology resulted in increased mean PM 2.5 (50.6 ± 24.3 μg m −3 ) during the lockdowns compared to the reference period (36.7 ± 9.7 μg m −3 ). Using RF‐SHAP the influence of meteorology and emission sources in driving secondary inorganic aerosol formation was examined. Secondary nitrate and residential fuel were identified as key contributors to exceedances of the Indian National Ambient Air Quality Standards (60 μg m −3 , 24‐hr average) at the study site. Integrating DN‐PMF with RF‐SHAP (driver analysis) enhanced source attribution by linking source contributions with their driving factors, establishing a framework for assessing pollution dynamics. This framework can help strengthen improved air quality initiatives in India, including the national Smart Cities Mission.
Pullokaran et al. (Sun,) studied this question.