Unravelling the Nexus of emission sources and meteorology on Regional PM­2.5: A Comprehensive Analysis Using Source Apportionment Model and Machine Learning for Effective Pollution Mitigation Strategies

The PM_2.5 concentrations in Northern India are one of the highest in the world, posing significant risks to human health and affecting climate and air quality. This study aims to assess the influence of different sources of emissions and meteorological conditions on the levels of PM_2.5 at a regional background location in Northern India. The study employed a combination of the Random Forest model with Shapley Additive exPlanations (RF-SHAP) and Partial Dependence Plot (RF-PDP), together with Positive Matrix Factorization (PMF), to assess the effects of various factors on PM­_2.5 pollution. The RF model accurately captured the variation of PM_2.5 (R² = 0.95) during the sampling period. The results show that emissions sources and meteorology accounted for approximately 79% (99.8 μg/m³ ± 68.9) and 21% (26.5 ± 18.3 μg/m³) of the variability in PM_2.5 levels, respectively. Secondary aerosols (SA) had the most significant influence among all sources, accounting for around 45.7% and having a SHAP value of approximately 23.6 μg/m³. Biomass burning had the second highest impact, contributing around 23.1% and having a SHAP value of approximately 19.3 μg/m³. The RF-PDP approach was utilized to assess the sensitivity of the combined influence of secondary aerosols and biomass burning on PM_2.5 concentrations. The results suggest that controlling concentrations of secondary aerosols below 25 μg/m³ and biomass burning below 15 μg/m³can reduce the overall PM_2.5 concentration by over 2.5 times. It is to be noted that even after the strategic control measures, PM_2.5 concentrations are predicted to be over 100 μg/m³. Given the critical role of secondary aerosols in PM_2.5 pollution and the complexity of their generation mechanisms, the temporal variations of SA concentrations and their drivers were also analyzed via RF-SHAP during the study period. The model results highlight that secondary aerosol formation is mostly driven by meteorological conditions (64% ~ 13.6 ± 18.5 μg/m³) than primary emissions (36% ~ 7.7 ± 10.4  μg/m³), making it difficult to implement control strategies due to dependence on meteorological conditions. However, the sensitivity analysis using RF-PDP suggests that under favourable meteorological conditions, strategic control of primary emissions like biomass burning and coal combustion can reduce the secondary aerosol concentration and consequently reduce particulate pollution. In conclusion, the findings aid in uncovering approaches to effectively mitigate particulate pollution by managing emissions during favourable meteorological situations. Thus, the integration of machine learning algorithms with expert decisions and existing methodology can assist in effectively addressing ambient air pollution and find extensive use in the field of air pollution.