A study on the impact of meteorological and emission factors on PM_2.5 concentrations based on machine learning

PM_2.5 pollution, a major environmental and health concern, is influenced by a complex interplay of emission sources and meteorological conditions. Accurately identifying these factors and their contributions is essential for effective pollution management. This study applies Positive Matrix Factorization (PMF) to identify primary sources of PM_2.5 and uses the Light Gradient Boosting Machine (LightGBM) model, SHapley Additive exPlanations (SHAP), and Partial Dependence Plots (PDP) to quantitatively assess the impact of meteorological and emission factors on PM_2.5 concentrations. SHAP results reveal that meteorological factors contribute 16.6% (5.3 μg/m³) to PM_2.5, with humidity being the most influential, while emission sources account for 83.4% (26.8 μg/m³), with secondary particulate matter being the dominant factor. Secondary particulate matter and biomass burning significantly impacted PM_2.5 in the first and fourth quarters, while dust sources became more influential in the second quarter, and coal emissions were most prominent in the second and third quarters. Two-dimensional PDP analysis indicated that in the first and fourth quarters, secondary particulate matter concentration increased with air pressure, and the atmospheric oxidation process was more pronounced under high-humidity conditions during the day. Strong transport conditions, with wind direction shifting from north to east, also influenced secondary particulate matter levels. This study demonstrates that the LightGBM model effectively captures the nonlinear relationships between PM_2.5 and meteorological and emission factors, providing a reliable approach for analyzing the causes of PM_2.5 pollution.