Unraveling decadal changes in PM2.5 sources and secondary aerosol formation drivers using integrated receptor model and machine learning.

Long-term evaluation of PM2.5 sources and elucidation of the formation mechanisms of secondary aerosols are essential for developing targeted and effective air pollution control strategies under the evolving atmospheric environment. This study conducted PM2.5 sampling and chemical analysis from 2014 to 2024 in Shenzhen, China. Positive Matrix Factorization (PMF) and machine learning were applied to source apportionment and the identification of drivers of secondary aerosol formation. Annual mean PM2.5 declined from 37.7 to 17.3 μg m-3, primarily attributed to anthropogenic emission reductions. PMF identified secondary sulfate as the historically dominant source, exhibiting the fastest decline, with vehicle emissions becoming the largest contributor after 2019. Secondary nitrate declined only marginally, emerging as a critical bottleneck for further PM2.5 improvement. Machine learning revealed distinct formation regimes among secondary aerosols. Declines in secondary nitrate and secondary organic aerosol benefited from reduced emissions of precursors NOx and volatile organic compounds (VOCs); however, increasing atmospheric oxidation capacity tempered this decreasing trend, underscoring its vital regulatory role in secondary aerosol formation. Secondary sulfate was predominantly controlled by atmospheric oxidation with regional transport playing a greater role than local SO2 emissions. These findings provide guidance for future PM2.5 management, prioritizing NOx and VOCs emission controls and regional collaborative governance.