Spatial heterogeneity and evolutionary pathways of PM₂.₅ pollution driven by urbanization in China to 2100

With the continued acceleration of global urbanization, PM₂.₅ pollution has become a critical environmental constraint to sustainable urban development. This study applied machine learning to predict the impacts of urbanization on PM₂.₅ concentrations and to examine long-term trends from 2020 to 2100 under four SSP scenarios. In addition, SHAP (Shapley Additive Explanations), ICE (Individual Conditional Expectation), and PDP (Partial Dependence Plots) analyses were used to elucidate the mechanisms through which urbanization influences PM₂.₅ levels. The results showed that population density, Urban area, and secondary industry output were the three dominant drivers of spatial PM₂.₅ differences, each exhibiting pronounced nonlinearities and synergistic effects. Moreover, except under SSP3-7.0, national PM₂.₅ concentrations generally declined, though substantial regional disparities remained. In terms of spatial dynamics, LISA time-path analysis indicated strong spatial spillover effects in the east, the greatest spatial structural instability in the west, and pronounced coordinated declines in the central and northeastern regions. Furthermore, the spatiotemporal transition matrix demonstrated that PM₂.₅ spatial structures exhibited strong path dependence and limited structural change across all scenarios. Overall, this study clarifies the key mechanisms through which urbanization drives PM₂.₅ evolution and reveals the spatial dynamics of pollution under different future scenarios, providing a scientific basis for region-specific and long-term air pollution control strategies.