A multi-evidence framework extended for interpreting spatiotemporal AQI dynamics and future risks in Qinghai province, China.

Air pollution remains a major global environmental health risk, and clarifying how the Air Quality Index (AQI) evolves and responds to coupled meteorological and human drivers is essential for effective governance, especially in high-altitude and ecologically sensitive regions such as Qinghai Province. Here we propose and apply "a multi-evidence framework for interpreting regional air quality dynamics in Qinghai Province", integrating spatiotemporal characterization, Morlet wavelet-based periodicity diagnosis, machine-learning-based machine-learning prediction, and interpretability-based mechanism inference, together with CMIP6-SSP scenario projection. The results show: (1) Qinghai's AQI exhibits pronounced spatial heterogeneity during 2014-2023, with persistently higher levels in the eastern urban corridor, lower levels in the southern plateau, and marked interannual variability in the Haixi region. (2) Wavelet analysis confirms a dominant annual-scale periodicity for AQI and major pollutants, indicating strong seasonal forcing with episodic strengthening in some years. (3) Among multiple candidate models, the best-performing boosting model provides robust prediction and serves as the basis for interpretation. (4) By jointly using SHAP attribution and interaction interpretation with two-dimensional partial dependence analysis, we identify temperature as the primary driver, with wind speed and precipitation acting mainly as conditional regulators; the responses are distinctly nonlinear and reveal clear "high-risk windows" under specific joint meteorological ranges. (5) SSP-based projections suggest that spatial hotspots are likely to persist; particulate matter and primary gaseous pollutants tend to decline or stabilize under mitigation-oriented pathways, while ozone increases emerge as an increasingly important constraint, particularly under higher-forcing scenarios. These findings support climate-sensitive, region-targeted strategies combining clean-energy transition, coordinated multi-pollutant control with an ozone-oriented focus, and sustained management of persistent hotspot regions.