Spatio-temporal variation and determinants of urban air quality in Chinese cities during 2014-2023: an interpretable machine learning analysis.

China launched its clean-air action plans in 2013, which have played a crucial role in driving one of the world's most significant and rapid improvements in urban air quality. However, the relative roles of financial, governance, and structural determinants remain poorly understood. This study developed a national-scale, interpretable machine-learning framework to model and explain air quality dynamics across 273 prefecture-level cities from 2014 to 2023. We used city-level AQI as the target variable and assembled a multidimensional panel of socioeconomic, financial, technological, energy, and governance indicators. We fitted a TPE-tuned XGBoost model and interpreted it using SHAP (global and local attributions) and ALE (non-linear response curves and two-way interactions). Nationally, more than 85% of cities exhibit substantial AQI declines, but improvements are uneven across regions and provinces. SHAP results identify green finance, population density, government intervention, openness, and digital finance as the most influential determinants, with energy consumption, green technology innovation, foreign direct investment, industrial structure, and urbanization also playing important roles. ALE analysis reveals strong nonlinearities: green and digital finance, as well as government intervention, are pollution-increasing at low and intermediate levels but become markedly pollution-reducing beyond critical thresholds, while density and energy use rapidly become strong positive drivers once modest levels are exceeded. Interactions show that green and digital finance yield significant air-quality benefits only when combined with strong governance and less energy-intensive and industry-intensive structures. City-level decompositions further distinguish improving from deteriorating cities in terms of the evolution of these drivers. The findings highlight that sustained governance, well-targeted green and digital finance, and accelerated structural change are mutually necessary to secure durable gains in urban air quality.