Abstract
Rapid urban advancement has exacerbated air‑pollution risks and calls for evidence‑based refined urban environmental governance. Based on panel observational data from 33 cities in China’s Yangtze River Economic Belt (YREB) spanning 2014‑2022, this study establishes a three‑tier sequential descriptive‑evidence‑chain framework combining XGBoost‑SHAP interpretable machine‑learning, Moran’s I spatial‑autocorrelation diagnosis, and SAR spatial‑econometric estimation. Moving beyond simple technical combination of existing tools, this workflow enables systematic cross‑comparison between variable predictive performance and spatial‑statistical patterns. Analytical results show that: (1) PM2.5, agricultural fertilizer application, electricity consumption, and NO2 possess high predictive importance for AQI variations; (2) Urban pollution sources exhibit significant spatial dependence, particularly for PM2.5 and electricity consumption; (3) Spatial‑effect decomposition identifies strong positive local statistical associations of PM2.5 and SO2 concentrations with urban AQI, while their indirect spatial‑statistical associations are counter‑intuitively negative. Agricultural‑fertilizer application presents offsetting negative‑direct and positive‑indirect spatial‑statistical associations, whereas electricity consumption, NO2, and natural gas show negligible overall spatial‑statistical associations. This work offers empirical insights to support differentiated and coordinated air‑pollution governance within the YREB.
Keywords:
- Keyword: Air pollution; Machine learning; Extreme Gradient Boosting; Spatial econometrics; Yangtze River Economic Belt
How to Cite:
Wang, X., Liang, K., Fan, C., Appolloni, A. & Wang, H., (2026) “Spatiotemporal Evolution and Spatial Spillover Effects of Urban Air Pollution Sources in the Yangtze River Economic Belt of China”, Journal of Intelligent and Sustainable Systems (JISS) 2(3).