Attribution, zonation, and simulation of spatiotemporal distribution of negative oxygen ions in mountainous megacity: A case study of Chongqing, China

Negative oxygen ions (NOI) are key indicators of urban ecological quality. However, NOI dynamics in mountainous megacities remain poorly understood. A key challenge lies in translating complex environmental observations into interpretable knowledge for environmental planning rather than prediction alone. Therefore, we integrated few-shot learning, explainable AI (e.g., SHapley Additive exPlanations), and causal inference within an explainable GeoAI framework to identify NOI drivers and predict spatial patterns in Chongqing, China. Results revealed pronounced spatiotemporal heterogeneity, with low concentrations in southwestern urbanized areas and high values in forested mountains and river valleys of northeastern Chongqing. Attribution analysis identified elevation, curvature, relative humidity (RH), NO₂, and aerosol optical depth (AOD) as dominant drivers with model-derived nonlinear response thresholds (AOD > 273.67; NO₂ > 15.42; RH > 90.95%), beyond which predicted NOI responses changed markedly. Additionally, DML-based analysis estimated a positive treatment effect of middle elevations on NOI (+0.179 per unit, p < 0.001). Scenario simulations demonstrated that humidity enhancement and the integrated scenario yielded stronger responses than pollution mitigation or greening alone, and achieved a mean ΔNOI of 130.12 cm −3 with 76.25% of the area displaying gains. However, the integrated scenario increased inequality despite maximizing population-weighted benefits (106.99 cm −3 ), favoring sparsely populated mountainous regions (Gain = −28.76 cm −3 , Gini = 0.0835). These findings underscore the capacity of explainable GeoAI for adaptive urban planning by linking environmental response interpretation, intervention evaluation, and spatial equity assessment.