Climate change and urban development are among the most significant drivers of hydrological shifts in urban watersheds; however, disentangling their relative contributions to the changing rainfall–runoff relationship remains a major challenge in urban hydrology. This study aims to analyze the evolution of the rainfall–runoff relationship in the Kan River Basin, Tehran, and evaluate the relative roles of climate variability and urban expansion using an Explainable Artificial Intelligence (XAI) framework. To this end, hydrological and climatic datasets were integrated with remote sensing-derived indices and urban development indicators. Following an assessment of temporal trends and system non-stationarity, machine learning models were developed to simulate runoff responses under varying environmental conditions. Subsequently, the SHAP (SHapley Additive exPlanations) method was employed to quantify feature importance and interpret the influence patterns of variables on model outputs.

Moving beyond conventional trend analyses, the results indicate that the rainfall–runoff relationship in the Kan Basin has undergone a non-linear structural decoupling. The primary innovation of this study lies in the weighted attribution of variables through SHAP analysis, revealing how physical development—by altering infiltration thresholds—has significantly increased the basin’s sensitivity to extreme precipitation compared to previous decades. These findings underscore the critical need to revise the design of hydraulic structures based on non-stationary conditions. Nevertheless, the results should be interpreted as model-based evidence rather than definitive proof of causal relationships. The findings of this study offer valuable insights for urban water resources management, assessing the impact of physical expansion on hydrological responses, and designing environmental adaptation strategies.