Abstract: Objectives: Debris flows are typical rapid-onset geohazards in mountainous regions, and a scientific characterization of their spatiotemporal risk dynamics and driving mechanisms is of great importance for disaster prevention and mitigation. However, the evolution mechanisms of debris-flow risk remain insufficiently understood. This study aims to develop a dynamic debris-flow risk assessment framework that integrates watershed-based units with interpretable machine-learning analysis to elucidate the spatiotemporal evolution and driving forces of debrisflow risk. Methods: A dynamic debris-flow risk assessment method coupling watershed units and SHAP (Shapley Additive Explanations) interpretability analysis is proposed and applied to Tanchang County, Gansu Province, China. First, a debris-flow inventory was constructed, watershed units were delineated, and potential conditioning factors were screened and preprocessed. Subsequently, four models—Random Forest (RF), Logistic Regression (LR), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost)—were employed to quantitatively evaluate debris-flow risks for four periods (2005, 2010, 2015, and 2020), and the corresponding spatiotemporal evolution patterns were analyzed. Finally, the SHAP method was introduced to quantitatively evaluate the marginal contributions and interaction effects of driving factors, thereby revealing the intrinsic mechanisms governing the spatiotemporal differentiation and dynamic evolution of debris-flow risk. Results: RF model achieved the best predictive performance, with an Area Under the Curve (AUC) value exceeding 0.94, significantly outperforming the other models. The spatial pattern of debris-flow risk in Tanchang County exhibited an accelerated bipolar aggregation trend characterized by the expansion of high-risk areas and the contraction of extremely low-risk areas, with the proportion of extremely high-risk zones increasing from 23.84% in 2005 to 26.31% in 2020, while that of extremely low-risk zones decreased from 35.45% to 33.21%. The driving mechanisms of debris-flow risk displayed pronounced stage-dependent characteristics, with anthropogenic disturbances (e.g., road construction and economic development) dominating in 2010, a re‑enhanced fundamental role of natural factors in 2015, and a relatively balanced synergistic effect of natural and human factors emerging by 2020. Conclusions: This study advances the understanding of the spatiotemporal evolution of debris-flow risk in mountainous regions and clarifies the stagedependent transitions of its driving mechanisms. The proposed framework provides robust methodological support for dynamic risk assessment and targeted prevention and mitigation of debris-flow hazards in mountainous areas.