Abstract: The disaster-forming environment of post-earthquake debris flows in mountainous areas exhibits staged evolution. Traditional debris flow susceptibility assessments mostly rely on random sampling or cross-validation, which is difficult to test the inter-temporal prediction ability and transferability of models. To solve this problem, this paper introduces a temporal-extrapolation validation framework to carry out dynamic susceptibility assessment of post-earthquake debris flows. Taking Yingxiu Town, Wenchuan County, Sichuan Province as the study area, three clustered debris flow events (the “August 13” event in 2010, the “July 10” event in 2013, and the “August 20” event in 2019) were selected to construct an event-scenario susceptibility index system including stable factors and time-varying factors. The Random Forest model was used, and key factors were screened through feature importance. A decision threshold optimization strategy integrating recall and precision was proposed. The model was trained with samples from 2010 and 2013, and extrapolated to the 2019 event for independent validation. The results show that after threshold optimization, the model performance on the test set is: Re =0.90, Pre=0.41, F1 = 0.56. Compared with the original threshold (Pre = 0.33, F1 = 0.48), the Pre is increased by 24 % and F1 by 17 % after calibration. This strategy effectively improves the reliability of early warning information and can significantly reduce unnecessary early warnings and resource consumption. In addition, the spatial distribution of high and very high susceptibility zones varies significantly over time, and actual debris flow are mostly located in such watersheds. The susceptibility assessment framework proposed in this study provides a feasible technical solution for cross-temporal identification of post-earthquake debris flows, and can provide a reference for dynamic susceptibility assessment of post-earthquake debris flows.