<p>Flash floods have often occurred in small-scale scattered areas that often lack hydrological, rainfall and geotechnical data. Under extreme rainfall conditions, evaluating flash flood susceptibility in this region has been a major challenge in current research. Building on this, high-resolution DEM data, combined with the random forest (RF) model optimized by grid search (GS) and a feature selection algorithm, are used to identify small-scale scattered flash floods. At the same time, the prediction effect of the model established by low-resolution and high-resolution DEM data on the flash flood in Taibai Creek small watershed is compared. The results showed that 13 conditioning factors influence the occurrence of flash flood, among which distance to ravine (D2R) is the most important factor affecting the flash flood sensitivity of small watersheds. Evaluating flash floods in small watersheds using high-resolution DEM data combined with the random forest algorithm is feasible. The model demonstrates strong predictive performance, achieving an AUC value of 97.2% in the Taibai Creek small watershed. Low-resolution DEM data lead to inaccurate hazard assessment results. The spatial distribution characteristics of the susceptibility map constructed using high-resolution DEM data are highly consistent with observations in the small watershed. This study helps improve the assessment of geological disasters in small watersheds and addresses the over-identification issue of high-risk areas in previous susceptibility analysis.</p>