Evaluating DEM Refinement Strategies and Flow-Depth Threshold Criteria for Debris Flow Simulation in a Data-Limited Catchment

Debris flows can severely damage downstream areas, but predicting their impacts is often hindered by coarse topographic data. This study evaluates whether digital elevation model (DEM) refinement improves debris flow simulations in a data-limited catchment using the Hyper KANAKO system. It also examines whether flow-depth threshold analysis provides a more standardized basis for interpreting simulation results. Three DEM configurations were evaluated using a documented debris flow event in South Korea: (1) incorporation of building elevations, (2) refinement of channel and road topography, and (3) a combination of both approaches. Simulated areas were compared with the observed affected area using receiver operating characteristic (ROC)-based performance measures and confusion matrix indices. Refinement of channel and road topography (DEM configuration 2) yielded the most consistent improvement, with an accuracy of 0.961 and the highest F1 score (82.6%). Flow-depth threshold analysis identified 0.3 m as a practical threshold for delineating deposition-area boundaries across the DEM configurations. When both overprediction and underprediction were considered, Case 2, which applied the ROC-derived, application-specific threshold (Youden’s J cut point), achieved the highest Net Success Index. The proposed workflow offers a potentially transferable approach to debris flow hazard mapping and erosion-control planning in areas where high-resolution topographic data are unavailable.