A surrogate-assisted preliminary screening workflow for hypothetical fixed-location check-dam height scenarios in debris-flow mitigation

Abstract. Early-stage debris-flow mitigation planning often requires comparison of multiple countermeasure scenarios before detailed engineering design information is available. This study presents a surrogate-assisted preliminary screening workflow for comparing hypothetical fixed-location check-dam height scenarios using simplified proxy metrics. A depth-integrated debris-flow simulation model was applied to a terrain dataset from the area affected by the 2021 Atami debris-flow disaster in Japan. The analysis is not intended as a forensic reproduction of the event or as an assessment of existing or planned facilities. In total, 4,877 simulation cases were generated by varying the heights of six candidate check dams using Latin hypercube sampling. The simulation results were used to train surrogate models relating dam-height scenarios to the maximum downstream debris-flow depth, used here as a hazard-intensity proxy, while the sum of dam heights was used as a height-based proxy for construction-related effort. A multilayer perceptron provided the best coarse-grained approximation and enabled repeated surrogate-based searches under different trade-off weights. The screened scenarios formed a limited number of representative trade-off regimes. However, physics-based re-evaluation showed non-negligible discrepancies between surrogate-predicted and simulated depths, indicating that the workflow should be used for candidate screening rather than final quantitative design evaluation.