Prediction of Spatial Distribution of Debris-Flow Hit Probability Considering the Source-Location Uncertainty

Abstract

Debris-flow affected area is typically predicted using runout simulations, often estimating the hydrograph from rainfall conditions. However, rainfall is rarely considered when predicting initiation locations, which influence the occurrence number and location. This study proposes a hybrid method combining statistical source-location prediction based on rainfall conditions and runout simulations inputting the predicted source locations. First, logistic regression is used to predict the spatial probability of debris-flow initiation with rainfall as an input. Next, Monte Carlo simulation based on the initiation location generated from the rainfall-based probability yields the spatial distribution of the debris-flow hit probability. Simulation parameters are systematically determined in advance based on topographic change obtained via aerial LiDAR observations. This method was successfully employed to estimate the spatial distribution of the debris-flow hit probability at 1-m resolution for a debris-flow disaster that occurred in Hiroshima prefecture, Japan, using rainfall data obtained by radar. The simulation time indicated that hit probability can be issued prior to the event for early warning, owing to the adequate lead time of rainfall forecasts and recent developments in computational technology. The hit probability obtained in this study can be also applied to risk quantification based on rainfall conditions.