Prediction of spatial distribution of debris-flow hit probability considering the source-location uncertainty

TL;DR

This paper presents a method combining logistic regression and numerical simulation to predict debris-flow hit probability spatially, considering source-location uncertainty, for improved risk assessment and real-time hazard prediction.

Contribution

It introduces a novel integrated approach using logistic regression and simulation to accurately predict debris-flow distribution at high resolution, accounting for source uncertainty.

Findings

Achieved 1-m resolution debris-flow probability mapping.

Developed a real-time hazard prediction system.

Validated the model with actual disaster records.

Abstract

Prediction of the extent and probability of debris flow under rainfall conditions can contribute to precautionary activities through risk quantification. To this end, quantifying the debris-flow risk against rainfall involves three components: predicting the debris-flow initiation locations under rainfall conditions, setting appropriate physical parameters related to debris-flow transportation, and evaluating the affected area using numerical simulation. In this study, we developed a logistic regression method that includes rainfall and topographic parameters as explanatory variables to quantify the probability of debris-flow initiation in an actual area with disaster record. Moreover, an objective parameter-set selection was introduced by evaluating the agreement between the simulation results with multiple parameters and the erosion/deposition area determined using the aerial light detection and ranging difference data after debris flow. Finally, by combining these results, we conducted a predictive debris-flow transport simulation using the initiation datasets generated by the logistic model. Therefore, the spatial distribution using the probability of the effects of debris-flow, which can be applied for risk quantification and evacuation optimization, was successfully obtained at 1-m resolution. Furthermore, a real-time hazard probability prediction system could be developed based on the presented simulation cost.