Na\xefve Bayes Classifier for Debris Flow Disaster Mitigation in Mount Merapi Volcanic Rivers, Indonesia, Using X-band Polarimetric Radar

Abstract

Debris flow triggered by rainfall that accompanies a volcanic eruption is a serious secondary impact of a volcanic disaster. The probability of debris flow events can be estimated based on the prior information of rainfall from historical and geomorphological data that are presumed to relate to debris flow occurrence. In this study, a debris flow disaster warning system was developed by applying the Naïve Bayes Classifier (NBC). The spatial likelihood of the hazard is evaluated at a small subbasin scale by including high-resolution rainfall measurements from X-band polarimetric weather radar, a topographic factor, and soil type as predictors. The study was conducted in the Gendol River Basin of Mount Merapi, one of the most active volcanoes in Indonesia. Rainfall and debris flow occurrence data were collected for the upper Gendol River from October 2016 to February 2018 and divided into calibration and validation datasets. The NBC was used to estimate the status of debris flow incidences displayed in the susceptibility map that is based on the posterior probability from the predictors. The system verification was performed by quantitative dichotomous quality indices along with a contingency table. Using the validation datasets, the advantage of the NBC for estimating debris flow occurrence is confirmed. This work contributes to existing knowledge on estimating debris flow susceptibility through the data mining approach. Despite the existence of predictive uncertainty, the presented system could contribute to the improvement of debris flow countermeasures in volcanic regions.