Combination of machine learning model (LR-FR) for flash flood susceptibility assessment in Dawuan Sub watershed Mojokerto Regency, East Java

Abstract

The Dawuan Sub-watershed in Mojokerto Regency is a prone area to floods. There were flash floods in this area in 2002 and 2019, which caused casualties and property losses. As one of the mitigation efforts, this study aims to map a flash flood’s susceptibility using the LR-FR combination machine learning technique (logistic regression and frequency ratio). 11 conditioning factors are used to assess landslide susceptibility, namely: slope, aspect, TWI (Topographic Wetness Index), TPI (Topographic Position Index), SPI (Stream Power Index), profile curvature, distance to drainage, rainfall, geological unit, and land use. The results of the flash flood susceptibility mapping show that areas with very high levels of susceptibility have the following characteristics: slope < 8-35°; aspect east and southwest; TWI >16; TPI <(-3,39)-(-0,06); SPI <50-200; profile curvature (-0,001)-0,0; distance to drainage <10-40; rainfall <2000; geological unit Qvwl, Qvlw3, Qvlp3, Qvlp4, Qvwl, Qvf3, Qvf4 and Qvf8; and agricultural land use. The validation results show that the quality of the LR-FR model used has very good quality, as indicated by the AUC value = 0.93.