Integrated approaches for flash flood susceptibility mapping: spatial modeling and comparative analysis of statistical and machine learning models. A case study of the Rheraya watershed, Morocco

Abstract

Flash floods are highly destructive disasters, posing severe threats to lives and infrastructure. In this study, we conducted a comparative analysis of bivariate and multivariate statistical models and machine learning to predict flash flood susceptibility in the flood-prone Rheraya watershed. Six models were utilized, including frequency ratio (FR), logistic regression (LR), random forest (RF), extreme gradient boosting (XGBoost), K-nearest neighbors (KNN), and naïve Bayes (NB). We considered 12 flash flood conditioning variables, such as slope, elevation, distance to the river, and others, as independent variables and 246 flash flood inventory points recorded over the past 40 years as dependent variables in the modeling process. The area under the curve (AUC) of the receiver operating characteristic was used to validate and compare the performance of the models. The results indicated that distance to the river was the most contributing factor to flash floods in the study area. Moreover, the RF outperformed all the other models, achieving an AUC of 0.86, followed by XGBoost (AUC = 0.85), LR (AUC = 0.83), NB (AUC = 0.76), KNN (AUC = 0.75), and FR (AUC = 0.72). The RF model effectively pinpoints highly susceptible zones, which is critical for establishing precise flash flood mitigation strategies within the region.