Flood Susceptibility Mapping Using Random Forest Machine Learning and Generalized Bayesian Linear Model

Today, the phenomenon of flooding is one of the most complex hazardous events that, more than any other natural disaster, causes deaths and finances every year in different parts of the world. Therefore, flood susceptibility mapping is the first step in a flood management program. The purpose of this study was to identify flood susceptible areas using two methods of random forest (RF) and Bayesian generalized linear model (GLMbayesian) machine learning in the Tajan watershed in Mazandaran province, Sari. Past flood distribution maps were prepared to predict future floods. Of the 263 flood locations, 80% (210 flood locations) was used for modeling and 20% (53 flood locations) was used for validation. Based on previous studies and surveying of the study area, 13 conditional factors were selected for flood zoning. The results showed that three factors of elevation (21.55), distance from the river (15.28) and slope (11.18) had the highest impact on flood occurrence in the study area, respectively. The results also showed that the AUC values for RF and GLMbayesian models were 0.91 and 0.847, respectively, indicating the superiority of the RF model and the accuracy of this model in flood susceptibility mapping in the study area. The highest flood susceptibility area in the RF model is in the very low class and the high class in the GLMbayesian model.