Predicting flood prone areas using advanced machine learning models (Birjand plain)

Research on flood predicting models is one of the first steps in reducing flood damage and managing future floods in catchments. The aim of this study was to evaluate flood susceptibility in Birjand plain catchment through four machine learning models including support vector machine (SVM), J48 decision tree, random forest (RF) and Adaptive neuro fuzzy inference system (ANFIS). Therefore, in order to implement and validate the mentioned models, a list of flood-prone areas in the study area was prepared (42 flood-prone locations). In addition, 19 hydrogeological, topographical, geological and environmental criteria affecting flood occurrence in the study area were extracted to be used to predict flood susceptibility map. The results showed that the highest accuracy was related to the RF model (0.845) and the lowest accuracy was related to the SVM model (0.791). In addition, the validation of the results using the ROC curve showed that the most accurate values of flood susceptibility belong to the RF model (AUC = 0.958). The results of this study can be used to manage vulnerable areas and reduce flood damage.