Novel Bayesian Additive Regression Tree Methodology for Flood Susceptibility Modeling

Identifying areas prone to flooding is a key step in flood risk management. The purpose of this study develop and present novel susceptibility model based on Bayesian Additive Regression Tree (BART) methodology. predictive performance the new assessed via comparison with Naïve Bayes (NB) Random Forest (RF) methods that were previously published literature. All models tested real case Kan watershed Iran. following fifteen climatic geo-environmental variables used as inputs into all models: altitude, aspect, slope, plan curvature, profile drainage density, distance from river road, stream power index (SPI), topographic wetness (TPI), position curve number (CN), land use, lithology rainfall. Based existing field survey other information available for analyzed area, total 118 locations identified potentially flooding. data divided two groups 70% training 30% validation models. receiver operating characteristic (ROC) parameters evaluate accuracy area under (AUC) BART (86%) outperformed NB (80%) RF (85%) Regarding importance input variables, results obtained showed location’s altitude are most important assessing susceptibility.