Assessment of the Contribution of Geo-environmental Factors to Flood Inundation in a Semi-arid Region of SW Iran: Comparison of Different Advanced Modeling Approaches

Floods are a hazard for artificial structures and humans. From natural hazard management point of view, present the new techniques to assess the flood susceptibility is considerably important. The aim of this research is on one hand to evaluate applicability of different machine learning and advanced techniques (MLTs) for flood susceptibility analysis and on the other hand to investigate of the contribution of geo-environmental factors to flood inundation in a semi-arid part of SW Iran. Here, we compare the performance of six modeling techniques namely random forest (RF), maximum entropy (ME), multivariate adaptive regression splines (MARS), general linear model (GLM), generalized additive model (GAM), and classification and regression tree (CART)for first time to spatial predict the flood prone-area at Tashan Watershed, southwestern Iran. In the first step of study, a flood inventory map with 169 flood events was constructed through field surveys. These flood locations were then spatially randomly split into train, and validation sets with two different proportions of ratio 70 and 30%. Ten flood conditioning factors such as landuse, lithology, drainage density, distance from roads, topographic wetness index (TWI), slope aspect, distance from rivers, slope angle, plan curvature and altitude were considered in the analysis. In addition, learning vector quantization (LVQ) was used as a new supervised neural network algorithm to analyse thevariable importance. The applied models were evaluated for performance appliyng the area under the receiver operating characteristic curve (AUC). The result demonstrated that CART had the AUC value of 93.96%. It was followed by ME (88.58%), RF (86.81%), GAM (81.35%), MARS (75.62%), and GLM (73.66%).