Identification of Critical Flood Prone Areas in Data-Scarce and Ungauged Regions: A Comparison of Three Data Mining Models

Abstract

Flood is one of the most devastating natural disasters with socio-economic consequences. Thus, preparation of the flood prone areas (FPA) map is essential for flood disaster management, and for planning further development activities. The main goal of this study is to investigate new applications of the evidential belief function (EBF), random forest (RF), and boosted regression trees (BRT) models for identifying the FPA in the Galikesh region, Iran. This research was conducted in three main stages such as data preparation, flood susceptibility mapping using EBF, RF, and BRT models and validation of constructed models using receiver operating characteristic (ROC) curve. At first, a flood inventory map was prepared using documentary sources of Iranian Water Resources Department (IWRD) and extensive field surveys. In total, 63 flood locations were identified in the study area. Of these, 47 (75%) floods were randomly selected as training/model building and the remaining 16 (25%) cases were used for the validation purposes. The flood conditioning factors considered in the study area are altitude, slope aspect, slope angle, topographic wetness index, plan curvature, geology, landuse, distance from rivers, drainage density, and soil texture. Subsequently, the FPA maps were prepared using EBF, RF, and BRT models in a GIS environment. Finally, the results were validated using ROC curve and area under the curve (AUC) analysis. From the analysis, it was seen that the EBF (AUC = 78.67%) and BRT models (AUC = 78.22%) performed better than RF model (AUC = 73.33%). Therefore, the resultant FPA maps can be useful for researchers and planner in flood mitigation strategies.