Flood potential mapping by integrating the bivariate statistics, multi-criteria decision-making, and machine learning techniques

This research aims to determine the flood potential mapping within Golestan Province in Iran, applying six novel ensemble techniques guided by the multi-criteria decision-making (MCDM), bivariate statistics, and artificial neural network methods. The combinations of Combined Compromise Solution (COCOSO), Multi-Attributive Border Approximation Area Comparison (MABAC), and multilayer perceptron (MLP) with Frequency Ratio (FR), and Weights of Evidence (WOE) were then generated. It is noted that this is the first application of COCOSO method in flood susceptibility assessment and its efficiency had not been evaluated before. In this regard, 10 flood influential criteria namely altitude, slope, aspect, plan curvature, distance from rivers, Topographic Wetness Index (TWI), rainfall, soil type, geology, and land use, 240 flood points, and 240 non-flood points were employed for the modeling process, of which 70% of such data were chosen for training and remaining 30% for validating. The accuracy of proposed methods was tested by the area under the receiver operating characteristic (AUROC) curve. MABAC-WOE obtained the largest predictive precision (0.937), followed by MLP-WOE (0.934), COCOSO-WOE (0.923), MABAC-FR (0.921), MLP-FR (0.919), and COCOSO-FR (0.892), respectively. The high accuracy of all proposed models represents their capability in flood susceptibility assessment and can guide future flood risk management in the study location.