Flood Hazard Potential Evaluation Using Decision Tree State-of-The-Art Models

Flash floods are becoming an increasingly recurrent natural hazard globally. Romania has been devastated severely by heavy floods. Storms that cause heavy floods in mountain and hilly river catchments are generally triggered by heavy rainfall. In addition to temporal flash flood forecasting, models for recognizing risky places may greatly contribute to disaster risk reduction and policymaking. The need for more precise modes has arisen as a result of flash flood hazard mapping. Hence, the present research proposes four state-of-the-art hybrid models for the simulation of flood hazard potential in the basin of the Trotus River in Romania. The next four algorithms were used to obtain the desired results: Forest by Penalizing Attribute – Weights of Evidence (Forest-PA-WOE), Best First Decision Tree – Weights of Evidence (BFT-WOE), Alternating Decision Tree – Weights of Evidence (ADT-WOE), and Logistic Regression – Weights of Evidence (LR-WOE). As input data, we used 12 flood-influencing factors and 172 flood locations. This sample was split into training (70%) and validating (30%) datasets. Using the training points, for each class/category of flood predictors, coefficients were calculated by using the WOE. The methodological workflow was implemented with the help of Excel, Weka, SPSS, and ArcGIS software. Forest-PA-WOE achieved the best training procedure results, with an accuracy of 0.942, a k-index of 0.883, a specificity of 0.965, and a sensitivity of 0.921. The models revealed that a percentage between 16.22% and 25.84% of the Trotuş River basin has high and very high values of flood potential. In terms of flood hazard maps, it was also observed that Forest-PA-WOE obtained the most accurate result with an AUC-ROC Curve greater than 0.953. The findings of this research may substantially map the risky areas and further aid watershed managers in limiting and remediating flood damage in the data-scarce regions