Artificial neural network and multi-criteria decision-making models for flood simulation in GIS: Mazandaran Province, Iran

Abstract

Flood is one of the most destructive natural disasters globally and is a concern due to its high vulnerability. In this study for identification of flood susceptible areas, artificial neural network (ANN) and Multi-Attributive Border Approximation Area Comparison (MABAC) combined with Weights of Evidence (WoE) and Analytical Hierarchy Process (AHP) Models were used in Mazandaran province, Iran. MABAC method was used for the first time to evaluate the flood-prone areas in this study, and Attempts have been made for evaluate the performance of this new method by comparing with ANN model. The output of the neural network was discharge values in hydrometric stations. Using Geographic Information System (GIS) with eight effective factors including rainfall, distance from rivers, slope, soil, geology, elevation, drainage density, and land use, a flood model developed. Three precision parameters containing $${R}^{2}$$ , RMSE and MAE were applied to show the performance of the ANN model which yielded the values of 0.89, 0.0024 $${m}^{3}/s$$ , and 0.0018 $${m}^{3}/s$$ , respectively for testing data. The verification results indicated satisfactory agreement between the predicted and the real hydrological records. Also, based on flood inventory map and using the area under receiver operating curve, predictive power of the MABAC-WoE-AHP model was evaluated. The AUC value for prediction rate of this model was 86.1% which indicates the very good accuracy in predicting flood-prone areas. Comparison of flood susceptibility maps for ANN and MABAC-WoE-AHP models showed the good agreement between two models, that clarifies the efficiency of the new proposed method for future preventive measures.