Comparative Analysis of Multi-Linear Lag Cascade Model with Advanced Hydrological Techniques for Flood Prediction in Zarineh Rud River, Iran

The study focuses on the detailed methods of flood prediction in the ZarinehRud River within Urmia Lake. It compared five different methods: the Multi-Linear Lag Cascade model, Saint-Venant equations, and three soft computing methods, namely Artificial Neural Networks, Adaptive Neuro-Fuzzy Inference System, and Support Vector Machines. In this study, the data of 2022 flood recorded at Alasaql and Safakhaneh stations were used. The performances of models were then evaluated in terms of various statistical criteria such as the Nash-Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Peak Flow Ratio (PFR), and Percent Error in Peak (PEP). In general, it was found that the soft computing techniques, in particular ANN and ANFIS, are representing the best performance with NSE values of 0.938 and 0.935, respectively. Similarly, the MLLC model showed competitive performance with a value of NSE equal to 0.922 but with much lower computational time. The Saint-Venant model was somewhat less accurate, with an NSE value of 0.901 but with higher robustness against input uncertainty. For all models, results are better for the high flow range that is of importance for flood forecasting. Sensitivity analysis has shown that soft computing methods are more sensitive to input data errors than physically based Saint-Venant. This work underlines several critical trade-offs, when optimizing model accuracy, computational efficiency, and robustness to uncertainty for flood prediction. The results highlight the importance of considering different performance criteria and particular prerequisites of the chosen application when selecting an appropriate flood prediction model for the similar hydrological systems.