Developing an Adaptive Neuro-fuzzy Model to Predict the Maximum Daily Discharge Using 5-day Cumulative Rainfall

Rainfall is one of the factors involved in increasing soil moisture. Soil moisture, in turn, is a key parameter in the rise and fall of water in the soil which plays an important role in the rainfall-runoff process. It, therefore, requires to be carefully investigated in order to determine its effect on peak flood discharge. One method commonly used for this purpose is the CN-NRCS (curve-number method). Based on this approach, the sum of rainfalls during the 5 days preceding the flood event is taken to represent the soil moisture conditions prior to the event. Given the fact that natural phenomena are always associated with different degrees of uncertainty due to the involvement a multitude of factors, an efficient method for investigating their behavior is the Adaptive Neuro-Fuzzy Intelligent System (ANFIS). Here, we used ANFIS for determining the effect of rainfalls over the five days prior to the flood event in order to predict the maximum daily flood discharge. The model employed the two training algorithms of Back Propagation and Hybrid, which were then tested using different statistical tests and the results were analyzed for each model. The results indicate that the hybrid method outperformed the back propagation method. The best correlation coefficient of the 5-day model was 0.985 and the RMSE (Root Mean Squared Error) was 0.162 in the hybrid method.