ANFIS and NNARX based rainfall-runoff modeling

Modeling of non-linearity and uncertainty associated with rainfall-runoff process has received a lot of attention in the past years. Recently artificial intelligence techniques are used for hydrological time series modelling. Earlier studies showed this approach is effective, still there are concerns about how these techniques perform efficiently to predict the run-off with high standard of accuracy. To this end, this paper explores the ability of two artificial intelligence techniques, namely neural network auto regressive with exogenous input (NNARX) and adaptive neuro-fuzzy inference system, to model the rainfall-runoff phenomenon effectively from antecedent rainfall and runoff information. Specifically, to illustrate applicability of these techniques, two year (1994-1995) rainfall-runoff data from Brue catchment of The United Kingdom were used. The models having various input structures were constructed and the best structure was investigated with help of the proposed technique, called gamma test. Training data length selection and best input combination were carried out prior to modeling with help of gamma test. The performance of the ANFIS model in training and testing sets were compared with that of NNARX model with help of several statistical parameters. The results of the study have shown that both ANFIS and NNARX could work efficiently in rainfall-runoff modeling and can provide high accuracy and reliability in runoff prediction.