Application of Artificial Neural Network Model to Forecast Runoff for Waikato River Catchment

As we know that over the past decade or so the Artificial Intelligence (AI) techniques (e.g. ANN - Artificial Neural Network & FIS - Fuzzy Interference System) have been used as an alternative modelling tools in water resources management studies. Runoff generated from a catchment as a result of a rainfall event is a very complex hydrological process as it depends on climatological (i.e. rainfall depth, duration and intensity, etc.) and geographical (i.e. soil type, infiltration rate, evapotranspiration, etc.) factors of the catchment. The present study is about the application of Artificial Neural Network (ANN) model to forecast runoff from the Waikato River catchment areas of New Zealand. Similar to other modelling approaches, successful application of ANN is also dependant on the selection of appropriate input factors. To investigate this, the study applied three different approaches for the selection of appropriate input vectors to be used for the ANN model. The study demonstrated that ANN can successfully forecast the runoff generated from a catchment using antecedent rainfall and runoff data series identified on the basis of cross-correlation and auto-correlation coefficients. The ANN models developed using three approaches (i.e. sequential, pruned and non-sequential time series) were able to predict runoff generated from the Waikato River catchment using antecedent rainfall/runoff data. The study showed that the ANN models were sensitive to the selection of appropriate input vector. The ANN model developed using the non-sequence approach performed well, and gave the highest R2 and NSE values (i.e. 97-98 %) during the validation and testing phases of this modelling exercise.