Artificial Neural Networks and Fuzzy Inference System

Conceptual models have been often used in short- or mid-term hydrologic forecasting. In this paper a framework for combining two conceptual climatic and hydrologic models is used in order to generate long-lead Ensemble Streamflow Prediction (ESP) of streamflow to Zayandeh-rud reservoir. In the proposed approach, two models based on a Fuzzy Inference System (FIS) for seasonal rainfall forecasts and Artificial Neural Networks (ANNs) for mapping hydroclimatic variables to streamflow data are used. Illusions such as clustering of rainfall and streamflow data, a proper calibration procedure as well as using a stopped training approach in ANN calibration, improve the accuracy of the forecasts. The results of the proposed approach are assessed by various criteria. Further, the results are compared with an ANN-based streamflow forecast, which uses the observed hydroclimatic data in monthly streamflow forecasting. The results show that using the proposed approach has the advantage of generating proper long-lead point and ensemble forecasts, which could be potentially used to reflect the uncertainty of future available water resources.