Abstract:
Streamflow prediction serves as an important tool in water resource management to mitigate disasters and managing water to fulfil people’s needs. In field of hydrology, streamflow prediction is commonly done using hydrological models, i.e. a rainfall-runoff model. The application of data-driven models have also been developed in recent years for predicting streamflow. Both type of models have different approach and have their own strengths and shortcomings. In this research, hydrological model and data-driven models are used to predict monthly discharge in Pemali River Basin, Indonesia. The hydrological model used is a rainfall-runoff model NRECA. The data-driven models used are Seasonal Autoregressive Integrated Moving Average (SARIMA) and optimized Support Vector Machine (SVM). Model performance is assessed with different performance metrics and evaluated by Flow Duration Curve (FDC). The data-driven models, as an alternative approach for predicting streamflow, are able to give prediction result with decent accuracy. Furthermore, the sophisticated Artificial Intelligence (AI) technique in optimized SVM method have better accuracy than the conventional time series model SARIMA. The optimized SVM method, however, needs proper input combination in order to enhance prediction accuracy. In this case, the best input combination for optimized SVM method are rainfall and potential evapotranspiration (PET). The main advantage data-driven models is the ability to forecast future values through time series analysis, where monthly discharge prediction are based solely on previous values. On the contrary, the main disadvantage of data-driven models is the incapability of describing basin’s characteristic since no hydrological processes are described within the process of predicting monthly streamflow. Ultimately, model selection depends on the needs and purpose of prediction.
