A hybrid model for daily flow forecasting

Having a simple structure, the Tank Model has been widely used in modelling the rainfall-runoff process of many watersheds. In this study, the model was slightly modified to make it suitable for forecasting daily discharges. Two basins in Vietnam, namely the Da Nhim and La Nga Basins were considered. The daily discharges were forecast one day ahead using daily rainfall and daily evaporation. It was found that forecast discharges for the smaller basin (Da Nhim) were not as good as those for the larger basin (La Nga). In order to improve the forecasting performance of the Tank Model, a first-order autoregressive model, AR(1) was introduced to represent the error time series. The resulting hybrid model is the combination of the Tank Model and the AR(1) Model. Thus the forecast discharge for any day is obtained as the forecast value obtained from the Tank Model plus the forecast value of the error for that day. Application of the hybrid model to the above two basins was found to provide more accurate forecast values even though the overall performance of the hybrid model largely depends on the efficiency of the Tank Model.