Short-term Forecast of Inflows in a Reservoir using Seasonal ARIMA and ANN

Water resources planning and management, involves as one of the most important aspects, forecasting the inflows in the reservoirs with the highest possible accuracy. The aim of this paper is to forecast the inflow for the next 1, 2, 3, 5, 365 days, based on historical recorded data. Two different methods were applied: one belongs to the classical time series methods, autoregressive integrated moving average (ARIMA), and the second one is part of artificial intelligence, Artificial Neural Network (ANN), both methods having toolboxes and predefined functions developed in MATLAB and being suitable for forecasting non-stationary time series. Thus, the capabilities of seasonality components integrated in ARIMA method and the ability of ANN to give good results were explored using MATLAB. Inflow values recorded at a large reservoir in Romania were used for calibrating the model. Coefficient of determination, root mean square error (RMSE) and Nash Sutcliffe efficiency have been calculated to estimate the performance of the models. ARIMA, with the seasonal component removed, is well-suited to short-term forecasts. By increasing the forecast time horizon, the accuracy of the predicted values decreases. ARIMA can be used to forecast the inflows for one year prediction horizon but the accuracy is low. For one step ahead forecasting, both methods provide results in good agreement with the original data.