A comparative study in long‐term river flow forecasting models

Abstract

River flow forecasting experiments are carried out for rivers located in the Karun Basin and its sub‐basins situated in the Southwestern Iran, because of the potential importance of these rivers for supplying relatively large amounts of water. More specifically, multiple linear regressions (MLR), autoregressive integrated moving average (ARIMA), deseasonalized autoregressive‐moving average (DARMA), and Thomas‐Fiering (TF) models were fitted to monthly, bimonthly, and seasonal river flow series. One‐step‐ahead forecasts for the test portion of the time series were generated using the selected set of candidate models. Forecasting performance of the models was compared based on the mean absolute error, root mean square error, normalized mean bias error and correlation coefficient between observed and forecasted values. The results indicate that the ARIMA model is a more reliable model for monthly river flow forecasting applications in the basins under study. For bimonthly and seasonal river flow forecasting, MLR models perform better.

Keywords:

• Karun Basin
• regression analysis
• river flow forecasting
• time series models