A "consensus" real-time river flow forecasting model for the Blue Nile River

The efficacy of using a consensus real-time river flow-forecasting model for the Blue Nile River is investigated. The selected consensus model combines the river flow forecasts of two individual multiple-input single-output river flow routing models, both operating in simulation non-updating mode, the first being a non-parametric linear storage model and the second having the parametric structure of a multi-layer feed-forward neural network. The upstream inflow to the Blue Nile and the outflows of its two major tributaries are used as inputs to both models in order to provide the simulation-mode river flow forecasts just upstream of Khartoum, the capital city of Sudan. The weighted average method (WAM) is used to combine the simulation-mode forecasts of these two models. The consensus real-time river flow forecasts are obtained by updating the combined simulation-mode forecasts using an autoregressive (AR) model error updating procedure. Disappointingly, the results show that the performance of the consensus model, operating in the simulation mode, is not different from that of the best individual model, i.e. that the linear model is given practically zero weight in the consensus model. However, significant improvements in the forecasting performance are obtained after updating the simulation-mode consensus forecasts.