Rainfall-Runoff Model(s) Performance in the Blue Nile River

Abstract

This paper presents the performance of some tested rainfall-runoff models to forecast river flows of the Blue Nile catchment up to Eddeim. Flow forecasting is an important step in river basin management in particular and water resources management in general. River flow models are used as components in actual flow forecasting schemes. They are also used in providing for efficient operation of storage reservoirs. Usually, flow forecasts are obtained in real time by transforming the input into a discharge using models. These forecasts may subsequently be modified or updated in accordance with the errors observed in the previous forecasts up to the time of making the new forecast. The system analysis or black box approach depends on a prior assumption of flexible linear and time invariant relationship the expression of which can be obtained by the application of systems analysis approach to records. The conceptual model provides an alternative approach in which the input-output transformation goes through a series of steps. In this study, systems and conceptual modelling techniques are applied to the Blue Nile catchment up to Eddeim of the Ethiopian high lands. The models were applied in non-parametric and parametric forms. Parameter optimisation is carried out by ordinary least squares, Rosenbrock, Simplex and genetic algorithm. The areal rainfall which is the main input to these models was estimated using arithmetic mean. However, attempts to estimate the areal rainfall by the Thiesen polygon method was made but the improvement in the model performance can not justify the amount of work involved in making Thiesen’s estimate. It is shown that the simple assumption of linearity is not adequate in modelling the rainfall runoff transformation. However, in the Blue Nile catchment, which exhibits a marked seasonal behaviour, good results were obtained with Linear Perturbation Model (LPM) which involves the assumption of linearity between the departures from seasonal expectations in input and output series. The application of the GFFS (collection of systems and conceptual models) software proved to be possible with variable efficiencies. The LPM in non-parametric or parametric form, the LVGF model, the ANN and the SMAR model can be used to forecast (reproduce) Ed-deim flows with high performance. Within the range of the tested models LPM was found to be the best candidate model that can forecast the flows under a wide range of conditions of marked seasonality by accounting for more than 90% of the initial variance.