Evaluating the performance of a deterministic daily rainfall-runoff model in a low-flow context

The characterization of a stream's low-flow regime is required for ecological purposes, water quality studies and various other water projects. If observed stream flow records are insufficient, low-flow characteristics may need to be estimated from simulated daily stream flow time-series. The model employed should conceptualize low-flow generation mechanisms and surface-subsurface interactions adequately. The ability of the model to simulate low-flow regimes may be assessed by means of various low-flow analysis techniques. This paper illustrates the approach using the example of the physically based, semi-distributed VTI daily rainfall-runoff model. The model has been applied to five perennial headwater catchments in South Africa, which are located in different parts of the country, represent different physiographical environments and are characterized by different baseflow responses. The model simulations are evaluated in terms of both conventional goodness-of-fit criteria and several low-flow measures such as recession characteristics, baseflow volumes, flow duration curves and continuous low-flow events below specified threshold discharges. For all the catchments considered the model has been found to perform successfully in terms of conventional fit statistics and flow duration curves. However, its ability to reproduce recession characteristics and continuous low-flow spells appears to be less satisfactory. This suggests that daily model simulations should be evaluated by low-flow criteria, which are frequently ignored in water resource assessment practices. (C) 1998 John Wiley & Sons, Ltd.