Thinking inside the box

Using reservoir levels to improve a conceptual rainfall-runoff model in the Umbeluzi River Basin

More Info


The development of a conceptual rainfall-runoff model generally requires calibration to obtain non-observable parameters. This becomes especially difficult in river basins where data are absent, frequently intermittent or inaccurate. Rainfall-runoff models are usually calibrated on discharge data that are derived from stage observations by means of a rating curve. However, this method is error-prone, as it is not able to get around problems such as a changing geometry of the river channel, flows over floodplain and a variable roughness coefficient due to vegetation. Moreover, uncertain stage observations can be a severe source of error as well, because of incorrect placement of the staff gauges or incorrect and infrequent stage readings. Calibration on uncertain discharge data can affect modeled parameters and consequently be cause for erroneous model predictions. This works uses water levels of reservoirs to calibrate a conceptual rainfall-runoff model in the Umbeluzi River Basin. The conceptual model provides the inflow to the reservoir, subsequently reservoir storage is modeled and compared to the observed storage. The proposed method has several advantages over the conventional method. First of all, stage observations are accurately observed because of the social and economic importance of the reservoir. Moreover, the method is well suited to observe floods, as the shape of the reservoir is well confined and its capacity generally more than adequate. It also captures the entire volume of a flood, in contrary to the conventional method where effects of hysteresis can be cause for misconception of the river discharge. An additional advantage is that reservoirs can integrate multiple tributaries, allowing to model tributaries that are not or scarcely gauged. Lastly, this method may have some economic benefits as the water levels and the surface area of the reservoir can in principal be derived by radar altimetry through satellites. This allows for the system to be remotely calibrated. By incorporating irrigation into the hydrological model, the effects of artificial reservoirs on the basin hydrology are assessed. The conceptual model is calibrated with inclusion and exclusion of irrigation and subsequently best parameter sets are compared. Results show that the excess irrigation water that percolates to the ground water has a significant influence on the resulting base flow, such that model parameters have to be compensated when irrigation is excluded. Moreover, a changing crop need due to the varying lengths of dry spells, is cause for variability in irrigation over the years. This makes modeling base flow challenging. It appears that inclusion of irrigation into the hydrological model is the only way to obtain a consistent ground water parameter. This thesis shows a validation approach whereby multiple system components are validated on various independent data-sources. The modeled evaporation is compared to evaporation that is developed with use of satellite data. Both time-series correspond well, which implies that the lumped modeled water storage of the catchment is plausible. In addition, it implicitly validates the amount of transpiration from irrigation, which is included in the actual evaporation. To validate on smaller system components, the water levels of the conceptualized floodplain storage are compared to observed river stage. Although absolute values are not comparable, the emptying and filling of the flood plains harmonizes with the rate of change of observed stage levels. Lastly, the model is tested on a tributary of the catchment, generating reasonable model prediction, despite the poor data quality. The author argues that this way of validation is more valuable than validating on time-series outside the calibration period, as it takes multiple aspects of the system into account. For this study both a lumped and a topography driven model is tested. During the calibration, both models give a similar model performance. However the optimum parameter set of the topography driven mode was better applicable in other catchments. Moreover, the parameters derived where also more realistic. It is therefore that the topography driven model works better for the Umbeluzi catchment.