Modeling erosion and reservoir impacts in data sparse regions in South Africa

Reservoir sedimentation is a major problem in South Africa that impacts drinking water supplies, irrigation, recreation, hydropower production, and flood control. Although South Africa has been actively involved in research and monitoring of erosion and sedimentation, an ongoing challenge with assessing reservoir sedimentation is that many basins are ungauged and information required to guide policy and decision making is not available. Thus, there is a need to better understand methods to estimate sediment yields in these areas. Hydrological models can provide detailed spatial and temporal estimates of sediment fluxes. However, when modeling data sparse regions, it can be challenging to find a balance between model complexity and data availability to not only optimize simulation performance, but also to not overparameterize the model.

In this study, we used the open source, semi-distributed Hydrological Predictions for the Environment (HYPE) model to investigate how model routines with different data requirements affected simulations of long-term reservoir sediment yields within the uMngeni River basin in South Africa. Specifically, we assessed the impacts on model performance due to selection of the erosion routine and the stream sedimentation/resuspension method. An ensemble of models for the uMngeni basin was calibrated to sediment yields calculated from reservoir surveys and suspended sediment sampling data for seven reservoirs. In total, the ensemble consisted of 12 model setups representing unique combinations of two erosion routines (Morgan-Morgan-Finney model and HBV-SED index method), three HBV-SED seasonal erosion correction methods (no adjustment, values from earth observation indices, and calibrated values), and three stream sedimentation/resuspension routines (based on bankfull flow, based on bankfull flow with calibrated correction factor, and the Bagnold equation). Upon completion of the ensemble calibration, a matrix of model fit statistics was created to evaluate how sediment simulation performance varied according to model routine complexity and input data requirements. Erosion rates, sources of sediment fluxes, and sediment yields were also compared between the model setups to provide guidance on ranges of potential reservoir impacts. Results from this study can inform the selection of modeling techniques during the development of sediment flux models for regions with limited data.