Assessing the performance of techniques for disaggregating daily rainfall for design flood estimation in South Africa.

Abstract

Design Flood Estimation (DFE) and other hydrological modelling methods are used to limit the risk of failure and ensure the safe design of infrastructure and for the planning and management of water resources. The temporal distribution of rainfall has a significant impact on the magnitude and timing of flood peak discharges. Rainfall temporal distributions are therefore an important component of DFE approaches. In order to improve DFE methods which are based on event or continuous simulation rainfall-runoff models, it is generally necessary to use sub-daily time step rainfall hyetographs as input. However, the number of recording raingauges which provide sub-daily timesteps in South Africa is relatively scarce compared to those which provide daily data. Rainfall Temporal Disaggregation (RTD) techniques can be used to produce finer resolution data from coarser resolution data. Several RTD approaches have been applied in South Africa. However, application of RTD approaches locally is relatively limited, both in terms of diversity of approaches and cases of application, compared to those developed and applied internationally. Therefore, a need exists to further assess the performance of locally applied approaches as well update the list of available approaches through inclusion of internationally developed and applied RTD techniques. A pilot study was performed in which selected locally applied and internationally applied approaches were applied to disaggregated daily rainfall data. Some approaches were applied in their original form while others were modified. Temporal distributions of rainfall were represented by dimensionless Huff curves, which served as the basis for comparison of observed and disaggregated rainfall. It was found that for daily rainfall, the SCS3, SCS4 and Knoesen model approaches performed considerably better than the other approaches in the pilot study. The RTD approaches were further assessed using data from 14 additional rainfall stations. For the additional stations, the Knoesen model disaggregated depths provided the most realistic temporal distributions overall, followed by the SCS-SA approach. In additional, an adapted form of the Triangular distribution was found to show potential for disaggregation when a generalised value for the timing of the peak was utilised.