Assessing the performance of regional flood frequency analysis methods in South Africa.

Abstract

In engineering and flood hydrology, the estimation of a design flood refers to procedures whereby the magnitude of a flood is associated with a level of risk at a given site (Pegram and Parak, 2004). The use of a Regional Flood Frequency Analysis (RFFA) approach improves the accuracy and reliability of estimates of design floods. However, no RFFA method is currently widely used in South Africa, despite a number of RFFA studies having been undertaken, that include South Africa. Hence, the performance of the current RFFA approaches needs to be assessed in order to determine the best approaches to use and to determine if a new RFFA approach needs to be developed for use in South Africa. Through a review of the relevant literature it was found that the Meigh et al. (1997) Method, the Mkhandi et al. (2000) Method, the Görgens (2007) Joint Peak-Volume (JPV) Method, which uses a K-Region regionalisation, as well as a Veld zone regionalisation, and the Haile (2011) Method are most suitable for application in a nationwide study. Each regional approach was assessed by comparing their design flood estimates with those estimated from an at-site flood frequency analysis of the observed flood data, using both the General Extreme Value (GEV) and Log Pearson Type 3 (LP3) distributions. However, due to the LP3 distribution producing inconsistent design flood estimates, it was removed from further analysis and only the GEV distribution was assessed. Annual Maximum Flood (AMF) data were obtained from the Department of Water and Sanitation (DWS) for 1458 stations across the entire country. In addition to these datasets, 89 synthesised dam inflow records were obtained from the DWS and incorporated into the study. Due to a thorough data screening process, the final number of stations and dam inflow records analysed was reduced to 407 stations. In order to determine the overall accuracy of the RFFA methods, Relative Errors (RE) (%) were calculated at each station. Box plots and frequency plots were utilised to represent the distribution of relative errors and the degree of bias was measured using a ratio of the estimated and observed design floods. The results of the study show that the Haile Method generally performs better than the other RFFA methods, however it also consistently under-estimates. The Mkhandi Method generally over-estimates. The Meigh Method generally performs the worst, consistently over-estimating. For the JPV Methods, the K-Region regionalisation generally performs better than the Veld zone regionalisation; however, they both consistently over-estimate design floods. The poor overall performance of the RFFA methods are due to a number of reasons. In the case of the Mkhandi et al. (2000) Method, the

tests for homogeneity that were developed were too lenient, which may have incorrectly defined regions as being homogeneous. In the case of the Meigh et al. (1997) Method, the regionalisation of homogeneous flood regions were too broad, where only two flood regions have been identified for South Africa. For the Haile (2011) Method, the logarithmic regressions developed for a number of regions were not able to determine index floods for all catchment areas. Therefore, power regressions were developed in this study. In the case of the JPV Methods, the Kovacs K-Regions and Veld zone regions were used, which have not been updated in the past several years. In response to the generally poor performance of the RFFA methods assessed in this study, it has been recommended that a new method be developed for application in design flood practice in South Africa.