Assessment of the performance of deterministic design flood estimation methods in South Africa

Abstract

A design flood may be defined as the peak discharge value that corresponds to an assigned non-exceedance probability, typically expressed in terms of a return period. The estimation of design floods focuses on the frequency analysis of peak flows and/or runoff volumes for the design of hydraulic structures and flood inundation delineation. Furthermore, realistic estimates are crucial to ensure an acceptable correlation between expected flood magnitudes and associated risk levels, not only to enable the planning and design of hydraulic structures, but more importantly, to preserve human life and infrastructure. The primary aim of this research was to assess the performance of the event-based deterministic design flood estimation (DFE) methods currently used in South Africa in 396 gauged catchments scattered across South Africa. This assessment was intended to contribute to the development of best practice guidelines for the application of event-based deterministic DFE methods in South Africa, consistent with the objectives set out by the National Flood Studies Programme (NFSP). Recognising the practitioners’ dilemma in selecting a single, justifiable event-based deterministic DFE method from the suite of DFE methods and alternative options available, the specific objectives of this research sought to establish a performance assessment and ranking system. This system was designed to guide practitioners in choosing the most appropriate event-based deterministic DFE method for application in specific primary drainage regions (PDRs) and return periods, thereby promoting the consistent adoption and application amongst engineering practitioners and hydrologists. The methodology adopted to assess and compare the performance of the event-based deterministic and at-site probabilistic DFE methods was based on the core components of DFE, namely, input (observed and design rainfall), transfer functions (catchment characteristics), and output (runoff estimation). Event-based deterministic DFE was carried out using the Rational Method-Alternative 3 (RM3), Soil Conservation Services (SCS), Soil Conservation Services-South Africa (SCS-SA), Lag-routed Hydrograph (LRH) and Synthetic Unit Hydrograph (SUH) methods. Both the veld-type and time of concentration (TC) approaches associated with the LRH method were considered in this study. Apart from the SCS and SCS-SA methods, all other DFE methods incorporated the use of the areal reduction factor (ARF) estimation methods recommended for South Africa as developed by Alexander (2001) and Pietersen (2023), and designated as ARF (A) and ARF (P), respectively. The standard application procedures associated with each DFE method were followed by employing relevant software tools and acknowledging the relevant assumptions, limitations, and intended applications of each DFE method. Results from these methods were consolidated to facilitate the comparison of the deterministic (QTi) flood peaks against the benchmark probabilistic (QPi) flood peaks. The performance assessment was based on a ranking procedure to assess the relative accuracy and biases of each event-based deterministic DFE method across the 396 gauged catchments using an array of goodness-of-fit (GOF) criteria. The DFE method with the lowest composite ranking value was considered as the best performer. Additionally, graphical tools such as box plots and scatter plots were used to visualise discrepancies and identify areas for improvement. The combination of statistical rankings and visual assessments enabled a comprehensive evaluation of the various DFE methods. Overall, when considering the different ranking and grouping procedures adopted, it was evident that the SCS and SCS-SA methods consistently emerged as the most reliable and robust event-based deterministic DFE methods across all catchments and return periods within the 20 PDRs distributed across South Africa. In contrast, the LRH-Veld-ARF (P) method was regarded as the least reliable. Both the LRH-TC-ARF (A) and LRH-TC-ARF (P) methods also demonstrated a consistent performance, frequently ranking amongst the top methods, while the SUH-ARF (A), SUH-ARF (P), LRH-Veld-ARF (A), and LRH-Veld-ARF (P) methods generally performed less satisfactorily. The RM demonstrated moderate performance, with RM3-ARF (P) often ranking better than RM3-ARF (A). However, both these methods demonstrated higher uncertainty at lower return periods. Ultimately, the selection of hydrological input parameters and methods, along with the application of various DFE methods, can lead to significantly different estimates. This variability arises from factors such as differences in data quality and sources, uncertainties in input parameter values, variations in spatial and temporal resolution, distinct calibration and validation procedures, as well as methodological frameworks inherent to each DFE method. A comparison between literature, the results obtained in this study, and results from other studies revealed several key factors that may contribute to the observed discrepancies, including: (i) the deterioration of the current rainfall monitoring and flow-gauging networks, (ii) overall quality of rainfall, streamflow, and catchment parameter data sets, (iii) characteristics of the annual maximum series (AMS) data sets, particularly record lengths and outlier handling methods, (iv) selection of plotting positions and theoretical probability distributions, and (v) GOF criteria and employed ranking procedure(s). The above-listed factors underscore the need for rigorous assessments and standardisation in DFE, all of which were achieved in this research at a national scale. Subsequently, the spatial and qualitative performance rankings of the event-based deterministic DFE methods across all PDRs and return periods can be incorporated for the development of a best practice guideline framework for DFE in South Africa.