Estimation of Extreme Rainfall in South Africa and Development of Methods to Account for Non-stationary Climate Data

Abstract

<p>The estimation of design rainfalls and design floods are required by engineers and hydrologists to design and quantify the risk of failure of hydraulic structures. Extreme design rainfall quantities such as high-return period rainfalls and the probable maximum precipitation (PMP) are needed to design high-hazard hydraulic structures. In South Africa, previous design rainfall estimates have been produced up to the 200 year return period. PMP estimates were last determined nearly 50 years ago based on only 30 years of data. Most studies on extreme rainfall reported are based on frequency analysis assuming stationary conditions. Previous studies in South Africa have assumed a stationary climate. However, the assumption of a stationary climate in rainfall and flood frequency analysis has been challenged owing to evidence of climate change. Recent literature indicates that the magnitude and frequency of extreme rainfall events has been changing and this is likely to continue in future. Hence, methods to account for trends in extreme rainfall events in a changing environment need to be developed. In addition, the concept of PMP, particularly as used for the design and safety evaluation of large dams in South Africa, is being challenged with the recommendation that high-return period design rainfalls be used in these assessments. The aims of this study are: (i) to estimate extreme design rainfall values, with a focus on return periods greater than 200 years, (ii) to update PMP estimates using updated data and modernised methods, and (iii) to account for non-stationary climate data in the estimation of these extreme rainfall events in South Africa. Frequency analysis using LH-moments, which more accurately fit the upper tail of distributions, have been used to estimate high-return period design rainfalls. Regular L-moments are shown to overestimate the extreme rainfall quantities when compared to LH-moments by giving undue favour to outliers. PMP estimates have been determined using a storm maximisation and transposition approach. Radial Basis Functions (RFBs) have been used to transpose PMP estimates to ungauged locations, producing PMPs for the entire country. Approximately 80 % of the new PMPs are greater than the previous estimates. This is probably due to the many limitations of the old approach and differences used in the new approach, indicating that the new approach undertaken in this study may provide improved estimates. The PMP represents the upper limit of extreme rainfall, however, comparisons of high-return period rainfalls to the PMP show that the PMP is sometimes exceeded by the high-return period rainfalls. To develop methods to estimate extreme design rainfall events in a non-stationary climate, this study explores the impacts of climate drivers, such as the Southern Oscillation Index (SOI), and changes in atmospheric variables, such as dew point temperature, on high-return period rainfalls and the PMP.</p>