Development of peaks-over-threshold based flood estimation methods

Abstract

Flood Frequency Analysis (FFA) serves as a cornerstone for water resource management, disaster preparedness, and infrastructure planning—especially in hydrologically diverse and flood‐prone regions such as southeast Australia. Traditional FFA methods based on the Annual Maximum (AM) model, which consider only the single highest flood event per year, inherently underrepresent frequent flood occurrences. The Peaks-over-Threshold (POT) modelling approach, in contrast, offers greater flexibility and theoretical accuracy but has been underutilized due to challenges related to threshold selection, independence criteria, parameter estimation, and regional adaptation. The current regional flood frequency analysis (RFFA) approach in Australian Rainfall and Runoff (ARR) is based on AM model and it does not support design flood estimates smaller than 2-year return period. This thesis confronts these challenges by developing and applying innovative POTbased methodologies. An initial comparative analysis using data from 188 gauged stations across southeast Australia reveals significant differences in flood quantile estimates between the AM and POT modelling approaches. Discrepancies are further explained by key catchment characteristics—mean annual rainfall (MAR), mean annual evapotranspiration (MAE), and stream density (SDEN)—thus laying a robust foundation for tailoring advanced design flood estimation techniques to diverse hydrological settings. Overall, the thesis establishes that POT-based FFA is superior to traditional AM methods by addressing pivotal challenges in threshold selection, parameter estimation, and spatial interpolation. The resulting, flexible framework improves flood quantile estimation across gauged and ungauged catchments, thereby informing robust flood risk management strategies and guiding updates to national guidelines. This superiority arises because the POT framework makes more efficient use of the available information by retaining multiple peaks above a threshold in each year, thereby increasing sample size and reducing sampling variability, particularly for frequent flood quantiles. Consistent with these theoretical advantages, the POT-based models developed in Chapters 3–5 and 8 consistently yield lower median relative errors than the AM-based RFFA currently recommended in the ARR guideline for very frequent to frequent events.

Date of Thesis submission	2025
Original language	English
Awarding Institution	Western Sydney University
Sponsors	Australian Government Research Training Program Scholarship
Supervisor	Ataur Rahman (Supervisor) & Sathaa Arumugam Sathasivan (Supervisor)