THE ROLE OF EXTREME EVENTS FOR WATER SECURITY IN A NONSTATIONARY ENVIRONMENT

About 25% of the global population currently lacks adequate access to safe and secure water, while 50% experience severe water scarcity for at least part of the year. These figures are expected to rise due to a non-stationary climate, where extreme weather events are intensifying and making water cycles increasingly uncertain. Visible water crises often lead to new infrastructure funding, though planned expansions can stall due to conflicting projections of water availability. To address this challenge, this research explores a novel approach for assessing water supply security in a non-stationary environment by investigating the relationship between extreme events, i.e. extreme precipitation and floods, and water supply. The thesis focuses on three key objectives: (1) identifying and characterising relationships between floods and annual streamflows, (2) developing robust methods for modelling floods and streamflows in a non-stationary climate, and (3) evaluating how these relationships can improve water availability modelling. A significant relationship was found between floods and annual streamflow volumes that is well described by catchment streamflow variability and aridity. In regions with high variability and aridity, such as Australia, southern Africa, and the Mediterranean, total annual streamflow volumes often concentrate in a small number of low-frequency events, with over 40% of annual volume coming from a single flood event. This suggests water supply can be characterized by just a few significant, low-frequency events. Furthermore, a reliable and parsimonious alternative for rainfall projections was developed using a non-parametric, stochastic, subdaily rainfall generator. This tool produces ensembles of continuous rainfall that preserve Intensity-Frequency-Duration relationships and better capture low-frequency persistence than Regional Climate Model (RCM) downscaling. A robust calibration methodology was also introduced to improve parameter selection for extreme events. Projections of flooding were evaluated at climatologically unique sites across Australia using a large ensemble of the most recent RCM products. Variability in low frequency events was observed to be better captured compared to higher frequency events, while evidence of less bias in lower frequency events showed strong dependence on climate. Significant uncertainty persists in rainfall representation by RCMs, limiting confidence in their projections for water supply modelling.