Flood frequency analysis at multiple durations

Abstract

Flooding currently affects more people than almost any other natural hazard (Van Loenhout et al., 2020) and the proportion of the world’s population that lives in flood-exposed areas is growing rapidly (Tellman et al., 2021). Part of the way society manages exposure to flood risk is through estimation of flood design values. These values give estimates of flood magnitude within a given return period and are essential to making adaptive decisions for a variety of hydrologic applications, e.g., infrastructure design, land use planning, and water resource management. For flood retention-specific applications—e.g. floodplain management and reservoir design—we often need design values at multiple durations. Here our focus is the retention capacity of a man-made or natural basin. We are therefore concerned with the total flow volume we can expect to see over a short duration like one hour vs a longer one like one day, regardless of whether that volume of water comes from a single event or multiple consecutive events. This thesis explores and develops statistical methods for obtaining design values at different durations for flood retention-specific applications. Specifically, we propose an extension to an existing flood-duration-frequency model that allows for more realistic modeling of the relationship between design values of different duration at individual locations. A Bayesian inference framework for these local models is also proposed, allowing for accessible uncertainty estimation and estimation of a mixture model that helps establish the importance of the model extension. We also assess the suitability of regression-based regional flood frequency analysis models for estimating design values at multiple durations at out-of-sample locations, and offer recommendations for regional model structure if design value estimation at multiple durations is the goal.