Date of Award

Summer 2024

Document Type

Open Access Dissertation

Department

Civil and Environmental Engineering

First Advisor

Michael Meadows

Second Advisor

Vidya Samadi

Abstract

Extreme flood events have been persistent throughout history and have challenged human populations around the globe. Over time, scientists and engineers have developed tools and technology to support flood risk management. Most, if not all, of these tools rely on some form of the field of stochastic hydrology. Stochastic hydrology is a broad discipline, but the overarching principle is to use statistical and data-driven techniques to develop an understanding of one or more hydrologic variables (e.g., flood flow or peak storm tide). However, the need for large sample sizes of data that offer insight into all possible combinations of outcomes routinely offers challenges when implementing stochastic techniques to estimate and predict hydrologic variables. If large data sets are available, the next challenge is developing a parsimonious model and/or tool that offers both historical and future predictive capabilities with confidence. Herein, the core of this research was turned to applications of stochastic hydrology with respect to quantifying and characterizing riverine peak flood flow and multivariate coastal compound flooding. Three distinct studies are presented using data from the southeastern United States to better quantify and address challenges in the field of stochastic hydrology with respect to stationary flood frequency analyses, non-stationary flood frequency analyses, and stationary multivariate time series analyses. The first study adopted a stationary peaks-over-threshold and block maxima approach to characterize and estimate peak stream flow of four major subcatchments affected by flooding from Hurricane Joaquin in South Carolina. Goodness-of-fit statistics were used to select a parent flood distribution and check for changing distribution tails while return periods were estimated and complemented with bootstrapped confidence intervals to evaluate model accuracy. Results of the first study provided insight into extreme flooding experienced during Hurricane Joaquin as well as the complexity of estimating peak flood flows, especially when using a peaks-over-threshold approach. The second study expanded on the first study by developing non-stationary block maxima flood frequency models of the same subcatchments with a focus on winter and spring peak discharge. Various climate signals (i.e., SOI, AMO, and NAO) and a modified reservoir index were considered as possible sources of non-stationarity. Comparisons were made between baseline stationary models and competing non-stationary models using bootstrapped return period and design life risk of failure metrics. Results of the study suggested that consideration of non-stationarity had insignificant effects on design discharge and risk of failure point estimates considering overall model confidence, except for design discharge estimates for the Saluda River spring model. As a result, the use of classical stationary flood models may still be appropriate in South Carolina, and careful consideration should be taken before adopting a more sophisticated non-stationary approach. The third study focused on developing a framework to jointly characterize observed rainfall and storm tide time series using LOWESS-based autoregressive moving average (ARMA) time series models and bivariate copulas. The proposed framework is tested using data observed during Hurricane Irma along the coast of Florida, Georgia, and South Carolina. A Monte Carlo approach was used to estimate model significance (i.e., two-sided p-values) while a classical time series forecasting approach was used to check for model suitability based on random realizations from the developed storm copulas. Storm intensities were then evaluated using "AND" and Survival Kendall (SK) non-exceedance probability definitions. The proposed framework was successful in using bivariate copula-based time series models to characterize the dependence between observed rainfall and storm tide in coastal watersheds, thereby offering another resource for coastal compound flood risk management.

Rights

© 2024, Ryne Phillips

Download

Share

COinS