Development of a Bayesian Copula-based Nonstationary Framework for Compound Flood Risk Assessment

Flooding is one of the most important natural hazards that threatens social stability and economic growth in flood-prone regions. Over the last decades, a significant shift in the frequency and intensity of climate extremes has amplified the flood hazard. As the first step to addressing flood hazards for future planning, flood frequency analysis is widely utilized. While older studies primarily focused on a single flooding driver, recent studies have emphasized the importance of considering two or more flooding variables in an interactive framework, i.e., considering the potential for compound flooding. However, the transition from a univariate and stationary framework into a multivariate structure that accounts for non-stationarity can drastically increase the uncertainty, making the inference arduous, if not impossible. This study develops a non-stationary and copula-based Bayesian framework that incorporates the impact of dependence between flooding drivers, i.e., sea level and precipitation, and their existing trend into flood frequency analysis. The framework will allow investigating how the individual and combined effects of dependence and non-stationarity will influence the frequency and magnitude of extreme events. Furthermore, the Bayesian framework will allow for the incorporation of uncertainty, which may arise from the shortage of data, model selection, and parameter estimation, into the flood frequency analysis. Eight station pairs along US coastlines that represent different types of behavior based on marginal trends, the strength of dependence, and the length of joint data are selected for this study. Analysis of these station pairs shows that the inter-play between dependence and trends can reduce the joint return period by more than five times in locations with high positive trends in sea level and precipitation and strong dependence while increasing the joint return period in locations with a negative trend in precipitation. In addition, preliminary results reveal that uncertainty is highly dependent on the length of joint data. This study highlights the importance of hydrological dependence and trends in the quantification of return periods and emphasizes the necessity of accounting for uncertainty to permit reliable estimation of flood hazards.