Investigation of the Robustness of Probabilistic Coastal Hazard Analysis Results to Different Joint Distribution Models of Storm Parameters
Abstract
The United States Army Corps of Engineers has developed the Probabilistic Coastal Hazard Analysis (PCHA) framework (https://doi.org/10.2112/SI95-235.1) to extend and advance the joint probability method (JPM) used to establish coastal hazard curves. Beginning with the analysis of historical storm-related climatology records, the framework builds a joint distribution model of tropical cyclone (TC) parameters (i.e., central pressure deficient, forward velocity, radius of maximum wind, and heading direction). The current framework uses a meta-Gaussian copula (MGC) to characterize the dependence among TC parameters. However, the MGC has a limitation associated with modeling of circular-variables such as heading direction. This research investigates the robustness of storm surge hazard curves generated from the PCHA framework to the approaches and assumptions used to characterize the occurrence frequency of storm parameter combinations, with a particular emphasis on modeling the joint distribution of linear and circular variables. A von Mises kernel function (VKF) is proposed to replace Gaussian kernel function (GKF) in the computation of storm recurrence rate (DSRR), which is used to represent the probability model of heading direction. This study then builds a series of joint distribution models based on assumptions of parameter independence, partial dependence (e.g., dependence between and ), and full dependence. Full dependence models consider a range of copula models (e.g., MGC and vine copulas combining linear-circular copulas with Gaussian or Frank copulas). The sensitivity of the results is assessed by comparing hazard curves at representative locations. The stability of hazard curves generated using an MGC assumption to the selection of the zero-degree convention is also assessed. The instability issue induced by the circular variable can be overcome by introducing the VKF in the DSRR and using the linear-circular vine copula. The tail dependence associated with various copula assumptions are also compared. Preliminary results suggest that the vine model based on linear-circular copula and Frank copula is able to generate robust hazard curves.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMNH21A..03L