A Bayesian Network Method for Deaggregation: Identification of Tropical Cyclones Driving Coastal Hazards

Abstract

Bayesian networks (BN) have advantages in visualizing causal relationships and performing probabilistic inference analysis, making them ideal tools for coastal hazard analysis and characterizing the compound mechanisms of coastal hazards. Meanwhile, the Joint Probability Method (JPM) has served as the primary probabilistic assessment approach used to develop hazard curves for tropical cyclone (TC) induced coastal hazards in the past decades. To develop hazard curves that can capture the breadth of TC-induced coastal hazards, a large number of synthetic TCs need to be simulated, which is computationally expensive. Given that low exceedance probability (LEP) coastal hazards are likely to result in the most significant damage to coastal communities, it is practical to focus efforts on identifying and understanding TC scenarios that are dominant contributors to LEP coastal hazards. This study developed a BN-based framework incorporating existing JPM for multiple TC-induced coastal hazards deaggregation. Copula-based models capture dependence among TC atmospheric parameters and generate CPTs for corresponding BN nodes. Machine learning surrogates model the relationship between TC parameters and coastal hazards, providing conditional probability tables (CPTs) for hazard nodes. Case studies are applied to the Greater New Orleans region in Louisiana (USA). Deaggregation is a method for identifying dominant scenarios for a given hazard, which was first established in the field of probabilistic seismic hazard analysis. The objective of this study is to leverage BN to develop a deaggregation method of multiple LEP coastal hazards to better understand the dominant drivers of coastal hazards to refine storm parameter set selection to more comprehensively represent multiple forcings.