A Multi-Tiered Bayesian Network Coastal Compound Flood Analysis Framework

TL;DR

This paper introduces a multi-tiered Bayesian network framework for analyzing coastal compound floods, enabling probabilistic risk assessment with varying data and resource levels, demonstrated through a case study in New Orleans.

Contribution

It develops a flexible, multi-tiered Bayesian network framework for coastal flood analysis, accommodating different data availabilities and complexities, with a practical case study implementation.

Findings

The framework effectively models complex flood interactions.

Probabilistic hazard curves were successfully constructed.

The approach adapts to varying data and resource constraints.

Abstract

Coastal compound floods (CCFs) are triggered by the interaction of multiple mechanisms, such as storm surges, storm rainfall, tides, and river flow. These events can bring significant damage to communities, and there is an increasing demand for accurate and efficient probabilistic analyses of CCFs to support risk assessments and decision-making. In this study, a multi-tiered Bayesian network (BN) CCF analysis framework is established. In this framework, conceptual designs of multiple tiers of BN models with varying complexities are developed for application with varying levels of data availability and resources. A case study is conducted in New Orleans, LA, with three tiers of BN models constructed to demonstrate this framework. In the Tier-1 BN model, storm surges and river flow are incorporated based on hydrodynamic simulations. A seasonality node is used to capture the dependence between concurrent river flow and tropical cyclone (TC) parameters. In the Tier-2 BN model, joint distribution models of TC parameters are built for separate TC intensity categories. TC-induced rainfall is modeled as input to hydraulic simulations. In the Tier-3 BN model, potential variations of meteorological conditions are incorporated by quantifying their effects on TC activity and coastal water level. Flood antecedent conditions are also incorporated to more completely represent the conditions contributing to flood severity. In this case study, a series of joint distribution, numerical, machine learning, and experimental models are used to compute conditional probability tables needed for the BNs. A series of probabilistic analyses is performed based on these BN models, including CCF hazard curve construction and CCF deaggregation. The results of the analysis demonstrate the promise of this framework in performing CCF hazard analysis under varying levels of resource availability and project needs.