Advancements in Uncertainty Quantification for Coastal Hazard Assessment

The vulnerability of coastal communities has been gradually increasing in the past decade owing to migration trends and the intensification due to climate change of the natural-hazard threats they face. This has led to the demand for comprehensive hazard assessment tools that can support decision-making regarding pre-disaster planning, emergency management, and post-disaster recovery. This demand has motivated researchers to develop high-fidelity, numerical simulation models to accurately predict hazard exposure and fragility for coastal infrastructures. However, the numerical complexity of these models entails a substantial computational burden, making their utilization in supporting comprehensive hazard/risk assessment challenging. To address this challenge, researchers have focused on computational statistics and machine learning tools that aim to reduce the number of high-fidelity simulations needed within such assessments. This work establishes numerous advancements on this topic, focusing specifically on storm surge hazard assessment. A novel global sensitivity analysis framework is developed to gain a better understanding of the relationship between storm features and surge responses, accommodating a potential reduction of the uncertainties that need to be considered in hazard estimation. For real-time surge hazard estimation (during landfalling events), two distinct methodologies are introduced for improving computational efficiency by sharing numerical simulation results across the hazard assessments performed for different storm advisories: (i) adaptive importance sampling and (ii) adaptive multi-fidelity Monte Carlo. For long-term regional hazard assessment, the benefits of leveraging surrogate models for the surge response approximation are examined in detail, revealing for the first time the true computational benefits that these surrogate models can offer. Also, the selection of representative storm scenarios that can achieve a significant reduction in computational burden for the hazard assessment without compromising the accuracy of the hazard map estimation across large coastal regions is examined. Finally, to improve the accuracy of surrogate models developed for the storm surge approximation, novel data imputation and dimensionality reduction techniques are introduced, extracting the spatio-temporal correlation of surge responses and leveraging the information to establish higher accuracy with respect to the surrogate model calibration and deployment.