Regional Infrastructure Resilience Assessment and Urban Planning Applications
Advisor: Professor Paolo Gardoni
ABSTRACT
Infrastructure is fundamental to the functioning of cities, providing essential services that support economic activity, mobility, and public well-being. However, natural disasters can severely disrupt infrastructure systems (e.g., transportation, potable water, and wastewater). Such disruptions may delay emergency response, extend recovery, and increase socioeconomic losses. Understanding infrastructure risk and resilience is essential for strengthening urban resilience and improving decision-making processes. Urban land use planning is crucial in strengthening infrastructure resilience by guiding development patterns, reducing exposure to hazards, and supporting resilience strategies. Effective planning supports the long-term functionality and adaptability of infrastructure systems, ensuring they can withstand and recover from disruptions more effectively. With prediction models widely used in decision-making processes, understanding model uncertainty and updating models is also significant to refining predictions and improving the robustness of decision-making. These models usually involve comprehensive prediction models, including nested models in complex multi-step procedures. Providing a robust infrastructure risk assessment and integrating it into urban planning while accounting for model uncertainty is essential to enhancing urban resilience. However, achieving these goals requires rigorous and accurate models to evaluate infrastructure performance. Incorporating resilience into practical urban planning remains challenging due to the complexity of interdependent systems and the lack of comprehensive quantitative methodologies. Furthermore, uncertainties in prediction models are difficult to measure due to lack of systematic validation procedure. A systematic approach is needed to integrate multi-level dependencies and incorporate heterogeneous data into model updates for risk analysis. This dissertation addresses the fundamental challenges and focuses on transportation infrastructure, particularly roads, as an example among various critical infrastructure. The contributions of this dissertation cover four main parts, specifically focusing on (1) how to estimate infrastructure risk, particularly road blockage risk due to falling debris after disasters, (2) modeling the impact of infrastructure failure on emergency services and developing mitigation strategies using urban planning tools, (3) understanding prediction model uncertainty through validation, and (4) improving model accuracy as new data become available.
This dissertation develops a probabilistic formulation for estimating the road blockage due to building damaged following an earthquake. The proposed model considers the relevant factors, including debris distance, building damage and road characteristics. The road blockage probability at a given road section is estimated for the four road section types, considering buildings on only one side of the road or both sides, and with or without a raised traffic median. The probability of road blockage for an entire road is then calculated by system and parallel reliability analysis.
This dissertation then proposes a network-based approach for land use optimization, leveraging novel applications of pathfinding algorithms to assess emergency services accessibility. The approach aims to integrate the quantification analysis of infrastructure resilience into this optimization process to enhance decision-making. The two objectives are minimizing the post-disaster accessibility risk while maximizing housing and urban development in the area.
This dissertation proposes three measures to validate the predictive ability of models used in regional risk analysis (i.e., Accuracy Likelihood, Prediction Error, and Distribution Match). Accuracy Likelihood quantifies the probability of observing the recorded data under the predictive model's hypotheses/assumptions. Prediction Error measures the difference between the recorded value and the values predicted by the models. Distribution Match measures the similarity between the probability distributions of the predicted quantities and the corresponding empirical distributions of the recorded data. As an example, we assess the predictive validity of seismic risk and resilience analysis models using data from the 2016 Kumamoto earthquake in Mashiki City, Kumamoto, Japan. This comparison highlights the predictive performance of available models and informs future research on crucial improvements.
Finally, the dissertation proposes a probabilistic formulation using the Bayesian approach to update the model parameters and reduce the uncertainties as data becomes available. As an example, this dissertation applies the proposed methodology to updates the seismic risk analysis in the HAZUS model using data from the 2016 Kumamoto earthquake in Mashiki, Japan. Three levels of updates (i.e., hazard, vulnerability, and functionality) are applied to the risk analysis models for bridges, potable water infrastructure, and wastewater infrastructure. The proposed methodology enables consistent updates across modeling levels when data become available.