Disaster risk modeling and assessment towards enhanced infrastructure resilience under extreme events
High-impact, low-frequency (HILF) events such as tsunamis, floods, and earthquakes cause inevitable damages to the built environments. Those extreme events are hard to predict as they depend on a complex mixture of variables. Alternative efforts to control disaster risk have been paid to rapid asses...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Thesis-Doctor of Philosophy |
Language: | English |
Published: |
Nanyang Technological University
2024
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/175523 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | High-impact, low-frequency (HILF) events such as tsunamis, floods, and earthquakes cause inevitable damages to the built environments. Those extreme events are hard to predict as they depend on a complex mixture of variables. Alternative efforts to control disaster risk have been paid to rapid assessment and mitigation strategies. Computational approaches leveraging a deluge of multimodal data require high precision, short processing time, and informed uncertainty. This thesis presents several studies on quantitative risk modeling and assessment methods throughout the disaster risk management circle including mitigation, preparedness, response, and recovery phases, each focusing on a specific issue contextualized for several representative infrastructure systems.
Short-term prediction of overall disaster loss helps make proper disaster management plans, e.g., increasing investment in advance to resist possible shocks. In Chapter 3, three key indicators of disaster loss (i.e., death rate, direct economic loss, internally displaced persons) are identified and assessed of their spatio-temporal patterns aggregating all the available historical records, which contain data with a temporal resolution of per year and a spatial resolution of per country. Besides, factors from economic, social, and environmental dimensions with the same resolution are included for predictive analysis.
Following the panoramic view of disaster loss, the lens is shifted to impact assessment of a specific key infrastructure system. Rapidly assessing the building conditions plays a critical role in sheltering the affected persons and supporting developmental activities after disasters. In Chapter 4, a deep learning framework with a data imposition strategy is proposed to automatically identify the physical damages from satellite imagery for those small point infrastructures, i.e., buildings. The experimental study on a recent open-source dataset named xBD validates its effectiveness in solving the extremely imbalanced classification problem for building damages.
Urban functionality will not be possible if places are isolated from each other. The connectivity provided by transportation network heavily affects yet not guarantee its ability in satisfying travel demand. In Chapter 5, a new system performance metric based on a localized measure derived from ridership data is developed for resilience modeling of networked infrastructure systems, i.e., urban rail transit network. Applied to the Hangzhou Metro system, service restoration relying on the new metric is 2~3 times faster and more robust than that relying on the common recovery cost metric.
Efficient repair operations for networked infrastructure systems depend on propositioning of crew members and resources. In Chapter 6, a multiplex network approach with cluster analysis is proposed to further address the propositioning and operation optimization problem of emergency rescue stations in repairing metro transit network. Modeling the metro stations and its supporting emergency rescue stations as a coupled network, the optimized operation from the proposed cluster-based multiplex network has much better or comparable recovery resilience (i.e., reduces the disrupted performance over time by 0.25 to 0.5) than its four centrality- based competitors.
In summary, those studies harness the power of data-driven approaches incorporating multimodal data (i.e., disaster loss, remote sensing, and ridership data) to improve the precision, speed, and reliability of disaster risk modeling of infrastructure systems concerning tasks such as proactive risk planning, rapid impact assessments, resilient restoration strategies, and optimal restoration operations. |
---|