Session Abstracts:
Session 7A
Professor
Stanford University
Presentation Title: Stochastic sampling strategies for infrastructure risk assessment
Abstract: Risk assessment of distributed infrastructure systems often utilizes stochastic simulations of hazardous events. These simulations are comprised of maps of loading intensity (e.g., ground motion intensity or flood inundation depth) and associated occurrence rates. Given the uncertainty in the characteristics of future hazardous events, a large number of simulations are produced to represent the distribution of possible loading. However, the downstream analyses of impacts are often computationally expensive, making it undesirable or infeasible to consider many simulations. One approach to address this problem is to select a subset of the simulations that approximates the distribution of loading represented by the full set. Algorithms for subset selection will be presented, and their advantages and limitations discussed. This work facilitates expanded and risk-consistent studies of the risk and resilience of distributed systems and communities.
Professor
University College London, UK; & University of British Columbia, Vancouver, Canada
Presentation Title: Dynamic cities, dynamic natural-hazard risk: representing urban changes and hazard interactions in regional risk modeling for decision making under deep uncertainty
Abstract: The world’s changing climate and rapidly evolving societies are exacerbating the risk posed by natural hazards to our infrastructure and communities. The complex interplay between built, natural, and social systems, coupled with the potential amplification of impacts due to hazard interactions and climate change, presents significant challenges in assessing urban resilience. Yet, there is a lack of tools available in the literature for comprehensively assessing (and supporting related decision-making on) the performance of the built environment under future multi-hazard conditions, considering the compounding and cascading impacts caused by hazard interactions (potentially amplified by climate change) and systems’ interdependencies. This talk will discuss some of the challenges and opportunities involved in dynamic regional risk modeling to support decision-making under deep uncertainty. It will also describe recent advances to address such challenges, including 1) the use of urban growth modeling to estimate the uncertain future expansion of urban areas and their exposure to natural hazards; 2) a methodology based on the theory of competing Poisson processes and sequential Monte Carlo sampling to efficiently simulate multi-hazard event sets (i.e., sequences of hazard events and associated characteristics throughout a system’s lifecycle); 3) a discrete-time, discrete-state Markovian framework for practical multi-hazard lifecycle consequence analysis of deteriorating engineering systems, appropriately accounting for hazard interactions and their cumulative effects on the system’s performance. These insights aim to aid in formulating robust mitigation strategies and policies under various climate and societal development scenarios.
Professor
University of Notre Dame
Presentation Title: Promoting computational efficiency for regional risk assessment applications
Abstract: Comprehensive regional risk assessment applications entail a significant computational burden. This may originate, for example, from the aspiration to accurately describe the response of a large number of assets, or the need to address high-dimensional input and output variables, to properly capture correlations across the quantities of interest (QoIs). This presentation focuses on UQ advances that can promote computational efficiency within these settings. A number of different topics are briefly presented to demonstrate recent developments and foster discussions. The use of reduced order modeling (physics-based approximation) and surrogate modeling (data-driven techniques) is examined for supporting the response estimation for large number of assets. Subtopics on the importance of information fidelity for building reduced order models, or the need to accommodate alleatoric hazard uncertainties when adopting surrogate models are discussed. For accommodating high-dimensional variables in regional risk assessment, the use of projection and dimensionality reduction methods is promoted, offering improved numerical efficiency while simultaneously reducing memory requirements in the computational workflow. Finally, the implications of the high-dimensionality of the QoIs is discussed. Such QoIs may represent competing priorities for many UQ algorithms with adaptive characteristics (such as, importance sampling, Multi-Fidelity Monte Carlo or intelligent selection of simulation scenarios), creating the need for efficient solutions to find a balanced compromise across them. It is shown that the higher the degree of adaptivity of the algorithm (meaning the higher the potential to provide substantial benefits in computational efficiency for single QoI), the greater the reduction of its efficiency in setting involving multiple, competing QoIs.