The question
Multiscale resilience analysis stitches together sub-models for hazard, deterioration, performance, and consequences. Beyond the usual parameter uncertainty, the choices among those sub-models (for example, selecting their fidelity or the quantity of interest) introduce a subjectivity that is rarely quantified. This modeling uncertainty compounds across scales, can overshadow careful sub-model refinement, and shifts outcomes for the very populations for whom decisions are made.
What we do
We investigate computational methods to propagate input uncertainties across scales. Ranging from analytical to machine-learning-based approaches, we explore ways to accelerate UQ for different quantities of interest. We also quantify the uncertainty arising from the modeling process itself, not only from model inputs, by studying how modeling choices interact and compound. We are interested in understanding how the resulting outcome shifts reach distinct communities. We develop decision-theoretic frameworks that translate this uncertainty, including the uncertainty introduced by our own modeling choices, into concrete model-selection guidance. The goal is guidance that trades model efficiency against the risk of prediction error, or against the social cost of decisions made on incorrect estimates, making the science of measuring resilience more objective, equitable, and fit-for-purpose.
Publications
- Identifying submodel fidelity combinations that yield efficient multiscale models for infrastructure performance estimationReliability Engineering and System Safety, in review (April 2026)
- Coupling Effects of Fragility Fidelity and Network Resolution in Infrastructure ResilienceEngineering Mechanics Institute Conference and Probabilistic Mechanics & Reliability Conference (EMI/PMC 2024), Chicago, IL, USA1st Place, EMI Objective Resilience Student Paper/Presentation Competition