Computational Science Initiative Event

Uncertainties about future climate change substantially influence the long-range resilience strategies adopted by regional planners. I survey some recent methodological work targeted at quantifying computer model "structural" uncertainties, which are those arising from subjective choices of physics and numerical approximations made by different model developers, leading to multi-model uncertainty. These methods include hierarchical Bayesian approaches for combining prediction from multiple models; Bayesian networks of computer model emulators for combining information from different types of models and data sources; and "quasi-intrusive" techniques for system identification and reduced model construction to explore the space of model structural uncertainties.

I then demonstrate how climate projection uncertainties may be used to redesign infrastructure networks for resilience to climate impacts. Sea level rise projections, combined with a stochastic hurricane generator and propagated through a physical hydrodynamic model, result in a probability distribution of local storm surge flooding. These flood impacts are evaluated against a toy model of an electrical power grid in a coastal setting. Formal optimization methods, formulated as a chance-constrained, multi-stage mixed-integer program, are used to redesign the power network for resilience against the gradually increasing vulnerability to short-term extreme weather events induced by climate change.