Extreme Values from Spatiotemporal Chaos: Precipitation Extremes and Climate Variability

Characterizing the underlying data generation processes for extreme values in complex spatiotemporal dynamical systems, and developing predictive insights on these extremes, can improve our fundamental understanding of fluid dynamics including turbulence and eddies. The translational benefits are expected to span disciplines such as earth or atmospheric sciences and space sciences all the way to water resources engineering or electrical power grids and the propagation of diseases in humans or information in social media. However, each discipline may offer unique challenges. One challenge in understanding and projecting mean change and extremes patterns in climate, and translating to climate adaptation, is the inherent natural variability of the climate system, which dominates in the stakeholder-relevant 0-30 year near-term. The nonlinear dynamical climate system exhibits high sensitivity to initial conditions. Climate model simulations attempt to capture the inherent natural variability in climate systems through multiple initial condition ensembles. In addition, the variability resulting from gaps in our process understanding are encapsulated through multi model ensembles based on parametric and structural differences in models, while uncertainties in emissions trajectories are described via what-if scenario ensembles. While extreme value theory has been found useful for studying climate extremes, the multiplier of ensembles may disproportionately increase the uncertainty in projections. Here we examine the hypothesis that initial condition ensembles, which are generated from identical dynamical simulations, can be examined collectively to narrow the uncertainty in our assessments of precipitation extremes under climate variability. Our findings may be directional for climate studies and potentially relevant for related disciplines.