The recent occurrence of low-likelihood, high-impact events (LLHIs)—impactful weather events with return intervals that are comparable to, or longer than, the length of the observational record—raises important and unanswered questions related to the relative effects of forcing and variability on these events. 

For example, Hurricane Harvey in 2017 and the Summer 2021 heatwave in the Pacific Northwest (PNW) are two high-impact events with no modern analog. LLHIs challenge two classes of tools that might be used to answer such questions: statistical methods and climate models. 

The Summer 2021 PNW heatwave had temperatures that were so much higher than previous heatwaves that statistical analysis of the event essentially fails because the temperatures appear to be above the bounds predicted by time-varying generalized extreme value theory. 

Further, only high-resolution climate model simulations have demonstrated fidelity in simulating tropical cyclones. Still, computation costs make it infeasible to run the large ensembles of simulations necessary to make inferences about the statistics of extremely rare hurricanes like Harvey. In addition to these methodological challenges, there is a clear need to understand the weather phenomena and physical processes that lead to LLHIs in the present, and that might cause LLHIs in the future.

CASCADE Planning

CASCADE will develop new machine-learning (ML) approaches to generate the huge ensembles of simulations needed to study LLHIs. This will be coupled with advances in statistical approaches for better characterizing the observational record, including a driver-focused framework for evaluating LLHI risk. In the longer term, CASCADE will: 

• Develop a world-leading ML framework for climate emulation and causal inference;
• Transform the observation, simulation, and theory of LLHIs by leading a seamless integration of cutting-edge ML, high-performance computing, and novel statistical methods; and
• Increase confidence and reduce uncertainty in subseasonal to decadal LLHI forecasts by drawing evidentiary strength across this seamless integration.