Abstract. Climate science needs more efficient ways to study high-impact, low-probability extreme events, which are rare by definition and costly to simulate in large numbers. Rare event sampling (RES), including ensemble boosting, offers a novel strategy to extract more information from those occasional simulated events, by applying small perturbations to turn a moderate event into a severe one which otherwise might not come for many more simulation-years. But how severe the events can become, and their estimated probabilities, depend sensitively on the details of the perturbation. In particular, for sudden and transient events like precipitation, performance of boosting depends sensitively on the choice of advance split time (AST) of the perturbation. Heuristically, the perturbation must come early enough before the event to let the ensemble of simulations diversify, but not so early that they forget the special initial conditions that led to the extreme. In pursuit of guidelines for choosing the AST, we study the effect of AST in the task of sampling extreme fluctuations of a passive tracer in a quasigeostrophic turbulent channel flow. This model system is idealized, but captures key elements of midlatitude storm track dynamics while exposing similar algorithmic challenges. We formulate AST selection as a concrete optimization problem for statistical accuracy against a ground truth. Given that such a ground truth would not generally be available, we propose a proxy objective function to optimize in practice: thresholded entropy, which rewards ensembles with both a high mean and a large spread. We show that ensemble boosting, when given a well-chosen AST and equipped with methods to estimate probabilities, can accurately sample extremes at long return periods. We furthermore find evidence that thresholded entropy successfully identifies an optimal AST, which is roughly 1–3 ddy turnover timescales in the quasigeostrophic system. Moreover, this proxy captures the variation of AST with the target location of the tracer within the flow field, suggesting it can generalize to more general chaotic systems including realistic climate models. Applying our boosting methodology at scale will require further development in adaptive optimization strategies, but our work here is an essential first step for establishing what must be optimized.
Finkel et al. (Thu,) studied this question.