Large-scale circulation patterns rule the predictability of extreme events

Extreme events in geophysical flows have a strong impact in human life and in diverse economical activities, and their

prediction is crucial to mitigate their consequences. Particularly now, in the context of climate change, interest has

emerged to develop models and indicators that allows for early detection and warning of extreme events. Some of these

models have proved successful, but it is unclear if they can be improved because the predictability limit of extreme

events is unknown. In this talk, we show that it is now possible to measure the predictability of extreme events directly by probing phase

space with a computationally-intensive approach. This allows to assess the exact potential of predictive models and to

determine precisely the conditions under which they may be effective. We analysed the predictability of extreme bursts

of the dissipation in a two-dimensional Kolmogorov flow by producing massive ensembles of initial conditions perturbed

around independent base flows. We produced millions of realisations to cover the full attractor of the Kolmogorov flow,

and used the Kullback—Leibler divergence, an information-theoretical tool, to assess predictability. This analysis

shows that extreme bursts of the dissipation may be successfully predicted beyond a few Lyapunov times, but that their

predictability depends strongly on the phase-space region from which they emerge. Specifically, we reveal that predictable

and unpredictable events evolve from two distinctly different large-scale circulation patterns. These results

open the possibility of improving predictive models by tuning them to the large-scale dynamics. Our approach could

be adapted with the available compute power to more complex flows.