Complexity of extreme-event prediction in turbulent flows

Predicting extreme events in turbulent flows represents a challenge due to chaos or modeling uncertainty. This paper addresses a fundamental but unexplored limitation in extreme-event forecasting: the minimum computational cost of producing accurate forecasts. The information bottleneck method is applied to massive ensembles of Kolmogorov flow simulations to construct optimal predictive models of dissipation bursts. It is shown that, to maintain relative predictive skill, the minimum model complexity (size) and the cost of predictions increase exponentially with the forecast horizon. This limitation is connected to causal uncertainty, whereby the number of states from which an extreme event may emerge increases with the forecast horizon. It is argued that, to maintain accuracy, data-driven models need to encode theses states, increasing their size and the amount of data necessary for training.

Complexity of extreme-event prediction in turbulent flows Articles