Joint Seminar: Simulation of heat waves in climate models using large deviation algorithms

The study of extremes in the climate system is hindered by both a lack of past observational data for events with a return time larger than decades or centuries, and by the large computational cost required to perform a proper sampling of extreme statistics with state of the art climate models. The study of the dynamics leading to extreme events is especially difficult as it requires hundreds or thousands of realizations of the dynamical paths leading to similar extremes.

We will discuss here a new numerical algorithm, based on large deviation theory, that allows to efficiently sample very rare events in complex climate models. A large ensemble of realizations are run in parallel, and selection and cloning procedures are applied in order to oversample the trajectories leading to the extremes of interest. The statistics and characteristic dynamics of the extremes can then be computed on a much larger number of events. This kind of importance sampling method belongs to a class of genetic algorithms that have been successfully applied in other scientific fields (statistical mechanics, complex biomolecular dynamics), allowing to decrease by orders of magnitude the numerical cost required to sample extremes with respect to standard direct numerical sampling.

We study the applicability of this method to the computation of the statistics of European surface temperatures with the Planet Simulator (Plasim), an intermediate complexity general circulation model of the atmosphere. We demonstrate the efficiency of the method by comparing its performances against standard approaches, focusing on the realization of temperature extremes corresponding to heat waves. We then discuss the feasibility of this method for applications with state of the art climate models in more realistic settings.

Datum

26.01.2016

Uhrzeit

15:15 Uhr

Ort

Bundesstr. 53, room 022/023
Seminar Room 022/023, Ground Floor, Bundesstrasse 53, 20146 Hamburg, Hamburg

ReferentIn

Francesco Ragone, École normale supérieure de Lyon

Chair

Gualtiero Badin

Zur Übersicht