Abstract

Historically, most of the efforts to parallelise the weather and climate models are put into spatial discretisation, where the atmosphere is divided into a number of overlapping domains with data being shared across the overlap. There is a limit to this way of parallelization eventually, when there is not enough computational work to be done on a very small domain while waiting for the data from the neighboring domain to arrive.

Recently the parallelism in the time discretisation have been identified as a possible way to increase concurrency in climate models. There are a couple of ways to provide parallelism in the time direction for the numerical solution of ordinary and partial differential equations. In this talk, we’ll explore a phase averaged model of the rotating shallow-water equations which computes the solution in parallel within each timestep. This is a different approach to a usual parallel-in-time methods where solutions for multiple time steps are computed in parallel. I will present a phase-averaging framework for the rotating shallow water equations, a time integration methodology for it, proof-of-concept results and error analysis in order to examine the impact of the phase averaging on the model solutions.

The zoom link is https://cam-ac-uk.zoom.us/j/89029983641?pwd=Zyt5ODZ4eThkcWtsd2pjbjRqK3hYUT09 (Password 309683)