event
Monday 17 May 2021: 1st year Phd talks
Tim Andrews and Nell Hartney -
online 09:30-10:30
Speaker: Tim Andrews
Title: Utilising low frequency dynamics to improve time-stepping capabilities in weather and climate models
Abstract:The ability to take larger time steps is a major aspiration for weather and climate modellers. We seek to develop a method for doing so by considering features that are commonly considered a hinderance. Our approach utilises the fast waves as an exponential mapping to a system with a smoother gradient. This ‘modulation variable method’ will ideally be an initial predictor step in a multi-stage algorithm. This will be followed by correction steps, which should adjust for any deterministic phase error generated by the large time step. An application of this algorithm to a ‘swinging spring’ system at resonance has already proved effective. A physical understanding of low-frequency behaviours will ensure this solution retains the important dynamics. Analysis of power spectra in wavenumber-frequency space is proposed as a means of identifying these key features. Specifically, it is hypothesised that near-resonance interactions from quadratic non-linearities are crucial for shifting energy into low frequencies. This analysis procedure will also enable comparisons between numerical methods that use a larger time step.
Speaker: Nell Hartney
Title: Next generation time-stepping schemes for weather and climate prediction
Abstract: Producing timely and accurate weather and climate forecasts is dependent on the effective use of supercomputing power. It is anticipated that future developments in computing performance will come, not from the use of more powerful processors as has been the trend until now, but from the addition of more processors, meaning that future supercomputers will make their gains from exploiting routes to parallelism. Parallel systems with a latitude-longitude spatial grid, though, suffer communication bottlenecks related to grid geometry. Compatible finite element methods provide spatial discretisations that solve this problem. We are interested in how, using this spatial discretisation, the time discretisation can be structured to make most efficient use of parallel computing systems. Developing such a time-stepping technique with a model that includes moisture allows moist physics to be incorporated in the scheme. Using the Firedrake finite element library, we have are investigating frameworks for the inclusion of moisture in the shallow water system, with the aim of extending the test suite of the shallow water system to more realistic situations that incorporate moisture, and using moist shallow water models to investigate different time-stepping schemes.
