Tuesday 16 Jul 2013: CliMathNet seminar: Multiscale Methods for the Multiscale Environment of Climate Simulation
Rob Scheichl - University of Bath
Harrison 209 16:00-17:00
To combine high resolution atmospheric flow models with state-of-the-art data assimilation and statistical analyses is one of the biggest current challenges for numerical weather and climate prediction. In this talk we will show how hierarchical multiscale or multilevel methods can be designed to tackle this challenge. They are essential as scalable PDE solvers for individual simulations, but their biggest potential lies in accelerating standard Monte Carlo (or ensemble-type) methods for statistical analyses and for data assimilation, including Markov chain Monte Carlo methods which are of great interest also in the atmospheric sciences. Monte Carlo methods are notorious for their slow convergence and high computational cost. In this talk we will present revolutionary recent developments to mitigate and overcome this serious problem using novel multilevel strategies that can lead to a more than 100-fold acceleration of existing ensemble-type methods. The talk will focus on methodology. Most of the applications are so far mainly coming from subsurface flow.
(seminar given via access grid)