Atmosphere Ocean Science Colloquium

Stochastic Superparameterization

Speaker: Ian Grooms, NYU/Courant

Location: Warren Weaver Hall 1302

Date: Wednesday, November 6, 2013, 3:30 p.m.

Synopsis:

Two of the primary challenges in climate-atmosphere-ocean science are to develop efficient models without resolving the entire range of scales of the turbulent dynamics, and to develop robust methods of state estimation and prediction with quantified uncertainty. Many existing methods for developing efficient multiscale models rely on strong scale separation in space and/or time, but such scale separation can be limited or non-existent in applications like eddy-permitting ocean modeling. Existing models for data assimilation and prediction incur errors by equating the variables of coarsely-resolved models with the variables of the true system. Superparameterization is a multiscale method where the dynamics of the unresolved scales are simulated on local domains embedded within the coarse computational grid, and the solutions are used to compute the feedback to the large scales. This talk presents a formal mathematical framework for superparameterization, and extends it by developing quasi-linear stochastic models of the small-scale dynamics on the embedded domains, which helps mitigate the lack of scale separation. The method is tested in two idealized but difficult turbulent systems, with great success. Superparameterization provides estimates of the small-scale variable at each coarse grid point. A novel ensemble kalman filtering framework is developed to take advantage of this extra small-scale information: the algorithm is then tested with good results on one of the previous test problems for stochastic superparameterization.