[GTP] Reminder- GTP Seminar at NCAR on June 24- Will be webcast and recorded

[Silvia Gentile]

***This session will be webcast and recorded 
http://www.fin.ucar.edu/it/mms/ml-live.htm****
Monday, June 24, 2013
Mesa Laboratory, Main Seminar Room
Lecture at 2:30pm

STOCHASTIC SUPERPARAMETERIZATION
Ian Grooms
Center for Atmosphere Ocean Science
Courant Institute of Mathematical Sciences, New York University


Efficient modeling of unresolved small-scale turbulence is of primary 
importance in simulations of large-scale geophysical fluid dynamics. In 
many geophysical and astrophysical settings the unresolved turbulence is 
not homogeneous/isotropic, being affected by rotation, stratification, 
moist processes, magnetism, etc., and the multiscale interactions with 
the resolved large scales are complex and consist of more than 
inertial-range energy transfer. Furthermore, small-scale feedback to the 
resolved scales is not completely determined by the resolved large 
scales. The random nature of the small scales requires stochastic 
models, which in turn can improve the robustness of ensemble-based 
prediction and state estimation algorithms.
Superparameterization is a multiscale framework that models unresolved 
scales by PDEs evolving on pseudo-physical domains embedded into the 
coarse grid of a general circulation model. Although the small-scale 
PDEs are deterministic, their chaotic/turbulent dynamics generate an 
effectively stochastic feedback to the large scales. Though successful 
in modeling tropical atmospheric moist convection, superparameterization 
remains computationally costly, and of limited generality.
We develop an improved framework for superparameterization that models 
the small-scale turbulent dynamics by stochastic, quasilinear PDEs 
rather than nonlinear, deterministic ones. This greatly improves the 
efficiency of the algorithm, and our mathematical framework for 
developing the large- and small-scale PDEs increases the generality of 
superparameterization. The resulting algorithm is developed and tested 
in two idealized turbulent models: the one-dimensional complex-scalar 
MMT equation, and two-layer quasigeostrophic turbulence. In both 
settings the algorithm achieves several orders of magnitude of reduction 
in computational cost compared to direct simulation of all scales, and 
produces qualitatively accurate results. This is particularly impressive 
in the quasigeostrophic tests where the algorithm successfully 
parameterizes the inverse cascade of kinetic energy from unresolved to 
resolved scales. Future directions include more realistic applications, 
and optimization of the numerical algorithm.



-- 
Silvia Gentile
NCAR IMAGe
1850 Table Mesa Drive
Boulder, CO 80305
303 497 2480
www2.image.ucar.edu