CSAR Seminar
SPEAKER: Aaron Herrnstein, University of Louisville
TITLE:
Adaptive Mesh Refinement Applied in a Numerical Ocean Model
DATE: Wednesday, August 2, 2006
TIME: 12:00 Noon
PLACE: 2240 DCL
1304 W. Springfield Ave., Urbana, IL
ABSTRACT
Numerical models of the global ocean span inherently large time and
space scales resulting in extremely long run times (~months) for high
resolution simulations, even on the fastest supercomputers available
today. Uniform high resolution is often unnecessary in applications
where areas of interest are localized. In such situations, computation
can be reduced by implementing local increases in resolution. Adaptive
Mesh Refinement (AMR) is often used to achieve such tasks in
computational fluid dynamics. Data are maintained in a hierarchy of
refinement levels which are subdivided into a collection of refinement
patches that change size and position as the simulation evolves in
time. The application of AMR in ocean modeling has not progressed far,
in part due to the use of numerical methods which are not suitable for
the traditional AMR methodology.
An ocean model with AMR capabilities was developed a Lawrence Livermore
National Laboratory using the software library SAMRAI (Structured
Adaptive Mesh Refinement Application Infrastructure) as
its foundation. Amendments were made to the library which allow
numerical methods common in standard ocean models to be used in AMR
simulations. The added capabilities include leapfrog time integration,
B-grid staggering of tracer and momentum variables, and time refinement
of coupled systems with unequal time steps. The AMR Ocean Model
operates on top of the amended SAMRAI library using basic components of
pre-existing ocean models. This presentation will discuss the AMR
algorithms and strategies developed in this research followed by
results of the AMR Ocean Model applied in practical applications such
as carbon sequestration via direct injection. Small scale features are
revealed that are unattainable with the model at a uniform high
resolution.