| 12 Jul 2001 | http://www.cgd.ucar.edu/oce/kauff/mywork.shtml |
| Brian Kauffman | NCAR | CGD | OCE | B. Kauffman |
My first project at NCAR involved building an interactive 3D graphics
utility. The Oceanography Section was interested in more advanced
graphics capabilities, so they acquired a Stardent graphics
mini-supercomputer and hired three CU students (myself included)
to build a software tool we called "OceanVu".
On the right is an example of some output.
OceanVu was written entirely in C,
used an object oriented graphics library called Dore',
and used netCDF as a data format.
Next I began working with Oceanographer/Scientist Peter Gent, who, among other things, was interested in ENSO studies. My work here involves the continued development and execution of computer code which is a high resolution Pacific Basin ocean model . At first this code was run uncoupled, forced with climatological data. Later we (Gent, Tribbia, Lee, Kauffman) began coupling this ocean model with NCAR's CCM2 atmospheric model. While we had some significant success with this coupled model, we also learned quite a bit about the deficiencies of the computer models and the standard coupling strategies of the time. Drawing on our experiences, we starting forming ideas about how to combine the various components of a coupled climate system. We (also with McWilliams, Bryan, et. al.) began designing and building a proto-type modular coupled climate model, but it became clear that such a project would require a sustained, coordinated effort from a larger group of researchers.
As a larger group of NCAR scientists became interested in coupled climate models,
our proto-type model was adopted as the basis for a new, NCAR-wide proposal,
the Climate System Model
(CSM) Project.
The CSM project formally began in January 1994 with the long-term goal of building,
maintaining, and continually improving a comprehensive model of the climate system.
The comprehensive CSM is presently composed of a set of four independent models,
atmosphere, ocean, land surface, and sea ice,
each using message passing to communicate with a "Flux Coupler."
Most of my time is now spent on the CSM project, in particular,
developing the Flux Coupler code (more info below).
As I've said, I'm developing the Flux Coupler (aka "Coupler" or "driver") code. The Flux Coupler is software that allows separate climate model components (eg. atmosphere, ocean, sea ice, land surface) to be connected together in a modular fashion. The Coupler coordinates and controls the integration of, and interaction (flux) between, component models, dealing with disparities in spatial grids and time stepping schemes. Data is passed between the Coupler and component models using a message passing technique. This allows the component models to be developed as separate executable codes that may be distributed amongst several computers.
In conjunction with this code development, I ran some of the initial coupled simulations. Initial CSM simulations involved coupling:
Note that this was an initial configuration --
a main point of the Flux Coupler concept is that alternate component
models can be swapped in with relative ease.
This configuration requires roughly 50MW of memory
and 50 CPU hours per simulated year on a sixteen processor
Cray C90. Thanks to the
Climate System Labratory,
(CSL) at NCAR,
we have the computing resources necessary to support such an undertaking.
The initial configuration (code & simulation results) was presented at the
1st annual CSM workshop held in Breckenridge, Colorado in May of 1996.
Since then CSM models have been continually improved upon.
Improvements include upgrades in model physics,
more choices of model resolutions,
and adapting the codes to run on a variety of computer archetectures.
New and improved CSM components are continually under development
and are released periodically.