Cecelia DeLuca, Byron Boville, Anthony Craig,
Nancy Collins, Erik Kluzek, John Michalakes,
David Neckels, Earl Schwab, Shepard Smithline, Jon Wolfe
National Center for Atmospheric Research
and
Arlindo da Silva, Max Suarez, Carlos Cruz,
William Sawyer, Atanas Trayanov, Leonid Zaslavsky
Global Modeling and Assimilation Office, NASA
and
Mark Iredell, Weiyu Yang
National Center for Environment Prediction, NOAA
and
Robert Hallberg, Balaji
Geophysical Fluid Dynamics Laboratory, NOAA
and
Chris Hill
Massachusetts Institute of Technology
and
Robert Jacob, Jay Larson
Argonne National Laboratory
and
William Sawyer (presenter)
Global Modeling and Assimilation Office, NASA, and
Institute for Atmospheric and Climate Science, ETH Zurich
email: will.sawyer@gmao.gsfc.nasa.gov
The Earth System Modeling Framework (ESMF) project is developing a standard software platform for Earth system models. The standard, which defines a component architecture and a support infrastructure, is being developed using open-software practices by major research and operational modeling institutions in the US.
ESMF allows earth scientists to develop complex software models with numerous components in a coordinated fashion by defining a standard, but highly flexible, interface. The interface concept developed by the ESMF group is both self-describing and extensible, easing collaboration between model component developers without being overly prescriptive.
ESMF also supplies a standard infrastructure to run the models efficiently on high performance computers which can use different paradigms to attain parallelism, e.g. message-passing and multithreading. The infrastructure offers capabilities that are commonly needed in Earth Science applications, for example, support for a broad range of discrete grids (such as finite-difference, finite volume and spectral grids), regridding functions which map data from one grid to another in a consistent manner, data streams for observational data, and a distributed grid class which represents the data's decomposition into sub-domains to support parallel computation.
A first implementation of ESMF has been available since May, 2003 (see http://www.esmf.ucar.edu). Already numerous development groups from existing research and operational enterprises have begun to adopt ESMF.
In this paper we present first an overview of the ESMF architecture, and subsequently two model configurations employing ESMF are discussed. The first is an atmospheric model consisting of an ESMF component using finite volumes to calculate the state of the atmospheric dynamics at a given time step, and another component which describes the physics processes in the atmosphere using Held-Suarez forcings. The second application is a data assimilation system integrating a General Circulation Model (GCM) ESMF component with a data analysis component. The emphasis in these experiments is on the parallel computing aspects - integration effort and parallel efficiency - of these models. ESMF programming examples are taken from these experiments, and some empirical conclusions are drawn about ESMF's impact on future model development practices.