University of Colorado
ESMF was used to introduce a new atmospheric dynamical core into the GEOS-5 atmospheric general circulation model. The dynamical core is the part of the atmospheric model that solves the governing fluid dynamics equations. The new core was developed by S.J. Lin (NOAA GFDL) and W. Putman (NASA GSFC). It can be run in either hydrostatic or non-hydrostatic modes and scales well to thousands of processors. The GEOS-5 model continues to support its original latitude-longitude finite-volume core.
The new component was implemented in GEOS-5 by embedding it in an ESMF wrapper that makes it plug-compatible with the rest of the GEOS-5 system, including the atmospheric physics and chemistry components and the land hydrology model. In this wrapped form it also relies on the ESMF-based machinery of GEOS-5 for its checkpoint restart capability and for all diagnostic output, making it also compatible with other parts of the system, such as the ocean model and the data assimilation components.
The figure above shows cloud fields associated with a cold air outbreak on 1 January 2009. It compares 1-day forecasts with various resolutions of the cubed sphere version of GEOS-5 with satellite imagery from that day. The three resolutions use roughly 28km, 14km and, at the lower left, 7km grid boxes. At the highest resolution (7 km), the core is run in non-hydrostatic mode. At the low resolution (28 km), the model scales to ~10,000 cores, if one ignores I/O associated with initialization and diagnostics. At the higher resolutions the model has been run with over 20,000 cores before problems occurred with the core's communication performance.
Since these are short forecasts done from a low-resolution (1/2-degree) analysis, we do not expect much change in large-scale features with increased resolution. Even after a few hours, however, the model begins to show much more realistic small scale structures. Particularly exciting is the simulation of organized shallow convective clouds in the cold air outbreak south and east of Japan. Models capable of producing such fine-scale structures will allow us to extract much more information from the high resolution data coming from satellites.
These runs were performed on the Cray XT5 at Oak Ridge National Labs. Longer forecasts and free-running climate runs to study the behavior of the model at these resolutions are now in progress.
Image courtesy of the GEOS-5 team at NASA GMAO.