GEOS Integration Results

Four components of GEOS has been ported to NDSL:

Finite Volume Cubed (FV3) Dynamical Core
Geophysical Fluid Dynamics Lab Single-Moment Microphysics
Grell-Freitas Convection Parameterization
University of Washington Shallow Convection And Moist Turbulence Scheme

These ports maintain numerical accuracy, and strive for numerical equivalence. Throughout the process, redundant code was consolidated, unused code was removed, and poorly organized code was restructured. These improvements often depart from numerical equivalence; however, it has been verified that - at a given timestep - a field has errors of no more than six orders of magnitude less than the observed value or 10^-11 (whichever is greater), and if such a condition is met it can be said that numerical accuracy is maintained.

Ported code is tested for numerical accuracy by running "translate tests" - a numerical comparison of the ported code and the original Fortran code on a single timestep in isolation from the rest of the model. Validation has been performed using Single Column Model (SCM) runs and General Circulation Model (GCM) runs. For SCM experiments, the ported code is verified independently on each of the first 20 timesteps. For GCM runs (at C24 horizontal resolution), only the first ten timesteps are verified. A successful, or "passed" test signals that numerical accuracy has been maintained in the ported code.

Next, scientific validity is assessed by comparing the output of a full model runs (SCM or GCM) using either the Fortran and NDSL component. This analysis has been performed with three SCM experiments ("bomex", "armtwp-ice", and "armtwp-july97") and a GCM run initialized on 14 April 2000. SCM experiments ran for their entire durations, while GCM runs were limited to 10 days.

More information on the benchmarking process coming soon...