University of Colorado
In trying to spin up a development version of CESM1 using the spectral element dynamical core and the latest CAM physics package (CAM5), unexpected behavior was discovered with the ocean component. It was not clear whether this behavior was caused by the differences in the grid in the polar regions, when compared to the finite volume dynamical core (and the corresponding effect on the resulting ocean surface stresses). To determine if this was the case, we decided to try an experiment where the grid was refined by a factor of 4 in the top and bottom cubed sphere faces. We wanted to run a fully active 1850 model configuration with the atmospheric and land components running on this refined grid and the ocean and sea-ice running on the standard displaced pole grid. With the availability of the ESMF offline generation utility and the corresponding grid generation infrastructure that CSEG has created on top of this utility, the CESM Software Engineering group was able to get this configuration up and running in about 1 week (with most of the effort coming from working with CAM scientists to get atmosphere correctly stabilized on startup). This task would have been impossible without the regridding capability that ESMF has provided. The end result was that the refinement did not make a difference in the spinup problem and we were able to show this at the AMWG meeting. Overall a great success story in my opinion.
Mariana Vertenstein, NCAR Climate and Global Dynamics Division
GrADS has a requirement that all gridded data must be handled internally on arectilinear grid. In order to access data on curvilinear grids in their native format,I use the ESMF_RegridWeightGen tool to generate a set of interpolation weights. I then use a short customized C program which translates the ESMF weights into a specific format that GrADS requires in order to perform the interpolation on the fly, at the moment each I/O request is made. ESMF's tool has made it possible to access a large class of data from ocean and sea ice models that could not otherwise be handled by GrADS. This process was made even easier when RegridWeightGen was modified to accept a CMIP5-formatted data file as input.
p.s. In case you didn't know this already, the technical support I got from Peggy Li was exemplary and invaluable.
Jennifer Adams, Center for Ocean-Land-Atmosphere Studies
NCL's implementation of the ESMF regridding methods has been very beneficial to the scientific community using model and/or observational data sets. The source and destination grids can be rectilinear, curvilinear or unstructured. Further, the ESMF regridding tool allows users to choose from three interpolation methods (bilinear, patch and conservative) to better address scientific needs. For example: (a) it is being used in CESM post processing tools to regrid conventional and experimental, ‘complicated’ model grids developed for computational and physical reasons; (b) satellite swath data, which often has random missing data on curvilinear grids, is being regridded to facilitate comparisons with forecast model output.
Dennis Shea, NCAR Climate Analysis Section
Before we used ESMF regridding in NCL, we had an adhoc collection of regridding/interpolation routines, all with different interfaces, certain restrictions about the kinds of grids they could handle, and limited success with the results.
Now that we have ESMF regridding, this has provided us with a unified interface with support for rectilinear, curvilinear, and unstructured grids. We get a lot of questions on regridding, and having this ESMF interface is allowing us to provide users with much more functionality than we could before, and sometimes with better numerical results. I don't think we've even tapped into the full functionality of ESMF regridding, as we haven't used some of the special pole options (although we do make it available).
Finally, the email support that we get from the ESMF group is crucial to the success of this software. The close collaboration has worked out really well and I hope it continues to do so.
Mary Haley, NCAR NCL Lead
In our work to implement the MPAS models -- both the ocean and the non-hydrostatic atmosphere -- in CESM, we've made particular use of the ESMF_RegridWeightGen program to generate the remapping weights between MPAS's unstructured spherical centroidal Voronoi tessellation meshes and the grids of other CESM components. The flexibility and parallel capability of ESMF_RegridWeightGen has been a great benefit, and the ongoing development and very responsive support have ensured that our progress hasn't been slowed down by any issues in remapping fields between Earth system components.
Michael Duda, NCAR Mesoscale and Microscale Meteorology Division
The ESMF utility for generating mapping weights has permitted us to generate bilinear mapping weights from the CAM atmosphere HOMME cubed sphere grid to the POP ocean displaced pole grid. This is a capability that did not previously exist and that was required for upcoming simulations. We are leveraging it to carry out the first coupled simulation using all active components with HOMME as the atmospheric dynamical core. We have carefully verified that the mapping weights (both conservative and bilinear) are correct and the tool seems very robust and is definitely helping CESM development efforts.
Mariana Vertenstein, NCAR Climate and Global Dynamics Division
ESMF regridding tools [via ESMP] have become indispensable in the support that PMEL provides to the climate community through the Ferret and Live Access Server (LAS) applications. Ferret is a model visualization and analysis application for the desktop; LAS is an analogous tool for Web users. The ESMF software allows PMEL's users to intercompare models on different coordinate systems -- curvilinear in the horizontal and sigma in the vertical. The ESMF tools are several times faster than the utility code that we had previously been using. Future versions of Ferret and LAS will depend upon the ability of the ESMF tools to handle unstructured grids, as well.
Steve Hankin, NOAA Pacific Marine Environmental Library
The ESMF regridder quickly maps values between grids of different types (structured to unstructured, and back again) and is the only regridder in the [CSDMS] framework that is mass conservative. In addition, it is the only regridder that can be run in parallel. Much of the CSDMS framework is written in Python so the Python bindings for the ESMF regridding tools (ESMP) have been invaluable.
Eric Hutton, Community Surface Dynamics Modeling System
The primary goal of the NASA Global Modeling and Assimilation Office (GMAO) is to develop a set of multi-component GEOS-5 systems and bring them to “operational” readiness. GEOS-5 systems provide products for NASA instrument teams, for the Modern-Era Retrospective Analysis for Research and Applications (MERRA) and for the conduct of quasi-operational experimental weather and seasonal climate forecasts. These goals have been achieved by implementing the Earth System Modeling Framework (ESMF) across all GMAO models and assimilation systems. GEOS-5 has a comprehensive ESMF implementation, allowing relatively easy incorporation of other ESMF-compliant modules (e.g., GFDL MOM4, LANL CICE, GOCART, GMI chemistry).
Arlindo da Silva, NASA Global Modeling and Assimilation Office
The Earth System Modeling Framework (ESMF) has been (and remains) critical in the development and transition of the Navy’s Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPSTM). Integration of current operational models (atmosphere, ocean, wave, and sea ice) into COAMPS has been enabled through the component and data type abstractions provided by ESMF. Key to this integration was ability to implement a “coupling” layer in each model while retaining the “stand-alone” capability in the same model source code. The parallel sparse-matrix multiply and regridding capabilities provided by ESMF have been important in providing efficient data transformation between models in COAMPS. New developments in COAMPS and the Navy Earth System Prediction Capability (ESPC) are focused on adopting the National Unified Prediction Capability (NUOPC) Common Model Architecture. Implementation of the NUOPC Layer into the model components will further improve interoperability of these models with multiple coupled systems.
Tim Campbell, Naval Research Laboratory