NG006: Ensemble Methods for Combining Climate and Weather Models
Software Engineering Designs for Super-Modeling Different Versions of CESM Models using DART
*Kluzek E.
erik@ucar.edu
National Center for Atmospheric Research, PO Box 3000, Boulder, CO 80307, USA
Duane G.
gregory.duane@Colorado.EDU
2University of Colorado, Department of Atmospheric and Oceanic Sciences, Boulder CO, USA, and Macedonian Academy of Sciences and Arts, Skopje,
Republic of Macedonia
Tribbia J.
tribbia@ucar.edu
National Center for Atmospheric Research, PO Box 3000, Boulder, CO 80307, USA
Vertenstein M.
mvertens@ucar.edu
National Center for Atmospheric Research, PO Box 3000, Boulder, CO 80307, USA
We explore different designs to connect different versions of the IPCC class climate model the Community Earth System Model (CESM) together. We use the Data Assimilation Research Test-bed (DART) software to provide data assimilation as well as provide a software framework to link different model configurations together. We also show results from some simple experiments that demonstrate the ability to synchronize different model versions together.
The CESM (version 1.2.0) is made up of different component models that couple ocean, atmosphere, land, sea-ice, glacier, and ocean wave models together. Each component can be different versions of either active, data, or stub models. Although there is usually only one active model for a component, most components are build-time or run-time configurable enough that you can come up with a total model system that is significantly different from other model configurations. Hence, you can run the atmosphere model with CAM3, CAM4, and CAM5 (Community Atmosphere Model versions 3, 4 and 5) physics for example as well as run the land model with CLM4 and CLM4.5 (Community Land Model versions 4 and 4.5) geophysics while also choosing between: SP, CN, and BGC (Satellite Phenology, Carbon-Nitrogen, and CLM45. -biogeochemistry) biogeochemistry modes. The sea-ice model can be run in prognostic or prescribed ice concentration modes, while the POP (Parallel Ocean Model) ocean model can be run with different settings of the vertical and horizontal mixing. You can see where a mix of these different options gives you substantially different model versions especially when you consider the system en tot ale.
The data assimilation part is separate from the synchronization of the models and can be done at a different temporal frequency or not at all. We explore different designs for the data assimilation all the way from simple nudging to using the full capability of the Ensemble Kalman Filter (EKF) that is the central part of DART. DART’s EKF can be used to synchronize multiple fields of the models together or simple nudging can be used to synchronize a single field.
A significant hurdle to the project is that the linkage between DART and CESM for this version of DART and CESM is via files. Also the different model configurations of CESM HAVE to be different at build-time. This means you have different programs that need to initialize and then run writing out data that other programs then read in and operate on, and then feed back to other programs all controlled by high level scripts. Since, climate models normally run for long lengths of time, initialization is typically inefficient and may not even be fully parallel. By starting and stopping the model so often the inefficiency of the models initialization becomes more apparent and has a large impact on workable configurations for climate studies. We explore different options to make this workable by using low resolutions, and limiting the temporal frequency for both data assimilation as well as the frequency that the models are synchronized together.