|
NEMS
Version 1.1.0
This is a living document and may be updated at any time.
|
|
|
NEMS
Version 1.1.0
This is a living document and may be updated at any time.
|
|
This page describes the main NEMS mediator. There is also a NEMS space weather mediator.
In the NUOPC Layer architecture, the mediator (often called the coupler) handles the scientific and customized aspects of coupling for a modeling system. For example, it may handle transfer of coupling fields, regridding, merging, treatments of coastlines or custom field transformations.
In NEMS, the mediator is a separate gridded component with multiple phases. Phases can be thought of as different subroutines within the mediator that can be called in some sequence to carry out the work of the mediator. The mediator runs on its own set of PETs (persistent execution threads, similar to processors). Often the PETs chosen for the mediator overlap with other components to optimize resource utilization. The PETs and run sequence are specified by the compset. Mediator phases might have names like prep_atm, prep_ocn, restart, accumulate, and so forth. The way the mediator phases are implemented is somewhat arbitrary and will probably evolve over time as sequencing requirements change.
The mediator PETs are set in NEMS in the nems.configure file. Other mediator attributes are also defined there in a section that looks like this,
MED_model: nems MED_petlist_bounds: 76 135 MED_attributes:: Verbosity = 0 DumpFields = true DumpRHs = false coldstart = false restart_interval = 0 ::
The main NEMS mediator is capable of technically coupling atmosphere, ocean, sea ice, land, hydrology, and wave components. All of these component types have demonstrated techically correct passage of fields through the mediator. Only the behavior of atmosphere, ocean, and sea ice components has been examined to assess correct physical coupling.
All components do not have to be present to operate the NEMS mediator.
The mediator and all components advertise fields during initialization. NUOPC reconciles those field names and then coupling fields are realized and connected. In order for a field to be connected between two components, it has to be advertised in the two components as well as in the mediator. Field names that match between components are automaticallly coupled in the system. Any field that is exported from one component and imported to another component with the same standard name is coupled between those components.
There is a slow and a fast coupling period in the NEMS mediator. The slow coupling period is associated with the ocean model and allows the ocean model to be coupled at a lower frequency than other components. The fast coupling period is for the atmosphere and ice model. They are coupled at the same frequency in the system.
The mediator accumulates all fields between coupling periods for all components. For example, the atmosphere and ice coupling fields are accumulated and averaged between calls to the ocean model. At the same time, the ocean fields coupled to the atmosphere and ice models are held static between the longer ocean coupling periods.
Model grids are passed to the mediator at initialization. The mediator receives those grids and instantiates a decomposition of those grids on its PETs (persistent execution threads, similar to processors). The mediator is then able to receive all model fields on the native model grids.
The ocean and ice components will generally be run on the same grid for science reasons, but the mediator is implemented such that this is not a requirement.
Regridding is performed by the mediator. The regridding weights are computed at initialization and depend on the grids and regridding method. The regridding method is defined in the mediator on a field-by-field basis when the field is defined. In general, fluxes are mapped conservatively, and states are mapped bilinearly. The higher order finite element patch method was not used for any fields because of an observed reproducibility issue.
In the current revision, fields transferred from the ocean to the sea ice component are copied rather than interpolated. This can be done because the ocean and sea ice components are on the same grid. When different grids are detected for these components, the interpolation method defaults to bilinear.
The run sequence is evolving as new requirements are defined. There are several mediator phases currently implemented in the mediator and a typical UGCS-Seasonal run sequence is set up as follows as shown in a typical nems.configure file,
runSeq::
@1800.0
MED MedPhase_prep_ocn
MED -> OCN :remapMethod=redist
OCN
@900.0
MED MedPhase_prep_ice
MED MedPhase_prep_atm
MED -> ATM :remapMethod=redist
MED -> ICE :remapMethod=redist
ATM
ICE
ATM -> MED :remapMethod=redist
ICE -> MED :remapMethod=redist
MED MedPhase_atm_ocn_flux
MED MedPhase_accum_fast
@
OCN -> MED :remapMethod=redist
MED MedPhase_write_restart
@
::
In the file above, the sequence of component run calls (ie. OCN, ICE, ATM), field coupling via connectors (ie. ATM to MED), and mediator phases (ie. "MedPhase_prep_ice", "MedPhase_atm_ocn_flux", MedPhase_write_restart) are indicated. The coupling periods are also defined (ie. @1800.0, @900.0) for the ocean and atmosphere/ice models.
The current implementation of the mediator phases does the following,
MedPhase_prep_ocn - prepares the ocean coupling fields for the ocean model by averaging the atm/ice coupling fields, interpolating fields to the ocean grid, computing any merges or custom field calculations, and then filling the ESMF State sent to the ocean model.
MedPhase_prep_atm - prepares the atmosphere coupling fields for the atmosphere model by interpolating fields to the atmosphere grid, computing any merges or custom field calculations, and then filling the ESMF State sent to the atmosphere model.
MedPhase_prep_ice - prepares the sea ice coupling fields for the sea ice model by interpolating fields to the sea ice grid, computing any merges or custom field calculations, and then filling the ESMF State sent to the sea ice model.
MedPhase_atm_ocn_flux - computes the atmosphere/ocean fluxes from atmosphere and ocean fields. The computation is done on the ocean grid and data from the atmosphere model are interpolated to the ocean grid. These fluxes can be used in the atmosphere and ocean model.
MedPhase_accum_fast - accumulates the atmosphere and ice coupling fields for coupling to the ocean model.
MedPhase_write_restart - writes mediator restart files.
The land mask implementation is described in more detail here.
The NEMS mediator uses the convention that heat/water/momentum flux and wind stress is positive downward. There is also a hierachy of "down" with respect to models, which from top to bottom is:
If a flux in the coupler is positive, that means it's transferring heat/water/momentum downward from a higher component to a lower component.
Model components have to have consistent conventions for fluxes that NEMS mediator knows about so that it can adjust signs when passing data from one component to another.
The NEMS system couples heat, water, and momentum fluxes. In general, fluxes are computed at the interface between two components and those fluxes are often computed in one of the two components or in the mediator. Fluxes are normally interpolated conservatively, but the best regridding method for a particular application is a science issue.
The mediator is able to compute fluxes as well as regrid and couple fluxes computed in components. There is no specific constraint on how or where the fluxes should be computed. Often the choice depends on the relative coupling frequency, the amount of information required to be coupled to compute the fluxes, the sophistication of the flux calculation, the relative grid resolution, and whether an exchange grid is used.
For some components, fluxes need to be merged. For instance, on the atmosphere grid, the ice fraction might evolve in time and the land fraction might be static. For conservation, the merging of fluxes is as important as the regridding and reuse of fluxes. Again, this is a science issue.
To compute fluxes,
1.8.13