CMIP6-BMBF

In Stephen Griffies paper (http://www.geosci-model-dev-discuss.net/gmd-2016-77/) are tables which variable are to be delivered in addition on a spherical grid.

The WIP has been asked about this issue. The following is an extract of a WIP draft paper (courtesy of Karl Taylor, 17 Nov 2016):
" ...

Interpolation to standard grids¶

Data on a regular grid is preferable for many users, but there are other users who require data on the native grid of the model.

In the CMIP6 data request, the MIPs have specified for each variable whether they have a preferred or required output grid.

In 01.beta.37 of the data request API, two "grid policy" switches are added to support different approaches to dealing with the range of different requests.

The first is a global default to be used when there is no explicit preference expressed for a variable (note that there may be different requests for different experiments).

With 01.beta.37, having a global default of "native" leads to a volume estimate of 483Tb, with more than 50% coming from ocean model level data.

If the default is set to a 1 degree regular grid, the volume estimate drops to 295.5Tb. ...".

And here a comment by Karl Taylor:
" ... At the WGCM meeting last month there was no agreement to regrid data to some standard grid, so currently modeling groups may choose to report data on native grids and/or regrid it to one or more target grids of their choice. There was support (but no commitments) for providing a file containing mapping "weights" needed to regrid data (via matrix multiplication) from the grid chosen by the modeling group to a standard grid. I think this is both the right approach and will happen, although perhaps not immediately.

There was talk that "someone" (ESGF? DDC?, ????) might take on the task of regridding some subset of the contributed model output to the standard grid(s) using the weights provided by the modeling groups. That would especially be helpful for non cartesian latxlon grids. ..."

And a comment by myself:
I guess that for operational workflows this issue is not so urgent, because the interpolation will presumably not be done before a simulation is finished. However, we should ask the WIP to ask the MIPs to give advice about the interpolation method to be used for any requested interpolated variable (smoothness, conservation, positiveness, preservation of gradients, etc.). The best would be if the interpolation method could be prescribed. What Karl describes as 'right approach' is about what was required for CMIP5 variables on irregular grids, though not followed by all groups: provide a script that interpolates data on irregular grids with the data.

Maybe AP6 (ESMValTool) has different requirements?

e-mail from Julie to OMIP-list (Feb 14, 2018):

IPSLCM6 is just about to start production of netcdf files to be archived on ESGF nodes, and I am hitting back on the question of which variables to interpolate on a regular grid.
Engineers from my group extracted from the Data Request xls files, the list below of variables to be interpolated on 1degree and 2 degree regular grids (a mix of ocean and biogeochemical variables). None of the varialbes on this list seems relevant to me, for interpolation…

In addition, this list contains NO variable which I had in mind to provide on a regular grid, ie
sea surface temperature (tos)
sea surface salinity (sos)
sea surface height (zos)
total heat fluxes at the surface (hfds)
barotropic streamfunction (msftbarot).

Providing these variables on a regular grid from IPSLCM6 simulations, only makes sense if other models provide those as well...
so any feedback on which variables will be provided on a 1deg regular grid in your respective groups / models, will be greatly appreciated !

email from Aurore to Julie (Feb 16, 2018):

I am a bit surprised by your message. At CNRM, according to DR.00.20, we obtain grid interpolation for tos and sos. For tos interpolation is automatically done for daily and monthly outputs, for sos only monthly outputs are interpolated. We have disactivated interpolation for all other fields.

e-mail from Gokhan to Julie (Feb 18, 2018):

The information about regridding from a model's native grid to a standard lon-lat grid is provided in Appendix A3 of Griffies et al. (2016). Section A3.3 states that i) 2D priority 1 tracer fields and their corresponding surface fluxes and ii) 3D priority 1 tracer fields are recommended to be regridded. At NCAR, we will indeed regrid the majority of our 2D fields and all 3D tracer fields, including BGC fields. Whether or not to regrid and for which fields to do so are actually upto each modeling group as we have not reached a consensus as some groups do not wish to provide regridded data for various reasons. The carrot is that if they provide regridded data, their data will likely be used / included in more publications and analysis.

e-mail from Stephen to Julie (Feb 19, 2018):

GFDL is also aiming to respond to the OMIP paper request, so that all of our submitted scalar fields will be mapped to the sphere.

email from Helmuth to Stephanie (March 21, 2018):

Mir scheint es gibt dabei kein einheitliches Vorgehen, ich persönlich finde es ausreichend ein cdo basiertes „Kochrezept“ zur Verfügung zustellen.

e-mail from Stephanie to Helmuth (March 21, 2018):

soweit wir am DKRZ mit der Datenaufbereitung befasst sind, werden wir uns zunächst (d.h. vor oder während des Modelllaufs)
nicht um die Interpolation kümmern. Wenn ein Experiment durchgelaufen ist, und die Daten auf den native-Gittern am
ESGF zur Publikation eingereicht sind, werden wir Skripte mit der Interpolation laufen lassen.
Diese Skripte können zumindest im WDCC als 'ADDITIONAL INFO' abgelegt werden.
Was mir Sorgen macht, ist, ob alle Leute, die mit der Interpolation befasst sind, hinreichend Erfahrung mit
Interpolationsalgorithmen haben....
Es sollten eigentlich die MIPs, die die Interpolation von einer Variablen verlangen, auch angeben mit welcher Methode.
Und eigentlich sollte das alles jetzt schon im DR stehen, damit man seine Workflows entsprechend aufsetzen kann.

E-mail Fabian Wachsmann von Sep 3, 2018:

" ... in folgendem Dokument https://docs.google.com/document/d/1kZw3KXvhRAJdBrXHhXo4f6PDl_NzrFre1UfWGHISPz4/edit steht:

Für Ozean:
- 'All two-dimensional and three-dimensional diagnostic fields produced by unstructured grid models should be archived on a spherical coordinate grid '
- '.. tracer-related diagnostics be archived on both the native and spherical grids'

Regridding Methoden:
- 'at least one of the methods must be "conservative"'

Auflösung des target-grid ist modellabhängig, aber es werden target-grids vorgegeben. "