Possibly re-frame this as deciding on a solution for templating. E.g. do we ignore existing datasets, or have a special "template" recipe for CMEW that has a placeholder for datasets to be added.
CMEW modifies the first two datasets in a template recipe. For some recipes we will want to see the performance of model test runs in the context of many CMIP runs, and not replace any of the datasets in the source recipe.
Related, a user may wish to denote a CMIP run as the reference, rather than a model development run. We might or might not assume that such a CMIP reference was already one of the recipes in the template. This potentially needs one or two separate issues.
The REF may have capabilities that help in this space.
A concept that ESMValTool uses that may help is to specify the reference via a special variable that gets passed to diagnostic codes e.g. reference=HadGEM3-GC31-LL, rather than having separate CMEW variables for test and reference runs.
Possibly re-frame this as deciding on a solution for templating. E.g. do we ignore existing datasets, or have a special "template" recipe for CMEW that has a placeholder for datasets to be added.
CMEW modifies the first two datasets in a template recipe. For some recipes we will want to see the performance of model test runs in the context of many CMIP runs, and not replace any of the datasets in the source recipe.
Related, a user may wish to denote a CMIP run as the reference, rather than a model development run. We might or might not assume that such a CMIP reference was already one of the recipes in the template. This potentially needs one or two separate issues.
The REF may have capabilities that help in this space.
A concept that ESMValTool uses that may help is to specify the reference via a special variable that gets passed to diagnostic codes e.g.
reference=HadGEM3-GC31-LL, rather than having separate CMEW variables for test and reference runs.