search

WCRP-CORDEX / cmip6-for-cordex

CMIP6 input data selection for CORDEX
https://wcrp-cordex.github.io/cmip6-for-cordex
0 stars 2 forks source link

Member "available & plausible" but not "available"? #24

Closed jesusff closed 2 months ago

jesusff commented 3 months ago

Regarding the MED table

S. Somot wrote:

I found a problem in the synthetic table for the model ACCESS-CM2. The r4 appears in the category: available and plausible but not in the category available. strange.

Another strange point (possibly related) is that in the MED synthetic table, we have not anymore the whole list of all members for all GCMs. What corresponds to the list "All members with 2 or more scenarios and/or some metric available" in the EUR version of the table. Could you add it again for MED too ?

jesusff commented 3 months ago

@sam-somot, this is due to the "single member" policy, which selects a single member independently for each category table, to avoid many members for a given model showing up in the summary tables. Currently, the most evaluated member is selected (that the smallest amount of missing metrics), unless this member would be filtered out in the corresponding table:

https://github.com/WCRP-CORDEX/cmip6-for-cordex/blob/15902ee4a8e56673f7db720648794f49a933cc45/CMIP6_studies_table.py#L186-L194

In the particular case you mention, ACCESS-CM2 r4 appears as plausible, since it was evaluated only for a few metrics, showing plausible values in all of them. In the available list, ACCESS-CM2 r1 is shown, because this is the member most thoroughly evaluated, even though 2 of the metrics show implausible values. In fact, members r1, r4 and r5 are available, but only r1 is shown due to the single member filter. Finally, for the available and plausible table, only r4 survived both filters. [Note: this was before 15902ee, where r4 was set to fail the Black Sea SST metric, due to missing this sea entirely]

As you mention, this is clearer if we bring back for MED the full table with all members available and or evaluated, that was removed in #9. Also, we could remove the single member filter for all tables, or change the policy to select this single member. We used to take the member with the lowest number of implausible values (synthesis closer to 0), but this favoured members evaluated for a single (plausible) score.

jesusff commented 3 months ago

I disabled the single member option for MED temporarily in https://github.com/WCRP-CORDEX/cmip6-for-cordex/commit/34f69500c179ce76b079f554c24db0907aa0a68d to show why we needed it in the first place.

jesusff commented 2 months ago

Finally reverted to single member after clarifying policy with S. Somot. The extended table with all members can still be accessed here.