Constraints on {instanceId} format?

[geira]

Just mentioning before I forget it...

GeoWEB reported an issue where they for any given data source need to present different model runs in a dropdown box with timestamps. Currently EDR only allows three text fields that are relevant: Id, Title and Description, which are not standardized between different data publishers. E.g.:

interpol.met.no

• ISO timestamp ("2024-09-25T15:00Z") as Id
• same in title and description with increasing level of verbosity

FMI

• uses compressed timestamp ("20240925T000000") as Id
• Title is too long for dropdown box ("Origintime: 2024-09-25T00:00:00Z Starttime: 2024-09-25T03:00:00Z Endtime: 2024-10-01T00:00:00Z Timestep: 180")
• no Description
• has a custom field with extra temporal metadata (probably non-standard)

MET Office

(Probably not relevant for GeoWEB, but shows how usage differs)

• has even more compressed timestamps ("2022070900") as Id
• rather generic Title and Description, but then again only one instance given in this prototype

[havardf]

Can we solve this by rather using the start time in temporal extent interval?

So for an instance of a collection like this:

"title": "Foo",
"temporal": {
          "interval": [
            [
              "2022-07-12T12:00Z",
              "2022-07-14T18:00Z"
            ]
          ],
          "values": [
            "R54/2022-07-12T12:00Z/PT1H"
          ],
          "trs": "TIMECRS[\"DateTime\",TDATUM[\"Gregorian Calendar\"],CS[temporal,1],AXIS[\"Time (T)\",future],TIMEUNIT[\"day\",86400]]"
        },

you would present in a menu of instances as:

Foo (start time: 2024-10-09T12:00Z)
Bar (start time: 2024-10-09T18:00Z)

If so maybe we do not need any specific rules on instanceid.

[geira]

No, because:

1. the instanceId is the reference time (run time) of the model, not the forecast time
2. you can't use complex data structures in a dropdown menu

Also, apart from specific models there is no guarantee that the instanceId actually reflects model run time (could be ensemble members, e.g.).

[havardf]

I know, I formulated that badly. My point was, is the model run time actually important anymore? The models will run more and more frequently, so the start time or end time of the model might anyway be more informative. At least good enough for us not to have to solve this problem....