Marker scenario versions incorrect?

[maritsandstad]

With the current out-of-the-box-main, I do not get version matches for the marker scenarios:

• REMIND what is in the database are versions 4,3,6 and the file is looking for version 5
• WITCH what is in the database are versions [37 34 36 38 35 39 32], the file is looking for 31
• Model MESSAGE has versions [ 3 1 7 14 20 15 12], the file is looking for version 19
• IMAGE works (version 25 which is prompted for exists)
• GCAM works (version 3 which is prompted for exists)
• COFFEE works (version 13 which is prompted for exists, but I assume this is still not necessarily the final version for COFFEE?)
• AIM does have the asked for version (23), but the known version file is looking for an older version (22), this might be fine, though.

@mzecc do you have explanation for this?

[mzecc]

REMIND, WITCH, and MESSAGE have been updated in the meanwhile so we are lagging behind 1 version.

WITCH what is in the database are versions [37 34 36 38 35 39 32], the file is looking for 31

For WITCH what I see in the database is this:

I fail to see where the other versions are coming from.

AIM does have the asked for version (23), but the known version file is looking for an older version (22), this might be fine, though.

There has been an AIM submission on 2026-02-05 (version 23) that has been used but the correspondent versions.json sits on PR#164, and it has not been merged yet.

[maritsandstad]

The versions I am printing are for all scenarios available for that model, so the 32 one is for the correct marker scenario, I assume.

So, if I wanted the correct versions it would be 6 for REMIND, 32 for WITCH, 20 for MESSAGE and 23 for AIM?

[mzecc]

These versions are indeed the latest submissions of the marker scenarios. In theory they should be very very similar to the previous one that, I believe, we are going to keep "fixed" as are the ones we have run in the final ScenarioMIP round.

[maritsandstad]

Are the previous one to be run as the in the final ScenarioMIP round?

Can one still get the pipeline here to run with those?

[jkikstra]

@maritsandstad;

Reasons for why there are different versions, but not different results for you.

• REMIND: added detail on solid biomass emissions reporting
• MESSAGE: implemented some small fixes (tiny negative values where they shouldn't be) that were already implemented/checked for
• WITCH: updated reporting of electricity (not emissions)

Can one still get the pipeline here to run with those?
This should be possible, needs a change to not only return default versions in the scenario list but also non-default versions.
Perhaps it's as easy as changing a default (see ixmp.Platform.scenario_list) - I've asked @mzecc to have a look, but if you find it before, that'd be great too, of course.

For your current version, as we speak, it shouldn't matter (as you have confirmed). And teams shouldn't be uploading scenarios anymore that change this data.
But still, it would be a good option to have.

[maritsandstad]

@jkikstra : There are in fact different results for me...

[maritsandstad]

I'll have a look at your tips for how to return all the version for the scenario list. IMO this should be on the main version of the repo so that the MARKER-versions that are fixed can be ran to get the correct version even if the latest default has changed, also for backward compatibility...

[maritsandstad]

Ok, think I have a solution, will test properly and make a PR after, though:

If we change:

props = conn_ssp.properties().reset_index()

to:

props = conn_ssp.properties(default_only=False).reset_index()

I think that does the trick, and we should be able to run with whatever fixed version we want...

[maritsandstad]

(In the 5090_download-scenarios.py notebook). Sound ok @mzecc and @jkikstra ?

[maritsandstad]

Ok PR that I think solves this here:

#167

I have tested that this works and pulls the original Marker versions, I will do a little more testing to see if that means I get the data from the official ScenarioMIP zenodo upload (or at least something that looks the same in what I look at) later today...

[jkikstra]

@jkikstra : There are in fact different results for me...

What scenarios are different, between the versions you mentioned above?

props = conn_ssp.properties(default_only=False).reset_index()

Great that's exactly what I had in mind and what I hoped would work.

Thanks for making the PR, i agree this should be on the main.
I will try to run and confirm and then respond on the PR

[maritsandstad]

@jkikstra : There are in fact different results for me...

What scenarios are different, between the versions you mentioned above?

AIM, MESSAGE and WITCH all differ for CO2 Energy and Industrial Processes. (see plots in my email from the 9th of February).

I'll rerun with the fix from the PR to get the fixed versions and not the latest and see if I can make them all agree.

[maritsandstad]

Ok, I think I've ran with the correct Marker versions now, and I still see discrepancies in CO2|Energy and Industrial Processes for AIM, MESSAGE and WITCH (and these seem to be the same with the default versions):

It might definitely be methodological (i.e. mea culpa), but I am not quite sure... The blue "off"-line is the data downloaded from the official ScenarioMIP zenodo account, the stippled orange line "em" is what I get from the pipeline with the current version (I am reading in from a csv that the extensions notebook makes called continous_emissions_timeseries_1750_2500.csv) and then the green dotted line is a version of infilled_emissions that @jkikstra shared with Chris a little while ago (for AFOLU that one is clearly outdated for AIM). All other components show good matches and errors in actual numbers that are likely just round-off.

I will fix the remaining functionality thing for the #167 PR, and the mystery of the data discrepancies remain...

[znichollscr]

@mzecc or @jkikstra can you please post the versions.json file which sits underneath https://www.doi.org/10.5281/zenodo.18497404

Can someone confirm if my understanding, written below, of what we're doing is correct please?

1. Only the harmonised and infilled data is available via Zenodo/the scenario explorer, so we're only ever comparing harmonised and infilled emissions?
   • if yes, that makes our life a bit harder as the error could come from differences in the raw emissions, or the harmonisation, or the infillling, but ok
2. The only variable where we're seeing any difference is CO2|Energy and Industrial Processes? Every other variable has things lining up?
   • if yes, that's weird, but ok

[znichollscr]

I can confirm that, using the setup in https://github.com/iiasa/emissions_harmonization_historical/tree/compare-against-zenodo, which uses DOWNLOAD_SCENARIOS_ID = "202602051053" and therefore the versions below, I see the same issue as Marit (assuming that the data file I grabbed from an email from Jarmo is what ended up on zenodo, I'm still waiting for access)

{
  "AIM 3.0": {
    "SSP2 - Low Overshoot_a": 23
  },
  "GCAM 8s": {
    "SSP3 - High Emissions": 3
  },
  "IMAGE 3.4": {
    "SSP2 - Medium Emissions": 25
  },
  "MESSAGEix-GLOBIOM-GAINS 2.1-M-R12": {
    "SSP2 - Low Emissions": 19
  },
  "REMIND-MAgPIE 3.5-4.11": {
    "SSP1 - Very Low Emissions": 5
  },
  "WITCH 6.0": {
    "SSP5 - Medium-Low Emissions_a": 31
  }
}