Multi-model ensembles are the standard download now?

Datasets and Usage C3S - Datasets and Usage
1

Hello everyone,

Let me preface this by saying I am not a climate scientist, but in the past, downloads from the CDS have greatly aided me.

I just noticed that when looking at climate projections downloads (e.g. Gridded dataset underpinning the Copernicus Interactive Climate Atlas )
it seems no longer possible to select individual General Circulation Models to download. Instead, I assume we are now downloading a multi-model ensemble?

Apart from the fact that I personally am fine with using an ensemble rather than selecting individual GCMs, I was curious about this change and the rationale behind it, and want to make sure I am not overlooking something.

Thank you!

2

Dear Merijn,

Thanks for your query. I confirm, first that for the C3S Atlas, we always provided all available models for the various data sources. At the same time, we have the inidividual climate projection entries (like CMIP5, CMIP6 or CORDEX for instance), where you can select any model for download. Maybe that is what you refer to.

I also mention that generally the preference is not to stick to a single model, when you look at the climate, but ensemble of models to quanitfy uncertainties in the climate projections. So ideally you should use more (all) models for evaluating the climate change signal for instance.

I hope it helps a bit, thank you!

Best regards

            Andras Horanyi
3

Thank you Andras, I do have a follow-up question;

I note that when downloading .nc files from the C3S atlas or IPCC 6 interactive atlas, for the CORDEX-11 geographic region, there is a dimension to the files ‘Member’ which I suppose is the single model, and the .NC file then contains all these Members, making it the full multi-model ensemble.

Because I lack the knowledge needed to pick and choose the ‘best’ or ‘most relevant’ single members, would it be an option to show the uncertainty inherent to the projections by extracting the mean, along with say a 10th and 90th percentile?

If this is a valid approach, how do I weigh the individual members, given that some GCMs have had more individual RCMs/variants than others? Say there’s 19 members across 5 GCMs, do I take the mean across the 19 members, or should I first take mean across all the RCMs/variants per GCM?

All the best

Merijn

4

Dear Merijn,

See the available models for the CORDEX-EUR-11 domain and resolution: Gridded data underpinning the Copernicus Interactive Climate Atlas: Description of the datasets and variables - Copernicus Knowledge Base - ECMWF Confluence Wiki

As you can see we have variants (versions) for the GCMs , but not for the RCMs.

We consider all the RCMs as equally possible realisations and therefore ideally you can use all multi-modell ensemble members and compute mean, standard deviation and percentiles. At the same time, it might be better to use the median instead of the mean as it is for the C3S Atlas viewr for instance: Copernicus Interactive Climate Atlas

So you don’t need to weight individual members, but to use them as equally probable outcomes. This takes into account that the uncertainties are more coming from the GCMs (though boundary conditions for the RCMs) than for the individual RCMs and the RCM uncertainty is sampled by the use of more RCM models.

I hope it helps, best regards

                             Andras
5

Thank you Andras, that’s extremely helpful! I only need to figure out the meaning of the extreme SPEI-6 values in the .NC files (- and + 8.209961) and I can move ahead!