EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Balanced subsampling of future regional climate ensembles of opportunity

Jesús Fernández and María Dolores Frías
Jesús Fernández and María Dolores Frías
  • Meteorology Group, Dept. Applied Mathematics and Comp. Sci., Universidad de Cantabria, Santander, Spain (

International model intercomparison initiatives, such as CORDEX or CMIP5, along with several relatively recent projects at international and national level, provide a wealth of model simulations of future regional climate. In a recent work, Fernandez et al (2019) collected 196 different future climate change projections over Spain, considering data from ENSEMBLES, ESCENA, EURO- and Med-CORDEX, along with their driving global climate projections from CMIP3 and CMIP5. This ensemble mixed different multi-model initiatives in an ensemble of opportunity, in the sense that it does not respond to any scientific design beyond the exploration of multi-model uncertainty. This ensemble of opportunity is not only the result of the mixture of different initiatives, but also responds to the lack of a balanced experimental design within most of the initiatives. Many of the initiatives -especially those unfunded, such as CORDEX- are carried out on a voluntary basis, with no strong constraint in the global climate models (GCMs) used as boundary conditions or in the number of contributing members per regional climate model (RCM).

Fernandez et al (2019) found in this ensemble a strong influence of the driving GCM on the regional climate change signal, along with favored GCMs, selected by many regional climate modelling groups to the detriment of GCMs publishing their output later or not at all. In this work, we quantitatively assess the impact of unbalanced GCM-RCM ensembles. For this purpose, we subsampled the ensemble of opportunity to obtain balanced sets of members according to different “what-if” situations: What if all RCMs had contributed a single member to the ensemble? What if each GCM had been dynamically downscaled only once? What if a given GCM/RCM had not contributed to the ensemble? For each hypothesis, there are a number of alternative sub-ensembles, which are used to evaluate uncertainty.


This work is partially funded by the Spanish government through MINECO/FEDER co-funded projects INSIGNIA (CGL2016-79210-R) and MULTI-SDM (CGL2015-66583-R). 


Fernández, J., et al. (2019) Consistency of climate change projections from multiple global and regional model intercomparison projects. Clim Dyn 52:1139.

How to cite: Fernández, J. and Frías, M. D.: Balanced subsampling of future regional climate ensembles of opportunity, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-20052,, 2020

Comments on the presentation

AC: Author Comment | CC: Community Comment | Report abuse

Presentation version 2 – uploaded on 04 May 2020
Added new slide (pag. 24) with results for additional model families (ALADIN, RACMO, REMO).
  • AC1: Reply to CC1, Jesús Fernández, 05 May 2020

    Hi Samuel,

    Thanks for commenting. Regarding the local changes, yes, I expect to have some enhanced signal at least over the Pyrenees where delta change differences of up to 4 degrees were found for some models due to a too persistent snow cover.

    It is also likely that the differences were larger in a far future period. Note, however, that in this study we completely disregarded the forcing scenario. Members for a given family arise from different resolutions and RCP scenario. I assume that, in a far future period, all PDFs would tend to be bimodal (due to the scenario separation) and it could be the case that model-family signals increase. We'll see. This was a fast check, using data we already had from a previous work. This limited us to Spain and also to the near future period. Our plan is to load the current ensemble, extending it to other European regions and to the end of the century.

    I'll keep an eye on ALADIN over central Europe as you mention. In the meantime, I added a new slide with other model families, including ALADIN for your reference. It is by far the model showing less spread among members (ALADIN/ALARO, 011/044, RCP26/45/85) in this region. This is probably due to nesting in a single GCM, which is the main driver for the mean delta.

    I'll keep you posted.

    Best regards,


    • CC1: Reply to AC1, Samuel Somot, 07 May 2020

      thanks for the reply and the new slide. I agree with your interpretation for ALADIN weak spread. ALADIN63 nudged into HadGEM is now available on ESGF, soon MPI and NorESM is running

Presentation version 1 – uploaded on 30 Apr 2020
  • CC1: Comment on EGU2020-20052, Samuel Somot, 04 May 2020


    I really find your study very interesting. Thanks for having done it. Even if this is a good news for the robustness of the results produced with Euro-CORDEX, I'm still surprised by the weak differences obtained when you remove models or when you balance the ensemble. You answered me during the chat that locally the changes can be larger, I imagine at least over islands or the Pyrenees. Ok fine.

    I'm also wondering if you would obtain stronger differences for the far-futur period ? 

    There is at least one situation in which you would obtain a big difference: removing ALADIN run(s) for the rsds scenario (Summer, central Europe) following results from Boé et al. 2020 (doi:10.1007/s00382-020-05153-1) and Gutiérrez et al. 2020 (

    Keep me updated when published

    best regards