EGU2020-13441, updated on 12 Jun 2020
https://doi.org/10.5194/egusphere-egu2020-13441
EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Effects of GCM selection for regional climate modelling illustrated by the interactive tool GCMeval

Oskar A. Landgren1, Kajsa Parding1, Andreas Dobler1, Carol F. McSweeney2, Rasmus Benestad1, Helene B. Erlandsen1, Abdelkader Mezghani1, Hilppa Gregow3, Olle Räty3, Elisabeth Viktor4, Juliane El Zohbi4, Ole Bøssing Christensen5, and Harilaos Loukos6
Oskar A. Landgren et al.
  • 1Norwegian Meteorological Institute, Model and climate analysis, OSLO, Norway (oskar.landgren@met.no)
  • 2Met Office Hadley Centre, Exeter, United Kingdom
  • 3Finnish Meteorological Institute, Helsinki, Finland
  • 4Climate Service Center Germany (GERICS), Helmholtz-Zentrum Geesthacht, Germany
  • 5Danish Meteorological Institute, Copenhagen, Denmark
  • 6The Climate Data Factory, Paris, France

With the increasing number of global climate models available, regional modellers have to make choices to select a manageable subset for downscaling. This limits the representation of both present day climate and future climate change compared to the full GCM ensemble.

We present the interactive web-based tool called “GCMeval”, available at https://gcmeval.met.no. This tool lets you assign weights to different regions, seasons, climate variables, and skill scores and presents a ranking with model performance for a historical period. We demonstrate how the tool can be used to, for example, remove models with the largest historical biases for the selected criteria, or to optimise the spread. The weighting can be used to illustrate the sensitivity of the results to model choice.

Based on the choice of regions and weights, the tool produces scatter plots of projected future temperature and precipitation and shows how the selected sub-ensemble compares to the full ensemble. The tool can also be used to evaluate ensemble selections "post-hoc", as demonstrated with examples from CORDEX.

How to cite: Landgren, O. A., Parding, K., Dobler, A., McSweeney, C. F., Benestad, R., Erlandsen, H. B., Mezghani, A., Gregow, H., Räty, O., Viktor, E., El Zohbi, J., Bøssing Christensen, O., and Loukos, H.: Effects of GCM selection for regional climate modelling illustrated by the interactive tool GCMeval, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-13441, https://doi.org/10.5194/egusphere-egu2020-13441, 2020

Comments on the presentation

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

Presentation version 1 – uploaded on 29 Apr 2020
  • AC1: Response to questions during live chat, Oskar Landgren, 04 May 2020

    We have put the questions from the live chat and our answers here:
    https://docs.google.com/document/d/1eSO_-4iRbMJuF-E1sYHr5_wMiAgGxZozmFBQc97Y8SI/

  • CC1: Comment on EGU2020-13441, Samuel Somot, 04 May 2020

    Hi,

    thank for the slides and the video (quite clear). On one of your slides, you mention the 2 members for CNRM-CM5. I would like to mention 2 issues with the CNRM-CM5 members. Both issues are  completely independant.

    1. r8i1p1 does not exist officially. So every simulation labelled with r8i1p1 should have been labelled r1i1p1. This error is due to the fact that, initially the run was r8 but it was re-named r1 before the ESG publication. By the way, do you know who is using this r8 naming ? This error is very common in Med-CORDEX or CORDEX Africa but not in Euro-CORDEX

    2. Later, an error has been detected in the LBC files provided for the r1 member of CNRM-CM5 (this error is documented here (https://www.medcordex.eu/warnings/Communication-Issue-Files_CNRM-CM5_historical_6hLev_en.pdf). It means that the r1i1p1 member for the historical period is not really a r1 but an unknown member. Since the communication of this error and the providing of the correct LBCs, various RCM groups in Euro-CORDEX have re-run the CNRM-CM5 driven simulations. The correct ones are labelled v2 (RACMO, ALADIN63, ...) and are the most recent, the one using the unknown member are often labelled v1. Hope it helps.

    • AC2: Reply to CC1, Andreas Dobler, 05 May 2020

      Hi Samuel,

      thanks for the feedback and pointing out the issues with CNRM-CM5.

      Considering point 1: The list of GCMs is based on the information provided in the EURO-CORDEX Simulation list available at https://www.euro-cordex.net/imperia/md/content/csc/cordex/20180130-eurocordex-simulations.pdf
      CNRM-CM5 r8i1p1 is still listed there (page 5). Good to know this is not correct, we try to keep this in mind for future presentations etc.

      Point 2: Yes, we are aware of that. However, we are only using tas and pr data from the GCMs (no RCM or LBC data). Thus, the information provided in the tool should not be affected.

      Kind regards

      Andreas

  • CC2: Comment on EGU2020-13441, Samuel Somot, 04 May 2020

    Hi,

    after re-reading the questions/answers in the google, I think that my question still holds. From your point of view and taking into account the fact that many tests can be done with GCMeval and that the tas-pr criteria can be discussed, are you able to give a list of 2 or 3 CMIP5 GCMs that would have improved the current Euro-CORDEX ensemble, that is to say: good models and allowing to better cover the future spread. Thanks  

    • AC3: Reply to CC2, Andreas Dobler, 05 May 2020

      Hi Samuel,
      I see the point. We did a similar exercise for the “Nordic Convection Permitting Climate Projections” and concluded that the GFDL-CM3 model is a good candidate as it shows good rankings and high dT & dP (in agreement with a very strong sea-ice retreat in the Arctic). Using the tool, we calculated now, for the three European subregions and globally, the spread covered by the GCMs used for EURO-CORDEX and when adding GFDL-CM3 (see table below). In summary, it would better span the warming side together with strong drying in the South and wetting in the North.

      % of full spread
      EUR-11 GCMs EUR-11 GCMs + GFDL-CM3 + GFDL-CM3
      Region dT dP dT dP
      NEU
      27%
      38% 36% 68%
      CEU 61% 61% 73% 61%
      MED 47% 72% 74% 100%
      Global 64% 68% 64% 68%

      We also tried other models, but there it is less clear. Some of them are

      • CCSM4: very good performance, but changes are relatively similar to MPI-ESM-LR (a bit more to the edges)
      • HadGEM2_AO: good performance, precip. changes at the edge
      • GFDL-ESM2(M&G): OK performance, temp. changes at the low edge

      We found no model that improves the cold side of the spread of projections in all three European subdomains.

      Of course, there can be other issues with these particular models that are not covered by the metrics in GCMeval. Additional evaluation of the spread, for example regarding extremes and inter-annual variability would also be useful before making a final selection. And, one should probably ask why these models are at the edges of the scatter. How are they different from the others? Why is the change in the atmosphere-ocean version (AO) of HadGEM (so) different than in the full ESM version? etc.

      Does this answer help?

      Best regards,
      Andreas

      • CC3: Reply to AC3, Samuel Somot, 14 May 2020

        Hi Andreas

        thanks for he very constructive response. This kind of information could be of interest for the Euro-CORDEX community if new runs have to be done with CMIP5 GCMs or to get feedbacks to understand why they have not been used (technical issues ?).

        By the way, GFDL-ESM2G has been used by REM2015 (to my knowledge) but only for rcp26 what is strange. I hope that the next CMIP6-based coordinated matrice will be better prepared

        samuel