Journal cover Journal topic
Earth System Dynamics An interactive open-access journal of the European Geosciences Union
Journal topic

Journal metrics

Journal metrics

  • IF value: 4.351 IF 4.351
  • IF 5-year value: 5.124 IF 5-year
    5.124
  • CiteScore value: 4.44 CiteScore
    4.44
  • SNIP value: 1.250 SNIP 1.250
  • IPP value: 4.10 IPP 4.10
  • SJR value: 2.203 SJR 2.203
  • Scimago H <br class='hide-on-tablet hide-on-mobile'>index value: 29 Scimago H
    index 29
  • h5-index value: 31 h5-index 31
Discussion papers
https://doi.org/10.5194/esd-2019-32
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.
https://doi.org/10.5194/esd-2019-32
© Author(s) 2019. This work is distributed under
the Creative Commons Attribution 4.0 License.

Submitted as: research article 19 Jun 2019

Submitted as: research article | 19 Jun 2019

Review status
This discussion paper is a preprint. It is a manuscript under review for the journal Earth System Dynamics (ESD).

Improving weather and climate predictions by training of supermodels

Francine Schevenhoven1,2, Frank Selten3, Alberto Carrassi4,1,2, and Noel Keenlyside1,2 Francine Schevenhoven et al.
  • 1Geophysical Institute, University of Bergen, Bergen, Norway
  • 2Bjerknes Centre for Climate Research, Bergen, Norway
  • 3Royal Netherlands Meteorological Institute, De Bilt, The Netherlands
  • 4Nansen Environmental and Remote Sensing Center, Bergen, Norway

Abstract. Recent studies demonstrate that weather and climate predictions potentially improve by dynamically combining different models into a so called "supermodel". Here we focus on the weighted supermodel – the supermodel's time derivative is a weighted superposition of the time-derivatives of the imperfect models, referred to as weighted supermodeling. A crucial step is to train the weights of the supermodel on the basis of historical observations. Here we apply two different training methods to a supermodel of up to four different versions of the global atmosphere-ocean-land model SPEEDO. The standard version is regarded as truth. The first training method is based on an idea called Cross Pollination in Time (CPT), where models exchange states during the training. The second method is a synchronization based learning rule, originally developed for parameter estimation. We demonstrate that both training methods yield climate simulations and weather predictions of superior quality as compared to the individual model versions. Supermodel predictions also outperform predictions based on the commonly used Multi-Model Ensemble (MME) mean. Furthermore we find evidence that negative weights can improve predictions in cases where model errors do not cancel (for instance all models are warm with respect to the truth). In principle the proposed training schemes are applicable to state-of-the-art models and historical observations. A prime advantage of the proposed training schemes is that in the present context relatively short training periods suffice to find good solutions. Additional work needs to be done to assess the limitations due to incomplete and noisy data, to combine models that are structurally different (different resolution and state representation for instance) and to evaluate cases for which the truth falls outside of the model class.

Francine Schevenhoven et al.
Interactive discussion
Status: open (until 24 Sep 2019)
Status: open (until 24 Sep 2019)
AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment
[Subscribe to comment alert] Printer-friendly Version - Printer-friendly version Supplement - Supplement
Francine Schevenhoven et al.
Francine Schevenhoven et al.
Viewed  
Total article views: 443 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
359 76 8 443 7 10
  • HTML: 359
  • PDF: 76
  • XML: 8
  • Total: 443
  • BibTeX: 7
  • EndNote: 10
Views and downloads (calculated since 19 Jun 2019)
Cumulative views and downloads (calculated since 19 Jun 2019)
Viewed (geographical distribution)  
Total article views: 279 (including HTML, PDF, and XML) Thereof 278 with geography defined and 1 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Cited  
Saved  
No saved metrics found.
Discussed  
No discussed metrics found.
Latest update: 15 Sep 2019
Publications Copernicus
Download
Short summary
Weather and climate predictions potentially improve by dynamically combining different models into a 'supermodel'. A crucial step is to train the supermodel on the basis of observations. Here we apply two different training methods to the global atmosphere-ocean-land model SPEEDO. We demonstrate that both training methods yield climate and weather predictions of superior quality as compared to the individual models. Supermodel predictions can also outperform the commonly used Multi-Model mean.
Weather and climate predictions potentially improve by dynamically combining different models...
Citation