Verjans et al. (2026) Large potential of performance-based model weighting to improve decadal climate forecast skill

Identification

• Journal: npj Climate and Atmospheric Science
• Year: 2026
• Date: 2026-04-01
• Authors: Vincent Verjans, Markus G. Donat, Carlos Delgado-Torres, Timothy DelSole
• DOI: 10.1038/s41612-026-01397-6

Research Groups

• Earth Sciences Department, Barcelona Supercomputing Center, Barcelona, Spain
• Institució Catalana de Recerca i Estudis Avançats (ICREA), Barcelona, Spain
• Department of Atmospheric, Oceanic, and Earth Sciences, George Mason University, Fairfax, VA, USA

Short Summary

This study implements a performance-based model weighting scheme for decadal climate predictions, focusing on sea-surface temperature, demonstrating its potential to improve forecast skill, particularly when predicting pseudo-observations, but revealing challenges in validating these gains against real-world observations.

Objective

• To implement and evaluate a performance-based model weighting scheme, utilizing a novel deviance statistic, to improve the skill of decadal sea-surface temperature climate predictions by favoring models consistent with observations in climate forced response and stationary dynamics.

Study Configuration

• Spatial Scale: Global ocean (sea-surface temperature).
• Temporal Scale: Decadal predictions (10-year forecasts) and hindcasts evaluated over a 95-year period.

Methodology and Data

• Models used: Large climate model ensemble simulations, specifically CMIP6 models.
• Data sources:
  • Climate model ensemble simulations (CMIP6)
  • Pseudo-observations (model realizations)
  • Real-world sea-surface temperature observations: Extended Reconstructed Sea Surface Temperature, version 6 (ERSSTv6) and HadISST data.

Main Results

• Performance-weighted predictions of pseudo-observations showed a large potential for decadal forecast skill improvement compared to unweighted predictions.
• Skill gains were also observed in decadal hindcasts of 95-year real-world sea-surface temperature observations, though at considerably lower levels.
• The discrepancy between potential skill improvement in pseudo-observations and lower gains in real-world hindcasts is attributed to limited intrinsic predictability, similarity between unweighted and weighted ensembles, and inherent skill sampling uncertainties.

Contributions

• Introduction and demonstration of a novel deviance statistic for performance-based model weighting in decadal climate predictions.
• Quantification of the large potential skill improvement of performance-based weighting when predicting model realizations (pseudo-observations).
• Identification of previously unrecognized challenges in validating performance-based model weighting for climate forecasting against real-world observations, highlighting factors like intrinsic predictability limits and sampling uncertainties.

Funding

• European Union Horizon project EXPECT (grant 101137656)
• US National Oceanic and Atmospheric Administration (grant NA23OAR4310606-T1-01)

Citation

@article{Verjans2026Large,
  author = {Verjans, Vincent and Donat, Markus G. and Delgado-Torres, Carlos and DelSole, Timothy},
  title = {Large potential of performance-based model weighting to improve decadal climate forecast skill},
  journal = {npj Climate and Atmospheric Science},
  year = {2026},
  doi = {10.1038/s41612-026-01397-6},
  url = {https://doi.org/10.1038/s41612-026-01397-6}
}