Suárez‐Gutiérrez et al. (2025) Temperature variability projections remain uncertain after constraining them to best performing Large Ensembles of individual Climate Models

Identification

• Journal: Nature Communications
• Year: 2025
• Date: 2025-12-13
• Authors: Laura Suárez‐Gutiérrez, Nicola Maher
• DOI: 10.1038/s41467-025-67005-y

Research Groups

• Institute for Atmospheric and Climate Science, ETH Zürich, Zurich, Switzerland
• Laboratoire des Sciences du Climat et de l’Environnement, Institut Pierre-Simon Laplace, Paris, France
• Meteorology and Air Quality Group, Wageningen University & Research, Wageningen, The Netherlands
• Research School of Earth Sciences, The Australian National University, Canberra, ACT, Australia
• ARC Centre of Excellence for the Weather of the 21st Century, The Australian National University, Canberra, ACT, Australia

Short Summary

This study evaluates the historical performance of eleven climate models in simulating temperature variability and uses the best-performing models to constrain future projections, finding that significant uncertainties in the intensity and sign of temperature variability changes persist globally.

Objective

• To evaluate the ability of single model initial-condition large ensembles (SMILEs) to adequately represent historical regional temperature variability under current climate conditions.
• To assess if constraining future temperature variability projections to historically best-performing SMILEs reduces projection uncertainty.

Study Configuration

• Spatial Scale: Global, 24 land regions, 9 ocean regions (based on IPCC SREX definitions), grid-cell level.
• Temporal Scale: Historical period (1900-2024 for observations), future projections based on 5 global warming levels (relative to 1850-1899 baseline), monthly mean temperature anomalies, seasonal (December-January-February (DJF) and June-July-August (JJA)).

Methodology and Data

• Models used: 11 Single Model Initial-condition Large Ensembles (SMILEs): ACCESS-ESM1.5, CanESM2, CanESM5, CESM-LE, CESM2-LE, CSIRO-MK3.6, GFDL-ESM2M, GFDL-SPEAR-MED, MIROC6, MPI-GE5, MPI-GE6.
• Data sources:
  • Observational surface air temperature: GISTEMPv4 (1880-2024) over land.
  • Observational sea surface temperature: ERSSTv5 (1854-2024) over ocean.
  • Multi-Model Large Ensemble Archive version 2 (MMLEAv2) for SMILE data.
  • Rank-frequency analysis framework (expanded from Suarez-Gutierrez et al., 2021) with two formally defined evaluation criteria (perfect-model rank range test and threshold-based grid-cell performance).
  • Detrending: subtracting ensemble mean for model data, least squares quadratic trend for observations.
  • Warming levels: computed as the year when the global mean surface temperature from the SMILE ensemble mean crosses an individual warming level.

Main Results

• Some regions are systematically poorly represented by most models across both seasons (e.g., Arctic, Antarctic, Southern Oceans, Indian Peninsula, Northern Australia, Amazon basin, Northern/Eastern Africa, Southern Australia in local summer).
• Local winter seasons are generally simulated better by most models than local summer seasons.
• The top 3 best performing models for isolated temperature variability (detrended data) are CESM-LE (47 regions), CESM2-LE (43 regions), and GFDL-SPEAR-MED (40 regions).
• Models typically overestimate temperature variability in the current climate.
• Constraining projections to best-performing models generally decreases the multi-model spread (uncertainty) in the magnitude of future temperature variability change, particularly in poorly represented regions (e.g., Amazon, Southern Australia, Tropical Pacific Ocean).
• The constraint does not substantially improve model agreement on the sign of the projected change, though some patches of improvement exist (e.g., India, Australia in DJF; parts of South America, Africa, Northern Australia in JJA).
• In regions like South America, Africa, and Australia, where current variability is overestimated, the constrained ensemble projects a larger increase in temperature variability from 1 °C to 3 °C global warming compared to the unconstrained ensemble, suggesting unconstrained projections may underestimate future variability increases.
• Significant uncertainty in the magnitude of change remains over key regions even with constraint (e.g., Tropical Pacific Ocean, Southern Ocean, Northern South America, Northern Hemisphere extratropical land surface).

Contributions

• Provides the first performance-constrained projections of global temperature variability and its change using a multi-model ensemble of SMILEs.
• Introduces an expanded rank-frequency evaluation framework with two formally defined criteria, including a perfect-model rank range test, to rigorously assess model performance in simulating historical temperature variability while accounting for internal variability sampling limitations.
• Offers a comprehensive multi-model evaluation of 11 state-of-the-art SMILEs, identifying best-performing models and systematically misrepresented regions globally for historical summer and winter temperature variability.
• Demonstrates that while performance-based constraint can reduce uncertainty in the magnitude of future variability change, it does not eliminate it, nor does it drastically improve agreement on the sign of the change, highlighting the continued necessity of diverse multi-model ensembles.
• Reveals that unconstrained multi-model mean projections may underestimate future temperature variability increases in poorly modeled regions, with implications for extreme event planning.

Funding

• Australian Research Council Discovery Early Career Researcher Award DE230100315 (N.M.)
• European Union’s Horizon Europe Framework Programme under the Marie Skłodowska-Curie grant agreement No. 101064940 (L.S.G.)
• Deutsches Klimarechenzentrum (DKRZ) for computational resources.
• US CLIVAR Working Group on Large Ensembles for Multi-Model Large Ensemble Archive version 2 (MMLEAv2) data.
• World Climate Research Programme (WCRP) Working Group on Coupled Modelling for CMIP6 coordination.
• Climate modelling groups, Earth System Grid Federation (ESGF), and funding agencies supporting CMIP6 and ESGF.
• Open access funding provided by Swiss Federal Institute of Technology Zurich.

Citation

@article{SuárezGutiérrez2025Temperature,
  author = {Suárez‐Gutiérrez, Laura and Maher, Nicola},
  title = {Temperature variability projections remain uncertain after constraining them to best performing Large Ensembles of individual Climate Models},
  journal = {Nature Communications},
  year = {2025},
  doi = {10.1038/s41467-025-67005-y},
  url = {https://doi.org/10.1038/s41467-025-67005-y}
}