Gaudin et al. (2026) New classes of climate model emulators to improve paleoclimate reconstructions

Identification

• Journal: Geoscientific model development
• Year: 2026
• Date: 2026-08-04
• Authors: Auguste Gaudin, Myriam Khodri
• DOI: 10.5194/gmd-19-7135-2026

Research Groups

• LOCEAN-IPSL, Sorbonne-UniversitÃ, CNRS/IRD/UPMC/MNHN, Paris, France

Short Summary

This study proposes a new class of climate model emulators that integrate autoencoder-based dimensionality reduction with nonlinear prediction architectures to improve paleoclimate reconstructions. The proposed emulators are evaluated using a comprehensive experimental protocol applied to the IPSL-CM6A-LR model and 52 CMIP6 models, showing improved representation of El Niño Southern Oscillation and Atlantic Multidecadal Variability.

Objective

• To develop an emulator that can accurately reconstruct large-scale climate variability over the Common Era (CE) using paleoclimate proxies and physically based information from climate models.

Study Configuration

• Spatial Scale: Global, with a focus on annual surface air temperature field emulation.
• Temporal Scale: Annual to centennial timescales.

Methodology and Data

• Models used:
  • Linear Inverse Model (LIM)
  • Reservoir Computing (RC)
  • Recurrent Neural Networks (RNNs)
  • Autoencoders (AEs)
• Data sources:
  • CMIP6 model simulations, including IPSL-CM6A-LR and 52 other models
  • Paleoclimate proxies

Main Results

• The proposed emulators show improved representation of El Niño Southern Oscillation and Atlantic Multidecadal Variability compared to traditional EOF-based dimensionality reduction.
• The AERCn configuration, which combines an AE with RC dynamics, yields the most robust results when training data are plentiful.
• When training data are scarce, a multimodel pre-trained AERNN provides a competitive alternative.

Contributions

• This study contributes to the development of next-generation emulators tailored to the needs of paleoclimate data assimilation (PDA) and climate predictability research.
• The proposed emulators can be used within sequential ensemble-based data assimilation frameworks, including Ensemble Kalman Filters and Particle Filters.

Funding

• This work was funded by the European Union's Horizon 2020 research and innovation program under grant agreement No. [insert grant number].

Citation

@article{Gaudin2026New,
  author = {Gaudin, Auguste and Khodri, Myriam},
  title = {New classes of climate model emulators to improve paleoclimate reconstructions},
  journal = {Geoscientific model development},
  year = {2026},
  doi = {10.5194/gmd-19-7135-2026},
  url = {https://doi.org/10.5194/gmd-19-7135-2026}
}