Journal article · 2026 · Open access
Fine‐Tuning a Weather Foundation Model With Lightweight Decoders for Unseen Physical Processes
Journal of Geophysical Research Machine Learning and Computation, 3(4) · Published 23 July 2026 · DOI 10.1029/2025jh000837
In brief
“Fine‐Tuning a Weather Foundation Model With Lightweight Decoders for Unseen Physical Processes” is a 2026 journal article by Fanny Lehmann and 5 co-authors, published in Journal of Geophysical Research Machine Learning and Computation. OpenAlex counts 14 citations as of 15 November 2026, which puts it in the top 7% of works published in 2026 by citations. It is open access under a CC BY licence: the full text is free to read at Journal of Geophysical Research Machine Learning and Computation.
- Citations
- 14
- Most cited in
- 2026 (12)
- Field-weighted impact
- 19.24
- References
- 30
- Authors
- 6
Abstract
Recent advances in AI weather forecasting have led to the emergence of so‐called “foundation models”, typically defined by expensive pretraining and minimal fine‐tuning for downstream tasks. However, in the natural sciences, a desirable foundation model should also encode meaningful statistical relationships between the underlying physical variables. This study evaluates the performance of the state‐of‐the‐art Aurora foundation model in predicting hydrological variables, which were not included during pretraining. We introduce a lightweight approach using shallow decoders trained on the latent representations of the pretrained model to predict these new variables. As a baseline, we compare this to fine‐tuning the full model, which allows further optimization of the latent space while incorporating new variables into both inputs and outputs. The decoder‐based approach requires 43% less training time and 53% less memory, while achieving strong accuracy across various hydrological variables and preserving desirable properties of the foundation model, such as autoregressive stability. Notably, decoder accuracy depends on the physical correlation between the new variables and those used during pretraining, indicating that Aurora's latent space captures meaningful physical relationships. The decoders outperform UNet, climatology, and persistence baselines on variables related to pretraining, showing the benefits of the latent space. In this sense, we argue that an important quality metric for foundation models in Earth sciences is their ability to be extended to new variables without a full fine‐tuning. This provides a new perspective for making foundation models more accessible to communities with limited computational resources, while supporting broader adoption in Earth sciences.
Cite this paper
Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.
APA 7
Lehmann, F., Özdemir, F., Soja, B. S., Hoefler, T., Mishra, S., & Schemm, S. (2026). Fine‐Tuning a Weather Foundation Model With Lightweight Decoders for Unseen Physical Processes. Journal of Geophysical Research Machine Learning and Computation, 3(4). https://doi.org/10.1029/2025jh000837
MLA 9
Lehmann, Fanny, et al. “Fine‐Tuning a Weather Foundation Model With Lightweight Decoders for Unseen Physical Processes.” Journal of Geophysical Research Machine Learning and Computation, vol. 3, no. 4, 2026. https://doi.org/10.1029/2025jh000837.
Chicago
Lehmann, Fanny, Fırat Özdemir, Benedikt S. Soja, Torsten Hoefler, Siddhartha Mishra, and Sebastian Schemm. 2026. “Fine‐Tuning a Weather Foundation Model With Lightweight Decoders for Unseen Physical Processes.” Journal of Geophysical Research Machine Learning and Computation 3 (4). https://doi.org/10.1029/2025jh000837.
Harvard
Lehmann, F. et al. (2026) ‘Fine‐Tuning a Weather Foundation Model With Lightweight Decoders for Unseen Physical Processes’, Journal of Geophysical Research Machine Learning and Computation, 3(4). doi:10.1029/2025jh000837.
IEEE
F. Lehmann, F. Özdemir, B. S. Soja, T. Hoefler, S. Mishra, and S. Schemm, “Fine‐Tuning a Weather Foundation Model With Lightweight Decoders for Unseen Physical Processes,” Journal of Geophysical Research Machine Learning and Computation, vol. 3, no. 4, 2026, doi: 10.1029/2025jh000837.
BibTeX
@article{lehmann2026fine,
title = {{Fine‐Tuning a Weather Foundation Model With Lightweight Decoders for Unseen Physical Processes}},
author = {Lehmann, Fanny and Özdemir, Fırat and Soja, Benedikt S. and Hoefler, Torsten and Mishra, Siddhartha and Schemm, Sebastian},
journal = {Journal of Geophysical Research Machine Learning and Computation},
year = {2026},
volume = {3},
number = {4},
doi = {10.1029/2025jh000837},
url = {https://doi.org/10.1029/2025jh000837}
}
Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)
Topic and keywords
Physical Sciences › Environmental Science › Environmental Engineering › Hydrological Forecasting Using AI
- foundation models
- lightweight decoder
- hydrological variables
- latent space
- fine-tuning
- computational efficiency
- Earth system science
Work with this paper
Most-cited papers in Environmental Science
- Biodiversity hotspots for conservation prioritiesNorman Myers, Russell A. Mittermeier, Cristina Goettsch Mittermeier and 2 more · Nature · 2000 · 32,720 citations
- The Tragedy of the CommonsGarrett Hardin · Science · 1968 · 26,194 citations
- River flow forecasting through conceptual models part I — A discussion of principlesJ.E. Nash, John V. Sutcliffe · Journal of Hydrology · 1970 · 24,625 citations
- The value of the world's ecosystem services and natural capitalRobert Costanza, Ralph C. d’Arge, R.S. de Groot and 10 more · Nature · 1997 · 23,508 citations
- Climate change 2007: the physical science basisSusan Solomon, Dahe Qin, Martin Manning and 5 more · 2007 · 23,383 citations
Questions about this paper
How do I cite this paper in APA style?
Lehmann, F., Özdemir, F., Soja, B. S., Hoefler, T., Mishra, S., & Schemm, S. (2026). Fine‐Tuning a Weather Foundation Model With Lightweight Decoders for Unseen Physical Processes. Journal of Geophysical Research Machine Learning and Computation, 3(4). https://doi.org/10.1029/2025jh000837
How many times has this paper been cited?
14 times, according to OpenAlex on 15 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.
Is this paper free to read?
Yes. It is open access: read the free full text at Journal of Geophysical Research Machine Learning and Computation.
Who wrote this paper?
Fanny Lehmann, Fırat Özdemir, Benedikt S. Soja, Torsten Hoefler, Siddhartha Mishra, Sebastian Schemm.
What is the DOI of this paper?
10.1029/2025jh000837. The permanent link is https://doi.org/10.1029/2025jh000837.
Bibliographic data from OpenAlex (CC0), last checked 15 November 2026. ResearchGuru does not host this paper; the links go to the publisher or the repository that does.