Differentiable programming for Earth system modeling
Open accessPublished version of a preprint
The DOI matches its Crossref record; 72 of the 72 references it cites were found in a registry or an index.
Abstract
Earth system models (ESMs) are the primary tools for investigating future Earth system states at timescales from decades to centuries, especially in response to anthropogenic greenhouse gas release. State-of-the-art ESMs can reproduce the observational global mean temperature anomalies of the last 150 years. Nevertheless, ESMs need further improvements, most importantly regarding (i) the large spread in their estimates of climate sensitivity, i.e., the temperature response to increases in atmospheric greenhouse gases; (ii) the modeled spatial patterns of key variables such as temperature and precipitation; (iii) their representation of extreme weather events; and (iv) their representation of multistable Earth system components and the ability to predict associated abrupt transitions. Here, we argue that making ESMs automatically differentiable has a huge potential to advance ESMs, especially with respect to these key shortcomings. First, automatic differentiability would allow objective calibration of ESMs, i.e., the selection of optimal values with respect to a cost function for a large number of free parameters, which are currently tuned mostly manually. Second, recent advances in machine learning (ML) and in the number, accuracy, and resolution of observational data promise to be helpful with at least some of the above aspects because ML may be used to incorporate additional information from observations into ESMs. Automatic differentiability is an essential ingredient in the construction of such hybrid models, combining process-based ESMs with ML components. We document recent work showcasing the potential of automatic differentiation for a new generation of substantially improved, data-informed ESMs.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“Differentiability enables end-to-end ‘online training’27, with machine-learning components optimized in the context of interactions with the governing equations for large-scale dynamics, which we find enables accurate and stable forecasts.”
From Neural general circulation models for weather and climate (2024) · cites this paper as [27]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Raissi, M., Perdikaris, P., & Karniadakis, G. E. (2019) · Journal of Computational Physics · doi:10.1016/j.jcp.2018.10.045 · Cited by 20,384 · cited 1×
“A number of approaches have been proposed to combine physical laws with ML. Physics-informed neural networks (Raissi et al., 2019) penalize physically inconsistent solutions during training, the penalty acting as a regularizer which favors, but does not guarantee, physically consistent choices of…”
BackgroundEyring, V., Bony, S., Meehl, G. A., Senior, C. A., Stevens, B., Stouffer, R. J., & Taylor, K. E. (2016) · Geoscientific Model Development · doi:10.5194/gmd-9-1937-2016 · Cited by 11,981 · cited 2×
“Comprehensive ESMs, such as those used for the projections of the Coupled Model Intercomparison Project (CMIP) (Eyring et al., 2016), are highly complex numerical algorithms consisting of hundreds of thousands of lines of Fortran code, which solve the relevant equations (such as the equations of…”
MethodBayesian calibration of computer models
Kennedy, M. C., & O'Hagan, A. (2001) · Journal of the Royal Statistical Society Series B: Statistical Methodology · doi:10.1111/1467-9868.00294 · Cited by 4,222
Duane, S., Kennedy, A. D., Pendleton, B. J., & Roweth, D. (1987) · Physics Letters B · doi:10.1016/0370-2693(87)91197-x · Cited by 3,872
Automated Solution of Differential Equations by the Finite Element Method: The FEniCS Book
(2012) · Lecture Notes in Computational Science and Engineering · doi:10.1007/978-3-642-23099-8 · Cited by 2,204
Neural ordinary differential equations
Chen, R. T. Q., Rubanova, Y., Bettencourt, J., & Duvenaud, D. (2018) · arXiv · doi:10.48550/arxiv.1806.07366 · Cited by 2,097 · cited 2×
“Universal differential equations (UDEs) can be such a physics-aware form of ML, as they can combine primitive physical equations directly with ANNs to minimize model errors effectively. de Bézenac et al. (2019) developed a hybrid model to forecast high-resolution sea surface temperatures by using the output of an ANN…”
BackgroundDeep learning to represent subgrid processes in climate models
Rasp, S., Pritchard, M. S., & Gentine, P. (2018) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1810286115 · Cited by 1,044
Machine learning–accelerated computational fluid dynamics
Kochkov, D., Smith, J. A., Alieva, A., Wang, Q., Brenner, M. P., & Hoyer, S. (2021) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.2101784118 · Cited by 1,042 · cited 8×
“Based on these tools, partial differential equation (PDE) solvers for fluid dynamics have recently been developed (Holl et al., 2020; Kochkov et al., 2021; Bezgin et al., 2023), and advances regarding AD in other fields, such as molecular dynamics (Schoenholz and Cubuk, 2020) and cosmology…”
BackgroundMethodInfluentialRecipes for adjoint code construction
Giering, R., & Kaminski, T. (1998) · ACM Transactions on Mathematical Software · doi:10.1145/293686.293695 · Cited by 809 · cited 6×
“However, considerable efforts have also been spent on frameworks such as dolfin-adjoint for finite-element models (Mitusch et al., 2019) as well as the source-to-source AD tools Transforms of Algorithms in Fortran (TAF) (Giering and Kaminski, 1998) and Tapenade (Hascoët and Pascual, 2013) that already enabled differen…”
BackgroundMethodInfluentialUniversal Differential Equations for Scientific Machine Learning
Rackauckas, C., Ma, Y., Martensen, J., Warner, C., Zubov, K., Supekar, R., Skinner, D., Ramadhan, A., & Edelman, A. (2020) · doi:10.21203/rs.3.rs-55125/v1 · Cited by 643
Research works printed but not linked: 7
- Chizat, L., Oyallon, E., and Bach, F.: On Lazy Training in Differentiable Programming, in: Advances in Neural Information Processing Systems, edited by: Wallach, H., Larochelle, H., Beygelzimer, A., d'Alché-Buc, F., Fox, E., and Garnett, R., Curran Associates, vol. 32, Inc., https://proceedings.neurips.cc/paper/2019/file/ae614c557843b1df326cb29c57225459-Paper.pdf (last access: 30 May 2023), 2019.
- Frezat, H., Sommer, J. L., Fablet, R., Balarac, G., and Lguensat, R.: A posteriori learning for quasi-geostrophic turbulence parametrization, ArXiv, 10.48550/ARXIV.2204.03911, 2022. · DOI
- Ge, H., Xu, K., and Ghahramani, Z.: Turing: A Language for Flexible Probabilistic Inference, in: Proceedings of the Twenty-First International Conference on Artificial Intelligence and Statistics, edited by: Storkey, A. and Perez-Cruz, F., Proc. Mach. Learn. Res., 84, 1682–1690, https://proceedings.mlr.press/v84/ge18b.html (last access: 30 May 2023), 2018.
- Holl, P., Thuerey, N., and Koltun, V.: Learning to Control PDEs with Differentiable Physics, International Conference on Learning Representations, https://openreview.net/forum?id=HyeSin4FPB (last access: 31 May 2023), 2020.
- Michalak, K. and Ollivier-Gooch, C.: Differentiability of slope limiters on unstructured grids, in: Proceedings of fourteenth annual conference of the computational fluid dynamics society of Canada, https://scholar.google.com/scholar?hl=en&as_sdt=0%2C5&q=Differentiability+of+Slope+Limiters+on+Unstructured+Grids&btnG= (last access: 31 May 2023), 2006.
- Moses, W. and Churavy, V.: Instead of Rewriting Foreign Code for Machine Learning, Automatically Synthesize Fast Gradients, in: Advances in Neural Information Processing Systems, edited by: Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M. F., and Lin, H., 33, 12472–12485, Curran Associates, Inc., https://proceedings.neurips.cc/paper/2020/file/9332c513ef44b682e9347822c2e457ac-Paper.pdf (last access: 30 May 2023), 2020.
- Schoenholz, S. and Cubuk, E. D.: JAX MD: A Framework for Differentiable Physics, in: Advances in Neural Information Processing Systems, edited by: Larochelle, H., Ranzato, M., Hadsell, R., Balcan, M., and Lin, H., 33, 11428–11441, Curran Associates, Inc., https://proceedings.neurips.cc/paper_files/paper/2020/file/83d3d4b6c9579515e1679aca8cbc8033-Paper.pdf (last access: 30 May 2023), 2020.
Cited by
Neural general circulation models for weather and climate
Kochkov, D., Yuval, J., Langmore, I., Norgaard, P., Smith, J., Mooers, G., Klöwer, M., Lottes, J., Rasp, S., Düben, P., et al. (2024) · Nature · doi:10.1038/s41586-024-07744-y · Cited by 454
“26 and 27.”
BackgroundWillard, J., Varadharajan, C., Jia, X., & Kumar, V. (2025) · Environmental Data Science · doi:10.1017/eds.2024.14 · Cited by 59
“…to induce a model, (2) transductive transfer learning where the source and target tasks are the same but from different feature space domains and zero labeled data is available from the target domain, and (3) unsupervised transfer learning where no labeled data is available in both the source and…”
Pahlavan, H. A., Hassanzadeh, P., & Alexander, M. J. (2024) · Geophysical Research Letters · doi:10.1029/2023gl106324 · Cited by 34
“For online learning, various methods such as reinforcement learning (Mojgani et al., 2023; Novati et al., 2021), differentiable programming (Frezat et al., 2022; Gelbrecht et al., 2023), and ensemble Kalman inversion (EKI) (Iglesias et al., 2013; Lopez-Gomez et al., 2022) can be used.”
MethodResultWesselkamp, M., Chantry, M., Pinnington, E., Choulga, M., Boussetta, S., Kalweit, G., Bödecker, J., Dormann, C. F., Pappenberger, F., & Balsamo, G. (2025) · Geoscientific model development · doi:10.5194/gmd-18-921-2025 · Cited by 22
Reconstructing historical climate fields with deep learning
Bochow, N., Poltronieri, A., Rypdal, M. W., & Boers, N. (2025) · Science Advances · doi:10.1126/sciadv.adp0558 · Cited by 20
Self-citation
Metadata
- Authors
- Gelbrecht, M., White, A., Bathiany, S., & Boers, N.
- Affiliations
- Potsdam Institute for Climate Impact Research (DE)
- University of Exeter (GB)
- Corresponding author
- Gelbrecht, Maximilian
- Venue
- Geoscientific Model Development
- Publisher
- Copernicus GmbH
- Published
- 2023-06-02
- Volume
- 16
- Issue
- 11
- Pages
- 3123-3135
- ISSN
- 1991-959X (ISSN-L)
- 1991-9603
- OpenAlex
- W4379230313
- Semantic Scholar
- dcb49fd04f350e4bbb6db20d00e82bac693adcce
- Cited by
- 31
- References
- 71
- Citations by year
- 2026: 13 · 2025: 12 · 2024: 5
- FWCI
- 3.36
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- Geoscientific model development (publishedVersion, cc-by)
- arXiv (Cornell University) (submittedVersion)
- Publication Database PIK (Potsdam Institute for Climate Impact Research (PIK)) (submittedVersion, other-oa)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- arXiv (Cornell University) (cc-by)
- Funders
- Horizon 2020
- HORIZON EUROPE Marie Sklodowska-Curie Actions (956170)
- Bundesministerium für Forschung, Technologie und Raumfahrt (01LS2001A)
- Volkswagen Foundation (-)
- Topics
- Physical Sciences › Environmental Science › Global and Planetary Change › Climate variability and models
- Physical Sciences › Earth and Planetary Sciences › Atmospheric Science › Meteorological Phenomena and Simulations
- Physical Sciences › Environmental Science › Global and Planetary Change › Climate Change and Geoengineering
- Keywords
- Earth system models, automatic differentiation, differentiable programming, climate sensitivity, model calibration, extreme weather events, abrupt climate change, climate change projections
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Differentiable programming for Earth system modeling. https://app.cochat.ai/papers/10.5194/gmd-16-3123-2023
- The paper
- Gelbrecht, M., White, A., Bathiany, S., & Boers, N. (2023). Differentiable programming for Earth system modeling. Geoscientific Model Development. https://doi.org/10.5194/gmd-16-3123-2023
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.