pith. machine review for the scientific record. sign in

arxiv: 2605.10455 · v1 · submitted 2026-05-11 · 💻 cs.LG

Recognition: 2 theorem links

· Lean Theorem

AxiomOcean: Forecasting the Three-Dimensional Structure of the Upper Ocean

Sensen Wu , Yifan Chen , Guantao Pu , Xiaoyao Sun , Yijun Chen , Jin Qi , Ming Kong , Keyi Yang
show 4 more authors
Lichen Xu Wenguan Wang Xiaofeng Li Zhenhong Du
Authors on Pith no claims yet

Pith reviewed 2026-05-12 04:09 UTC · model grok-4.3

classification 💻 cs.LG
keywords ocean forecastingAI ocean modelthree-dimensional structureupper oceantemperature salinity currentseddy kinetic energyglobal predictionphysical fidelity
0
0 comments X

The pith

Explicitly preserving three-dimensional upper-ocean structure improves AI forecast accuracy and physical fidelity.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents AxiomOcean as a global AI model that forecasts upper-ocean temperature, salinity, and currents by maintaining the full vertical hierarchy down to 643 m depth. It claims that conventional AI ocean models lose this structure and produce overly smooth subsurface fields that weaken under strong atmospheric forcing. By using a three-dimensional encoder-backbone-decoder architecture together with surface forcing, AxiomOcean delivers lower errors and higher anomaly correlations over ten-day forecasts while retaining eddy kinetic energy and variance. A sympathetic reader would care because upper-ocean structure controls heat storage, stratification, and the ocean's response to weather, directly affecting marine forecasts and climate-relevant quantities. The central demonstration is that these gains appear across variables, depths, and regions when the vertical dependence is modeled explicitly.

Core claim

AxiomOcean is a fully three-dimensional AI ocean forecasting model that jointly predicts temperature, salinity, and three-dimensional currents at global 1/12° resolution. In ten-day forecasts it reduces day-1 RMSE by 20–35 % relative to an advanced comparison model while preserving higher anomaly correlation and better maintaining eddy kinetic energy, temperature-salinity variance, and upper-ocean heat content. The advantage holds through the water column and across the equatorial Pacific, Kuroshio Extension, and Southern Ocean.

What carries the argument

The fully three-dimensional encoder-backbone-decoder architecture that explicitly represents vertical hierarchy and cross-layer dependence within the water column.

If this is right

  • Forecasts maintain realistic subsurface features instead of over-smoothing eddies and heat content.
  • Physical consistency improves under strong atmospheric forcing across variables and lead times.
  • Advantages extend through the water column and appear in major current systems such as the equatorial Pacific and Kuroshio Extension.
  • Upper-ocean heat content is reconstructed more realistically than in models that ignore vertical structure.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same vertical-hierarchy design could be tested in coupled ocean-atmosphere systems to check whether surface flux consistency also improves.
  • If the 3D representation reduces bias accumulation, similar architectures might help longer-range climate simulations that rely on accurate subsurface heat storage.
  • A controlled scaling study that varies only the vertical connectivity while holding total parameters fixed would isolate the contribution of hierarchy from raw capacity.

Load-bearing premise

The reported forecast gains arise specifically from the explicit three-dimensional vertical hierarchy rather than from differences in model capacity, training data volume, or hyperparameter choices.

What would settle it

Train two models of identical capacity on the same data volume, one with and one without the three-dimensional vertical hierarchy, then compare their independent-test RMSE, anomaly correlation, and eddy kinetic energy preservation at multiple lead times and depths.

read the original abstract

Short-term ocean forecast skill depends strongly on the three-dimensional ocean structure of the upper ocean, which governs stratification, subsurface heat storage, and the response of the ocean to atmospheric forcing. However, AI ocean forecasting models often fail to preserve this vertical structure, resulting in over-smoothed subsurface features and weak physical consistency under strong forcing. Here, we present AxiomOcean, a global AI ocean forecasting model that explicitly represents vertical hierarchy and cross-layer dependence within the water column. By combining a fully three-dimensional encoder-backbone-decoder architecture with surface atmospheric forcing, AxiomOcean jointly predicts upper-ocean temperature, salinity, and three-dimensional currents at global 1/12{\deg} resolution down to 643 m depth. In 10-day forecasts, AxiomOcean outperforms an advanced AI comparison model across variables and lead times, reducing day-1 RMSE by approximately 20 to 35% while maintaining higher anomaly correlation. The gain is not achieved through excessive smoothing: AxiomOcean better preserves eddy kinetic energy, temperature and salinity variance. Its advantage also extends through the water column and remains evident across the equatorial Pacific, Kuroshio Extension, and Southern Ocean, yielding a more realistic reconstruction of upper-ocean heat content. These results show that explicitly preserving upper-ocean three-dimensional structure can improve both forecast accuracy and physical fidelity in AI ocean prediction.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces AxiomOcean, a global AI ocean forecasting model using a fully three-dimensional encoder-backbone-decoder architecture with surface atmospheric forcing. It jointly predicts upper-ocean temperature, salinity, and 3D currents at 1/12° resolution down to 643 m depth. The central claim is that explicitly preserving vertical hierarchy and cross-layer dependence yields 20-35% lower day-1 RMSE, higher anomaly correlation, and better preservation of eddy kinetic energy, temperature/salinity variance, and upper-ocean heat content compared to an advanced AI baseline, with advantages extending through the water column in regions like the equatorial Pacific, Kuroshio Extension, and Southern Ocean.

Significance. If the performance and fidelity gains can be shown to arise specifically from the 3D vertical structure rather than capacity or training differences, the work would provide evidence that architecture choices enforcing physical vertical dependence improve both accuracy and consistency in AI ocean models. The emphasis on preserving EKE and variance (rather than over-smoothing) is a constructive direction, though the current evidence does not yet isolate this factor.

major comments (2)
  1. [Abstract] Abstract: The reported 20-35% RMSE reductions and improved anomaly correlation are presented without error bars, statistical tests, or any description of the comparison model's vertical resolution, parameter count, training data volume, or optimizer settings. This prevents attribution of gains to the 'explicitly preserving upper-ocean three-dimensional structure' (the central claim) versus unstated differences in model capacity or regularization.
  2. [Abstract] Abstract and methods (inferred from architecture description): No ablation studies, size-matched 2D baselines, or flattened equivalents trained under identical conditions are described. Without these controls, the claim that the 3D encoder-backbone-decoder plus cross-layer dependence is responsible for the preserved EKE, variance, and heat-content realism cannot be isolated from confounding factors.
minor comments (2)
  1. [Abstract] Abstract: The comparison model is referred to only as 'an advanced AI comparison model' without citation or name; this should be specified with a reference for reproducibility.
  2. [Abstract] Abstract: The depth range 'down to 643 m' and resolution '1/12°' are stated without clarifying the exact vertical levels or grid staggering used in the 3D architecture; adding this detail would aid physical interpretation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive feedback. We agree that stronger controls are required to isolate the contribution of the three-dimensional architecture from potential differences in model capacity or training. We will revise the manuscript to address both major comments by adding the requested statistical details, model specifications, and ablation studies.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The reported 20-35% RMSE reductions and improved anomaly correlation are presented without error bars, statistical tests, or any description of the comparison model's vertical resolution, parameter count, training data volume, or optimizer settings. This prevents attribution of gains to the 'explicitly preserving upper-ocean three-dimensional structure' (the central claim) versus unstated differences in model capacity or regularization.

    Authors: We acknowledge that the current manuscript does not provide error bars, statistical tests, or a full specification of the comparison model, which limits the strength of attribution. In the revised version we will add: (i) error bars derived from multiple independent training runs with different random seeds; (ii) paired statistical significance tests (e.g., Wilcoxon signed-rank) on the day-1 RMSE and anomaly-correlation differences; and (iii) a dedicated table and methods paragraph detailing the baseline model's vertical resolution (surface-focused 2D layers with no explicit cross-depth connections), parameter count, training data volume, optimizer, and learning-rate schedule. These additions will allow readers to evaluate whether the reported gains arise primarily from the 3D vertical hierarchy rather than capacity or regularization differences. revision: yes

  2. Referee: [Abstract] Abstract and methods (inferred from architecture description): No ablation studies, size-matched 2D baselines, or flattened equivalents trained under identical conditions are described. Without these controls, the claim that the 3D encoder-backbone-decoder plus cross-layer dependence is responsible for the preserved EKE, variance, and heat-content realism cannot be isolated from confounding factors.

    Authors: We agree that the absence of controlled ablations prevents definitive isolation of the architectural contribution. We will add a new subsection presenting ablation experiments trained under identical conditions (same dataset, optimizer, batch size, and number of epochs). These will include: (1) a size-matched 2D baseline obtained by replacing the 3D convolutions and attention layers with equivalent-capacity 2D layers applied independently at each depth; and (2) a 'flattened' 3D variant in which vertical connections are removed while preserving total parameter count. All variants will be evaluated on the same metrics (RMSE, EKE spectra, temperature/salinity variance, and upper-ocean heat content) to quantify the specific benefit of explicit cross-layer dependence. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical performance claims rest on external model comparisons

full rationale

The paper presents AxiomOcean as a 3D encoder-backbone-decoder architecture for ocean forecasting and reports empirical gains (20-35% RMSE reduction, higher anomaly correlation, better EKE/variance preservation) versus an external advanced AI comparison model. No equations, fitted parameters, or derivations appear in the abstract or described text that would reduce any prediction or result to the inputs by construction. The central claim attributes improvements to explicit vertical hierarchy and cross-layer dependence, but this is framed as an observed outcome of the architecture rather than a self-referential definition or renamed known result. No self-citations are invoked as load-bearing uniqueness theorems, and no ansatzes are smuggled via prior work. The comparison metrics are falsifiable against benchmarks and do not collapse into tautology. This is a standard empirical AI modeling paper whose derivation chain is self-contained.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The performance claim rests on a trained neural network whose weights constitute thousands of fitted parameters; no explicit invented physical entities are introduced, but standard ML assumptions about data representativeness and generalization are required.

free parameters (1)
  • neural network weights and biases
    All trainable parameters in the 3D encoder-backbone-decoder are fitted to ocean reanalysis or observational data during training.
axioms (1)
  • domain assumption Training and test data distributions are representative of future ocean states
    Implicit in any data-driven forecasting claim; stated in the abstract's evaluation setup.

pith-pipeline@v0.9.0 · 5579 in / 1266 out tokens · 33013 ms · 2026-05-12T04:09:24.644911+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages

  1. [1]

    R. A. Weller, J. P. Dean, J. F. Price, E. A. Francis, J. Marra, D. C. Boardman, Three-dimensional flow in the upper ocean. Science 227, 1552-1556 (1985)

    work page 1985

  2. [2]

    de Boyer Montegut, G

    C. de Boyer Montegut, G. Madec, A. S. Fischer, A. Lazar, D. Iudicone, Mixed layer depth over the global ocean: An examination of profile data and a profile-based climatology. J. Geophys. Res. 109, C12003 (2004). https://doi.org/10.1029/2004JC002378

    work page doi:10.1029/2004jc002378 2004

  3. [3]

    K. S. Smith, R. Ferrari, The production and dissipation of compensated thermohaline variance by mesoscale stirring. J. Phys. Oceanogr. 39, 2477-2501 (2009). https://doi.org/10.1175/2009JPO4103.1

    work page doi:10.1175/2009jpo4103.1 2009

  4. [4]

    C. Bian, Z. Jing, H. Wang, et al., Oceanic mesoscale eddies as crucial drivers of global marine heatwaves. Nat. Commun. 14, 2970 (2023). https://doi.org/10.1038/s41467-023-38811-z

    work page doi:10.1038/s41467-023-38811-z 2023

  5. [5]

    Zhang, Y

    Y. Zhang, Y. Du, M. Feng, et al., Vertical structures of marine heatwaves. Nat. Commun. 14, 6483 (2023). https://doi.org/10.1038/s41467-023-42219-0

    work page doi:10.1038/s41467-023-42219-0 2023

  6. [6]

    Q. He, W. Zhan, M. Feng, et al., Common occurrences of subsurface heatwaves and cold spells in ocean eddies. Nature 634, 1111-1117 (2024). https://doi.org/10.1038/s41586-024-08051-2

    work page doi:10.1038/s41586-024-08051-2 2024

  7. [7]

    D. B. Chelton, P. Gaube, M. G. Schlax, et al., The influence of nonlinear mesoscale eddies on near-surface oceanic chlorophyll. Science 334, 328-332 (2011). https://doi.org/10.1126/science.1208897

    work page doi:10.1126/science.1208897 2011

  8. [8]

    D. J. McGillicuddy Jr., Mechanisms of physical-biological-biogeochemical interaction at the oceanic mesoscale. Annu. Rev. Mar. Sci. 8, 125-159 (2016)

    work page 2016

  9. [9]

    Zhang, H

    H. Zhang, H. He, W.-Z. Zhang, et al., Upper ocean response to tropical cyclones: A review. Geosci. Lett. 8, 1 (2021). https://doi.org/10.1186/s40562-020-00170-8

    work page doi:10.1186/s40562-020-00170-8 2021

  10. [10]

    Lin, C.-H

    I.-I. Lin, C.-H. Chen, I.-F. Pun, et al., Warm ocean anomaly, air-sea fluxes, and the rapid intensification of tropical cyclone Nargis (2008). Geophys. Res. Lett. 36, L03817 (2009). https://doi.org/10.1029/2008GL035815

    work page doi:10.1029/2008gl035815 2008

  11. [11]

    Storto, A

    A. Storto, A. Alvera-Azcarate, M. A. Balmaseda, et al., Ocean reanalyses: Recent advances and unsolved challenges. Front. Mar. Sci. 6, 418 (2019). https://doi.org/10.3389/fmars.2019.00418

    work page doi:10.3389/fmars.2019.00418 2019

  12. [12]

    Barnier, G

    B. Barnier, G. Madec, T. Penduff, et al., Impact of partial steps and momentum advection schemes in a global ocean circulation model at eddy-permitting resolution. Ocean Dyn. 56, 543- 567 (2006). https://doi.org/10.1007/s10236-006-0082-1

    work page doi:10.1007/s10236-006-0082-1 2006

  13. [13]

    Jean-Michel, E

    L. Jean-Michel, E. Dombrowsky, C. Cabanes, et al., The Copernicus Global 1/12° Oceanic and Sea Ice GLORYS12 reanalysis. Front. Earth Sci. 9, 698876 (2021). https://doi.org/10.3389/feart.2021.698876

    work page doi:10.3389/feart.2021.698876 2021

  14. [14]

    Hersbach, B

    H. Hersbach, B. Bell, P. Berrisford, et al., The ERA5 global reanalysis. Q. J. R. Meteorol. Soc. 146, 1999-2049 (2020). https://doi.org/10.1002/qj.3803

    work page doi:10.1002/qj.3803 1999

  15. [15]

    W. G. Large, S. G. Yeager, The global climatology of an interannually varying air-sea flux data set. Clim. Dyn. 33, 341-364 (2009). https://doi.org/10.1007/s00382-008-0441-3

    work page doi:10.1007/s00382-008-0441-3 2009

  16. [16]

    K. Bi, L. Xie, H. Zhang, et al., Accurate medium-range global weather forecasting with 3D neural networks. Nature 619, 533-538 (2023). https://doi.org/10.1038/s41586-023-06185-3

    work page doi:10.1038/s41586-023-06185-3 2023

  17. [17]

    R. Lam, A. Sanchez-Gonzalez, M. Willson, et al., Learning skillful medium-range global weather forecasting. Science 382, 1416-1421 (2023). https://doi.org/10.1126/science.adi2336

    work page doi:10.1126/science.adi2336 2023

  18. [18]

    L. Chen, X. Zhong, F. Zhang, et al., FuXi: A cascade machine learning forecasting system for 15-day global weather forecast. npj Clim. Atmos. Sci. 6, 190 (2023). https://doi.org/10.1038/s41612-023-00512-1

    work page doi:10.1038/s41612-023-00512-1 2023

  19. [19]

    Allen, S

    A. Allen, S. Markou, W. Tebbutt, et al., End-to-end data-driven weather prediction. Nature 641, 1172-1179 (2025). https://doi.org/10.1038/s41586-025-08897-0

    work page doi:10.1038/s41586-025-08897-0 2025

  20. [20]

    Nature637, 84–90 (2025) https://doi.org/10.1038/s41586-024-08252-9

    I. Price, A. Sanchez-Gonzalez, F. Alet, et al., Probabilistic weather forecasting with machine learning. Nature 637, 84-90 (2025). https://doi.org/10.1038/s41586-024-08252-9

    work page doi:10.1038/s41586-024-08252-9 2025

  21. [21]

    Chattopadhyay, M

    A. Chattopadhyay, M. Gray, T. Wu, et al., OceanNet: A principled neural operator-based digital twin for regional oceans. Sci. Rep. 14, 21181 (2024). https://doi.org/10.1038/s41598-024- 72145-0

    work page doi:10.1038/s41598-024- 2024

  22. [22]

    Holmberg, E

    D. Holmberg, E. Clementi, I. Epicoco, et al., Accurate Mediterranean Sea forecasting via graph-based deep learning. Sci. Rep. 15, 45051 (2025). https://doi.org/10.1038/s41598-025- 31177-w

    work page doi:10.1038/s41598-025- 2025

  23. [23]

    X. Wang, R. Wang, N. Hu, et al., XiHe: A data-driven model for global ocean eddy-resolving forecasting. Preprint (2024)

    work page 2024

  24. [24]

    El Aouni, Q

    A. El Aouni, Q. Gaudel, C. Regnier, et al., GLONET: Mercator's End-to-End Neural Global Ocean Forecasting System. J. Geophys. Res. Mach. Learn. Comput. 2, e2025JH000686 (2025). https://doi.org/10.1029/2025JH000686

    work page doi:10.1029/2025jh000686 2025

  25. [25]

    Y. Cui, R. Wu, X. Zhang, et al., Forecasting the eddying ocean with a deep neural network. Nat. Commun. 16, 2268 (2025). https://doi.org/10.1038/s41467-025-57389-2

    work page doi:10.1038/s41467-025-57389-2 2025

  26. [26]

    P. G. Thoppil, S. Frolov, C. D. Rowley, et al., Ensemble forecasting greatly expands the prediction horizon for ocean mesoscale variability. Commun. Earth Environ. 2, 89 (2021). https://doi.org/10.1038/s43247-021-00151-5

    work page doi:10.1038/s43247-021-00151-5 2021

  27. [27]

    B. A. Storer, M. Buzzicotti, H. Khatri, et al., Global energy spectrum of the general oceanic circulation. Nat. Commun. 13, 5314 (2022). https://doi.org/10.1038/s41467-022-33031-3

    work page doi:10.1038/s41467-022-33031-3 2022

  28. [28]

    Frenger, N

    I. Frenger, N. Gruber, R. Knutti, et al., Imprint of Southern Ocean eddies on winds, clouds and rainfall. Nat. Geosci. 6, 608-612 (2013). https://doi.org/10.1038/ngeo1863

    work page doi:10.1038/ngeo1863 2013

  29. [29]

    A. G. Ryan, C. Regnier, P. Divakaran, et al., GODAE OceanView Class 4 forecast verification framework: Global ocean inter-comparison. J. Oper. Oceanogr. 8, s98-s111 (2015). https://doi.org/10.1080/1755876X.2015.1022330

    work page doi:10.1080/1755876x.2015.1022330 2015

  30. [30]

    https://doi.org/10.48670/moi-00021

    Copernicus Marine Service, Global Ocean Physics Reanalysis (GLORYS12V1). https://doi.org/10.48670/moi-00021

    work page doi:10.48670/moi-00021

  31. [31]

    Balaguru, P

    K. Balaguru, P. Chang, R. Saravanan, et al., Ocean barrier layers' effect on tropical cyclone intensification. Proc. Natl. Acad. Sci. U.S.A. 109, 14343-14347 (2012). https://doi.org/10.1073/pnas.1201364109

    work page doi:10.1073/pnas.1201364109 2012

  32. [32]

    F.-F. Jin, J. Boucharel, I.-I. Lin, Eastern Pacific tropical cyclones intensified by El Nino delivery of subsurface ocean heat. Nature 516, 82-85 (2014). https://doi.org/10.1038/nature13958

    work page doi:10.1038/nature13958 2014

  33. [33]

    Balaguru, G

    K. Balaguru, G. R. Foltz, L. R. Leung, et al., Global warming-induced upper-ocean freshening and the intensification of super typhoons. Nat. Commun. 7, 13670 (2016). https://doi.org/10.1038/ncomms13670

    work page doi:10.1038/ncomms13670 2016

  34. [34]

    Huang, I.-I

    P. Huang, I.-I. Lin, C. Chou, et al., Change in ocean subsurface environment to suppress tropical cyclone intensification under global warming. Nat. Commun. 6, 7188 (2015). https://doi.org/10.1038/ncomms8188

    work page doi:10.1038/ncomms8188 2015

  35. [35]

    Roemmich, G

    D. Roemmich, G. C. Johnson, S. Riser, et al., The Argo Program: Observing the global ocean with profiling floats. Oceanography 22, 34-43 (2009). https://doi.org/10.5670/oceanog.2009.36

    work page doi:10.5670/oceanog.2009.36 2009

  36. [36]

    S. Zhao, J. Tang, D. Wang, et al., On the breakdown of ENSO's relationship with thermocline depth in the central-equatorial Pacific. Geophys. Res. Lett. 48, e2020GL091665 (2021). https://doi.org/10.1029/2020GL091665

    work page doi:10.1029/2020gl091665 2021

  37. [37]

    L. Xu, X. Li, S. R. Ramp, et al., Observing mesoscale eddy effects on mode-water subduction and transport in the North Pacific. Nat. Commun. 7, 10505 (2016). https://doi.org/10.1038/ncomms10505

    work page doi:10.1038/ncomms10505 2016

  38. [38]

    T. Du, Y. Xu, S. Guo, et al., Three-dimensional observations of a mesoscale eddy in the Kuroshio Extension based on multiple platforms. Sci. Data 12, 742 (2025). https://doi.org/10.1038/s41597-025-05298-z

    work page doi:10.1038/s41597-025-05298-z 2025

  39. [39]

    K. L. Sheen, D. P. Stevens, H. L. Johnson, et al., Eddy-induced variability in Southern Ocean abyssal mixing on climatic timescales. Nat. Geosci. 7, 577-582 (2014). https://doi.org/10.1038/ngeo2200

    work page doi:10.1038/ngeo2200 2014

  40. [40]

    X. Ma, Z. Jing, P. Chang, et al., Western boundary currents regulated by interaction between ocean eddies and the atmosphere. Nature 535, 533-537 (2016). https://doi.org/10.1038/nature18640

    work page doi:10.1038/nature18640 2016

  41. [41]

    Bonavita, On some limitations of current machine learning weather prediction models

    M. Bonavita, On some limitations of current machine learning weather prediction models. Geophys. Res. Lett. 51, e2023GL107377 (2024). https://doi.org/10.1029/2023GL107377

    work page doi:10.1029/2023gl107377 2024

  42. [42]

    A. J. Charlton-Perez, et al., Do AI models produce better weather forecasts than physics-based models? A quantitative evaluation case study of Storm Ciaran. npj Clim. Atmos. Sci. 7, 93 (2024). https://doi.org/10.1038/s41612-024-00638-w

    work page doi:10.1038/s41612-024-00638-w 2024

  43. [43]

    G. I. Marchuk, V. P. Kochergin, V. I. Klimok, V. A. Sukhorukov, On the dynamics of the ocean surface mixed layer. J. Phys. Oceanogr. 7, 865-875 (1977)

    work page 1977

  44. [44]

    Y. Wang, H. Cai, X. Wang, et al., Enhanced wind mixing and deepened mixed layer in the Pacific Arctic shelf seas with low summer sea ice. Nat. Commun. 15, 10389 (2024). https://doi.org/10.1038/s41467-024-54733-w

    work page doi:10.1038/s41467-024-54733-w 2024

  45. [45]

    J. F. Price, Upper ocean response to a hurricane. J. Phys. Oceanogr. 11, 153-175 (1981). https://doi.org/10.1175/1520-0485(1981)011<0153:UORTAH>2.0.CO;2

    work page doi:10.1175/1520-0485(1981)011 1981

  46. [46]

    Radfar, T

    A. Radfar, T. A. Magee, H. An, et al., Rapid intensification of tropical cyclones in the Gulf of Mexico is more likely during marine heatwaves. Commun. Earth Environ. 5, 421 (2024). https://doi.org/10.1038/s43247-024-01578-2

    work page doi:10.1038/s43247-024-01578-2 2024

  47. [47]

    Addressing fairness in artificial intelligence for medical imaging.Nature Communications, 13:4581, 2022

    K. Bhatia, A. Baker, W. Yang, et al., A potential explanation for the global increase in tropical cyclone rapid intensification. Nat. Commun. 13, 6626 (2022). https://doi.org/10.1038/s41467-022- 34321-6

    work page doi:10.1038/s41467-022- 2022

  48. [48]

    https://doi.org/10.24381/cds.143582cf

    Copernicus Climate Data Store, Complete ERA5 global atmospheric reanalysis. https://doi.org/10.24381/cds.143582cf

    work page doi:10.24381/cds.143582cf