Pith. sign in

REVIEW 5 major objections 7 minor 1 cited by

Ocean-E2E: Hybrid Physics-Based and Data-Driven Global Forecasting of Extreme Marine Heatwaves with End-to-End Neural Assimilation

T0 review · 5 major / 7 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a hybrid physics-plus-learned-closures ocean model, initialized from observations by a neural assimilation network, can forecast marine heatwaves for 40 days with lower error than operational numerical forecasts.

desk verdict A plausible hybrid framework whose SOTA claim currently depends on scoring the model against its own training reanalysis; send it to review but demand independent validation. read the letter →

arxiv 2505.22071 v3 pith:KL6ISMY3 submitted 2025-05-28 physics.geo-ph

Ruiqi Shu , Yuan Gao , Hao Wu , Ruijian Gou , Kun Wang , Yanfei Xiang , Fan Xu , Qingsong Wen
show 1 more author
Xiaomeng Huang
This is my paper · ORCID
classification physics.geo-ph
keywords marineheatwavesseasurfacetemperatureanomalieshybridphysics-AIforecastingneuraldataassimilationsubseasonal-to-seasonalforecastGM90bolusvelocityextremeeventglobaloceanmodeling
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

Marine heatwaves harm ecosystems and economies, yet existing forecasts either run costly numerical ocean models or train deep networks that smooth away extremes and depend on those models for initial conditions. Ocean-E2E tries to get the best of both: keep a simple physical transport equation for the temperature anomaly as the backbone, and let small neural networks learn the two processes the equation cannot resolve—subgrid mesoscale advection and air-sea heat exchange. A separate neural assimilation step starts directly from sparse observations, so the whole pipeline runs without a numerical ocean model. If the reported numbers hold, the framework produces 40-day global and regional forecasts with lower overall error and better detection of extreme events than the operational S2S numerical system and the listed deep-learning baselines.

What carries the argument

The central object is the SSTA transport identity $C_t=\int_0^t\left[-(u_g+u_\theta(u_g,C))\nabla C+\dot S_\theta(A,C)\right]dt+C_0$, discretized by forward Euler. The named machinery is the GM90 bolus velocity, the classical parameterization of subgrid eddy transport that the network $u_\theta$ is trained to emulate, so the advection term carries the mesoscale dynamics that otherwise smooths or destabilizes pure AI rollouts. The source term $\dot S_\theta$ absorbs air-sea interaction, mainly sensible and latent heat fluxes, and ageostrophic mixing, driven by atmospheric variables, while separate pretrained networks $M_\theta$ and $N_\theta$ evolve the geostrophic current and atmospheric fields that serve as boundary conditions. A neural assimilation network, built on a Kirsch-guided reparameterized convolution, maps sparse observations plus a perturbed background ensemble to the analysis field that initializes the forecast. The load-bearing structural choice is that the physics equation stays explicit and differentiable, so the learned terms act as closures rather than autoregressive replacements.

What would settle it

Evaluate Ocean-E2E and the S2S system for 2020-2021 lead times of 10-40 days against independent in-situ measurements such as moored buoys and Argo profiles, or against a different blended SST analysis, and check whether Ocean-E2E's average RMSE remains about 10 percent lower; if the advantage disappears or reverses, the central claim of independence from numerical models is not established.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that marine heatwave forecasting reduces to predicting the evolution of sea-surface temperature anomaly, and that this evolution can be written as a transport equation with two learned closure terms: an advective term $(u_g+u_\theta)\nabla C$, where the neural network $u_\theta$ plays the role of the GM90 bolus velocity that captures unresolved mesoscale eddy transport, and a source term $\dot S_\theta(A,C)$ that models air-sea heat fluxes and mixing given surface wind, temperature, and humidity. The forecast model integrates these equations with a forward-Euler scheme, driven at the boundaries by pretrained networks for geostrophic currents and atmospheric states. A second neural network performs data assimilation in one step, merging sparse observations with a background field produced by the same forecast model, which is what allows fully end-to-end forecasts without numerical initialization. Against the GLORYS12V1 reanalysis fields, the authors report an average 10 percent RMSE reduction over the S2S system across 40-day subseasonal-to-seasonal forecasts, larger RMSE and CSI gains against AI baselines in 40-60 day simulations, and stable regional 1/12-degree forecasts.

Load-bearing premise

The load-bearing premise is that the best-effort historical reconstruction of ocean temperatures that the model trains on is also an unbiased measure of reality, so comparing a model trained on that reconstruction against an independent numerical forecast system is fair.

Editorial extensions

If this is right

  • Global 40-day marine heatwave forecasts can run end-to-end from observations with an average 10% RMSE reduction against the operational S2S system in 2020-2021.
  • Extreme-event detection, measured by CSI, improves against all listed AI baselines in 40-60 day simulations, with the largest gains at longer lead times.
  • Both learned closures matter: removing the source term hurts CSI more, and removing the advection closure leaves the pure numerical kernel numerically unstable at strong gradients.
  • Regional 1/12-degree forecasts of the western Atlantic preserve the same physics-based structure and outperform baselines at high resolution.
  • Because assimilation errors stabilize within days and remain stable over a year, the same network can serve as an online analysis tool rather than only a forecast initializer.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, the same dynamic kernel should transfer to other passive oceanic tracers such as surface salinity, chlorophyll, or dissolved oxygen, because the framework only assumes tracer advection plus learnable sources and sinks.
  • A natural external test would compare Ocean-E2E not against GLORYS-trained baselines but against in-situ mooring and Argo temperatures, since training on GLORYS12V1 means an independent reference is the only way to separate physical skill from reanalysis-to-reanalysis consistency.
  • The learned bolus-velocity field $u_\theta$ could itself be diagnosed: if it resembles the classical GM90 streamfunction, the network is recovering a physically interpretable subgrid closure rather than an opaque correction term.
  • If the approach scales to near-real-time observations, it could provide a lightweight operational analysis and forecast pathway for regions without local numerical ocean model infrastructure.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 7 minor

Summary. The paper proposes Ocean-E2E, a hybrid physics-based/data-driven framework for 40-day global sea surface temperature anomaly (SSTA) and marine heatwave (MHW) forecasting with end-to-end neural data assimilation. The forecast model decomposes SSTA evolution into a learned geostrophic bolus-velocity advection term and a learned air-sea interaction source term, forced by a pretrained AI weather model (OneForecast) and a learned surface geostrophic current model. A neural assimilation module maps sparse observations plus a forecast background into an analysis field. The authors report improved RMSE and CSI over the ECMWF S2S system and several AI baselines for 2020-2021, plus regional 1/12° simulations in the western Atlantic, and they provide open-source code.

Significance. The hybrid decomposition is physically well motivated, and the end-to-end assimilation from sparse observations is a genuinely useful capability that could make MHW forecasting operational from observations without a numerical ocean model at run time. If the reported skill were established on independent data, the paper would be a significant advance in subseasonal MHW prediction. Strengths include the principled GM90-style learned bolus velocity, a clear ablation showing that both the advection and source terms contribute to skill, and the availability of code and detailed appendices. However, the headline state-of-the-art claims are not yet supported by the evaluation as designed: the main comparison is scored against a reanalysis used for training, the test period is only two years with no uncertainty quantification, and the abstract's '10%' claim does not match Table 2.

major comments (5)
  1. [Experiments, Table 2 and Figure 4] The headline comparison against S2S is scored against GLORYS12V1, which is also the training target for uθ, Sθ, and the assimilation network ϕaθ; S2S is an independent numerical system not calibrated to GLORYS, so the reported RMSE reductions may reflect closer agreement with the training reference rather than superior skill against the real ocean. Appendix G (Table 9) partially mitigates this by using S2S analysis fields as truth, but those fields are also model-derived and the forecast networks were still trained on GLORYS. A concrete remedy is to verify at least one full forecast season against an independent observational analysis such as OISST or HadISST, or to train on one reanalysis and verify on another (e.g., ORAS5).
  2. [Abstract and Table 2] The abstract's claim of 'an average 10% reduction in RMSE compared to ... across 40-day subseasonal-to-seasonal forecasts' is not what Table 2 shows: the 40-day promotions are 12.2% (2020) and 11.7% (2021), while the 10- to 30-day promotions are much larger on average (29.4%, 31.8%, 21.0%, 22.4%, 16.3%, 13.7%), giving an overall mean of about 19.8%. Please state explicitly which average is being reported and make the abstract and table consistent.
  3. [Table 2 / RQ2] The S2S comparison rests on only two initialization years (2020 and 2021) and no uncertainty intervals or significance tests are reported; with two starts, the differences are not sufficient to establish a general state-of-the-art claim, especially for extreme-event metrics that depend on the particular MHW events in those years. Please report per-initialization results with bootstrap or ensemble spread, and ideally extend the test period beyond two years.
  4. [Experiments, RQ2] The text states that 'We conducted two sets of assimilation experiments using analysis fields from January 1, 2020 and January 1, 2021 as initial conditions', yet Table 2 reports 10-, 20-, 30-, and 40-day RMSE for each year; it is unclear whether these are single-start forecasts from those two dates or averages over multiple starts per year. If they are single starts, the comparison is highly sensitive to the chosen initial date; please clarify the evaluation protocol and report the number of forecast initializations.
  5. [Abstract and Methods] The abstract's claim that Ocean-E2E can 'operate completely independently of numerical models' is stronger than what is demonstrated: the learned terms uθ, Sθ, and ϕaθ are trained on GLORYS12V1, which is a numerical model reanalysis, and the atmospheric boundary condition A is provided by OneForecast, which is trained on ERA5, also a numerical reanalysis. Independence holds only at inference time (initialization from observations); the claim should be qualified to avoid overstating the result.
minor comments (7)
  1. [Equation (5)] The equation w|z=0 = ηt is dimensionally inconsistent unless it means w|z=0 = ∂η/∂t; please correct the notation.
  2. [Equation (14)] The term '∇·dGug' appears to be a typo for the earlier '∇·cGu'; please unify the notation.
  3. [Equation (18)] The parentheses in the integral expression for Ct appear unbalanced; please check the matching of brackets in the integrand.
  4. [Table 1] In the FourCastNet row, RMSE is replaced by '—' for lead times 40-60 days, but CSI values are still listed for those lead times; please clarify how CSI was computed when the forecast was deemed unstable.
  5. [Appendix C, Equation (31)] The boundary mask M multiplies only the advection term and not the source term ˙Sθ; please state explicitly whether the source network is applied everywhere or only in ocean cells, and why.
  6. [Table 8] The table reports inference times for Ocean-E2E 'w/ Dynamical Kernel' and 'w/o Dynamical Kernel', but the 'w/o' variant is not otherwise defined; please specify what is removed in that variant.
  7. [Figure 1] The caption mentions 'The flame icon indicates that the model parameters are trainable', but no flame icon is visible in the figure; please adjust either the caption or the figure.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the physics-AI derivation and forecast evaluation are not equivalent to the training inputs by construction, though the S2S comparison uses a same-reanalysis reference.

full rationale

The forecast model's learned components uθ and ˙Sθ are optimized with the MSE loss L = ||Ĉt − ∫(...)dt||² (Eq. 19) against GLORYS12V1 fields, and the headline comparison in Table 2 scores forecasts against 'GLORYS12V1 reanalysis as ground truth.' This is a standard held-out supervised evaluation, not a circular reduction: the test years (2020–2021) are disjoint from training (1993–2018), the test RMSE is not algebraically equal to the training loss, and all AI baselines were 'trained under identical experimental configurations' (Appendix E). The same-reanalysis reference is a real external-validity limitation—GLORYS is a model-data blend, so matching it is not identical to matching the real ocean—but it does not make the forecast equal to the training target by construction. Appendix G partially addresses this concern by taking S2S analysis fields as both initialization and ground truth and again finding lower RMSE for Ocean-E2E; those S2S fields were not used in training. The dynamical-core equations (Eqs. 7–20) are standard advection and source/sink decompositions with learned subgrid closures; no equation is defined in terms of the headline metric. Self-citations (Shu et al. 2025 for SSTA construction; Gao et al. 2025 for OneForecast) are used as data-preprocessing or pretrained-component references, not as the justification of the SOTA claim. The abstract's '10%' versus Table 2's lead-time-specific promotions is an internal-consistency issue, not evidence of circularity. A genuine omission is that the assimilation network φaθ's training objective is not stated in the main text or Appendix D; this affects reproducibility and validity, but it is not a circularity because the forecast claim does not reduce to that network's training target.

Assumptions & free parameters 9 free parameters · 7 assumptions · 0 invented entities

The central claims rest on the choice of GLORYS12V1 as both training target and evaluation truth, on the adequacy of the learned bolus-velocity and source-term parameterizations, and on the accuracy of two pretrained networks (Mθ for ocean currents, Nθ/OneForecast for atmosphere). The paper does not quantify the error introduced by these pretrained components and does not show sensitivity to time step, ensemble size, or the scaling factor εgm.

free parameters (9)
  • uθ network parameters (learned bolus velocity) = not reported
    Neural network approximating subgrid advective transport; trained with MSE on GLORYS SSTA and satellite-derived currents (Eq. 11).
  • Sθ network parameters (learned source/mixing term) = not reported
    Neural network for air-sea heat flux and mixing effects; trained with MSE on GLORYS and ERA5 forcing (Eq. 13).
  • Mθ network parameters (surface geostrophic current forecaster) = not reported
    Pretrained autoregressive model for ug; frozen during final training (Eq. 15); no accuracy or error analysis is provided.
  • Nθ parameters (OneForecast atmospheric model) = not reported
    Pretrained AI weather model providing atmospheric boundary conditions (Eq. 16); not released by the paper.
  • εgm scaling factor = 0.1
    Hand-chosen scaling on the learned bolus velocity to stabilize training (Appendix C, Eq. 31).
  • Global time step Δt (global model) = 24 hours
    Chosen for the 1/2° global simulations; no CFL analysis or sensitivity test is shown (Appendix C).
  • Regional time step and neural update interval = 1800 s; NN applied every 48 steps
    Chosen for 1/12° western Atlantic to balance stability and cost (Appendix C).
  • Perlin noise ensemble size N = 10
    Used to generate background ensemble in assimilation (Algorithm 1); no sensitivity study is reported.
  • Normalization statistics reference period = 1991-2018 (stated in Appendix B; main text says 1993-2018)
    Statistical moments for standardizing inputs; the discrepancy in the stated reference period is unresolved.
assumptions (7)
  • domain assumption GLORYS12V1 reanalysis faithfully represents true SSTA, including extreme MHW magnitudes.
    Serves as ground truth in Tables 1-2 and Figure 4; no verification against independent observations for extremes is provided.
  • domain assumption MHW climatology and seasonal cycle are computed without using 2020-2021 test information.
    Appendix E applies the Hobday 90th-percentile definition without stating the reference period; leakage would inflate CSI.
  • domain assumption Subgrid advective transport can be represented by a learned bolus velocity depending only on surface ug and C.
    Eq. (11) replaces the GM90 bolus velocity, which normally requires subsurface density data, with a surface-only neural network.
  • domain assumption Vertical advection w∂C/∂z is negligible for SSTA evolution.
    Dropped in Eq. (7) because surface vertical velocity is small, but mixed-layer vertical exchange can still affect SSTA.
  • ad hoc to paper Surface geostrophic velocity evolves autoregressively as ∂ug/∂t = Mθ(ug).
    Eq. (15) is asserted after the wave equation Eq. (14); the step to a learned autoregressive operator is not derived.
  • domain assumption Forward Euler integration at Δt=24h (global) is stable because the learned bolus term damps large gradients.
    The ablation shows pure numerical advection collapses; stability depends on learned damping, not on a numerical analysis guarantee.
  • domain assumption Nθ (OneForecast) provides sufficiently accurate atmospheric forcing over 40-day leads for ocean SSTA forecasting.
    Atmospheric boundary condition in Eq. (16); the paper does not evaluate Nθ's forecast error contribution to SSTA error.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Ocean-E2E: Hybrid Physics-Based and Data-Driven Global Forecasting of Extreme Marine Heatwaves with End-to-End Neural Assimilation." pith.science (2026). https://pith.science/paper/KL6ISMY3

@misc{pith2026250522071,
  author       = {Pith},
  title        = {Pith review of: Ocean-E2E: Hybrid Physics-Based and Data-Driven Global Forecasting of Extreme Marine Heatwaves with End-to-End Neural Assimilation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KL6ISMY3}},
  note         = {Machine review of arXiv:2505.22071}
}
read the original abstract

This work focuses on the end-to-end forecast of global extreme marine heatwaves (MHWs), which are unusually warm sea surface temperature events with profound impacts on marine ecosystems. Accurate prediction of extreme MHWs has significant scientific and financial worth. However, existing methods still have certain limitations in forecasting general patterns and extreme events. In this study, to address these issues, based on the physical nature of MHWs, we created a novel hybrid data-driven and numerical MHWs forecast framework Ocean-E2E, which is capable of 40-day accurate MHW forecasting with end-to-end data assimilation. Our framework significantly improves the forecast ability of MHWs by explicitly modeling the effect of oceanic mesoscale advection and air-sea interaction based on a dynamic kernel. Furthermore, Ocean-E2E is capable of end-to-end MHWs forecast and regional high-resolution prediction, allowing our framework to operate completely independently of numerical models while outperforming the current state-of-the-art ocean numerical/AI forecasting-assimilation models. Experimental results show that the proposed framework performs excellently on global-to-regional scales and short-to-long-term forecasts, especially in those most extreme MHWs. Overall, our model provides a framework for forecasting and understanding MHWs and other climate extremes. Our codes are available at https://github.com/ChiyodaMomo01/Ocean-E2E.

Figures

Figures reproduced from arXiv: 2505.22071 by the authors.

Figure 1
Figure 1. Our proposed Ocean-E2E framwork. a) Overview of our framework. b) Details of our hybrid physics and data-driven [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Governing Mechanisms of MHWs The second equal sign is due to the fact that, accord￾ing to equation (5), w|z=0 is small (10−6 ∼ 10−5 m/s) compared to other terms. Equation (7) implies two key governing mechanisms of MHWs: 1) advective transport by geostrophic currents that redistribute thermal energy, 2) mixing and external sink/source encompassing convective mixing and external ageostrophic forcing through surface b… view at source ↗
Figure 3
Figure 3. Snapshots of our framework and other baseline [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (12 more)
Figure 4
Figure 4. Figure 4: Results of our end-to-end assimilation framework. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Results of our regional high resolution MHWs [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Core components of the GABlock. (Top) Core computational flow of the Multilayer Perceptron (Mlp) mod￾ule, including a 1x1 convolution, a depthwise convolution, GELU activation, and Dropout. (Bottom) Core computa￾tional flow of the Spatial Attention (SA) module, illustr…
Figure 7
Figure 7. Figure 7: Global Simulation of MHWs Initialized on January 13, 2020 [PITH_FULL_IMAGE:figures/full_fig_p019_7.png]
Figure 8
Figure 8. Figure 8: Global Simulation of MHWs Initialized on May 12, 2020 [PITH_FULL_IMAGE:figures/full_fig_p020_8.png]
Figure 9
Figure 9. Figure 9: Global Simulation of MHWs Initialized on July 11, 2020 [PITH_FULL_IMAGE:figures/full_fig_p021_9.png]
Figure 10
Figure 10. Figure 10: Global Forecast of MHWs Initialized on January 13, 2020 [PITH_FULL_IMAGE:figures/full_fig_p022_10.png]
Figure 11
Figure 11. Figure 11: Global Forecast of MHWs Initialized on May 12, 2020 [PITH_FULL_IMAGE:figures/full_fig_p023_11.png]
Figure 12
Figure 12. Figure 12: Global Forecast of MHWs Initialized on July 11, 2020 [PITH_FULL_IMAGE:figures/full_fig_p024_12.png]
Figure 13
Figure 13. Figure 13: Regional Simulation of West Atlantic Initialized on January 13, 2020 [PITH_FULL_IMAGE:figures/full_fig_p025_13.png]
Figure 14
Figure 14. Figure 14: Regional Simulation of West Atlantic Initialized on March 13, 2020 [PITH_FULL_IMAGE:figures/full_fig_p026_14.png]
Figure 15
Figure 15. Figure 15: Regional Simulation of West Atlantic Initialized on May 12, 2020 [PITH_FULL_IMAGE:figures/full_fig_p027_15.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Frequency-Aligned Knowledge Distillation for Lightweight Spatiotemporal Forecasting

    cs.LG 2025-06 reject novelty 4.0 of 10

    The proposed spectral distillation method is not actually evaluated in the experiments, and the abstract's headline error reductions contradict the reported tables.

Reference graph

Works this paper leans on

70 extracted references · 49 canonical work pages · cited by 1 Pith paper

  1. [1]

    Arcucci, R.; Zhu, J.; Hu, S.; and Guo, Y.-K. 2021. Deep data assimilation: integrating deep learning with data assimilation. Applied Sciences, 11(3): 1114

    work page 2021

  2. [2]

    A.; Smith, G

    Benthuysen, J. A.; Smith, G. A.; Spillman, C. M.; and Steinberg, C. R. 2021. Subseasonal prediction of the 2020 Great Barrier Reef and Coral Sea marine heatwave. Environmental Research Letters, 16(12): 124050

    work page 2021

  3. [3]

    Bi, K.; Xie, L.; Zhang, H.; Chen, X.; Gu, X.; and Tian, Q. 2022. Pangu-weather: A 3d high-resolution model for fast and accurate global weather forecast. arXiv preprint arXiv:2211.02556

    arXiv 2022

  4. [4]

    Bi, K.; Xie, L.; Zhang, H.; Chen, X.; Gu, X.; and Tian, Q. 2023. Accurate medium-range global weather forecasting with 3D neural networks. Nature, 619(7970): 533--538

    2023

  5. [5]

    Boudier, P.; Fillion, A.; Gratton, S.; G \"u rol, S.; and Zhang, S. 2023. Data assimilation networks. Journal of Advances in Modeling Earth Systems, 15(4): e2022MS003353

    work page 2023

  6. [6]

    Brajard, J.; Carrassi, A.; Bocquet, M.; and Bertino, L. 2020. Combining data assimilation and machine learning to emulate a dynamical model from sparse and noisy observations: A case study with the Lorenz 96 model. Journal of computational science, 44: 101171

    work page 2020

  7. [7]

    J.; Hazen, E

    Brodie, S.; Pozo Buil, M.; Welch, H.; Bograd, S. J.; Hazen, E. L.; Santora, J. A.; Seary, R.; Schroeder, I. D.; and Jacox, M. G. 2023. Ecological forecasts for marine resource management during climate extremes. Nature Communications, 14(1): 7701

    work page 2023

  8. [8]

    Chen, K.; Bai, L.; Ling, F.; Ye, P.; Chen, T.; Luo, J.-J.; Chen, H.; Xiao, Y.; Chen, K.; Han, T.; et al. 2023 a . Towards an end-to-end artificial intelligence driven global weather forecasting system. arXiv preprint arXiv:2312.12462

    arXiv 2023

Show all 70 references
  1. [9]

    Chen, K.; Han, T.; Gong, J.; Bai, L.; Ling, F.; Luo, J.-J.; Chen, X.; Ma, L.; Zhang, T.; Su, R.; et al. 2023 b . Fengwu: Pushing the skillful global medium-range weather forecast beyond 10 days lead. arXiv preprint arXiv:2304.02948

    2023

  2. [10]

    Chen, Y.; Jin, Y.; Liu, Z.; Shen, X.; Chen, X.; Lin, X.; Zhang, R.-H.; Luo, J.-J.; Zhang, W.; Duan, W.; et al. 2025. Combined dynamical-deep learning ENSO forecasts. Nature Communications, 16(1): 3845

    2025

  3. [11]

    Cui, Y.; Wu, R.; Zhang, X.; Zhu, Z.; Liu, B.; Shi, J.; Chen, J.; Liu, H.; Zhou, S.; Su, L.; et al. 2025. Forecasting the eddying ocean with a deep neural network. Nature Communications, 16(1): 2268

    2025

  4. [12]

    Delaunay, A.; and Christensen, H. M. 2022. Interpretable deep learning for probabilistic MJO prediction. Geophysical Research Letters, 49(16): e2022GL098566

    2022

  5. [13]

    Fablet, R.; Chapron, B.; Drumetz, L.; M \'e min, E.; Pannekoucke, O.; and Rousseau, F. 2021. Learning variational data assimilation models and solvers. Journal of Advances in Modeling Earth Systems, 13(10): e2021MS002572

    2021

  6. [14]

    S.; Durand, C.; Farchi, A.; Bocquet, M.; Rampal, P.; and Carrassi, A

    Finn, T. S.; Durand, C.; Farchi, A.; Bocquet, M.; Rampal, P.; and Carrassi, A. 2024. Generative diffusion for regional surrogate models from sea-ice simulations. Journal of Advances in Modeling Earth Systems, 16(10): e2024MS004395

    2024

  7. [15]

    Gao, Y.; Wu, H.; Shu, R.; Dong, H.; Xu, F.; Chen, R.; Yan, Y.; Wen, Q.; Hu, X.; Wang, K.; et al. 2025. OneForecast: A Universal Framework for Global and Regional Weather Forecasting. arXiv preprint arXiv:2502.00338

    2025

  8. [16]

    C.; Kersting, D

    Garrabou, J.; Coma, R.; Bensoussan, N.; Bally, M.; Chevaldonn \'e , P.; Cigliano, M.; D \' az, D.; Harmelin, J.-G.; Gambi, M. C.; Kersting, D. K.; et al. 2009. Mass mortality in Northwestern Mediterranean rocky benthic communities: effects of the 2003 heat wave. Global change ...

    2009

  9. [17]

    R.; and Mcwilliams, J

    Gent, P. R.; and Mcwilliams, J. C. 1990. Isopycnal mixing in ocean circulation models. Journal of Physical Oceanography, 20(1): 150--155

    1990

  10. [18]

    Giamalaki, K.; Beaulieu, C.; and Prochaska, J. 2022. Assessing predictability of marine heatwaves with random forests. Geophysical Research Letters, 49(23): e2022GL099069

    2022

  11. [19]

    Ham, Y.-G.; Kim, J.-H.; and Luo, J.-J. 2019. Deep learning for multi-year ENSO forecasts. Nature, 573(7775): 568--572

    2019

  12. [20]

    Hao, R.; Zhao, Y.; Zhang, S.; and Deng, X. 2025. Deep Learning for Ocean Forecasting: A Comprehensive Review of Methods, Applications, and Datasets. IEEE Transactions on Cybernetics

    2025

  13. [21]

    Hatfield, S.; Chantry, M.; Dueben, P.; Lopez, P.; Geer, A.; and Palmer, T. 2021. Building tangent-linear and adjoint models for data assimilation with neural networks. Journal of Advances in Modeling Earth Systems, 13(9): e2021MS002521

    2021

  14. [22]

    He, K.; Zhang, X.; Ren, S.; and Sun, J. 2016. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, 770--778

    2016

  15. [23]

    Hersbach, H.; Bell, B.; Berrisford, P.; Hirahara, S.; Hor \'a nyi, A.; Mu \ n oz-Sabater, J.; Nicolas, J.; Peubey, C.; Radu, R.; Schepers, D.; et al. 2020. The ERA5 global reanalysis. Quarterly Journal of the Royal Meteorological Society, 146(730): 1999--2049

    2020

  16. [24]

    J.; Spillman, C

    Hobday, A. J.; Spillman, C. M.; Paige Eveson, J.; and Hartog, J. R. 2016. Seasonal forecasting for decision support in marine fisheries and aquaculture. Fisheries Oceanography, 25: 45--56

    2016

  17. [25]

    J.; Scannell, H

    Holbrook, N. J.; Scannell, H. A.; Sen Gupta, A.; Benthuysen, J. A.; Feng, M.; Oliver, E. C.; Alexander, L. V.; Burrows, M. T.; Donat, M. G.; Hobday, A. J.; et al. 2019. A global assessment of marine heatwaves and their drivers. Nature communications, 10(1): 2624

    2019

  18. [26]

    D.; and Hoefler, T

    Huang, L.; Gianinazzi, L.; Yu, Y.; Dueben, P. D.; and Hoefler, T. 2024. Diffda: a diffusion model for weather-scale data assimilation. arXiv preprint arXiv:2401.05932

    2024 arXiv

  19. [27]

    P.; Anderson, K

    Hughes, T. P.; Anderson, K. D.; Connolly, S. R.; Heron, S. F.; Kerry, J. T.; Lough, J. M.; Baird, A. H.; Baum, J. K.; Berumen, M. L.; Bridge, T. C.; et al. 2018. Spatial and temporal patterns of mass bleaching of corals in the Anthropocene. Science, 359(6371): 80--83

    2018

  20. [28]

    P.; Kerry, J

    Hughes, T. P.; Kerry, J. T.; \'A lvarez-Noriega, M.; \'A lvarez-Romero, J. G.; Anderson, K. D.; Baird, A. H.; Babcock, R. C.; Beger, M.; Bellwood, D. R.; Berkelmans, R.; et al. 2017. Global warming and recurrent mass bleaching of corals. Nature, 543(7645): 373--377

    2017

  21. [29]

    G.; Alexander, M

    Jacox, M. G.; Alexander, M. A.; Amaya, D.; Becker, E.; Bograd, S. J.; Brodie, S.; Hazen, E. L.; Pozo Buil, M.; and Tommasi, D. 2022. Global seasonal forecasts of marine heatwaves. Nature, 604(7906): 486--490

    2022

  22. [30]

    Kim, H.; Ham, Y.; Joo, Y.; and Son, S. 2021. Deep learning for bias correction of MJO prediction. Nature Communications, 12(1): 3087

    2021

  23. [31]

    Kurth, T.; Subramanian, S.; Harrington, P.; Pathak, J.; Mardani, M.; Hall, D.; Miele, A.; Kashinath, K.; and Anandkumar, A. 2023. Fourcastnet: Accelerating global high-resolution weather forecasting using adaptive fourier neural operators. In Proceedings of the platform for ad...

    2023

  24. [32]

    Legler, S.; and Janji \'c , T. 2022. Combining data assimilation and machine learning to estimate parameters of a convective-scale model. Quarterly Journal of the Royal Meteorological Society, 148(743): 860--874

    2022

  25. [33]

    Lellouche, J.-M.; Le Galloudec, O.; Greiner, E.; Garric, G.; Regnier, C.; Drevillon, M.; Bourdall \'e -Badie, R.; Bricaud, C.; Drillet, Y.; and Le Traon, P.-Y. 2018. The Copernicus Marine Environment Monitoring Service global ocean 1/12 physical reanalysis GLORYS12V1: descript...

    2018

  26. [34]

    B.; Azizzadenesheli, K.; liu, B.; Bhattacharya, K.; Stuart, A.; and Anandkumar, A

    Li, Z.; Kovachki, N. B.; Azizzadenesheli, K.; liu, B.; Bhattacharya, K.; Stuart, A.; and Anandkumar, A. 2021. Fourier Neural Operator for Parametric Partial Differential Equations. In ICLR

    2021

  27. [35]

    S.; Ide, K.; and Boukabara, S

    Liang, X.; Garrett, K.; Liu, Q.; Maddy, E. S.; Ide, K.; and Boukabara, S. 2022. A deep-learning-based microwave radiative transfer emulator for data assimilation and remote sensing. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 15: 8819--8833

    2022

  28. [36]

    Lin, Z.; Zhang, S.; Zhang, Z.; Yu, X.; and Gao, Y. 2023. The Rossby Normal Mode as a Physical Linkage in a Machine Learning Forecast Model for the SST and SSH of South China Sea Deep Basin. Journal of Geophysical Research: Oceans, 128(9): e2023JC019851

    2023

  29. [37]

    Liu, Q.; and Ma, J. 2024. Foundation Models for Geophysics: Review and Perspective. arXiv preprint arXiv:2406.03163

    2024 arXiv

  30. [38]

    W.; Lu, Y.; Gao, Y.; Guo, Y.; Ren, W.; Zhu, F.; and Wang, F.-Y

    Liu, R. W.; Lu, Y.; Gao, Y.; Guo, Y.; Ren, W.; Zhu, F.; and Wang, F.-Y. 2024. Real-Time Multi-Scene Visibility Enhancement for Promoting Navigational Safety of Vessels Under Complex Weather Conditions. IEEE Transactions on Intelligent Transportation Systems

    2024

  31. [39]

    Liu, Y.; Zheng, Z.; Cheng, J.; Tsung, F.; Zhao, D.; Rong, Y.; and Li, J. 2025. Cirt: Global subseasonal-to-seasonal forecasting with geometry-inspired transformer. arXiv preprint arXiv:2502.19750

    2025 arXiv

  32. [40]

    J.; Siedlecki, S

    Malick, M. J.; Siedlecki, S. A.; Norton, E. L.; Kaplan, I. C.; Haltuch, M. A.; Hunsicker, M. E.; Parker-Stetter, S. L.; Marshall, K. N.; Berger, A. M.; Hermann, A. J.; et al. 2020. Environmentally driven seasonal forecasts of Pacific hake distribution. Frontiers in Marine Scie...

    2020

  33. [41]

    Melinc, B.; and Zaplotnik, Z . 2024. 3D-Var data assimilation using a variational autoencoder. Quarterly Journal of the Royal Meteorological Society, 150(761): 2273--2295

    2024

  34. [42]

    E.; Pershing, A

    Mills, K. E.; Pershing, A. J.; and Hern \'a ndez, C. M. 2017. Forecasting the seasonal timing of Maine's lobster fishery. Frontiers in Marine Science, 4: 337

    2017

  35. [43]

    C.; Benthuysen, J

    Oliver, E. C.; Benthuysen, J. A.; Darmaraki, S.; Donat, M. G.; Hobday, A. J.; Holbrook, N. J.; Schlegel, R. W.; and Sen Gupta, A. 2021. Marine heatwaves. Annual review of marine science, 13(1): 313--342

    2021

  36. [44]

    Pathak, J.; Subramanian, S.; Harrington, P.; Raja, S.; Chattopadhyay, A.; Mardani, M.; Kurth, T.; Hall, D.; Li, Z.; Azizzadenesheli, K.; et al. 2022. Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators. arXiv preprint arXiv:2...

    2022 arXiv

  37. [45]

    R.; Danabasoglu, G.; Keenlyside, N.; Matei, D.; Miesner, A

    Payne, M. R.; Danabasoglu, G.; Keenlyside, N.; Matei, D.; Miesner, A. K.; Yang, S.; and Yeager, S. G. 2022. Skilful decadal-scale prediction of fish habitat and distribution shifts. Nature Communications, 13(1): 2660

    2022

  38. [46]

    marine heat wave

    Pearce, A. F.; Lenanton, R.; Jackson, G.; Moore, J.; Feng, M.; and Gaughan, D. 2011. The "marine heat wave" off Western Australia during the summer of 2010/11. Western Australian Fisheries and Marine Research Laboratories Hillarys, WA …

    2011

  39. [47]

    Peebles, W.; and Xie, S. 2023. Scalable diffusion models with transformers. In Proceedings of the IEEE/CVF international conference on computer vision, 4195--4205

    2023

  40. [48]

    Pujol, M.-I.; Faug \`e re, Y.; Taburet, G.; Dupuy, S.; Pelloquin, C.; Ablain, M.; and Picot, N. 2016. DUACS DT2014: the new multi-mission altimeter data set reprocessed over 20 years. Ocean Science, 12(5): 1067--1090

    2016

  41. [49]

    Raonic, B.; Molinaro, R.; De Ryck, T.; Rohner, T.; Bartolucci, F.; Alaifari, R.; Mishra, S.; and de B \'e zenac, E. 2023. Convolutional neural operators for robust and accurate learning of PDEs. Advances in Neural Information Processing Systems, 36: 77187--77200

    2023

  42. [50]

    Rasp, S.; Hoyer, S.; Merose, A.; Langmore, I.; Battaglia, P.; Russell, T.; Sanchez-Gonzalez, A.; Yang, V.; Carver, R.; Agrawal, S.; et al. 2024. Weatherbench 2: A benchmark for the next generation of data-driven global weather models. Journal of Advances in Modeling Earth Syst...

    2024

  43. [51]

    Ronneberger, O.; Fischer, P.; and Brox, T. 2015. U-net: Convolutional networks for biomedical image segmentation. In Medical image computing and computer-assisted intervention--MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part II...

    2015

  44. [52]

    Shi, B.; Feng, L.; He, H.; Hao, Y.; Peng, Y.; Liu, M.; Liu, Y.; and Liu, J. 2024. A physics-guided attention-based neural network for sea surface temperature prediction. IEEE Transactions on Geoscience and Remote Sensing

    2024

  45. [53]

    Shi, X.; Chen, Z.; Wang, H.; Yeung, D.-Y.; Wong, W.-K.; and Woo, W.-c. 2015. Convolutional LSTM network: A machine learning approach for precipitation nowcasting. Advances in neural information processing systems, 28

    2015

  46. [54]

    Shi, Z.; Zheng, H.; and Dong, J. 2024. OceanVP: A HYCOM based benchmark dataset and a relational spatiotemporal predictive network for oceanic variable prediction. Ocean Engineering, 304: 117748

    2024

  47. [55]

    Shin, N.-Y.; Kim, D.; Kang, D.; Kim, H.; and Kug, J.-S. 2024. Deep learning reveals moisture as the primary predictability source of MJO. npj Climate and Atmospheric Science, 7(1): 11

    2024

  48. [56]

    Shu, R.; Wu, H.; Gao, Y.; Xu, F.; Gou, R.; Xiong, W.; and Huang, X. 2025. Advanced forecasts of global extreme marine heatwaves through a physics-guided data-driven approach. Environmental Research Letters, 20(4): 044030

    2025

  49. [57]

    G.; Johnson, B.; Moradi, I.; Karpowicz, B.; and McCarty, W

    Stegmann, P. G.; Johnson, B.; Moradi, I.; Karpowicz, B.; and McCarty, W. 2022. A deep learning approach to fast radiative transfer. Journal of Quantitative Spectroscopy and Radiative Transfer, 280: 108088

    2022

  50. [58]

    Sun, W.; Zhou, S.; Yang, J.; Gao, X.; Ji, J.; and Dong, C. 2023. Artificial intelligence forecasting of marine heatwaves in the South China Sea using a combined U-Net and ConvLSTM system. Remote Sensing, 15(16): 4068

    2023

  51. [59]

    Tan, C.; Gao, Z.; Li, S.; and Li, S. Z. 2022. Simvp: Towards simple yet powerful spatiotemporal predictive learning. arXiv preprint arXiv:2211.12509

    2022 arXiv

  52. [60]

    A.; Burkholder, D

    Thomson, J. A.; Burkholder, D. A.; Heithaus, M. R.; Fourqurean, J. W.; Fraser, M. W.; Statton, J.; and Kendrick, G. A. 2015. Extreme temperatures, foundation species, and abrupt ecosystem change: an example from an iconic seagrass ecosystem. Global change biology, 21(4): 1463--1474

    2015

  53. [61]

    Verma, Y.; Heinonen, M.; and Garg, V. 2024. ClimODE: Climate and weather forecasting with physics-informed neural ODEs. arXiv preprint arXiv:2404.10024

    2024 arXiv

  54. [62]

    Wang, X.; Wang, R.; Hu, N.; Wang, P.; Huo, P.; Wang, G.; Wang, H.; Wang, S.; Zhu, J.; Xu, J.; et al. 2024. Xihe: A data-driven model for global ocean eddy-resolving forecasting. arXiv preprint arXiv:2402.02995

    2024 arXiv

  55. [63]

    Wu, H.; Hu, T.; Luo, H.; Wang, J.; and Long, M. 2023. Solving High-Dimensional PDEs with Latent Spectral Models. arXiv preprint arXiv:2301.12664

    2023 arXiv

  56. [64]

    Wu, H.; Weng, K.; Zhou, S.; Huang, X.; and Xiong, W. 2024 a . Neural Manifold Operators for Learning the Evolution of Physical Dynamics. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 3356--3366

    2024

  57. [65]

    Wu, H.; Xu, F.; Chen, C.; Hua, X.-S.; Luo, X.; and Wang, H. 2024 b . Pastnet: Introducing physical inductive biases for spatio-temporal video prediction. In Proceedings of the 32nd ACM International Conference on Multimedia, 2917--2926

    2024

  58. [66]

    Xiao, Y.; Bai, L.; Xue, W.; Chen, H.; Chen, K.; Han, T.; Ouyang, W.; et al. 2024. Towards a self-contained data-driven global weather forecasting framework. In Forty-first International Conference on Machine Learning

    2024

  59. [67]

    Xiong, W.; Xiang, Y.; Wu, H.; Zhou, S.; Sun, Y.; Ma, M.; and Huang, X. 2023. Ai-goms: Large ai-driven global ocean modeling system. arXiv preprint arXiv:2308.03152

    2023 arXiv

  60. [68]

    Yu, H.; Wang, H.; Yuan, C.; and Xing, Q. 2024. Assessing the predictability of the marine heatwave in the Yellow Sea during the summer of 2018 based on a deterministic forecast model. Weather and Climate Extremes, 44: 100663

    2024

  61. [69]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all...

  62. [70]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.