arxiv: 2605.08935 · v2 · submitted 2026-05-09 · 💻 cs.AI · cs.LG

Recognition: 2 theorem links

· Lean Theorem

PnP-Corrector: A Universal Correction Framework for Coupled Spatiotemporal Forecasting

Hao Wu , Fan Xu , Yuxu Lu , Penghao Zhao , Fan Zhang , Hao Jia , Yuxuan Liang , Ruijian Gou

show 4 more authors

Qingsong Wen Xian Wu Xiaomeng Huang Yuan Gao

Authors on Pith no claims yet

Pith reviewed 2026-05-14 21:17 UTC · model grok-4.3

classification 💻 cs.AI cs.LG

keywords coupled spatiotemporal forecastingerror correctionreciprocal error amplificationplug-and-play frameworkocean-atmosphere modelDSLCastclimate predictionlong-term forecast stability

0 comments

The pith

By freezing pre-trained physics engines and training only a correction agent, the PnP-Corrector framework counters reciprocal error amplification to improve long-term accuracy in coupled spatiotemporal forecasts.

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

Coupled systems such as ocean-atmosphere models suffer from reciprocal error amplification, where small mistakes in one simulator grow through mutual feedback and destroy long-range predictions. The paper introduces PnP-Corrector to separate the physics simulation from error fixing: the original engines stay frozen while a new agent learns to offset the biases that arise only when the engines run together. This plug-and-play design avoids retraining entire simulators. On a 300-day global ocean-atmosphere test, the approach cuts baseline error by 29 percent and beats prior methods on several metrics. The same separation is presented as a general solution for any set of interacting dynamical simulators.

Core claim

The PnP-Corrector framework decouples physical simulation from error correction by freezing pre-trained physics engines and training a dedicated correction agent, supported by the DSLCast backbone, to proactively counteract the systematic biases produced by reciprocal error amplification, thereby extending the stable horizon of coupled forecasts.

What carries the argument

PnP-Corrector, which freezes pre-trained physics simulators and trains only a correction agent to offset interaction biases, using DSLCast as its predictive backbone.

Load-bearing premise

That a correction agent trained separately on frozen simulators can offset the biases created by their mutual error feedback without any retraining or modification of the simulators themselves.

What would settle it

A long-horizon coupled run in which the added correction agent produces no measurable drop in accumulated error or actually increases error relative to the frozen baseline.

Figures

Figures reproduced from arXiv: 2605.08935 by Fan Xu, Fan Zhang, Hao Jia, Hao Wu, Penghao Zhao, Qingsong Wen, Ruijian Gou, Xian Wu, Xiaomeng Huang, Yuan Gao, Yuxuan Liang, Yuxu Lu.

**Figure 1.** Figure 1: The PnP-Corrector framework enables long-term stability in coupled forecasting. This figure compares a 200- day 2-meter temperature (T2M) forecast initialized from the state shown top-left. While the standard DSLCast baseline (bottom-left) accumulates significant errors and drifts from the true state, our PnP-Corrector framework (bottom-right) effectively corrects these systematic biases, producing a forec… view at source ↗

**Figure 2.** Figure 2: Taming the vicious cycle of errors in coupled prediction with our PnP-Corrector framework. (a) Ideal Uncoupled Simulation: A single simulator performs well when driven by perfect external forcing. (b) Coupled Prediction Collapse: In an autoregressive coupled mode, errors from each simulator feed into the other, leading to an exponential error growth (Reciprocal Error Amplification) that ultimately collapse… view at source ↗

**Figure 3.** Figure 3: Overview of the DSLCast Architecture. (Left) Our core innovation is the Differentiable Semi-Lagrangian Advection Block (DSL-Block), which explicitly models the physical advection process. The block first predicts a Flow Field, which defines a Sampling Grid via Backward Tracing from a Base Grid. A differentiable Grid operation then warps the input features. (Center) The architecture is built upon the effici… view at source ↗

**Figure 4.** Figure 4: The latitude-weighted RMSE and MAE results of several important variables. Across these representative variables, PnP-Corrector consistently reduces errors over long lead times, highlighting its improved rollout stability [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗

**Figure 5.** Figure 5: 100-day forecast results of different models using our proposed PnP-Corrector framework. Our method achieves better physical consistency and yields results that are closest to the ground truth. 7 [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗

**Figure 6.** Figure 6: Performance evaluation on extreme event. The bar charts compare the CSI and SEDI for baselines (GraphCast and DSLCast) against their counterparts enhanced by PnP-Corrector. of autoregressive rollout. The spectral decay revealed in our analysis has direct, detrimental qualitative consequences [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗

**Figure 7.** Figure 7: Comprehensive Spectral Fidelity of 300-day Forecasts. Our PnP-Corrector framework restores physically realistic energy spectra across multiple key atmospheric variables (Z500, T850, U10M, T2M, MSLP). The corrected models (solid lines) consistently align with the Ground Truth, demonstrating a universal improvement over the uncorrected baselines (dashed lines). 120°W 60°W 60°E 120°E Initial Condition 120°W 6… view at source ↗

**Figure 8.** Figure 8: Qualitative comparison of a 100-day MSLP forecast over the Southern Hemisphere. The standard GraphCast model suffers from significant smoothing, failing to preserve the structure of the low-pressure system highlighted in the insets. Our PnP-Corrector counteracts this degradation, maintaining a prediction that aligns closely with the ground truth. competitive efficiency in terms of Params and MACs. 5.7. A… view at source ↗

**Figure 9.** Figure 9: In the case study of expanding the PnP-Corrector framework to more spheres, we show 100-day forecast results of different models using our proposed PnP-Corrector framework for land variables. Our method still achieves better physical consistency and yields results that are closest to the ground truth. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗

**Figure 10.** Figure 10: The latitude-weighted RMSE (lower is better) results of several important atmosphere and ocean variables. The corrected models that use PnP-Corrector framework (solid lines) achieve lower RMSE, demonstrating a universal improvement over the uncorrected baselines (dashed lines). 19 [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗

**Figure 11.** Figure 11: The latitude-weighted MAE (lower is better) results of several important atmosphere and ocean variables. The corrected models that use PnP-Corrector framework (solid lines) achieve lower MAE, demonstrating a universal improvement over the uncorrected baselines (dashed lines). 20 [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗

**Figure 12.** Figure 12: 60-day forecasting results of different models. 21 [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗

**Figure 13.** Figure 13: 90-day forecasting results of different models. 22 [PITH_FULL_IMAGE:figures/full_fig_p022_13.png] view at source ↗

**Figure 14.** Figure 14: 120-day forecasting results of different models. 23 [PITH_FULL_IMAGE:figures/full_fig_p023_14.png] view at source ↗

**Figure 15.** Figure 15: 150-day forecasting results of different models. 24 [PITH_FULL_IMAGE:figures/full_fig_p024_15.png] view at source ↗

**Figure 16.** Figure 16: 180-day forecasting results of different models. 25 [PITH_FULL_IMAGE:figures/full_fig_p025_16.png] view at source ↗

**Figure 17.** Figure 17: 210-day forecasting results of different models. 26 [PITH_FULL_IMAGE:figures/full_fig_p026_17.png] view at source ↗

**Figure 18.** Figure 18: 240-day forecasting results of different models. 27 [PITH_FULL_IMAGE:figures/full_fig_p027_18.png] view at source ↗

**Figure 19.** Figure 19: 270-day forecasting results of different models. 28 [PITH_FULL_IMAGE:figures/full_fig_p028_19.png] view at source ↗

**Figure 20.** Figure 20: 300-day forecasting results of different models. 29 [PITH_FULL_IMAGE:figures/full_fig_p029_20.png] view at source ↗

read the original abstract

Coupled spatiotemporal forecasting is important for predicting the future evolution of multiple interacting dynamical systems, such as in climate models. However, existing methods are severely constrained by the persistent bottleneck of compounding errors. In coupled systems, errors from each subsystem simulator propagate and amplify one another, a phenomenon we term Reciprocal Error Amplification, leading to a rapid collapse of long-range predictions. To address this challenge, we propose a universal framework called PnP-Corrector (Plug-and-Play Corrector). The core idea of our framework is to decouple the physical simulation from the error correction process: it freezes pre-trained physics simulation engines and exclusively trains a correction agent to proactively counteract the systematic biases emerging from the coupled system. Furthermore, we design an efficient predictive model architecture, DSLCast, to serve as the backbone of this framework. Extensive experiments demonstrate that our method significantly enhances the long-term stability and accuracy of coupled forecasting systems. For instance, in the challenging task of a 300-day global ocean-atmosphere coupled forecast, our PnP-Corrector framework reduces the prediction error of the baseline model by 29% and surpasses state-of-the-art models on several key metrics.

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.

Desk Editor's Note private letter to a colleague

PnP-Corrector cleanly separates frozen physics engines from a trainable correction agent, but the 29% error drop on 300-day coupled runs rests on thin experimental detail that leaves the reciprocal-amplification claim unproven.

read the letter

The paper's core move is to freeze existing simulators and train only a correction module that sees the coupled outputs. DSLCast is offered as the efficient backbone for that module. This separation is a useful framing for anyone who already has working physics engines and wants to add error control without retraining the whole stack. The 300-day ocean-atmosphere test is a realistic stress case, and the reported 29% error cut plus gains on a few metrics show the idea can move the needle in practice. That part is worth noting for groups doing operational coupled forecasting. The main weakness is the missing experimental controls. The abstract gives no information on whether the correction agent was trained on full bidirectional coupled trajectories or on single-engine rollouts, no error bars, and no clear statement of held-out splits. Without those, it is hard to tell whether the gains come from interrupting the amplification loop or from ordinary bias correction that any post-processor might achieve. The stress-test concern therefore stands: if the training data never let the two engines feed errors back to each other, the method has not been shown to solve the problem it claims to target. For readers already working on climate or environmental coupled models, the paper is worth skimming for the architecture and the test-bed choice. It is not yet strong enough to change practice on its own. A serious editor should send it to review so the authors can supply the missing training details, ablation runs, and statistical reporting; the underlying problem is real and the plug-and-play framing is testable.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces PnP-Corrector, a plug-and-play framework for coupled spatiotemporal forecasting that freezes pre-trained physics engines and trains a separate correction agent (with DSLCast backbone) to counteract reciprocal error amplification without modifying the simulators. It reports a 29% error reduction versus baseline on a 300-day global ocean-atmosphere coupled forecast and claims superiority over state-of-the-art models on key metrics.

Significance. If the empirical gains are shown to arise from correction of coupled amplification loops rather than marginal bias removal, the modular separation of simulation and correction would offer a practical route to longer stable horizons in climate and multi-physics forecasting without retraining expensive engines.

major comments (2)

[Abstract] Abstract: the 29% error reduction on the 300-day coupled forecast is presented without any description of experimental controls, error-bar reporting, data splits, or confirmation that the correction agent was trained and evaluated on trajectories containing reciprocal subsystem feedback; this information is required to distinguish the claimed proactive decoupling from ordinary single-subsystem bias correction.
[Methods] Methods / Experimental Setup: the training protocol for the correction agent on frozen engines must explicitly state whether the input trajectories allow the two physics engines to feed errors back to each other; if training occurs only on decoupled or single-subsystem rollouts, the learned corrections cannot address the amplification loops that emerge exclusively in joint coupled runs and the reported long-horizon gain would be explained by standard bias mitigation.

minor comments (2)

Provide full architectural details and hyper-parameter settings for DSLCast so that the backbone can be reproduced independently.
[Results] All result tables and figures should include error bars, number of runs, and statistical significance tests for the claimed metric improvements.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which help clarify the distinction between our framework and standard bias correction. We have revised the manuscript to explicitly address the experimental details in both the abstract and methods sections.

read point-by-point responses

Referee: [Abstract] Abstract: the 29% error reduction on the 300-day coupled forecast is presented without any description of experimental controls, error-bar reporting, data splits, or confirmation that the correction agent was trained and evaluated on trajectories containing reciprocal subsystem feedback; this information is required to distinguish the claimed proactive decoupling from ordinary single-subsystem bias correction.

Authors: We agree the abstract was too concise on these points. In the revised version we have added a sentence noting that the 29% reduction is measured on fully coupled 300-day rollouts generated from the joint ocean-atmosphere model, using a standard 70/15/15 temporal split, with error bars reported over five independent runs. The correction agent is trained and evaluated exclusively on trajectories that contain reciprocal subsystem feedback, allowing it to target amplification loops rather than isolated bias. revision: yes
Referee: [Methods] Methods / Experimental Setup: the training protocol for the correction agent on frozen engines must explicitly state whether the input trajectories allow the two physics engines to feed errors back to each other; if training occurs only on decoupled or single-subsystem rollouts, the learned corrections cannot address the amplification loops that emerge exclusively in joint coupled runs and the reported long-horizon gain would be explained by standard bias mitigation.

Authors: The training protocol uses fully coupled rollouts in which the ocean and atmosphere engines exchange state variables at every time step, so errors propagate bidirectionally. We have inserted a new paragraph in the Methods section that describes the coupled data-generation procedure, confirms that the correction agent receives the joint state at each step, and contrasts this with single-subsystem ablations we performed to isolate the effect of reciprocal amplification. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical claims rest on experimental results rather than definitional reduction

full rationale

The paper's central contribution is a methodological framework (PnP-Corrector) whose performance is demonstrated through experiments on coupled forecasting tasks, including a reported 29% error reduction on 300-day ocean-atmosphere forecasts. No equations, derivations, or self-referential definitions are present in the provided text that equate the claimed gains to fitted inputs or prior self-citations by construction. The decoupling of simulation from correction is introduced as an architectural choice validated empirically, without load-bearing uniqueness theorems, ansatzes smuggled via citation, or renaming of known results. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract does not enumerate free parameters, axioms, or invented entities; the framework implicitly assumes that systematic biases in coupled runs are learnable from data without altering the physics engines.

pith-pipeline@v0.9.0 · 5545 in / 1094 out tokens · 37988 ms · 2026-05-14T21:17:58.580018+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Lean theorems connected to this paper

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

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

The core idea of our framework is to decouple the physical simulation from the error correction process: it freezes pre-trained physics simulation engines and exclusively trains a correction agent to proactively counteract the systematic biases emerging from the coupled system.
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We term this cross-system, iterative contamination of predictive distributions Reciprocal Error Amplification (REA).

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

300 extracted references · 300 canonical work pages · 8 internal anchors

[1]

Applied Mathematics Letters , volume=

Continuity of ODE solutions , author=. Applied Mathematics Letters , volume=. 1994 , publisher=

work page 1994
[2]

a comprehensive and broader view of instruction learning , author=

Is prompt all you need? no. a comprehensive and broader view of instruction learning , author=. arXiv preprint arXiv:2303.10475 , year=

work page arXiv
[3]

2018 , publisher=

Partial differential equations and complex analysis , author=. 2018 , publisher=

work page 2018
[4]

arXiv preprint arXiv:2505.22071 , year=

Ocean-E2E: Hybrid Physics-Based and Data-Driven Global Forecasting of Extreme Marine Heatwaves with End-to-End Neural Assimilation , author=. arXiv preprint arXiv:2505.22071 , year=

work page arXiv
[5]

Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

Learning to prompt for continual learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

work page
[6]

2001 , publisher=

Partial differential equations in several complex variables , author=. 2001 , publisher=

work page 2001
[7]

GPT-4 Technical Report

Gpt-4 technical report , author=. arXiv preprint arXiv:2303.08774 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[8]

Advances in neural information processing systems , volume=

Dynamic graph neural networks under spatio-temporal distribution shift , author=. Advances in neural information processing systems , volume=

work page
[9]

IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

Graph theory based large-scale machine learning with multi-dimensional constrained optimization approaches for exact epidemiological modelling of pandemic diseases , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , year=

work page
[10]

and Musciotto, F

Battiston, F. and Musciotto, F. and Wang, D. and Barabási, A. L. and Szell, M. and Sinatra, R. , title =. Nat. Rev. Phys. , volume =

work page
[11]

arXiv preprint arXiv:2202.06709 , year=

How do vision transformers work? , author=. arXiv preprint arXiv:2202.06709 , year=

work page arXiv
[12]

Journal of Machine Learning Research , volume=

Neural operator: Learning maps between function spaces with applications to pdes , author=. Journal of Machine Learning Research , volume=

work page
[13]

IEEE Transactions on Multimedia , year=

SimVPv2: Towards simple yet powerful spatiotemporal predictive learning , author=. IEEE Transactions on Multimedia , year=

work page
[14]

DeepONet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators

Deeponet: Learning nonlinear operators for identifying differential equations based on the universal approximation theorem of operators , author=. arXiv preprint arXiv:1910.03193 , year=

work page internal anchor Pith review arXiv 1910
[15]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Masked autoencoders are scalable vision learners , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

work page
[16]

arXiv preprint arXiv:2505.19038 , year=

Turb-L1: Achieving Long-term Turbulence Tracing By Tackling Spectral Bias , author=. arXiv preprint arXiv:2505.19038 , year=

work page arXiv
[17]

Bert: Pre-training of deep bidirectional transformers for language understanding , author=. Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers) , pages=

work page 2019
[18]

arXiv preprint arXiv:2509.17955 , year=

Breaking the Discretization Barrier of Continuous Physics Simulation Learning , author=. arXiv preprint arXiv:2509.17955 , year=

work page arXiv
[19]

Quarterly journal of the royal meteorological society , volume=

The ERA5 global reanalysis , author=. Quarterly journal of the royal meteorological society , volume=. 2020 , publisher=

work page 2020
[20]

Journal of Fluid Mechanics , volume=

The emergence of isolated coherent vortices in turbulent flow , author=. Journal of Fluid Mechanics , volume=. 1984 , publisher=

work page 1984
[21]

arXiv preprint arXiv:2503.13502 , year=

Foundation Models for Spatio-Temporal Data Science: A Tutorial and Survey , author=. arXiv preprint arXiv:2503.13502 , year=

work page arXiv
[22]

npj climate and atmospheric science , volume=

FuXi: A cascade machine learning forecasting system for 15-day global weather forecast , author=. npj climate and atmospheric science , volume=. 2023 , publisher=

work page 2023
[23]

International conference on machine learning , pages=

Spherical fourier neural operators: Learning stable dynamics on the sphere , author=. International conference on machine learning , pages=. 2023 , organization=

work page 2023
[24]

Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

Simvp: Simpler yet better video prediction , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

work page
[25]

Proceedings of the IEEE/CVF international conference on computer vision , pages=

Scalable diffusion models with transformers , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

work page
[26]

Kockum, A. F. and Miranowicz, A. and De Liberato, S. and Savasta, S. and Nori, F. , title =. Nat. Rev. Phys. , volume =

work page
[27]

Communications Earth & Environment , volume=

High-latitude Southern Ocean eddy activity projected to evolve with anthropogenic climate change , author=. Communications Earth & Environment , volume=. 2025 , publisher=

work page 2025
[28]

npj Climate and Atmospheric Science , volume=

The futures of climate modeling , author=. npj Climate and Atmospheric Science , volume=. 2025 , publisher=

work page 2025
[29]

Diffusion generative models for weather forecasting , author=

work page
[30]

Friis, Nicolai and Vitagliano, Giuseppe and Malik, Mehul and Huber, Marcus , title =. Nat. Rev. Phys. , volume =

work page
[31]

International Conference on Machine Learning , year=

Solving High-Dimensional PDEs with Latent Spectral Models , author=. International Conference on Machine Learning , year=

work page
[32]

Convolutional Neural Operators for robust and accurate learning of

Bogdan Raonic and Roberto Molinaro and Tim De Ryck and Tobias Rohner and Francesca Bartolucci and Rima Alaifari and Siddhartha Mishra and Emmanuel de Bezenac , booktitle=. Convolutional Neural Operators for robust and accurate learning of. 2023 , url=

work page 2023
[33]

Sakurai, J. J. , title =

work page
[34]

Juncai He and Xinliang Liu and Jinchao Xu , booktitle=. Mg. 2024 , url=

work page 2024
[35]

Research in the mathematical sciences , volume=

Approximation properties of deep ReLU CNNs , author=. Research in the mathematical sciences , volume=. 2022 , publisher=

work page 2022
[36]

Theoretical and Computational Fluid Dynamics , volume=

Data-driven modeling of the chaotic thermal convection in an annular thermosyphon , author=. Theoretical and Computational Fluid Dynamics , volume=. 2020 , publisher=

work page 2020
[37]

Progress in oceanography , volume=

Global observations of nonlinear mesoscale eddies , author=. Progress in oceanography , volume=. 2011 , publisher=

work page 2011
[38]

arXiv preprint , eprint =

Schroedinger, Erwin , title =. arXiv preprint , eprint =

work page
[39]

Annual Review of Fluid Mechanics , volume=

Submesoscale dynamics in the upper ocean , author=. Annual Review of Fluid Mechanics , volume=. 2023 , publisher=

work page 2023
[40]

Journal of Physical Oceanography , volume=

The submesoscale kinetic energy cascade: Mesoscale absorption of submesoscale mixed layer eddies and frontal downscale fluxes , author=. Journal of Physical Oceanography , volume=

work page
[41]

Science Advances , volume=

The future of Earth system prediction: Advances in model-data fusion , author=. Science Advances , volume=. 2022 , publisher=

work page 2022
[42]

Nature , volume=

Neural general circulation models for weather and climate , author=. Nature , volume=. 2024 , publisher=

work page 2024
[43]

3 and PALM v5

Towards multiscale modeling of ocean surface turbulent mixing using coupled MPAS-Ocean v6. 3 and PALM v5. 0 , author=. Geoscientific Model Development , volume=. 2021 , publisher=

work page 2021
[44]

Deep learning for the Earth sciences: A comprehensive approach to remote sensing, climate science, and geosciences , pages=

Deep learning for the parametrization of subgrid processes in climate models , author=. Deep learning for the Earth sciences: A comprehensive approach to remote sensing, climate science, and geosciences , pages=. 2021 , publisher=

work page 2021
[45]

Journal of Advances in Modeling Earth Systems , volume=

A perspective on climate model hierarchies , author=. Journal of Advances in Modeling Earth Systems , volume=. 2017 , publisher=

work page 2017
[46]

Journal of physical oceanography , volume=

Ocean eddy dynamics in a coupled ocean--atmosphere model , author=. Journal of physical oceanography , volume=

work page
[47]

Journal of Computational Physics , volume=

Multiscale eddy simulation for moist atmospheric convection: Preliminary investigation , author=. Journal of Computational Physics , volume=. 2014 , publisher=

work page 2014
[48]

Science china mathematics , volume=

MgNet: A unified framework of multigrid and convolutional neural network , author=. Science china mathematics , volume=. 2019 , publisher=

work page 2019
[49]

Nonlinear Dynamics , volume=

Coupling between high-frequency modes and a low-frequency mode: Theory and experiment , author=. Nonlinear Dynamics , volume=. 1996 , publisher=

work page 1996
[50]

Mgno: Efficient parameterization of linear operators via multigrid.arXiv preprint arXiv:2310.19809, 2023

MgNO: Efficient parameterization of linear operators via multigrid , author=. arXiv preprint arXiv:2310.19809 , year=

work page arXiv
[51]

Journal of Fluid Mechanics , volume=

Energy cascade and spatial fluxes in wall turbulence , author=. Journal of Fluid Mechanics , volume=. 2004 , publisher=

work page 2004
[52]

Advances in geophysics , volume=

Energy cascade in large-eddy simulations of turbulent fluid flows , author=. Advances in geophysics , volume=. 1975 , publisher=

work page 1975
[53]

arXiv preprint arXiv:2402.16845 , year=

Neural operators with localized integral and differential kernels , author=. arXiv preprint arXiv:2402.16845 , year=

work page arXiv
[54]

Annual review of fluid mechanics , volume=

Shell models of energy cascade in turbulence , author=. Annual review of fluid mechanics , volume=. 2003 , publisher=

work page 2003
[55]

Tellus A: Dynamic Meteorology and Oceanography , volume=

How reliable are climate models? , author=. Tellus A: Dynamic Meteorology and Oceanography , volume=. 2007 , publisher=

work page 2007
[56]

2000 , publisher=

Spectral methods in MATLAB , author=. 2000 , publisher=

work page 2000
[57]

Nature Machine Intelligence , volume=

Physically constrained generative adversarial networks for improving precipitation fields from Earth system models , author=. Nature Machine Intelligence , volume=. 2022 , publisher=

work page 2022
[58]

Science , volume=

Systems integration for global sustainability , author=. Science , volume=. 2015 , publisher=

work page 2015
[59]

2007 , publisher=

Finite difference methods for ordinary and partial differential equations: steady-state and time-dependent problems , author=. 2007 , publisher=

work page 2007
[60]

2019 , publisher=

Computational methods for fluid dynamics , author=. 2019 , publisher=

work page 2019
[61]

Tellus , volume=

The predictability of a flow which possesses many scales of motion , author=. Tellus , volume=. 1969 , publisher=

work page 1969
[62]

2010 , publisher=

Numerical methods for fluid dynamics: With applications to geophysics , author=. 2010 , publisher=

work page 2010
[63]

Measurement Science and Technology , volume=

Turbulent flows , author=. Measurement Science and Technology , volume=

work page
[64]

Journal of physical oceanography , volume=

An estimate of the Lorenz energy cycle for the world ocean based on the STORM/NCEP simulation , author=. Journal of physical oceanography , volume=

work page
[65]

Journal of Computational Physics , volume=

Exact conservation laws for neural network integrators of dynamical systems , author=. Journal of Computational Physics , volume=. 2023 , publisher=

work page 2023
[66]

Nature communications , volume=

Stable machine-learning parameterization of subgrid processes for climate modeling at a range of resolutions , author=. Nature communications , volume=. 2020 , publisher=

work page 2020
[67]

How Do Vision Transformers Work? , author=

work page
[68]

Nature Communications , volume=

Forecasting the eddying ocean with a deep neural network , author=. Nature Communications , volume=. 2025 , publisher=

work page 2025
[69]

2007 , publisher=

Parameterization schemes: keys to understanding numerical weather prediction models , author=. 2007 , publisher=

work page 2007
[70]

2013 , publisher=

Geophysical fluid dynamics , author=. 2013 , publisher=

work page 2013
[71]

Quarterly Journal of the Royal Meteorological Society , volume=

A nonlinear dynamical perspective on model error: A proposal for non-local stochastic-dynamic parametrization in weather and climate prediction models , author=. Quarterly Journal of the Royal Meteorological Society , volume=. 2001 , publisher=

work page 2001
[72]

Universality in Chaos, 2nd edition , pages=

Deterministic Nonperiodic Flow 1 , author=. Universality in Chaos, 2nd edition , pages=. 2017 , publisher=

work page 2017
[73]

Bulletin of the American Meteorological Society , volume=

The coupled model intercomparison project (CMIP) , author=. Bulletin of the American Meteorological Society , volume=. 2000 , publisher=

work page 2000
[74]

Nature , volume=

The quiet revolution of numerical weather prediction , author=. Nature , volume=. 2015 , publisher=

work page 2015
[75]

Geophysical Research Letters , volume=

Earth system modeling 2.0: A blueprint for models that learn from observations and targeted high-resolution simulations , author=. Geophysical Research Letters , volume=. 2017 , publisher=

work page 2017
[76]

Nature , volume=

Deep learning and process understanding for data-driven Earth system science , author=. Nature , volume=. 2019 , publisher=

work page 2019
[77]

Nature Machine Intelligence , volume=

Towards neural Earth system modelling by integrating artificial intelligence in Earth system science , author=. Nature Machine Intelligence , volume=. 2021 , publisher=

work page 2021
[78]

nature , volume=

Deep learning , author=. nature , volume=. 2015 , publisher=

work page 2015
[79]

FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators , author=. arXiv preprint arXiv:2202.11214 , year=

work page internal anchor Pith review Pith/arXiv arXiv
[80]

Science , volume=

Learning skillful medium-range global weather forecasting , author=. Science , volume=. 2023 , publisher=

work page 2023

Showing first 80 references.