Pith. sign in

REVIEW 3 major objections 5 minor 50 references

Machine-learning weather models mostly copy the initial-value framing of numerical weather prediction; the better choice is to align the model with either physics structure or data structure.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 09:38 UTC pith:H2VWKBET

load-bearing objection Clean genealogy and operator taxonomy for MLWP; interpretive, not empirical, and the free-form superiority claim is untested. the 3 major comments →

arxiv 2607.05045 v1 pith:H2VWKBET submitted 2026-07-06 physics.ao-ph

On the Genealogy of Machine Learning Weather Prediction

Mohammad Hassan Erfani This is my paper
classification physics.ao-ph
keywords machine-learning weather predictionscientific surrogate modelingfree-form data-driven modelinginitial-value problemstate-conditioned operatorevolution operatorprimitive equationsspatiotemporal sequence prediction
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Modern machine-learning weather prediction has largely taken over the initial-value-problem framing of classical numerical weather prediction. That inheritance produces a dominant style of learned autoregressive time-stepping and quietly steers which architectures get chosen. The paper makes the inheritance explicit and contrasts two traditions: scientific surrogate modeling, which embeds machine learning inside a physical system and must respect its structure, and free-form data-driven modeling, which treats atmospheric fields as spatiotemporal sequences and learns latent dynamics without explicit physical constraints. By reviewing the primitive equations, surveying recent literature, and working through concrete physical examples (friction, precipitation, tracer transport), it maps each tradition onto either a state-conditioned operator or an evolution operator. The practical claim is that model selection becomes principled only when architecture and training objective are deliberately aligned with one structure or the other rather than defaulting to the historical numerical-solver template.

Core claim

The dominant autoregressive, initial-value framing of machine-learning weather prediction is an historical inheritance from numerical weather prediction, not a necessity of the data; the field therefore bifurcates into scientific surrogate modeling (physics-structure-preserving) and free-form data-driven modeling (statistical-structure-driven), and each maps cleanly onto a state-conditioned versus an evolution-operator formulation. Principled selection requires consciously matching architecture and objective to one of those two structures.

What carries the argument

The distinction between state-conditioned operator evaluation (memoryless map Yi = Gθ(Xi) at a single time level) and evolution operator learning (transition or tendency map that advances the state, Xt+Δt = Mθ(Xt) or Xt+Δt = Xt + Δt Fθ(Xt)), which the paper uses to classify both physical processes and machine-learning approaches.

Load-bearing premise

That the two philosophical traditions remain cleanly separable in practice and that free-form spatiotemporal sequence models would systematically differ in stability and fidelity from current learned time-steppers.

What would settle it

A controlled head-to-head experiment that trains both a pure sequence-to-sequence spatiotemporal model and a physics-structure-preserving autoregressive stepper on identical atmospheric data and forecasts, then compares long-horizon stability and physical fidelity side by side.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The manuscript argues that modern machine-learning weather prediction (MLWP) has inherited the initial-value-problem (IVP) framing of numerical weather prediction (NWP), producing a dominant paradigm of learned autoregressive time-stepping. It contrasts two traditions—scientific surrogate modeling (ML embedded inside a physical system and required to respect its structure) versus free-form data-driven modeling (atmospheric fields treated as unconstrained spatiotemporal sequences)—and maps both onto either state-conditioned operator evaluation or evolution-operator learning. The argument is developed by reviewing the primitive equations, surveying recent MLWP literature, and working through three concrete physical examples (shallow-water friction, precipitation as a moisture sink, and atmospheric tracer transport). The conclusion is that principled architecture and objective selection requires explicit alignment with either the physical system structure or the statistical structure of the data.

Significance. If the framing holds, the paper supplies a clarifying taxonomy that the rapidly expanding MLWP community can use to decide when to preserve NWP-style time-marching versus when to treat the problem as generic sequence prediction. Strengths include a clean, equation-level mapping of prognostic versus diagnostic variables (Eqs. 1–6) onto the two operator classes (Eqs. 9–14), three well-chosen physical examples in §5 that illustrate the distinction in practice, and an independent (if informal) probe via chatbot recommendations in Appendix A. The work does not claim new forecast skill; its value is conceptual and genealogical, making the historical constraint explicit and urging deliberate rather than inherited problem formulation.

major comments (3)
  1. The central interpretive claim—that free-form spatiotemporal sequence models unconstrained by physical time-marching would systematically differ in long-term stability and fidelity from current learned NWP-style steppers—is advanced primarily via literature taxonomy (§2) and the chatbot exercise (Appendix A) rather than controlled head-to-head experiments. Because the manuscript presents itself as a clarifying framework rather than an empirical skill paper, this gap does not invalidate the operator taxonomy, but the claim should be explicitly labeled a hypothesis for future work and the language in the Introduction and Conclusion softened accordingly.
  2. §4.2 and the surrounding discussion assert that the two traditions are cleanly separable and that current MLWP architectures “rarely exhibit any difference in how they treat prognostic versus diagnostic fields.” In practice many published systems already hybridize the two (neural operators used as steppers, physics-informed losses inside free-form sequence models, etc.). A short paragraph acknowledging the continuum of hybrids and clarifying where the pure poles remain useful would strengthen the taxonomy without altering the main argument.
  3. The paper repeatedly asks “which approach better surrogates the underlying physical model in terms of stability and fidelity” (Introduction, Conclusion) yet never supplies even a schematic comparison or a proposed evaluation protocol. Given that the operator definitions themselves are sound, the manuscript would be more complete if it either (a) dropped the comparative language or (b) outlined a minimal set of diagnostics (e.g., conservation drift, spectral slope, multi-year free-run statistics) that future work could use to adjudicate the two traditions.
minor comments (5)
  1. Equation (2) contains an obvious typesetting error: the first advection term is written −u ∂v/∂t instead of −u ∂v/∂x. Correcting it will avoid confusing readers who are checking the primitive equations.
  2. In §3.1 the Euler-forward illustration (Eq. 7) is helpful, but a one-sentence reminder that operational NWP uses higher-order or semi-implicit schemes would prevent the impression that the analogy is limited to first-order stepping.
  3. Appendix A Table 1 ranks architectures by chatbot; a brief note on prompt sensitivity or temperature settings would make the informal probe more transparent.
  4. Several references appear with incomplete or slightly inconsistent formatting (e.g., arXiv identifiers mixed with journal citations). A uniform style pass would improve polish.
  5. The distinction between weather (IVP) and climate (BVP) is correctly noted early on, yet later sections sometimes use “MLWP” for both. Clarifying the scope when climate-oriented examples appear would reduce ambiguity.

Circularity Check

0 steps flagged

No significant circularity; conceptual taxonomy of operators and inheritance with no self-referential predictions or fitted claims.

full rationale

The paper is a historical and philosophical review that maps NWP primitive equations (Eqs. 1–6) and the prognostic/diagnostic distinction onto two standard supervised-learning formalisms (state-conditioned operator evaluation Eq. 9 versus evolution/transition operators Eqs. 10–11 and sequence-to-sequence Eq. 14). Concrete examples (SWE friction, precipitation, tracer transport) simply instantiate those definitions; nothing is fitted and then re-labeled a prediction. Self-citations (e.g., the author’s own spatiotemporal tracer-transport papers) appear only as illustrative instances of free-form modeling, not as load-bearing uniqueness theorems or unverified premises. Appendix A’s chatbot probe is an independent informal check, not a circular derivation. The central claim—that MLWP’s dominant autoregressive IVP framing is an inherited constraint rather than a necessity, and that architecture choice should align with either physical structure or data statistics—is therefore interpretive and self-contained against external literature, not forced by construction from its own inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 2 invented entities

Load-bearing content is almost entirely standard NWP mathematics and existing MLWP practice. No free parameters are fitted. The two named ‘traditions’ and the state-conditioned/evolution operator split are classificatory inventions of the paper; they rest on domain assumptions that weather is an IVP governed by the primitive equations and that diagnostic variables lack autonomous dynamics.

axioms (4)
  • domain assumption Atmospheric evolution is governed by the primitive equations (prognostic momentum, temperature, moisture; diagnostic continuity and hydrostatic balance).
    Section 3.1; standard NWP foundation used to define prognostic vs diagnostic variables.
  • domain assumption Weather prediction is classically an initial-value problem whose numerical solution is discrete time-stepping of prognostic fields.
    Introduction and §3 (Bjerknes framing); used to explain why MLWP inherited autoregressive rollouts.
  • domain assumption Diagnostic variables and parameterization outputs are instantaneous functions of the concurrent prognostic state and therefore have no autonomous temporal memory.
    §3.1–3.2 and §5.1–5.2; underpins the claim that they map to state-conditioned operators.
  • ad hoc to paper If atmospheric fields are stripped of physical semantics they become ordinary multichannel spatiotemporal tensors whose natural ML treatment is sequence-to-sequence prediction.
    Introduction and Appendix A; motivates free-form data-driven modeling as a distinct tradition.
invented entities (2)
  • Scientific surrogate modeling vs free-form data-driven modeling (the two traditions) no independent evidence
    purpose: Organize existing MLWP practice into two philosophical camps that map onto different operators.
    Named and contrasted throughout; useful taxonomy but not an independently measured physical object.
  • State-conditioned operator evaluation vs evolution operator learning no independent evidence
    purpose: Formalize the two learning problems that correspond to the two traditions.
    Defined in §4.1 (Eqs. 9–11, 14); restates standard regression vs dynamical-system learning under new labels.

pith-pipeline@v1.1.0-grok45 · 17248 in / 2667 out tokens · 25902 ms · 2026-07-11T09:38:23.896434+00:00 · methodology

0 comments
read the original abstract

Modern machine-learning weather prediction (MLWP) has largely inherited the initial-value-problem (IVP) framing of numerical weather prediction (NWP). This inheritance leads to a dominant paradigm of learned autoregressive time-stepping and constrains how the learning problem is defined and architectures are favored. In this study we make the inheritance explicit, contrast two philosophical traditions: "scientific surrogate modeling," where machine learning (ML) is embedded within a physical system and must respect its structural constraints, and "free-form data-driven modeling," where atmospheric fields are treated as spatiotemporal sequences and models learn latent dynamics without explicit physical constraints. By reviewing the governing primitive equations, surveying recent literature, and analyzing concrete physical examples, we map each modeling paradigm to either a state-conditioned or evolution operator formulation. We conclude that principled model selection requires explicitly aligning architecture and training objectives with either the physical system structure or the statistical structure of the data.

discussion (0)

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

Reference graph

Works this paper leans on

50 extracted references · 8 linked inside Pith

  1. [1]

    Vivit: A video vision transformer

    Anurag Arnab, Mostafa Dehghani, Riccardo Ramos, and Lin Bai. Vivit: A video vision transformer. InInternational Conference on Computer Vision (ICCV), 2021

    2021

  2. [2]

    An empirical evaluation of generic convolu- tional and recurrent networks for sequence modeling.arXiv preprint arXiv:1803.01271, 2018

    Shaojie Bai, J Zico Kolter, and Vladlen Koltun. An empirical evaluation of generic convolu- tional and recurrent networks for sequence modeling.arXiv preprint arXiv:1803.01271, 2018

    Pith/arXiv arXiv 2018

  3. [3]

    A regional high resolution ai weather model for the prediction of atmospheric rivers and extreme precipitation.npj Climate and Atmospheric Science, 8(1):385, 2025

    Jorge Ba˜ no-Medina, Agniv Sengupta, Daniel Steinhoff, Patrick Mulrooney, Thomas Nipen, Mario Santa-Cruz, Yanbo Nie, and Luca Delle Monache. A regional high resolution ai weather model for the prediction of atmospheric rivers and extreme precipitation.npj Climate and Atmospheric Science, 8(1):385, 2025

    2025

  4. [4]

    Atmospheric transport modeling of co2 with neural networks.Journal of Advances in Modeling Earth Systems, 17(2):e2024MS004655, 2025

    Vitus Benson, Ana Bastos, Christian Reimers, Alexander J Winkler, Fanny Yang, and Markus Reichstein. Atmospheric transport modeling of co2 with neural networks.Journal of Advances in Modeling Earth Systems, 17(2):e2024MS004655, 2025

    2025

  5. [5]

    Is space-time attention all you need for video understanding? InInternational Conference on Machine Learning (ICML), 2021

    Gedas Bertasius, Heng Wang, and Lorenzo Torresani. Is space-time attention all you need for video understanding? InInternational Conference on Machine Learning (ICML), 2021

    2021

  6. [6]

    Accurate medium-range global weather forecasting with 3d neural networks.Nature, 619(7970):533– 538, 2023

    Kaifeng Bi, Lingxi Xie, Hengheng Zhang, Xin Chen, Xiaotao Gu, and Qi Tian. Accurate medium-range global weather forecasting with 3d neural networks.Nature, 619(7970):533– 538, 2023

    2023

  7. [7]

    Spherical Fourier neural operators: Learning stable dynamics on the sphere

    Boris Bonev, Thorsten Kurth, Christian Hundt, Jaideep Pathak, Maximilian Baust, Karthik Kashinath, and Anima Anandkumar. Spherical Fourier neural operators: Learning stable dynamics on the sphere. InInternational conference on machine learning, pages 2806–2823. PMLR, 2023

    2023

  8. [8]

    Springer, 2008

    M Hanif Chaudhry.Open-channel flow. Springer, 2008

    2008

  9. [9]

    Fuxi: A cascade machine learning forecasting system for 15-day global weather forecast.npj climate and atmospheric science, 6(1):190, 2023

    Lei Chen, Xiaohui Zhong, Feng Zhang, Yuan Cheng, Yinghui Xu, Yuan Qi, and Hao Li. Fuxi: A cascade machine learning forecasting system for 15-day global weather forecast.npj climate and atmospheric science, 6(1):190, 2023

    2023

  10. [10]

    Empirical evaluation of gated recurrent neural networks on sequence modeling.arXiv preprint arXiv:1412.3555, 2014

    Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, and Yoshua Bengio. Empirical evaluation of gated recurrent neural networks on sequence modeling.arXiv preprint arXiv:1412.3555, 2014

    Pith/arXiv arXiv 2014

  11. [11]

    Impact of model structure and dynamics, version 2

    COMET MetEd. Impact of model structure and dynamics, version 2. Available athttps: //www.meted.ucar.edu/, 2020. Accessed: 2026-06-25. 11

    2020

  12. [12]

    Challenges and design choices for global weather and climate models based on machine learning.Geoscientific Model Development, 11(10):3999–4009, 2018

    Peter D Dueben and Peter Bauer. Challenges and design choices for global weather and climate models based on machine learning.Geoscientific Model Development, 11(10):3999–4009, 2018

    2018

  13. [13]

    Mohammad H Erfani, Kara D Lamb, Susanne E Bauer, Kostas Tsigaridis, Marcus van Lier- Walqui, and Gavin Schmidt. Spatiotemporal machine learning approaches for atmospheric composition emulation in nasa giss modele.Journal of Geophysical Research: Machine Learn- ing and Computation, 3(3):e2025JH001011, 2026

    2026

  14. [14]

    Interactive atmospheric composition emulation for next- generation earth system models.arXiv preprint arXiv:2510.10654, 2025

    Seyed Mohammad Hassan Erfani, Kara Lamb, Susanne Bauer, Kostas Tsigaridis, Marcus van Lier-Walqui, and Gavin Schmidt. Interactive atmospheric composition emulation for next- generation earth system models.arXiv preprint arXiv:2510.10654, 2025

    arXiv 2025

  15. [15]

    Prolongation of smap to spatiotem- porally seamless coverage of continental us using a deep learning neural network.Geophysical Research Letters, 44(21):11–030, 2017

    Kuai Fang, Chaopeng Shen, Daniel Kifer, and Xiao Yang. Prolongation of smap to spatiotem- porally seamless coverage of continental us using a deep learning neural network.Geophysical Research Letters, 44(21):11–030, 2017

    2017

  16. [16]

    Simvp: Simpler yet better video prediction

    Zhangyang Gao, Cheng Tan, Lirong Wu, and Stan Z Li. Simvp: Simpler yet better video prediction. InProceedings of the IEEE/CVF conference on computer vision and pattern recog- nition, pages 3170–3180, 2022

    2022

  17. [17]

    Lucie: A lightweight uncoupled climate emulator with long-term stability and physical consistency

    Haiwen Guan, Troy Arcomano, Ashesh Chattopadhyay, and Romit Maulik. Lucie: A lightweight uncoupled climate emulator with long-term stability and physical consistency. Journal of Advances in Modeling Earth Systems, 17(11):e2025MS005152, 2025

    2025

  18. [18]

    Disentangling physical dynamics from unknown factors for unsupervised video prediction

    Vincent Le Guen and Nicolas Thome. Disentangling physical dynamics from unknown factors for unsupervised video prediction. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 11474–11484, 2020

    2020

  19. [19]

    Long short-term memory.Neural computation, 9 (8):1735–1780, 1997

    Sepp Hochreiter and J¨ urgen Schmidhuber. Long short-term memory.Neural computation, 9 (8):1735–1780, 1997

    1997

  20. [20]

    Cambridge uni- versity press, 2003

    Eugenia Kalnay.Atmospheric modeling, data assimilation and predictability. Cambridge uni- versity press, 2003

    2003

  21. [21]

    Forecasting global weather with graph neural networks.arXiv preprint arXiv:2202.07575, 2022

    Ryan Keisler. Forecasting global weather with graph neural networks.arXiv preprint arXiv:2202.07575, 2022

    Pith/arXiv arXiv 2022

  22. [22]

    Rainfall–runoff modelling using long short-term memory (lstm) networks.Hydrology and Earth System Sciences, 22(11):6005–6022, 2018

    Frederik Kratzert, Daniel Klotz, Claire Brenner, Karsten Schulz, and Mathew Herrnegger. Rainfall–runoff modelling using long short-term memory (lstm) networks.Hydrology and Earth System Sciences, 22(11):6005–6022, 2018

    2018

  23. [23]

    Recent advances in wavelet analyses: Part 2—amazon, parana, orinoco and congo discharges time scale variability.Journal of Hydrology, 314(1-4):289–311, 2005

    David Labat, Josyane Ronchail, and Jean Loup Guyot. Recent advances in wavelet analyses: Part 2—amazon, parana, orinoco and congo discharges time scale variability.Journal of Hydrology, 314(1-4):289–311, 2005

    2005

  24. [24]

    Learning skillful medium-range global weather forecasting.Science, 382(6677):1416–1421, 2023

    Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, Peter Wirnsberger, Meire Fortunato, Ferran Alet, Suman Ravuri, Timo Ewalds, Zach Eaton-Rosen, Weihua Hu, et al. Learning skillful medium-range global weather forecasting.Science, 382(6677):1416–1421, 2023

    2023

  25. [25]

    Aifs–ecmwf’s data-driven forecasting system.arXiv preprint arXiv:2406.01465, 2024

    Simon Lang, Mihai Alexe, Matthew Chantry, Jesper Dramsch, Florian Pinault, Baudouin Raoult, Mariana CA Clare, Christian Lessig, Michael Maier-Gerber, Linus Magnusson, et al. Aifs–ecmwf’s data-driven forecasting system.arXiv preprint arXiv:2406.01465, 2024. 12

    Pith/arXiv arXiv 2024

  26. [26]

    Self-attention convlstm for spatiotemporal prediction

    Zhihui Lin, Maomao Li, Zhuobin Zheng, Yangyang Cheng, and Chun Yuan. Self-attention convlstm for spatiotemporal prediction. InProceedings of the AAAI conference on artificial intelligence, volume 34, pages 11531–11538, 2020

    2020

  27. [27]

    The origins of computer weather prediction and climate modeling.Journal of computational physics, 227(7):3431–3444, 2008

    Peter Lynch. The origins of computer weather prediction and climate modeling.Journal of computational physics, 227(7):3431–3444, 2008

    2008

  28. [28]

    Climax: A foundation model for weather and climate.arXiv preprint arXiv:2301.10343, 2023

    Tung Nguyen, Johannes Brandstetter, Ashish Kapoor, Jayesh K Gupta, and Aditya Grover. Climax: A foundation model for weather and climate.arXiv preprint arXiv:2301.10343, 2023

    Pith/arXiv arXiv 2023

  29. [29]

    Regional data-driven weather modeling with a global stretched grid.Artificial Intelligence for the Earth Systems, 5(2):250001, 2026

    Thomas Nils Nipen, H˚ avard Homleid Haugen, Magnus Sikora Ingstad, Even Marius Nordha- gen, Aram Farhad Shafiq Salihi, Paulina Tedesco, Ivar Ambjørn Seierstad, Jørn Kristiansen, Simon Lang, Mihai Alexe, et al. Regional data-driven weather modeling with a global stretched grid.Artificial Intelligence for the Earth Systems, 5(2):250001, 2026

    2026

  30. [30]

    Fourcastnet: A global data-driven high-resolution weather model using adaptive Fourier neural operators.arXiv preprint arXiv:2202.11214, 2022

    Jaideep Pathak, Shashank Subramanian, Peter Harrington, Sanjeev Raja, Ashesh Chattopad- hyay, Morteza Mardani, Thorsten Kurth, David Hall, Zongyi Li, Kamyar Azizzadenesheli, et al. Fourcastnet: A global data-driven high-resolution weather model using adaptive Fourier neural operators.arXiv preprint arXiv:2202.11214, 2022

    Pith/arXiv arXiv 2022

  31. [31]

    Skilful precip- itation nowcasting using deep generative models of radar.Nature, 597(7878):672–677, 2021

    Suman Ravuri, Karel Lenc, Matthew Willson, Dmitry Kangin, Remi Lam, Piotr Mirowski, Megan Fitzsimons, Maria Athanassiadou, Sheleem Kashem, Sam Madge, et al. Skilful precip- itation nowcasting using deep generative models of radar.Nature, 597(7878):672–677, 2021

    2021

  32. [32]

    Toward data-driven weather and climate forecasting: Approximating a simple general circulation model with deep learning.Geophysical Research Letters, 45(22):12–616, 2018

    Sebastian Scher. Toward data-driven weather and climate forecasting: Approximating a simple general circulation model with deep learning.Geophysical Research Letters, 45(22):12–616, 2018

    2018

  33. [33]

    Convolutional lstm network: A machine learning approach for precipitation nowcasting

    Xingjian Shi, Zhourong Chen, Hao Wang, Dit-Yan Yeung, Wai-Kin Wong, and Wang-chun Woo. Convolutional lstm network: A machine learning approach for precipitation nowcasting. Advances in neural information processing systems, 28, 2015

    2015

  34. [34]

    Deep learning for precipitation nowcasting: A benchmark and a new model

    Xingjian Shi, Zhihan Gao, Leonard Lausen, Hao Wang, Dit-Yan Yeung, Wai-kin Wong, and Wang-chun Woo. Deep learning for precipitation nowcasting: A benchmark and a new model. Advances in neural information processing systems, 30, 2017

    2017

  35. [35]

    Short-term and long-term streamflow forecasting using a wavelet and neuro-fuzzy conjunction model.Journal of Hydrology, 394(3-4):486–493, 2010

    Jalal Shiri and Ozgur Kisi. Short-term and long-term streamflow forecasting using a wavelet and neuro-fuzzy conjunction model.Journal of Hydrology, 394(3-4):486–493, 2010

    2010

  36. [36]

    Openstl: A comprehensive benchmark of spatio-temporal predictive learning

    Cheng Tan, Siyuan Li, Zhangyang Gao, Wenfei Guan, Zedong Wang, Zicheng Liu, Lirong Wu, and Stan Z Li. Openstl: A comprehensive benchmark of spatio-temporal predictive learning. Advances in Neural Information Processing Systems, 36:69819–69831, 2023

    2023

  37. [37]

    Learning spatiotemporal features with 3D convolutional networks

    Du Tran, Lubomir Bourdev, Rob Fergus, Lorenzo Torresani, and Manohar Paluri. Learning spatiotemporal features with 3D convolutional networks. InProceedings of the IEEE interna- tional conference on computer vision, pages 4489–4497, 2015

    2015

  38. [38]

    Wavenet: A generative model for raw audio.arXiv preprint arXiv:1609.03499, 12(1), 2016

    Aaron Van Den Oord, Sander Dieleman, Heiga Zen, Karen Simonyan, Oriol Vinyals, Alex Graves, Nal Kalchbrenner, Andrew Senior, Koray Kavukcuoglu, et al. Wavenet: A generative model for raw audio.arXiv preprint arXiv:1609.03499, 12(1), 2016. 13

    Pith/arXiv arXiv 2016

  39. [39]

    Attention is all you need.Advances in neural information processing systems, 30, 2017

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need.Advances in neural information processing systems, 30, 2017

    2017

  40. [40]

    PredRNN: Recurrent neural networks for predictive learning using spatiotemporal lstms.Advances in neural information processing systems, 30, 2017

    Yunbo Wang, Mingsheng Long, Jianmin Wang, Zhifeng Gao, and Philip S Yu. PredRNN: Recurrent neural networks for predictive learning using spatiotemporal lstms.Advances in neural information processing systems, 30, 2017

    2017

  41. [41]

    PredRNN++: Towards a resolution of the deep-in-time dilemma in spatiotemporal predictive learning

    Yunbo Wang, Zhifeng Gao, Mingsheng Long, Jianmin Wang, and Philip S Yu. PredRNN++: Towards a resolution of the deep-in-time dilemma in spatiotemporal predictive learning. In International conference on machine learning, pages 5123–5132. PMLR, 2018

    2018

  42. [42]

    Memory in memory: A predictive neural network for learning higher-order non-stationarity from spatiotemporal dynamics

    Yunbo Wang, Jianjin Zhang, Hongyu Zhu, Mingsheng Long, Jianmin Wang, and Philip S Yu. Memory in memory: A predictive neural network for learning higher-order non-stationarity from spatiotemporal dynamics. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 9154–9162, 2019

    2019

  43. [43]

    Machine learning for weather and climate are worlds apart.Philo- sophical transactions

    Duncan Watson-Parris. Machine learning for weather and climate are worlds apart.Philo- sophical transactions. Series A, Mathematical, physical, and engineering sciences, 379(2194): 20200098, 2021

    2021

  44. [44]

    Climatebench v1

    Duncan Watson-Parris, Yuhan Rao, Dirk Olivi´ e, Øyvind Seland, Peer Nowack, Gustau Camps- Valls, Philip Stier, Shahine Bouabid, Maura Dewey, Emilie Fons, et al. Climatebench v1. 0: A benchmark for data-driven climate projections.Journal of Advances in Modeling Earth Systems, 14(10):e2021MS002954, 2022

    2022

  45. [45]

    Ace: A fast, skillful learned global atmospheric model for climate prediction.arXiv preprint arXiv:2310.02074, 2023

    Oliver Watt-Meyer, Gideon Dresdner, Jeremy McGibbon, Spencer K Clark, Brian Henn, James Duncan, Noah D Brenowitz, Karthik Kashinath, Michael S Pritchard, Boris Bonev, et al. Ace: A fast, skillful learned global atmospheric model for climate prediction.arXiv preprint arXiv:2310.02074, 2023

    Pith/arXiv arXiv 2023

  46. [46]

    Can machines learn to predict weather? using deep learning to predict gridded 500-hpa geopotential height from historical weather data

    Jonathan A Weyn, Dale R Durran, and Rich Caruana. Can machines learn to predict weather? using deep learning to predict gridded 500-hpa geopotential height from historical weather data. Journal of Advances in Modeling Earth Systems, 11(8):2680–2693, 2019

    2019

  47. [47]

    Improving data-driven global weather prediction using deep convolutional neural networks on a cubed sphere.Journal of Advances in Modeling Earth Systems, 12(9):e2020MS002109, 2020

    Jonathan A Weyn, Dale R Durran, and Rich Caruana. Improving data-driven global weather prediction using deep convolutional neural networks on a cubed sphere.Journal of Advances in Modeling Earth Systems, 12(9):e2020MS002109, 2020

    2020

  48. [48]

    Deep learning for spatiotemporal forecasting in earth system science: a review.International Journal of Digital Earth, 17(1):2391952, 2024

    Manzhu Yu, Qunying Huang, and Zhenlong Li. Deep learning for spatiotemporal forecasting in earth system science: a review.International Journal of Digital Earth, 17(1):2391952, 2024

    2024

  49. [49]

    Efficient and information-preserving future frame prediction and beyond

    Wei Yu, Yichao Lu, Steve Easterbrook, and Sanja Fidler. Efficient and information-preserving future frame prediction and beyond. InInternational Conference on Learning Representations, 2020

    2020

  50. [50]

    CNN encoder + temporal Transformer

    Zhenghao Zhang, Qiang Zhang, and Vijay P Singh. Univariate streamflow forecasting using commonly used data-driven models: literature review and case study.Hydrological Sciences Journal, 63(7):1091–1111, 2018. 14 A ML Architecture Recommendations via AI Chatbot The following prompt was designed and presented to several AI chatbots to elicit model recommen-...

    2018