Pith. sign in

REVIEW 4 major objections 5 minor 59 references

Physics-Informed Teleconnection-Aware Transformer for Global Subseasonal-to-Seasonal Forecasting

T0 review · 4 major / 5 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 TelePiT claims that a model reading only latitude-averaged atmosphere, evolved by a physics-informed ODE with teleconnection-aware attention, produces the most skillful global 3-to-6-week forecasts yet reported, cutting 2-meter…

desk verdict A clearly written architecture paper whose headline skill claim collapses on the input bottleneck: the model only sees zonal means, so the teleconnection awareness it advertises cannot exist; still worth a referee's time as a cautionary case. read the letter →

arxiv 2506.08049 v3 pith:XFQ4JDRE submitted 2025-06-08 stat.ML cs.AIcs.LG

Tengfei Lyu , Weijia Zhang , Hao Liu This is my paper
classification stat.MLcs.AIcs.LG MSC 68T0786A1062M20
keywords subseasonal-to-seasonalforecastingglobalclimatepredictionphysics-informedneuralODEteleconnection-awareattentionsphericalharmonicembeddingmulti-scalewaveletdecompositiondeeplearningweatherERA5reanalysis
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

This paper claims that subseasonal-to-seasonal forecasts — climate outlooks three to six weeks ahead — can be made more accurately by a model that reads only the average atmospheric state around each latitude circle, discarding all longitude structure. The proposed architecture, TelePiT, feeds these zonal means through a learned multi-band decomposition, evolves each band with a physics-inspired differential equation containing advection, diffusion, forcing, and a neural correction, and applies a transformer whose attention is biased by learned teleconnection patterns. Evaluated on the ERA5 reanalysis under the ChaosBench protocol, it reports lower error than every data-driven baseline and every operational numerical weather prediction system across all tested variables and both forecast horizons, with 2-meter temperature RMSE cut by over 57% relative to the previous best model and spectral divergence roughly an order of magnitude smaller. If these results hold, a lightweight model running in milliseconds would set the standard for global subseasonal-to-seasonal skill.

What carries the argument

The argument rides on three interlocking mechanisms. First, the Spherical Harmonic Embedding reduces each day's 63-variable globe to a sequence of 121 latitude tokens by zonal averaging — formally a projection onto the $m=0$ spherical harmonics — destroying all longitude structure while keeping the latitude profile. Second, a learnable wavelet decomposition splits these tokens into $L+1$ frequency bands, each evolved by a physics-informed neural ODE whose right-hand side is $\gamma\,\tanh\!\big(\boldsymbol{\nu}\odot(\mathbf{x}_{i+1}-2\mathbf{x}_i+\mathbf{x}_{i-1}) + \boldsymbol{\mu}\odot(\mathbf{x}_{i+1}-\mathbf{x}_{i-1})/2 + \mathbf{f} + \alpha\,\mathrm{MLP}(\mathbf{x}_i)\big)$, i.e. explicit diffusion, advection, forcing, and a learned correction. Third, the Teleconnection-Aware Transformer adds a bias to the attention logits: a learned combination of teleconnection pattern vectors $\mathbf{P}_j$ forms a query whose dot products with all latitude keys are added to the standard attention scores, letting one latitude band attend to others according to the active global climate mode. A cross-scale fusion layer then mixes all frequency bands before a shared decoder emits full longitude-resolved fields for both target windows.

What would settle it

Collect pairs of initialization dates whose latitude-averaged fields are nearly identical but whose longitude-resolved states differ sharply, for example opposite Madden–Julian Oscillation phases. TelePiT will issue nearly identical forecasts for the two members of every such pair, because longitude information never enters the model. If the verifying weather then diverges between the paired cases, as MJO theory predicts for winter temperatures and precipitation over North America, the model is provably blind to a predictable, physically important S2S signal.

Watch

Extended reading notes

Core claim

TelePiT's central claim is that explicitly structuring a forecasting model around three physical ideas — spherical geometry, scale-separated atmospheric dynamics, and long-range climate couplings — produces global S2S forecasts more accurate than any existing data-driven or operational system at both weeks 3–4 and weeks 5–6. Concretely, the paper reports z500 RMSE of 48.67 gpm at weeks 3–4 versus 53.81 gpm for CirT and 63.67 gpm for ECMWF; 2-meter temperature RMSE of 12.06 K versus 28.53 K for CirT and 60.89 K for ClimaX, with ACC of 0.996; and SpecDiv values of 0.0018–0.018 across key variables, one to two orders of magnitude below the best baselines. Ablation results attribute the gains to all three components, with the spherical harmonic embedding most critical for near-surface temperature, the wavelet decomposition most important for geopotential height, and the teleconnection bias increasingly valuable at the longer lead time. The authors interpret the results as evidence that a single model can capture the multi-scale physical processes and teleconnections that govern predictability beyond the two-week deterministic limit.

Load-bearing premise

The model sees only the average state around each latitude circle, so any forecast signal that lives in the east–west arrangement of the atmosphere — such as the traveling tropical rain clusters of the Madden–Julian Oscillation — is erased before training begins, and both the reported skill and the teleconnection story rest on that information being unnecessary.

Editorial extensions

If this is right

  • If the reported skill is real, operational S2S outlooks could be issued at a fraction of current cost: TelePiT runs at 14.5 GFLOPs with 37 million parameters, roughly 7,500 times fewer FLOPs than GraphCast and 11,500 times fewer than Pangu-Weather, with inference on the order of tens of milliseconds.
  • Each architectural component earns its place according to the ablations: removing the spherical harmonic embedding raises t2m RMSE from 12.1 K to 27.2 K, removing the wavelet decomposition costs 7–8% on geopotential height, and removing the teleconnection bias hurts most at weeks 5–6.
  • Spectral fidelity is the paper's largest margin: SpecDiv values of 0.0018–0.018 across variables are one to two orders of magnitude below every baseline, implying forecast power spectra much closer to the verification fields.
  • The advantage persists out-of-sample on the 2019 holdout year, which the paper takes as evidence that the model captures generalizable atmospheric structure rather than training-period patterns.
  • A single instantaneous zonal-mean snapshot suffices for both target windows at once, a direct corollary of the design that makes the reported skill a statement about how much predictable S2S signal sits in the latitude profile alone.

Reading between the lines

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

  • My reading: the measured gains may come chiefly from a strong smoothing prior — the zonal-mean tokenization plus the transport-form ODE suppresses high-wavenumber noise — rather than from the named mechanisms; the paper itself concedes in its appendix that the ODE is 'inspired by, rather than derived from' atmospheric equations, and the teleconnection patterns are simply trainable vectors never ch
  • Testable extension: replacing the single zonal mean per latitude with a few longitude sectors would directly measure how much remaining predictable signal lives in the east–west structure the model currently erases, and would show whether MJO-phase information adds or subtracts skill.
  • If the zonal-mean bet is right, S2S predictability research could shift toward the axisymmetric state — polar-vortex strength, meridional temperature gradients, tropical heating asymmetries — as the natural input variables for weeks 3–6, with the full three-dimensional field as the decode target.
  • Because 2-meter temperature in ERA5 is itself a model-derived field, the headline t2m numbers (RMSE 12.06 K, ACC 0.996) deserve an independent check against station-based or otherwise observation-grounded temperature analyses before being read as skill against the real near-surface atmosphere.
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

4 major / 5 minor

Summary. TelePiT is a deep-learning architecture for global subseasonal-to-seasonal (S2S) forecasting from ERA5 reanalysis data. The model encodes the initial state as per-latitude zonal means, evolves them through a multi-scale physics-informed neural ODE and a teleconnection-aware transformer, and outputs full longitude-resolved forecasts for weeks 3-4 and 5-6. The paper claims that this lightweight model outperforms both data-driven baselines and operational numerical weather prediction systems across all forecast horizons, with the largest gains for near-surface variables. The evaluation uses a ChaosBench-based protocol with RMSE, ACC, SpecDiv, MS-SSIM, and SpecRes, and includes an ablation study and an out-of-sample 2019 robustness check.

Significance. If the central claim held, this would be a striking result: a model that sees only zonal averages at a single initial time would produce the most skillful global S2S forecasts available, at a fraction of the computational cost of existing models. The paper deserves credit for its broad multi-metric evaluation, the explicit 2019 out-of-sample test, and its honest acknowledgment in Appendix A.3 that the latent ODE is inspired by, rather than derived from, atmospheric equations. However, the architecture irreversibly removes longitude-resolved information from the input, which is inconsistent with the paper's central narrative of teleconnection awareness and with the claimed full-field forecast skill. The omission of FuXi-S2S, a cited state-of-the-art baseline, further weakens the comparative claim. As a result, the findings are not currently established.

major comments (4)
  1. [Section 3.1, Eqs. (2)-(5); Section 3.4, Eq. (19)] The model's only input is the zonal mean per latitude, u_i = (1/W) sum_j X_{i,j}, and the longitude embedding is averaged before being added to the latitude embedding (Eqs. 4-5). The output head then maps each latitude token to all longitudes via Eq. (19). Consequently, the forecast is a function of latitude and the zonal-mean state only; it is invariant under arbitrary longitude shifts of the initial condition. This removes any mechanism for the teleconnection awareness claimed in Section 3.3.1: the MJO, NAO, and Rossby wave trains are defined by their zonal phase structure, which the architecture cannot see. A concrete falsifiable test is to shift the initial field by a longitude offset and confirm that the forecast does not change; the paper should also compare against a per-longitude climatology or persistence baseline. Until such a test is provided, the RMSE/ACC/SpecDiv gains in Tables 1-4 could be entirely dominated by the zonal-mean/climatological component of the targets.
  2. [Table 1 and Section 4.1] The paper cites FuXi-S2S (ref [8]) as a strong global S2S model but does not include it in the comparison tables. The abstract and Section 4.1 claim that TelePiT 'significantly outperforms state-of-the-art data-driven baselines' and 'across all forecast horizons.' Without FuXi-S2S, which is a directly relevant published S2S model, the state-of-the-art claim is not supported by the evidence presented.
  3. [Table 1, Section 4.1.1] The near-surface metrics are implausibly large in magnitude: u10 RMSE for TelePiT is 0.491 m/s versus 4.0 m/s for FourCastNetV2 and PanguWeather, and t2m RMSE is 12.1 K versus 60.9 K for ClimaX. For a model that ingests only zonal means, these gaps are suspicious and suggest either different evaluation conventions (for example, anomaly RMSE versus absolute RMSE, or different weighting/gridding) or a trivial solution such as predicting climatology. The paper should state explicitly whether these are absolute or anomaly metrics, and should add a climatology baseline to the comparison.
  4. [Appendix A.4, Proposition A.8 (Eq. 31); Appendix A.2, Proposition A.4 (Eq. 25)] The theoretical claims that the learned teleconnection patterns converge to a basis spanning the EOF subspace (Prop. A.8) and that the frequency bands preserve (1-delta) of the mutual information with the target (Prop. A.4) are asserted with proof sketches that assume the conclusion. Specifically, Prop. A.8's proof says the optimization 'is equivalent' to variance maximization without establishing that equivalence, and Prop. A.4 does not prove the existence of a band decomposition satisfying the stated mutual-information lower bound. These propositions are presented as theoretical support but do not provide it; they should be corrected or removed.
minor comments (5)
  1. [Table 2] The header reads 'SepcDiv' instead of 'SpecDiv'.
  2. [Appendix B.1, Eq. (36)] The ACC formula does not include the latitude weight w(theta_i) in the numerator, while the denominator includes it. Please verify whether this is a typo or an intentional definition.
  3. [Table 6] The table contains '9404' and '0556' where '0.9404' and '0.0556' are intended; there are also inconsistencies between 'SpecRec' and 'SpecRes' in Tables 7 and 8.
  4. [Section 3.2.1 and Eq. (9)] The term 'learnable wavelet decomposition' is used, but Eq. (9) is an MLP followed by a split; no wavelet properties such as orthogonality or multiresolution analysis are enforced. Clarify the terminology.
  5. [Section 3.2.2 vs. Appendix A.3] Section 3.2.2 states zero-padding boundary conditions (x_0 = x_{H+1} = 0), while Appendix A.3 states 'we apply periodic boundary conditions.' One of these descriptions is incorrect.

Circularity Check

2 steps flagged · score 5.0 of 10

Partially circular theoretical layer: Appendix A.4 assumes the multi-scale decomposition preserves target information without proof, and Proposition A.8's proof sketch restates the training objective as the claimed EOF-convergence result; the empirical evaluation itself is standard held-out skill and is not circular.

  1. other [Appendix A.2, Proposition A.4]
    "Proposition A.4 (Information Preservation). Each frequency band ℓ preserves mutual information with the forecast target, satisfying: Σ_{ℓ=0}^{L} I(x^{(ℓ)}; y) ≥ (1−δ) I(x; y) for arbitrarily small δ > 0 given sufficient model capacity."

    This proposition is stated without a proof or derivation. The multi-scale bands are produced by MLPs trained to minimize the forecast loss, so the claim that they retain (1−δ) of the target-relevant mutual information is exactly the desired conclusion, not a consequence of the stated capacity assumption. The 'theorem' is an assumption masquerading as a first-principles result, and it is load-bearing for the paper's claim that the learned decomposition does not discard predictable information.

  2. other [Appendix A.4, Proposition A.8 and its proof sketch]
    "During training, the patterns P_j and projection matrix W_p are jointly optimized to minimize prediction error. This optimization is equivalent to finding patterns that maximize the predictive information content, which necessarily includes the variance-maximizing directions captured by linear EOFs."

    The proof of the claimed convergence to an EOF-spanning basis consists of restating the training objective ('minimize prediction error') as 'maximizing predictive information' and then asserting that this 'necessarily includes' linear EOFs. No covariance analysis or spectral argument is supplied, and minimizing prediction error does not by itself imply that the learned span contains the linear EOF subspace. The theorem's conclusion is effectively the objective restated, so it provides no independent mathematical support for the teleconnection claim.

full rationale

The core empirical result—TelePiT trained on ERA5 (1979–2016), validated on 2017, and evaluated on 2018 plus out-of-sample 2019 against standard data-driven and operational baselines—is a conventional supervised evaluation on held-out data. No fitted parameter is renamed as a weather forecast, and the reported skill numbers are not forced by the loss construction, so the empirical derivation chain is not circular. The circularity is confined to the paper's theoretical/narrative layer. Appendix A.2's Proposition A.4 assumes the learned decomposition preserves target information, and Appendix A.4's Proposition A.8 rests on a proof sketch that restates the predictive-information objective as the conclusion. These propositions are load-bearing for the paper's claims of 'theoretical rigor' and 'explicit teleconnection modeling,' although they do not affect the numerical tables, which remain independent. A related but distinct concern is architectural: the input is a zonal mean and the longitude embedding is averaged (Eqs. 2–5), so longitude-resolved modes such as MJO phase cannot reach the model; this is a correctness/physical-mechanism mismatch rather than a circular reduction. Self-citations in the manuscript ([26,27,54,55]) are not load-bearing. Overall score 5: partial circularity in the justification layer, with independent empirical content.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The central claim depends on a long list of learned and hand-chosen parameters. The physics terms and teleconnection patterns are fitted on the training data, and several key hyperparameters (alpha, n_p, L) are unreported. The architecture also relies on domain assumptions about zonal averaging and single-snapshot initialization that are not validated.

free parameters (5)
  • lambda (teleconnection influence) = 0.2
    Chosen by sensitivity analysis in Appendix B.8; controls teleconnection bias in Eq. (15).
  • gamma (ODE scaling factor) = 0.1
    Set for numerical stability in Section 3.2.2; moderates the rate of change.
  • alpha (neural correction weight)
    Learned parameter balancing physical terms and MLP correction in Eq. (12); value not reported in the paper.
  • n_p (number of teleconnection patterns)
    Hyperparameter in Eq. (14); not specified in the paper.
  • L (number of multi-scale decomposition levels)
    Hyperparameter controlling number of frequency bands; not specified in the paper.
assumptions (5)
  • domain assumption Zonal averaging preserves enough information for S2S forecasting.
    Section 3.1: 'This zonal averaging operation is physically justified as atmospheric patterns often exhibit stronger correlations along latitude bands.' The model discards all longitude structure.
  • domain assumption A single initial snapshot X_t1 is sufficient to initialize a 3-6 week forecast.
    Eq. (1) uses only the state at day t1; no temporal history. MJO and other low-frequency modes are phase-dependent and may require longer input sequences.
  • ad hoc to paper Learned teleconnection patterns P_j represent real physical teleconnections.
    Section 3.3.1: P_j are 'typically can be obtained by manual extraction or automatic learning from data'; here they are trained from scratch, with no validation against observed climate indices.
  • standard math Standard spherical harmonic expansion and zonal projection.
    Appendix A.1 Proposition A.1; standard mathematics, not controversial.
  • domain assumption Atmospheric signals satisfy Holder continuity with exponent alpha > 0.
    Appendix A.2 Assumption 1, used for the wavelet decomposition error bound.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-Informed Teleconnection-Aware Transformer for Global Subseasonal-to-Seasonal Forecasting." pith.science (2026). https://pith.science/paper/XFQ4JDRE

@misc{pith2026250608049,
  author       = {Pith},
  title        = {Pith review of: Physics-Informed Teleconnection-Aware Transformer for Global Subseasonal-to-Seasonal Forecasting},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/XFQ4JDRE}},
  note         = {Machine review of arXiv:2506.08049}
}
read the original abstract

Subseasonal-to-seasonal (S2S) forecasting, which predicts climate conditions from several weeks to months in advance, represents a critical frontier for agricultural planning, energy management, and disaster preparedness. However, it remains one of the most challenging problems in atmospheric science, due to the chaotic dynamics of atmospheric systems and complex interactions across multiple scales. Current approaches often fail to explicitly model underlying physical processes and teleconnections that are crucial at S2S timescales. We introduce \textbf{TelePiT}, a novel deep learning architecture that enhances global S2S forecasting through integrated multi-scale physics and teleconnection awareness. Our approach consists of three key components: (1) Spherical Harmonic Embedding, which accurately encodes global atmospheric variables onto spherical geometry; (2) Multi-Scale Physics-Informed Neural ODE, which explicitly captures atmospheric physical processes across multiple learnable frequency bands; (3) Teleconnection-Aware Transformer, which models critical global climate interactions through explicitly modeling teleconnection patterns into the self-attention. Extensive experiments demonstrate that \textbf{TelePiT} significantly outperforms state-of-the-art data-driven baselines and operational numerical weather prediction systems across all forecast horizons, marking a significant advance toward reliable S2S forecasting.

Figures

Figures reproduced from arXiv: 2506.08049 by the authors.

Figure 1
Figure 1. The framework of TelePiT for global subseasonal-to-seasonal forecasting. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. RMSE comparison on variable 𝑧, 𝑡, 𝑢, and 𝑣 of different pressure levels. this capability was not available in the ECMWF-provided imple￾mentations. Additionally, GraphCast encountered out-of-memory when attempting inference for Weeks 5-6 forecast period. Due to space constraints, we present results for key variables and primary metrics in the main text. More additional experiments and model complexity analysis can be… view at source ↗
Figure 3
Figure 3. The global RMSE distribution of 𝑡850 with lead times weeks 3-4 in the testing set. 4.2 Ablation Study The ablation study presented in [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (17 more)
Figure 4
Figure 4. Figure 4: Forecasting Performance of 𝑡850 on each day in the testing set. 4.3 Empirical Analysis Global visualization [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Experiments distribution between main paper and appendix. Note that GraphCast (GC) results are unavailable at [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: ACC comparison of variable 𝑧, 𝑡, 𝑢, and 𝑣 on all different pressure levels. The color scale indicates normalized ACC values (higher is better), with darker blue representing superior performance. TelePiT consistently shows higher ACC values across all variables and pre…
Figure 7
Figure 7. Figure 7: MS-SSIM comparison of variable 𝑧, 𝑡, 𝑢, and 𝑣 on all different pressure levels. The color gradient indicates normalized MS-SSIM values (higher is better), showing TelePiT’s superior ability to preserve structural features in forecasts compared to all baseline models, w…
Figure 8
Figure 8. Figure 8: SpecDiv comparison of variable 𝑧, 𝑡, 𝑢, and 𝑣 on all different pressure levels. This metric quantifies differences in power spectrum distributions between predicted and ground truth fields. Lower values (lighter colors) indicate better spectral match. TelePiT demonstra…
Figure 9
Figure 9. Figure 9: SpecRes comparison of variable 𝑧, 𝑡, 𝑢, and 𝑣 on all different pressure levels. This metric measures spectral energy distribution accuracy. Lower values (lighter colors) indicate better performance. TelePiT exhibits substantially lower SpecRes values compared to all co…
Figure 10
Figure 10. Figure 10: Global visualization of RMSE distribution of [PITH_FULL_IMAGE:figures/full_fig_p020_10.png]
Figure 11
Figure 11. Figure 11: Global visualization of RMSE distribution of [PITH_FULL_IMAGE:figures/full_fig_p021_11.png]
Figure 12
Figure 12. Figure 12: Global visualization of RMSE distribution of [PITH_FULL_IMAGE:figures/full_fig_p021_12.png]
Figure 13
Figure 13. Figure 13: Global visualization of RMSE distribution of [PITH_FULL_IMAGE:figures/full_fig_p022_13.png]
Figure 14
Figure 14. Figure 14: Global visualization of RMSE distribution of [PITH_FULL_IMAGE:figures/full_fig_p022_14.png]
Figure 15
Figure 15. Figure 15: Global visualization of RMSE distribution of [PITH_FULL_IMAGE:figures/full_fig_p023_15.png]
Figure 16
Figure 16. Figure 16: Global visualization of RMSE distribution of [PITH_FULL_IMAGE:figures/full_fig_p023_16.png]
Figure 17
Figure 17. Figure 17: Daily forecasting performance (RMSE) of 𝑡500 in 2018. For improved visualization clarity, a Gaussian smoothing (with 𝜎 = 1.5) was applied to the daily values for models exhibiting high variability (Pangu, GraphCast, FourCastNetV2, CMA, ECMWF, NCEP, and UKMO). Note tha…
Figure 18
Figure 18. Figure 18: Daily forecasting performance (RMSE) of 𝑧850 in 2018. TelePiT CirT ClimaX RMSE 40 60 80 100 Jan Mar May Jul Sep TelePiT FourCastNet-V2 PanguWeather GraphCast RMSE 50 100 Jan Mar May Jul Sep TelePiT CMA ECMWF NCEP UKMO RMSE 50 100 Jan Mar May Jul Sep TelePiT CirT Clima…
Figure 19
Figure 19. Figure 19: Daily forecasting performance (RMSE) of 𝑧500 in 2018 [PITH_FULL_IMAGE:figures/full_fig_p024_19.png]
Figure 20
Figure 20. Figure 20: Parameter sensitivity analysis of teleconnection coefficient [PITH_FULL_IMAGE:figures/full_fig_p026_20.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 52 canonical work pages

  1. [8]

    Lei Chen, Xiaohui Zhong, Hao Li, Jie Wu, Bo Lu, Deliang Chen, Shang-Ping Xie, Libo Wu, Qingchen Chao, Chensen Lin, et al . 2024. A machine learning model that outperforms conventional global subseasonal forecast models. Nature Communications 15, 1 (2024), 6425

    work page 2024

  2. [1]

    2012.Spherical harmonics and approximations on the unit sphere: an introduction

    Kendall Atkinson and Weimin Han. 2012.Spherical harmonics and approximations on the unit sphere: an introduction . Vol. 2044. Springer Science & Business Media

    work page 2012

  3. [2]

    Mark P Baldwin and Timothy J Dunkerton. 2001. Stratospheric harbingers of anomalous weather regimes. Science 294, 5542 (2001), 581–584

    work page 2001

  4. [3]

    Mark P Baldwin, David B Stephenson, David WJ Thompson, Timothy J Dunkerton, Andrew J Charlton, and Alan O’Neill. 2003. Stratospheric memory and skill of extended-range weather forecasts. Science 301, 5633 (2003), 636–640

    work page 2003

  5. [4]

    Peter Bauer, Alan Thorpe, and Gilbert Brunet. 2015. The quiet revolution of numerical weather prediction. Nature 525, 7567 (2015), 47–55

    2015

  6. [5]

    Tom Beucler, Michael Pritchard, Stephan Rasp, Jordan Ott, Pierre Baldi, and Pierre Gentine. 2021. Enforcing analytic constraints in neural networks emulating physical systems. Physical review letters 126, 9 (2021), 098302

    work page 2021

  7. [6]

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

    work page 2023

  8. [7]

    Amy H Butler, Andrew Charlton-Perez, Daniela IV Domeisen, Isla R Simpson, and Jeremiah Sjoberg. 2019. Predictability of Northern Hemisphere final stratospheric warmings and their surface impacts. Geophysical Research Letters 46, 17-18 (2019), 10578–10588

    work page 2019

Show all 59 references
  1. [9]

    Hwangyong Choi, Jeongwhan Choi, Jeehyun Hwang, Kookjin Lee, Dongeun Lee, and Noseong Park. 2023. Climate modeling with neural advection–diffusion equation. Knowledge and Information Systems 65, 6 (2023), 2403–2427

    2023

  2. [10]

    Judah Cohen, Laurie Agel, Mathew Barlow, Chaim I Garfinkel, and Ian White

  3. [11]

    Judah Cohen, Dim Coumou, Jessica Hwang, Lester Mackey, Paulo Orenstein, Sonja Totz, and Eli Tziperman. 2019. S2S reboot: An argument for greater inclusion of machine learning in subseasonal to seasonal forecasts. Wiley Inter- disciplinary Reviews: Climate Change 10, 2 (2019), e00567

    2019

  4. [12]

    Taco S Cohen, Mario Geiger, Jonas Köhler, and Max Welling. 2018. Spherical CNNs. In International Conference on Learning Representations (ICLR)

    2018

  5. [13]

    Yoo-Geun Ham, Jeong-Hwan Kim, and Jing-Jia Luo. 2019. Deep learning for multi-year ENSO forecasts. Nature 573, 7775 (2019), 568–572

    2019

  6. [14]

    Sijie He, Xinyan Li, Laurie Trenary, Benjamin A Cash, Timothy DelSole, and Arindam Banerjee. 2022. Learning and dynamical models for sub-seasonal climate forecasting: Comparison and collaboration. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 36. 4495–4503

    2022

  7. [15]

    Hans Hersbach, Bill Bell, Paul Berrisford, Shoji Hirahara, András Horányi, Joaquín Muñoz-Sabater, Julien Nicolas, Carole Peubey, Raluca Radu, Dinand Schepers, et al. 2020. The ERA5 global reanalysis. Quarterly journal of the royal meteorolog- ical society 146, 730 (2020), 1999–2049

    2020

  8. [16]

    Brian J Hoskins, Michael E McIntyre, and Andrew W Robertson. 1985. On the use and significance of isentropic potential vorticity maps. Quarterly Journal of the Royal Meteorological Society 111, 470 (1985), 877–946

    1985

  9. [17]

    Jessica Hwang, Paulo Orenstein, Judah Cohen, Karl Pfeiffer, and Lester Mackey

  10. [18]

    George Em Karniadakis, Ioannis G Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. 2021. Physics-informed machine learning. Nature Reviews Physics 3, 6 (2021), 422–440

    2021

  11. [19]

    Karthik Kashinath, M Mustafa, Adrian Albert, JL Wu, C Jiang, Soheil Es- maeilzadeh, Kamyar Azizzadenesheli, R Wang, Ashesh Chattopadhyay, A Singh, et al. 2021. Physics-informed machine learning: case studies for weather and climate modelling. Philosophical Transactions of the ...

    2021

  12. [20]

    Joseph Kidston, Adam A Scaife, Steven C Hardiman, Daniel M Mitchell, Neal Butchart, Mark P Baldwin, and Lesley J Gray. 2015. Stratospheric influence on tropospheric jet streams, storm tracks and surface weather. Nature Geoscience 8, 6 (2015), 433–440

    2015

  13. [21]

    Dmitrii Kochkov, Jamie A Smith, Ayya Alieva, Qing Wang, Michael P Bren- ner, and Stephan Hoyer. 2021. Machine learning–accelerated computational fluid dynamics. Proceedings of the National Academy of Sciences 118, 21 (2021), e2101784118

    2021

  14. [22]

    Solomon Kullback and Richard A Leibler. 1951. On information and sufficiency. The annals of mathematical statistics 22, 1 (1951), 79–86

    1951

  15. [23]

    Thorsten Kurth, Shashank Subramanian, Peter Harrington, Jaideep Pathak, Morteza Mardani, David Hall, Andrea Miele, Karthik Kashinath, and Anima Anandkumar. 2023. Fourcastnet: Accelerating global high-resolution weather forecasting using adaptive fourier neural operators. InPro...

    2023

  16. [24]

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

    2023

  17. [25]

    Yang Liu, Zinan Zheng, Jiashun Cheng, Fugee Tsung, Deli Zhao, Yu Rong, and Jia Li. 2025. CirT: Global Subseasonal-to-Seasonal Forecasting with Geometry- inspired Transformer. In The Thirteenth International Conference on Learning Representations

    2025

  18. [26]

    Tengfei Lyu, Jindong Han, and Hao Liu. 2025. NRFormer: Nationwide Nuclear Radiation Forecasting with Spatio-Temporal Transformer. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    2025

  19. [27]

    Tengfei Lyu, Weijia Zhang, Jinliang Deng, and Hao Liu. 2025. AutoSTF: Decou- pled Neural Architecture Search for Cost-Effective Automated Spatio-Temporal Forecasting. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1. 985–996

    2025

  20. [28]

    Andrew J Majda. 2007. Multiscale models with moisture and systematic strategies for superparameterization. Journal of the Atmospheric Sciences 64, 7 (2007), 2726– 2734

    2007

  21. [29]

    William J Merryfield, Johanna Baehr, Lauriane Batté, Emily J Becker, Amy H But- ler, Caio AS Coelho, Gokhan Danabasoglu, Paul A Dirmeyer, Francisco J Doblas- Reyes, Daniela IV Domeisen, et al. 2020. Current and emerging developments in subseasonal to decadal prediction. Bullet...

    2020

  22. [30]

    Franco Molteni, Roberto Buizza, Tim N Palmer, and Thomas Petroliagis. 1996. The ECMWF ensemble prediction system: Methodology and validation.Quarterly journal of the royal meteorological society 122, 529 (1996), 73–119

    1996

  23. [31]

    Soukayna Mouatadid, Paulo Orenstein, Genevieve Flaspohler, Miruna Oprescu, Judah Cohen, Franklyn Wang, Sean Knight, Maria Geogdzhayeva, Sam Levang, Ernest Fraenkel, et al. 2023. SubseasonalclimateUSA: A dataset for subseasonal forecasting and benchmarking. Advances in Neural I...

    2023

  24. [32]

    Juan Nathaniel, Yongquan Qu, Tung Nguyen, Sungduk Yu, Julius Busecke, Aditya Grover, and Pierre Gentine. 2024. ChaosBench: A Multi-Channel, Physics-Based Benchmark for Subseasonal-to-Seasonal Climate Prediction. In The Conference on Neural Information Processing Systems Datase...

    2024

  25. [33]

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

    2023 arXiv

  26. [34]

    Tung Nguyen, Rohan Shah, Hritik Bansal, Troy Arcomano, Romit Maulik, Rao Kotamarthi, Ian Foster, Sandeep Madireddy, and Aditya Grover. 2024. Scaling Conference acronym ’XX, June 03–05, 2018, Woodstock, NY Tengfei Lyu, Jindong Han, and Hao Liu. transformer neural networks for s...

    2024

  27. [35]

    Maziar Raissi, Paris Perdikaris, and George E Karniadakis. 2019. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations. Journal of Computa- tional physics 378 (2019), 686–707

    2019

  28. [36]

    Stephan Rasp, Peter D Dueben, Sebastian Scher, Jonathan A Weyn, Soukayna Mouatadid, and Nils Thuerey. 2020. WeatherBench: a benchmark data set for data-driven weather forecasting. Journal of Advances in Modeling Earth Systems 12, 11 (2020), e2020MS002203

    2020

  29. [37]

    Stephan Rasp, Stephan Hoyer, Alexander Merose, Ian Langmore, Peter Battaglia, Tyler Russell, Alvaro Sanchez-Gonzalez, Vivian Yang, Rob Carver, Shreya Agrawal, et al. 2024. Weatherbench 2: A benchmark for the next generation of data-driven global weather models. Journal of Adva...

    2024

  30. [38]

    Markus Reichstein, Gustau Camps-Valls, Bjorn Stevens, Martin Jung, Joachim Denzler, Nuno Carvalhais, and F Prabhat. 2019. Deep learning and process understanding for data-driven Earth system science. Nature 566, 7743 (2019), 195–204

    2019

  31. [39]

    Andrew W Robertson, Arun Kumar, Malaquias Peña, and Frederic Vitart. 2015. Improving and promoting subseasonal to seasonal prediction. Bulletin of the American Meteorological Society 96, 3 (2015), ES49–ES53

    2015

  32. [40]

    Andrew W Robertson, Frederic Vitart, and Suzana J Camargo. 2020. Subseasonal to seasonal prediction of weather to climate with application to tropical cyclones. Journal of Geophysical Research: Atmospheres 125, 6 (2020), e2018JD029375

    2020

  33. [41]

    Suranjana Saha, Shrinivas Moorthi, Xingren Wu, Jiande Wang, Sudhir Nadiga, Patrick Tripp, David Behringer, Yu-Tai Hou, Hui-ya Chuang, Mark Iredell, et al

  34. [42]

    Kevin E Trenberth. 1998. Atmospheric moisture residence times and cycling: Implications for rainfall rates and climate change. Climatic change 39 (1998), 667–694

    1998

  35. [43]

    Geoffrey K Vallis. 2017. Atmospheric and oceanic fluid dynamics . Cambridge University Press

    2017

  36. [44]

    Yogesh Verma, Markus Heinonen, and Vikas Garg. 2024. ClimODE: Climate and Weather Forecasting with Physics-informed Neural ODEs. In The Twelfth International Conference on Learning Representations

    2024

  37. [45]

    Frédéric Vitart. 2014. Evolution of ECMWF sub-seasonal forecast skill scores. Quarterly Journal of the Royal Meteorological Society 140, 683 (2014), 1889–1899

    2014

  38. [46]

    Frederic Vitart, Constantin Ardilouze, Axel Bonet, Anca Brookshaw, M Chen, C Codorean, M Déqué, L Ferranti, E Fucile, M Fuentes, et al. 2017. The subseasonal to seasonal (S2S) prediction project database. Bulletin of the American Meteorological Society 98, 1 (2017), 163–173

    2017

  39. [47]

    Frédéric Vitart and Andrew W Robertson. 2018. The sub-seasonal to seasonal prediction project (S2S) and the prediction of extreme events. npj climate and atmospheric science 1, 1 (2018), 3

    2018

  40. [48]

    John M Wallace and David S Gutzler. 1981. Teleconnections in the geopotential height field during the Northern Hemisphere winter. Monthly weather review 109, 4 (1981), 784–812

    1981

  41. [49]

    Christopher J White, Henrik Carlsen, Andrew W Robertson, Richard JT Klein, Jeffrey K Lazo, Arun Kumar, Frederic Vitart, Erin Coughlan de Perez, Andrea J Ray, Virginia Murray, et al. 2017. Potential applications of subseasonal-to-seasonal (S2S) predictions. Meteorological appli...

    2017

  42. [50]

    Jared Willard, Xiaowei Jia, Shaoming Xu, Michael Steinbach, and Vipin Kumar

  43. [51]

    KD Williams, CM Harris, A Bodas-Salcedo, J Camp, RE Comer, D Copsey, D Fereday, T Graham, R Hill, T Hinton, et al. 2015. The met office global coupled model 2.0 (GC2) configuration. Geoscientific Model Development 88, 55 (2015), 1509–1524

    2015

  44. [52]

    Tongwen Wu, Yixiong Lu, Yongjie Fang, Xiaoge Xin, Laurent Li, Weiping Li, Weihua Jie, Jie Zhang, Yiming Liu, Li Zhang, et al . 2019. The Beijing climate center climate system model (BCC-CSM): The main progress from CMIP5 to CMIP6. Geoscientific Model Development 12, 4 (2019), ...

    2019

  45. [53]

    Chidong Zhang. 2005. Madden-julian oscillation. Reviews of Geophysics 43, 2 (2005)

    2005

  46. [54]

    Weijia Zhang, Chenlong Yin, Hao Liu, and Hui Xiong. 2025. Unleashing the power of pre-trained language models for irregularly sampled time series. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining

    2025

  47. [55]

    Weijia Zhang, Chenlong Yin, Hao Liu, Xiaofang Zhou, and Hui Xiong. 2024. Irregular multivariate time series forecasting: A transformable patching graph neural networks approach. In Forty-first International Conference on Machine Learning. Physics-Informed Teleconnection-Aware ...

    2024

  48. [2014]

    Journal of climate 27, 6 (2014), 2185–2208

    The NCEP climate forecast system version 2. Journal of climate 27, 6 (2014), 2185–2208

    2014

  49. [2019]

    In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining

    Improving subseasonal forecasting in the western US with machine learning. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining . 2325–2335

  50. [2021]

    Science 373, 6559 (2021), 1116–1121

    Linking Arctic variability and change with extreme winter weather in the United States. Science 373, 6559 (2021), 1116–1121

    2021

  51. [2022]

    Integrating scientific knowledge with machine learning for engineering and environmental systems. Comput. Surveys 55, 4 (2022), 1–37

    2022

Pith tools

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