Pith. sign in

REVIEW 4 major objections 5 minor 15 references

Solaris: A Foundation Model of the Sun

T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A 109-million-parameter transformer pre-trained on 13 years of multi-wavelength Sun images forecasts the solar atmosphere and transfers to an unseen wavelength.

desk verdict Plausible first solar foundation model with a useful public dataset, but the transfer claim is under-evidenced: no baselines, no error bars, and an undisclosed 1700 Å split. read the letter →

arxiv 2411.16339 v1 pith:ZQBVXI5X submitted 2024-11-25 astro-ph.SR astro-ph.IMcs.LGphysics.space-ph

Harris Abdul Majid , Pietro Sittoni , Francesco Tudisco This is my paper · ORCID
classification astro-ph.SRastro-ph.IMcs.LGphysics.space-ph
keywords solaratmosphereforecastingfoundationmodelSwinTransformertransferlearningmulti-wavelengthimagingspaceweatherDynamicsObservatorypre-training
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

The paper introduces Solaris, which it calls the first foundation model trained to forecast the Sun's atmosphere. The model is pre-trained on 13 years of full-disk, multi-wavelength images from the Solar Dynamics Observatory, spanning a complete solar cycle, and learns to predict the solar state 12 hours ahead from two prior states. The central evidence for the foundation-model claim is a transfer test: fine-tuning Solaris for 25 steps on the data-poor 1700 Å wavelength, which was never seen in pre-training, outperforms an identical model trained from scratch for 775 steps. If this holds, a single pre-trained model could capture generalizable solar dynamics and be adapted cheaply to wavelengths with scarce observations.

What carries the argument

The architecture follows the Aurora atmospheric model: a perceiver-based encoder tokenizes each wavelength image into patches, uses cross-attention to aggregate the variable number of wavelengths into a fixed set of latent vectors, and adds Fourier spatial and temporal embeddings; a 3D Swin Transformer U-Net processor with shifted-window self-attention evolves this representation through time; a perceiver-based decoder projects back to pixel space for any requested wavelength subset. A per-wavelength scaling factor normalizes intensities into [0,2], and a learnable blend of the raw value and its logarithm balances sensitivity to bright and faint features. Pre-training used two 12-hour-separated input states at eight wavelengths to forecast the next state, with a weighted MAE loss, AdamW, and 7750 steps.

What would settle it

Repeat the 1700 Å experiment with a from-scratch model trained for the same 7750 steps as pre-training, with the 987 samples split by time so that no fine-tuning frame shares a date with the test set, and with the exact split published; if the from-scratch model reaches the fine-tuned RMSE, the transfer advantage is not established.

Watch

Extended reading notes

Core claim

Solaris's central claim is that pre-training on eight well-sampled extreme-ultraviolet wavelengths gives the model a general representation of solar atmospheric dynamics that transfers to a wavelength it never saw. The concrete result is that fine-tuned SolarisS, a 117M-parameter 3D Swin Transformer, reaches lower RMSE on the 1700 Å forecasting task after 25 fine-tuning steps than an identical architecture trained from scratch for 775 steps. The paper also reports that the pre-trained model forecasts emerging active regions rotating into view, which suggests an internal representation of the Sun's global state rather than mere image interpolation. The model predicts all eight pre-training wavelengths with consistent error despite large intensity differences, and the larger SolarisS configuration beats the smaller SolarisT.

Load-bearing premise

That the 1700 Å fine-tuning comparison is a fair and uncontaminated test of transfer: the 987 samples are split so test frames never appear in fine-tuning or pre-training, and the from-scratch baseline receives enough training steps to converge.

Editorial extensions

If this is right

  • A pre-trained Solaris can be adapted to a new solar wavelength with only a handful of training steps, making forecasting feasible for wavelengths with sparse archival data.
  • The 1700 Å experiment suggests that missing wavelength channels in observational archives could be filled by fine-tuning rather than by collecting new data.
  • Consistent performance across eight wavelengths indicates that the normalization scheme prevents bright wavelengths from dominating training, a precondition for multi-wavelength foundation models.
  • The apparent ability to forecast emerging regions implies the model encodes a global, not just disk-facing, representation of the Sun.
  • Larger SolarisS outperforms SolarisT, suggesting further gains from scaling model size and input resolution.

Reading between the lines

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

  • The 25-versus-775-step comparison likely measures sample efficiency more than final accuracy; a longer from-scratch run could narrow the gap, so the paper's strongest evidence is the speed of convergence, not necessarily the asymptotic skill.
  • A direct testable extension is to fine-tune Solaris on a synthetic wavelength generated by corrupting one of the pre-training channels, such as adding noise or masking, and compare against a scratch model on the same synthetic data; this would control for the unknown 1700 Å data distribution.
  • The claimed internal representation of the Sun's far side could be probed by checking whether the model's latent representation predicts the magnetogram of a region once it rotates onto the visible disk; a null result would suggest the emerging-region forecasts come from temporal persistence of rotating features.
  • If transfer is real, the same approach should work for other sparsely observed AIA wavelengths or for EVE spectral irradiance, providing a cheap way to extend a foundation model without pre-training on new data.
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. The paper presents Solaris, a 3D Swin Transformer-based foundation model for forecasting the Sun's atmosphere. Solaris is pre-trained on 13 years of SDO/AIA synoptic images at eight wavelengths with a 12-hour forecasting objective, and the authors report multi-wavelength forecast RMSEs and qualitative predictions. The central claim is that fine-tuning the pre-trained SolarisS model for only 25 steps on the 1700 Å wavelength, which was not used in pre-training, outperforms an identical model trained from scratch for 775 steps. The authors argue this demonstrates transfer from well-sampled wavelengths to a data-poor, unseen wavelength and positions Solaris as a generalizable solar foundation model.

Significance. If the transfer claim is established, the paper would provide a valuable demonstration of foundation-model transfer in heliophysics, with practical implications for forecasting at sparsely observed wavelengths and for completing missing archive data. The paper also contributes a publicly released machine-learning dataset covering a full solar cycle, which is a useful community resource. The transfer setup, where the target wavelength is held out from pre-training and evaluation is on a temporally future period, is a sensible and well-posed experimental design. However, the current evaluation does not quantitatively support the headline claims: the fine-tuning result lacks split disclosure, error bars, and a converged baseline, and the main forecasting results are presented as raw RMSE without persistence or skill baselines.

major comments (4)
  1. [4.2] The central transfer claim is not quantitatively established because the fine-tuning data split is undisclosed. The paper states that 987 samples are available for 1700 Å, but it does not report how many fall in the 2010-2022 training period, the Jan-Jun 2023 validation period, and the Jul-Dec 2023 test period, nor how many test frames are evaluated. Without this information, and without repeated-seed statistics or error bars, the sentence 'Finetuning SolarisS for only 25 training steps ... outperformed an identical model trained from scratch for 775 steps' cannot be assessed for statistical significance.
  2. [4.2, Figure 5] The from-scratch baseline may not be converged. At the stated effective batch size of 32, 775 steps corresponds to roughly 25 epochs if all 987 samples are available for training, and Figure 5 shows the from-scratch RMSE still declining at the final step. The paper does not report the fine-tuning hyperparameters (learning rate, schedule, loss weights) beyond saying the configuration is 'the same', nor does it provide final RMSE values. A longer training run, ideally with early stopping or a convergence criterion, is needed to distinguish a training-speed advantage from a genuine transfer advantage.
  3. [Table 2] The quantitative forecasting results are reported as raw-intensity RMSE with no baseline. No persistence forecast, autoregressive baseline, or scale-invariant skill metric is provided, so the reader cannot tell whether the RMSE values represent skillful prediction or merely a model that reproduces slowly evolving solar features. Because intensity scales differ by orders of magnitude across the eight wavelengths, comparing raw RMSE across rows is not meaningful; normalized error metrics or skill scores should be added.
  4. [4.1, Figures 2-3] The claim that accurate predictions of emerging regions imply 'an internal representation of the Sun's global state' is supported only by visual inspection. A quantitative test is needed, such as evaluating forecast error as a function of heliographic longitude or limb distance, or comparing predicted far-side emergence against HMI magnetogram evolution. Without such an analysis, the statement over-interprets the figures.
minor comments (5)
  1. [4.1, Table 2, Figures 2-3] The text and Table 2 use 355 Å while Figures 2 and 3 use 335 Å; the standard AIA passband is 335 Å. Please correct the typo throughout.
  2. [Figures 6-9] Figures 6-9 are not referenced in the text and appear after the references; either cite them in Section 4.1 or remove them.
  3. [Figure 1, Section 3.1] Figure 1 and the encoder description mention EVE and HMI inputs, but the experiments use only AIA data; clarify that EVE and HMI are planned extensions rather than part of the current model.
  4. [3.4] The transform equation applies min(x, 2.5) and max(x, 10^-3) but the choice of these constants is not justified, and the behavior for x = 0 should be stated explicitly.
  5. [4.1] The weighted MAE loss weights are not specified; report how the per-wavelength weights are chosen.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation: the forecasting target is temporally future data and no fitted parameter is renamed as a prediction; the main caveats are experimental-soundness concerns, not circularity.

full rationale

I walked the paper's claimed derivation chain. The central claim is that Solaris, pretrained on eight AIA wavelengths over 2010–2022, can forecast 12-hour-ahead solar atmospheric states, and that fine-tuning on the data-poor 1700 Å wavelength outperforms an identical model trained from scratch. No step in this chain reduces to its own inputs by construction. The forecast target X_{t+1} is on the temporally future side of the chronological train/validation/test split described in Section 4, so the evaluation is not a re-statement of the training loss. The normalization scaling factor in Section 3.4 is computed from the training set, but it is an input-preprocessing constant, not the predicted quantity, and it is applied uniformly to inputs and targets. The fine-tuning experiment compares a pretrained-then-fine-tuned model against a from-scratch model on the same 1700 Å task; this is a legitimate (if not fully controlled) comparison of transfer learning, not a fitted parameter renamed as a prediction. The paper contains no self-citations, and the architecture choices are attributed to external prior work (Aurora, Swin Transformer, Perceiver). The reader's and skeptic's concerns—undisclosed fine-tuning/test split overlap and the short from-scratch training budget—are validity and robustness issues, not circularity. They do not exhibit an equation or construction where the output equals an input by definition. Accordingly, the circularity score is very low; the paper's derivation is self-contained with respect to circularity, and the caveats belong to experimental soundness rather than circular reasoning.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claims rest on the data representation, the temporal split, the architecture's inductive bias, and the evaluation metric. The free parameters are the normalization constants and loss weights that shape the reported errors; they are fitted or unreported, not derived. No circularity is built into the model, but the evaluation does not independently calibrate the error metric.

free parameters (4)
  • per-wavelength normalization scaling factor = half the average maximum pixel value per wavelength (computed on training data)
    Section 3.4: used to map all wavelengths into [0,2]; fitted to the training set and affects both inputs and the RMSE numbers reported after unscaling.
  • learnable transform coefficients c1 and c2 = initialized to 0.5, learned during training
    Section 3.4: parameters in the intensity-balancing transformation; final values are not reported.
  • weighted MAE loss weights = not reported
    Section 4.1: the objective is called 'weighted Mean Absolute Error' but the per-wavelength weights are never specified, so the model's objective is not fully defined.
  • patch size = 8 (8x8 patches)
    Section 5: patch size is fixed at 8x8 due to computational constraints; it determines the tokenization in the encoder.
assumptions (5)
  • domain assumption AIA 512x512 synoptic images preprocessed as in Galvez et al. capture the atmospheric state needed for 12-hour forecasting.
    Section 2: the model only sees full-disk images from the Earth-facing side; if this partial view is insufficient, the forecasting problem is not well posed.
  • domain assumption The temporal split prevents data leakage and represents real forecasting conditions.
    Section 4: pretraining uses 2010-2022 and test 2023, but the 1700 Å fine-tuning split is not described, so clean evaluation is assumed rather than demonstrated.
  • domain assumption A perceiver-encoded 3D Swin Transformer U-Net is a sufficient inductive bias for solar atmospheric dynamics.
    Sections 3.1-3.2: the architecture is adopted from Aurora and Swin-UNet without domain-specific validation or ablations.
  • domain assumption RMSE on raw pixel intensity is a meaningful forecast-skill metric across wavelengths.
    Section 4.1, Table 2: dimmer wavelengths naturally produce smaller RMSE, so comparing RMSE across wavelengths is only meaningful under this assumption.
  • domain assumption The model can infer unobserved far-side structures from rotating observed images.
    Section 4.1: the claim that Solaris has an 'internal representation of the Sun's global state' presumes the model can learn hidden dynamics from partial observations, which is not independently verified.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Solaris: A Foundation Model of the Sun." pith.science (2026). https://pith.science/paper/ZQBVXI5X

@misc{pith2026241116339,
  author       = {Pith},
  title        = {Pith review of: Solaris: A Foundation Model of the Sun},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZQBVXI5X}},
  note         = {Machine review of arXiv:2411.16339}
}
read the original abstract

Foundation models have demonstrated remarkable success across various scientific domains, motivating our exploration of their potential in solar physics. In this paper, we present Solaris, the first foundation model for forecasting the Sun's atmosphere. We leverage 13 years of full-disk, multi-wavelength solar imagery from the Solar Dynamics Observatory, spanning a complete solar cycle, to pre-train Solaris for 12-hour interval forecasting. Solaris is built on a large-scale 3D Swin Transformer architecture with 109 million parameters. We demonstrate Solaris' ability to generalize by fine-tuning on a low-data regime using a single wavelength (1700 {\AA}), that was not included in pre-training, outperforming models trained from scratch on this specific wavelength. Our results indicate that Solaris can effectively capture the complex dynamics of the solar atmosphere and transform solar forecasting.

Figures

Figures reproduced from arXiv: 2411.16339 by the authors.

Figure 1
Figure 1. Illustration of Solaris’ encoder module. The encoder processes multi-wavelength solar [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Multi-wavelength forecasting results from [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Multi-wavelength forecasting results from [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: RMSE error during training of SolarisS and SolarisT [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: RMSE error during finetuning of SolarisS. 4.2 Finetuning A significant challenge in solar atmospheric forecasting is the uneven availability of data across different wavelengths. Specifically, the 1700 Å wavelength is observed in only 987 of our samples, compared to ap…
Figure 6
Figure 6. Figure 6: Multi-wavelength forecasting results from [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Multi-wavelength forecasting results from [PITH_FULL_IMAGE:figures/full_fig_p012_7.png]
Figure 8
Figure 8. Figure 8: Multi-wavelength forecasting results from [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: Multi-wavelength forecasting results from [PITH_FULL_IMAGE:figures/full_fig_p014_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

15 extracted references · 4 canonical work pages

  1. [1]

    Chemberta-2: Towards chemical foundation models

    Walid Ahmad, Elana Simon, Seyone Chithrananda, Gabriel Grand, and Bharath Ramsundar. Chemberta-2: Towards chemical foundation models. arXiv preprint arXiv:2209.01712, 2022

    arXiv 2022

  2. [2]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Dosovitskiy Alexey. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv: 2010.11929 , 2020

    arXiv 2010

  3. [3]

    Aurora: A foundation model of the atmosphere

    Cristian Bodnar, Wessel P Bruinsma, Ana Lucic, Megan Stanley, Johannes Brandstetter, Patrick Garvan, Maik Riechert, Jonathan Weyn, Haiyu Dong, Anna Vaughan, et al. Aurora: A foundation model of the atmosphere. arXiv preprint arXiv:2405.13063, 2024

    arXiv 2024

  4. [4]

    On the opportunities and risks of foundation models

    Rishi Bommasani, Drew A Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, et al. On the opportunities and risks of foundation models. arXiv preprint arXiv:2108.07258, 2021

    arXiv 2021

  5. [5]

    Swin-unet: Unet-like pure transformer for medical image segmentation

    Hu Cao, Yueyue Wang, Joy Chen, Dongsheng Jiang, Xiaopeng Zhang, Qi Tian, and Manning Wang. Swin-unet: Unet-like pure transformer for medical image segmentation. In European conference on computer vision, pages 205–218. Springer, 2022

    2022

  6. [6]

    A machine- learning data set prepared from the nasa solar dynamics observatory mission

    Richard Galvez, David F Fouhey, Meng Jin, Alexandre Szenicer, Andrés Muñoz-Jaramillo, Mark CM Cheung, Paul J Wright, Monica G Bobra, Yang Liu, James Mason, et al. A machine- learning data set prepared from the nasa solar dynamics observatory mission. The Astrophysical Journal Supplement Series, 242(1):7, 2019

    work page 2019

  7. [7]

    Poseidon: Efficient foundation models for pdes

    Maximilian Herde, Bogdan Raoni ´c, Tobias Rohner, Roger Käppeli, Roberto Molinaro, Em- manuel de Bézenac, and Siddhartha Mishra. Poseidon: Efficient foundation models for pdes. arXiv preprint arXiv:2405.19101, 2024

    arXiv 2024

  8. [8]

    Perceiver: General perception with iterative attention

    Andrew Jaegle, Felix Gimeno, Andy Brock, Oriol Vinyals, Andrew Zisserman, and Joao Carreira. Perceiver: General perception with iterative attention. In International conference on machine learning, pages 4651–4664. PMLR, 2021

    2021

Show all 15 references
  1. [9]

    Graphcast: Learning skillful medium-range global weather forecasting

    Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, Peter Wirnsberger, Meire Fortunato, Ferran Alet, Suman Ravuri, Timo Ewalds, Zach Eaton-Rosen, Weihua Hu, et al. Graphcast: Learning skillful medium-range global weather forecasting. arXiv preprint arXiv:2212.12794, 2022

    2022 arXiv

  2. [10]

    Predicting solar flares using a novel deep convolutional neural network

    Xuebao Li, Yanfang Zheng, Xinshuo Wang, and Lulu Wang. Predicting solar flares using a novel deep convolutional neural network. The Astrophysical Journal, 891(1):10, 2020

    2020

  3. [11]

    Language models of protein sequences at the scale of evolution enable accurate structure prediction

    Zeming Lin, Halil Akin, Roshan Rao, Brian Hie, Zhongkai Zhu, Wenting Lu, Allan dos Santos Costa, Maryam Fazel-Zarandi, Tom Sercu, Sal Candido, et al. Language models of protein sequences at the scale of evolution enable accurate structure prediction. BioRxiv, 2022:500902, 2022

    2022

  4. [12]

    Swin transformer: Hierarchical vision transformer using shifted windows

    Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In Proceedings of the IEEE/CVF international conference on computer vision , pages 10012–10022, 2021

    2021

  5. [13]

    Exploring possibilities for solar irradiance prediction from solar photosphere images using recurrent neural networks

    Amita Muralikrishna, Rafael Duarte Coelho dos Santos, and Luis Eduardo Antunes Vieira. Exploring possibilities for solar irradiance prediction from solar photosphere images using recurrent neural networks. Journal of Space Weather and Space Climate , 12:19, 2022

    2022

  6. [14]

    Climax: A foundation model for weather and climate

    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

    2023 arXiv

  7. [15]

    Solar flare prediction with the hybrid deep convolutional neural network

    Yanfang Zheng, Xuebao Li, and Xinshuo Wang. Solar flare prediction with the hybrid deep convolutional neural network. The Astrophysical Journal, 885(1):73, 2019. 10 0094 Å Xt 1 Xt Targ: Xt + 1 Pred: Xt + 1 Erro: |Xt + 1 Xt + 1| 0131 Å 0171 Å 0193 Å 0211 Å 0304 Å 0335 Å 1600 Å ...

    2019

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.