HelioFill: History-Conditioned Diffusion Estimation of Solar Far-Side EUV Emission

arXiv:2510.17012 · astro-ph.SR, physics.space-ph · Submitted 2025-10-19 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Astrophysics Radio. Generated commentary on the latest astrophysics papers.

Vera: I'm Vera.

Jocelyn: And I'm Jocelyn, here with Subrahmanyan.

Vera: We're looking at the paper: "HelioFill: Diffusion-Based Model for EUV Reconstruction of the Solar Farside".

Jocelyn: Here's the gist of the paper. HelioFill presents diffusion-based inpainting models to restore full-Sun EUV coverage by synthesizing missing data from Earth-side and STEREO views This model restores full-Sun EUV coverage by synthesizing the STEREO-B sector from Earth-side (SDO) and STEREO-A views Pruning redundant reasoning tokens The loss of STEREO-B in 2014 created a persistent blind spot in Extreme UltraViolet (EUV) imaging of the solar farside HelioFill is the first denoising diffusion inpainting model that restores full-Sun EUV coverage by synthesizing the STEREO-B sector from Earth-side (SDO) and STEREO-A views It couples a latent diffusion backbone with domain-specific additions such as spectral gating, confidence weighting, and auxiliary regularizers to produce operationally suitable 304 ˚A reconstructions On held-out data, the model preserves the observed hemisphere with mean SSIM 0.871 and mean PSNR 25.56 dB while reconstructing the masked hemisphere with mean SSIM 0.801 and mean PSNR 17.41 dB The generated maps maintain cross-limb continuity and coronal morphology, supporting synoptic products and cleaner inner-boundary conditions for coronal/heliospheric models How the training signal is built The training dataset was constructed from the 2011-2014 interval when SDO/AIA and both STEREO/EUVI spacecraft provided nearly continuous 360◦ coverage AIA images were processed to Level 1.5 using the aiapy calibration tools which update pointing information and register the images to a common plate scale EUVI images from both STEREO spacecraft were calibrated using the secchi prep routine in SolarSoft including flat-fielding, despiking, and limb fitting Synchronic maps were generated by reprojecting each image to Carrington equalarea (CEA) longitude–latitude mosaics To emulate post-2014 observing conditions, input/output training pairs were constructed by masking the unobserved farside sector due to the STEREO-B coverage gap Synthetic data was augmented by applying masks directly to synchronic Carrington mosaics to generate N = 10 masked variants for each real mosaic Where the gains come from HelioFill achieves substantial improvements across all evaluation metrics compared to the PixelHacker baseline Key heliophysics features are preserved with high fidelity, including coronal holes appearing with appropriate contrast and boundary definition The ablation study reveals that mask expansion contributes most to seam-error reduction while maintaining perceptual quality The spectral gating with confidence weighting (SG+CW) provides additional hole reconstruction fidelity HelioFill raises the mean PSNR from 13.85 to 17.41 dB with corresponding mean SSIM improvement (0.757 → 0.801) in the reconstructed farside The seam-band L2 error drops from 0.01101 to 0.00874, indicating that seamless cross-limb blending is essential for operational synoptic products Inference and feathered composition To minimize visible seams at inpainting boundaries, a post-processing feathered blending is applied during inference A soft alpha mask is constructed by applying average pooling to the binary hole mask The final output is xˆfinal = αsoft ⊙ xˆ + (1 − αsoft) ⊙ xˆpreserve This post-process complements the training-time seam loss and yields smooth, measurement-consistent mosaics Discussion HelioFill demonstrates that domain-specific diffusion models can successfully bridge critical observational gaps in solar physics The ability to reconstruct farside EUV observations addresses a fundamental limitation in space weather forecasting that has persisted since the loss of STEREO-B This approach restores global context essential for understanding coronal evolution by tracking coronal holes as they rotate behind the limb The preservation of spacecraft measurements with minimal perturbation is crucial for operational deployment The success of HelioFill comes from several key design choices that address the unique challenges of scientific image inpainting Acknowledgments Firas Ben Ameur acknowledges support from the KAUST Global Fellowship Program (Award No.

Vera: RFS-KGFP2023-6044) Rayan Dhib acknowledges support from the FWO SB PhD fellowship (No.

Jocelyn: 1S66825N) For computational resources, this research used the Ibex cluster, managed by the Supercomputing Core Laboratory at KAUST This work was funded by KAUST via the student visiting program (Grant No.

Vera: BAS/1/1663-01-01) References Barnes, W.

Title and authors: Vera: Now let's look at the specific title and authors of this paper to get a feel for what they tackled. The paper is titled "HelioFill: Diffusion-Based Model for EUV Reconstruction of the Solar Farside". The authors are Firas Ben Ameur, Rayan Dhib, Yahia Battach, Andrea Lani, Matteo Parsani, Omar Knio and Stefaan Poedts.

Jocelyn: It sounds very specific to what they're doing—a diffusion-based model for EUV reconstruction. That tells you right away that the technique involves generative modeling rather than just simple interpolation or fitting existing data points.

Subrahmanyan: The implication here is that they are trying to generate physically plausible missing data, not just fill in gaps with blurry averages, which is important when you're trying to model solar structures like coronal holes.

Vera: That's right. They are aiming to restore the full-Sun EUV coverage by synthesizing the STEREO-B sector using Earth-side SDO and STEREO-A views. It’s about bridging that two thousand fourteen gap where we lost that full context of how things rotate around the sun <ref:2510.17012#pg1>.

Jocelyn: So, if you think about it, before this paper, we were limited in tracking features as they moved across the solar disk because of that missing piece from STEREO-B. This work aims to fix that observational limitation.

Subrahmanyan: From a theoretical standpoint, restoring this context is crucial for understanding the evolution of coronal holes over time because you need those boundary conditions to run models like surface flux transport.

The paper's summary: Vera: So, what they actually did in HelioFill is present a denoising diffusion inpainting model that restores full-Sun EUV coverage by synthesizing the STEREO-B sector from Earth-side SDO and STEREO-A views.

Jocelyn: They coupled a latent diffusion backbone with some domain specific additions like spectral gating, confidence weighting, and auxiliary regularizers to produce reconstruction results that are operationally suitable for three hundred four degree reconstructions <ref:2510.17012#pg3>.

Subrahmanyan: These additions suggest they weren't just using a vanilla diffusion model; they tailored the process specifically for solar data constraints, which is a smart way to improve the accuracy of the synthesis.

Vera: And looking at their results on held-out data, they preserved the observed hemisphere with a mean SSIM of zero point eight seven one and a mean PSNR of twenty-five point five six dB, while reconstructing the masked hemisphere with a mean SSIM of zero point eight zero one and a mean PSNR of seventeen point four one dB.

Jocelyn: Those numbers show they managed to keep the real data high quality while still generating a reconstruction that’s fairly good, though clearly less accurate than what was actually observed in the preserved area.

Subrahmanyan: The fact that they maintain cross-limb continuity and coronal morphology in the generated maps is important because it means these synthetic images support synoptic products and provide cleaner inner boundary conditions for coronal or heliospheric models.

The paper's improvements: Vera: They point out a few key improvements compared to the PixelHacker baseline. They show that mask expansion contributes most to reducing seam error while maintaining perceptual quality, which means expanding the masked area helps smooth out the transitions between what was seen and what was generated.

Jocelyn: Then they highlight that spectral gating with confidence weighting provides extra fidelity for hole reconstruction, which is a specific fix for getting those features right in the missing data.

Subrahmanyan: The ablation study shows that this spectral gating with confidence weighting raises the mean PSNR from thirteen point eight five to seventeen point four one dB, and it improves the mean SSIM from zero point seven five seven up to zero point eight zero one in the reconstructed farside region.

Vera: The seam-band L2 error drops from a value of zero point zero one one zero one down to zero point zero zero eight seven four, which is a significant reduction, indicating that seamless cross-limb blending is essential for operational synoptic products.

Jocelyn: And during inference, they use a post processing feathered blending step where they apply an average pooling to the binary hole mask to create a soft alpha mask before combining the preserved and reconstructed images.

Subrahmanyan: This post processing complements the training time seam loss and yields smooth, measurement consistent mosaics, which is critical for getting reliable outputs when you actually deploy this in a forecasting system.

Conclusion: Vera: So, to wrap up, HelioFill demonstrates that domain specific diffusion models can successfully bridge critical observational gaps in solar physics by synthesizing the STEREO-B sector from Earth-side SDO and STEREO-A views.

Jocelyn: The ability to reconstruct farside EUV observations addresses a fundamental limitation in space weather forecasting that has persisted since the loss of STEREO-B, restoring global context for understanding coronal evolution.

Subrahmanyan: This approach is crucial because it restores the global context needed for tracking coronal holes as they rotate behind the limb and ensures that spacecraft measurements are preserved with minimal perturbation, which is essential for operational deployment.

Vera: The success comes down to careful design choices like spectral gating and confidence weighting, which address the unique challenges of scientific image inpainting.

Jocelyn: It shows how these techniques can successfully fill in observational gaps when you have the right training data and model architecture for complex physical systems like the solar atmosphere.

Subrahmanyan: The paper "HelioFill: Diffusion-Based Model for EUV Reconstruction of the Solar Farside" gives us a tool to get better, more complete pictures of the sun's dynamics across its entire surface.

King Abdullah University of Science and Technology (KAUST) · Centre for Mathematical Plasma-Astrophysics, Department of Mathematics, KU Leuven · Institute of Physics, University of Maria Curie-Sk lodowska

astro-ph.SR, physics.space-ph

Submitted: 2025-10-19

Updated: 2026-11-12

DOI: 10.3847/2041-8213/aeb378

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 79/100

The gist: HelioFill presents diffusion-based inpainting models to restore full-Sun EUV coverage by synthesizing missing data from Earth-side and STEREO views This model restores full-Sun EUV coverage by…

Key concepts

Diffusion-Based Inpainting
This is a type of AI model that learns to fill in missing parts of an image by gradually removing noise. Instead of just guessing, the model starts with a noisy version of the incomplete image and learns how to iteratively refine it until it generates a realistic, coherent reconstruction that matches the surrounding observed data.
STEREO-B Sector Synthesis
The paper focuses on reconstructing the STEREO-B sector, which was lost in 2014, by intelligently combining observations from Earth-side (SDO) and STEREO-A spacecraft. The model uses this combination to synthesize the missing data, effectively creating a complete picture of the Sun's surface that was previously unobtainable.
Domain-Specific Additions
HelioFill enhances the standard diffusion process with specialized tools like spectral gating and confidence weighting. These additions guide the AI to focus on specific physical features, ensuring that the reconstructed EUV maps are not just visually plausible but also scientifically accurate and operationally useful for solar physics models.

Terminology

Summary

HelioFill presents diffusion-based inpainting models to restore full-Sun EUV coverage by synthesizing missing data from Earth-side and STEREO views This model restores full-Sun EUV coverage by synthesizing the STEREO-B sector from Earth-side (SDO) and STEREO-A views

Pruning redundant reasoning tokens

The loss of STEREO-B in 2014 created a persistent blind spot in Extreme UltraViolet (EUV) imaging of the solar farside HelioFill is the first denoising diffusion inpainting model that restores full-Sun EUV coverage by synthesizing the STEREO-B sector from Earth-side (SDO) and STEREO-A views It couples a latent diffusion backbone with domain-specific additions such as spectral gating, confidence weighting, and auxiliary regularizers to produce operationally suitable 304 ˚A reconstructions On held-out data, the model preserves the observed hemisphere with mean SSIM 0.871 and mean PSNR 25. p. 0

Where the paper says it

On held-out data, the model preserves the observed hemisphere with mean SSIM 0. 871 and mean PSNR 25.

56 dB while reconstructing the masked hemisphere with mean SSIM 0. p. 0
Where the paper says it

871 and mean PSNR 25. 56 dB, while reconstructing the masked hemisphere with mean SSIM 0.

801 and mean PSNR 17. p. 0
Where the paper says it

56 dB, while reconstructing the masked hemisphere with mean SSIM 0. 801 and mean PSNR 17.

41 dB The generated maps maintain cross-limb continuity and coronal morphology, supporting synoptic products and cleaner inner-boundary conditions for coronal/heliospheric models

How the training signal is built

The training dataset was constructed from the 2011-2014 interval when SDO/AIA and both STEREO/EUVI spacecraft provided nearly continuous 360◦ coverage AIA images were processed to Level 1.5 using the aiapy calibration tools which update pointing information and register the images to a common plate scale EUVI images from both STEREO spacecraft were calibrated using the secchi prep routine in SolarSoft including flat-fielding, despiking, and limb fitting Synchronic maps were generated by reprojecting each image to Carrington equalarea (CEA) longitude–latitude mosaics To emulate post-2014 observing conditions, input/output training pairs were constructed by masking the unobserved farside sector due to the STEREO-B coverage gap Synthetic data was augmented by applying masks directly to synchronic Carrington mosaics to generate N = 10 masked variants for each real mosaic

Where the gains come from

HelioFill achieves substantial improvements across all evaluation metrics compared to the PixelHacker baseline Key heliophysics features are preserved with high fidelity, including coronal holes appearing with appropriate contrast and boundary definition The ablation study reveals that mask expansion contributes most to seam-error reduction while maintaining perceptual quality The spectral gating with confidence weighting (SG+CW) provides additional hole reconstruction fidelity HelioFill raises the mean PSNR from 13.85 to 17.41 dB with corresponding mean SSIM improvement (0. p. 8

Where the paper says it

In the reconstructed farside, HelioFill raises the mean PSNR from 13. 41 dB with corresponding mean SSIM improvement (0.

757 → 0.801) in the reconstructed farside The seam-band L2 error drops from 0. p. 8
Where the paper says it

This gain represents a significantly better recovery of the EUV intensity structure in areas where no direct observations exist. Crucially, boundary consistency improves dramatically: the seam-band L2 error drops from 0.

01101 to 0. p. 7
Where the paper says it

Main results on full-Sun CEA maps (304 ˚A) Method Hole PSNR [dB] ↑ Known PSNR [dB] ↑ Hole SSIM ↑ Known SSIM ↑ Seam L2 ↓ LPIPS (hole) ↓ Coverage PixelHacker 13. 01101 ± 0.

00874, indicating that seamless cross-limb blending is essential for operational synoptic products

Inference and feathered composition

To minimize visible seams at inpainting boundaries, a post-processing feathered blending is applied during inference A soft alpha mask is constructed by applying average pooling to the binary hole mask The final output is xˆfinal = αsoft ⊙ xˆ + (1 − αsoft) ⊙ xˆpreserve This post-process complements the training-time seam loss and yields smooth, measurement-consistent mosaics

Discussion

HelioFill demonstrates that domain-specific diffusion models can successfully bridge critical observational gaps in solar physics The ability to reconstruct farside EUV observations addresses a fundamental limitation in space weather forecasting that has persisted since the loss of STEREO-B This approach restores global context essential for understanding coronal evolution by tracking coronal holes as they rotate behind the limb The preservation of spacecraft measurements with minimal perturbation is crucial for operational deployment The success of HelioFill comes from several key design choices that address the unique challenges of scientific image inpainting

Acknowledgments

Firas Ben Ameur acknowledges support from the KAUST Global Fellowship Program (Award No. RFS-KGFP2023-6044) Rayan Dhib acknowledges support from the FWO SB PhD fellowship (No. p. 9

Where the paper says it

RFS-KGFP2023- 6044). Rayan Dhib acknowledges support from the FWO SB PhD fellowship (No.

1S66825N) For computational resources, this research used the Ibex cluster, managed by the Supercomputing Core Laboratory at KAUST This work was funded by KAUST via the student visiting program (Grant No. p. 9
Where the paper says it

For computational resources, this research used the Ibex cluster, managed by the Supercomputing Core Laboratory at KAUST. This work was funded by KAUST via the student visiting program (Grant No.

BAS/1/1663-01-01)

References

Barnes, W. T., Cheung, M. p. 9

Where the paper says it

REFERENCES Barnes, W. , Cheung, M.

C. M., Bobra, M. p. 9
Where the paper says it

, Cheung, M. , Bobra, M.

G., et al., Journal of Open Source Software, 5, 2801 Bentely, R. p. 9
Where the paper says it

, Bobra, M. 2020, Journal of Open Source Software, 5, 2801, doi: 10.

D., & Freeland, S. p. 2
Where the paper says it

Bentely & S. Freeland 1998), including flat-fielding, despiking, and limb fitting.

L., in ESA Special Publication, Vol. p. 9
Where the paper says it

, & Freeland, S. 1998, in ESA Special Publication, Vol.

417, Crossroads for European Solar and Heliospheric Physics Bertello, L. p. 9
Where the paper says it

1998, in ESA Special Publication, Vol. 417, Crossroads for European Solar and Heliospheric Physics.

, Pevtsov, A. p. 9
Where the paper says it

Priest, 225 Bertello, L. , Pevtsov, A.

, Petrie, G. p. 9
Where the paper says it

, Pevtsov, A. , Petrie, G.

, & Keys, D. p. 9
Where the paper says it

, Petrie, G. , & Keys, D.

, Solar Physics, 289, 2419 Caplan, R. p. 9
Where the paper says it

, & Keys, D. 2014, Solar Physics, 289, 2419 Caplan, R.

, Downs, C. p. 9
Where the paper says it

2014, Solar Physics, 289, 2419 Caplan, R. , Downs, C.

, & Linker, J. p. 9
Where the paper says it

, Downs, C. , & Linker, J.

, The Astrophysical Journal, 823, 53 DeVore, C. p. 9
Where the paper says it

, & Linker, J. 2016, The Astrophysical Journal, 823, 53 DeVore, C.

R., Sheeley Jr, N. p. 0
Where the paper says it

INTRODUCTION Magnetic fields on the solar surface evolve continuously, shaping the global structure of the corona and heliosphere (N. Sheeley Jr 2005; T.

, & Boris, J. p. 9
Where the paper says it

, Sheeley Jr, N. , & Boris, J.

, Solar physics, 92, 1 Felipe, T. p. 9
Where the paper says it

, & Boris, J. 1984, Solar physics, 92, 1 Felipe, T.

, & Ramos, A. p. 1
Where the paper says it

Felipe & A. Ramos (2019) trained a convolutional network on farside helioseismic holography phase-shift maps, calibrated against delayed Helioseismic and Magnetic Imager (HMI) magnetograms, to produce probability maps of farside active-region presence.

A., Astronomy & Astrophysics, 632, A82 Hang, T. p. 9
Where the paper says it

, & Ramos, A. 2019, Astronomy & Astrophysics, 632, A82 Hang, T.

, Gu, S., Li, C., et al., in Proceedings of the IEEE/CVF international conference on computer vision Ho, J. p. 9
Where the paper says it

2019, Astronomy & Astrophysics, 632, A82 Hang, T. 2023, in Proceedings of the IEEE/CVF international conference on computer vision, 7441–7451 Ho, J.

, Jain, A., & Abbeel, P. p. 9
Where the paper says it

2023, in Proceedings of the IEEE/CVF international conference on computer vision, 7441–7451 Ho, J. , & Abbeel, P.

, Advances in neural information processing systems, 33, 6840 Jarolim, R. p. 9
Where the paper says it

, & Abbeel, P. 2020, Advances in neural information processing systems, 33, 6840 Jarolim, R.

, Veronig, A. p. 9
Where the paper says it

2020, Advances in neural information processing systems, 33, 6840 Jarolim, R. , Veronig, A.

, P¨otzi, W. p. 9
Where the paper says it

, Veronig, A. , P¨otzi, W.

, & Podladchikova, T. p. 9
Where the paper says it

, P¨otzi, W. , & Podladchikova, T.

, Nature Communications, 16, 3157 Jeong, H. p. 9
Where the paper says it

, & Podladchikova, T. 2025, Nature Communications, 16, 3157 Jarolim, R.

-J., Moon, Y.-J., Park, E., & Lee, H. p. 9
Where the paper says it

2024, The Astrophysical Journal Letters, 961, L31 Jeong, H. , & Lee, H.

, The Astrophysical Journal Letters, 903, L25 Jeong, H. p. 9
Where the paper says it

, & Lee, H. 2020, The Astrophysical Journal Letters, 903, L25 Jeong, H.

-J., Moon, Y.-J., Park, E., Lee, H., & Baek, J. p. 9
Where the paper says it

2020, The Astrophysical Journal Letters, 903, L25 Jeong, H. , & Baek, J.

-H., The Astrophysical Journal Supplement Series, 262, 50 Kaiser, M. p. 9
Where the paper says it

, & Baek, J. 2022, The Astrophysical Journal Supplement Series, 262, 50 Jeong, H.

L., Kucera, T. p. 9
Where the paper says it

2025, The Astrophysical Journal Letters, 989, L31 Kaiser, M. , Kucera, T.

, Davila, J. p. 9
Where the paper says it

, Kucera, T. , Davila, J.

, et al., Space Science Reviews, 136, 5 Leka, K. p. 9
Where the paper says it

, Davila, J. 2008, Space Science Reviews, 136, 5 Leka, K.

, & Barnes, G. p. 0
Where the paper says it

Strong, complex fields in active regions (ARs) drive flares and coronal mass ejections (CMEs) that can disturb near-Earth space and impact ground-based and space-borne systems (K. Barnes 2003; S.

, The Astrophysical Journal, 595, 1277 Lemen, J. p. 9
Where the paper says it

, & Barnes, G. 2003, The Astrophysical Journal, 595, 1277 Lemen, J.

R., Title, A. p. 9
Where the paper says it

2003, The Astrophysical Journal, 595, 1277 Lemen, J. , Title, A.

M., Akin, D. J., et al., Solar Physics, 275, 17 Lugmayr, A. p. 9
Where the paper says it

, Title, A. 2012, Solar Physics, 275, 17 Lugmayr, A.

, Danelljan, M. p. 9
Where the paper says it

2012, Solar Physics, 275, 17 Lugmayr, A. , Danelljan, M.

, Romero, A. p. 9
Where the paper says it

, Danelljan, M. , Romero, A.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved AI system can perform full-Sun EUV reconstruction by synthesizing missing solar sectors using Earth-side and STEREO-A views, restoring full-Sun EUV coverage as stated in the abstract.

  2. The system will generate high fidelity reconstructions of coronal structures, such as sharp coronal loop morphology and coherent active region boundaries in the unobserved farside sector, which supports improved forecasting of eruptive events.

  3. It enables continuous global solar monitoring by maintaining cross-limb continuity and providing cleaner inner-boundary conditions for coronal/heliospheric models, complementing helioseismic and SFT estimates.

  4. The system maintains observational integrity by using a post-processing feathered blending step, which ensures that the output is smooth, measurement-consistent mosaics and avoids introducing systematic biases compared to interpolation methods.

  5. The system incorporates domain-specific enhancements like spectral gating and confidence weighting to produce results that are operationally suitable, specifically targeting the reconstruction of missing regions with a mean SSIM of 0.801.

Abstract

The loss of STEREO-B in 2014 created a persistent blind spot in Extreme Ultraviolet (EUV) imaging of the solar farside. We present HelioFill, to the authors' knowledge, the first denoising-diffusion inpainting model that restores full-Sun EUV coverage by synthesizing the STEREO-B sector from Earth-side (SDO) and STEREO-A views. Trained on full-Sun maps from 2011-2014 (when SDO+STEREO-A+B provided 360 degrees coverage), HelioFill couples a latent diffusion backbone with domain-specific additions: spectral gating, confidence weighting, and auxiliary regularizers, to produce operationally suitable 304 Angstrom reconstructions. On held-out data, the model preserves the observed hemisphere with mean SSIM 0.871 and mean PSNR 25.56 dB, while reconstructing the masked hemisphere with mean SSIM 0.801 and mean PSNR 17.41 dB and reducing boundary error by approximately 21 percent (Seam L2) compared to a state-of-the-art diffusion inpainting model. The generated maps maintain cross-limb continuity and coronal morphology (loops, active regions, and coronal-hole boundaries), supporting synoptic products and cleaner inner-boundary conditions for coronal/heliospheric models. By filling observational gaps with observationally consistent EUV emission, HelioFill maintains continuity of full-Sun monitoring and complements helioseismic farside detections, illustrating how diffusion models can extend the effective utility of existing solar imaging assets for space-weather operations.

Sources

Related papers