Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HelioFill: Diffusion-Based Model for EUV Reconstruction of the Solar Farside

Firas Ben Ameur, Rayan Dhib, Yahia Battach, Andrea Lani, Matteo Parsani, Omar Knio, Stefaan Poedts

TL;DR

HelioFill introduces a domain-specific latent-diffusion inpainting framework to reconstruct the solar farside EUV view lost with STEREO-B. By mapping 512×1024 CEA maps through a frozen VAE into a latent 64×128 grid and applying a conditioned UNet with spectral gating, the model fills farside regions from Earth-side and STEREO-A data while preserving observed data. It achieves higher Hole PSNR and SSIM and substantially reduces seam boundary errors, producing physically consistent full-Sun EUV maps suitable for synoptic and space-weather applications. This approach demonstrates how diffusion-based generative models, when tailored to solar optics and observational geometry, can extend the utility of existing imaging assets and improve operational monitoring of coronal evolution and space weather.

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.

HelioFill: Diffusion-Based Model for EUV Reconstruction of the Solar Farside

TL;DR

HelioFill introduces a domain-specific latent-diffusion inpainting framework to reconstruct the solar farside EUV view lost with STEREO-B. By mapping 512×1024 CEA maps through a frozen VAE into a latent 64×128 grid and applying a conditioned UNet with spectral gating, the model fills farside regions from Earth-side and STEREO-A data while preserving observed data. It achieves higher Hole PSNR and SSIM and substantially reduces seam boundary errors, producing physically consistent full-Sun EUV maps suitable for synoptic and space-weather applications. This approach demonstrates how diffusion-based generative models, when tailored to solar optics and observational geometry, can extend the utility of existing imaging assets and improve operational monitoring of coronal evolution and space weather.

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.
Paper Structure (18 sections, 16 equations, 3 figures, 2 tables)

This paper contains 18 sections, 16 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Data
  3. Dataset Description
  4. Synthetic Data Generation
  5. Model
  6. overview
  7. Mask-aware standardization and VAE interface.
  8. Diffusion forward process and denoiser input
  9. Conditioned UNet with spectral gating
  10. Losses
  11. Auxiliary regularizers
  12. Inference and feathered composition
  13. Training and Evaluation Protocol
  14. Results
  15. Experimental design
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Illustration of the dataset construction workflow. Left column: raw 304 Å observations from SDO/AIA and STEREO/EUVI-A/B. Middle column: preprocessed images after calibration, disk clipping, intensity rescaling, and normalization. Right column: CAE synchronic maps. The upper map shows the input configuration with the STEREO-B coverage gap masked (used as model input), while the lower map shows the corresponding full three-view mosaic serving as the target ground truth.
  • Figure 2: HelioFill architecture. A 9-channel UNet input concatenates the noised latent $z_t$, the latent mask $m_\ell$, and the masked latent $z_{\mathrm{in}}$. Token embeddings condition the UNet. Three spectral-gating sites (down_1, mid, up_1) apply FFT-based residual enhancement. The VAE decodes to $\hat{x}$, and a compose step preserves known pixels.
  • Figure 3: Example inpainting on a full-Sun CEA 304 Å map. Left: masked input (black region = missing data). Middle: Output (Prediction) produced by HelioFill. Right: Ground truth. All panels use the same color map and dynamic range; no intensity renormalization between panels.