Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AI Foundation Model for Heliophysics: Applications, Design, and Implementation

Sujit Roy, Talwinder Singh, Marcus Freitag, Johannes Schmude, Rohit Lal, Dinesha Hegde, Soumya Ranjan, Amy Lin, Vishal Gaur, Etienne Eben Vos, Rinki Ghosal, Badri Narayana Patro, Berkay Aydin, Nikolai Pogorelov, Juan Bernabe Moreno, Manil Maskey, Rahul Ramachandran

TL;DR

This paper provides a perspective on the criteria for designing an FM for heliophysics and associated challenges and applications using the Solar Dynamics Observatory (SDO) dataset and believes that this is the first study to design an FM in the domain of heliophysics.

Abstract

Deep learning-based methods have been widely researched in the areas of language and vision, demonstrating their capacity to understand long sequences of data and their usefulness in numerous helio-physics applications. Foundation models (FMs), which are pre-trained on a large-scale datasets, form the basis for a variety of downstream tasks. These models, especially those based on transformers in vision and language, show exceptional potential for adapting to a wide range of downstream applications. In this paper, we provide our perspective on the criteria for designing an FM for heliophysics and associated challenges and applications using the Solar Dynamics Observatory (SDO) dataset. We believe that this is the first study to design an FM in the domain of heliophysics.

AI Foundation Model for Heliophysics: Applications, Design, and Implementation

TL;DR

This paper provides a perspective on the criteria for designing an FM for heliophysics and associated challenges and applications using the Solar Dynamics Observatory (SDO) dataset and believes that this is the first study to design an FM in the domain of heliophysics.

Abstract

Deep learning-based methods have been widely researched in the areas of language and vision, demonstrating their capacity to understand long sequences of data and their usefulness in numerous helio-physics applications. Foundation models (FMs), which are pre-trained on a large-scale datasets, form the basis for a variety of downstream tasks. These models, especially those based on transformers in vision and language, show exceptional potential for adapting to a wide range of downstream applications. In this paper, we provide our perspective on the criteria for designing an FM for heliophysics and associated challenges and applications using the Solar Dynamics Observatory (SDO) dataset. We believe that this is the first study to design an FM in the domain of heliophysics.

Paper Structure

This paper contains 17 sections, 7 equations, 12 figures.

Table of Contents

  1. Introduction
  2. Applications & State-of-the-Art Machine Learning Models
  3. Data and Preparation
  4. AIA
  5. HMI
  6. Temporal Alignment of HMI and AIA Data
  7. Dataset Statistics
  8. Methodology
  9. Sampling strategy
  10. Data Normalization
  11. Training objectives
  12. Vision Transformers
  13. Long-Short Spectral Transformer
  14. Diffusion Models
  15. Graph methods
  16. ...and 2 more sections

Figures (12)

  • Figure 1: Data captured by HMI and AIA instruments aboard SDO on 2012-01-30 at 22:12 UT. The top left panel shows the line-of-sight magnetogram from HMI and the rest of the panels show Sun in the seven EUV wavelengths from AIA. The wavelengths in $\textup{\AA}$ are mentioned in the bottom right corner of each panel.
  • Figure 2: Visualization of selected downstream tasks with their spatial and temporal scales
  • Figure 3: An example of the ML-ready data preparation steps for AIA 171 Å and HMI LOS magnetogram on 2012-01-30 at 22:12 UT. Contours illustrate the image center, solar disk center, disk radius, and solar disk boundary. The top row shows the original AIA Level 1 image, HMI Level 1.5 magnetogram downloaded from JSOC, and HMI overlaid on AIA. The disk centers are misaligned with the image center (unregistered), and one dataset has a 180° roll, with noticeable plate scale differences. The middle row displays the registered AIA Level 1.5 image, HMI aligned with AIA, and HMI overlaid on AIA, showing corrected disk centers and plate scales. The bottom row presents the final ML-ready AIA and HMI images after exposure time normalization and orbital corrections for AIA, with the overlaid image showing proper alignment and a fixed disk radius of 976 arcsecs.
  • Figure 4: Mean of full-disk AIA images per second in seven wavelengths over time for before (left panel) and after (right panel) the degradation correction.
  • Figure 5: Bad AIA measurements due to a variety of reasons.
  • ...and 7 more figures