Diffusion Modeling of the Solar Corona
Description
Diffusion models such as DeepMind’s GenCast have demonstrated powerful performance in terrestrial weather forecasting, achieving results on-par and surpassing leading medium-range numerical weather simulations. We present our initial results to train a Denoising Diffusion Probabilistic Model (DDPM) of the solar corona, aimed as a first step at generating synthetic solar magnetic fields based on conditioning inputs. Conditioning inputs include multi-spectral imagery, magnetograms, and other measurements commonly used to frame coronal inverse problems. Our initial experiment targets the generation of synthetic global magnetic field configurations of the solar coronal basedon conditioning with the solar cycle phase. The model is trained on 10 years of WSA potential field source surface(PFSS) model runs, augmented with 15° to 345° azimuthal rotations to increase data diversity. Training is conducted inthe spherical harmonics domain, leveraging concepts from Fourier Neural Operators (FNOs) and Spherical Fourier Neural Operators (SFNOs). A physics-informed loss function, built on a differentiable spherical harmonic expansion, isused to maximize generation of realistic 3D magnetic potentials.
Detailed example
3D Magnetic Field Structures in the Solar Corona
AI / analytics pattern
Generative AI: AI that generates new or synthetic content (e.g., images, videos, audio, text, code).
Automation level / stage
a) Pre-deployment – The use case is in a development or acquisition status.
Expected benefit
First step towards probabilistic modeling of the solar corona
Controls / human review
ATO: Not reported; PIA: Not published