Aug 17 – 21, 2026
National Institute for Space Research, São José dos Campos, SP, Brazil
America/Sao_Paulo timezone

Estimating the photospheric magnetic field using Physics-Informed Neural Networks (PINNs) with Hinode/SP data

Aug 18, 2026, 4:50 PM
20m
Fernando de Mendonça - LIT (National Institute for Space Research, São José dos Campos, SP, Brazil)

Fernando de Mendonça - LIT

National Institute for Space Research, São José dos Campos, SP, Brazil

Av. dos Astronautas, 1758 - Jardim da Granja, São José dos Campos - SP, 12227-010
Oral Heliophysics & Space Weather Oral Contributions

Speaker

José Matheus Rocha (INPE)

Description

Understanding the solar photospheric magnetic field is vital for space weather forecasting. While Milne-Eddington (ME) inversion of spectropolarimetric data (Stokes I, Q, U, V for Fe I lines) is the standard for retrieving atmospheric parameters, it is computationally expensive, often taking hours per active region map on standard computers. To overcome this bottleneck, we apply a Physics-Informed Neural Network (PINN) to estimate the solar magnetic field from high-resolution observations. Our architecture embeds analytical ME radiative transfer equations into a Multilayer Perceptron (MLP) as physical constraints, ensuring predictions are physically grounded rather than strictly data-driven. Evaluations demonstrate that the PINN-generated synthetic Stokes profiles match observational data with high accuracy and noise resilience. This approach drastically reduces computational time, offering a robust and efficient alternative to classical inversions for rapid, large-scale solar data analysis.

Author

Co-authors

Presentation materials

There are no materials yet.