Using data science to interpret ground-breaking new observations and simulations of the Sun

Contact person: Tiago Pereira     
Keywords: Sun, stars, machine learning, magnetohydrodynamics, radiative transfer
Research group: Rosseland Centre for Solar Physics
Institute of Theoretical Astrophysics

Understanding the Sun is of crucial importance to understand other stars, the occurrence of life-supporting exoplanets, and predicting powerful storms that can be affect communications and life on Earth. Solar physics is on the cusp of a revolution made possible by a new generation of advanced telescopes and simulations reaching exascales. However, existing methods are no longer sufficient to make the most out of these new tools. The latest telescopes generate highly-dimensional data at volumes larger than 3 TB per hour, and 3D MHD simulations are too large to analyse in a single workstation. We are interested in ground-breaking approaches making use of machine learning and data mining techniques to enable a meaningful analysis of state-of-the-art solar observations and simulations. We invite research proposals that may combine traditional and machine learning methods to work in the topics listed below (or closely related).

Methodological research topics:

  • Data mining: spectral clustering at 5-10 dimensions
  • Data mining: visualisation of highly dimensional data sets
  • Data denoising: fast processing of images to remove atmospheric perturbations
  • Super-resolution spectral imaging: enhance datasets to transcend the diffraction limit of one or more sources
  • Deep learning: accelerate MHD simulations
  • Deep learning: accelerate 3D radiative transfer and ray tracing 

Topics in solar physics:

  • Understanding the formation of spectral lines in 3D simulations via spectral clustering
  • Near real time image deconvolution of ground-based observations
  • Combine and augment spectral data from multiple telescopes
  • Enable 3D non-LTE spectral synthesis of large datasets to compare observations and simulations
  • Use deep learning to predict surface magnetic field configurations for 3D MHD simulations
  • Use deep learning to accelerate 3D radiative transfer in MHD simulations"    

External partners:

  • Instituto de Astrofísica de Canarias
  • University of Glasgow
  • Lockheed Martin Solar and Astrophysics Laboratory (LMSAL)