• DOI:10.1051/0004-6361/201936910
• Corpus ID: 232258067

Machine learning initialization to accelerate Stokes profile inversions

@article{Gafeira2021MachineLI,
  title={Machine learning initialization to accelerate Stokes profile inversions},
  author={Ricardo Gafeira and David Orozco Su'arez and Ivan Mili{\'c} and Carlos Quintero Noda and Basilio Ruiz Cobo and Han Uitenbroek},
  journal={ArXiv},
  year={2021},
  volume={abs/2103.09651}
}

• R. Gafeira, D. O. Su'arez, +3 authors H. Uitenbroek
• Published 16 March 2021
• Environmental Science
• ArXiv

Context. At present, an exponential growth in scientific data from current and upcoming solar observatories is expected. Most of the data consist of high spatial and temporal resolution cubes of Stokes profiles taken in both local thermodynamic equilibrium (LTE) and non-LTE spectral lines. The analysis of such solar observations requires complex inversion codes. Hence, it is necessary to develop new tools to boost the speed and efficiency of inversions and reduce computation times and costs.
Aims. In this work we discuss the application of convolutional neural networks (CNNs) as a tool to advantageously initialize Stokes profile inversions.
Methods. To demonstrate the usefulness of CNNs, we concentrate in this paper on the inversion of LTE Stokes profiles. We use observations taken with the spectropolarimeter on board the Hinode spacecraft as a test bench mark. First, we carefully analyse the data with the SIR inversion code using a given initial atmospheric model. The code provides a set of atmospheric models that reproduce the observations well. These models are then used to train a CNN. Afterwards, the same data are again inverted with SIR but using the trained CNN to provide the initial guess atmospheric models for SIR.
Results. The CNNs allow us to significantly reduce the number of inversion cycles when used to compute initial guess model atmospheres (‘assisted inversions’), therefore decreasing the computational time for LTE inversions by a factor of two to four. CNNs alone are much faster than assisted inversions, but the latter are more robust and accurate. CNNs also help to automatically cluster pixels with similar physical properties, allowing the association with different solar features on the solar surface, which is useful when inverting huge datasets where completely different regimes are present. The advantages and limitations of machine learning techniques for estimating optimum initial atmospheric models for spectral line inversions are discussed. Finally, we describe a python wrapper for the SIR and DeSIRe codes that allows for the easy setup of parallel inversions. The tool implements the assisted inversion method described in this paper. The parallel wrapper can also be used to synthesize Stokes profiles with the RH code.
Conclusions. The assisted inversions can speed up the inversion process, but the efficiency and accuracy of the inversion results depend strongly on the solar scene and the data used for the CNN training. This method (assisted inversions) will not obviate the need for analysing individual events with the utmost care but will provide solar scientists with a much better opportunity to sample large amounts of inverted data, which will undoubtedly broaden the physical discovery space.