Accurately constraining velocity information from spectral imaging observations using machine learning techniques

Abstract

Determining accurate plasma Doppler (line-of-sight) velocities from spectroscopic measurements is a challenging endeavour, especially when weak chromospheric absorption lines are often rapidly evolving and, hence, contain multiple spectral components in their constituent line profiles. Here, we present a novel method that employs machine learning techniques to identify the underlying components present within observed spectral lines, before subsequently constraining the constituent profiles through single or multiple Voigt fits. Our method allows active and quiescent components present in spectra to be identified and isolated for subsequent study. Lastly, we employ a Ca ɪɪ 8542 Å spectral imaging dataset as a proof-of-concept study to benchmark the suitability of our code for extracting two-component atmospheric profiles that are commonly present in sunspot chromospheres. Minimization tests are employed to validate the reliability of the results, achieving median reduced χ 2 -values equal to 1.03 between the observed and synthesized umbral line profiles. This article is part of the Theo Murphy meeting issue ‘High-resolution wave dynamics in the lower solar atmosphere’.

Spectropolarimetric fluctuations in a sunspot chromosphere

“…In order to ensure the reliability of the results, here we restrict our attention to the intensity and CP fluctuations only. This choice is also motivated by the fact that since the Ca ɪɪ 854.2 nm spectral line often goes into emission, the estimation of the Doppler velocity may lead to inconsistencies [64]. Inversion techniques may provide a solid context for the reliable estimation of the Doppler velocity even in these conditions, but such a technique will be addressed and explored in future work on the subject.…”

MCALF: Multi-Component Atmospheric Line Fitting

“…This dataset typically contains spectra with multiple atmospheric components and this package supports the isolation of the individual components such that velocity information can be constrained for each component. The method implemented in this IBIS model has been discussed extensively in MacBride et al (2020). There are also several ongoing research projects using this model to extract velocity measurements.…”

High-resolution wave dynamics in the lower solar atmosphere

“…The interplay between source functions and opacity effects often requires the fitment of two (or more) profiles to any given spectral line in order to best constrain the underlying dynamics. This is a time consuming endeavour to perform manually, prompting MacBride et al [47] to apply machine learning techniques to this challenging problem to accurately, efficiently and repeatedly extract the key components of the observed line profiles. The authors employed a training dataset based on IBIS Ca ɪɪ 8542 Å sunspot observations, and when applied to high-cadence spectral scans it was able to accurately fit over 600 000 spectral profiles in approximately 2 h, with the potential to further parallelize via multiple CPU cores and/or the implementation of GPUs.…”

“…The authors employed a training dataset based on IBIS Ca ɪɪ 8542 Å sunspot observations, and when applied to high-cadence spectral scans it was able to accurately fit over 600 000 spectral profiles in approximately 2 h, with the potential to further parallelize via multiple CPU cores and/or the implementation of GPUs. MacBride et al [47] highlight the potential of their techniques to be applied to upcoming observations from the National Science Foundation’s Daniel K. Inouye Solar Telescope (DKIST; [9]) in real time, thus providing important wave diagnostics live at the observing site to help pinpoint regions of interest for specific wave studies.…”