Stellar activity correction using PCA decomposition of shells

Abstract

Context. Stellar activity and instrumental signals are the main limitations to the detection of Earth-like planets using the radialvelocity (RV) technique. Recent studies show that the key to mitigating those perturbing effects might reside in analysing the spectra themselves, rather than the RV time series and a few activity proxies. Aims. The goal of this paper is to demonstrate that we can reach further improvement in RV precision by performing a principal component analysis (PCA) decomposition of the shell time series, with the shell as the projection of a spectrum onto the spacenormalised flux versus flux gradient. Methods. By performing a PCA decomposition of shell time series, it is possible to obtain a basis of first-order spectral variations that are not related to Keplerian motion. The time coefficients associated with this basis can then be used to correct for non-Dopplerian signatures in RVs. Results. We applied this new method on the YARARA post-processed spectra time series of HD10700 (τ Ceti) and HD128621 (α Cen B). On HD10700, we demonstrate, thanks to planetary signal injections, that this new approach can successfully disentangle real Dopplerian signals from instrumental systematics. The application of this new methodology on HD128621 shows that the strong stellar activity signal seen at the stellar rotational period and one-year aliases becomes insignificant in a periodogram analysis. The RV root mean square on the five-year data is reduced from 2.44 m s −1 down to 1.73 m s −1 . This new approach allows us to strongly mitigate stellar activity, however, noise injections tests indicate that rather high signal-to-noise ratio (S/N> 250) is required to correct for the observed activity signal on HD128621.

Statistical Methods for Exoplanet Detection with Radial Velocities

“…A second approach, template matching, consists in measuring the RV based on a template or model spectrum and a Taylor expansion for the spectrum as a function of velocity (Connes 1985, Bouchy et al 2001, Anglada-Escudé & Tuomi 2012, Astudillo-Defru et al 2015, Jones et al 2022. Related work has been done by Cretignier et al (2022). This approach is based on the approximation…”

“…It is unclear whether the greater flexibility of neural networks will outweight their added complexity and difficulty of interpretation relative to linear regression or Scalpels, particularly given the limited size of datasets available for stars other than the Sun. Cretignier et al (2022) transform the spectra into "shells" instead of CCFs, in an effort to reduce the amount of information lost when averaging lines of different depths. The PCA scores for the shell are used as spectral indicators, but are first orthonormalised with respect to the shell corresponding to a pure Doppler shift.…”

Impact of stellar variability on exoplanet detectability and characterisation

“…• Use of activity indicators computed from the same spectra than the RVs, for example correlation with the bisector span [38], line depth [39], or chromospheric emission [40][41][42], gaussian processes based on those indicators [43][44][45][46], PCA analysis [47], Doppler imaging techniques [48], CCF shape [49], or magnitude-squared coherence comparison [50]. • More complex computation of RVs, for example of subsets of spectral lines to produce independent RV time series [51], use of selected lines with different sensitivity to magnetic field [52][53][54], PCA analysis of the spectra [55], and wavelength dependence of the signal [56]. • Use of external activity indicators, mostly photometry, as in the ff' method proposed by [57].…”

ABORAS: polarimetric, 10cm/s RV observations of the Sun as a star

“…3 A SNR of 300 is the recommended SNR to ensure that spectral line asymmetries can be well studied. 24 The signed magnetic flux of a quiet solar-type star varies within 10G, so we require to measure Integration time and SNR 5 minutes, SNR of >300 at 5500 Å and of >100 at 4000 Å.…”