skip to main content

gaia data release 3 documentation

11.2 Properties of the input data

11.2.4 Use of RVS spectra in CU8

Author(s): Yves Frémat

The RVS spectra processed by the Apsis pipeline have been averaged for each source over all valid CCDs and transits, by CU6’s MTA (Multiple Transit Analysis) pipeline. Before combining the data, MTA places all the CCD spectra in the target’s rest frame, then they are rebinned to a wavelength scale that ranges from 846 to 870 nm, with a bin size of 0.01 nm. The normalisation at the local (pseudo-)continuum is performed on the CCD spectra. Due to the presence of molecular bands in M stars, and the difficulty to locate the continuum for those stars, all the spectra belonging to targets having Teff < 3500 K (the Teff value adopted is the same as the one used by CU6 to select the RV template) are normalized to their median flux rather than to their (pseudo-)continuum. Because of the radial velocity correction applied to put all the spectra in rest frame, or the cosmic clipping etc., some flux values may be missing (especially at the edges) and replaced by NaNs. The data is provided to CU8 in the same form as the one adopted for DR3 publication. A more detailed description is provided in the corresponding CU6 section of the online documentation (Section 6.4.9) as well as in Seabroke et al. (2022).

The Apsis modules that use RVS spectra are GSP-Spec, ESP-CS, and ESP-HS. The data are usually normalized to the continuum and therefore all possible issues with the RVS passband filter response and/or flux calibration can be ignored at first. Hence, the key information for a proper comparison with synthetic spectra remains the spectroscopic LSF (i.e. Along-scan line-spread function). In practice, this function varies with time and with the light path (e.g. on the CCD strip and on the field of view). For this reason, the co-added data (used by CU8) produced by CU6 are a mixture of spectra broadened with different LSFs. On the other hand, the information available that would allow to reconstruct the instrument profile in detail for CU8 purposes is technically difficult to use in the Apsis framework. For the DR3 processing, we assumed the ”co-added” instrument broadening function to be Gaussian by applying the GenerateTemplate tool used by CU6 to construct the templates, and we adopted a median resolving power (R=11 500, Cropper et al. 2018) equal to the one measured on single transit/CCD spectra. This median value may however also be expected to vary from target to target, depending on the CCD/transit LSF mixture, or on the accuracy of the RV measurements used to put all the spectra in rest frame before co-addition.

During Cycle 3, a particular effort was made by CU6 to deblend, when needed, the transit spectra (Seabroke et al. 2022). Hence, a significant fraction of the data results from the combination of de-blended and non-blended spectra. While some spectra might be harder to deblend and introduce errors, the global impact of using these combinations has been proven to be significantly positive for the purpose of CU8’s processing.

Although RVS CU8 users re-process the data in a way adapted to their needs, they all perform multiple iteration steps of re-normalisation to the synthetic spectra. In addition, both GSP-Spec and ESP-HS are rebinning the data by a factor of 3 (Section 11.3.8, Gaia Collaboration et al. 2023h).

Finally, in addition to the co-added spectra, other CU6 data are used:

  • Radial velocity estimates and uncertainties are used by GSP-Spec to assess the quality of the co-added spectrum and construct the processing flags

  • vsini (i.e. vbroad, Frémat et al. 2023): FGK stars are usually expected to have a low vsini and, depending on the assumptions made, a higher vsini value may limit the accuracy of the derived parameters. GSP-Spec also uses the CU6 estimates to assess the quality and pertinence of the results. In ESP-CS, the value is taken into account before extracting the activity index (Section 11.3.9).

Synthetic spectra libraries

All CU8 RVS users adopt methods based on the comparison of the observed to theoretical data. A list of the synthetic spectral libraries, with the number of spectra (col.2) as well as the AP coverage (cols.3-5) and module consumer (col.6), is provided in Table 11.14. The atmosphere models used to make the A and OB libraries have the same origin as those used to simulate the BP/RP spectra (Section 11.2.3), the three other grids of spectra are module specific (Section 11.3.9, Section 11.3.8, Gaia Collaboration et al. 2023h).

Table 11.14: CU8 synthetic stellar spectra libraries used to analyse the RVS data.
Name Nb. Teff [K] logg [Fe/H] Consumer
SYNSPEC/ATLAS* 229 07000 – 20 000 -0.50 – 5.00 +0.00 ESP-HS
OB 216 15 000 – 55 000 -1.75 – 4.75 +0.00 ESP-HS
A 2332 06000 – 16 000 -2.50 – 4.50 -1.50 – +0.50 ESP-HS
MARCS/BSYN* 125 03000 – 07000 -3.00 – 5.00 -0.50 – +0.75 ESP-CS
GSP-spec Grid* 51 373 02750 – 08000 -0.50 – 5.50 -5.00 – +1.00 GSP-Spec
Notes. *Module specific grids that are not part of the DR3 StarNormal libraries used by CU6 to derive RVs.

Known issues

Spectra of O, B, and A-type stars:

During the validation runs (Blomme et al. 2023), CU6 reported issues with the radial velocity determination of hot stars which is, at least partially, linked with observations-to-template mismatches (e.g. due to the adopted APs or/and synthetic spectra). This issue impacts the AP determination of stars hotter than 7500 K. Before its main processing, ESP-HS therefore estimates the RV “residual” of the star, and puts the spectrum in the “true” rest-frame. For the cooler stars, GSP-Spec is performing a similar step of RV determination then flags or discards the obtained astrophysical parameters depending on the value that is obtained.

Mixed LSF:

We assumed the “co-added” instrument broadening function to be Gaussian by adopting a median resolving power as discussed in Section 11.2.4 . However this assumption may be inaccurate. To mitigate its impact and assess the correctness of the assumptions, the GSP-Spec module (Gaia Collaboration et al. 2023h) checks the value of line broadening (vbroad) and the accuracy of the radial velocity, and, where needed, either flags or discards the resulting astrophysical parameters. On the other hand, ESP-HS re-derives the vsini (Section 11.3.8), which then aggregates the impact not only of stellar rotation but also of all additional, astrophysical or instrumental, line broadening mechanisms. This issue does not directly impact the measurement of the stellar activity index by ESP-CS (Section 11.3.9).