====== Challenge 2: Reference stars ====== 

===== Objectives =====

  * Employ high-resolution spectra from X-shooter archive to investigate how the Gaia spectra will look like.
  * Investigate GSP-Phot performance on X-shooter spectra.

===== Comparison to RVS and BP/RP spectra =====

  * Gaia RVS spectra and corresponding segments of x-shooter spectra look very similar. Ca-Triplet lines are easily identified and other prominent lines are also in obvious correspondence.
  * x-shooter spectra contain telluric bands around 760nm and 950nm. In fact, BP/RP shows prominent dips there, which are NOT reflected by space-based instrument models. That must be a problem when modelling those spectra with GSP-Phot.

{{:tests:compare-spectra.png|}}



===== GSP-Phot performance on x-shooter spectra =====

  * Paula compiled mean parameters for all stars from Pastel. (These are no "true" values but mean literature values.)
  * Paula's data contains a lot of giants. GSP-Phot lacks giants in its training data, i.e., it is unlikely to perform well on giants (c.f. [[:tests:astropars:challenge4|Challenge 4]]).
  * GSP-Phot performs very poorly on x-shooter spectra: Teff is loosely related for some stars, but logg looks rather random and GSP-Phot tends to assign too low [Fe/H].

{{:compare-pastel-masked-tellurics.png|}}

== Possible explanations for poor GSP-Phot performance ==

  * x-shooter spectra are ground-based spectra, including telluric features, that are not part of the Gaia spectra.
  * Lack of giants in GSP-Phot training data.
  * x-shooter spectra were processed with GOG version 14, whereas GSP-Phot training data was processed with GOG version 8.



===== Andy's SICK performance on x-shooter spectra =====