Gaia DR2 known issues - Gaia
Known issues with the Gaia DR2 data
This page lists the issues in our second data release that have been discovered after the release of the Gaia data and related documentation. The Gaia DR2 contents page contains a summary of limitations that were known, and documented, already at the release date. Tips on how to better make use of the Gaia Archive can be found here.
- Astrometry: 2- versus 5-parameter solutions
- Astrometry: Considerations for the use of DR2 astrometry
- Astrometry: Systematic effects in Gaia DR2 parallaxes for very bright stars
- Crossmatch: Hipparcos2
- Radial Velocities: Potential contamination in crowded regions
- Photometry: Systematic effects and response curves
Gaia DR2 provides astrometry for 1.69 billion objects. The vast majority of them, 1.33 billion sources, have 5-parameter solutions, i.e., in addition to right ascension and declination, also the proper motion (in right ascension and declination) and the parallax are given. The remaining 361 million objects have 2-parameter solutions (right ascension and declination only). The criterion for receiving either a 2- or a 5-parameter solution has been based on the quality of the astrometric and photometric data, as set out in Equations (11) and (12) in Lindegren et al. (2018).
The decision between these 2- and 5-parameter solutions was based on preliminary (non-published) photometry used early in the Gaia DR2 data processing. During the final stages of the photometric data processing, the so called 'bronze' sources (see Riello et al. (2018) and Evans et al. (2018) for explanation) received updated photometry, which has been published as part of Gaia DR2. However, the astrometric selection between 2- and 5-parameter solutions was not re-done using these new data. As a consequence, the published photometry and the astrometric selection criteria between 2- and 5-parameter solutions are inconsistent in some 9 million cases.
Of these approximately 9 million cases, there are some 7 million sources in Gaia Data Release 2 with a 2-parameter solution while the criteria for a 5-parameter solution are met. In addition, there are some 2 million sources with a 5-parameter solution while a 2-parameter solution would be expected based on the documented criteria and published data.
An overview of Gaia DR2 astrometry was presented by Lindegren et al. at the IAU GA Division A meeting in Vienna on 27 August 2018. Topics covered include random and systematic errors, quality indicators, and spurious parallaxes.
An extended version of the presentation, containing detailed tips and recommendations for the best uses of the DR2 astrometry, is given here. Among other things it contains recipes for estimating the systematic uncertainty of the mean parallax or proper motion of a cluster, and for calculating the re-normalised unit weight error (RUWE).
The recipes use auxiliary data: a table of spatial covariances (DR2_spatialCov_V1.zip), and tables for calculating the RUWE (DR2_RUWE_V1.zip). Each zipped directory comes with a readme.txt file. Further information on the re-normalised unit weight error is in GAIA-C3-TN-LU-LL-124-01 available from the DPAC Public Documents page.
When using the data given in "DR2_spatialCov_V1.zip" we ask you to follow our Gaia Data Release 2 credit and citation guidelines and cite the "Gaia Data Release 2: the astrometric solution" paper. When using the data given in "DR2_RUWE_V1.zip" we ask you to cite the technical note GAIA-C3-TN-LU-LL-124-01.
Following a decision made by the Gaia Data Processing and Analysis Consortium and ESA, an official release of the RUWE data is available from the Gaia Archive along with the Gaia DR2 data. To ensure consistency of the RUWE data provided through ESA/Gaia/DPAC, also Gaia Partner Data Centres will be serving the same set of RUWE data. In cases where Gaia Partner Data Centres served RUWE values before the official release, a comparison was performed between initially provided RUWE values and the official release, showing a reasonably small difference.
When computing the RUWE data for distribution through the Gaia Archive and Gaia partner data centres, the choice was made to not provide RUWE values for two-parameter solutions. Currently the RUWE data can be found in a seperate table, but the RUWE data is expected to be provided in the gaia_source table from EDR3 onwards.
RUWE data is now available from the Gaia Archive
RUWE data is described in the Gaia Data Release Documentation
RUWE data is now part of the Gaia Data Release 2 and when using this data we ask you to follow the credit and citation instructions for Gaia Data Release 2.
In addition to larger uncertainties (as described in the topic "Astrometry: Considerations for the use of DR2 astrometry"), care should be taken when using the parallaxes of very bright (i.e. G < 5) stars, as they may have additional systematic errors due to calibration issues. A description of this topic can be found in this research note.
While we expect to find Gaia counterparts for most of the Hipparcos2 sources (van Leeuwen F., 2007, A&A 474, 653), with the exception of the brightest ones, the cross-match results include only about 2/3 of them. This means that according to the adopted cross-match algorithm only ∼2/3 of Hipparcos2 objects have a Gaia counterpart compatible within position errors (i.e. have at least one good neighbour). This issue is discussed in Section 6.1 of Marrese et al. 2018.
We provide in the file "Hipparcos2GaiaDR2coneSearch.zip" the results of a 1 arcsec cone search described in the paper. The table contains three columns: the Gaia and Hipparcos2 identifiers and the angular distance (in arcsec) for each nearest neighbour.
When using the data given in "Hipparcos2GaiaDR2coneSearch.zip” we ask you to follow our Gaia Data Release 2 credit and citation guidelines and cite the "Gaia Data Release 2: Cross-match with external catalogues - Algorithms and results” paper.
Boubert et al. (2019) obtained spectroscopic follow-up of Gaia DR2 5932173855446728064 and found a very different median radial velocity of -56.5 ± 5.3 km/s compared to the Gaia DR2 value of -614.3 ± 2.5 km/s.
Using the Gaia Observation Forecasting Tool (GOST), these authors speculate that Gaia DR2 5932173855446728064’s spectra are dominated by the light from a brighter star 4.3 arcsec away, and that due to the slitless, time delay integration of Gaia’s Radial Velocity Spectrometer (RVS), this angular offset corresponds to a spurious 620 km/s shift in the calcium triplet of the second star. The RVS team confirms this to be the case using unpublished Gaia DR2 information.
The Gaia DR2 RVS pipeline processed rectangular RVS windows only but these could be truncated as well as non-truncated. A window can be truncated in the across scan (AC) direction but still be rectangular if the truncation is the same at each along scan (AL) sample. This occurs when two sources are aligned in AL by less than 6.4 arcsec. The Gaia DR2 RVS pipeline let through rectangular truncation because many bright sources were truncated by windows containing negligible flux from spurious sources associated with the brighter source (Sartoretti et al. 2018).
An unintended consequence of this is letting through rectangular truncation due to two different sources, which can result in contamination like in the case of Gaia DR2 5932173855446728064. This is most likely to occur in crowded regions (Gaia DR2 5932173855446728064 has Galactic latitude b = -2.7 degrees).
Using GOST to assess the potential contamination of each Gaia DR2 source with a radial velocity (7.2 million) is not feasible. Boubert et al. (2019) cut Gaia DR2 sources with a radial velocity that have a companion in the full Gaia DR2 catalogue within 6.4 arcsec that either itself has a radial velocity or that is brighter in GRP or G magnitudes. The resulting list of 70,365 Gaia DR2 sources with potentially contaminated radial velocities are available here (CSV-file download here). The RVS team is investigating the issue further in order to improve the Gaia RVS pipeline for future releases.
The Gaia Data Release 2 photometry is affected by some systematic errors, as already mentioned by Evans et al. (2018) and Arenou et al. (2018). These effects include complex features for faint sources (G greater than about 17), likely caused by problems in the background calibration and contamination from nearby sources. For bright sources (G less than about 6), saturation causes systematic dependencies for which Evans et al. (2018) provide an empirical correction.
A systematic trend with magnitude has been detected in the G band at magnitudes brighter than about 16.5 by comparing Gaia DR2 data with synthetic photometry of CALSPEC sources (Casagrande & VandenBerg 2018, Weiler 2018). This trend is approximately linear. Maíz Apellániz & Weiler (2018), based on their own high-quality spectral library of 122 stars including CALSPEC, have proposed a linear correction of 3.2 +/- 0.3 mmag/mag over the interval 6 < G < 16 to adjust the Gaia G flux scale. However the trend could be more complex than this. Users should be careful that this correction is only valid in this magnitude range and should not be extrapolated, i.e. the value at G = 16 should be used for sources fainter than this.
Furthermore, a small systematic inconsistency in the BP photometric system has been spotted (Weiler 2018, Maíz Apellániz & Weiler 2018). This inconsistency is likely caused by insufficient convergence of the BP calibration for sources brighter than about G = 10.87. The inconsistency depends on colour, being more significant for sources bluer than about BP-RP = 0.5. Maíz Apellániz & Weiler (2018) mitigate this effect by providing two different BP response curves for the magnitude ranges brighter and fainter than G = 10.87.
As already mentioned in Evans et al. (2018), photometric passbands are not fully constrained by any given set of calibration sources (see Weiler 2018 for a detailed discussion). Synthetic photometry derived with such response curves for sources whose spectral energy distributions are not well represented by those calibration sources may be inaccurate.
Maíz Apellániz & Weiler (2018) have presented an alternative set of response curves for the Gaia DR2 photometric system, together with the corresponding zero points. These response curves are based on a library of high-quality spectra that include additional types of spectral energy distributions as compared to the Spectro-Photometric Standard Stars set used by the DPAC calibration (Pancino et al. 2012). These response curves, which must be used with their own zero points, need to be coupled with the suggested modification of the G flux scale. The combination of the suggested corrections and their set of response curves, by mitigating inconsistencies in the published photometry, allows the user to obtain more accurate synthetic photometry and is a valuable attempt to obtain the best possible results from the Gaia DR2 data.
Recipe for the use of Gaia DR2 photometry with the passbands by Maíz ApellÁniz & Weiler (2018):
- 2 < G ≤ 6: compute corrected magnitudes as Gcorr = -0.047344 + 1.16405*G - 0.046799*G2 + 0.0035015*G3 (following the saturation correction detailed in Evans et al. 2018, Appendix B)
- 6 < G ≤ 16: compute corrected magnitudes as Gcorr = G - 0.0032*(G-6) (following Maíz Apellániz & Weiler 2018)
- G > 16: compute corrected magnitudes as Gcorr = G - 0.032
The G referred to above is the G magnitude available via the Gaia DR2 archive.
mag = -2.5* log10(synthetic_flux)+synthetic_ZP
if synthetic_flux is expressed in photons per second per square metre, the values of the synthetic_ZP are:
- 26.3914 for G,
- 26.1297 for GBP (bright),
- 26.1698 for GBP (faint), and
- 25.3255 for GRP.
Note that these zeropoints are different to the ones used in Gaia DR2 and should only be used with synthetic fluxes derived using the Maíz Apellániz & Weiler (2018) passbands. They should not be used with the Gaia DR2 fluxes available from the Gaia archive. There will be small inconsistencies at about the 0.01 mag level between Gaia DR2 magnitudes and those generated using the Maíz Apellániz & Weiler (2018) passbands. This is due to the Gaia DR2 zeropoints and the Maíz Apellániz & Weiler (2018) passbands having been calculated in a different way.