3.5.4 Known issues in cyclic pre-processing

The image parameter determination is poor for SM windows containing bright sources. The success rate of this process falls from approximately 100% at $G=12$ mag to 0% for $G\lesssim 8$ mag. This is caused by the use of a sub-window of fixed size which needs to be expanded when many of the samples contained are saturated. The initial flux estimate is also too far from its true value when many samples are saturated, leading to IPD failure. Finally, when a source is bright enough to saturate the core the IPD can lock on to the first along-scan diffraction peak.

The PSF model for long gate calibrations, in particular NOGATE but also GATE12 and GATE11 to lesser degrees, is affected by a significant systematic error that has a strong variation with AC rate. Extensive investigations carried out since the calibration pipeline was executed have revealed this to be caused largely by a missing dependence in the PSF model, that of the along scan rate. A detailed explanation is given in section 6.1 of Rowell et al. (2021). This systematic error affects the measurements of source location and flux, and thus potentially affects both the astrometry and photometry over the magnitude range covered by the gates, which is $11.5\lesssim G\lesssim 13$.

The module that identifies spurious detections analysing the image profiles has been reviewed in order to avoid rejecting detections of close pairs (separation below 400 mas), as described in Section 3.4.13. In Cycle-02 processing, there were indications that this module was picking out valid detections of close objects when more than one peak was included in a single window and any of the peaks was well centred. Under these conditions, the module may classify incorrectly some detections affecting the quality of the final astrometric and photometric solutions. We estimate that a non-negligible fraction of valid detections of close pairs remains rejected as spurious detections (see Section 3.4.13) in Gaia EDR3 and furthermore some of these close sources may have few matched detections only. For this reason, a major update will be carried out in the next processing cycle to include the prompt identification of these detections and improve their treatment in the cross-match and subsequent pipelines (Torra et al. 2021).

As explained in Torra et al. (2021), several very bright sources () had a close-by source at less than 1${.}^{\mathrm{\prime \prime }}$5 distance with a similar magnitude in the input source catalogue for Gaia EDR3. Specifically, about 45% of these very bright sources in Gaia DR2 were marked as duplicates. Thanks to the updates in the cluster-source assignation module of the resolution stage (see Section 3.4.13), the duplicates in this magnitude range have been largely reduced in Gaia EDR3. This reduction has implied that a large fraction of sources have been superseded by a merge or deletion action (so non-persisting). Note that since most of the detections of these bright sources have persisted, these sources have not been lost and basically Gaia EDR3 includes all of them but with a new source identifier.