4 IDU Scientific Overview
4.2 Detection Classifier
The Gaia on board detection software was build to detect point like sources and it is in principle capable of autonomously discriminating stars from false detections i.e. cosmic rays. For this, parametrised criteria of the shape of the LSF/PSF are used, which need to be calibrated and tuned. Furthermore, this criteria of point like sources was relaxed as otherwise Gaia would not get moving asteroids, little bit extended galaxies or pe- culiar double star configurations. A study of the detection capability, in
Figure 4.5: A scene record parametrizes the object transit providing three predicted knots over the trajectory of the transit over the focal
particular non-saturated stars, double stars, unresolved external galaxies, and asteroids is provided in de Bruijne et al. [2015].
However, during Gaia commissioning, we detected several kinds of spuri- ous detections and related issues, and in much larger quantities than we expected [Fabricius, 2014b]. In fact, the number of spurious detections was largely increased when it was decided to update the detection parameters on board to make possible the observation of sources fainter than magni- tude 20 with few rejections of real sources. Some mitigation procedures were also introduced in the on board detection software [Fabricius, 2014a] beginning 2015 but they are not enough when such faint observations are pursued.
The main problem with the spurious detections arises from the fact that each of them may lead to the creation of a new source in the Cross-Match. Therefore, the goal of the Detection Classifier (IDU-DC) task is precisely to avoid that these detections result in new sources in the catalogue, clas- sifying detections in genuine and spurious and by maintaining a list of blacklisted detections. In other words, IDU-DC results will prevent that spurious sources are created in the Cross-Match and consequently that spurious sources enter other calibration pipelines from other downstream processes.
Here is a brief description of the several categories of spurious detections found in the data so far:
• Spurious detections due to cosmic rays. These are relatively harmless because they happen randomly across the sky.
• Spurious detections due to background noise or CCD cosmetics de- fects (i.e. CCD bad columns). These are also relatively harmless and normally rare, but depending on the detection criteria, they can lead to a huge number of spurious detections. A study of the probability of this kind of detections is available in Azaz [2014].
• Duplicated detections (essentially double detections) produced from slightly asymmetric SM images where more that one local maximum
is detected. In this case the acquired windows are basically containing the same samples.
• Spurious detections around and along the diffraction spikes of bright sources. Bright sources may easily lead to numerous (from hundreds to thousands) of spurious detections in each transit, especially near the source centre and along the diffraction spikes in the AL direction (example included in Figure 4.7).
• Spurious detections appearing on the other FoV originated from un- expected light paths and reflections within the spacecraft for very bright sources and very close planets. This group of spurious can be seen as ghost detections from those on the original FoV (Figure 4.7). • Spurious detections from major SSO, mainly planets. These tran- sits can easily pollute arbitrary sky regions with thousands spurious detections (Figure 4.8).
• Spurious detections from extended and diffuse objects. One clear example is the Cat’s Eye Planetary Nebula or NGC 6543 shown in Figure 4.6. This case was detected during an IDU test campaign at Data Processing Center Barcelona (DPCB) and it was published as image of the week on December 2014 showing that Gaia is actually detecting not only stars but also high surface brightness filamentary structures.
The detections in the filamentary structures looks point like enough, then Gaia SHOULD detect it
As commented above, the big impact of the spurious detections is an issue recently identified. The current mitigation measures and software modules implemented are still under active development in IDT and IDU. However, it seems clear that in the long run, an effective mitigation scheme should form part of the iteration from DRC to DRC with input from several down- stream processes, mainly the SEA from CU5 and the results from CU4.
Figure 4.6: The Cat’s Eye Planetary Nebula or NGC 6543 ob- served with the Hubble Space Telescope (left image) and as Gaia de- tections (the 84,000 blue points on middle and right images) (Credit: Photo: NASA/ESA/HEIC/The Hubble Heritage Team/STScI/AURA;
Gaia Observation Plot: Gaia/DPAC/DPCB)
Currently the baseline is that each CU will provide its own list of black- listed or whitelisted detections (list reverting the blacklisted detections) which will have to be combined for the ultimate filtering of the detections. Each spurious detection case listed above has its own complications and particularities and most of them are still not being treated. Currently only IDT and IDU implement the classification and the filtering of the spurious detections in the Cross-Match.
The current implementation in IDT is just identifying the spurious detec- tions in predefined regions or boxes around the actual observed bright stars [Bestard, 2015]. The process basically looks for the brighter observations in the object log (provided by the ASD7 packets) and select all the observa- tions falling in a predefined set of boxes centred in the parent observation coordinates. The selected observations are then analysed and classified as spurious detections if given distance and magnitude decay conditions are satisfied. These predefined boxes have been parametrized with the features and patterns seen in the real data according to the parent magnitude. This
Figure 4.7: Spurious detections around a bright source of magnitude 5.4 on the top panel. Very bright source of magnitude -1.4 on bottom panel where it can be seen in blue the spurious detection structures
Figure 4.8: Spurious detections around Jupiter and Venus (red dots on top and bottom panel respectively). For Jupiter, some of its satellites are also recognizable. In the case of Venus, although the planet is not directly observed by Gaia (being located far below the bottom CCD row) it is producing a large amount of spurious detections in both FoVs
Figure 4.9: Schematic data flow for the IDU-DC task, showing the main inputs and outputs
implementation is quite limited and fails to identify quite large numbers of spurious detections. This implementation will hardly be improved in IDT due to processing restrictions, in both resources and introduction of addi- tional dependencies such as the prior computation of some kind of transit predictions as done in the IDU-SCN task.
For IDU the situation is quite different, and a more ambitious solution is being implemented (see Figure 4.9 for a schematic diagram of the pro- cessing and data dependencies of this task). First of all, the IDU-SCN results are available which will enable the possibility of identifying more spurious detections cases. The IDU-SCN removes the limitation of only treating spikes of actually observed bright sources, even adding informa- tion of sources transiting the CCD edges, which may produce orphan spikes (without a parent observation to trigger the classification as in IDT). Additionally, the IDU-SCN provides information of the far too bright sources, the SSOs transits and the diffuse objects. These scene records will trigger the corresponding tailored classifications. As an example, for the VBSs and major SSOs we currently plan to filter all faint observations around the predicted transits – even from both FoVs for the transits of Venus.
The treatment of spurious created by faint sources is more tricky since no observable structures like the spikes shown in Figure 4.7 can be detected. In these cases, a multi–epoch treatment might be required to know if they are genuine or spurious detections - i.e. checking if more transits are com- patible or resolve to the same new source entry. Additionally, some kind
of feed back from downstream processes such as SEA from CU5 and CU4 solution could be of great help to resolve such cases.
For the remaining cases; cosmic rays, background noise and CCD bad columns, the damage caused is quite limited and we may consider a process that check the observed samples for the presence of any useful signal. This is in any case a low priority task, which will be for sure revisited as the DRCs progress.
In addition, spurious new sources can also be caused by attitude large ex- cursions leading to misplaced detections. These detections are not strictly spurious detections but they are considered as well since they may cause similar problems in the Cross-Match. Consequently, it is highly desirable to identify and clean up these detections during the on ground data pro- cessing.
The design and implementation of the IDU-DC task is lead by the CU3- UB, and preliminary results have been also included in Chapter 6.