# 3.3.4 Low-resolution background

Author(s): Michael Davidson, Nigel Hambly

The astrophysical background incident on the focal plane of Gaia is dominated by stray light (Gaia Collaboration et al. 2016b), in part because the compact and folded design (Safa et al. 2004) of the various optical instruments leaves little opportunity for stray light path baffling. The lower rows (1–4) of the focal plane have a background signal that is dominated by sunlight scattered around the DSA while in the higher rows (5, 6, and 7) the diffuse optical background from the Zodiacal Light and Milky Way makes a significant contribution. An overview of the typical background across the focal plane is seen in Figure 3.8.

Due to the multiple contributions the background manifests as a high amplitude, rapidly changing photo-electric component that only approximately repeats with the satellite spin period. It evolves in amplitude with the L2 elliptical orbital solar distance, and phase, as the scanning attitude changes with respect to the ecliptic and galactic planes. Superposed on this are transient spikes in background due to very bright stars and bright solar system objects transiting across highly efficient regions of the straylight field. There is also some evidence of a slight reduction in the solar straylight, possibly caused by yellowing of the loose fibres on the sunshield due to UV exposure. The photo-electric background signal varies routinely over three orders of magnitude depending on instrument and time with values as low as $\sim 0.1$ and up to $\sim 100$ electrons per pixel per second.

Previous processing cycles parameterised the complex astrophysical background via a two-dimensional spline surface. This failed to follow rapid changes in the background structure, suffered from poorly constrained behaviour at the edges, and required manual configuration. From Gaia EDR3 onwards a new approach has been developed. We obtain low-resolution background measures from virtual objects and the lowest samples in faint science windows, after subtraction of bias, dark signal and the charge release trail, using iterative filtering based on Poisson noise properties. These background measures can then be sampled using a $k$ nearest neighbours algorithm at any OBMT and across-scan position. The value of $k$ is a configurable parameter in the processing chain and is set to $k=30$, with a down-sampling scheme used to limit the maximum rate of contributing background measures.

While the improvements offered by the new sampling model are welcome this background calibration is less critical than in earlier data releases. In those releases the spline background was used as a fixed input in the image parameter determination (Section 3.3.6). In Gaia EDR3 a local background per window is fitted for the majority of windows as an image parameter. This low-resolution background calibration is therefore used only for the production of the calibrator observations (Section 3.3.5), as an initial estimate for the fitted value, or in case insufficient samples are available to attempt a local solution. An example illustrating the various calibration methods is given in Figure 3.9.