F. Drizzling

The problems related to mapping were recently studied in detail to fully exploit the possibilities offered by HST WFPC2 images. Due to the large size on the sky of the CCD pixels with respect to the width of the PSF delivered by the telescope optics, in these images the optical PSF is severely undersampled, leading to a decrease in the angular resolution achievable by the instrument. The techniques that have been developed in the HST community in order to recover the optical resolution are invariably based on dithering, i.e. on the superposition of several images of the same sky region displaced of a fraction of a pixel with respect to each other. In particular, a technique for the stacking of dithered images was developed for use in the Hubble Deep Field North (HDF-N) project ([Williams et al. 1996], and called drizzling. Since all BBP sample sizes adopted in the simulation of GAIA galaxy observations severely undersample GAIA Astro optical PSF, and since the centers of different GAIA observations of the same sky region are displaced with respect to each other, it is interesting to try and apply this technique to our case.

The drizzling algorithm, which is also more formally known as variable-pixel linear reconstruction, is conceptually simple and similar to the stacking algorithm described in Section 6.4. In practice, the only difference is that each sample of each observation is shrinked of a factor of two along both directions before carrying out the subsampling^F.1. Drizzling, however, is known to produce artifacts in the output image, and this effect, negligible in the HDF-N, increases in size with the ratio between the sample and flux map element sizes. Since in our case this ratio is particularly high, e.g. 24 for a sample size of $6\times4$ pixels against a ratio of about 4 used in the HDF-N, the artifacts may become unmanageably frequent and of large spatial extent. This is indeed the case, as illustrated by Figure F.1, where M100 flux maps obtained with the baseline stacking technique described in Section 6.4 and with the drizzling technique are compared. Throughout the drizzled flux map, conspicuous artifacts appear that could be easily confused with point-like features, should the overall signal-to-noise ratio be lower.

Figure F.1: Drizzling and artifacts. GAIA BBP flux maps of M100 reconstructed from 50 simulated observations, i.e. with an effective total exposure time of 43.09 s, obtained with $6\times4$ pixels/sample. Left: flux map obtained through the baseline stacking technique. Right: flux map obtained through the drizzling stacking technique. The side of each flux map is about 16 arcsec.

On the other hand, Figure F.2 illustrates how drizzling could increase the angular resolution of flux maps.

Figure F.2: Drizzling and angular resolution of flux maps. A detail of GAIA BBP flux maps of M100 reconstructed from 50 simulated observations, i.e. with an effective total exposure time of 43.09 s, obtained with $6\times4$ pixels/sample. Left: flux map obtained through the baseline stacking technique. Right: flux map obtained through the drizzling stacking technique. The side of each image is about 2 arcsec.

For these reasons, the drizzling algorithm, at least in its present form, is not to be considered for application in stacking of GAIA galaxy observations, but may be an interesting starting point for the development of dedicated stacking techniques aimed at fully recovering the optical resolution of GAIA observations.