Forum Springer Astron. Astrophys.
Forum Whats New Search Orders

Astron. Astrophys. 358, 30-44 (2000)

Previous Section Next Section Title Page Table of Contents

2. Description of the data

The difficulty to get a wide angle coverage of the sky in good conditions is the reason why there is not yet a clear detection of cosmic shear. For this work, we decided to get the widest angular field possible, which was done at the expense of homogeneity of the data set. However this does not impact our primary goals which are the detection of a weak lensing signal and the test of the control of systematics.

We use in total eight different pointings mixing CFH12K and UH8K data sets (see Table 1). They are spread over five statistically independent areas, each separated by more than 10 degrees. The total field covers about [FORMULA], and contains [FORMULA] galaxies (with a number density [FORMULA] gal/arcmin2). Note that the galaxies are weighted as discussed in Sect. 2.2, and parts of the fields are masked, so the effective number density of galaxies is about half.


Table 1. List of the fields. Most of the exposures were taken in the I band at CFHT. The total area is 1.7 deg2, and the 8 fields are uncorrelated.

All the data were obtained at the CFHT prime focus. We used observations spread over 4 years from 1996 to 1999, with two different cameras: the UH8K (Luppino et al. 1994), covering a field of 28[FORMULA]28 square arc minutes with 0.2 arc-second per pixel and the CFHT12K 2 (Cuillandre et al. 2000) covering a field of 42[FORMULA]28 square arc-minutes with 0.2 arc-second per pixel as well. Because these observations were initially done for various scientific purposes, they have been done either in I or in V band. Table 1 summarizes the dataset. The SA57 field was kindly provided by M. Crézé and A. Robin who observed this field for another scientific purpose (star counts and proper motions). The UH8K Abell 1942 data were obtained during discretionary time. The F14 and F02 fields are part of the deep imaging survey of 16 square-degrees in BVRI being conducted at CFHT jointly by several French teams. This survey is designed to satisfy several scientific programs, including the DESCART weak lensing program, the study of galaxy evolution and clustering evolution, clusters and AGN searches, and prepare the spectroscopic sample to be studied for the VLT-VIRMOS deep redshift survey (Le eFvre et al. 1998). CFDF-03 is one of the Canada-France-Deep-Fields (CFDF) studied within the framework of the Canada-France Deep Fields, with data collected with the UH8K (Mc Cracken et al. in preparation).

The observations were done as usual, by splitting the total integration time in individual exposures of 10 minutes each, offsetting the telescope by 7 to 12 arc-seconds after each image acquisition. For the I and the V band data, we got between 7 to 13 different exposures per field, all with seeing conditions varying by less than [FORMULA] 0.07 arc-seconds (the others were not co-added). The total exposure times range from 1.75 hours in V to 5 hours in I.

The total field observed covers 2.05 square degrees, including 0.88 square degrees in V and 1.17 square degrees in I. However, one CCD of the UH8K and two CCDs initially mounted on the CFH12K of the May 1999 run have strong charge transfer efficiency problems and are not suitable for weak lensing analysis. Therefore, the final area only covers 1.74 square degrees: 0.64 square degrees in V and 1.1 square degrees in I. As we can see from Table 1 each field has different properties (filter, exposure time, seeing) which makes this first data set somewhat heterogeneous.

The data processing was done at the TERAPIX data center located at IAP which has been created in order to process big images obtained with these panoramic CCD cameras 3. Its CPU (2 COMPAQ XP1000 with 1.2 Gb RAM memory each equipped with DEC alpha ev6/ev67 processors) and disk space (1.2 Tbytes) facilities permit us to handle such a huge amount of data efficiently.

For all but the CFDF-03 field, the preparation of the detrending frames (master bias, master dark, master flats, superflats, fringing pattern, if any) and the generation of pre-reduced and stacked data were done using the FLIPS pre-reduction package (FITS Large Image Pre-reduction software) implemented at CFHT and in the TERAPIX pipeline (Cuillandre et al. in preparation). In total, more than 300 Gbytes of data have been processed for this work.

The CFHT prime focus wide-field corrector introduces a large-scale geometrical distortion in the field (Cuillandre et al. 1996). Re-sampling the data over the angular size of one CCD (14 arc-minutes) cannot be avoided if large angular offsets ([FORMULA] arc-seconds) are used for the dithering pattern (like for the CFDF-03 data). Since we kept the offsets between all individual exposures within a 15 arc-seconds diameter disk, the contribution of the distortion between objects at the top and at the bottom of the CCD between dithered exposures is kept below one tenth of a pixel. With the seeing above 0.7 arc-second and a sampling of 0.2 arc-second/pixel, the contribution of this effect is totally negligible. A simulation of the optical distortion of the instrument shows that the variation from one field to another never exceeds 0.3%, which confirms what we expected from the CFHT optical design of the wide-field corrector. We discuss this point in Sect. 5, in particular by confirming that the sensitivity of the shear components with radial distances is negligible. Also not correcting this optical distortion results in a slightly different plate scale from the center to the edge of the field (pixels see more sky in the outside field). But this is also of no consequence for our program since the effect is very small as compared to the signal we are interested in.

The stacking of the non-CFDF images has been done independently for each individual CCD (each covering 7[FORMULA]14 arc-minutes). We decided not to create a single large UH8K or CFH12K image per pointing since it is useless for our purpose. It complicates the weak lensing analysis, in particular for the PSF correction, and needs to handle properly the gaps between CCDs which potentially could produce discontinuities in the properties of the field. The drawback is that we restricted ourselves to a weak lensing analysis on scales smaller than 7 arc-minutes (radius smaller than 3.5 arc-minutes, as shown in the next figures); this is not a critical scientific issue since the total field of view is still too small to provide significant signal beyond that angular scale. In the following we consider each individual CCD as one unit of the data set.

The co-addition was performed by computing first the offset of each CCD between each individual exposure from the identification of common bright objects (usually 20 objects) spread over one of the CCD's arbitrary chosen as a reference frame. Then, for each exposure the offsets in the x- and y- directions are computed using the detection algorithm of the SExtractor package (Bertin & Arnouts 1996) which provides a typical accuracy better than one tenth of a pixel for bright objects. The internal accuracy of this technique is given by the rms fluctuations of the offsets of each reference object. Because our offsets were small the procedure works very well and provides a stable solution quickly. We usually reach an accuracy over the CCDs of 0.25 pixels rms (0.05 arc-second) in both directions for offsets of about 10 arc-seconds (50 pixels). Once the offsets are known the individual CCDs are stacked using a bilinear interpolation and by oversampling each pixel by a factor of 5 in both x- and y- directions (corresponding to the rms accuracy of the offsets). The images are then re-binned 1[FORMULA]1 and finally a clipped median procedure is used for the addition. The procedure requires CPU and disk space but works very well, provided the shift between exposures remains small. We then end up with a final set of stacked CCDs which are ready for weak lensing analysis.

The twelve separate pointings of the CFDF-03 field were processed independently using a method which is fully described elsewhere (McCracken et al. 2000, in preparation). Briefly, it uses astrometric sources present in the field to derive a world coordinate system (WCS; in this work we use a gnomic projection with higher order terms). This mapping is then used to combine the eight CCD frames to produce a single image in which a uniform pixel scale is restored across the field. Subsequent pointings are registered to this initial WCS by using a large number of sources distributed over the eight CCDs to correct for telescope flexure and atmospheric refraction. For each pointing the registration accuracy is [FORMULA] rms over the entire field. The final twelve projected images are combined using a clipped median, which, although sub-optimal in S/N terms, provides the best rejection for cosmic rays and other transient events for small numbers of input images.

Previous Section Next Section Title Page Table of Contents

© European Southern Observatory (ESO) 2000

Online publication: June 26, 2000