2. Observations and data reduction
The observations of patch B were carried out over several months in the period July 1997 to December 1997, using the red channel of the EMMI camera on the 3.5m New Technology Telescope (NTT) at La Silla. The red channel of EMMI is equipped with a Tektronix 2046 2046 chip with a pixel size of 0.266 arcsec and a useful field-of-view of about . EIS uses a special set of BVI filters and the response function of the system can be found in paper I.
As described in paper I, the observations were carried out by a sequence of overlapping exposures (hereafter referred to as even/odd frames) 150 sec each, with each position on the sky being sampled at least twice. A total of 701 frames were obtained in the area with 200 in B, 282 in V and 219 in I bands. Only 150 frames in each band were required to cover the field but poor weather conditions required several frames to be re-observed. The strong variations in the observing conditions can be seen in Fig. 1 which shows, for each band, the seeing distribution of all observed frames. For comparison the shaded histograms show the seeing distribution of the frames finally accepted, with the solid vertical line in each panel indicating the median seeing for each band. The B-band is the worst overall with a median seeing of 1.2 arcsec and with a few frames extending to very large seeing ( arcsec). The upper and lower quartiles are, as can be seen in Fig. 1 1.0 and 1.3 in B, 0.8 and 1.2 for V and 0.9 and 1.2 in I. In the analysis below 9 frames with seeing arcsec were discarded because of their incompleteness at faint magnitudes. Fig. 2 shows, again for each band, the limiting isophote within 1 arcsec. The transparency of the nights also showed significant variations especially for the I-band images, with one frame reaching 21 mag/arcsec2 (not shown in the figure). For this frame, which was removed from the analysis, the depth reached is considerably shallower than the remaining frames leading not only to a bright limiting magnitude but also to the detection of a significant number of spurious objects. Other frames with bright limiting isophotes have no significant impact in the analysis presented in Sect. 3.
Figs. 3 and 4 show, for each band, the two-dimensional distribution of the seeing and limiting isophote as determined from the even frames. Similar results are obtained for the odd frames which alternate with the even ones. Such maps allow the potential user of the derived catalogs to evaluate their reliability. The final data are reasonably homogeneous with the median seeing in all bands arcsec. However, some poor images do exist and for some applications must be removed, as discussed above. The area that is affected is square degree.
2.2. Data reduction
Finally, it is worth mentioning that since the completion of paper I, the V images for patch A over an area square degrees have also been reduced and are being made available together with the catalogs extracted from them which are used below.
The data were processed by the EIS pipeline being developed to handle large imaging programs and described in detail in paper I. The software development is still in progress with new functionalities being constantly added to the pipeline as well as enhanced features in Skycat driven by the survey needs, in particular to facilitate the visual inspection of the target lists being produced. In addition, new tools are being developed to handle color information, which adds a new level of complexity especially in the preparation of object catalogs. Computation of colors require reliable association of objects detected in different passbands, which may be affected differently by the seeing, astrometric errors, the morphology of the objects and the performance of the de-blending algorithm. One also needs a proper definition for the measurement of colors for faint sources and upper limits for non-detections. Currently, preliminary color catalogs are being produced only for point-sources, which are being systematically inspected to verify the catalogs and identify any peculiarities (Zaggia et al. 1999). This is an important first step towards the preparation of the final color catalog for patch B. For instance, during the visual inspection of objects selected by their peculiar colors, it became evident that ghost images, observed near relatively bright stars () in the B and V images, contaminate the catalogs.
2.3. Color transformation
Using all the standard star observations carried out in photometric nights at the NTT with the EIS filters, in the period July 1997-March 1998, the color transformation between the EIS and the Johnson-Cousins systems has been determined. In Fig. 5 the observed transformations for all the three bands are shown, as a function of color in the Johnson-Cousins system. The fits are given by the relations:
Note that the transformation given here for is slightly different from that determined in paper I. This is because more standards have been included since and a more careful pruning of the data has been performed. The determination of color corrections include 284 measurements in B, 255 in V and 209 in I, with the formal errors in the color terms estimated to be 0.02 mag in all three bands. In general, the color term is small except for the B-band. In this case the data also suggests a possible departure from linearity at the red end. As a final note, it is worth mentioning that in the process of examining all the standard star observations, errors in positions and the presence of variable stars in the Landolt lists were found. A complete list of these problems will be reported elsewhere.
The photometric calibration of the patch was carried out by first bringing all frames to a common zero-point as determined from the relative magnitudes of objects in overlap regions, within a pre-selected magnitude range. This was done by a global least-square fit to all the relative zero-points, constraining their sum to be equal to zero. The internal accuracy of the derived photometric solution is mag (Paper I). Second, absolute zero-points are found for frames observed in photometric conditions. The zero-points for these frames were determined using a total of 36 frames of 7 fields containing standard stars taken from Landolt (1992 a,b), observed over 5 nights. These frames were also reduced through the pipeline, which identified the standard stars and measured magnitudes through Landolt apertures automatically (see paper I). Altogether 148 independent measurements of standards were used in the calibration.
Two solutions are then determined: one which computes a single zero-point offset, based on the weighted average of the zero-points of the calibrated frames, and the other using a first-order polynomial in both right ascension and declination. Comparison with external data suggests that a zero-point offset provides an adequate photometric calibration for the entire patch (see below).
External photometric data come from the Dutch 0.9m telescope at La Silla and from overlaps with DENIS data and with frames taken by Lidman & Peterson (1996). The regions of overlap of these data are shown in Fig. 6. In the figure the regions observed under photometric conditions are also indicated. Comparison of this figure with its counterpart in paper I, demonstrates that the data for patch B is clearly of superior quality with a much larger fraction of frames taken under photometric conditions. Comparison with these external data is important in order to look for possible gradients in the photometric zero-point, introduced by the relative photometry which implicitly assumes that there are no systematic errors in the flatfield from frame to frame.
2.5. Object catalogs
During the processing of a patch through the pipeline, object catalogs extracted from single frames are merged together into a "patch" catalog for each passband. This is the parent catalog which consists of multiple entries of objects detected in overlapping frames. For each detection, the seeing and noise of the frame in which the object was found are also stored. The parent catalog is used to derive different types of single-entry catalogs detected from 150 sec exposures such as the odd/even catalogs described in paper I. Alternatively, it has also been used to derive a unique catalog (hereafter "best" catalog) defined by examining the characteristics of the frames where a given object was detected, saving only the entry associated with the best seeing frame. Details regarding the methodology of association will be described elsewhere (Deul et al. 1999). From the flag information available in the single-entry catalog, filtered catalogs have been produced for analysis purposes. The filtering is required in order to eliminate truncated objects and objects with a significant number of pixels affected by cosmics and/or other artifacts. The parameters adopted in the filtering are the same as those given in paper I.
In general, this single-entry patch-wide catalog is the one used below, while the odd/even are used to estimate the magnitude errors directly from the data, by cross-identifying the objects. For point-like sources, a preliminary attempt has also been made to produce a color catalog combining the information of the catalogs derived from each passband. Using the same association scheme mentioned above, a cross-identification of objects is made and colors are computed using the mag_auto estimator of SExtractor (e.g., Paper I) which should be adequate for point sources. For non-detections in a given band, 1 limiting magnitudes are computed from the seeing and noise properties of the best seeing frame available at the expected position of the object. Even though still rudimentary, this derived catalog serves for verification purposes and for a first cut analysis of the data. The final color catalog will only be derived from the co-added images. In this case, colors will be computed using the detection area determined from a reference image (one band, e.g., I, or the summed images of different bands,e.g., ), but measuring the flux in the respective images. Even though the required software is available it is only now being integrated into the pipeline.
It is important to emphasize the complexity of handling and merging information extracted from different passbands. For instance, each object may have a different SExtractor stellarity-index which may impact the galaxy/star classification, close pairs may be de-blended in one passband and not in another, depending on the seeing. Clearly a complete description of all the possible pitfalls and the overall performance of the software is beyond the scope of the present paper, and will instead be discussed in Deul et al. (1999). The current work also shows the shortcomings of handling catalogs and points out the need for the implementation of an object database with a flexible user interface to allow for the full exploration of the data by different groups.
For the purposes of the present paper galaxies are objects with stellarity index if brighter than the star/galaxy classification limit (B=22, V=22, I=21) or any object, regardless of the stellarity index, fainter than this limit. Stars are objects with stellarity index . Note that this definition leads to some cross-contamination but it has no significant impact on the conclusions. In evaluating the data in Sect. 3 the derived star and galaxy catalogs were, for simplicity, trimmed at the edges and the bad frames discussed above were removed. After trimming the covered areas are: 1.3, 1.4 and 1.37 square degrees in , respectively. The corresponding two-dimensional distributions of stars, down to the star/galaxy classification limits, and galaxies, down to estimated 80% completeness limits (discussed below), are shown in Fig. 7. The total number of objects in these plots are: 2290 stars brighter than and 33133 galaxies brighter than ; 3378 stars brighter than and 58590 galaxies brighter than ; and 4297 stars and 44546 galaxies brighter than and , respectively. Recall that the distribution shown is for the "best" catalog. If one wishes to work with the odd/even catalogs their distribution have to be examined in the same way. In order to avoid extraneous colors due to bad data in one or more bands, the color catalog examined in Sect. 3 corresponds to the common areas covered in the different passbands and has a total area of 1.27 square degrees.
2.6. Completeness, contamination and magnitude Errors
As in paper I, the completeness of the derived catalogs has been established by using a single field where several exposures have been made over the period of observations of patch B. The corresponding catalogs in each passband were then compared with that of a typical frame with an exposure time of 150 sec. A total of 9 exposures in B band, 6 exposures in V band and 8 in I band, were selected discarding others taken in less favorable conditions. These exposures were coadded and the derived catalogs are at least one magnitude deeper than the typical survey frame. Comparing the identified objects in a single exposure frame with those extracted from the co-added images one can derive the expected completeness for typical 150 sec survey frames. The results of these comparisons are shown in Fig. 8 showing that the catalogs are 80% complete at , and . These limits should apply for survey frames with a seeing close to the median seeing. Also shown in the figure is the number of false positives computed from the fraction of objects identified in the single exposure frames which were not detected in the co-added image.
In order to estimate the accuracy of the magnitudes the odd and even catalogs were compared and a lower limit estimate of the photometric errors was obtained from the repeatability of the magnitudes for paired objects. The estimated errors from this comparison are given in Fig. 9, which shows that in the interval they range from 0.02 to 0.1 mag, reaching 0.3 mag at . Similar values are found for B and V brighter than . These values correspond well to those estimated by SExtractor.
It is also of interest to obtain a completely independent estimate of the errors in the magnitudes. This can be done by comparing the objects detected by EIS with those found by Lidman & Peterson (1996), who have used a different object detection algorithm. This is shown in Fig. 10 where the objects detected in two separate fields (see Fig. 6), with 1354 and 1299 objects each, are compared. Note that because of differences in the astrometry this comparison was done using the astrometric solution found by the EIS pipeline but the magnitudes as determined in the original catalog. Even adopting this procedure misidentifications are still present, leading to outliers, over the entire magnitude range, with significant magnitude differences. Nevertheless, the zero-point offset is typically 0.05 mag for , consistent with the zero-point correction proposed by these authors to bring their measurements into the Johnson-Cousins system. Beyond this limit, the Lidman & Peterson catalog becomes increasingly incomplete leading to a biased offset. The scatter in this comparison is less than 0.3 mag down to , consistent with our internal estimates if one attributes comparable errors to the Lidman & Peterson measurements. The zero-point offset and the scatter of the magnitude differences is the same for the two fields considered, suggesting that there are no strong gradients (in right ascension) in the photometric zero-point of the patch, at least on a 0.5 degree scale, corresponding to the separation of the two Lidman & Peterson fields.
In order to further investigate possible systematic errors in the photometric zero-point over the scale of the patch, the EIS catalogs were also compared with object catalogs extracted from the two DENIS strips that cross the survey region (see Fig. 6). This allows one to investigate the variation of the zero-point as a function of right ascension and, especially, of declination. The results are shown in Fig. 11. The domain in which the comparison can be made is relatively small because of saturation of objects in EIS at the bright end () and the shallow magnitude limit of DENIS (). Still, within the two magnitudes where comparison is possible one finds a roughly constant zero-point offset of less than 0.02 mag for both strips and a scatter that can be attributed to the errors in the DENIS magnitudes (Deul 1998).
Finally, similar comparisons can be made between the EIS magnitudes and those measured from the images obtained at the 0.9m Dutch telescope at La Silla, in this case, for all three passbands. Fig. 12 shows these comparisons, combining all the three fields that overlap patch B. Even though the total number of objects is relatively small (112 in B, 180 in V and 204 in I) preventing an accurate comparison, one finds a reasonable agreement in the zero-point and a scatter that can be accounted for by magnitude errors in the Dutch data ( at , and ). The observed zero-point offset between the Dutch and EIS data ( in B, 0.1 in V, in I) can be explained by the color term corrections required for the EIS and Dutch measurements to bring both measurements into the Cousins system. As the fields are well separated in right ascension, this result gives further evidence that there are no significant gradients in the photometric zero-point in any of the passbands.
In summary, comparison of the EIS magnitudes with available external data shows no indication of gradients in the photometric zero-point of the patch. However, the external data are mainly overlapping the EIS data at low declination (), especially for the and band, giving weaker constraints in that direction for these bands. Instead, the consistency of the zeropoints for these bands were investigated using the for halo stars in the turn-off region around . It was found that the of these stars showed a gradient with declination. To localize the problem the and were also checked for gradients in the direction of declination and the former was found to have gradient. Therefore, a linear relation (Eq. 5) was fitted to remove this gradient.
This correction has been applied to the catalogs and a second version of the data will be released in 1999, to take this correction into account.
© European Southern Observatory (ESO) 1999
Online publication: April 19, 1999