Source detection has been performed in the z band. The primary parameters that affect the detection process, namely the minimum connected area, the isophotal threshold (in unit of the background fluctuation rms) and the convolution kernel have been set with the help of numerical simulations to maximize sensitivity to faint and relatively compact sources while reducing the number of spurious sources to a minimum. In the simulations we inserted a number of artificial galaxies in the images and retrieved them with the same procedures adopted for real sources. The morphology of these galaxies was, with equal probability, either an exponential disk or a de Vaucouleur spheroid. Apparent magnitudes and half-light radii were extacted from uniform distribution functions, covering the range 20 <= z_850 <= 28 (AB) and PSF <= r_1/2 <= 2.0 arcsec, respectively. The simulated disk sample (viewed from arbitrary directions) is drawn from a population of oblate optically-thin spheroids with a Gaussian distribution of intrinsic axial ratios with mean b/a = 0.05 and sigma = 0.01. The spheroids are drawn from a population of oblate spheroids with intrinsic axial ratios uniformly populating the range 0.3 < b/a < 0.9. Position angles are randomly distributed.
We fine-tuned the SExtractor parameters to maximize the number of detected sources, while keeping the number of spurious sources, estimated with the method of the "negative images", essentially negligible.
One should keep in mind, however, that there is no one single catalog that is perfect for all scientific problems and applications. While we believe that the released catalogs are good for most investigations on faint galaxies, users with specific problems (e.g. low-surface brightness galaxies) are strongly urged to experiment with different settings of the SExtraction detection parameters.
Three types of photometric apertures for each source have been defined during the detection process, namely a suite of 11 circular apertures, the isophotal aperture and the SExtractor MAG_AUTO aperture. The radii of the circular apertures are 2.93, 4.17, 5.87, 8.33, 11.77, 16.67, 23.57, 33.33, 47.13, 66.67, 94.27 pixels, corresponding to a geometrical series of aperture areas with ratio equal to 2.
The estimate of the photometric errors takes into account the noise correlation intrinsic to drizzled images.
5.0 Modifications to SExtractor
In the process of building the catalogs, we realized that two small modifications to the SExtractor software (version 2.2.2) resulted in somewhat better catalogs. These modifications affect the splitting/merging behavior and the computation of the local sky background, and are described below.
- 5.1 Local sky background
Tests of early versions of the catalogs indicated that the photometry of faint galaxies was being somewhat underestimated and that this problem was, in part, due to the local sky background being overestimated when the wings of faint galaxies were incorrectly included in the sky measurement.
To compute the local sky, SExtractor selects pixels from a rectangular annulus around the object. The inner "radius" of this annulus is located G pixels away from the edge of the object (as defined by its min/max extent in x and y), where G = (object extent in x or y)/4. Since we used a minimum detection threshold of 16 pixels, this formula produced an inner radius abutting the detected object edge for most faint, compact objects.
To remedy this problem, we modified the code to enforce a minimum gapsize between the inner edge of the square annulus and the "end" of the source equivalent to 1 arcsec. For the GOODS v1.0 release data with scale 0.03 arcsec/pixel, this produced a minimum gap of 33 pixels around the detected object extent before selecting pixels for the sky calculation. For very large objects, where (extent/4 pixels) exceeds 1 arcsec, the original formula is used.
Comparisons between this modified version of the code and the original SExtractor run showed no difference in the sky background for large objects, and a systematic brightening of the faintest objects, as expected. Comparison of the new sky values against sky values measured with the IRAF task PHOT for faint, compact sources, run as an independent check, shows no discrepancies and, in fact, shows excellent agreement between the background estimates of the two packages.
- 5.2 Splitting/merging
As documented in the SExtractor manual, after the splitting is performed, a CLEANing algorithm is run to attempt to determine false splits. Each split child is analyzed to determine whether it would have been detected had its parent object not been present. If not, the pixels from the child object are merged back into the parent object.
As has been discussed on the SExtractor mail list, this procedure sometimes results in dramatically discontiguous objects, with pixels assigned to the same object that are clearly (to the human eye) unrelated.
To resolve this problem, we modified the code so that, instead of merging such false detections back in with the parent objects, they are simply discarded. Tests indicate that this eliminates nearly all discontiguous objects with little or no impact on photometry.
The GOODS Team conducted a number of tests of the completeness of the catalogs and of the accuracy of the measures that they include throughout their development. Here we show relevant plots that should be of help to investigators in assessing the completeness and photometric accuracy and precision of the catalogs for most (but certainly not all!) applications.
- 6.1 Completeness
- 6.2 Photometric Precision
- 6.2 Comparison with the HDF Wide-Field Planetary Camera data
The table below shows the mean F606W magnitude offsets (GOODS-HDF), along with the number of sources matched within 0.25 arcsec, and the standard deviation and standard error of the magnitude difference. GOODS Isophotal magnitudes are systematically brighter at the faint end than HDF isophotal magnitudes (reflecting the fact that the GOODS images are shallower). There is much less bias for MAG_AUTO.
MAG_ISO Mag range N mean stdev stderr --------- --- ----- ----- ------ 21 22 4 0.094 0.077 0.0192 22 23 6 0.057 0.089 0.0149 23 24 9 0.114 0.098 0.0109 24 25 22 0.156 0.138 0.0062 25 26 52 0.223 0.221 0.0042 26 27 116 0.485 0.435 0.0037 27 28 82 0.627 0.463 0.0056 MAG_AUTO Mag range N mean stdev stderr --------- --- ----- ----- ------ 21 22 4 0.059 0.075 0.0189 22 23 6 0.006 0.055 0.0092 23 24 10 -0.001 0.026 0.0026 24 25 22 0.007 0.095 0.0043 25 26 55 0.047 0.230 0.0041 26 27 129 0.038 0.181 0.0014 27 28 78 0.071 0.277 0.0035
- 6.2 Photometric Accuracy
The difference between the observed scatter and the SExtractor error is large, because the observed scatter is driven by outliers (due to crowding) and the distribution is non-gaussian. A fairer comparison, plotted in Figure 11, is between the SExtractor errrors and the semi 2-quartile width (ie. the half width of the distribution once the smallest and largest quartiles have been clipped), which indeed shows a much better agreement. At bright magnitudes SExtractor errors are underestimates because they do not account for variations in aperture corrections with galaxy shape, crowding, and other non-shot-noise effects. A reasonable "floor" to use for the errors is 0.07 magnitudes. Uncertainties in colors will be somewhat better because crowding and aperture-correction variations contribute less scatter in a differential comparison between two bands.
# 1 ID_IAU # 2 ALPHA_J2000 # 3 DELTA_J2000 # 4 SECT_REFNUM # 5 X_SECT # 6 Y_SECT # 7 X_MOSAIC # 8 Y_MOSAIC # 9 XPEAK_MOSAIC # 10 YPEAK_MOSAIC # 11 XPEAK_WORLD # 12 YPEAK_WORLD # 13 XMIN_MOSAIC # 14 YMIN_MOSAIC # 15 XMAX_MOSAIC # 16 YMAX_MOSAIC # 17 ISOAREA_IMAGE # 18 THETA_IMAGE # 19 ELLIPTICITY # 20 ELONGATION # 21 ERRTHETA_IMAGE # 22 KRON_RADIUS # 23 FLUX_RADIUS # 26 FWHM_IMAGE # 27 CLASS_STAR # 28 FLAGS # 29 IMAFLAGS_ISO # 30 NIMAFLAGS_ISO # 31 BACKGROUND # 32 FLUX_MAX # 33 MAG_ISO # 34 MAGERR_ISO # 35 FLUX_ISO # 36 FLUXERR_ISO # 37 MAG_ISOCOR # 38 MAGERR_ISOCOR # 39 FLUX_ISOCOR # 40 FLUXERR_ISOCOR # 41 MAG_AUTO # 42 MAGERR_AUTO # 43 FLUX_AUTO # 44 FLUXERR_AUTO # 45 MAG_BEST # 46 MAGERR_BEST # 47 FLUX_BEST # 48 FLUXERR_BEST # 49 MAG_APER # 60 MAGERR_APER # 71 FLUX_APER # 82 FLUXERR_APER # 93 X2_IMAGE # 94 Y2_IMAGE # 95 XY_IMAGE # 96 ERRX2_IMAGE # 97 ERRY2_IMAGE # 98 ERRXY_IMAGE # 99 A_IMAGE # 100 B_IMAGE # 101 ERRA_IMAGE # 102 ERRB_IMAGE # 103 ID_MOSAIC
A description of the GOODS observations and data products is given by Giavalisco and the GOODS Team, 2004, ApJ, January 20. Another paper describing in more details the GOODS data and the catalog is in preparation and, fort he time being, should be referred to as "Giavalisco and the GOODS Team, 2004 in preparation".