Information is available about the following topics:
Total exposure time: 151074 sec Mean background level: 9.57e-4 cts/s/pixel Wavelength Continuum S/N 2300-2600A: 1.5 per pixel 2700-3100A: 4 per pixel
In most cases, target acquisitions were done without peaking up on the QSO. The standard CCD target acquisition acquires targets to an accuracy of better than 0.01 arcsec, which is 5% of the slit width. Typical thermal drifts over a typical 45 minute exposure are expected to be less than 0.1 pixels. Frequent calibration lamp observations insure that the wavelength scale is accurately calibrated. Because the observations were pointed near the pole of the HST orbit, the change in velocity of HST during an exposure was no more than 5 km/s. On-board processing corrects for this shift to a precision of 5 km/s in this observing mode.
The spectrum was flatfielded using a preliminary NUV flat derived from internal calibration lamp observations. This flat has been verified on observations of spectrophotometric standards. The flatfielding was carried out by the standard STSDAS "calstis" pipeline. The normal "Doppler smoothing" option was omitted during the pipeline processing. This is not expected to degrade the data, due to the small on-board doppler correction.
Wavelength correction was carried out via standard procedures in calstis, using contemporaneous observations of the internal calibration lamp. The wavelength accuracy for this observing mode was verified using archival STIS observations of the spectrophotometric standard star BD +28 4211, which was observed with STIS both with the 0.2x0.2" FP-SPLIT aperture and with the more standard 0.2x0.2" single aperture. Wavelengths for narrow interstellar MgII absorption features agreed to within 0.03 Angstroms (3 km/s). The observations of this star were also compared to high-resolution IUE observations. The mean shift relative to three IUE observations was -0.08 +- 0.07 Angstroms.
Comparison with the UCLES optical spectrum of the QSO (Outram et al. 198) shows good agreement. The radial velocities of the higher-order Lyman series lines agree to within 10 km/s with those measured for the Lyman alpha line in the UCLES spectrum.
The mean background in the STIS exposures is very low (9.6x10^-4 cts/s), and is almost entirely due to phosphorescent emission from the detector window. The mean background level varies with time, and the time dependence varies with position across the detector. To subtract the background from each exposure, a low-order spline fit was performed using pixels in the inter-order regions. This fit was subtracted from the data prior to extracting the one-dimensional spectra. The smooth fitted background level was used in the computation of the errors.
4. One-dimensional spectral extraction
The spectrum was extracted using a modified version of the STSDAS stis.x1d task. This program finds the position of the orders via cross-correlation. The task was modified to reject orders with low S/N and apply to the whole spectral extraction table a global shift that is determined from the orders with high S/N. Typically about 10 orders were used to find the position of the spectrum, and the rms of the measured shifts of these 10 orders was less than 0.15 pixels. After this centroiding step, the spectrum was extracted along a "trace," determined from observations of standard stars. The spectral extraction box was 3 pixels wide. At each wavelength, pixels within this extraction box were summed together with uniform weights. The S/N of the final spectrum could possibly be improved by extracting with "optimal" weights.
Conversion of counts to flux was also carried out in the x1d task. Fluxes were corrected for the wavelength-dependent throughput of the 0.2x0.2" aperture, and for the light lost outside the extraction box, using the standard values in the APT and PCT reference files.
At the end of one-dimensional extraction, the spectrum consists of 1656 separate pieces, which need to be summed together in a way that maximizes the signal-to-noise ratio of the final spectrum. This "optimal" combination was carried out using a prototype version of the STSDAS routine "splice." This program uses weighting arrays that specify the weight to be used when averaging fluxes from orders that overlap in wavelength. We chose to weight the flux in each order by the (S/N)^2 vs. pixel, as determined from a smooth fit to the continuum; prefering not to have the weights vary sharply around the edges of absorption lines. The first and last 30 pixels of each order were masked during the combination.
This process was iterative. The individual extracted orders were first combined through a simple average in regions of wavelength overlap; the result was an initial estimate of the total echelle spectrum. This estimate was then fit with a cubic spline to produce a continuum estimate. The continuum was then interpolated onto the wavelength array of each order in each exposure, and converted into counts (through multiplication by exposure time and sensitivity). A weighting array for each order was then computed as
w = (continuum counts)^2 / (continuum counts + background counts).
Arrays of wavelength, flux, statistical errors, and weight for all
orders of all frames were then combined by splice into a second
iteration of the total echelle spectrum. The process was then
iterated once more (continuum estimation --> weight determination
--> splice combination) to produce the third and final iteration of
the total echelle spectrum.
Total exposure time : 18480 sec Mean background level : cts/s/pixel (excluding airglow lines) Wavelength Continuum S/N 1167-1590A: >10 per pixel in the continuum, peaking at ~ 18 at 1320A
In most cases, target acquisitions were done without peaking up on the QSO. The standard CCD target acquisition acquires targets to an accuracy of better than 0.01 arcsec, which is 5% of the slit width. Typical thermal drifts over a mean 38 minute exposure are expected to be negligible. Frequent calibration lamp observations insure that the wavelength scale is accurately calibrated.
Because the observations were pointed near the pole of the HST orbit, the change in velocity of HST during an exposure was no more than 5 km/s. On-board processing corrects for this shift to a precision of 5 km/s in this observing mode. The final wavelength scale has been reduced to the heliocentric frame.
Note that the final spectrum shows the residual effect of the presence of geocoronal Ly-alpha over the range ~1210-1220 Angstroms and, to a lesser extent, the effect of OI over the range 1301-1310 Angstroms. Airglow emission from the OI] 1356 Angstrom line is weak and almost completely absent in the final spectrum.
The spectrum was flatfielded using a preliminary FUV flat derived from internal calibration lamp observations. This flat has been verified on observations of spectrophotometric standards. The flatfielding was carried out by the standard STSDAS "calstis" pipeline. The normal "Doppler smoothing" option was omitted during the pipeline processing. This is not expected to degrade the data, due to the small on-board doppler correction.
Wavelength correction was carried out via standard procedures in calstis, using contemporaneous observations of the internal calibration lamp.
The mean background in the STIS exposures is extremely low and is almost entirely due to the detector itself. The mean background level varies with time, and shows a temperature dependence. Background counts vary with position across the detector and vary from a mean of ~6x10^-5 ct/s/pixel near the "hot spot" centered close to pixel location (222,615), to ~1x10^5 ct/s/pixel at the long wavelength edge of the detector.
4. One-dimensional spectral extraction
The spectrum was extracted using the standard version of the STSDAS stis.x1d task. After preliminary testing, a spectral extraction box of 11 pixels in cross-dispersion width was found to better S/N that smaller boxes. This is larger than the boxes used in the G230L and E230M cases, but is consistent with the lower background in this detector. As the background is so weak with this detector/grating combination, we were unable to obtain a good two- dimensional fit. We therefore used the standard background measurement of the x1d task in which background regions either side of the spectrum are averaged and subtracted from each spectrum pixel. As the background is weak, we chose large background measurement regions than normal, offset by ten pixels either side of the center of the spectrum, and extending for 280 pixels further away. The background box size was limited to this size by the desire to use a similar background region size for all spectra. For three spectra (root names: o52f41040, o52f41050 and o52f41060) the spectra were offset towards the top of the detector, limiting the maximum region available for background measurement.
The spectral intensity was high enough to permit centring of the extraction box by the x1d task. After this centroiding step, the spectrum was extracted using a standard "trace" from the calibration database (the 1DT file). This trace corrects for instrument-induced distortions of the spectrum and is determined from observations of standard stars. At each wavelength, pixels within this extraction box were summed together with uniform weights. The S/N of the final spectrum could possibly be improved by extracting with "optimal" weights.
Conversion of counts to flux was also carried out in the x1d task. Fluxes were corrected for the wavelength-dependent throughput of the 52x0.2" aperture, and for the light lost outside the extraction box, using the standard values in the APT and PCT reference files.
At the end of one-dimensional extraction, we needed to combine the twelve separate exposures in a way that maximizes the signal-to-noise ratio of the final spectrum. This "optimal" combination was carried out using a prototype version of the STSDAS routine "splice." This program uses weighting arrays that specify the weight to be used when averaging fluxes from spectra that overlap in wavelength. We chose to weight the flux in each spectrum by the (S/N)^2 per pixel, smoothed over five pixels to reduce sharp variations in weighting value from pixel-to-pixel. No other corrections were made to the final weights.
Total exposure time : 18424 sec Mean background level : 1.08e-3 cts/s/pixel (excluding airglow lines) Wavelength Continuum S/N 2200-2640A: ~20 per pixel in the continuum 2700-2900A: ~45 per pixel in the continuum
In most cases, target acquisitions were done without peaking up on the QSO. The standard CCD target acquisition acquires targets to an accuracy of better than 0.01 arcsec, which is 5% of the slit width. Typical thermal drifts over a mean 25 minute exposure are expected to be negligible. Frequent calibration lamp observations insure that the wavelength scale is accurately calibrated.
Because the observations were pointed near the pole of the HST orbit, the change in velocity of HST during an exposure was no more than 5 km/s. On-board processing corrects for this shift to a precision of 5 km/s in this observing mode. The final wavelength scale has been reduced to the heliocentric frame.
The spectrum was flatfielded using a preliminary NUV flat derived from internal calibration lamp observations. This flat has been verified on observations of spectrophotometric standards. The flatfielding was carried out by the standard STSDAS "calstis" pipeline. The normal "Doppler smoothing" option was omitted during the pipeline processing. This is not expected to degrade the data, due to the small on-board doppler correction.
Wavelength correction was carried out via standard procedures in calstis, using contemporaneous observations of the internal calibration lamp.
The mean background in the STIS exposures is very low (~1x10^-3 cts/s/pixel), and is almost entirely due to phosphorescent emission from the detector window. The mean background level varies with time, and the time dependence varies with position across the detector. To subtract the background from each exposure, a low-order two-dimensional polynomial fit was performed to pixels at both sides of the spectrum. Each of the fit regions covered the whole width of the detector, and one hundred pixels perpendicular to the dispersion direction. This fit was subtracted from the data prior to extracting the one-dimensional spectra. As over 2.0x10^5 pixels were used in the fitting process, this is assumed to be a negigible source of error. Hence the error array from the original (gross) counts was carried through this process and used in the computation of the final errors due to the one-dimensional extraction.
4. One-dimensional spectral extraction
The spectrum was extracted using the standard version of the STSDAS stis.x1d task. After preliminary testing, a spectral extraction box of five pixels in the cross-dispersion direction was found to give a good compromise between extracted signal and error, resulting in a peak S/N. The spectrum was strong enough to permit centring of the extraction box by the x1d task. After this centroiding step, the spectrum was extracted using a standard "trace" from the calibration database (the 1DT file). This trace corrects for instrument- induced distortions of the spectrum and is determined from observations of standard stars. At each wavelength, pixels within this extraction box were summed together with uniform weights. The S/N of the final spectrum could possibly be improved by extracting with "optimal" weights, though our choice of a small extraction box probably comes close to an "optimal" result.
Conversion of counts to flux was also carried out in the x1d task. Fluxes were corrected for the wavelength-dependent throughput of the 52x0.2" aperture, and for the light lost outside the extraction box, using the standard values in the APT and PCT reference files.
At the end of one-dimensional extraction, we needed to combine the twelve separate exposures in a way that maximizes the signal-to-noise ratio of the final spectrum. This "optimal" combination was carried out using a prototype version of the STSDAS routine "splice". This program uses weighting arrays that specify the weight to be used when averaging fluxes from spectra that overlap in wavelength. We chose to weight the flux in each spectrum by the (S/N)^2 per pixel, smoothed over five pixels to reduce sharp variations in weighting value from pixel-to-pixel. A test using combined, unweighted, spectra shows that the use of smoothed weights does not affect the line profiles significantly.
As noted above in step 3, the errors carried through to the final one-
dimensional extracted spectra are based on the counts prior to the two-
dimensional background subtraction. Therefore no other corrections were made
to the final weights.
The following are the main characteristics of the final data products for the STIS G430M data:
Wavelength coverage: 3025 A to 3565 A.
Total exposure time: 54,892 sec (~15.2 hrs)
(29,292 sec for cenwave=3165 A, and
25,600 sec for cenwave=3423 A)
Dispersion: 0.28 A/pix
Wavelength resolution: ~0.5 A
S/N ~5 to 15 per pixel.
Observations were taken with a CR-split of 2 to aid cosmic-ray rejection. The spectrum was also 'dithered' by about 2.5arcsec (50 pixels) in the cross-dispersion direction in order to be able to correct for the numerous hot-pixels in the data reduction stage. Frequent calibration lamp exposures were taken to ensure proper wavelength calibration. Since the observations were made near the pole of the HST-orbit, the change in the velocity of HST is less than 5 km/sec. Since the velocity resolution of the observations is about 50 km/sec, the smearing to to the change is the velocity of HST is not significant.
Follow this link for a listing of the individual datasets, including exposure times, that were used used to create the final combined spectra.
Wavelength correction was carried out via standard procedures in calstis, using contemporaneous observations of the internal calibration lamp. Comparison with the UCLES optical spectrum of the QSO (Outram et al. 1998) shows agreement to 1/5th of the resolution element of the G430M spectrum.
The spectrum was extracted using a modified version of the STSDAS stis.x1d task. A profile of the spectrum was made in the cross dispersion direction which shows almost all the flux within 5 pixels. Spectra were extracted using different extraction heights, but the best S/N was obtained from an extraction slit height of 5, which was subsequently used for the analysis. The S/N of the final spectrum could possibly be improved by extracting with "optimal" weights.
Conversion of counts to flux was also carried out in the x1d task. Fluxes were corrected for the wavelength-dependent throughput of the 52x0.2" aperture, and for the light lost outside the extraction box, using the standard values in the APT and PCT reference files.
At the end of one-dimensional extraction, the spectrum consists of 59
separate pieces, which need to be summed together in a way that
maximizes the signal-to-noise ratio of the final spectrum. We chose
to weight the flux by the (S/N)$^2$ vs. pixel. Since the spectrum is
mostly readnoise dominated, the weighting is thus proportional to
square of the exposure time. This was done by introducing a separate
keyword into the header, exp2. The the weighting in the 'splice' task
was set to this 'header keyword' to apply the appropriate weight.
WARNING:Please read the section below on the window reflection before trying to identify or analyze objects close to the quasar!!!
We present the imaging observations made with the Space Telescope Imaging Spectrograph of the Hubble Deep Field -- South. The field was imaged in 4 bands, a clear CCD bandpass for 156 ksec, a long-pass filter for 22-25 ksec per pixel typical exposure, a Near-UV bandpass for 23 ksec and a Far-UV bandpass for 52 ksec. The clear visible image is the deepest observation ever made in the UV-optical-NIR wavelength region. The field contains QSO J2233-606, the target of the STIS spectroscopy, and extends 50 arcsec for the visible images, and 25 arcsec for the ultraviolet images. We present the images and catalogs of objects in the field.
The images and catalogs presented here were taken with STIS as part of the HDF-S campaign between 1998 September 29 and October 10. For a description of the HDF-S, see Williams et al (1999), and on the web at http://www.stsci.edu/ftp/science/hdf/hdfsouth/hdfs.html. For more information about STIS, see Kimble et al (1998), Woodgate et al (1998), and Walborn & Baum (1998). This README file provides a preliminary description of the observations, the data reduction and the catalog. A more complete description will appear in Gardner et al (1999).
The images presented here were taken in 4 different modes, FUVQTZ, NUVQTZ, F28X50LP, and 50CCD. The FUVQTZ and NUVQTZ used the MAMA detectors as imagers with the quartz filter. The MAMA field of view is a square, 25 arcseconds on a side, and was dithered so that the observations include data on a field approximately 30 arcseconds square. The 50CCD is filterless imaging with a CCD. The field of view is a square 50 arcseconds on a side, and the dithering extends to a square 60 arcseconds on a side. The F28X50LP is a long-pass filter which vignettes the field of view of the CCD to a rectangle 28 by 50 arcseconds. The observations were dithered to image the entire field of view of the 50CCD observations, although the exposure time "per pixel" is thus approximately half the total exposure time spent in this mode. The original pixel scale is 0.0244 arcsec/pixel for the MAMA images, and 0.05071 arcsec/pixel. All of the images have been drizzled onto a new scale of 0.025 arcsec/pixel. Table 1 describes the observations. The filterless 50ccd observations correspond roughly to V+I, and reach a depth of 29.4 AB magnitudes at 10 sigma in a 0.2 square arcsecond aperture (320 drizzled pixels). This is the deepest exposure ever made in the UV-optical-NIR wavelength region.
Table 1 -- description of the observations Mode Cent Wave FWHM(A) Det. FOV Tot FOV Tot Exp Exp./pix Depth FUVQTZ 1370A 320A 25"x25" 30" 52124 sec same 27.8 NUVQTZ 2220A 1200A 25"x25" 30" 22616 sec same 27.5 F28X50LP 7230A 2720A 28"x52" 60" 49768 sec 22494 27.4 50CCD 5850A 4410A 52"x52" 60" 155590 sec same 29.4 Notes to Table 1: See Walborn & Baum for filter tracings. The detector fields of view has been clipped slightly to remove vignetting. The total fields are approximate. Depths are AB magnitudes at 10 sigma in a 0.2 square arcsecond aperture.
The QSO is at RA=22 33 37.5883, Dec=-60 33 29.128 (J2000). The errors on this position are estimated to be less than 40 milli-arcseconds (Zacharias et al 1999). The position of the QSO on the 50CCD and F28X50LP images is x=1206.61, y=1206.32, and on the MAMA images is x=806.61, y=806.32.
Test observations of the field were made on 1997 October 29 through 31. These data are not used in the present analysis.
The observations were made under the following proposal ID numbers:
7633 Test Data 1997-Oct-29 through 1997-Oct-31 8071 PSF observations 1998-Sep-20 and 1998-Oct-19 8071 Flanking Fields 1998-Sep-27 through 1998-Oct-29 8058 Visits 5-15 1998-Sep-28 through 1998-Sep-30 8073 Visits 17-23 1998-Oct-01 through 1998-Oct-02 8074 Visits 24-30 1998-Oct-02 through 1998-Oct-04 8075 Visits 32-39 1998-Oct-04 through 1998-Oct-06 8076 Visits 40-50 1998-Oct-06 through 1998-Oct-09The detailed phase 2 proposal, formatted listing of the observations, and archive information can be obtained from http://presto.stsci.edu/public/propinfo.html. The raw and pipeline processed data are non-proprietary, and are available through the STScI archive.
The images were dithered in RA and Dec in order to sample the sky at the sub-pixel level. In addition, variations in rotation of about +/- 1 degree were used to provide additional dithering for the WFPC2 and NICMOS fields during the STIS spectroscopic observations. The STIS imaging observations were interspersed with the STIS spectroscopic observations, therefore all of the images were dithered in rotation as well as RA and Dec. Imaging with the CCD was only done in the dark part of the orbits, while the bright part was typically spent doing G430M spectroscopy. The STIS MAMAs were only used in orbits which did not cross the South Atlantic Anomaly at any point in the orbit.
2.5. CR-SPLIT and pointing strategy
The CCD exposures were split into 2 or 3 "CR-SPLITS" which each have the same RA, Dec and rotation. This facilitates cosmic ray removal, although as discussed below, this was only used in the first iteration of the data reduction. The final 50ccd image is the combination of 193 readouts making up 67 "CR-SPLIT" pointings. The final F28X50LP image is the combination of 66 readouts making up 23 "CR-SPLIT" pointings. The F28X50LP image included 12 pointings at the northern part of the field, 1 pointing at the middle of the field, and 10 pointings at the southern half of the field.
In order to allow for PSF subtraction of the QSO present in the center of the STIS 50CCD image, two SAO stars of about 10mag were observed in the filterless 50ccd mode before and after the main HDF-S campaign. The stars are SAO 255267, a G2 star, and SAO 255271, a F8 star, respectively. These targets have spectral energy distributions in the STIS CCD sensitivity range similar to that of the QSO. For each star, 32 different CR-SPLIT exposures were taken (PID 8071, visits 1 and 13 respectively). The following strategy was used: (i) four different exposure times between 0.1 s and 5 s for each CR-SPLIT frame, to ensure high signal-to-noise in the wings while not saturating the center; (ii) a four-position dither pattern with quarter-pixel sampling and CR-SPLIT at each pointing with each exposure time; (iii) use of gain=4, to insure no saturation in the A-to-D conversion. During the observations for SAO255267, a failure in the guide star acquisition procedure caused the loss of its long-exposure (5s) images. Gain=4 has a well-documented large scale pattern noise that has to be removed, e.g., by Fourier filtering, before a reliable PSF can be produced. We have not reduced these observations. The raw and pipeline-processed data are available through the archive.
2.7. Flanking Field observations
The flanking field observing program (Lucas et al 1999) was designed to use WFPC2 to image a contiguous region containing the main deep fields of all three instruments. STIS was used in parallel to these observations to obtain 5100sec images in 50CCD mode. In general, there is very little overlap between the STIS flanking field observations and other HDF-S data. The exposures were made up of 4 dither positions, each CR-SPLIT into 2, for a total of 8 readouts. These images were reduced in a manner similar to that of the main field. We have not checked the astrometry on the STIS flanking field data. Therefore, the world coordinate system in the headers is only accurate to the astrometry of the Guide Star Catalog, i.e. of order 0.5-1.0 arcsec. We have not yet cataloged the flanking field observations. Astrometry and catalogs will be made available through this site when they are available.
There were an additional 9 orbits of data in 50CCD mode taken with the STIS centered on the NICMOS deep field. These data are presented with the NICMOS observations (Fruchter et al 1999).
1. Bias, Darks, Flats and Masks
Standard processing of CCD images involves bias and dark subtraction, flatfielding, and masking of detector defects. The bias calibration file used for the HDF-S was constructed from 285 individual exposures, combined together with cosmic-ray and hot-pixel trail rejection.
The dark file was constructed from a "superdark" frame and a "delta" dark frame. The superdark is the cosmic-ray rejected combination of over 100 individual 1200 second dark exposures taken over the several months preceeding the HDF-S campaign. The delta dark adds into this high S/N dark frame the pixels that are more than 5-sigma from the mean in the superdark-subtracted combination of 14 dark exposures taken during the HDF-S campaign. Calibration of the images with this dark frame removes most of the hot pixels but still leaves several hundred in each image.
An image mask was constructed to remove the remaining hot pixels and detector features. The individual cosmic-ray rejected HDF-S 50CCD exposures were averaged together without registration. The remaining hot pixels were identified with the iraf "cosmicrays" task. These pixels were included in a mask that was used to reject pixels during the "drizzling" phase. Pixels that were more than 5-sigma below the mean sky background were also masked. The unilluminated portions of the detector around the edges were also masked out. The 30 worst hot pixel trails were also masked. These are features that run along columns caused by high dark current in a single pixel along the column.
Flatfielding was carried out by the STSDAS calstis pipeline using two reference files. The first, the "PFLAT" corrects for small-scale pixel-to-pixel sensitivity variations. This file was created from ground-test data but comparisons to a preliminary version of the on-orbit flat revealed only a few places where the difference was more than 1%. The PFLAT corrects for small-scale variations, but is smooth on large scales. The CCD also shows a 5-10% decrease in sensitivity near the edges due to vignetting. This illumination pattern was corrected by a low-order fit to a sky flat constructed from the flanking field observations.
After pipeline processing, the CCD images were reduced using the IRAF/STSDAS package "dither", and test versions called "xdither", and "xditherII". These packages include the "drizzle" software (Fruchter & Hook 1998; Fruchter et al 1998; Fruchter 1998). We used "drizzle" version 1.2, dated 1998 February. The beta versions differ from the previously released version primarily in their ability to remove cosmic rays from each individual readout, and include tasks which have not yet been released.
The rotations used in combining the data were determined from the ROLL_AVG parameter in the jitter files, using the program, "bearing". We did not seek to improve on these rotations via cross-correlation or any other method. However, since we did use cross-correlation to determine the X and Y shifts, small errors in rotation would be removed in this iterative process, to first order.
Determination of the X and Y shifts was done with an iterative procedure. The first iteration was obtained by determining the centroid of the bright point source just east of the QSO, using the pipeline cosmic-ray rejected "crj" files. We could not use cross-correlation in this first iteration, since the very bright star on the southern edge of the field was present on images taken at some, but not all, dither positions, which corrupted the cross-correlation. The source we used for centroiding was clearly visible on all of the 50CCD and F28X50LP frames.
Using these shifts (which were accurate to better than 1 pixel), we drizzled the "crj" files onto individual outputs, which we combined using a median in the task "imcombine". This output image was placed back in the frame (x, y, rotation) of each individual read-out using the xdither task, "blot". After cosmic ray rejection, the blotted image was cross-correlated with the original data, corrected according to the cosmic ray mask, to determine the X & Y shifts used in the final combination.
Each readout, ("flt" file science extension), was compared to the blotted image, and a cosmic-ray mask was created from all of the pixels which differed (positively or negatively) by more than a given threshold from the blotted image. In the version 1.0 released 50CCD image, this threshold was set to be 5.0 sigma. However, we believe that a small error in the sky level determination, introduced by the amplifier ringing correction discussed below, meant that this level corresponded to a "real" rejection at approximately the 3.0 sigma level. The cosmic ray masks were multiplied by the hot pixel masks discussed above, and resulted in about 8% of the pixels being masked as either cosmic rays or hot pixels. This is, perhaps, overly conservative. A less conservative cut (correcting the error in the sky value) would result in slightly higher exposure time per pixel, and thus an improvement of 1-2% in the signal to noise ratio.
This problem with the sky value was corrected in the F28X50LP image, and a 3.0 sigma level was used in the cosmic ray rejection.
4. Amplifier ringing correction
Horizontal features due to amplifier ringing, varying in pattern from image to image, were present in most of the STIS CCD frames. These were caused by the fact that when a pixel saw a very high signal, the bias level was depressed in the readout for the next few rows. The very high signals causing this ringing came from hot pixels and from the saturated QSO. We removed them with an iterative procedure that subtracted on a row-by-row basis, from each individual image, the weighted average of the background as derived from the innermost 800 columns after masking and rejecting "contaminated" pixels. The masks included all visible sources, hot pixels and cosmic-ray hits. For each readout the mask for the sources was obtained from the "master blot image" derived by combining together all images available in the filter, blotted to the position of the considered frame. A "minmax" rejection criterion was applied to the remaining pixels, which excluded the highest and lowest 6% of the unmasked pixels from the average. The IRAF script implementing this procedure, developed for the HDFS, will be made available to the community.
Heavily smoothing the images reveals very slight horizontal residuals which were not removed by the current choice of smoothing/rejection parameters.
The final image combination was done by drizzling the amplifier-ringing corrected pipeline products together onto a single output image. The exposures were weighted by the square of the exposure time, divided by the variance, which is (sky+rn^2+dark). The rotations were corrected so that North is in the +y direction, and the scale used was 0.492999 original CCD pixels per output pixel so that the final pixel scale is exactly 0.025 arcsec/pixel. For the 50CCD data we used a pixfrac=0.1, which is approximately equivalent to interleaving, where each input pixel falls on a single output pixel. For the F28X50LP data we used pixfrac=0.6, as a smaller pixfrac left visible holes in the final image. See Fruchter & Hook (1998) for a discussion of the meaning of the drizzle parameters. The final image is given in counts per second, which can be converted to STMAG using the PHOTFLAM parameter supplied in the header. We also supply the weight image, which is the sum of the weights falling on each pixel. For the F28X50LP image, we supply an exposure-time image, which is the total exposure time contributing to each pixel. We have multiplied this image by the area of the output pixels. The world coordinate system in the headers was corrected so that North is exactly in the +y direction, and the pixel scale is exactly 0.025 arcsec/pixel. However, we have provided the original header information in a separate file.
A window in the STIS CCD reflects light from bright objects slightly out-of-focus to the +x, -Y direction (SE on the HDF-S images). The window reflection of the QSO (which is saturated in every 50CCD and F28X50LP exposure) is clearly visible in the F28X50LP image, but has been partially removed from the 50CCD image by the cosmic-ray rejection procedure. We wish to emphasize that it has only been partially removed -- there remain residuals. These residuals should not be mistaken for galaxies near the QSO, nor should they be mistaken for the host galaxy of the QSO. There is additional reflected light from the QSO (and from the bright star at the southern edge) evident in the images. It is the belief of the authors of this document that the version 1.0 released images are not appropriate for searching for objects very close to or underlying the QSO, and that such a search would require re-processing the raw data with particular attention paid to the window reflection, other reflected light, and to the PSF of the QSO. The diffraction spikes of the QSO are smeared in the final images by the rotation of the individual readouts. The QSO is not saturated in the MAMA images.
Currently, geometrically corrected NUVMAMA/F25QTZ and FUVMAMA/F25QTZ frames do not have the same plate scale; thus, before registration, all near-UV and far-UV frames were geometrically corrected, rescaled to 0.025'' pix-1, and rotated to align North with the +y image axis. The roll angle specified in the jitter files was used to determine the relative roll between frames, and the mean difference between the planned roll and the jitter roll determined the absolute rotation. All near-UV and far-UV frames were then cross-correlated against one of the far-UV frames (o52f44uwq) to provide shifts in the output coordinate system. Note that centroiding on the quasar in all far-UV and near-UV frames yields the same shifts as cross-correlation, within 0.1 pixel. The calibrated frames were then drizzled, including the above corrections, rescaling, rotations and shifts, to a 1600x1600 pixel image. The world coordinate system in the image headers has been updated to exactly reflect the plate scale and alignment, and also the NRL position of the quasar.
For both the far-UV and near-UV frames, individual pixels in each frame are weighted by the ratio of the exposure time squared to the dark count variance; this weights the exposures by (S/N)2 for sources that are fainter than the background. Although the far-UV dark profile is smooth, the near-UV dark profile is an actual sum of dark frames, and so we smoothed the near-UV dark profile when determining the drizzle weights. With this weighting algorithm, pixels in the upper left-hand quadrant of a given far-UV image contribute less when the dark glow is high, and contribute more when it is low. The statistical errors (counts sec-1) in the final drizzled image, for objects below the background (e.g., not the quasar), is given by the final drizzled weights file raised to the -1/2 power.
The drizzle "dropsize" (a.k.a. pixfrac) was 0.6, thus improving the resolution over a dropsize of 1.0 (which would be equivalent to simple shift-and-add). The 1600x1600 pixel format is the smallest ``standard'' image size that can contain all dither positions; pixels outside of the dither pattern are at a count rate of zero. The pixel mask for each near-UV input frame included the occulted corners of the detector, a small number of hot pixels, and pixels with relatively low response (those with values <=0.75 in the high-resolution flat field). The pixel mask for each far-UV frame included hot pixels and all pixels flagged in the data quality file for that frame. When every input pixel drizzled onto a given output pixel has been masked, that pixel has been set to zero.
Casertano, S., et al 1999, BAAS, 193, 75.05, and in preparation. Ferguson, H. C., et al. 1999, BAAS, 193, 75.02, and in preparation. Fruchter, A. S., & Hook, R. N. 1998, astro-ph/9808087. Fruchter, A. S. et al. 1998, in 1997 HST Calibration Workshop, ed. S. Casertano et al (STScI). Fruchter, A. S. 1998, http://www.stsci.edu/~fruchter/dither/dither.html Fruchter, A. S. et al 1999, BAAS, 193, 75.04, and in preparation. Gardner, J. P. et al. 1999, BAAS, 193, 75.03, and in preparation. Kimble, R. A., et al. 1997, ApJ, 492, L83. Lucas, R. A., et al. 1999, BAAS, 193, 75.06, and in preparation. Walborn, N., & Baum, S. 1998, STIS Instrument Handbook, version 2.0, (Baltimore: STScI), http://www.stsci.edu/instruments/stis. Williams, R. E., et al. 1996, AJ, 112, 1335. Williams, R. E., et al. 1999, BAAS, 193, 75.01, and in preparation. Woodgate, B. E., et al. 1998, PASP, 110, 1183. Zacharias, N., et al. 1999, BAAS, 193, 75.09, and in preparation.
This page was last updated on November 24, 1998.
Copyright Notice.