[Top] [Prev] [Next] [Bottom]

32.7 Flatfield Calibration

The structure in the FOS flatfields is dominanted by photocathode granularity. Target location on the photocathode and time changes in the photocathode itself are the dominant source of uncertainty in the FOS flatfields. The flatfields limit the signal-to-noise achievable (in the limit of infinite counting statistics) and the reliability with which line features can be identified. The major factors affecting the accuracy with which the FOS flat field reference files will correct the flatfield granularity in your observations are:

32.7.1 Flatfield Summary

Since photocathode granularity is the dominant component of FOS flatfield structure, factors which affect the illumination of the photocathode dominate the degree to which a calibration flatfield is appropriate for your data. Given the small spatial scale of photocathode variation, the best flatfield correction is obtained only for those targets that were acquired with accuracies comparable to those used in STScI calibration programs. Secondly, given the temporal variation present in some spectral regions, the accuracy of flatfield correction is also affected by the ability of flatfield calibration observations to track the variation.

The overall flatfield correction generally has an accuracy of better than 3%. Features in certain specific dispersers, apertures, spectral regions, and times of observation can be somewhat poorer as changes may not have been adequately tracked by calibration observations. Limiting accuracy of order 1% is possible for observations taken close to flatfield calibration observations and for which nearly exact alignment of photocathode granularity is accomplished (see Figure 32.23 as an example of some of the best flatfield correction possible with the FOS).

High-precision (S/N > 30) spectroscopic measurements with the FOS required that the observations sample the same portion of the photocathode as did the flatfield calibration observations. This sampling repeatability was limited by the target acquisition accuracy (for calibration and high S/N observations, typically 0.04" which corresponds to 0.025" in each coordinate) and filter-grating wheel positional repeatability (approximate 1 deviation of 0.04"). The photocathode granularity could vary by 5 to 10% over distances as small as 0.2". Small x-offsets, typically 1 pixel or less, could occur in the granularity pattern from one observing epoch to another. Paired aperture granularity was substantially different from that for the single apertures.

Post-COSTAR extended source flatfielding can be improved by using the last epoch of pre-COSTAR flatfields for the detector-disperser combination, due to the more extended nature of the pre-COSTAR PSF.

For a limited subset of gratings, mapping of the post-COSTAR granularity in the direction perpendicular to dispersion is available at one epoch which you can examine by reference to the special set of drift-scan FOS calibration observations made in calibration program 6916 (December 1996 and January 1997).

Working with Flatfields

You should be particularly wary of unusual features in your data. In order to assess the impact of the flatfield used in the data reduction process for any particular FOS observation, and as a careful check:

32.7.2 Flatfield Details

FOS flatfields are prepared from precisely pointed (£0.025" pointing accuracy in each coordinate), high S/N observations (S/N 200 per pixel, typically) of two relatively featureless spectrophotometric standard stars, BD+28D4211 and G191B2B (also known as WD0501+527). All FOS flatfield reference files are actually unity-normalized inverse flatfields, that are applied as multiplicative operators in calfos data reduction to correct for small-scale (less than 10 diode width) differences in the sensitivity. These pixel-to-pixel sensitivity variations resulted mainly from small-scale inhomogeneities in the photocathode (the so-called granularity). A much less important contributing factor was diode-to-diode variation in sensitivity, as this effect was greatly diminished by the OVERSCAN data-taking process which caused each output pixel to contain contributions from five different diodes.

If the appropriate observational strategies were employed, the overall flatfield correction generally has an accuracy of better than 2-3%. However, flatfielding for certain specific dispersers, apertures, spectral regions, and epochs of observation can be somewhat poorer. For example, substantial (5 to 10%) flatfield variation did occur on spatial size-scales as small as 0.2" for several detector and disperser combinations, so that precise flatfield correction is only possible for science targets acquired with the same high pointing accuracy used for calibration observations. The standard recommendation of the FOS team was that any program requiring S/N >30 should use the same target acquisition accuracy used in the STScI calibration observations.

So-called superflats and superflat-derived flats produced by comparison of raw data with superspectra (highly accurate countrate distributions from the FOS gratings obtained via a cross-correlation technique)2 provide the best quality flatfields. Some SV epoch flats and all polarimetry flats were produced with a more subjective, much less accurate technique that involved normalizing a standard star spectrum with a continuum spline fit (see FOS ISR 075). These subjective flats often did not completely remove larger features and features with high local rates-of-change due to the uncertainty in the continuum normalization process required for their preparation.

Time Dependence

The FOS/RD G190H, G160L, and, to a lesser extent, G270H flatfields displayed marked time dependence throughout both the pre- and post-COSTAR periods. Several other gratings (e.g., FOS/RD G400H, G570H, and FOS/BL G130H, G160L displayed varying degrees of generally more modest temporal change. Figure 32.19 illustrates the striking changes in the FOS/RD G190H flatfield that took place within the first two years of operation. The strong feature at approximately 1950 Å. continued until mid-1995 when it began to disappear. By the end of the FOS lifetime, the granularity at that wavelength had returned to near launch levels.

Figure 32.19: FOS/RD G190H Flatfield Changes

Aperture Dependence

All pre-COSTAR flatfields are somewhat aperture dependent as the PSF filled many of the apertures. However, individual aperture-specific flatfield observations were not made for many apertures in the pre-COSTAR period. Little post-COSTAR aperture dependence existed between the single apertures or between the paired apertures for a particular disperser, but substantial differences did exist between single and paired apertures for any disperser because substantially different portions of the photocathode were sampled by the centers of the different aperture sets. Due to these substantial photocathode granularity spatial variations, flatfield reference files derived from single aperture observations should never be used to correct paired aperture observations and vice versa.

Figures 32.20 and 32.21 provide a characterization of the overall photocathode granularity corrections by showing, at low-scale, representative SINGLE aperture flatfields for high dispersion gratings for both FOS detectors.

Figure 32.20: Typical SINGLE Aperture Flatfield Granularity: FOS/BL

Figure 32.21: Typical SINGLE Aperture Flatfield Granularity: FOS/RD

Granularity Shifts Along Dispersion

Small x-shifts of the observed granularity did occur relative to the positions at the standard star epochs due to slight changes from time-to-time in the internal magnetic environment of the detectors. This observed effect was typically +/- 1 pixel or less, with one documented 2 pixel shift. As a result, shifting your observed spectrum relative to the flatfield prior to re-calibration can occasionally improve the granularity correction. Tests performed with standard star data have indicated, however, that small-scale, even sub-pixel, offsets between identical observations can introduce pattern noise at the 1% level. As a result, 1% rms is deemed the practical limit with post-COSTAR superflat-based flatfields, good aperture centering, and precise x-coordinate alignment.

FOS flats are not interpolated between standard star observation epochs. They are each prepared from an individual high S/N observation of a standard star and typically applied over a time-range that is inclusive of the date of standard star observation. As a result, you might be able to improve the flatfielding at a particular epoch by interpolating between two flatfield reference files, but the x-shift pattern noise considerations mentioned above may limit the benefit of this procedure.

Granularity Perpendicular to Dispersion

For a limited subset of gratings, mapping of the post-COSTAR granularity in the direction perpendicular to dispersion is available at one epoch which you can examine by reference to the special set of drift-scan FOS calibration observations made in calibration program 6916 (December 1996 and January 1997). These observations, taken in RAPID mode, sampled the entire height of either the 4.3 or 1.0 aperture at approximate 0.1" intervals (see proposal for details) for several commonly-used gratings. Comparison of the individual readouts allows characterization of the flatfield rate-of-change perpendicular to dispersion. Additionally, the influence of target miscentering (and Y-base misalignment) on wavelength-dependent aperture throughput can be evaluated from these data.

Application of Inappropriate Flatfield

If an incorrect, or inappropriate, flatfield reference file is used to flatfield the data, small emission-like or absorption-like features will appear in the spectrum corresponding to sensitivity variations that were introduced or left uncorrected by the flatfielding. Depending upon the scale of the shift or mismatch, these artifacts can be range from as little as 1-2% up to >20% in size.

Figure 32.22 shows an example of a spectrum that was flatfielded incorrectly. Panel (a) shows the count rate data in the .c5h file that results from using both correct (solid line) and incorrect (dotted line) flatfield files. Panel (b) shows a blowup of the region from pixels 300 to 1300 of the same data.

Figure 32.22: Incorrectly Flat-Fielded Spectrum of the Red G190H Grating

Flatfield Generation Rules

The following general recommendations summarize the guidelines that have been used for the applicability of FOS flatfield reference files and in generating the FOS flatfield reference file set from the various calibration observation datasets. They are provided as a guide for those interested in comparing their observations with other flats and with standard star observations other than those used for the generation of the flats.





Characterization of Flatfields for Individual Gratings

Pre-COSTAR flatfields were generally smoothed with a 3-point boxcar smoothing function. Post-COSTAR flats were not smoothed. As a general rule, features in post-COSTAR flatfields were narrower than their pre-COSTAR counterparts by an amount greater than the influence of the smoothing function.

The descriptions of various flatfields that follow are generally restricted to the SINGLE apertures unless otherwise indicated:



Figure 32.23: Raw and Flatfield-Corrected FOS/RD G270H

Figure 32.24: Photocathode Blemish at 1500-1560 Å

[Top] [Prev] [Next] [Bottom]

1 Detailed information about flatfields is provided in FOS ISRs 075, 078, 088, 134, 143, and 157.

2 The superflat observational and analysis procedures are explained in detail in FOS ISRs 088, 134, and 157.

3 FOS ISR 134.

Copyright © 1997, Association of Universities for Research in Astronomy. All rights reserved. Last updated: 01/14/98 14:55:10