1. Introduction and History of Background Problem

The extraction of background fluxes from high-dispersion International Ultraviolet Explorer (IUE) echellograms is the bane of all experienced users of data from this satellite. The accurate determination of the background spectrum, the subject of this paper, is important because it provides the first correction to the extracted ``gross" fluxes, resulting in ``net" spectroscopic fluxes which an investigator utilizes for scientific analysis. The difficulty of extracting these fluxes from IUE data arises from several causes, which are listed as follows:

(1) various electro-optical properties of the SEC vidicon (television) tube in each of the three IUE cameras changed with time. This caused a complicated time-dependence in the zeropoint and sensitivity which must be characterized empirically. These effects include the null-flux level, sensitivity to target illumination, shifts of the echelle format, and Cerenkov radiation as high energy particles from the outer Van Allen belt passed through the camera window;

(2) in any echelle spectrogram orders crowd toward short-wavelengths with the inverse square of the order number. Typical order numbers for the IUE are m $\approx$ 100, and the range of orders covered is about 60. Thus, for the SWP camera the spacings between orders varies by a factor of 3.6. In addition, scattering due to grating-groove irregularities increase as a high power of 1/$\lambda$ (Cardelli, Ebbets, & Savage 1990);

(3) the pixel Point Spread Functions (PSFs) may have different forms in the spatial and spectral directions. Typical PSFs are known to have an extended ``toe" from halation effects (Coleman et al. 1977; see also our Fig. 3). A scattered light component from the cross-dispersing grating has also been measured in reflected solar light (Clarke 1981, Crivelli & Praderie 1982), but the unmeasured scattered light contribution in the spectral direction from the echelle grating is probably of comparable importance (see Cardelli, Ebbets, & Savage 1993, for a discussion of its effect in GHRS spectra from the Hubble Space Telescope). This component means that even if the interorder component were completely removed the fluxes of saturated lines would not reach zero intensity;

Figure 3: Crosscut of background fluxes from a central Pass 1 data swath through a high-dispersion SWP image. The triangular region describes the local background fluxes in the Interorder Overlap Region where crowding of echelle orders is severe. The halation fluxes are indicated in the line range 388-590. The thin squiggly line and squares show the raw fluxes from the extraction, and the small crosses depict the working fluxes corrected for interorder contamination from an on-the-fly Point Spread Function. The solid line is the Pass 1 solution, which is 7th degree Chebyshev polynomial. Points nc, nf, & nd, indicated schematically by dashed lines, represent the three vertices of the ``interorder overlap region" discussed in the text. Arrow indicates the position of the order containing Ly $\alpha$.

(4) the image boundary is circular. A curved boundary limits the extraction of raw fluxes in the spatial direction to a small range of orders. This effect is most telling on the short wavelength side of the echelle blaze function where the efficiency is still high;

(5) nonuniformities in electron charge accumulation near the image boundary ring. These nonuniformities vary from camera to camera and with time. They include target ring glow both from reflected electrons on the edge of the SWP camera image and static discharges. One such mild discharge present early in the mission developed in 1983 into the well-known ``flare," which curtailed usage of the LWR camera;

(6) sudden changes in the pixel-to-pixel noise amplitude across the image;

(7) fixed point and transient (``cosmic ray," microphonic) defects on some images.

The IUE Project has dealt with this array of problems in a variety of ways. The ``IUESIPS" (IUE Spectral Imaging Processing System; see Turnrose & Thompson 1984) software extracted flux from pseudo-pixels locally along an interorder path adjacent to each echelle order. These raw fluxes are then smoothed through a succession of 63-point median and two 31-point mean filters (Turnrose, Bohlin, & Harvel 1979). The resulting filtered fluxes are adopted as background fluxes. The IUESIPS background-extraction scheme was criticized by Grady et al. (1989) who demonstrated that image artifacts such as a cosmic rays can produce spurious absorptions features in the net spectrum, e.g., as the result of smoothing over bright spots, which are comparable to the width of photospheric lines. In addition, and perhaps most importantly, the extraction of local flux from short-wavelength orders of the SWP camera of high dispersion images resulted in systematic overcorrections of fluxes, and often large negative fluxes, for deep absorption features. This effect results from IUESIPS's completely neglecting the overlap of echelle order flux onto pixels in the interorder lanes from which the local background is sampled. Two remedies were suggested to overcome these shortcomings. To avoid the generation of false features through the smoothing of artifacts, Grady, Smith, & Garhart (1989) rejected the filtering approach and suggested a fitting a polynomial function to extracted fluxes. Orthogonal polynomials were used because they are easy to compute and require no fittings of derivatives or weightings of special node points. These authors recommended Chebyshev polynomials in particular because any residual errors caused by narrow features with this function tend to be relatively distributed across the entire wavelength domain (Press et al. 1992). However, tests with other orthogonal functions in the development of our software have shown only small differences among different functions of a given degree.

A more empirical approach was taken by Bianchi & Bohlin (1984) who measured depths of saturated absorption features, i.e. those expected to have virtually no flux at line center, in IUE and Copernicus spectra of four OB stars. These authors determined an interorder flux-overlap correction factor as a function of wavelength. Although clearly a good first step, the Bianchi & Bohlin approach as implemented is not suitable for a variety of reasons (nor was it meant to be) for producing background fluxes for a general data archive : (1) the correction factors in this study suggested a dependence on spectral type which was not subsequently evaluated, (2) the technique assumes overlap corrections are only a function of wavelength, whereas in echellograms the fluxes are extracted from a two-dimensional region, (3) the flux corrections are likely to be dependent on a variety of observational parameters (e.g., time, camera-head temperature, focus, particle flux at the satellite, and target spectral type) the effects of which have not been calibrated. For example, the data sample given in the Bianchi & Bohlin paper consists of only four images, of which three are large aperture and one is small aperture; the effects of the difference in aperture were not evaluated, (4) the Copernicus archive containing the reference spectra is not comprehensive in spectral type or wavelength coverage. Despite a needed evaluation for targets with different spectral distributions, the Bianchi & Bohlin corrections have been used for high-dispersion SWP images with success for stars of a limited range in spectral type (e.g., Diplas & Savage 1994) and also have been used in routine processing of many high-dispersion SWP images from objects having a wide range of spectral types by the University of London's IUE Data Reduction IUEDR processing system (Giddings & Rees 1989).

The production of the IUE Final Archive (``NEWSIPS;" see Nichols & Linsky 1996 for details) for some 40,000 high-dispersion images required that any flux extractions be carried out with an automated data-processing system and that they be carried out with a minimum of pre-conditioning of the data (Garhart et al. 1997) To address these requirements, the author developed a subprogram called BCKGRD within NEWSIPS for extracting background fluxes from interpolation of the global properties in the on-the-fly processing of each image. For echellograms of sources with weak or no continua the routine samples local fluxes in the interorder lanes of each order. However, for the majority of high-dispersion images with strong continua, the algorithm extracts interorder fluxes first in the spatial direction. It then uses these solutions for a second, final pass in the spectral directions. Fits are performed to these extractions by means of a Chebyshev function of intermediate degree.

The current analysis was motivated by a report by Massa et al. (1998) of a time-dependence in the apparent strengths of the Lyman $\alpha$ absorption line in some 71 SWP high-dispersion spectra of the calibration star $\tau$ Scorpii (B0.2V). Each of these spectra was exposed for 6 seconds. These exposures are short enough that exposure-dependent blemishes (cosmic rays) only rarely complicate a background solution. The Lyman $\alpha$ core in this star's sharp-lined spectrum is determined by both by the photospheric interstellar components, so it should show no flux. Massa demonstrated that the line cores of this line in NEWSIPS data change from positive to slightly negative values at early (1979) and late (1995) epochs of the IUE mission. This trend is shown in Figure 1 for a subsample of the images used in his original analysis. The total range of spurious changes in this line is about 15% and occurs primarily during the late stages of the IUE mission. This is a highly significant trend (equal to several times the rms fluctuations of the locally sampled continuum). Discussions with Dr. Massa led to the conclusion that these time-dependent errors were caused by systematic changes in the determined background fluxes by NEWSIPS. This paper discusses the reasons for this circumstance and to make users aware of even more fundamental problems which have the potential to produce false time-dependences in all flux extractions of short-wavelength orders of the SWP camera, for low-dispersion as well as high-dispersion spectra.

Figure 1: Plot of fractional core depths of the Lyman $\alpha$ line in a sample of SWP high-dispersion spectra of $\tau$ Sco. The core depths should be constant (near zero), so the trend is spurious and is the subject of this paper. To convert the ordinate values to ``FN" units given by NEWSIPS, multiply them by 106.

We proceed by first reviewing in $\S$2 the general process of the background estimation scheme for continuum sources in high-dispersion images by the NEWSIPS module, BCKGRD. Readers wishing a more detailed description may consult the Appendix A of Garhart et al. (1997) or its posting on the Web at the URL address  http://archive.stsci.edu/iue/newsips/bckgrd/. A discussion of proof of concept tests has also been published (Smith 1990a). In $\S$3 we describe the analysis of a subgroup of the Massa et al. sample of images of $\tau$ Sco distributed in time and discuss the reason the BCKGRD algorithm produces time dependent errors. In $\S$4 aspects of an already known ``drift" of null-image fluxes with time are discussed. This drift has a crucial bearing on background flux solutions of $\tau$ Sco and other science images. Thus, in $\S$5 we extend the discussion to provide quantitative estimates of the effects of null drift on the continuum fluxes of SWP high-dispersion images of this star.