Overall, the BCKGRD routine does a good job of following these changes, especially in terms of the average interorder noise rms values, which is BCKGRD's reference figure of merit for judging aberrant fluxes and fitting errors. The scale of these changes happens to be largest in the short-wavelength region where the problem of interorder crowding is also the most severe. The evolution of the null surface relative to the fixed zeropoints of the FN unit system (defined by the ITF for a single epoch) results in BCKGRD's producing progressively lower background levels near the Ly line in late-epoch images of Sco. Yet, despite the comparatively large scope of this problem, the background-determining algorithm is not the primary source of the time-dependent background errors in this region of the camera. Rather, the errors are an indirect consequence of a rapid change in the null levels over a small area of the camera. The uneven loss of the null image flux, paradoxically, permits the intermediate degree Chebyshev polynomials used by BCKGRD to fit more readily the new, more evenly decayed surfaces. Consequently, better solutions in the short-wavelength region are produced late in the mission.
The original time-dependence of the background fluxes found by Massa, cast in terms of a time-dependent error of background fluxes for Sco images, is neither confined only to the background extractions nor to high-dispersion images. The progressive changes of the zeropoints of the ITF flux scale have a complicated, but generally smooth, dependence on image position and undoubtedly affect the extracted background and gross fluxes for all SWP images, albeit better handled for single-order, low-dispersion spectra. Consider additionally that the change of the null surface has a spatial dependence, whereas only a wavelength dependence was assumed in the calibration of the absolute fluxes (Cassatella 1996). These different dependences almost insure that errors will be introduced in the calibration of high-dispersion spectra which manifest themselves in the ripple correction step of processing.
Lastly, we believe that certain types of ``customized extractions," e.g., the Bianchi & Bohlin technique, may alleviate the problem of background extraction for specific applications, but probably only if the investigator recognizes that such corrections depend upon the epoch of the observation and preferably if he/she can find saturated absorption features in the same region of the spectrum. Empirical corrections might then be accurate, but they too depend upon spectral type and epoch. Even these correction techniques can be problematical for accurate background extractions at the toward short wavelengths where the background surface changes rapidly with position and time. However, such problems can probably be reduced by relying on empirical corrections of contemporaneous spectra of stars with saturated lines. A growing collection of archival data is becoming available from new-generation UV spectrometers such as ORFEUS, GHRS, and (shortly) FUSE which might lend itself to this purpose.
We are grateful to Dr. Joy Nichols for many technical discussions which have enhanced the quality of this investigation and Dr. Derck Massa for providing us with a list of the SWP images he used in his original analysis. It is also our pleasure to thanks Dr. Catherine Imhoff for her special processing of the null images discussed in 4 and Mr. Randall Thompson for assistance with technical aspects of IDL-programming. We are grateful to improvements in the original manuscript suggested by Drs. Imhoff, Nichols, and Jeff Linsky. We also thank an anonymous referee whose comments also visibly enhanced its quality. We appreciate several suggestions by Dr. Linsky which have led to the inclusion of 5 into the revised manuscript. The author is indebted to Dr. Carol Grady and Mr. Carl Cox for writing initial driver codes to compute smoothed Chebyshev solutions for background spectra.