The original LWP and SWP sensitivity degradation corrections included standard star data taken through early 1992 for SWP point and trailed sources (Garhart 1992) and mid-1993 for LWP point sources (Garhart 1993). In early 1995, concerns were raised about the accuracy of the extrapolations in time of the sensitivity corrections for NEWSIPS processing. While no errors due to the extrapolated corrections were seen in NEWSIPS data, the IUE Project felt that updating the old sensitivity degradation analysis would result in more accurate fluxes for post-1992 and 1993 images. As a result of these concerns, the LWP and SWP sensitivity degradation analysis was updated in 1995 using data taken though the beginning of 1995. These new corrections were applied to post-01 January 1990 images while the old corrections were used for pre-1990 images. This 1995 update represents the first sensitivity degradation "catch-up" calibration. Since the LWR was analyzed as a closed data set, no catch-up calibration was necessary.
Towards the end of 1996, a changing trend in the late epoch sensitivity degradation of the SWP camera was seen (Massa and Fitzpatrick 1996). Absolutely calibrated fluxes of post-1994 data showed systematic errors approaching -5% (i.e., lower fluxes) when compared with images taken prior to 1994. This indicated that the 1995 catch-up calibration was under-correcting the post-1994 data and could occur if the rate of SWP degradation was increasing. In the case of the LWP camera, there was faint evidence for a possible opposite effect, an over-correction of post-1995 data of approximately 1-2%. This situation could occur if the rate of LWP degradation was tapering off. Due to these new trends, the IUE Project decided to explicitly extend the sensitivity degradation analysis to include all available late-epoch data. The following analysis reports the results of this second calibration update.
The SWP low-dispersion point source sensitivity degradation correction analysis was updated to include data taken through August of 1996. This new analysis contains observations of the usual 5 low-dispersion standard stars obtained through October of 1995 as well as observations of the white dwarf standard CD-38° 10980 taken through August of 1996 (i.e., 19th episode data). Five images were excluded (using a list obtained from VILSPA) as they were subject to DMU corruption (SWP56912, 56913, 56914, 66915, and 56139). The images (73 total) used to update the point source sensitivity degradation are listed in Table 1.
Since the standard star data were normalized using fluxes taken during the 1985 epoch, the white dwarf data had to be treated specially, as no observations of the white dwarf were taken at that time. Each white dwarf spectrum was normalized by dividing by an average of several white dwarf spectra taken in a three-month time period beginning in 1991 (i.e., 1991.0-1991.25) and then binned at 5 Angstrom intervals. A similar average of the standard star data taken in this time period was also obtained and ratioed to the average white dwarf spectrum, yielding a white dwarf adjustment factor for each wavelength bin. The normalized white dwarf spectra were then adjusted to coincide with the standard star data by multiplying each individual white dwarf spectrum by the appropriate adjustment factor. As a result of this adjustment, the white dwarf spectra were placed on the same scale as the standard star data, extending the range of the sensitivity degradation analysis to August of 1996.
The set of sensitivity data (binned at 5 Angstrom intervals) was binned in time at six-month intervals. The time-binned ratios were then plotted for each wavelength bin. From these plots, it was clear that the linear fits from the previous analysis (which included data taken through early 1995) no longer applied to the post-1995 ratios of this new data set. Since the SWP sensitivity analysis data set is now a "closed" data set, the binned flux ratios could in principle be fitted with a fourth-order polynomial covering the entire range of observation dates. However, such a polynomial fit would not necessarily join smoothly with the old linear fit for all wavelengths, so it was not practical to use this polynomial for late-epoch corrections. In order to join a new late-epoch correction smoothly to the existing linear fit, a new linear fit was computed, joined to the old linear fit at a specific time and extended to a point defined by the average of the last three time-binned fluxes an average of 1995.5, 1996.0, and 1996.5 bins). This results in a set of segmented linear fits (one for each wavelength bin) which are splicings of the previous linear fits and a late-epoch linear fit.
Tests were performed which moved the "break point" between the original and second linear fit (i.e., the two segments) from 1994 to 1993 and finally to 1992 in order to assess the quality of the fit to the late-epoch data (Figs. 1-15). Checks were performed which computed the differences between the various segmented linear fits and the fourth-order polynomial fit (the polynomial fit was used strictly as a reference "best-fit overall solution"). One check computed the maximum observed difference between the segmented linear fits and polynomial fit, over the entire time range from 1992 to 1996. A second check computed the difference between the segmented linear fits and the polynomial fit at the adopted break point time. A third check computed an average over time of the difference between the segmented linear fits at each individual time bin and the polynomial fit. In each case, the differences were averaged over all wavelength bins. The percentage differences (corresponding to the three above-mentioned checks) between the segmented linear fit and the fourth-order polynomial fit are summarized in Table 2 and are interpreted as being reasonably representative of the expected errors associated with use of the various linear fits.
The 1994 break point correction was a clear loser so no further analysis was performed using that break point. In order to attempt to discriminate between using the 1992 break point and the 1993 break point, the individual differences for each wavelength bin (as opposed to differences averaged over all wavelength bins) were plotted as a function of wavelength, separately for the cases of a 1992 break point and a 1993 break point (Figs. 16 and 17). The 1993 data shows considerably smaller differences in the 1300 Angstrom region than the 1992 data and overall shows differences that are less dependent on wavelength (implying a "greyer" expected error). A final consistency test of the new segmented sensitivity degradation correction using the 1993 break point was performed (see Figs. 18 and 19). Two images (SWP56137 taken in October 1995 and SWP05741 taken in July 1979) were corrected using the new segmented solution and differenced. The same two images were also corrected using the old single linear fit and differenced. The old extrapolated linear fit produced an average difference of -4.8% while the new segmented solution yielded an average difference of +0.8%. Clearly, the new solution does indeed yield results which are closer to zero than the old results.
The SWP low-dispersion trailed sensitivity degradation correction analysis was also updated to include data taken through August of 1995. After this point in time, no additional trailed standard star observations were obtained, and only one Guest Observer requested trailed exposures. The spacecraft switched to one-gyro control mode in March of 1996 which resulted in the loss of the ability to perform trailed observations. Thus, although the data available for the analysis stops at April 1995, it in fact covers most of the time period over which trailed observations could possibly have been taken. Unlike the point-source analysis, the new trailed correction does not include observations of the white dwarf standard, as none were taken. The images (7 total) used to update the SWP trailed sensitivity degradation are listed in Table 3.
The method of analysis applied to the trailed data is the same as used in the case of point source data. Segmented linear fits were produced using the same trial break points (i.e., 1992, 1993, and 1994), except that the endpoint of the second (late-epoch) segment is an average of the 1994.5, 1995.0, and 1995.5 bins. Figures 20-34 display the segmented linear fits for the three break points. Tests of percentage differences between the segmented linear corrections and a fourth-order polynomial correction, identical to the tests used in the point source analysis, were performed on the trailed corrections. The results are summarized in Table 4.
As is the case with the point source test, the 1994 break point correction was clearly worse than the others. Therefore, it was excluded from further analysis. The individual differences for each wavelength bin plotted as a function of wavelength are shown in figures 35 and 36. The 1993 data show smaller differences overall than the 1992 data. A final consistency check of the new correction compared to an extrapolation of the old is demonstrated in figures 37 and 38. Despite the success of the segmented correction for point sources, the segmented trailed source correction yields poorer results than using an extrapolation of the old trailed sensitivity degradation correction. Tests were performed on a total of 21 post-1993 images, and in the majority of the cases the extrapolated original linear fit provided a better correction than the segmented solution.
The LWP low-dispersion point source sensitivity degradation correction analysis was updated to include data taken through August of 1996. This new analysis contains observations of the usual 5 low-dispersion standard stars obtained through February of 1996 as well as observations of the white dwarf standard CD-38° 10980 taken through August of 1996 (i.e., 19th episode data). Nine images were excluded (using a list obtained from VILSPA) as they were subject to DMU corruption (LWP31634, 31665, 32089, 32096, 32500, 32513, 32606, 32621, and 32632). The images (54 total) used to update the point source sensitivity degradation are listed in Table 5.
The method of analysis applied is the same as used in the case of the SWP point source data, except segmented linear fits were produced using slightly different final break points (i.e., 1993, 1994, and 1995), as they were more appropriate in the case of the LWP data. Figures 39-67 display the segmented linear fits for the three break points. Tests of percentage differences between the segmented linear corrections and a fourth-order polynomial correction, identical to the tests used in the SWP point source analysis, were performed on the LWP corrections. The results are summarized in Table 6, which implies that a 1994 break point is best among the four trial break points.
The individual differences for each wavelength bin plotted as a function of wave-length are shown in Figures 68-71. The 1994 data show smaller differences overall than the other break points, reinforcing the results presented in Table 6. The 1994 break point correction was better (less noisy) than the others. Therefore, it was chosen for the final analysis.
A final consistency check of the new segmented correction compared to an extrapolation of the old is demonstrated in Figures 72 and 73. Despite the success of the segmented correction for SWP point sources, the use of a segmented correction for the LWP was not better than using an extrapolation of the old sensitivity degradation correction. Tests were performed on a total of seven post-1995 images, and in each case, although the differences were small, the segmented solution was in fact equal to or poorer than the extrapolated linear fits. In these tests, the absolute value of the deviation between an early image and a late-epoch image of the same star using the new segmented-fit correction was computed at each wavelength bin. A similar computation was also made of the absolute value of the deviation between the early image and the late-epoch image of that star using the original, extrapolated correction. Then the differences between these absolute-valued deviations were calculated at each wavelength bin and plotted. Figures 74 and 75 illustrate these differences for two post-1996 images. The mean difference over all wavelengths is plotted in each figure as a horizontal line. In each of the seven cases looked at, mean differences greater than or equal to zero were found, indicating that the segmented-fit corrections lead to deviations that are, on average, either the same as or larger than the deviations obtained using the original sensitivity correction.
As a result of the analyses described above, the Calibration Group
recommends the following for NEWSIPS processing:
SWP Point Source: Adopt a new segmented sensitivity degradation
catch-up correction, using 1993 as the break point. AU images
obtained in 1993 and later would thus be processed using the new
correction.
SWP Trailed Source:
Continue to use an extrapolation of the 1995 trailed source sensitivity
degradation catch-up correction.
LWP Point Source:
Continue to use an extrapolation of the 1995 point source sensitivity
degradation catch-up correction.
