It is important to point out that the geometric distortion characteristic of
raw IUE images (Section
requires that the fitting of analytic dispersion
be performed in geometrically corrected image space.
Consequently, although spectral data images are no longer explicitly
geometrically corrected in standard IUESIPS processing, WAVECAL images
geometrically corrected and all derived dispersion relations refer to
geometrically correct space. The G-1 geometric mapping function
is then used throughout the software system to transfer the dispersion
relations to raw-image space.
The pixel locations of the platinum-neon emission lines are measured on the
geometrically corrected WAVECAL images by a cross-correlation search algorithm
(see Turnrose and Perry, 1977) like that used to find reseau positions. The
starting search positions are determined from the current set of mean disper
sion constants corrected for the temperature and, in the case of LWR and SWP,
the time of acquisition of the particular WAVECAL image (see Section
The measured platinum-neon line positions are then combined with laboratory
values for the wavelength and order number of each emission line and used in a
regression analysis to determine a set of dispersion constants (A and B)
relating wavelength and order number to pixel location according
to the following relations:
6.l. The implementation of mean calibration data eliminated the biweekly discontinuities in the way the IUE data were formerly reduced and also made possible the further improvements described in Section 6.3.1 of correcting the dispersion constants for temperature variations and secular effects. Continual monitoring of the biweekly calibration images is conducted to determine whether the implemented mean dispersion relations should be modified.
|Implementation||No. of Images||End Dates||Corrections|
|3-03-81||41||3-31-79||1-01-81||THDA & Time||2|
|9-21-82||46||1-01-80||8-10-82||THDA & Time||3|
|6-20-84||105||7-15-78||3-07-84||THDA & 2nd Order Time||5|
|3-03-81||40||3-31-79||1-01-81||THDA & Time||2|
|9-21-82||44||1-01-80||8-10-82||THDA & Time||3|
|6-20-84||107||9-30-78||3-11-84||THDA & 2nd Order Time||5|
|5-19-81||41||3-31-79||1-01-81||THDA & Time||2|
|9-21-82||47||1-01-80||8-10-82||THDA & Time||3|
|6-20-84||103||9-30-78||3-07-84||THDA & 2nd Order Time||5|
|5-19-81||41||3-31-79||1-01-81||THDA & Time||2|
|9-21-82||45||1-01-80||8-10-82||THDA & Time||3|
|6-20-84||109||9-11-78||3-11-84||THDA & 2nd Order Time||5|
|1||Thompson, et al. (1980)|
|2||Thompson, Turnrose and Bohlin (1982a)|
|3||Thompson and Turnrose (1983)|
|5||Gass and Thompson (1984)|
Corrections for thermal and temporal shifts based on the derived correlations were first implemented as part of the standard LWR and SWP data processing at GSFC on 3 March 1981 in low dispersion and on 19 May 1981 in high dispersion by a new IUESIPS applications program, TCCAL. Corrections for thermal (but not temporal) shifts for LWP images were implemented on 12 April 1983.
The corrections are based on the THDA temperature at the end of the exposure (denoted by T in the equation below, and expressed in centigrade degrees), and the time (t) expressed as the total number of elapsed days since 1 January 1978. Both values are generally obtained from data extracted from the image header label. The THDA at the end of exposure is normally extracted from the camera snapshot section of the image label, but if that THDA value is not available, then the THDA at the time of image read is extracted. If neither THDA is available, then the processing defaults to the mean dispersion constants unless either (a) a specific THDA value is manually specified by the image processing specialist when the image is processed, or, (b) for LWR and SWP images processed after 20 June 1984, a reasonable date of observation can be extracted from the image label in which case a correction for time only is applied. The last option above was implemented to improve the calibration of images which do not contain valid temperature data (e.g., history playback images and images obtained prior to about March 1979). See Section 9.3 for information about how these data are documented in the image processing history label.
The correction terms Ws and Wl representing a uniform shift applied to the mean dispersion constants are defined by the general expression
|LWP HIGH||LWR HIGH||SWP HIGH|
|A1||5.873462158066862E 03||-4.568022566378104E 03||5.240320204548078E 02|
|B1||1.722851374444825E 03||1.567990956548678E 04||-7.171777625701399E 03|
|W1(S)||-7.430500388145447E-01||5.459306716918945E 00||-2.977794647216797E 00|
|W1(L)||-4.000792503356934E 00||-8.628579139709473E 00||-2.841607093811035E 00|
|LWP LOW||LWR LOW||SWP LOW|
|A1||1.046282942865237E 03||-2.992355784397701E 02||9.833223402481985E 02|
|B1||-2.722748512318324E 02||-2.647551045134080E 02||-2.633234804632572E 02|
|W1(S)||-7.578814029693604E-01||5.142534255981445E 00||-3.452352523803711E 00|
|W1(L)||-2.995339393615723E 00||-8.595767974853516E 00||-1.659444808959961E 00|
Based on the statistical analysis of the WAVECAL images used to establish the time and temperature dependencies, the corrections described above reduced the average relative (i.e., intrinsic) error in high dispersion wavelength assignments to a velocity equivalent of less than 3 km s-1. Larger relative errors may be expected for specific wavelengths near the tube peripheries or affected by residual uncorrected geometric distortion (see Section 6.5) and for images exposed during times when large variations in spacecraft temperature exist. There is limited recent evidence that overall differential (i.e., non-uniform) motion of the spectral format may occur, which may also increase the expected errors. This as yet ill-understood motion has been observed to introduce non-uniformities of up to approximately 1 pixel. In addition, extrinsic errors caused, for example, by spacecraft pointing limitations (target decentering due to small errors in the initial acquisition, spacecraft roll drift during long exposures, etc.) and the uncertainty in centroiding spectral features in extracted spectra may exist which contribute to the absolute error in IUE wavelengths. A more detailed discussion of overall errors is presented in Section 6.5.4, are used by the spectral extraction routines described in Section 7 not only to assign wavelengths but also to determine the position of the extraction slit as it is passed numerically by the computer along each spectral order of the image. Accurate gross and background flux levels therefore require proper registration of the dispersion relation with the actual spectral orders, particularly in the region of the closely-spaced orders in high dispersion. Since the corrections described in Section 6.3.1 for determining the location of the spectral format are statistical in nature, automatic and manual registration methods are used to modify further the dispersion constants on an image-by-image basis to remove residual registration errors in the direction perpendicular to the dispersion. See Section 9.3 for information regarding documentation of registration shifts in the image label. 6-3. In high dispersion the registration shifts are first calculated in the region of the closely spaced orders (order 108) where precise registration is most critical. If these shifts do not pass certain built-in program constraints (described in Thompson and Bohlin, 1982), up to three progressively lower orders are searched in the same way. With the average of the 12 offset values, IUESIPS adjusts the A1 and B1 terms in the dispersion relations, in effect tailoring the dispersion relations for the particular target image in question. Algorithms currently exist for the automatic registration of both high and low dispersion point-source spectra and of low dispersion trailed spectra.
|Low dispersion wavelengths (Å)|
|ong wavelength||Short wavelength|
|High dispersion wavelengths (Å)|
|Long wavelength||Short wavelength|
|m= 108||100||86||77||m= 108||100||82||77|
Note that the GO can greatly facilitate the processing operation by specifying on the Observatory Record Sheet (see Section 1.4) that the manual shift routine should be used for those images which would not be suitable for the automatic algorithm such as weak or emission-line spectra.6.3.1, it is possible that even perfect registration in order 108 could leave registration errors as large as a pixel in the lower orders (since the differential motion error generally changes linearly across the image). Since the lower orders in high dispersion are farther apart and since the extraction slit is intentionally made slightly longer than the width of the spectral orders, these registration errors should not seriously affect the extracted spectral data. Although the errors involved in the manual registration routine will depend upon the accuracy of the image processing specialist and the quality of the raw spectral image, they are probably typically less than 0.5 pixel.
It may be noted that although the dispersion relation describes the spectral format location in geometrically correct space, the correction applied refers to a shift measured in raw image space. Theoretically the raw image space registration shift should be converted back to an equivalent geometrically correct shift to compensate for any expansion, contraction, or rotation of the image that would occur in the geometric correction process. Test results indicate, however, that this error is in fact less than a few percent of the calculated shift and is therefore insignificant.7), two additional corrections to the assigned wavelengths are routinely made. These corrections are described in the following subsections in order to consolidate the discussion of wavelength topics within Section 6. 126.96.36.199) or the image header label as described in Section 9.3.
Using the time of the midpoint of observation, the velocity components of the earth and IUE in a righthanded rectangular equatorial coordinate system (+x is toward the vernal equinox, +z is toward the north celestial pole) are computed using the routines described in Harvel (1980). Harvel has shown that by using the fixed orbital elements listed in that reference, the spacecraft velocity calculation was accurate to 0.25 km s-1 over the first 3 years of IUE operation. (Schiffer (1982) has indicated that an observed evolution of the IUE orbital elements may increase the error in the spacecraft velocity calculation by about 1 km s-1. Most of the change is in the Z component. The operations software now stores current IUE orbital elements in the science header; see Figure 9-1b.) Additional spacecraft velocity errors can result from errors in calculating the time of the midpoint of the observation. Currently the midpoint time is set equal to the end-of-exposure time minus half of the total exposure time, as read from the raw-image label. Depending on the camera procedures used, this method could cause an error in the spacecraft velocity calculation. However, since the magnitude of the spacecraft velocity is small compared to that of the earth's motion (4 km s-1 versus ~ 30 km s-1) the error will in general have a small effect on the overall wavelength correction.
The computed net radial velocity of the IUE spacecraft toward the object is then:
Note that the calculation is such that a positive net radial velocity correction indicates a net approach of the IUE spacecraft toward the target, following the standard convention. The individual IUE and earth velocity components and the net radial velocity correction used are documented in the image processing history label as discussed in Section 188.8.131.52.4.1. The formula used for wavelengths vac equal to or greater than 2000 Å is:
Note that under the old reduction software (i.e., prior to 3 November 1980 for low dispersion and 10 November 1981 for high dispersion) the vacuum-to-air correction was only applied to data from the long wavelength spectrograph (cameras LWR and LWP). It is now applied for all cameras at wavelengths greater than or equal to 2000 Å.Table 6-4 lists these average intrinsic errors for the various time/temperature compensation methods. (Note that for SWP and LWR, the column relevant to current production techniques is that headed "THDA & 2nd Order Time", whereas for LWP, which currently has no time dependence applied, the relevant column is that headed "THDA Only." These errors pertain to the limitations of modeling the behavior of dispersion relations fitted to the rather homogeneous set of WAVECAL images. The absolute accuracy of applying this internal wavelength scale to actual spectral images depends on a number of other, largely extrinsic, factors. Such factors include the extent to which the temperature and time pertinent to a given image fall within the correlation range defined by the WAVECAL data, the extent to which spectral- image-specific reseau motion and geometric distortion can be compensated, the extent to which the limited-term polynomial defining the dispersion relations can compensate for small-scale deviations from a smooth wavelength scale, the extent to which the laboratory wavelengths assigned to the WAVECAL spectra are correct, the extent to which the target object is centered in the aperture, and the extent to which a feature in an extracted spectrum can be accurately centroided. These various factors are addressed below.
|Dispersion Direction||No Correction||THDA Only||Time Only||THDA & Time||THDA & 2nd Order Time|
Prior to the implementation of temperature and time corrections (reference Table 6-1), systematic errors were introduced for images exposed during extreme spacecraft temperatures or at times much different than the average time for the mean dispersion relations. Further, now that the overall time dependence has been observed to be nonlinear (i.e., a second-order time dependence is used), it is apparent that systematic errors were also introduced by extrapolating the linear time correction beyond the end dates of the input calibration data (see Table 6-1) such that the greater the extrapolation in time, the larger the systematic error.
Because the dispersion relations, which are determined in geometrically corrected space, are now applied to spectral images in raw image space, errors in the mapping function G-1 represent an additional error in the assigned wavelengths. Although studies have shown that reseau positions vary with both temperature and image intensity (see Section 4.2, Oliver (1979), and Thompson 1983a), IUESIPS currently uses only mean reseau positions for LWP and LWR images and only temperature-corrected mean reseau positions for SWP images (see Section 4.3). The lack of any correction for image intensity can cause errors of 1 pixel or more in the assigned reseau positions (Thompson 1984c). Since the error occurs primarily in the line direction, it may result in a position-dependent wavelength error corresponding to up to f 0.7 pixels. The mean reseau positions currently used in IUESIPS are all based on an analysis of 60-percent UVFLOOD images. Although the data set included images taken over a fairly wide range of temperatures, they were all at approximately the same DN level. Since the 60-percent UVFLOODs have a central mean DN of approximately 120, the largest errors may occur for images with higher or lower exposure levels (and at extreme spacecraft temperatures for the uncorrected LWR and LWP cameras).
A possible source of absolute wavelength error is the inability of a limited-term polynomial to compensate for small-scale deviations from a smooth wavelength scale. Small-scale image distortions, when occurring in the direction perpendicular to the dispersion, have been observed and described as residual curvature; such distortion must certainly also occur along the dispersion and would result in localized small-scale (sub-pixel) wavelength errors.
Another possible explanation for errors in the assigned wavelengths, suggested by Ayres (1984), is that errors may exist in the library of laboratory wave lengths. In addition, Ayres suggested that the non-uniform distribution of emission lines used in generating the dispersion relation may be such that the fit favors the regions which contain the most emission lines. Recent work by de La Pena (1984) shows that the SWP high dispersion wavelength assignments show a wavelength-dependent variation of ~ 2 km s-1 which could be explained by Ayres' suggestion. It should be noted that all the above results were based on studies of WAVECAL images and therefore would not be subject to reseau-position errors due to the beam-pulling effects mentioned above.
Target de-centering within the aperture can lead to appreciable absolute wavelength errors, particularly in the large aperture, where the centering error can typically be about an arcsecond, even larger in the case of blind-offset acquisition. A pointing error of 1 arcsecond along the dispersion direction is 0.66 pixel, or about 5 km s-1 in high dispersion. As mentioned in Section 6.3.1, spacecraft drift during long exposures can also contribute to de-centering error. The small size (about 3 arcseconds) of the small aperture should result in better absolute wavelengths due to reduced target de-centering effects.
The ability to centroid a spectral feature in an extracted spectrum is a final limitation to the accuracy of assigned wavelengths. As is pointed out in Thompson, Turnrose, and Bohlin (1982a), for an isolated narrow feature with the instrumental resolution of 2.5 pixels in the cleanest IUE data, the best possible estimate of the measurement accuracy would be a 0.25 pixel, which in high dispersion corresponds to a 2 km s-1 velocity uncertainty. In many cases, the spectral data will not be of sufficient quality to approach a 0.25 pixel measurement error.
Finally, it should be remembered that the errors referred to in most of the above discussions are AVERAGE errors, particularly those in Table 6-4. Recent experimental results have, in general, confirmed these average errors although it was found that the quoted averages can be misleading if used to describe the accuracy of measuring a single spectral feature. One analysis that was performed involved extracting several LWR and SWP high dispersion WAVECAL spectra using standard production processing techniques and measuring the wavelength assignment accuracy of the Pt-Ne emission lines in several orders (Heckathorn 1984 and Thompson 1984a). It was concluded that although the average errors found were small (i.e., the overall mean wavelength error for each of six images was always less than 3 km s-1, larger errors were found in measuring the wavelengths of individual lines (± 6 km s-1).
The special calibration procedure has become somewhat simplified in that separate TFLOOD exposures are no longer necessary. Reseau positions can now be directly measured from the WAVECAL images themselves (thereby eliminating the need for the separate TFLOOD).
Some Guest Observers have found it useful to obtain WAVECAL images with their normal spectral images and have them extracted using the standard production processing procedure. The wavelength errors measured on the WAVECAL image can then be used as a guide to correcting the other extracted spectral images.
Guest Observers who wish to use a special calibration should fill in the last
two lines of the "Processing Specifications" portion of the observing script
indicating the image number of the special calibration to be
used. Guest Observers should also be warned that special processing requests
may cause some delay in the processing of their images.