spacer link to MAST page spacer logo image spacer
link to STScI page

NEWSIPS and its Handling of Emission-Line Spectra

Catherine L. Imhoff

1. Introduction

Since NEWSIPS was introduced, there have been some indications of sporadic errors in the extraction of emission-line spectra from low-dispersion spectra (see Rodriguez-Pascual et al., A&A, 139,183,1999). Although many spectra appear to be accurately extracted and calibrated, a few can be found in which the emission-line fluxes are significantly too low. An example is LWP 21062 (Figure 1). This article reports the results of a study of this problem, its likely cause, and some recommendations.

2. Investigation

Several examples of spectra which show low emission-line fluxes (as compared to IUESIPS) were studied to verify its occurrence and to locate the source of the problem. The first question: might the difference be that NEWSIPS is correctly excluding some anomalously high pixels, such as cosmic ray events, that were included in the IUESIPS extraction? An inspection of LWP 21062 shows that this can indeed occur. The apparent line at 2695 Å is a cosmic ray hit that falls well off the spectrum. The long rectangular slit in IUESIPS picked up the cosmic ray, but NEWSIPS rejected it. However, this does not explain the discrepancy of about a factor of 2 in the Mg II 2800 Å emission line flux between IUESIPS and NEWSIPS.

A further examination of several such spectra revealed a common element. All the spectra were weak enough that the NEWSIPS Signal-Weighted Extraction Algorithm ("SWET") algorithm did not find enough flux to determine the cross-dispersion profile of the spectrum. The processing software then uses a default point-source profile to extract the data. The default profiles were determined by the NEWSIPS team for each camera by analysis of a number of well-exposed continuum spectra. The processing software applies the default profiles to weight the extraction from the SILO file in the same manner that the weights are applied in the SWET extraction using empirically determined profiles. The default profiles, obtained from the NEWSIPS source code, are shown below (Figure 2 and Table 1).

Default Point-Source Profiles in NEWSIPS

0 4.0984E-3 2.108E-3 3.9511E-3
1 3.7522E-3 1.179E-3 7.1198E-3
2 5.6412E-3 5.627E-3 1.6838E-2
3 1.8079E-2 2.140E-2 4.8395E-2
4 7.3424E-2 7.301E-2 0.11658
5 0.25434 0.2283 0.23804
6 0.39403 0.3447 0.31888
7 0.19098 0.2359 0.18662
8 3.7127E-2 6.443E-2 4.4981E-2
9 1.0018E-2 1.247E-2 9.8457E-3
10 3.4538E-3 5.757E-3 4.1713E-3
11 2.6066E-3 2.473E-3 2.4216E-3
12 2.4546E-3 2.718E-3 2.1482E-3

This approach for a weak spectrum seems valid. However a problem can occur if the extraction is not well aligned with the spectral signal. The algorithm that determines the center of the spectrum was designed to avoid the locations of the major emission lines. This was done to avoid the difficulties encountered when a point source with a continuous spectrum is accompanied by an extended or offset emission line source, such as frequently occurs with planetary nebula spectra. For a rich emission-line source such as RR Tel, the algorithm can still determine the center of the spectrum and the extractions are reliable. However if the spectrum consists only of a few or weak emission lines, it may not find the center and thus defaults to the nominal center of the aperture.

It is not unusual for a faint source, perhaps acquired by blind offset, to fall an arcsec or two away from the center of the aperture. If the extraction uses the default point source profile and default centering, it can then miss much of the flux in the spectrum. In SWET, the pixels aligned in the cross-dispersion direction are weighted according to the determined profile. If this weighting is misaligned, then the pixels associated with the highest flux values may be underweighted, and the pixels associated with moderate or low flux values may be overweighted.

Figure 3 gives a simple illustration of this effect for a simulated emission line extraction; the flux scale is arbitrary. In this example, the point-source spectrum is offset by one pixel. This corresponds to a miscentering of 1 arcsec, a typical error for an exposure obtained via a blind offset. If the weighting follows the standard centering and profile, the resulting net flux is reduced from its correct value, depicted in the section portion of the figure. In this case, the resulting flux of the emission line is 77% of the correct value.

3. Which Data are Affected?

How can one determine if a given low-resolution spectrum has been affected? The NEWSIPS processing followed one of three paths, depending on how well the spectrum could be automatically located and described. These are as follows:

  • There is enough signal that the SWET algorithm can derive an empirical cross-dispersion profile having at least 3 spline nodes and a spatial centroid position (rounded to the nearest integer).
  • There is not enough signal to derive an empirical profile, but the centroid position can be determined. Then the default profile is used with the "found" centroid position in order to extract the flux.
  • There is not enough signal to derive either an empirical profile or the spatial centroid. In this case, the default profile and default line center are used to extract the flux.
The latter case is the one that may be affected by this problem, if the spectrum was also not well-centered in the instrument entrance aperture. In addition, spectra following in the second category could be affected if the "found" center of the spectrum is not accurate.

To determine how the spectrum was extracted, one should inspect the FITS keywords in the header. These are easily accessible in the file header if you have read the FITS data file on your computer or by inspecting the MAST IUE preview data. Below the plot, check the "MX FITS header" link. Under the large aperture keywords there are several that are generated during the processing. Below is an example for LWP 21062.

LDATABKG=                   55 / Estimated mean background level (DNs)          
LDATACNT=                   88 / Estimated maximum continuum level (DNs)        
LCNTRAPR=                 51.0 / Predicted center line of spectrum              
LXTRMODE= 'POINT   '           / Extraction mode                                
LXTRPROF= 'DEFAULT POINT'      / Profile used                                   
LXTRASYM= 'NO      '           / Asymmetrical profile in extraction             
LXTRCNTR=                 51.4 / Center line of extracted spectrum              
LFLUXAVE=                 10.8 / Average flux (FNs) 
This shows that the default point-source extraction profile was used. The nominal center line is 51.0, but the used value was 51.4, so this spectrum falls under the second category of SWET processing. This is made explicit in the processing history included in the full FITS header (below). In this case, even though there was apparently enough flux for the processing to detect and locate a center, it was too inaccurate for the extracted emission lines fluxes to be determined well.

HISTORY START SWET                                  26-MAY-1996 20:59:35        
HISTORY  NOISE MODEL USED: LWP  VERSION 1.0                                     
HISTORY  *********************LARGE APERTURE DATA***********************        
HISTORY       LINE 51, PEAK AT LINE 52, AVERAGE PEAK FN =    10.8               
HISTORY  *** WARNING: LOW NUMBER OF SPLINE NODES                                
HISTORY               DEFAULT PROFILE SUBSTITUTED                               
HISTORY               FOR EMPIRICAL FIT                                         
HISTORY  PROFILE CENTROID AT LINE  51.4                                         
HISTORY  EXTRACT FLUX FROM LINES 45 THROUGH  57                                 

4. Prospects for Correcting the Problem

If one determines that a given spectrum is affected by this problem, what can one do? The NEWSIPS SILO files were designed to be easy to use should a user wish to re-extract the spectrum. A simple "boxcar" slit (all lines evenly weighted) will conserve the flux, although it also includes more noise. One might also determine the center line of the spectrum interactively and re-extract the data using that center line and weighting, using either the default or similar profile.