3 INES data quality evaluation

 

3.1 Flux repeatability

The repeatability of the INES low resolution spectra has been tested on a large sample of spectra of some of the IUE standard stars. The only restriction imposed has been to include only non-saturated spectra of similar level of exposure (i.e. similar exposure times) in order to avoid the remaining non-linearity effects (see Sect. 3.3). The spectra cover all the range of observing epochs and camera temperatures. Therefore it must be taken into account that the repeatability, as defined here, includes implicitly the uncertainties in the camera time degradation and the temperature corrections.

The study has been performed in 100 Å wide bands. Table 2 lists the central wavelength of the bands and the repeatability, defined as the percent rms respect to the mean intensity of the band. The figures in brackets are the number of spectra considered in each case.

  

Table 2: Flux repeatability of INES low resolution spectra

SWP

LWP

LWR

As expected, the best repeatability is attained in the regions of maximum sensitivity of the cameras. In the SWP the repeatability is around 2% longward 1400 Å. For the LWP camera, values lower than 3% are reached in the central part of the camera, 2400-3000 Å. At the extreme wavelengths the repeatability is around 15%. The results are slightly worse for LWR most likely due to the instability of the camera after it ceased to be used for routine operations. The repeatability is between 3-4% in the region 2300-3000 Å. Particularly bad is the 3300 Å band, but at the shortest wavelengths (1850-1950 Å) the repeatability is substantially better than in the LWP camera. When considering only images taken when LWR was the prime long wavelength camera, the repeatability is similar to that of LWP in the central part of the camera.

  

Figure 15: Comparison of the extraction errors (Errors) in INES with the dispersion around the mean spectrum (STD) for three standard stars, after correction with the coefficients shown in Table 3 for the SWP camera

  

Figure 16: Same as Fig. 15, but for the LWP camera

3.2 Reliability of extraction errors

In addition to the flux spectrum, optimal extraction methods also provide an error spectrum. Formally, these errors only account for the uncertainties in the extraction procedure, based on the noise model of the detector. They do not include uncertainties driven by parameters affecting the image registration. During the processing, corrections are applied to account for the changes in temperature in the head amplifier of the cameras (THDA) and the loss of sensitivity of the detectors. All these corrections have their own uncertainties. There are yet other systematic errors that affect the absolute fluxes, as the uncertainty in the inverse sensitivity curve, but do not affect the comparison of different sets of IUE spectra. The extraction errors can be used to compare fluxes in different bands of the same spectrum or to compute weighted averages of a set of spectra, but they may not be appropriate to evaluate the variability of a source or an spectral feature, due to the considerations given above.

  

Figure 17: Same as Fig. 15, but for the LWR camera with a voltage of -5.0 kV

In order to check the statistical validity of the errors provided by the INES extraction, we have taken the same data set used in the previous section (i.e. a large sample of spectra of standard stars with similar level of exposure) and compared the rms around the mean with the average errors as given by the extraction procedure. In general, the extraction errors underestimate the errors represented by the rms in the three cameras, with the exception of the shortest wavelength end of the LWP camera. In the SWP camera the errors are underestimated by $\sim$20-40%, depending on the wavelength. In the LWR the ratio between extraction and actual errors is nearly constant (12%) all along the camera, while in the LWP the discrepancy can be as large as 40% at the longest wavelength. In this camera, shortward 2400 Å the extractions errors are too large by 15-20%. This region is very noisy and there are reasons to suspect that such a noise departs significantly from a gaussian behaviour.

It is also found that the dependency of the ratio STD/Error (where "STD'' is the standard deviation around the mean spectrum, and "Error'' is derived from the extraction errors) with wavelength can be well represented by a straight line with the coefficients shown in Table 3. Reliable values for these coefficients could not be obtained for the LWR camera operated at -4.5 kV due to the scarcity of data.

  

Table 3:

In order to compare fluxes in different spectra of the same object, the extraction errors must be modified according to

$$\varepsilon(\lambda)=a+b\varepsilon_{{\rm E}}(\lambda)$$ (7)

where a and b are the coefficients in Table 3 and $\varepsilon_{{\rm E}}(\lambda)$ are the extraction errors.

The results of the application of this correction are shown in Figs. 15, 16 and 17. The dispersion around the expected value of 1 is 0.15 for SWP, 0.17 for LWP and 0.18 for LWR. The structure still seen in these figures might be related to remaining non linearities in the ITF's (see below).

3.3 Flux linearity

  Despite the correction applied during the processing of the IUE data through the application of the Intensity Transfer Functions (ITFs), the final spectra are still affected by non-linearities to some degree. As a consequence, spectra of the same non-variable object observed with different exposure times might have slightly different flux levels.

  

Figure 18: Ratios of different exposure levels to the 100% for the three cameras. The thick horizontal line marks the saturated part in the overexposed spectra

  

Table 4: Linearity properties of INES low resolution spectra (ratios to 100% exposure)

SWP

LWP

LWR-ITF-B

In order to evaluate the importance of the remaining non-linearities we have chosen, for each camera, a set of low resolution spectra of the standard star BD+28 4211, extracted with the INES system, obtained close in time under similar temperature conditions and with different exposure times. In each set one of the spectra is defined as a 100% exposure, and all the other are referred to that one.

The summary of the data used for each camera is as follows:

• SWP: Nine spectra taken in December 1993, with exposure times ranging from 2 s (9%) to 40 s (150%).
• LWP: Nine spectra taken in October 1986, with exposure times ranging from 10 s (20%) to 100 s (200%).
• LWR: Five spectra taken in August 1980, with exposure times ranging from 20 s (30%) to 150 s (250%). All this spectra have been processed with ITF-B [Garhart et al.1997, (Garhart et al. 1997),] which is the one giving the best correlation coefficient in this case, as in most of the pre-1984 LWR spectra.

Each spectrum was binned into in 100 Å bands and divided by the corresponding reference exposure. The results are summarized in Table 4. Examples of the behaviour of different spectral bands for each of the cameras as a function of the level of exposure are shown in Fig. 18.

In the SWP camera the largest departures from linearity are found at the short wavelength end of the underexposed spectra, where flux can be underestimated by up to a 20%. Apart from this case, longward Lyman $\alpha$ ratios to the 100% spectrum are generally within$\pm$ 5%. The best results are achieved in the 1800 Å band, where linearity is within $\pm$3%.

For the LWP camera the largest non-linearities occur at the extreme wavelengths (1900, 3300 Å), where the flux is largely overestimated. Except for these bands, linearity is within $\pm$ 5% for spectra with exposure levels from 40% to 150%. In the saturated region of the most exposed spectrum the flux is overestimated by 10%. Excluding the saturated region, the bands which show the best linearity characteristics (within $\pm$ 3%) are those centered at 2800 and 3000 Å.

The LWR camera shows the largest non-linearities at the longest wavelengths of the underexposed spectra. Linearity remains within $\pm$ 5% for exposure levels above 60%. The most linear bands are those centered at 2500, 2900 and 3100 Å. In the saturated part of the 200% spectrum, the flux is underestimated by approximately 10%. However, the flux is correct in the 170% spectrum, which is also saturated.