next up previous contents
Next: 10.2.4 One-Dimensional Flag Spectrum Up: 10.2 Spectral Flux Extraction Previous: 10.2.2 Description of Spectral

10.2.3 Noise Models

Unlike the processing of low-dispersion images, the extraction of high-dispersion fluxes does not require a noise model. Nevertheless, NEWSIPS provides a noise vector at each wavelength as an indicator of its reliability. This vector is generated by using the camera-dependent high-dispersion noise model (discussed below) to estimate a noise value according to the fluxes along the extraction slit at each wavelength. The total noise amplitude for the wavelength is the sum of the individual noise values of all pixels along the slit, including flagged pixels. In this computation, pixels at the ends of the slit are given their corresponding fractional weights.

The noise models are derived empirically for each camera by measuring the scatter in the FNs around the mean FN in the background regions of several hundred science and flat-field images taken at a variety of exposure levels. These measurements are made in a 21 × 21 grid of regions, each region being 35 pixels on a side. At each grid point (region), the relation between noise level and FN is fit with a fourth order polynomial. This polynomial is then sampled uniformly at 50 points from 0 to 588 FN, and these sampled noise values for each grid point comprise the ``noise model''. This positional FN and noise-level information is stored as a static data cube for use in the processing of high-dispersion images. As a given image is processed, the noise level corresponding to a given pixel location and FN value is calculated by interpolation of the noise model data cube: bilinear spatial interpolation is done among the appropriate grid point locations, and linear interpolation in FN is done among the sampled noise values. Noise values for FNs below zero and above 588 FN are set to the noise values corresponding to these extrema, respectively.

Unlike the low-dispersion counterpart, the 1-D noise spectrum which is output to the high-dispersion MX is not in absolutely calibrated units. To achieve this, the user should multiply the noise spectrum for a given order by the ratio of the absolutely calibrated flux to the net flux. Although different in detail from the ``sigma'' vector produced by SWET in low dispersion, the high-dispersion noise vector is fundamentally analogous to the sigma vector in its origins from a ``noise model'' derived from rms measurements of flat fields and in the relationship of its construction to the spectral-flux extraction method used in each case.


next up previous contents
Next: 10.2.4 One-Dimensional Flag Spectrum Up: 10.2 Spectral Flux Extraction Previous: 10.2.2 Description of Spectral
Karen Levay
12/4/1997