6.1. Dark Subtraction
The majority of the FOS instrumental background is Cerenkov radiation, caused by cosmic rays striking the photocathode, rather than thermal dark current. Consequently, the instrumental background can vary by factors of two between exposures, a fact ignored in the pipeline calibration. As with the WFPC2, the relative dark rate is a function of instrument geometry and is quite stable over the detector. The G650L low-resolution grating illuminates only 144 of the 512 diodes; therefore, the unilluminated portion can be used to identify the dark rate during a given exposure. On the suggestion of L. Petro (private communication), we produced a "superdark" using 25 dark frames taken in the HDF parallel program (Program 6342, Freedman; Program 6339, Petro) and scaled this to the level indicated by a subset of the unexposed diodes (pixels 900-1100) in each of our spectra. The exposed and unexposed regions of the diode array are labeled in Figure 15, where we plot the averaged dark spectrum. To determine the accuracy of this dark-subtraction method, we have reduced darks which were taken during our own orbits expressly for this purpose. The test reduction of these darks shows no systematic error in the dark subtraction by this method, and statistical errors are dominated by shot noise (± 2.2 DN/pixel). The mean dark level can be determined in an individual frame to roughly ± 0.3DN/pixel. With a sky signal of roughly 30 DN/pixel in the low-dispersion configuration, this dark-subtraction method introduces a 0.5% random error to each spectrum overall and 0.4% systematic uncertainty to each resolution element.
Figure 15. Average of 25 FOS dark exposures (1300 sec), demonstrating the stability of the pixel-to-pixel structure of the dark signal. Strong features appear where diodes have been turned off due to poor performance. Sub-stepping along the diodes allows complete spectral coverage over these dead diodes.