Processing digital spectroscopic data : the point-spread function and second-order scaling errors

We discuss the analysis of some peculiarities found in semi-raw spectroscopic CCD data, and demonstrate that they result from effects that correlate with the zenith angle at the time of observation. We present evidence and arguments that suggest that the ultimate cause of such effects is alterations of the point-spread function of the stellar image that is delivered to the entrance aperture of the spectrograph by the combination of the telescope optics and the terrestrial atmosphere along different sight paths. Based upon this working hypothesis, we discuss strategy for coping with the problem. Our coping strategy for this problem leads to a scale-normalization procedure akin to fitting a (pseudo) continuum to stellar spectra. Ideally, such a procedure should produce stellar spectra that can be directly compared numerically for the purpose of detecting and measuring small variations of spectral features. However, small subjective (personal) errors in any such procedure introduce small systematic errors that can raise the detection threshold for small variations above the noise level of the data; or even produce chimerical variations that can be misinterpreted as being real. We discuss and demonstrate an objective method for detecting such systematic effects and eliminating them, so as to achieve the optimum detection capability that the signal-to-noise ratio of the data will permit.