Removal of particle hits

Cosmic ray  events pose a serious problem in long CCD exposures. Their removal is a rather delicate step, because in high-contrast images (such as well-exposed echelle spectra) there is always a danger of damaging the scientific contents of the frame. Particle hits can be removed from scientific exposures by splitting the exposure and comparing spectra of the same target obtained under the same instrumental configuration. Offsets of the target resulting from the positioning of the target on the entrance slit of the spectrograph and variations of exposure time must be accounted for. Commands AVERAGE/IMAGE and AVERAGE/WEIGHT offer number of options to compare the images and reject particle hits. In the case of echelle spectra of sources with very little variation of the spectral information along the slit, one can also exploit the knowledge provided by the order definition as to where in the frame the relevant data is located.

The removal of unwanted spikes, above an otherwise featureless background such as the inter-order space of echelle spectra, is done most easily with a median filter. Therefore, in a first step a median filter is applied to the entire frame. This then enables the true background to be determined as described in Section `Background Definition' of this chapter. The subtraction of the background calculated completes the second step, and, as far as the inter-order space is concerned, the final result is reached already. The removal of cosmics from the object spectrum forms the third step and is restricted to the regions covered by the spectral orders in the background-corrected, but otherwise raw frame. This step comprises the following operations performed within a sliding (along and separately for each spectral order) window of user-specified width:

• For all x_i of the window, normalize the spatial profile to the total flux in the associated slice (i.e. all y_j of the order considered).
• Form the `true' spatial profile as the median over the individual profiles at all x_i in the window.
• For all pixels ( x_c, y_j) in the central slice, c, of the window compare their re-normalised flux, f_c,j, with the properly scaled contents m_j of pixel j in the median profile. If the difference exceeds the expected statistical error of f_i,j (calculated from the number of photons detected and the readout noise) by a user-specified factor, replace f_c,j with m_j.

If the threshold for substitution by the median is set properly ($\sim$4 $\sigma$) and the spectral information within the spatial profile does not change (point sources are best), this procedure does not redistribute the flux or dilute the point spread function.

These three steps are the backbone of command FILTER/ECHELLE. The final output frame is the merger of the median-filtered inter-order domains with the spectral orders after having been subjected to step three. Note that the background has been subtracted already. A keyword BACKGROUND with contents SUBTRACTED is appended to the frame as a flag to subsequent high-level procedures so as not to have to go through the very time consuming step of the background modeling again. Delete that descriptor or change its contents if re-modeling of the residual background is desired.