The revisions provide:
The weights, when using a weighted average in all previous versions of the combine tasks (except the old onedspec.combine task), were based on estimates of the sigma (the square root of the variance). However, variance weighting is more commonly used so the weights are now based on variance estimates.
Specification of scaling and weighting factors was made more flexible by allowing the factors to be specified from text files (using the standard @file convention) and from image header keywords. The header keywords can now be different for the different factors and can be different than the exposure time. The documentation indicates that the scaling, both offset and gain, are to be numbers to add or multiply to the image. However, in V2.10.2 the code was incorrect and used the values to subtract and divide. In subsequent versions this was corrected so be aware that the interpretation of this type of input will change to that given in the documentation.
In V2.9 the clipping rejection algorithms were limited to rejecting a single pixel. In V2.10 this restriction was dropped and as many pixels as exceeded the clipping threshold could be rejected. However, this opened the possibility of rejecting all pixels. This might occur even with normal data by chance when the closest pair of pixels to the central median or average are above and below that value by a few sigma. This can become significant if the sigma factors are made small (2 sigma or less) or the sigmas are underestimated by the CCD noise parameters or the ``avsigclip'' estimate. This possibility is particularly a problem when combining flat-fields where it is better to preserve even a poor estimate of the flat-field level than using the blank value (which defaults to zero).
In this revision a new parameter, ``nkeep'', was added to allow specifying the minimum number of pixels to keep or, by using a negative value, the maximum number of pixels to reject. The default value is set at 1 which requires at least one pixel be retained by the clipping algorithms. Note that this parameter does not apply to bad pixel masking or threshold rejection.
In line with this change and the comments above the ccdred ccd-type specific combining scripts were modified to include some additional parameters. The ``nkeep'' parameter was added and the ``blank'' parameter was made a parameter of these tasks rather than being fixed at zero. For flatcombine the default ``blank'' value was set to one rather than zero. Also the new ``noise'' parameter, described below, was added.
The final change was to add a third CCD noise parameter called ``snoise''. This provides a sensitivity or flat-field noise component. This type of noise scales with the intensity rather than the square root as provided by the Poisson term. The default value is zero which preserves the previous noise model but does allow some additional flexibility which I believe is important in typical data. A 1% uncertainty in the flat-field correction can give uncertainties for high signals well in excess of the Poisson uncertainties. Without this the centers of bright stars could be clipped by the ``"ccdclip'' or ``crreject'' algorithms. I have found the task stsdas.hst_calib.wfpc.noisemodel a nice one to characterize the CCD noise parameters including the sensitivity noise from groundbased CCD data as well as WFPC data. A point to note is that the read noise is specified in DN in that task instead of electrons as is used by imcombine.
The most common error message encountered by sites which updated to V2.10.2 when using the revised version of imcombine and related tasks is ``parameter `snoise' not found''. This is due to an incomplete installation of the patch. The Frequently-Asked-Questions file describes this common error and its solution.
Frank Valdes
This document was translated by ms2html v1.8 on 21Jan95.