iraf.net

Dear all,

I am trying to reduce a longslit data(from SUBARU FOCAS, btw,) and I am having some problem that I couldn't understand.

When I invoke apall task with weight option set to "none," resulting spectra seems fine.
But if I set weight option to "variance," the optimally extracted spectrum(z=1) is significantly different from the standard spectrum(z=2, or the one with weight=none.)
Say, if I plot optimally extracted spectrum divided by standard spectrum, it resembles a polynomial function of wavelength with order of 4 or higher, and the value differes from nearly 0 to 30, contrary to my belief that it should be flat 1.

I tried changing some parameters. For example, I set saturation to 32000, which is maximum DN value for the FOCAS CCD minus the bias level. The resulting optimal spectrum tends to have similar values with standard spectrum, yet optimal spectrum still seems to be a bit curvy. Still the shape of the optimal divided by standard resembles polynomial function.

Could anyone give me some advice?

Yes extraction without weighting is pretty fool-proof since it is simply a sum of pixels with partial values at the edges. Weighted or "optimal" extraction involves finding a profile function from the data as a function along the dispersion and this can get tricky. Also when it goes wrong it does, indeed, look quite different from unweighted. There are various causes for this and it depends on the data; things like how tilted or rotated the 2D spectrum is from the rows and columns and whether there are bad or extreme values in the data. So I can't really tell you how to set the parameters without actually playing with the data myself.

By the way, using the maximum value is a good way to both clean cosmic rays and minimize the effect of extreme values. However, this should not be interpreted as the maximum of the detector but the highest value in the spectrum that is not due to cosmic rays or detector defects. Unless you are very interested in very bright lines (in which case optimal extraction may not be the best approach) you can set the value much lower.

Something not appreciated sometimes is that something better than an unweighted extraction over the full range of the profile but not specifically weighted is to use a narrow aperture along the core of the profile. This requires the profile to be well sampled but in this case the resulting spectrum is basically weighted by the good S/N and not by data way into the low S/N.

I hope this helps.

Frank Valdes