improving flattening/sky subtraction of Gemini GMOS images

[jlowenthal]

We're working with Gemini/GMOS r-band images and are seeking advice on improving the flat-fielding. Here's what we've done to our 55x5min science frames:

* gbias to overscan-subtract and coadd the biases to make a master bias
* giflat to make an OS-sub, bias-sub, trimmed, coadded master twilight flat
* gireduce to reduce our sci data
* gmosaic to mosaic each frame
* imcoadd to average the frames with avsigclip cosmic-ray rejection

The result shows significant residuals in the sky background, with large-scale gradients of amplitude about 1% of sky. They also show significant vertical strips where the gaps between the 3 GMOS chips are, even with dithering. Before co-adding, the mosaiced images (3 chips combined into one image ) also show (1) those gradients, (2) significant discontinuities across the 3 chips, and (3) very significant differences in sky background across the set of 55 images (nearly a factor of 10; some were dark sky, some in bright moonlight).

Q1: Is there any way to improve the flat-fielding beyond the twilight flats we got from the Gemini Science Archive? Is 1% the best we can do?

Q2: Any suggestions for removing residual sky gradients through means other than flat-fielding? We've tried subtracting low-order fits using imsurfit with various fit orders and median box sizes and rejection schema, but have not been satisfied with the results. Haven't tried masking bright objects explicitly, a la xdimsum, but that might be the next step. Either way this is cumbersome because it requires treating the 3 GMOS chips separately. Is there a GMOS task to do this on the MEF files?

Q3: Is there a Gemini or other task for combining images with automatic sky subtraction? Given the large range in sky backgrounds, we definitely want to sky-subtract individual frames before co-adding, but that doesn't seem to be an option in imcoadd. Imcoadd does make sky-subtracted geometrically transformed images (*_trn.fits) but it doesn't seem to use them in the final coadd.

Thanks for any tips,

James Lowenthal
Smith College