Photometric tests of high-ADU nonlinearities

Exposures 180638-654 were taken on 22 Feb 2013 with the g filter, looking at the Omega Cen outskirts and varying the exposure time: the sequence was (10,40,10,7,10,72,10,20,10,80,10,5,10,28,10,14,10) seconds. No (intentional) dithering was applied, so stars stay on the same CCDs. The goal is to see how measured fluxes of stars vary with brightness and exposure time; if everything is linear, then the measured fluxes will scale with exposure time, with perhaps some tweaks for clouds, but the ratio of (flux in 80s) / (flux in 10s) should be independent of stellar magnitude.

The images are processed through the standalone version of the DESDM pipeline, with (a) no nonlinearity corrections applied, or (b) Huan's nonlinearity lookup tables that were used in SVA1 processing, derived from studies of dome flat scaling with exposure time. SExtractor magnitudes in a 4"-diameter aperture are used below. Stars with MAGERR_AUTO < 0.1 and |SPREAD_MODEL<0.005| and FLAGS==0 were selected. Any pixel above 25,000 ADU was flagged as saturated for this study.

The SExtractor instrumental magnitudes were adjusted by a star flat and color terms derived from dithered observations taken on the same night. My PhotoFit code was used to assign a zeropoint to each exposure to minimize the variation in magnitudes of the stars between different exposures. This fitting suggests up to 0.04 mag of extinction variation during the sequence. PhotoFit then spits out a calibrated magnitude for each star in each exposure.

Next I look at each star and calculate its mean magnitude in the 10s exposures. I can then get a differential magnitude shift vs exposure time, with the 10s defined to be zero.

These residuals are then binned according to the brightness of the star and plotted below.

The signature of nonlinearity would be this: As the exposures get longer, the brighter stars get more nonlinear and accumulate bigger mag residuals. So if there is nonlinearity, we should see a fan-out of the different brightness bins as we depart from the reference exposure time. This is indeed what is seen for the data with no corrections applied. But this fanout is gone (to within our accuracy) for the data with the nonlinearity lookup table applied. So the high-ADU end of the LUT does seem to be doing its job.

For reference: the sky bgrnd at 10 s exposure is 100 ADU per pixel. The brightest bin here (~12.5 mag, not on a calibrated scale) has a peak flux of about 5000 ADU per pixel in the 10s images.

No linearity correction: (note incorrect title on this plot!)

SVA1 linearity correction applied: