GPC Dark Trends

I analyzed a set of 140 dark exposures taken during run 1c in the period from 02/08/08-02/11/08. The exposuretimes range from 12 to 100s, the detector temperatures range from -90 to -60. The figure below shows expodure time vs. temperature of the dataset:


ccd36/cell2 analysis

I selected a region x=461..510,y=81..130 in chip36/extension2 with a relatively prominent dark feature:


In the first step I took the average of the region and made a second order polynomial fit to the exposuretime (DARKTIME keyword). I found that the first two exposures of each of the 10 dark series have significantly lager counts that the other ones. I plot the residuals after subtracting the best fit as a function of exposure number withing series (different series color-coded):


The first blue point s not displayed and has a residual value of 200. From now on I dropped the first two exposures of each series. The two figures below shows the linear fit in exposure time of the remaining 120 dark images and the residuals as a function of temperature (DETTEM keyword):


There seems to be no clear 2nd order trend like you would expect it(monotonically rising with temperature), so I decided to look at the 30s exposures only and fitted a linear relation to the dark current as a function of temperature:


The next step is to do the analysis pixel-by-pixel. I used again the 120 images with different exposure times and fitted a linear relation in time and linear relation in temperature. The coefficients for each pixel are stored in a fits file:

A1_1.jpgA1_2.jpg A2_1b.jpgA2_2.jpg

Using this basis a dark of any arbitrary exposure time and temperature can be created: dark = a1_1 * time[s] + a1_2 + a2_2 * temperature[C] + a2_2. Of course, a1_2 and a2_2 could be combined. Below there are some example residual images:


Also interseting to see are the nchi-images which show the quality of both fits, nchi = sqrt( SUM[ dark_residual2 ] / N ):


Cosmics are clearly visible and should be removed to identify the pixels that are really bad. To do this I included an iterative kappa-sigma-clipping to the fit. This is the resulting nchi-image of the time-fit:


Ok, this particular cell (chip36/extension2) seems to be quite 'friendly', meaning correctable. But what about the others?

full detector improved analysis

In the following I show the results of a full-detector analysis. I dropped one more raw image, because it turned out to have zero values only is some of the chips. Also I changed from doing a linear fit in time and temperature to a 2-dimentional fit: dark = a0 + a1 * time + a2 * temperature.