9.2 Dark

There are 22 dark exposures, all with exposure time 1 hour, taken on 7 different nights, as listed in Table .

 

Table: The Dark Exposures

image #	dfsc #	night	level
1	1511	2	5.09
2	1512	2	5.70
3	1513	2	7.51
4	1642	3	4.58
5	1643	3	5.04
6	1644	3	5.76
7	2057	6	16.13
8	2058	6	17.07
9	2059	6	23.15
10	2189	7	17.89
11	2190	7	25.09
12	2191	7	40.16
13	2353	8	23.83
14	2354	8	20.96
15	2355	8	22.77
16	2415	9	16.63
17	2416	9	17.90
18	2417	9	22.47
19	2418	9	32.98
20	2509	10	21.68
21	2510	10	31.07
22	2511	10	47.11

 

Notes: ``level'' is the median value (after bias subtraction). The median is approximated by midpt from imstat, using an upper cut of 200 to avoid most of the cosmic-ray-events.

The level histograms of the 22 images is shown in Figure .

 

As is evident from Table  and Figure , the levels in the 22 images are not the same. Two effects are seen:

1.

There is a big difference of the levels before and after the change of CCD temperature, which happened between night 5 and 6.

2.

Within each night, there is a tendency towards higher levels for later times. Since the dark exposures were made in the morning, this indicates a light leak in the camera.

From this it was decided to make two dark frames, one for night 1-5, and one for night 6-14. The following frames were considered not affected by the light-leak, and therefore usable for constructing the two final dark images: 1, 2 (night 1), 4, 5, 6 (night 2), which were combined to the image dark1; and 7, 8 (night 6), 16, 17 (night 9), which were combined to the image dark2. The combination was done using a 2 sigma clipping (avsigclip) with a grow radius of 1. The images from night 8 was not used, since they all had a higher level, even though they did not show the time evolution the other did.

To get rid of the cosmic-ray-events, the values in the tails of the two combined images (dark1 and dark2) were replaced (using imreplace) by the median value, using the following values: dark1: values outside [INDEF,20.] replaced by 5.457, dark2: values outside [7.,27.] replaced by 17.14.

The two images (especially dark1) had a number of columns with higher or lower levels (the lower ones being fat zero, the higher ones probably being remanence), which were removed by linear interpolation of the unaffected neighboring pixels (using fixpix).

The dark images had now two kinds of structure left: some smooth variations, and a discontinuity between y = 512 and 513 (in the coordinates of the final 1024 $\times$ 1024 trimmed image). The two images, dark1 and dark2, where cut in halves according to the discontinuity, and to each of these 4 pieces a Legendre polynomial of third order in x and y with cross terms was fitted. Afterwards, the corresponding halves were joined. It was tested that these two new dark1 and dark2 images did not have any residuals or large scale gradients with respect to the ``unsmoothed'' ones. The two final dark images are shown in Figure .

 

The two images do not exactly scale, the relative standard deviation on the quotient of the two is 2%. Given the low absolute counts, this is a small difference.

The mean dark levels the final two dark images are: dark1: 5.6 ADU/hour, dark2: 17.3 ADU/hour. The low frequency pattern, characterized by the standard deviation of the two images, is 0.12 (dark1) and 0.24 (dark2). Since they are smooth legendre polynomials, there is no pixel-to-pixel noise in them.