9.7 Overflow and Remanence

When there is a very bright object (star) in the frame, there will be so much charge build up in the corresponding pixels, that the potential well can not hold on to it, and it overflows. For this CCD, it flows upwards (i.e. towards higher y's) along the column(s) (typically 1-5 columns, but it can be even more). To have such an overflow stripe is in itself a problem.

But the real problem comes in the following frames, where one can now see a remanence stripe going downwards from where the star was in the last frame (say, we are now looking at another field without any star at the point where the bright star was in the last frame). In approximately 20 frames (not including a similar number of frames used only for pointing the telescope) after the exposure of the bright star, this remanence stripe can be seen, only fading slowly. The strength of the remanence stripe increases with exposure time. We do not fully understand the physical processes involved with this phenomenon.

It is possible to measure how much extra signal there is in an overflow or remanence affected column compared to the background. For the remanence effect, this signal is (to a good approximation) a linear function of y, although sometimes 2 (or more) linear pieces are needed, with an abrupt jump usually at y = 512. The overflow signal falls off much faster than a linear function in y, but it is not easily approximated by a simple function. A many-piece cubic spline works satisfactory.

The remanence and overflow stripes were removed using the cl-script remanence, written by IJ. It works like this: On the displayed image, the user marks the position of all the stripes and their widths in pixels (only the interval 1-9 is supported, since the width is marked just by pressing one of the keys 1 to 9). Thereafter, for each stripe, the program plots the level along the stripe (i.e. along y). The user then marks a number of intervals in y where a correction needs to be applied. The choices are fitting a linear function, l, or a many-piece cubic spline, s. The appropriate key is pressed at the endpoints of the interval. The program then fits the appropriate function to the stripe, and fits the background on each side of the stripe using a linear function. After this, the difference between the stripe fit and the sky fit can be subtracted from the original image (if this difference is less than zero, no action is taken), and one has the cleaned image.

As the above might indicate, the process involves a lot of manual work, and is rather time consuming. However, it works remarkably well in most cases.

114 of the 123 images were corrected for one or (usually) more remanence and/or overflow stripes. No of the 5 JU images needed correction. 6 of the images had also horizontal stripes (very strong overflow), and this was removed by rotating the images 90 degrees, running remanence on it, and rotating them back again.