(This ticket is a consequence of an attempt to do finer granulation of PIPE2D-928). 

I am noticing problems in my creation of variance images. At the moment I am creating ``custom'' variance images which are created from the model images. To get the relation between the values in the  variance and science images I do a simple 2nd order fit and apply this relation to the model image to get the variance image. The mistake I have made is that I am reusing the previous variance image (instead of the original variance image) to create a new variance image. This leads to runaway values in the variance image and eventually to the crash. 
I have only noticed this now as I have modified the 2nd order fit procedure in order to avoid nonophysical values (which I believe was the origin of strange features in the image ``example_in_focus'' in PIPE2D-615)