varying the pause between subexposures does not help that much, unless you pause for a very long time. canon cameras sample the temperature at the beginning of the exposure, so your temperature will seem to be low but in fact the sensor heats up again pretty quickly once the shutter is open.
ian, you should shoot a bias frame at the end of your dark and see what the temperature is. the true temperture of your dark will be somewhere between what the dark EXIF says and the bias EXIF.
Yes I understand that the camera heats up over the duration of the exposure and that the EXIF temperature is only a sample taken at the start of the exposure. I don't need to re-sample the temperature again at the end of the exposure, since I am starting at the same temperature each time (9C in this case), the exposure length is the same (10 minutes) and the environmental temperature is constant (fairly close to 0C once the camera and fridge have been pre-cooled for a couple of hours). Since all these factors are equal, the temperature gain from start to end of the exposure is going to be equivalent each time.
You do indeed need to insert a reasonably long pause between exposures. In my case a 2 hour initial cool down, followed by a repeating cycle of one 10 minute dark followed by a 20 minute pause produces a constant EXIF temperature of 9C at the start of each exposure. After about 8 or 9 hours, the temperature starts to rise as the outer layer of the ice packs melt (they are solid in the core still, but not as effective at cooling). This is not an issue since I just set up APT with the relevant plan (bias exposure with a 2 hour pause, then a large number of 10 minute exposures with 20 minute pauses) and walk away.
Of course this is not as satisfactory as having a set point cooled camera, as the rate of dark current accumulation will increase as the exposure goes on, but this mimics the behaviour of light frames which have the same problem. Ultimately the dark scaling process tries to match the dark frame and the light frame by analysing the noise in each, so to my mind (perhaps erroneously) the best compromise is to start with a master dark frame that is created from a large number of matching frames rather than a bunch of darks that have been taken under different conditions.
The other wrinkle is the on-camera processing over which we have no control. The behaviour seems to be that in the first minute or two of exposure, the dark current decreases (when it should increase) but the dark current noise increases. Thereafter the behaviour seems more normal as the dark current and noise both increase fairly linearly with time. This is why I aim for 10 minute exposures rather than very short ones, in the hope that the dark scaling process has a better chance of succeeding.
