IRAC is incredibly stable over a timescale of a couple days with aparrent stochastic errors on the order 0.3-1 DN. There are, however, systematic effects (similar to the first-frame effect), and in the case of channel 3 these are quite large (10s of DN). These systematics appear well-behaved and could easily be removed by an observer. However, it is not entirely clear how repeatable they are, and hence whether a routine calibration curve could aid in removing them.

2. Test Data

During the three-day stability test a numbers of IERs were executed that had been designed to mimic the AORs to be used by the Legacy teams. One of these (designed to mimic the Evans team AOR) utilized high dynamic range obervations, specifically, a series of exposures like:

0.6 sec frametime
12  sec frametime
10 sec delay
0.6 sec frametime
12  sec frametime
10 sec delay
...

Illustration of the relative time gaps between runs of the Evans IER.

There were eight repetitions of this IER, generally separated by a couple hours, with some longer 12-hour timescale gaps. The lamps were off during this time, so these are basically darks. After the data was collected, it was processed with the SSC pipeline "pre-processor", which corrects for the InSb sign flipping, the wraparound effect (from bit truncation), and the fowler-sampling/barrel-shifting. The median of the pixels was calculated from a 50x50 box in the center of each array. For initial sorting the time of observation was determined from the DATEFILE keyword (since the other times such as SCET seem to have been reset several times during the test). The data was also sorted by channel and wait number.

3. Results

Channel 1

The following figure shows median DN values of the frames. The horizontal axis is the frame number increasing sequentially in time. The light grey lines demarcate individual runs of the IER.

The short frame times are obviously super-stable. Their scatter over the whole test is 0.3 DN. For the 12-second frames there is obviously something more complex happening. Intuitively, we woule have expected the first long and short frames to have been different, owing to the first-frame effect. Since the timing is always the same on the following exposures, we would have expected them to have nearly the same values all the time. The plan was to correct for the first frame effect, and then subtract the same pair of offsets from each HDR pair. We do actually see roughly two offsets for each exposure time. However, we see above that the first run after a long hiatus has a different offset than consecutive runs (this occurs after the 3rd and 6th run). We also see what looks like a drift downwards within a run and a slight drift upwards between all the runs. The difference is about 2 DN. If we just used a constant value all the time for the 12second frames, we would have a 1 sigma scatter of 1.1 DN.

Channel 2

Channel 2 appears to be the most stable array. The short frame times are the same over 3 days with a scatter of only 0.23 DN. Although there is no first-frame effect, the longer frametimes show a systematic downward trend when executed nearly sequentially, which seems to reset after long intervals. The total amplitude of this drift in a single AOR is only about 0.5 DN, and indeed if we just used the same offset all the time throughout the test the 1-sigma scatter would only be 0.5 DN.

Channel 3

Expectedly, channel 3 is the most pathological of the arrays. Both the long and short frame times exhibit the expected "first-frame effect", i.e. the first frame taken in the IER is markedly different from all of the remaining ones. Furthermore, the offset drift seen in the other channels within a series of identically timed HDR exposures is also seen here, but it's amplitude is much larger. For the short frame time it is about 1 DN peak-to-peak. For the longer frame time it is much larger. The first-frame effect is around 70 DN and there is a roughly 15 DN peak-to-peak additional drift. I don't know yet why the very first run was so different from all the following ones. Maybe the camera was still equilibrating after turn-on.

Channel 4

Channel 4 is also very stable. The HDR drift seen in all the other arrays does not seem to appear. The first-frame effect can be clearly seen in the short exposre times. Scatter in the 0.6 second frames is only 0.36 DN, and in the long 1.15 DN.

4. Bottom Line If We Did Nothing

It's obvious from the above that there are systematic effects in the IRAC arrays which we currently don't correct and for which the planned routine calibrations would not provide a correction. So what if we just ignored those systematics and used a single number for each frame time? We would get, 1-sigma scatter in DN: