If we take all our data and plot the flux for each image (total counts divided by nominal exposure time) against the lunar phase (0 phase is Full Moon) at the time of observation we get plots like this:
This is for the B filter, and the top panel shows the raw flux. We see the outline of the expected 'phase law' with some scatter. In the second panel we see the data corrected for extinction (the data are taken at different airmasses and must be extinction-corrected). The scatter is not really reduced, so we conclude that the scatter is mainly due to something else. In the third panel we have more or less taken out the phase law (determining the phase law is a matter of research - BBSO does it with empirical methods, we are going to use the latest Hapke et al phase laws, but at the moment merely correct for the 'geometric illumination fraction').
The factors that cause the scatter that remains could be
a) clouds - i.e. flux is reduced, by the passage of a thin cloud, compared to data at the same phase but without clouds.
b) The shutter failed so that we do not have access to the actual exposure time - merely the one we asked for.
c) The color filter that was used was simply not the one we asked for.
d) The Moon was not centred in the frame and part of the flux is missing because the Moon is outside the edges of the picture. Lunar eclipses also enter as a problem but happen, of course, only at full moon - we have examples of this and know how to eliminate these frames (and full moon data will not be used for DS/BS work anyway).
I think that data suffering from problem a could be eliminated by some sort of image analysis that looks for an uneven sky - but it will be a tough job to separate this effect from c.
Problem b will not affect the DS/BS ratio. Remember that the above is mainly the BS we are looking at! When the shutter fails it often causes 'dragged images' and they can quite easily be eliminated.
Problem c is worse - we know from the problems we encountered, in trying to find the focus for each filter, that the FW does not always select the filter we ask for. Due to the lack of 'polling' in the system design we are victims of 'timeout': proceedings may halt while the system waits for a filter to be selected, but after a preset time the system simply moves on and does the next command, such as an exposure. How can we detect and eliminate the exposures that were taken through the wrong filter? (a future design should have a method to query the system about the identity of the filter in the beam: some sort of coding system?).
In the plot above there are clearly two sequences of flux values for phases between +90 and +150 with a flux-ratio of about 3.5. Is it possible to understand if this is due to selection of a different filter (or no filter at all)?
From experience we know that the flux-ratio between the B filter and no filter at all is likely to be near 10, while the ratio between B and V is like 2 or 3. So - it may not be easy to assign a 'real filter' to each observation on the basis of flux (and phase-information) alone.
Since DS/BS is unaffected by shutter problems but is affected by 'wrong filter' issues we should be able to design a data-examination method. More to follow on this.
Seperately, a delicate analysis method for the presence of drifting clouds in the image is needed in order to eliminate problem a. Suggestions are welcome.
Problem d can be handled with image analysis, and this will be implemented as part of a data-validation filter.