We have been looking at the 'pedestal' which is a part of the model we fit to our observations. With a black sky and perfect bias and dark-current removal, the pedestal should be 0, and the fitted model should have a halo that sloped off from the Moon and perfectly explained all brightness seen on the sky.
In reality, there is background galactic light, zodiacal light, airglow and any terrestrial light (Moonshine reflected from the ground!!), to allow for. To this end we have operated with this pedestal in the model, which was a vertical offset of the model. In looking at fits we notice that the pedestal is degenerate with the halo-parameters in such a way that they tend to lie along a line when a lot of fits are compared, each started at slightly different points - even for the same image repeatedly fit. The use of this pedestal therefore leeds to a bias in the determined albedo. Which is not good.
We therefore have tried to evaluate expressions for the additional sky brightness so that the pedestal could be fixed at an assigned, physics-based, value and not interfere with the fit itself.
In doing that we had reason to revisit and elaborate on Henriette's error budget analysis. In particular, we performed the analysis allowing for error terms in DS and BS fluxes, dark currents, and exposure times. We introduced the uncertainty on the superbias factor and see now that it is the dominating factor when area-means are considered.
For area-means of 20x20 points, the error due to the superbias factor is as large as the error due to Poisson statistics on the DS. For smaller areas than 20x20 the DS Poisson error dominates.
Since levels of sky brightness are of the same order as the contribution due to factor uncertainty, or larger, we see that we have a problem.
We could try to estimate sky brightness better, but as we realize that fitted albedo varies, for a single image treated in different data-reduction ways (bias removal, and stacking), we need to also re-consider the use of 'anchoring' of the model halo in the DS sky part of the image. This is what BBSO achieve by fitting a line to the DS sky and subtracting it from the image.
Improvements in estimating sky brightness could also be pursued, but all constant contributions will disappear in 'anchored' procedures. The point is to not let the fit be biased by such terms.