In this entry below we suggested 4 things to try to improve the accuracy and precision of our fitting method. A fifth idea is that byt fitting small, sharp-edged regions on the DS we are allowing small shifts in the image to ‘pull’ thge fit. A similar comment applies to the effects of atmsopheric turbulence. We have elsewhere shown that the lunar disc ‘wiggles’ when you look at an animation of 100 images. Some regions (5-10 arcminutes in diameter) shift sideways by what looks like a large fraction of a pixel, relative to other parts. If we use small patches to fit on quality will suffer if we fit across a wiggly patch’s edges.

Using larger patches might therefore be an idea, and we test that here, while keeping all other thinsg fixed. That is, we use the exact same pixels as before but instead of concatenating 11 row-averaged bands we average the 11 bands into one profile. Again, we use the set of 10 ideal images to which realistic noise had been added.

We get:

mean albedo: 0.28031400
std. dev. of 10 fits:0.00021
SD expressed as percentage of mean: 0.08 %

Compare this to what we found before, using concatenated strips:

mean albedo: 0.28051
std. dev. of 10 fits: 0.00030
SD expressed as percentage of mean: 0.11%

It would seem we have about 30% less spread in the values found. As before the known fixed albedo of this ideal image was near 0.2808, so we have moved slightly away from this, and the bias is now about 2 SDs.

The use of averaging the 11 bands into one (they follow the contour of the lunar disc) does seems to ensure that flux is not lost outside a narrow band. We still use radial bins (5 pixels wide) so flux can be moved from bin to bin that way. We hesitate to make larger bins, as the degrees of freedom (surely?) would suffer. We have about 28 radial bins.

So, we have less scatter but more bias! Not sure this is a way it’s worth to go. Alsao recall that thgese are ideal images – we are not sure what happens with real images. One day we might fit all images using the concatenated and averaged versions of the fitting method and see which has the smaller scatter.

We jave more suggestions from last blog entry to try:

1) Use more sky pixels
2) Use more DS pixels
3) Use BS edge
4) Average several images

4. We can always do and is independent of other method choices. 3 is scary, so I will try 1 and 2 next.