For the night 2456003 we now have both EFM and FFM results (and BBSO method results, to be analysed later).
We look at the mean difference and standard deviation of various quantities determined from the images by the two algorithms. The differences are calculated as percentages thus: 100*(FFM-EFM)/((EFM+FFM)*0.5). Data are for about 40 observations with a mix of filter repetitions – since these are differences the mixture of filters should not matter as same-filter images are compared.
alfa Mean: -0.3805 SD: 0.2027
a Mean: -66.64 SD: 0.6907
BS Mean: -0.0004525 SD: 0.0001109
TOTall Mean: -5.082 SD: 0.5449
TOTnoPed Mean: -0.01426 SD: 0.003551
DS23 Mean: 30.68 SD: 7.404
DS45 Mean: -7.415 SD: 9.557
x0 Mean: 0.9189 SD: 1.491
y0 Mean: 0.06020 SD: 0.3928
radius Mean: -11.19 SD: 1.802
The difference in offset a is imposed by the codes and can be ignored. The two totals are: TOTall=sum of pixel intensities in image, TOTnoPed=sum of image pixels minus the estimated bias and sky brightness pedestal (i.e. most close to ‘total original light in the image without sky’). The two DS values are calculated at 2/3 and 4/5ths of the radius on the DS of the disc. The difference is radius is due to user setting and is and error I can see now. Ups. More later!
We notice that disc center coordinates (x0,y0) have a bit of variance – the algorithm does not find the same lunar center and radius each time.
Fixing the codes so that radius is determined in the same way we rerun the analysis and get:
DS23 Mean: 21.70 SD: 5.806
DS45 Mean: 1.648 SD: 6.742
x0 Mean: 0.7158 SD: 1.765
y0 Mean: 0.1147 SD: 0.2926
radius Mean: 0.2838 SD: 1.524
The problem on radius is solved now. The other results are more or less the same – DS23 is significantly biased while DS45 is not (given the scatter).
What can we learn from this analysis? The error on DS and total flux is very small. DS45 is not biased, but the scatter is very large – that is, the EFM and FFM methods arrive at the same answer but both are subject to noise.