Hello,
as this is the time of the year to re-calibrate, I did my homework and shot M67 last week with filters b, v, r and i. Applying what I learned in 2018 in CCD Photometry 2 CHOICE course, I processed by data using VPhot (with a long sequence of 147 stars) and calculated my 2020 coefficients in Excel. I am however a little perplex with the graphics for my transforms and wonder if there is something suspicious in my data. Can some old wise (wo)man look at this graph and tell me the result is acceptable? The first 3 graphs look good to me but the B-Mag and V-Mag transform graphs look a little too spread and I question the linearity of the transform.
Btw, I am using an Orion SSDS III mono camera (Sony ICX 285AL sensor).
Thanks a lot for your help and advise.
Normand
Hi Normand:
That scatter is not unusual. There are a lot of main sequence yellow stars (B-V=0.6) in the cluster. The important part is the presence of highly colored stars to help define the ends of the slope line.
IF you use TG to generate your coeffs with the same images, you may find that you can deselect some of these stars that may be beyond 2 SD. Try this and compare the results with your spreadsheet technique. It is an interesting exercise.
Ken
Hi Normand,
I think the "scatter" near B-V = 0.6 is mostly a perception issue. There are a lot of points here, with many of them partially overlapping in a dense mass near the linear fit line. At the same time, the large number of points also means that there are more outliers, making the data look noisier. Having too much data is rarely a problem, though! This lets you eliminate some of the outliers as Ken suggested. These outliers might be stars that are contaminated by neighbors in your images, or ones that had a noise/cosmic ray hit on an image (assuming you're averaging values), or even undetected variables (not likely in M67, though).
Shawn (DKS)
Thank you Ken and Shawn for your advice. Much appreciated. I didn't average the image, I used median to stack, so your first guess is the most probable cause of those outliers.
Often those outliers are just fainter stars and when limited overall number of standards is used, IMHO there is no good justification to use every possible datapoint.
I'd suggest:
And even better - fit the regression using weighted least squares method. Usually good weight is 1/merr2. The result of the weighted vs non-weighted (read: considering all datapoints having equal, perfect, quality) fitting on the same dataset can be quite dramatic. I'm dreaming about TG having that weighting built-in....
Best wishes,
Tõnis
It is:
Use a relatively small number of data points (in practice, about 9 to 12, but it could be more if targets are available); select targets with a wide range of B-V values, which are distributed across the entire range; select targets within a restricted magnitude range; use exposures to optimize the S/N ratio; delete obvious outliers after plotting (in practice, one or maybe two).
I don't use weighted regression, because I don't believe it is necessary. My understanding is that it is necessary is there is a systematic gradation of variance for a parameter. A theoretical example would be if the variance in B-V grandually increased as the B-V value increased. This sort of thing does not occcur with any of my transformation plots.
I determined TCs using the above principles recently, then used those TCs in measurments of 17 target/comparison star pairs (they were all standard stars). I achieved just over 14 of 17 V mag measurements that differed from catalogue by less than 0.01, and the remaining 3 differed by no more than 0.015. B and B-V were not quite as good, but only 4 of 17 measurements differed from catalogue by >0.03. Six of the B-V measurements and 7 of the B measurements differed from catalogue by <0.01.
Roy