When to use a single comp versus an ensemble
In another forum the question was asked, concerning VPHOT, "The video tutorial mentions that the ensemble method can "improve accuracy but increase the error rate". Could someone please elaborate a bit on the subject? Under what circumstances is either method preferable?"
I would like to post my reply to this forum as well and ask for feedback on another area of the tutorials, which we are now working on updating to present enhancements to VPHOT.
Here was my reply:
Your question is a very good one. When writing the tutorial, teaching how and when to use ensemble versus a single comp star was beyond the ken of the tutorials time and focus. So the disclaimer you mentioned was added so that it would alert you to seek more knowledge on this subject, which you have correctly done!
I'm sure there are many who can provide better details on resources to give a full explanation of this issue and a detailed answer to the many things to consider.
For the purposes of VPHOT, we thought it was necessary to point out that while the ensemble method is ideally preferable, it is not always practicle, and that always blindly applying it can cause more harm than good So the short answer to the question is that what to do depends, among other things, upon the quality and type of the comp stars available.
For example if there are ten comps in the field, but nine of them have a large error value or are very far from the magnitude of the target--whereas one comp which has a low error and is in the same ball park magnitude wise as the target--then that is a case where using just the one comp would yield a more accurate result in some ways--but there is always the risk that other things might bring down the overall confidence level of the measurement (not really quantifiable, though I may be incorrect).
On the other hand, when more than one to many comps are available with qualities such as a low error value, where they are close to the target as oppsed to near the edge of the FOV--which have ADU, SNR and magnitude values within certain parameters which others or I can describe later in more detail--then it is most always better to use an ensemble.
This is the fun part which requires some discretion and which requires a bit of experience.
So to begin with, never just blindly use all the comps in the field. If there are many from which to choose, they should ideally have an error of less than .01-.02 and certainly not more than .05, and ADU of at least 10,000 but not more than 30,000, an SNR of at least 100 as well as magnitude comparable to the target, (for example if the target is around 120 then try to avoid using a comp of 150).
I think I will stop there for now as a reasonable point of departure.
We are just starting to work on a follow up tutorial to address enhancements to VPHOT as well as a manual, so I am glad you raise the issue as I am still not sure if we should questions like this in the second tutorial in more detail--there is so much material to cram into the tutorial just on how to use the software, that I tend to think how to actually do photometry--which your point addresses, should be addressed in other materials.
But please everyone, let us know what you think. Also I would bne interested to know if people will prefer a voice over tutorial such in the same style as exists now, or whether showing the same type video with instructions and commentary written in printed captions is more desirable.
Ken Mogul (MQE)