[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
We have been discussing the uncerainty in photometric measurements
made with Mark IV cameras.
Michael K. wrote:
> I'm dumb, but isn't there an ensemble solution for each frame?
> Wouldn't the error in the fit be a suitable error for each star in
> that frame? (i.e. the zeropoint error).
I suggested a simple table of _empirical_ scatter from the
> 1) table of scatter of Mark IV database magnitudes from
> their mean value, as function of magnitude,
> something like this (I'm making up values here)
> V mag 7 8 9 10 ....
> V uncert 0.03 0.05 0.07 0.10 ....
> Well - that's an omnibus first guess but it throws away the
> information about variable sky brightness and variable
> aperture size.
This is a complicated issue. Let me try to describe (some of) the
various contributions to the uncertainty in each measurement.
a) photon noise -- to get this right, you need to know
the number of electrons measured within the stellar
aperture, the number of background electrons in
the aperture, the uncertainty in the background,
and so on. Andrew is focusing on this item.
b) instrument errors -- we know that the Mark IV units
suffered from a variation in sensitivity across
the (very wide 4x4-degree) field, which was
not perfectly corrected in the flatfielding.
This is independent of the stellar brightness
and aperture details.
c) calibration -- we compare the instrumental magnitudes
for each frame to the magnitudes of a subset of
stars in the Tycho-2 catalog. If those catalog
values are in error, or some of the relatively few
(20-40 per frame) calibration stars have bad measurements
on the image, or if the instrument passband
is not the same as the catalog passband, we'll
make errors which affect all stars in the frame.
Michael K. is thinking about this source of error.
Each of these sources is present in the Mark IV data. For the
bright stars, sources b) and c) dominate. For the faint sources,
type a) is usually most important. No single sort of error
is always the most important one.
The information needed to compute the photon-noise/electron-noise
errors is partially available in the on-line database. Information
on the size of apertures, readnoise, background sky values, is all
present in the .param files which can be grabbed from the
database. Some intermediate files with exact values for
number of counts from a star, number of counts from the
background annulus, uncertainty therein, are probably _not_
saved in the database. I'm not sure if they are easily
accessible anywhere -- I suspect not.
The information needed to evaluate the effect of large-scale
"flatfield" errors has not been dragged out of the data itself,
except for a very small first step described in Tech Note 97.
The frame-by-frame catalog-type errors are hard to isolate.
One can see overall systematic color terms, perhaps, by comparing
Mark IV photometry to catalogs other than the Tycho-2.
A few Tech Notes show how one can do this, and provide some
information. The .param files saved in the database do have
a sort of nightly summary of the accuracy of the photometric
solution for the night as a whole.
When I mentioned "ensemble photometry", I was referring
not to the one-frame-at-a-time reductions of the Mark IV pipeline,
but to a later analysis of all measurements of the stars in
a small region of the sky, from all nights over all years.
This can _sometimes_ reduce the large-scale "flatfield"
errors in the data, although Tom is certainly right to be
skeptical: there are occasions when it fails to help at all,
and we have only anecdotal evidence at this point for its
utility on a large scale.
I hope to do some work on this last item in the near
future. It produces as a by-product information which can
be used to make an overly-simplified table of "uncert vs. mag"
which I mentioned in my earlier E-mails. As Andrew states,
that very very broad sort of knowledge of uncertainty
just doesn't cut it for some types of research :-(