Blanton and I worked for a while on sky. It is interesting that such a simple problem (subtract the smooth, additive foreground) is so hard! We decided that the way you subtract sky is a strong function of the scientific question you are asking, and you make different choices in different cases. So to make a sky determination, you have to have a very specific question in mind. This may seem strange, but it is because the sky has structure on all scales, and sometimes you want to photometer a star in a dense field, and sometimes you want to detect low surface-brightness emission behind a field of Galactic stars. There is no one sky
that gives good signal-to-noise for both of these cases. For measuring the properties of huge, well-resolved galaxies, we were using a hack, but now we are getting close. Here's a comparison (before
is on top, after
is below):
The red swath is a single camcol
ReplyDeletewidth where the sky was improperly
subtracted? Why was it so much
worse in one bandpass?
No idea!
ReplyDeleteMy guess is that the g-band and r-band were over subtracted due to the galaxy size, making the i-band look redder. We see this all the time. I just made a nice sky-subtracted image of this guy 2 weeks ago (http://astro.berkeley.edu/~awest/katarina/64.jpeg) for a collaborator that has high res HI data for this system. I have a new technique too that combines my original method and Blanton's all in one. We should really sit doen and talk since I spend time on this too.
ReplyDeleteActually, since we set the zeros hard, it can't be that g and r are low. It has to be that i is high.
ReplyDeleteThe bad i-band sky subtraction
ReplyDeletehas to do with it not tracking
the variation in the sky on short
enough position OR time scales.
Some of the sky is scattered light
from stars and should be pretty
constant with time and vary only
with position in the run. So even
though the g, r, and i images are
taken at different times by a few
minutes, that sort of sky should
not be bad in only one band.
But the sky comes from other sources
than stars fixed in the sky too.
That it happens in one band and
not another might be because the
fluctuations in TIME from such
sources are large --- the sky is
not coming from scattered
light from nearby stars but
something else, that affects all
bands simultaneously in TIME, but
in different positions on the SKY.
So you can look for that signature.
In my sky fits to 4623, I find
exactly that pattern: an upward
fluctation a few percent of sky
first in r, then in i, NOT in u,
then in z, then NOT in g. The
spacing is exactly what you expect
it to be (around 2 fields between
r and i, 4 fields between i in z).
So that sounds like why it is in
one band but not another for this
object.