Hi,

I was listening to the discussion of wcsremap and hotpants, and I
would like to give a couple of comments:

- Only deprojecting the template, but not the image (i.e. leaving the
image 'pristine'):

* If the image is not deprojected, it still contains higher-order
distortions. One of the side effects of this is that the platescale is
varying across the CCD. If your flattened image is physically flat
(i.e.  the sky background has the same number of counts everywhere),
then this flattened image is photometrically not flat (since the pixel
area projected on the sky is variable). So when you project your
template onto the image, you have to make sure that the projected
template has the same photometrically 'un'-flatness, just to ensure
that the kernel sum is constant over the complete CCD.  The difference
image has then to be either deprojected, or a correction to the
photometry has to be applied on the dB level.

I don't know how much the platescale varies for LSST, but on the 4m it
varies enough that the effect on the photometry is in the % range.

* We did exactly that at the beginning of the Supermacho project: we
'remapped' the template to match the image. We had a failure rate of
several %, and we never could really get it down. By using the WCS we
completely eliminated this failure mode. I think one of the reasons is
that one has to find the transformation of one disorted image to
another distorted image. This is intrisically more difficult than
having two single transformation to a tangential plane.

Especially for LSST there will hopefully be a non-linear WCS solution
that is based on the real physical distortions (e.g. it has radial
terms etc), and not just a random polynomial representation. In that
case it will add another layer of complexity to try to find the
transformation from one distorted system to another (instead of from
the distorted to a simple tan. plane)

* In the Supermacho and ESSENCE project, we project ALL images of a
given field to the same tangential plane.  This has the big advantage
that a given star is always at the same x,y position, and all images
are physically and photometrically flat. Clearly, LSST cannot do it
this way since the exposures are dithered, but I thought I still 
point it out.

I think my bottom line is: I don't think the advantages gained
by not deprojecting the images outweigh the trouble it causes. 

- oversampling the template

One big reason we don't oversample is simply disk space: If both
template and image are oversampled, the required disk space can easily
double or quadruple. For the reasons stated above we opted to
deproject both, the image and template.

- in hotpants: why not use the complete stamp to determine the kernel,
why substamps?

besides the time issue:

* There are two extreme cases: sparse and crowded fields: In sparse
fields, you don't want to use too much sky, so it's better to use only
subareas with significant flux. For crowded fields, the saturated
stars are the problem: if your stamp is too big, then there is a very
high chance that there is a saturated star in the stamp. If you use
the complete stamp to determine the kernel, all stamps with saturated
pixels have to be kicked out. In crowded fields, that can be the
majority.  If you have several substamps for each stamp, there is a
very good chance that at least one substamp does not have saturated
pixels in it.

- hotpants failure modes:

As Andy pointed out, on very rare occasions a satellite can make it
fail.  There are some cases where an extremely bright star is in the
image. The halo of that star cannot be subtracted out correctly, and
hotpants then tries to compensate that with the background and
kernel. One solution to that is to have a list of very bright stars,
and pass a mask that indicates to hotpants to not use any pixels
within a big aperture around this bright star.

The most common reason for not catastrophic, but low-quality
difference images:

* bad seeing. At CTIO in september (end of winter), it is not uncommon
that there are nights with extremely bad seeing (>2.5").  These
difference images are just ugly. Don't know if it is worth to look
into that since these images are anyway of low scientific use.

* lots of saturated stars, or a star cluster.

Cheers,

-Armin


_______________________________________________
LSST-data mailing list
[email protected]
http://www.lsstmail.org/mailman/listinfo/lsst-data

Reply via email to