> Date: Wed, 22 May 2002 17:18:47 -0400 (EDT)
> From: JunXian Wang <email@example.com>
> To: <firstname.lastname@example.org>
> Subject: wavdetect detect source with 0.9 net_photons?
> Dear Colleagues,
> I runned wavdetect on a ACIS-I(0123) image using 10^-7 as
> the sigthresh. It's wierd that I got one source with
> only 0.9 net_photons. I also got several sources which
> have less than 3 net_photons. I cann't believe these
> sources are real. Have any of you met such problem?
Peter Freeman and I looked at your dataset, and I am replying
to your original mail here because others may be interested in
how this issue panned out.
First, the answer to the main question, from Peter:
Peter> The "0.9 count" source was detected at (4395.5,4424.5)
Peter> (logical coordinates [1596,1625]). The PSF here is
Peter> estimated to be 1.26 pixels, so the relevant fluxscale
Peter> is 1 (sigma = 1.0). The source itself was only detected
Peter> at scale 8.0, and barely at that (visually the source
Peter> looks fine, albeit weak).
Peter> The source cell was constructed using a scale of 1 pixel
Peter> for the smoothing function. Since this scale is too small
Peter> for analyzing a source only detected at scale 8 pixels,
Peter> the resulting source cell is too small, containing only
Peter> a fraction of the source's counts (in this case, 1 count).
Peter> Ergo net_counts is too small.
So in summary, the source is real, but the flux estimate is not.
Now, this does illustrate the danger of using a complex algorithm
such as WAVDETECT as a black box. Let me reiterate that WAVDETECT
is designed to be a source *detection* algorithm, and only secondarily
as a source flux measurement tool. There are actually 9 counts that
we can associate with this source if we were looking at it "by eye",
enough for a reliable detection, but at a much larger scale than
suggested by the PSF size. The *estimated flux* is however unreliable
because of the sparse distribution of the counts in this apparently
Note that flux estimation with WAVDETECT is generally quite reliable,
but there is a small fraction of cases where the algorithm will fail
to return useful source parameter estimates. This is inevitable.
There is no such thing as a "perfect source detection algorithm."
Peter also points out that:
Peter> This situation is noted in the WAVDETECT paper. In section
Peter> 3.2.2, we note "...situations where care must be exercised
Peter> in interpreting results...the source cell for an [apparently]
Peter> extended source may be too small if the smoothing scale
Peter> is ~ [the PSF scale]..."
Peter> The workaround for this is to rerun wavdetect with the
Peter> parameter fluxscales changed from "1 2 3 4 5" to "4 5" to get
Peter> a more proper accounting of the source in question.
This archive was generated by hypermail 2b29 : Tue May 21 2013 - 01:00:17 EDT