Simulated data, inconsistent solve

[Steve Foale]

Hi

We have an issue solving simulated fields (i.e. in a telescope simulator we generate realistic-ish fields from a catalog then send them to astrometry.net to solve).

Attached is a fits file of a simulated field and a list of extracted sources in binary table format. The binary table is just a copy of the .axy  obtained from astrometry.net version 0.93 by solving the fits file.

The resulting solutions are hugely different.

solve-field --crpix-x 1023.0 --crpix-y 1023.0  --no-tweak  --radius 5.0 --ra 300.0 --dec 10.0 --guess-scale --scale-units arcsecperpix --scale-low 0.36000000000000004 --scale-high 0.44000000000000006 --no-plots --no-remove-lines -N none --code-tolerance 0.03 --pixel-error 1 -d 1-200 --solved none --match none  --corr none --overwrite -X X -Y Y -s FLUX --width 2048 --height 2048 test.fits -b /etc/astrometry.cfg -D /tmp
Reading input file 1 of 1: "test.fits"...
Found an existing WCS header, will try to verify it.
Extracting sources...
simplexy: found 120 sources.
Reading sort column "FLUX"
Sorting sort column
mmapping input file
Copying table header.
Writing row 0
Done
Solving...
Reading file "/tmp/test.axy"...
Only searching for solutions within 5 degrees of RA,Dec (300,10)
Verifying WCS using indices with quads of size [1.36533, 13.6533] arcmin
Verifying WCS with index 3 of 10 (/data/Astronomy/astrometry.net/data//index-4205-07.fits)
Verifying WCS of field 1.
  log-odds ratio 70.3851 (3.69718e+30), 12 match, 0 conflict, 38 distractors, 12 index.
  RA,Dec = (300,10.0001), pixel scale 0.4 arcsec/pix.
  Hit/miss:   Hit/miss: -------------+-+-----++----+--+----+-+-+-----+--++(best)--------------------------------------------------
 --> log-odds 70.3851
Verifying WCS with index 4 of 10 (/data/Astronomy/astrometry.net/data//index-4205-03.fits)
Verifying WCS of field 1.
 --> log-odds 0
Verifying WCS with index 5 of 10 (/data/Astronomy/astrometry.net/data//index-4204-30.fits)
Verifying WCS of field 1.
  log-odds ratio 62.8933 (2.06166e+27), 14 match, 1 conflict, 49 distractors, 27 index.
  RA,Dec = (300,10.0001), pixel scale 0.4 arcsec/pix.
  Hit/miss:   Hit/miss: -------------+-+-----++----+c-+----+-+-++----+--++-------------+(best)----------------+-------------------
 --> log-odds 62.8933
Verifying WCS with index 6 of 10 (/data/Astronomy/astrometry.net/data//index-4204-12.fits)
Verifying WCS of field 1.
 --> log-odds 0
Verifying WCS with index 7 of 10 (/data/Astronomy/astrometry.net/data//index-4203-30.fits)
Verifying WCS of field 1.
  log-odds ratio 49.7924 (4.21264e+21), 14 match, 0 conflict, 38 distractors, 55 index.
  RA,Dec = (300,10.0001), pixel scale 0.4 arcsec/pix.
  Hit/miss:   Hit/miss: -------------+-+-----++----+--+----+-+-++----+--++-+(best)-----------+----------------+-------------------
 --> log-odds 49.7924
Verifying WCS with index 8 of 10 (/data/Astronomy/astrometry.net/data//index-4203-12.fits)
Verifying WCS of field 1.
 --> log-odds 0
Verifying WCS with index 9 of 10 (/data/Astronomy/astrometry.net/data//index-4202-30.fits)
Verifying WCS of field 1.
  log-odds ratio 35.4403 (2.46333e+15), 14 match, 0 conflict, 38 distractors, 120 index.
  RA,Dec = (300,10.0001), pixel scale 0.4 arcsec/pix.
  Hit/miss:   Hit/miss: -------------+-+-----++----+--+----+-+-++----+--++-+(best)-----------+----------------+-------------------
 --> log-odds 35.4403
Verifying WCS with index 10 of 10 (/data/Astronomy/astrometry.net/data//index-4202-12.fits)
Verifying WCS of field 1.
 --> log-odds 0
Got 4 solutions.
Field: test.fits
Field center: (RA,Dec) = (300.000169, 10.000167) deg.
Field center: (RA H:M:S, Dec D:M:S) = (20:00:00.041, +10:00:00.600).
Field size: 13.6533 x 13.6533 arcminutes
Field rotation angle: up is -0 degrees E of N
Field parity: neg

Versus the binary table

solve-field --crpix-x 1023.0 --crpix-y 1023.0  --no-tweak  --radius 5.0 --ra 300.0 --dec 10.0 --guess-scale --scale-units arcsecperpix --scale-low 0.36000000000000004 --scale-high 0.44000000000000006 --no-plots --no-remove-lines -N none --code-tolerance 0.03 --pixel-error 1 -d 1-200 --solved none --match none  --corr none --overwrite -X X -Y Y -s FLUX --width 2048 --height 2048 test.lis -b /etc/astrometry.cfg -D /tmp
Reading input file 1 of 1: "test.lis"...
Reading sort column "FLUX"
Sorting sort column
mmapping input file
Copying table header.
Writing row 0
Done
Solving...
Reading file "/tmp/test.axy"...
Only searching for solutions within 5 degrees of RA,Dec (300,10)
Field 1 did not solve (index index-4206-07.fits, field objects 1-200).
Field 1 did not solve (index index-4206-03.fits, field objects 1-200).
  log-odds ratio 58.4371 (2.39271e+25), 9 match, 0 conflict, 44 distractors, 13 index.
  RA,Dec = (300.231,9.99988), pixel scale 0.399995 arcsec/pix.
  Hit/miss:   Hit/miss: -++--------+----+-+-------+--+---------+------------+(best)-----------------------------------------------
Field 1: solved with index index-4205-07.fits.
Field: test.lis
Field center: (RA,Dec) = (300.231121, 9.999938) deg.
Field center: (RA H:M:S, Dec D:M:S) = (20:00:55.469, +09:59:59.778).
Field size: 13.6531 x 13.6531 arcminutes
Field rotation angle: up is -0.00378732 degrees E of N
Field parity: neg

The solved RA is around 800 arcsec different, which happens to be the approximate width of the simulated image.

The nova service solves the image fine - see here https://nova.astrometry.net/user_images/8245102#annotated

What am I doing wrong?

Cheers

Steve

[Dustin Lang]

That's bizarre!!  I don't have a trivial explanation for what you're seeing.

One observation in the log files you've sent here is that the first one is using the "verify" mode -- it has noticed that your input FITS image already has a WCS header, and it is checking whether detected star positions are consistent with that -- in other words, it's not doing a "blind" solve.

Another observation is that, looking at the nova "red-green" plot, there doesn't seem to be really strong overlap between your simulated stars and the index stars.

Is there any chance that your simulated star field actually includes stars from a wider field that are wrapped into your 2048-pixel box?

It is surprising to me that you get a very-nearly 2048 pixel shift in only one dimension.  If it was both x and y I would be more suspicious of a code error :)

cheers,

dustin

--
You received this message because you are subscribed to the Google Groups "astrometry" group.
To unsubscribe from this group and stop receiving emails from it, send an email to astrometry+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/astrometry/d165fdb0-d953-4284-b0c9-e93cfaebe679n%40googlegroups.com.

[Dustin Lang]

Also -- looks like maybe half your stars are very bright, with fluxes in the 16k-60k range, while the rest are in the 10-20 range.

[Dustin Lang]

Heh, so I think you've got something funky going on in your sims ....

If I crank down the --radius and tell it to look around 299.76, I get a solution

https://nova.astrometry.net/user_images/8245273#annotated

If I look around RA=300.0, I get a different solution

https://nova.astrometry.net/user_images/8245299#annotated

If I look around 300.24, I get a different solution:

https://nova.astrometry.net/user_images/8245301#annotated

cheers,

dustin

[Steve Foale]

Thanks for looking in to this, and you are right, there was something bad going on with the simulated images - now fixed and simulated images are now solving nicely for WCS.

[Dustin Lang]

Glad to hear it!  There's good precedent for this, by the way -- in early LSST sims, there was something like the stars were drawn from a catalog but the galaxies repeated every 5 degrees and they were surprised when the astrometric solutions were so far off!!

cheers,

dustin

To view this discussion on the web visit https://groups.google.com/d/msgid/astrometry/b90ce2a9-cc24-4ce6-b39b-66d386201033n%40googlegroups.com.