Astrometrical Accuracy Issue - Cross Match with Gaia DR1 TGAS

[Cédric Pereira]

Hello,

I'm working in an asteroid detection project and I need to calibrate my images with very high accuracy.

I have tried different methodologies but without the success that is needed.

The idea is calibrate the images astrometrically and then, cross-match the results with Gaia DR1 TGAS to check the accuracy.

For the astrometrical calibration we want to use "Astrometry.net" (local version). For the cross-match we use "VO TOPCAT". For extract more sources, we use "Sextractor".

For the sample image that we are using, there are 19 possible match stars with Gaia DR1 TGAS catalog.

Here is some results from different methods:

#1

Astrometry.net > XY-RADEC Converter > TGAS Cross-match
(standard astrometric calibration with "Astrometry.net" > convert xy-file given by "Astrometry.net" to new radec-file using "wcs-xy2rd" > cross-match new radec-file)

622 sources extracted - 19 sources matched

Stastistical from matched sources ["separation" - angular distance in arcsec]:

Standard Deviation = 0.459

Min. Value = 0.211

Max. Value = 1.794

Remarks: This is the best accurate result that we got, but with few sources (622). As we need more sources, we have tried to extract sources with "Sextractor", as you can see in the next method.

#2

Astrometry.net > Sextractor > TGAS Cross-match
(standard astrometric calibration with "Astrometry.net" > extract sources with "Sextractor" using *.new file given by "Astrometry.net"  > cross-match radec-file given by "Sextractor")

748 sources calibrated - 18 sources matched

Stastistical from matched sources ["separation" - angular distance in arcsec]:

SD = 2.342

Min. Value = 0.246

Max. Value = 8.609

Remarks: With "Sextractor" we got more sources (748), but the accuracy is worst than method #1. In the documentation says that we can use a command (--use-sextractor) for "Astrometry.net" use "Sextractor" to extract the sources. Let's try...

#3

Astrometry.net & Sextractor > XY-RADEC Converter > TGAS Cross-match

(Astrometry.net use Sextractor to extract sources and then, perform standard astrometric calibration > convert xy-file given by "Astrometry.net & Sextractor" to new radec-file using "wcs-xy2rd" > cross-match new radec-file)

778 sources calibrated - 18 sources matched

Stastistical from matched sources ["separation" - angular distance in arcsec]:

SD = 1.068

Min. Value = 0.299

Max. Value = 5.138

Remarks: With "Astrometry.net & Sextractor" we got even more sources (778), the accuracy is better than method #2, but worst than method #1. Let's try a similar method...

#4

Sextractor > Astrometry.net > XY-RADEC Converter > TGAS Cross-match

(extract sources with Sextractor > standard astrometric calibration with "Astrometry.net" > convert xy-file given by "Sextractor" to new radec-file using "wcs-xy2rd" > cross-match new radec-file)

748 sources calibrated - 18 sources matched

Stastistical from matched sources ["separation" - angular distance in arcsec]:

SD = 1.402

Min. Value = 0.098

Max. Value = 6.594

Remarks: This is only a similar method... From this point, we have searched and found that maybe "Sextractor" cannot deal with SIP distortions given by "Astrometry.net". Some searches and we have found this forum thread: https://groups.google.com/forum/#!topic/astrometry/qrNW5KtsNCc Let's try to understand better this issue...

--------------------------------sip2pv methods----------------------------------------

#2

Astrometry.net > Sextractor > TGAS Cross-match
(standard astrometric calibration with "Astrometry.net" > extract sources with "Sextractor" using *.new file given by "Astrometry.net"  > cross-match radec-file given by "Sextractor")

748 sources calibrated - 18 sources matched

Stastistical from matched sources ["separation" - angular distance in arcsec]:

SD = 2.342

Min. Value = 0.246

Max. Value = 8.609

Remarks: This is only the method #2... only for reference values.

#5 

Astrometry.net --tweak-order 4 > Sextractor > TGAS Cross-match
(astrometric calibration "--tweak-order 4" with "Astrometry.net" > extract sources with "Sextractor" using *.new file given by "Astrometry.net"  > cross-match radec-file given by "Sextractor")

748 sources calibrated - 15 sources matched

Stastistical from matched sources ["separation" - angular distance in arcsec]:

SD = 3.137

Min. Value = 0.689

Max. Value = 9.845

Remarks: To use the sip2pv converter given in the forum thread, It's necessary to perform the astrometric calibration with the "--tweak-order 4" command. The results are worst than method #2 and only 15 sources was matched. But now, let's try to convert the SIP distortions to PV and use Sextractor....

#6 

Astrometry.net --tweak-order 4 > sip2pv > Sextractor > TGAS Cross-match
(astrometric calibration "--tweak-order 4" with "Astrometry.net" > convert SIP distortions to PV > extract sources with "Sextractor" using *.new file given by "Astrometry.net"  > cross-match radec-file given by "Sextractor")

748 sources calibrated - 15 sources matched

Stastistical from matched sources ["separation" - angular distance in arcsec]:

SD = 1.153

Min. Value = 0.184

Max. Value = 5.531

Remarks: Okay, the results are still bad and only 15 sources matched. If we perform the astrometric calibration without "--tweak"? Let's try...

#7 

Astrometry.net --no-tweak > Sextractor > TGAS Cross-match
(astrometric calibration "--no-tweak" with "Astrometry.net" > extract sources with "Sextractor" using *.new file given by "Astrometry.net"  > cross-match radec-file given by "Sextractor")

748 sources calibrated - 18 sources matched

Stastistical from matched sources ["separation" - angular distance in arcsec]:

SD = 1.861

Min. Value = 0.319

Max. Value = 8.614

Remarks: Comparative with method #2, the standard deviation is better... why? If "Sextractor" cannot deal with SIP distortions, comparative with method #2 that include SIP, why the results are different?

-----------------------------------------------------------------------------

Conclusions: Maybe there are a compatible problem with "Astrometry.net" and "Sextractor". Maybe not. Maybe "Sextractor" is returning worst values oy "Astrometry.net" cannot reach the accuracy that we need. I don't know... Does someone have an idea that can improve our results? A different methodology?

Greetings and thank you in advance...

Cédric Pereira

[Dustin Lang]

- be sure that you are comparing the correct coordinates -- if your image is taken in 2018, you should be using the TGAS proper motions to move the stars to their 2018 positions, right?
- you should look at the images to see whether the source detection is working and giving you good centroids
- you can ask the built-in source detector in Astrometry.net to reduce the threshold (producing more stars) using the "solve-field --nsigma N" option -- by default N=8
- you haven't said what kind of accuracy you think your images should be able to provide :)
- un-clipped standard deviation is not really a good metric to use here -- consider that if you matched all but one of the available Gaia stars and then increased the matching radius until the last one matched, then that one large match distance would completely dominate the STD of the sample.  Instead, I would suggest plotting the results and looking at them.

Did you ever figure out the issue you reported before of the index_{ra,dec} coordinates not matching Tycho-2?  They are supposed to be very close (micro-arcsec) to the Tycho-2 catalog values.

cheers,
--dustin

[Cédric Pereira]

Hi Dustin,

- be sure that you are comparing the correct coordinates -- if your image is taken in 2018, you should be using the TGAS proper motions to move the stars to their 2018 positions, right?

> I think the software is comparing correctly the images. There are many stars without significant proper motions and as you can see in the next table, there are stars with high "angDist" and lower "pmRA pmDEC" and vice-versa.

- you should look at the images to see whether the source detection is working and giving you good centroids

> On going...

- you can ask the built-in source detector in Astrometry.net to reduce the threshold (producing more stars) using the "solve-field --nsigma N" option -- by default N=8

> Yeah, its works. Thank you :)

- you haven't said what kind of accuracy you think your images should be able to provide :)

> Using TGAS, we estimated a 3sigma error of 0.4 arcsec.

- un-clipped standard deviation is not really a good metric to use here -- consider that if you matched all but one of the available Gaia stars and then increased the matching radius until the last one matched, then that one large match distance would completely dominate the STD of the sample.  Instead, I would suggest plotting the results and looking at them.

> Yeah, I'm also doing that. But there are several sources with an error greater than 1 arcsec.

- Did you ever figure out the issue you reported before of the index_{ra,dec} coordinates not matching Tycho-2?  They are supposed to be very close (micro-arcsec) to the Tycho-2 catalog values.

> Nop, not yet. It's something very strange. I have cross-match the "index_ra" & "index_dec" from the *.corr file given by "Astrometry.net" with Tycho-2 catalog and as you can see, the are some significant differences. Can be the epoch? Maybe I'm doing something wrong...

Thank you for your help,

Cheers,

Cédric Pereira

[Cédric Pereira]

Regarding the last question:

In Aladin if I compare the positions from Tycho-2 catalog and *.rdls or *.corr (index_{ra,dec}) they are more similar if I change the epoch of Tycho-2 catalog to 1991/1992 :P