High-precision astrometry and image reconstruction with Chandra
Following are notes on improving the absolute astrometry
and image reconstruction of X-ray sources
in Chandra observations. This is based on documents written by Eric
Feigelson and the ACIS team, and we gratefully acknowledge their contribution.
Overview
-
Read the CXC memo on aspect caveats
to see if your dataset may be affected by offsets in absolute
astrometry that are correctable.
- For improved image reconstruction (i.e. X-ray image sharpness) read the CXC
thread regarding removal of ACIS pixel
randomization, which can slightly improve imaging resolution.
-
The Chandra mission requirement on absolute source
position (i.e. celestial location) accuracy is an RMS radial deviation
of less than 1.0 arcsec. Currently, the RMS radial deviation is less than 0.5
arcsec, and fewer than 1% of observations have an offset of more
than 1 arcsec (
Chandra absolute astrometric accuracy).
-
The aspect flight performance (as of 2003) is described
in a paper published in Experimental
Astronomy, v. 16, Issue 1, p. 1-68 (2003). The key result for astrometry is
that the overall 90% uncertainty circle of
Chandra X-ray absolute position has a radius of 0.6 arcsec, and the 99% limit on
positional accuracy is 0.8 arcsec.
-
Other aspects of the design and performance of the Chandra aspect
flight hardware are described in this SPIE
paper.
Improving absolute astrometry
Improved celestial location precision is possible
for some observations by cross-correlating detected X-ray sources (from
celldetect or other detection algorithms) with high-precision optical,
IR, or radio catalogs. This can be used to remove large offsets due
to processing problems, or to fine-tune the astrometry to well below the
typical 0.6 arcsec performance.
ACIS team members have used this technique to achieve absolute astrometry
accurate to +/-0.3" (90% confidence, Sgr A* field), +/-0.15" (Hubble Deep
Field), and +/-0.1" (Orion Nebula cluster).
Following is a suggested procedure for bringing
an image into the Hipparcos coordinate frame. Best results are expected
from high-S/N sources in the inner portion of the field where the PSF is
narrow.
-
Check to see whether any sources are associated with
in the Hipparcos/Tycho
catalog with ~1 mas precision for the ~100,000 Hipparcos stars (V<9)
and ~40 mas precision for the ~1M Tycho stars (V<11), and especially
the new Tycho-2
catalogue with ~2.5M stars.
-
Note 1: Do not use the HST Guide Star Catalog Version 1 with ~20M stars
due to its poor precision and pre-Hipparcos reference frame.
-
Note 2: It may be necessary to look at all columns of these databases;
e.g., column 40 of the Tycho catalogues gives the astrometric accuracy
of each star. See description of Tycho columns here
. When printing out Hipparcos/Tycho tables from Netscape Navigator,
you must "Save as" in "text" format; otherwise most columns are lost.
-
Note 3: The Hipparcos/Tycho-1 catalog is in J1991.25 coordinates, so proper
motion correction is potentially important. Tycho-2 is given in J2000,
so proper motion is less of an issue.
-
Check to see whether any sources have counterparts on the all-sky Schmidt
photographic plates. For declinations >-20 deg, search the
USNO-A2.0
catalogue of 526M objects from the Palomar Sky Survey, which is based on
the Hipparcos frame and has positional precisions around 0.3". For
more southerly declinations, try the ROE/NRL
COSMOS catalog of ~500M objects from the ESO/UK Southern Sky survey
plates, but note the frame is pre-Hipparcos.
-
Check the new 2-micron all-sky catalogues. 2MASS has >162M objects
over half the sky available at IPAC.
The astrometric
accuracy of these objects a standard deviation of +/- 0.1" with
respect to Tycho stars.
-
Though unlikely, check to see whether any source has a counterpart in the
FIRST
radio survey covering 15% of the sky. Currently at 549M,
these sources have precisions
ranging from 0.05-1 arcsec on the VLBI reference frame (which is precise
and nearly identical to the Hipparcos frame).
-
Check to see whether any sources are listed in NED
or SIMBAD
databases, but beware that their stated positions come from a variety of
sources with various precisions and reference frames.
-
If several astrometric counterparts are available, one can quickly find
the mean offset in (RA,Dec), or more elaborately one can make a least-squares
solution that includes a roll angle correction. The mean offsets
(without a roll correction) can easily be inserted into the FITS header
events or image files of interest, using the FTOOL fv:
-
View the Header of the binary EVENTS table, and find the keywords TCTYPdd
(where "dd" is a two-digit number) which have the values "RA--TAN" and
"DEC--TAN".
-
Use the Edit Insert/Delete menu options to correct the corresponding keywords
TCRVLdd. These keywords give the RA and Dec (degrees) of the central
pixel in the physical (detector) image.
-
For images (as opposed to event files), examine and edit the CTYPEdd and
CRVALdd keywords, respectively.
-
Note: Off-axis positions may have additional systematic positional errors
of unknown amplitude due to the asymmetric off-axis point spread function.
Statistical uncertainty of source locations
Individual source locations are subject to statistical uncertainties affecting
the centroiding algorithm and to the dispersion of photons due to the PSF.
This has not been studied thoroughly, but the ACIS team has done a detailed
astrometric analysis of 27 ACIS sources with 2MASS/VLA counterparts in the
Orion Nebula Cluster (Garmire et al. 2000, AJ submitted, Table 2). From this
they estimate 90% confidences of +/-0.5" for sources with ~10 counts, +/-0.2"
for 20-50 count sources, and negligible for >100 count sources.
Improved image reconstruction
Application of the Lucy-Richardson maximum-likelihood algorithm to
structures on the ACIS-S3 chip gives a reconstructed image with
resolution around 0.3"-0.35" FWHM. This was achieved on the PKS
0637-752 jet (Chartas et al. 2000, ApJL in press) and the SN1987a
supernova remnant (Burrows et al. 2000, Science submitted).
|
