The astrometric coordinate system of the raw ACS images is specified in the WCS keywords in the image headers. Its accuracy is ultimately limited by the accuracy of the catalogue used for pointing the telescope. For earlier data, the Guide Star Catalog I (Lasker et al. 1990) was used, and since 2000 the Guide Star Catalog II (McLean et al. 2000) is used. Both catalogues are based on scans of the photographic Sky Survey plates at various epochs and bandpasses. The absolute accuracy reaches 0.3'' over a large fraction of the sky, but errors can be as high as several arcseconds towards the edges of the scanned plates. Additionally, misidentification of sources or confusion effects can, in some cases, create inaccuracies in the raw ACS astrometry as large as 10".
In order to improve the astrometric accuracy of the ACS direct images, and thus of the sources associated to released spectra, we cross-correlated source catalogues extracted from the ACS fields with several reference astrometric catalogs: UCAC2, 2MASS, SDSS-DR6, USNO-B1, GSC2.3.2, and the GOODS WFI R-band catalog. When multiple matches were found, the most accurate of these catalogue was used. The fitting procedure allows for simple shifts of the astrometry, but not for rotation which is generally negligible in HST observations. We used several techniques to avoid false matches, including sigma-clipping and cluster analysis; we also introduced magnitude cuts for each astrometric catalog (so that, for example, a 2MASS star would not match an R ~ 22 mag object).
The internal-evaluated accuracy of the whole technique strongly depends on the number and quality of the matches, but for the large majority of the fields is below 0.6" (median error = 0.3"). This accuracy is confirmed by a check in the GOODS fields where accurate (50 mas) astrometry is available.
Independent methods were used to check the accuracy of the wavelength calibration and its stability across the associations. Extracted spectra of late type stars were compared with each other and with a library of spectral templates. In addition, ~ 50 spectra extracted in the CDF-South field were compared with spectra at resolution 200-600 available from public spectroscopy surveys. These tests yielded a wavelength accuracy within 1 pixel.
The flux calibration for the ACS/G800L grism was derived from observations of the HST primary standard star G191B2B and cross-checked with observations of the primary standard star GD153 (Kuntschner, Kuemmel & Walsh, 2008). For point sources the absolute flux calibration of the 1st order is accurate to better than 2% for wavelengths from 6000 to 9500 A. For extended objects the flux-calibration curve is adjusted to take the size of the target into account (Gaussian approximation), however, the flux calibration accuracy is reduced to about 10%.