Both approaches must deal with optical field angle distortions, lateral color shifts, and plate scales. However, FGS measurements are far more vulnerable to temporal variations that might occur during the observing sequence. The challenge is to assemble an astrometric plate by defining a common but arbitrary coordinate system onto which the individual observations are mapped. Observers must assume that the telescope's yaw, pitch, and roll might be slightly different for each observation, causing the sky to wobble about in FGS 3's detector space. Such motions can be detected and eliminated using guide star data and check star measurements. Corrections based on guide star data are referred to as POSITION mode dejittering, and those based on check star data are called drift corrections. Here we discuss the errors associated with each procedure.
The adjustments to the astrometry centroids from this POSITION mode dejittering correction are typically about 1 mas. However, the corrections occasionally can be as large as 5 mas for one or two observation of a visit. These large corrections arise most frequently when orbital day-to-night or night-to-day transitions excite HST's vibrational modes. During such events the residuals depend upon the amplitudes of these excited modes but are estimated to be typically about 1 mas. During quiet times, the residual of this correction is about +/- 0.3 mas.
12.3.2 Drift Correction Errors
As discussed on page 11-6, astrometry targets observed multiple times per visit typically drift across the FGS by about 6 to 12 mas when two FGSs guide the telescope and by up to 70 mas with only one FGS guiding. Because astrometry observations execute sequentially, the resulting errors in the measured angular separations between objects increase as the time between the measurements lengthens. The pipeline must then remove an effect that is typically 4 and not infrequently up to 25 times the overall astrometry error budget (2.7 mas).
To remove this drift, the calibration pipeline applies a model derived from the check star data to all the observations in the visit. The residuals from this correction are difficult to quantify in the usual way because the standard deviation of the data from a fit means little if only three to five points determine the fit. On the other hand, the success of the drift correction is clearly demonstrated by comparing the residuals of two plate overlays, one where the individual visits are drift corrected, the other not. Provided adequate check star data are available to generate a reliable model, plates with the drift correction applied correlate well, typically with 2 mas rms residuals. At minimum, two check stars should be observed three times each. Residuals between those same visits without drift correction range up to 15 mas rms.