Recent developments in VLBI instrumentation have been aimed at achieving
millimeter accurate station coordinates with one-day measurement durations.
The Mark III VLBI system consists of electronics to amplify and frequency
shift the very weak signals from extragalatic radio sources so that they can
be recorded on wide bandwidth tape recorders. Associated with the Mark III
VLBI system is a correlator system which can play back the recorded tapes,
cross correlate the signals, and determine accurately the differences in
arrival times (group delays), in phase (phase delays), and in Doppler shift
(phase delay rates) of the signals recorded on the tapes. The developments to
this system have been to improve (1) group-delay data accuracy; (2) observing
schedules; (3) atmospheric delay modeling; and (4) combination of data. The
primary enhancements to Mark III instrumentation to achieve higher-accuracy
group-delay measurements are widening of the synthesized bandwidth
[ Corey and Clark, 1991]; increasing the recorded bandwidth of
the signal, which also includes the Mark IV VLBI developments
[ Whitney et al., 1991]; and improving the delay calibrations
through the VLBI receiving and recording systems [ Rogers,
1991]. Overviews of these improvements are reported by Rogers
[1991] and Rogers et al. [1993]. The implementation of the
these system improvements have made group-delay measurements with
uncertainties of 
5 mm routine for modern VLBI systems.
The schedules for making VLBI observations (i.e., the choice of which quasars to observe, with which antenna, and for what duration for each observation) have also been improved over the last few years. Davis et al. [1991] reported on improved baseline length measurements through the use of observations made at low-elevation angles (as low as 5 degrees) and the use of low elevation angle observations is now routine in VLBI experiments. Another important enhancement in the scheduling has been making the azimuth coverage at each station as uniform as possible [ Niell, 1991].
With the increased use of low elevation angle data in VLBI experiments there has been recent improvements to atmospheric delay models. The models have sought to improve the low elevation angle performance of the functions that relate the atmospheric zenith delay to the line of sight delay; the so called mapping functions. Herring [1992 a] reports on a new series of mapping functions based on the analysis of radiosonde data that takes advantage of the continued fraction in sine of elevation angle to parameterize the mapping function. (The continued fraction has in its denominator sine of elevation angle plus a small parameterized number which has as its denominator sine of elevation angle plus another a small parameterized number or so on. Three levels of continued fraction is sufficient to parameterize the mapping function.) These mapping functions use the surface temperature projected from the temperature profile below the tropopause as a proxy for the seasonal variation in atmospheric parameters. These new mapping functions have root mean square (RMS) errors of typically 30 mm at 5 degrees elevation angle for the hydrostatic delay and 10 mm RMS at 5 degrees elevation for the wet delay. These errors are about half the size of the errors in previous generation models. There has also been increased study of the effects of atmospheric gradients with results being reported in [ Herring, 1992 a; Davis, 1992; Rogers et al., 1993; and Davis et al., 1993]. Some of these studies have used water vapor radiometers to study the variations in water vapor delay as the radiometer is scanned in azimuth at a fixed elevation angle. Others have used estimates of the gradients from the analysis of the VLBI data (i.e., gradient parameters are estimated along with other geodetic parameters). In both cases, the magnitudes of gradients are similar and typically result in a departure from azimuthal symmetry of 10 mm at 5 degrees elevation angle. In some cases the departure from azimuthal symmetry at 5 degrees elevation angle can reach 100 mm. Uncorrected, a persistent gradient of 10 mm at 5 degrees elevation angle could result in a horizontal errors of the same magnitude.
Several studies have also been carried out to assess the utility of water
vapor radiometers for calibrating the most variable and least predictable
component of the atmospheric delay. The two largest studies
[ Elgered et al., 1991; and Kuehn et al., 1991]
used water vapor radiometer data collected over intervals between 3 to 10
years and compared the RMS repeatability of baseline lengths obtained from the
best stochastic solutions not using the WVR data with the best analyses using
the WVR data. In both studies it was concluded that for WVR data to be
useful, accurate surface pressure measurements are needed to calibrate the
hydrostatic delay and no low elevation angle data (
20 degrees) should be
included in the analysis. In the analyses using WVR data, no atmospheric
parameters are estimated. The most recently collected WVR data improves the
baseline length RMS over the best stochastic solutions but the improvement is
less than 10%. In both studies, there were anomalous biases of up to 10 mm
amplitudes between the atmospheric delay estimates inferred from the WVR data
and those estimated from the VLBI data. These studies could not resolve the
influence of mapping function errors on the stochastic estimates and the
effects of biases in the WVR measurements and on the conversion of WVR sky
temperature measurements to delays. An additional study was carried comparing
VLBI and Global Positioning System (GPS) estimates of atmospheric delays
[ Tralli, et al., 1992]. Here the differences between the VLBI
and GPS estimates were typically 
10 mm and comparable to or smaller than the
differences seen between VLBI and WVR's. In general, all of these studies
indicate residual atmospheric delay modeling errors of 10 mm and less, but
errors of this magnitude can induce errors in height estimates by up 30 mm.
There have been no recent extensive comparisons of WVR's and VLBI primarily
because of lack of WVR data.
The performance of VLBI measurements as reported in Rogers et al. [1993] for circa 1992 measurements was 5 mm + 2 parts-per-billion (ppb) implying height measurements precisions of about 25 mm. Recent VLBI measurements from January 1994 indicate that the highest quality VLBI experiment have achieved precisions of 1 mm + 0.7 ppb implying height precisions of about 9 mm [ Clark et al., 1994]. All of the improvements discussed above seem to have played equal roles in this improvement. Results of this quality should be capable of addressing numerous questions about post glacial rebound and for defining a global reference systems for measuring global sea level rise.