Brunner and Welsch [1993] and Yunck [1993] present overviews of atmospheric effects on geodetic measurements and means of dealing with them. Tropospheric water vapor is the principal offender, and the water vapor radiometer (WVR) has been the traditional tool for calibrating its effect. Elgered et al [1992] present a detailed overview of WVRs; the evidence suggests that a well-tuned dual frequency WVR can recover the zenith delay to 8 mm or better. A now standard alternative is to estimate the zenith delay from the GPS data as part of the geodetic solution, which has the virtue of absorbing residual errors in the dry delay model as well. Possibly the most effective approach is to model the zenith delay as a stochastic process and estimate a constrained adjustment every few minutes. Tralli et al [1992] show that stochastic delay estimates made with GPS and very long baseline interferometry agree with one another, and with WVR measurements, at the 8--15 mm level. Mendes and Langley [1994] compare 14 zenith delay mapping functions (central to delay estimation) and find notable discrepancies below 20 elevation. Further study is needed to identify preferred mapping functions for different sites and seasons.
Several studies hint that the power of stochastic zenith delay estimation all but nullifies the value of non-pointing WVR data [e.g., Dixon et al, 1991]. Ware et al [1993] describe a study in which data from WVRs pointed towards the GPS satellites improve the vertical precision of a 50-km baseline measurement by 40% (to 2.6 mm) over that from stochastic estimation alone. The high cost of WVRs, however, may limit their use. The advent of dense regional arrays promises a further advance in atmospheric calibration. Preliminary results from the Scripps Institution of Oceanography show that data from a regional network can help identify the effects of local atmospheric changes; those trends can then be exploited in a ``spatial filtering'' to improve the stability of position estimates [ S. Wdowinski, unpublished results]. Tranquilla and Alrizzo [1993] show that snowstorms can cause significant height variations in baseline solutions. In another twist, VanDam et al [1994] report that variable atmospheric loading can cause as much as 5 mm of variation in actual vertical site positions and offer confirming evidence from the global network.
The flip side of error removal is error source observation, and the earth's atmosphere is a particularly rich source to observe. A nascent industry is now seeking to apply in practical ways the enveloping atmospheric grasp of GPS. Objectives range from regional mapping of tropospheric water vapor for weather prediction to global recovery of stratospheric temperatures for studying long term climate change. Bevis et al [1992, 1994] discuss the conversion of GPS-inferred zenith wet delays from ground data into estimates of precipitable water vapor (PWV) and suggest that GPS may yield PWV accuracies of 1% or better. Rocken et al [1993] show agreement between GPS estimates and WVR measurements to better than 1 mm of PWV (or roughly 7 mm of delay). Ground based GPS can give only the total value of the overlying PWV; knowledge of the vertical distribution of water vapor, however, is critical to many science applications. Space based techniques can provide that knowledge by a different approach. A receiver in orbit can trace the changing phase of GPS signals passing through the atmosphere to obtain precise vertical profiles of atmospheric density, pressure and either temperature (in the upper troposphere and stratosphere) or water vapor (lower troposphere). GPS occultation is described by Hajj et al [1994a] and Melbourne et al [1994]. Yuan et al [1993] consider how ground and space based techniques may complement one another in the long term study of climate change.