GPS gives precise and reasonably accurate fixes of position. However, as with all complex systems its limitations need to be understood so that it can be used appropriately.
In radio system planning getting a reasonable accurate three dimensional coordinate for the location of an antenna or the base of a mast can be useful for predicting signal propagation to help assess coverage or potential interference between systems.
Nowadays, with GPS selective availability turned off, we can typically expect accuracies of the order 20 metres or an order of magnitude better with secondary corrections from Wide Area Augmentation System (WAAS) or other ancillary systems.
The accuracy is a function of many things, including number of satellites in view, their geometry relative the receiver and radio propagation effects along the paths. The error is generally expressed with dilution of precision ratios in space and time, which when all combined result in positions being expressed as confidence within a circular area in the horizontal plane or within a sphere in three dimensions.
The first simple observation is that we have different metrics for error in the horizontal and combined horizontal/vertical planes. Conditions that lead to low error in the horizontal plane may not coincide with the vertical. For example, if all the satellites in view are on the horizon and none overhead then we may get a low horizontal error, but a high vertical error. Typically the vertical error is several times worse than the horizontal.
The size of error in each plane may have disproportionate effects when applied to propagation prediction. This is particularly true when considering diffraction geometry to terrain or clutter obstacles. In the horizontal plane, a 20 metre position error is often insignificant when we are looking at systems with separation distances of the order hundreds of metres or kilometres, especially when using general purpose gridded terrain models to model terrain obstacles surrounding the location of the fix. However, the same kind of error in the vertical plane can lead to gross changes in diffraction geometry, especially when considering antenna heights that are low in comparison to the terrain or local clutter.
The vertical height reported by default in most GPS receivers is the height above the WGS84 ellipse. This ellipse is a hypothetical construct that is an oblate sphere which is a coarse approximation to the Earth’s surface. In fact the Earth has a more complex shape than an oblate sphere, and a better datum for measuring height the is that of a geoid, which is defined as the line where gravity pulls equally.
Many mistake GPS heights for height above mean sea level, which can lead to gross errors. To give an example of the height difference between the GPS ellipse and a geoid model, taking EGM96, which is a global geoid model, then using the handy UNAVCO geoid height calculator page, the geoid height at say Newlyn harbour is 53.5 metres above WGS84. If the raw GPS height is used without correction the vertical height error will be at least 53.5 m along with the other sources mentioned. Coupled with a terrain elevation database for terrestrial propagation prediction and the results will be grossly errored, as most terrain data will be referenced a local datum which will be closer to the geoid. In the case of the Great Britain, the heights are most often quoted against the average tidal height datum measured in Newlyn from 1912-1921. Note that there are usually better local geoid models than the global EGM96, such as the OSGM02 which is generally applicable in Great Britain.
Are GPS heights too good to be true? Well not if used appropriately!