At times we have used GPS for survey work when we have had to do "non critical" coverage over large areas of very "rugged" land, and to ensure that the GPS transmitted, variable error could be eliminated as much as possible we would set up a radio beacon (line of sight accuracy only) on a prominent high point of known XYZ coordinates. By taking a continuous stationary series of downloaded GPS readings at the beacon (with the recorded time of each reading) then we could take both readings from the beacon, when in line of sight and readings from the GPS, and when out of line of sight, the GPS readings only (and at a minimum 5 GPS readings at each point) With the time of each reading known then it is relatively simply to reduce the error factor to an acceptable level in the final result calculations.
Even with this system the accuracy could only be calculated to cm accuracy, suitable for something like a gravity survey but a totally unacceptable accuracy for cadastral surveying for areas like CBD where a 1 mm error can mean thousands of dollars difference in “rental” for a multi storied building, or even out of boundary residential work.
With the fluctuating (anything but “constant” ) errors recorded using this sort of an accurate “checking” system, our results showed that for use as an accurate measurement for speed the GPS is “indicative” only having a low degree of accuracy.