of a GPS receiver in a moving vehicle, such as an aircraft,
renders the accuracy estimation more difficult or rather impos-
sible if a non redundant satellite constellation (i.e. 4 satel-
lites or less) is available or if an unsuitable receiver is
used. Then the suggested algorithm with ARI-processes can be
utilized for stochastic investigations.
As an example some results of an accuracy study on real GPS
data are presented in the following. The data for the analysis
originate from a photogrammetric test flight with a Sercel GPS
receiver TR5SB carried out by the Survey Departement of Rijks-
waterstaat in the Netherlands on June 10th and 12th 1987 (D.
Boswinkel, R.Witmer, J.W.v.d. Vegt 1988). The position coordi-
nates in the earth-fixed WGS system derived from the primary
phase measurements are subject of this analysis. Systematic
effects which influence the determination of the coordinates
are eliminated to a great extent in advance by simultaneous
measurements with a second GPS receiver at a stationary refe-
rence point. Both receivers recorded with a registration rate
of 1 measurement every 0.6 seconds. The data of the whole
flight were divided according to the 6 photo-strips with about
130 registrations each. Only the data within the strips are of
interest and the aircraft then shows a very uniform kind of
movement.
Each position coordinate is considered separatly by it's own
ARI-process. The mean ARI process order was found to be (5,2);
so 7 ARI process parameters fully describe the model. The
parameters of the stochastic model were estimated without any a
priori information. In table 1 the estimated standard devia-
tions of the noise-component in the GPS coordinates are listed.
The main result is that the internal accuracy of the positions
determined from a moving GPS receiver is about 0.02 m.
Table 1: Estimated noise o» [m] of GPS coordinates
Strip Date X Y Z
1 12.06.87 0.027 0.011 0.038
3 10.06.87 0.020 0.009 0.026
4 10.06.87 0.024 0.009 0.027
5 10.06.87 0.020 0.013 0.024
6 12.06.87 0.014 0.011 0.014
7 12.06.87 0.012 0.007 0.010
rms of strips 3,4,5 0.021 0.010 0.026
rms of strips 1,6,7 0.018 0.010 0.021
These results can be compared with an estimation of the inter-
nal accuracy out of the least squares adjustment of the GPS po-
sitions. The LS adjustment determines the weight coefficient
matrix Q and gives in general an estimate of the weight unit
sigma naught, i.e. the GPS range measurement accuracy. In the
case of a moving GPS receiver observing 4 satellites the sigma
naught value is not determinable because of the missing redun-
85
