Influence of satellite geometry and differential correction on GPS location accuracy

The location accuracy of Global Positioning System (GPS) units depends on l ocation type (3-dimensional, 3-D>2-dimensional, 2-D) and satellite geometry (indexed by horizontal dilution of precision, HDOP). To determine the best computation method for positions collected by CPS telemetry collars in hil ly terrain, we used a stationary GPS collar that attempted to calculate a p osition every 10 minutes for 5 days, and we evaluated the relationship betw een horizontal and vertical location accuracy and HDOP (range of HDOP, 3-D= 1.4 to 489.1, 2-D=1.4 to 4,891). The 50th and 95th percentiles of horizonta l and vertical location error were related linearly to HDOP. Location error depended mainly on accuracy of the collar altitude estimate used to comput e the position in 2-D. Most forced 2-D locations not differentially correct ed were more accurate than 3-D locations when collar altitude error was les s than or equal to 50 m. It was better to force the computation of 2-D posi tions from differentially corrected 3-D locations with HDOP>15 when the col lar altitude error was less than or equal to 10 m. We also used data collec ted on 10 free-ranging moose (Alces alces) for 12 months to examine whether moose altitude could be estimated accurately using prior 3-D locations. Ac curacy of moose altitude estimation was related inversely to the time elaps ed since the first 3-D location used to make the estimate. When animal alti tude is likely to vary greatly within small time periods, we suggest runnin g the differential correction program twice for a single time period, using an HDOP cutoff of 20-25 in the first run and 10-12 in the second. All 3-D positions computed during the first processing of the data should be kept b ur only the 2-D locations calculated in the second pass should be used beca use they were calculated using more accurate estimates of animal altitude. When applying this method to our data, only 9% of successful locations are discarded and we estimate that horizontal location error is <35 m 95% of th e time.