44 



FISH AND WILDLIFE TECHNICAL REPORT 30 



Our work with satellite telemetry data required frequent 

 development of programs designed to assist data analysis 

 (Fancy et al. 1988). We developed systems that were used 

 automatically to achieve specific objectives other pro- 

 jects using large quantities of satellite data have developed 

 similar systems. Merrick and Mate (1985) developed a 

 series of programs for dealing with satellite data for ceta- 

 ceans. Also the Wildlife- Wildlands Institute in Missoula, 

 Montana, has developed a series of programs to reformat 

 Argos data and produce files that can be manipulated on a 

 microcomputer by dB ASE III+. 



The system we developed had three components: a data 

 summary stage, in which Argos data were summarized 

 into files with all information from an overpass in a single 

 record; a differentiation stage, in which smaller files were 

 created consisting only of information from overpasses 

 fixing a location; and a formatting stage, in which these 

 smaller files were converted to GIS -compatible formats 

 for presentation either as location points or vectors be- 

 tween successive locations. In this last stage, summary 

 statistics were also computed. We also adapted a NASA 

 program for predicting the times and characteristics of 

 satellite orbits. Predictions were used to determine opti- 

 mum duty cycles for transmitters and to synchronize direct 

 observations of an animal with satellite overpasses to eval- 

 uate activity sensors or location accuracy. The program 

 calculated satellite azimuth, elevation, and range at all 

 times during each overpass. Description of an earlier ver- 

 sion of these programs is provided by Fancy et al. (1988). 



Sampling Concerns 



Selecting Locations 



As with conventional telemetry, error is always present 

 in location estimates obtained from satellite telemetry. A 

 clear example was when the two satellites passed over an 

 animal within 10-15 min of each other: animals some- 

 times appeared to make spectacularly quick "move- 

 ments" from one location to another. Estimates of ani- 

 mals' rates of movement would have become inflated if 

 these apparent movements (many of which were attributa- 

 ble to telemetry error) were included in analyses. We 

 review here two suggested algorithms for choosing among 

 competing locations in such situations. Both create an 

 objective set of rules to govern selection of data for analy- 

 sis, although neither solves the problem of error. 



The algorithm we used allowed us to specify a time 

 window during which only one location was to be selected 

 for inclusion with the resulting data. This window was 

 varied, depending on the objectives of the analysis and the 

 PTT's duty cycle. The algorithm identified the cluster of 



locations falling within the specified window and with 

 specified criteria chose the best location offered it then 

 found the next cluster of locations, beginning with the first 

 observation not in the previous cluster. Beginning in April 

 1987, criteria for choosing caribou locations were the 

 location with the highest LQ index (3 > 2 > 1 ) and, in case 

 of a tie, the location calculated from the greatest number of 

 messages. Other criteria that might be used include choos- 

 ing locations estimated by the best satellite overpass with 

 elevation closest to the optimum (see Fig. 8) or those 

 estimated when the PTT displayed minimum temperature 

 variation. 



Choice of a time window substantially altered the re- 

 sulting display of animal movements. Figure 41 portrays 

 the movements of an adult female caribou from the Porcu- 

 pine herd as she traveled from her wintering area toward 

 her eventual calving site. The general movement pattern 

 remained unchanged, regardless of which location fre- 

 quency was used, but short-term movements were pro- 

 gressively less evident as shorter time windows were used. 

 Fancy et al. (1989) used a 1-h window to assess movement 

 patterns of Alaskan caribou. 



Even after selecting among locations within a window, 

 the locations occasionally seemed to be biologically un- 

 reasonable. We incorporated algorithms that flagged a 

 location whenever the animal's calculated rate of move- 

 ment exceeded a specified tolerance, which was unique to 

 each species. When successive locations were closely 

 spaced in time, we found this method helpful. 



Independence of Successive Locations 



Independence of successive observations is critical for 

 some statistical analyses of animal movements, but inde- 

 pendence can be violated when observations are closely 

 clustered in time, as often occurs with satellite telemetry 

 data. Schoener (1981) devised a procedure to assess the 

 independence assumption. Swihart and Slade (1985a) de- 

 rived a test of significance for deviations from the ex- 

 pected value of Schoener 's ratio and, by doing so, devel- 

 oped a method to determine whether a given data set meets 

 the independence assumption. They also suggested that a 

 data set failing to achieve independence could still be used 

 by systematically excluding observations (thereby in- 

 creasing the elapsed time between successive observa- 

 tions) until the resulting series satisfied the independence 

 criterion. 



We examined some monthly series of satellite-obtained 

 locations from different species, calculating Schoener 's 

 ratio and Swihart and Slade's critical value each time. 

 Most data sets we examined failed the test of indepen- 

 dence, although considerable variation among species and 

 seasons was noted. For example, the movements of a 



