TRACKING WILDLIFE BY SATELLITE 



15 



Fig. 8. Relation between location error and 

 maximum satellite elevation during an over- 

 pass. Numbers in parentheses are sample 

 sizes. 



c 

 o 



CD 



13 



E 

 o 



it- 

 CD 

 O 



C 

 CO 



- 

 CO 



b 



600 



g 550 

 o 



500 



450 



400 



350 



(56) 



(104) 



(117) 



(150) 



10 20 30 40 50 60 70 80 

 Maximum Satellite Elevation () 



90 



(Table 9). Location estimates calculated for two PTT's 

 deployed on captive caribou consistently had greater error 

 than did location estimates calculated simultaneously 

 (same overpass) for two similar PTT's on an adjacent 

 fencepost (Fig. 11). (The remaining PTT's one each on a 

 caribou and a fencepost contributed too few locations to 

 be of use.) Movement of the caribou within the pen could 

 have contributed, at most, about 100 m of this error. Loca- 

 tions from the same overpass were estimated from signifi- 

 cantly fewer messages for the caribou PTT's than for the 

 PTT's on the fencepost (paired-f = 2.58, df = 1 8, P < 0.02). 

 Apparently, signal attenuation (VSWR effect) and subtle 

 changes in temperature or orientation of the PTT attached 

 to the caribou contributed to the failure of the satellite to 

 receive some messages, which resulted in decreased preci- 

 sion of location estimates. 



In a crossover experiment, however, we found indica- 

 tions that the reduction in precision associated with de- 

 ployment on an animal may be complicated by a 3-way 

 interaction between overpass, attachment status (on ver- 

 sus off the animal), and the specific PTT. A third-genera- 

 tion PTT that had been tested on a caribou later produced 

 location estimates with standard deviations 33 and 58% 

 lower when transmitting from the fencepost during identi- 

 cal daytime hours (Wilcoxon LMest of ranked straight-line 

 errors, z = 3.237, P< 0.001; Figs. 12Aand 12B). However, 

 when a second-generation PTT was also tested on both the 

 caribou and the fencepost, errors were not different (Wilc- 

 oxon t/-test, z = 1.383, P > 0.15; Figs. 12C and 12D). 

 These two PTT's did not differ from each other in vari- 

 ability of location estimates when both were on the 

 fencepost (Wilcoxon U- test, z = 1 . 133, P > 0.25), but they 



Fig. 9. Improvement in mean location error 

 for a FIT at Nome (solid line) as overpasses 

 (grouped by maximum elevation in 10 

 blocks) are progressively excluded from the 

 total sample. Dashed line shows the propor- 

 tion of overpasses remaining as each block is 

 excluded. Numbers above dashed line are the 

 maximum elevations of excluded blocks. 



LLJ 



C 

 03 

 0) 



500 1 



450 



400 



350 



300 



Locations Excluded (10 blocks) 



