26 



FISH AND WILDLIFE TECHNICAL REPORT 30 



Overpass 1 

 Score - 1.631 



0.2 



,- Overpass 1 

 Score 2.000 



Overpass 2 

 Score 1.184 



\ 



Running 



^ Overpass 1 

 Score 2.000 



D. 



Overpass 2 

 Score 0.999 



10 20 



40 50 60 10 20 

 Short-term Activity Index 



30 



40 50 



Fig. 22. Comparison of short-term activity indices (scaled as proportions) for a free-ranging caribou during two satellite overpasses 

 with indices obtained from captive caribou engaged in four different activities. A. Lying. B. Feeding. C. Walking. D. Running. 

 Activity counts for overpass 1 overlapped most with lying (A, lowest score = 0.097), whereas counts for overpass 2 overlapped 

 most with running (D, score = 0.999). 



anomalies in each mercury switch, sampling error, or 

 other causes. 



Discussion. The 60-s motion detector has potential as 

 a short-term activity index when behavioral categories 

 are coarsely defined. We are skeptical that attempts to 

 discriminate more finely between categories of activity 

 (e.g., browsing versus grazing) would be productive. Fan- 

 cy et al. (1988) found high overlap in sensor counts with a 

 finer delineation of activities. Gillingham and Bunnell 

 (1985) concluded that conventional tip-switch and vari- 

 able-pulse collars were not highly accurate in discriminat- 

 ing among nine behavior categories. We have noted lim- 

 itations in the present system; for example, in not one of 

 the four species tested could we distinguish standing still 

 from bedded (inactive). We can see no resolution for this 

 problem; in a motion-sensing device, the effect of stand- 

 ing motionless cannot be expected to differ from lying. 

 Beier and McCullough (1988) reported similar conclu- 



sions using a conventional tip-switch motion detector. 

 However, they also pointed out that standing behavior is 

 generally uncommon in the wild for most ungulates; thus, 

 the confounding of inactive and active periods would 

 probably be minor. 



Discrimination accuracy for captive animals was gener- 

 ally high when using the entire series of counts from an 

 overpass. Our simulation experiments for the caribou data 

 suggested that classification error was quite small. This 

 analysis assumed that successive counts from each ac- 

 tivity category were independent; examination of our data 

 supported this assumption. The assumption that animals 

 did not change behavior during the course of an overpass 

 was also implicit in this categorization system. Although 

 we had no method to assess the behavior of the categoriza- 

 tion system when animals switched activity types during 

 an overpass, it is likely that the selected type will be one 

 that actually occurred, most often the predominant one. 



