ALLEN ET AL: EFFECT OF DISTURBANCE ON HARBOR SEALS 



type II, where no seals left the area. A zero-seal 

 disturbance was any activity occurring in the area 

 when seals were not hauled out but which may 

 have prevented seals from doing so. 



To investigate the effect of actual disturbance on 

 the seals at KI, disturbances were classified by 

 four criteria: 1) seal response (yes, at least one 

 seal left; no, no seals left); 2) distance between the 

 seals and the disturbance source ( =£100 m, 101-200 

 m, 201-300 m); 3) day of week (weekend/holiday, 

 weekday); and 4) disturbance type (person/dog, 

 nonpower boat, power boat). Due to the low 

 number of disturbances during the breeding and 

 winter seasons, season was not used as a variable. 

 Only the 156 instances where disturbance type 

 and distances were known were used in the analy- 

 sis. The result was a2x3x2x3 contingency 

 table. We used log-linear models to examine the 

 effects of 2, 3, and 4 (explanatory variables) on 1 

 (response variable). Log-linear models are used to 

 analyze multidimensional contingency tables and 

 can be used to study two- and three-way interac- 

 tions between variables (Bishop et al. 1975). We 

 asked the question: "Is seal response indepen- 

 dent of the other variables?" We then asked: 

 "If not, what variables affect seal response?" 

 Exploration of models was done by backward and 

 forward selection of models (models were fit with 

 iterative proportional fitting). Both methods pro- 

 duced the same final model; only the backward 

 selection method is presented in the paper. 



Backward selection starts with a model that fits 

 the data. All interaction terms that are found to be 

 not significantly different from zero by using a 

 conditional likelihood ratio test (Fienberg 1981) 

 are removed from the model. The final model is 

 found when all nonsignificant terms are deleted. 

 Models were adjusted for marginal zeros (Bishop 

 et al. 1975: Ch. 3), and are displayed here using a 

 shorthand notation (Fienberg 1981). For example, 

 seal response independent of all other variables is 

 denoted [1] [234], and seal response dependent on 

 one of the variables, such as distance, is denoted 

 [12] [234]. Once the final model is selected, weights 

 or "u-terms" are calculated from the data, and are 

 given to each of the levels of each variable included 

 in the final model. The sign of the weight indicates 

 the effect of the explanatory variable [2, 3, or 4] on 

 the response variable [1]. The relative magnitude 

 of the weight indicates the importance of the 

 explanatory variable. 



The average time it took for seals to recover from 

 disturbance was based on the elapsed time be- 

 tween when the seals were flushed to when 50% of 



the original number had rehauled. A chi-square 

 test determined the significance of tide level on the 

 ability of seals to recover from disturbance at KI 

 (Snedecor and Cochran 1967). Correlation coeffi- 

 cients were used to detect the importance of PWI 

 as an alternate haul out site. 



RESULTS 



Camera Reliability 



A correlation of KI camera counts with 

 field counts revealed that the camera was not a 

 reliable indicator of the actual number of seals 

 present but was reliable for information on daily 

 trends. Correlation coefficients by season were as 

 follows: winter, r = 0.92, n = 19; breeding, r = 

 0.55, n = 40; summer, r = 0.75, n = 28. Discrepan- 

 cies between the two count methods were caused 

 by seals shifting along the haul out area and, 

 therefore, out of the camera's viewfinder, the incli- 

 nation of females with pups to haul out on the 

 fringe of the herd, and the difficulty in identifying 

 pups. On the other hand, the camera was very 

 reliable at PWI for the summer (r = 0.94, n = 25) 

 and breeding (r = 0.97, n = 13) seasons because 

 seals could not shift out of viewfinder range. Also, 

 this camera was slightly elevated, allowing for 

 better detection of seals hauled out close together. 

 During much of the winter season the PWI camera 

 was broken. Both cameras readily detected boats 

 and people on foot disturbing seals, but could not 

 detect aircraft. Dogs were seen on film twice and 

 on 13 occasions during field counts. 



Seasonal and Spatial Use Patterns 



Seals used the lagoon on 95-100% of the days 

 each month. More seals hauled out on KI during 

 the breeding and summer seasons than during 

 winter. Numbers on field counts averaged 31.2 ± 

 28.1 seals during the breeding season (range 0-101, 

 n = 78), 53.5 ± 28.5 during the summer (range 

 0-105, n = 48), and 19.6 ± 19.3 during the winter 

 (range 0-58, n — 28). The same trend was apparent 

 from field counts on PWI; the number of seals 

 averaged 10.6 ± 15.1 for the breeding season 

 (range 0-77, n = 107), 10.0 ± 18.4 for summer 

 (range 0-48, n = 41), and 7.7 ± 12.5 for winter 

 (range 0-55, n = 43). The PWI means were much 

 lower than those for KI, indicating that KI was the 

 preferred haul out site. 



During the breeding season, mother-pup pairs 

 hauled out on PWI and on exposed sand bars along 



495 



