Mathews et al.: Effects of vessel disturbance on Phoca vitulina on glacial ice in Tracy Arm, Alaska 
189 
Seals were categorized as nonpups or pups and as 
hauled out or in the water. We identified seals as pups 
by several criteria, including their small size, close 
association and positioning with an adult (presumed 
to be the mother), behavior (e.g., suckling, nuzzling), 
overall body shape, and pelage. In late June as pups 
approach or cease weaning, they typically are larger 
and more likely to be left unattended (Boness et al., 
1994). Therefore, we distinguished pups and yearlings 
in late June by pelage differences if they were close 
enough for pelage to be seen clearly. It is likely that 
we underestimated pup abundance in late June be- 
cause of difficulty in seeing pelage on unattended pups 
and yearlings far from our observation point, and that 
difficulty may have slightly reduced peak pup counts, 
slightly increased peak nonpup counts, and possibly 
caused a slight underestimate in the date of the peak 
pup count. In contrast, the estimated date of onset of 
pupping would not have been affected by these factors. 
We recorded percent ice cover (visually estimated 
in 10% increments), weather variables, the number 
of vessels present, and incidences of disturbance (i.e., 
seals entering the water in response to vessel activ- 
ity) during each count. Weather variables included sky 
condition (i.e., clear, partly cloudy, overcast), tempera- 
ture, precipitation (i.e., none, mist or light rain, heavy 
rain), wind speed (Beaufort scale), and wind direction 
(i.e., up, down, or across the fjord). When there were 
2 observers, we used the average of the estimates of 
ice cover. For 13 counts with ice cover recorded in 25% 
increments, we used the mid-point of each category for 
our estimate. 
During each count, we also recorded the number of 
icebergs with evidence of fresh blood or the presence of 
bald eagles (Haliaeetus leucocephalus ), which we used 
as a proxy for the number of recent seal births. Bald 
eagles are attracted to fjords during seal pupping be- 
cause they eat the seal afterbirth and stillborn pups 
(Calambokidis and Steiger, 1985). At the end of a count, 
each observer rated the quality of his or her count on 
a scale of 1-7, with 1 being excellent and 7 being poor. 
Counts with a quality rating of 7 were excluded from 
analyses. 
We used generalized linear models (GLMs; Poisson 
error, log link) to estimate the relationships between 
seal counts (i.e., nonpups, pups, and all seals) and the 
following explanatory variables: count quality, day of 
year (DOY), time of day (TOD), percent ice cover, sky 
conditions, temperature, precipitation, wind speed and 
direction, and the presence of vessels in the count area. 
We included vessels as a variable in 1 of 2 forms: 1) 
presence or absence or 2) the number of vessels in the 
inlet; only 1 of these vessel-related variables was used 
in a single model. For this and all subsequent analyses, 
DOY was centered on the median DOY of all counts, 
and TOD was centered on solar noon. Centering pre- 
dictors eliminates the correlation between linear and 
quadratic terms, facilitating model fitting (Draper and 
Smith, 1981). Quadratic terms for DOY and TOD were 
also included in initial models. 
To account for over-dispersion, we included a scale 
parameter estimated as the Pearson chi-square value 
divided by the degrees of freedom. All variables were 
included in the initial model and deleted one at a time 
until all remaining variables had Wald chi-square- 
based P-values of approximately <0.05. Mean counts 
adjusted for the other variables in the model (i.e., 
least-squares means; Littell et al., 2006) were comput- 
ed for the levels of categorical variables that remained 
in the final models. 
Vessel traffic 
We recorded all vessels that entered the inlet from 
the start of observations in the morning to the end of 
observations for that day. We recorded the time that 
each vessel entered (came into view) and departed (dis- 
appeared from view) from the inlet. Vessels that ap- 
proached the glacier face were likely to have had great- 
er effects on seals than vessels that did not; therefore, 
in 2001, we divided the study area into 2 sections (Fig. 
1, inset map, A and B) by drawing a line between our 
observation point and a large waterfall across the in- 
let, and we recorded if and when a vessel entered and 
left section A. Because the South Sawyer Glacier was 
receding, the A and B section categorization was not 
used after 2001. 
We categorized vessels into 6 types: 1) cruise ships 
(large, oceangoing vessels of 91 metric tons gross or 
more that carry passengers for hire; 2) tour boats (com- 
mercial vessels less than 91 metric tons gross that op- 
erate on a daily or weekly schedule); 3) power boats 
(chartered vessels and private vessels, including sail- 
boats under power and nonskiff auxiliary vessels from 
cruise ships [these subcategories were pooled because 
it was not always possible to distinguish among them]); 
4) inflatables (inflatable skiffs with an outboard motor), 
5) skiffs (hard-hulled skiffs with an outboard motor), 
or 6) kayaks. For groups of inflatables, skiffs, or kay- 
aks traveling together, the lead vessel was tracked if 
the group was monitored for seal disturbance, but each 
boat in the group was counted separately for vessel 
summaries. Because observations were concentrated 
during the day and vessels may have entered the inlet 
before or after observations, vessel counts were mini- 
mums; counts of tour boats likely were the most accu- 
rate because their times of daily entry and departure 
usually were within our observation periods. 
Seal-vessel interactions 
Randomized observations We conducted randomized 
observations of focal groups (Altmann, 1974) of seals to 
determine the rate at which seals entered the water — a 
rate that we modeled as a function of predictor vari- 
ables, including the presence of vessels. We conducted 
2-14 observations (median: 6 observations) of 10 min 
each per monitoring session. To spread sampling pro- 
portionately, the study site was divided into 10 zones. 
We used a computer-generated list to randomly select 1 
