Boldt et al.: Factors influencing cannibalism and abundance of Theragra chalcogramma on the eastern Bering Sea shelf 
295 
Study area on the eastern Bering Sea shelf (<200 m depth) showing the stations 
sampled annually in the National Marine Fisheries Service bottom trawl survey 
for groundfish species, including walleye pollock (Theragra chalcogramma ), during 
1982-2006 (excluding 1983 and 1984). Three depth domains of the shelf are delin- 
eated by depth contours (coastal <50 m, inner 50-100 m, and outer 100-180 m). 
number of fish per hectare) was estimated by size class 
by using the area swept by the net (average net width 
measured during the bottom trawl haul multiplied by the 
distance the net was towed over the seafloor). Based on 
diet information, the assumption was made that small 
pollock caught in the survey were potential prey (sum 
of small [60-200 mm SL] pollock CPUE; SmallPollock), 
and large pollock were assumed to be potential predators 
(sum of large [>200 mm SL] pollock CPUE; LgPollock). 
Predator-prey co-occurrence ( Overlap ) between Small- 
Pollock and LgPollock was determined for each station 
and designated a value of 1 (co-occurrence) if the CPUE 
values of both were greater than zero, otherwise a value 
of 0 was assigned (no co-occurrence). 
Bottom temperature, cold pool, and depth 
Water temperature and depth profiles were collected at 
each station by using bathythermographs attached to 
the headrope of the bottom trawl net (expendable models 
before 1993 and microbathythermographs in later years). 
Bottom depths and temperatures were recorded for each 
station. It was also noted whether the cold pool (tem- 
peratures <2°C) was present or absent at each station 
(binomial variable, ColdPool; see Buckley et al. [2009] 
for a description). 
Pollock abundance models 
General additive models (GAMs) were used to explore 
the relationship between pollock abundance and pre- 
dictor variables (covariates). The 3 response variables 
examined were large predatory pollock CPUE (>200 
mm SL, LgPollock ), small prey pollock CPUE (60-200 
mm SL, SmallPollock), and the co-occurrence of LgPol- 
lock and SmallPollock (Overlap). Pollock CPUE values 
(SmallPollock and LgPollock) were ln+1 transformed to 
achieve normality. Because the timing of the survey and 
overall abundance of pollock changed from year to year, 
“Year” was included as a categorical predictor. Pollock 
catch or the occurrence of cannibalism (e.g., inshore 
vs. offshore stations) may have been influenced by sta- 
tion location; therefore, station location (latitude and 
longitude) was included as a single, smoothed bivariate 
term, sOatitude, longitude), and referred to as “ Loca- 
tion ” in the GAMs. Latitude may limit the northerly 
distribution of pollock and longitude may be important 
in terms of distribution with depth. Because bottom 
temperature and bottom depth are related, a smoothed 
bottom temperature and bottom depth interaction term 
was included (TempDepth). The presence or absence of 
the cold pool was included as a binomial term (Cold- 
Pool) in the model as well. Survey duration typically 
extended over the entire summer period from June to 
August (i.e., fish likely grew or moved during the time it 
took to complete the survey and collect the data in one 
year), therefore Julian day was considered for inclusion 
in models. The groundfish survey, however, began in 
the southeast and proceeded to the northwest Bering 
Sea, and the day of year was significantly correlated 
with latitude and longitude (coefficient of determination 
[r 2 ] = 0.390 and 0.817, P<0.001 and <0.001, respectively). 
