298 
Fishery Bulletin 1 10(3) 
Table 1 
Final general additive models from a backward selection process and the associated adjusted r 2 values, deviance explained (%), 
and sample size ( n ). b=intercept, s=nonparametric smoothing function, E=error term. Response variables are: LgPollock- ln+1 
transformed predator walleye pollock (Theragra chalcogramma ) (>200 mm standard length [ SL] ) catch per unit of effort (CPUE) 
in the NMFS bottom trawl survey; Small Pollock =ln+l transformed prey pollock (60-200 mm SL) CPUE in the bottom trawl 
survey; PA = binomial presence or absence of cannibalism; %FO LgPollock , %BW LgPoUock , and PF LgPollock = \n+l transformed percent 
frequency of occurrence, percent body weight, and partial fullness of prey pollock in predator pollock stomachs, weighted by 
LgPollock , respectively. Covariates were the following: Location , a smoothed station location covariate, s( latitude, longitude); 
TempDepth, a smoothed bottom temperature and depth covariate, slbottemp, botdepth); ColdPool, a binomial covariate indicat- 
ing presence or absence of the cold pool; PredatorLength , a predator length categorical covariate; Ocerfap=binomial covariate 
indicating the spatial overlap between predator and prey pollock; s (SmallPollock), a smoothed SmallPollock covariate. 
Model 
type 
Final model formulation 
r 2 
(adjusted) 
Deviance 
explained (%) 
n 
CPUE 
LgPollock = b+s(Location )+s(TempDepth )+Year+ColdPool+E 
0.50 
51.0 
2754 
CPUE 
Small Pollock = b+s(Location)+s(TempDepth)+Year+E 
0.21 
22.2 
2754 
CPUE 
Overlap =b+s(Location)+s(TempDepth)+E 
0.19 
17.1 
2754 
Diet 
PA = b+s(Location)+s(TempDepth)+Year+PredatorLength + Overlap+E 
0.13 
21.1 
7479 
Diet 
%FO LaPollock = b+s(Location )+s(TempDepth )+s(SmallPollock)+E 
0.52 
55.1 
610 
Diet 
%BW LaPo[locli =b+s(Location)+s{TempDepth )+s(SmallPollock)+PredatorLength+E 
0.63 
65.5 
610 
Diet 
PF Lgp 0 u 0C k ~ b +s (Location )+s(TempDepth)+s(SmallPollock)+PredatorLength+E 
0.55 
58.3 
610 
SL were generally consumed by predators measuring 
>60 cm FL and prey pollock <60 mm SL were consumed 
in small proportions by all sizes of predator pollock 
(Fig. 5). Additional pollock were consumed but could 
not be included in the analyses because they were too 
digested to obtain length measurements (Fig. 5). The 
occurrence of cannibalism in the EBS varied spatially 
and temporally (Figs. 4 and 6). The proportion of sta- 
tions where cannibalism occurred ranged from 0.02 (in 
2003 and 2004) to 0.18 (in 1990) and decreased during 
the years examined (Figs. 4 and 6). Fewer samples were 
collected in 1982, 1985, and 1986 (Fig. 6). 
Pollock diet models 
The presence or absence of pollock cannibalism on age-1 
pollock (60-200 mm SL) was significantly affected by 
most covariates examined, TempDepth , Location , Year, 
PredatorLength, and Overlap (excluding ColdPool), and 
the model explained 21.1% of the deviance (Table 1). The 
occurrence of cannibalism was highest in the northwest 
EBS and covered portions of all depth domains, but 
primarily in the middle and outer domains (Fig. 7). The 
presence of cannibalism generally occurred most often 
at bottom temperatures between 0°C and 5°C (Fig. 7). 
Where cannibalism occurred, the amount of can- 
nibalism (%FO LgPollock , %BW LgPollock , and PF LgPollock ) 
on age-1 pollock (60-200 mm SL) was significantly 
affected by TempDepth, Location, SmallPollock, and 
PredatorLength (except PredatorLength in the %FO L 
^Pollock GAM). Covariates that were not significant in 
the diet GAMs were ColdPool (except in the %FO lgPol 
lock GAM) and Year. Significant covariates explained 
55.1%, 65.5%, and 58.3% of the deviance in models for 
%FO LgPoiiock • %BW LgPollock’ and FF LgPollock’ respectively 
(Table 1). Generally, the amount of cannibalism was 
highest in the northwest EBS, in the outer half of the 
middle depth domain and in the outer domain (Fig. 7). 
The amount of cannibalism increased with increasing 
temperatures between 1°C and 5°C and with increas- 
ing bottom depths (Fig. 7). In addition, the amount of 
cannibalism increased with increasing abundance of 
small pollock ( SmallPollock ) and increasing predator 
size ( PredatorLength ; Fig. 7). 
Discussion 
The GAM approach, used in this study, enabled us to test 
hypotheses regarding pollock distribution and cannibal- 
ism. Results from this study showed that the distribu- 
tions of large cannibalistic pollock and vulnerable age-1 
prey pollock were affected by environmental factors and 
varied among years. Large pollock were generally found 
in deeper waters of the outer and middle domains and 
they avoided the cold pool; whereas, age-1 pollock were 
more broadly distributed and were generally found in 
slightly shallower (but overlapping) depths in the north- 
west middle and outer domains and in cooler tempera- 
tures. These results are consistent with what has been 
found in previous studies regarding pollock distribution 
in the EBS (Swartzman et al., 1994; Kotwicki et ah, 
2005; Mueter et al., 2011). The area where cannibalism 
on age-1 pollock could potentially have occurred (area 
of overlap between predator and prey pollock) was in 
the middle and outer domains of the northwest EBS 
at depths less than 100 m and at bottom temperatures 
between 0°C and 5°C. 
Although diet samples represent snapshots in time, 
the spatial coverage over the 23 years examined pro- 
