278 
Fishery Bulletin 113(3) 
Table 3 
Backward stepwise term selection used in identification of the best-fitting generalized additive models (GAM) for predicting 
distribution and abundance of Pacific ocean perch ( Sebastes alutus ) on the basis of data from bottom trawl surveys conducted 
in the Aleutian Islands during 1997-2010, including the percent contribution of each predictor variable to the deviance 
explained by the best-fitting model and Akaike’s information criterion score (AIC). An asterisk (*) indicates the best-fitting 
GAM formulation. See Materials and methods section for initial model formulations and Table 1 for definitions of variable 
abbreviations. CPUE=catch per unit of effort. 
Initial 
Contribution by 
Deviance 
Response 
model 
Independent 
each variable 
explained 
Life stage 
variable 
formulation 
variables retained 
(%) 
(%) 
AIC 
Juvenile 
Presence- 
1 
F+G+Gp+M+U+Co+Br+ 
absence factor 
2 
s(Log.n)+s(Sl)+s(D,T)+s( C '%) 
F+G+Gp+M+U+Co+Br 
24.9 
1971.77 
+s(Long.)+s(Sl)+s(D, T)+s(V t ) 
2.7;3.1;0.9;0.6;3.0;1.0; 
1.1; 12.3;2.1;40.9;1.7 
25.0 
1970.10* 
3 
Sp+Co+s(Long.)+s(Sl)+s(D,T)+s(C x ) 
20.7 
2070.22 
4 
Sp+Co+s(Long.)+s(Sl)+s(D,T)+s(V t / 
20.9 
2068.65 
Juvenile 
Conditional 
1 and 2 
C+0+V+s(Log.n)+s(D,T) 
3.3;5.5;4.3;41.7;26.8 
24.9 
2358.05* 
CPUE 
3 and 4 
s(Long.)+s(D,T) 
22.0 
2372.99 
Adult 
Presence- 
1 
F+G+U+Br+s(Long.)+s(Sl)+s(D,T)+ 
absence factor 
2 
s(Ng) 
F+G+U+B7'+s(Long.)+s(Sl)+ 
s(D,T)+s(V r ) 
1.4;0.4;0.8;0.9;6.6;1.8; 
38.6 
2054.45 
68.8;0.9 
38.7 
2050.85* 
3 
Sp+Co+Br+s(Long.)+s(Sl)+s(D,T) 
37.7 
2079.74 
4 
Sp+Co+Br+s(Long.)+s(Sl)+s(D,T)+ 
s(V T ) 
38.0 
2073.70 
Adult 
Conditional 
1 and 2 
G+Co+s(Long.)+s(Sl)+s(D,T) 
1.4;0.7;14.2;1.4;84.0 
42.5 
5692.83* 
CPUE 
3 and 4 
Br+s(Lon)+s(Sl)+s(D,T) 
42.0 
5702.49 
32 m. Many of the deeper stations are near passes be- 
tween islands in the Aleutian archipelago, but, in gen- 
eral, deeper stations occur more frequently in the east- 
ern part of the survey area (around Samalga, Amukta, 
and Seguam passes) than in the west. The values for 
SI from the kriged bathymetry at trawl stations ranged 
from 0° to 15°. Tidal currents predicted at each bottom 
trawl haul from the tidal prediction model ranged from 
<1 cm/s to ~ 300 cm/s (around 3 m/s or 6 kn). The high- 
est current velocities were predicted around Seguam 
and Amukta passes, and some additional areas of high 
current were predicted on the east side of Amchitka 
Pass. Bottom temperatures (T) measured in situ during 
trawl hauls ranged from 3°C to 7°C. 
Results of generalized additive modeling 
Prediction of presence and absence The best-fitting 
GAMs for predicting the probability of presence of ju- 
venile and adult Pacific ocean perch in the Aleutian 
Islands accounted for a quarter of the deviance in the 
juvenile model and 38.7% of the deviance in the adult 
model (Table 3). The depth distribution of adults, cen- 
tered around 225 m, was deeper than that of juveniles, 
found at depths around 150 m. Model effects showed 
very little dependence on temperature (Figs. 3 and 
4). Model responses (GAM effects represented by the 
solid lines on graphs) indicate increased probability of 
encountering Pacific ocean perch life stages when >0 
and decreased probability when <0. The standard error 
generally increases around the predictions for which 
sample size decreases, signifying areas of lower confi- 
dence in the model. Geographically, the predicted odds 
of collecting either life stage increased at Unimak Pass 
(165°W), in the passes near the Islands of Four Moun- 
tains (170°W), in Amukta and Buldir passes (173°W 
and 177°E), and in Near Strait (175°E; Figs. 5 and 6). 
Effects due to local slope were similar for both juve- 
niles and adults and the probabilities of encountering 
either life stage increased over moderate slopes up to 
around 5° of incline. 
Increasing tidal velocities initially led to increased 
probability of encountering adults, but the probability 
of encounter for juveniles steadily decreased with in- 
creasing velocities. The presence of biogenic structures 
accounted for more of the deviance explained in the 
juvenile presence-absence model than with the adult 
model (11.3% versus 2.6%; Table 3). The biogenic struc- 
