Laman et al.: Correlating environmental and biogenic factors with abundance and distribution of Sebastes alutus in Alaska 
279 
Longitude 
Slope 
o 
CD 
it 
CD 
< 
0 
Figure 3 
From the best-fitting generalized additive model (GAM), predictions of presence of juvenile Pacific ocean 
perch ( Sebastes alutus ) at trawl stations in the Aleutian Islands during 1997-2010 in relation to (A) longi- 
tude, (B) kriged bottom slope, (C) predicted tidal current velocity, and (D) the interaction between depth and 
temperature. A change in number orientation indicates that either a maximum or a minimum was reached 
for the GAM effect. 
tures retained in the models were either erect forms 
(e.g., F or Co) or have been reported to be epizoic on 
erect sponges (e.g., G and Gp sponges; Stone et ah, 
2011). Although the presence of certain sponge morpho- 
groups and Co was associated with increasing probabil- 
ity of encountering juveniles and adults, the presence 
of Br was associated with decreasing probability of en- 
countering either life stage of Pacific ocean perch. This 
decrease may result from the shallower depth distribu- 
tion of Br (depths <110 m) compared with the depth 
distribution of Pacific ocean perch juveniles (depths 
-150 m) and adults (depths -225 m). More than half of 
the deviance explained in each presence-absence GAM 
was attributable to the D,T and Long, terms. 
Validation of presence-absence generalized additive mod- 
eling The success of the GAMs to predict Pacific ocean 
perch distribution (presence) varied with life stage 
(Figs. 5 and 6). Based on the scale proposed by Lan- 
dis and Koch (1977) for assessing model performance, 
the GAM predictions of presence of juvenile Pacific 
ocean perch from 2012 survey data displayed moder- 
ate agreement with trawl observations from the same 
year (£=0.52). GAM predictions of adult presence were 
more accurate, showing substantial agreement when 
predicted presence or absence of adults was compared 
with observed adult distribution for 2012 (£=0.70). 
Prediction of conditional abundance More deviance in 
the model was explained by the best-fitting condition- 
al abundance GAM for adults than by the best-fitting 
GAM for juveniles (Table 3). As with the presence-ab- 
sence GAMs, the D,T and Long, terms were the most 
influential predictors of conditional abundance. Unlike 
the results from the presence-absence GAMs, there 
were temperature-related optima in both the juvenile 
