Yasumiishi et al.: Effect of population abundance and climate on 2 populations of Oncorhynchus keta 
211 
Table 2 
Insample generalized least squares (GLS) models and the integrated GLS and vector autoregression 
error correction models for the growth of age0.3 male chum salmon ( Oncorhynchus keta) from Fish 
Creek, Alaska. Growth indices include middle juvenile growth (SWlb), late juvenile growth (SWlc), 1 st 
immature year growth (SW2), 2 nd immature year growth (SW3), and maturing growth (SW4). Length in- 
dices include early juvenile length (La) and total juvenile length (LI). JsE-APt,Pink,t = j uven ile pink salmon 
abundances from southeast Alaska to the Alaska Peninsula. IsEAP,t=i m mature chum salmon and matur- 
ing pink salmon from southeast Alaska to the Alaska Peninsula. Igc-SE. (immature chum salmon and 
maturing pink salmon from British Columbia to southeast Alaska. M AS chum t =maturing chum salmon 
from Asia. WF=fall wind speed. SST=sea surface temperature. PDO=the Pacific Decadal Oscillation in- 
dex. Model statistics included coefficient estimates (coeffi), P-values of the coefficients (P-value), the 
coefficient of determination (r 2 ), Pvalue of the Pstatistic, coefficient of variation (CV), and Schwarz 
information criterion (SIC). 
Variables Model statistics 
Response 
Predictors 
Coeff. 
P-value 
r 2 
P-value 
CV 
SIC 
SWlb 
La t .i 
0.618 
<0.001 
0.70 
<0.001 
0.039 
165 
JsEAP.Pink.t 
-0.538 
<0.001 
SWlc 
WF t 
-0.571 
<0.001 
0.39 
<0.001 
0.111 
205 
SW2 
Llt-i 
-0.582 
<0.001 
0.50 
<0.001 
0.049 
218 
IsEAP.t 
0.741 
<0.001 
SW2 
Lin 
-0.582 
<0.001 
0.77 
<0.001 
0.031 
193 
IsEAP.t 
-0.741 
<0.001 
SST t _ 2 
0.438 
<0.001 
PDO t . 4 
-0.321 
0.001 
SW3 
lBC-SE,t 
-0.605 
0.001 
0.35 
0.001 
0.066 
208 
SW4 
Mas, C hum, t 
0.602 
0.0016 
0.34 
0.002 
0.052 
178 
SW4 
■^AS, t, Chum, t 
0.556 
0.0016 
0.46 
<0.0001 
0.047 
175 
SST t _] 0.388 
PDO t . 3 -0.302 
related with SST, but the SIC was lower than for the 
model with abundance. 
SW4 was most strongly and inversely correlated 
with the abundance of maturing chum salmon from 
Asia (r 2 =0.34; P=0.002). SW4 was not significantly re- 
lated to other abundance or with SST and PDO during 
the year of growth. The SST at a 1 year lag and the 
PDO at a 3 year lag explained additional variation in 
SW4 and were both significant in the VAR model (r 
2 =0.46; P<0.001). 
Quilcene River chum salmon models 
Quilcene River chum salmon growth was inversely re- 
lated to chum and pink salmon abundance during im- 
mature and maturing life stages, but not during the 
juvenile stage (Table 3). All sets of residuals passed 
the tests for normality, homoscedasticity, and serial 
correlation. 
SW2 was inversely correlated with the abundance of 
immature age-0.1 chum salmon from Asia. Abundance 
explained 35% of the variability in growth in the GLS 
model (r 2 =0.35; PcO.OOl) (Table 3). 
SW3 was inversely correlated with the abundance 
of maturing age-0.1 pink salmon and immature age- 
0.2 chum salmon from BC to southern AP (r 2 =0.52; 
P<0.001) (Table 3). The GLS/VAR model explained an 
additional 21% in growth (r 2 =0.73; P<0.001) and re- 
duced the SIC. Similar to the 1 st immature year growth 
models for Fish Creek, growth was positively correlated 
with SST and negatively correlated with the PDO at a 
2-year lag. 
SW4 was negatively correlated with the abundance 
of maturing chum salmon from Asia (r 2 =0.52; PcO.OOl) 
(Table 3). Maturing growth was not correlated with the 
climate indices in this study. 
Model validation 
For the reserved observations, the growth models did 
not perform well in the model validations (Tables 4 and 
5). Positive r 2 values occurred for SWlb and SW4 Fish 
Creek and SW4 Quilcene River models. Negative r 2 val- 
ues indicated that the mean of the growth time series 
performed better in model validation, when model pre- 
dictions were tested against 20% of the data that were 
not used for model specification, than did the fitted val- 
ues for the juvenile and immature models. Significant 
model deterioration occurred for the SW2 GLS and 
GLS/VAR and SW3 GLS models for Fish Creek and for 
the SW2 GLS and SW3 GLS/VAR models for Quilcene 
River as indicated by the P-statistic for model dete- 
