Farley et aL: Early marine growth In relation to marine-stage survival rates for Oncorhynchus nerka 



123 



collected during each trawl included 

 the trawl speed obtained with a global 

 positioning system and the height and 

 width of the net opening obtained with a 

 Simrad FS900 (Simrad, Lynnwood, WA) 

 net sounder. The mean date of collection 

 of juvenile sockeye salmon sampled for 

 length differed slightly between years 

 (i.e., 26 August during 2000; 5 Septem- 

 ber during 2001; 1 September during 

 2002); lengths were adjusted to account 

 for these differences. 



Survival and early marine-stage growth 

 rates inferred from adult scales 



For each freshwater age group, we calcu- 

 lated an index of survival rate that nor- 

 malized the data and removed possible 

 density-dependent effects (i.e., Peterman 

 et al., 1998; Mueter et al., 2002). Specifi- 

 cally, our index of survival rate was the 

 time series of residuals from a Ricker 

 model defined by 



In 



"■i.a.2.t+2 "'"■"i,a.3,(+3 



(1) 



= a. 



■P.A..- 



where t = the first ocean year for sock- 

 eye salmon; 



S = the total number of spawners 

 within river system / (i rep- 

 resents Egegik or Kvichak); 



R = the total return (catch-i- 

 spawners) for each fresh- 

 water age group a (a repre- 

 sents freshwater age 1 or 2 ) 

 within river system /; 

 a and [i = model parameters represent- 

 ing the number of recruits 

 per spawner at low numbers of spawners and 

 the level of density dependence (Quinn and 

 Deriso, 1999); and 

 £,^ , = the normally distributed residuals of the 

 model. 



For our analysis, partitioning salmon brood-year produc- 

 tivity by freshwater age group was necessary to directly 

 compare our index of survival with our time series of 

 MSWl,^, growth. 



Analysis of covariance (ANCOVA) was used to ex- 

 amine the effect of MSWl, ^^ ^ on our indices of survival 

 (see Fig. 2, A-D for scatter plots of f, „ , and MSWl, „ , 

 and the addition of river system, age group and year 

 were used as factors in the model. The results indi- 

 cated that the year factor was highly significant (F-test, 

 P<0.001) and that MSWl, „ , was not significant (F-test, 



■60°0'0"N 



■55°0'0"N 



165°0'0"W 



160°0'0"W 



Figure 1 



Survey area of the annual August-September (2000-2002) Bering- 

 Aleutian Salmon International Survey (BASIS) within the coastal and 

 middle domains of the eastern Bering Sea. 



P=0.18). It was possible that during some years all fish 

 could have had excellent growth and attained a large 

 size, but the ANCOVA model would have attributed the 

 large size to the highly significant year factor. However, 

 when we removed the year factor from the ANCOVA 

 model. MSWl, ^, was less significant (F-test, P=0.27). 

 In addition, the residuals from these models contained 

 significant positive autocorrelation. 



Because our data contained significant autocorrelation 

 and showed a time series character, we created univari- 

 ate time series models (Wei, 1990) for both MSWl, ^ , 

 and f, Q , to determine whether autoregressive or moving 

 average components were present. The univariate mod- 

 els were developed by examining the sample autocor- 

 relation and partial autocorrelation functions for each 

 time series. Time series data were considered white 

 noise processes, i.e., uncorrelated random variables 



