Stewart: Defining migration rates of Parophrys vetulus 



477 



25 _, 



20 - 



15 



10 



0.0 



Nortfi 



November (model 2) 



January (model 3) 



October, May, or June (models 1,4,5) 



0.5 



Proportion of population moving 



Figure 2 



Prior (horizontal line) and posterior distributions for movement parameters 

 (proportion of the population moving) in the single-season models (1-5). Solid 

 lines indicate movement to the north and south in November (model 2), dashed 

 lines show movement only in January (model 3), and dotted lines show move- 

 ment in October, May. or June (nodels 1, 4, 5: not separated in the legend). 



north to be 0.10 (0.04-0.18) and to be 0.09 (0.04-0. 

 17) to the south. In the fall, parameter medians were 

 0.05 (0.02-0.09) to the north and 0.27 (0.17-0.38) to 

 the south. This model was strongly supported over the 

 best single-season model (TBF=49). The next best two- 

 season model (TBF=16) allowed movement in October 

 and May. Movement rates estimated from two-season 

 models showed a similar pattern to those estimated 

 from the single time-step models. Net movement to the 

 south was identified in both cases, primarily in the fall, 

 and although movement rates from two-season models 

 were somewhat reduced, they were applied twice per 

 year. The cumulative expected value for movement in 

 the best two-season model was 0.11 to the north and 

 0.30 to the south, very close to the values from the best 

 single-season model. 



When movement was allowed in each month of the 

 year, parameter estimates were much smaller per month 

 (implying a similar magnitude of annual movement), 

 but greater movement was still predicted to the south 

 than to the north (Fig. 4). The best monthly model (no. 

 11) included separate proportions moving north (0.02 

 [0.02-0.03]), and south (0.04 [0.03-0.04]). This model 

 received strong support over the best two-season model 



(TBF=101), but only slightly more support from the 

 data (TBF=3) than constant and equal movement all 

 year (modellO). The cumulative expected value of this 

 movement was 0.21 to the north and 0.29 to the south, 

 similar to the best two-season model, but with more net 

 movement to the north. With only a single movement 

 parameter, the median proportion moving in model 10 

 was 0.03 (0.02-0.04). Although more complex models 

 including movement in only some months were explored, 

 none were supported by the data (TBFs 16-34) over 

 models 10 or 11 (Table 5). 



Model parameters other than movement rates showed 

 no obvious restriction by their priors, although posterior 

 distributions included much of the marginal param- 

 eter space within the prior bounds. For all models, the 

 overdispersion parameters ik^.) for each data type had 

 substantial density below 1.0, indicating variability 

 far in excess of that expected from a Poisson distribu- 

 tion (Fig. 4). Reporting rate was generally less than 

 0.75 and was highly correlated with the median fishing 

 mortality rates. Tag loss rates were predicted to be high 

 (often greater than 0.5), but quite uncertain. Sensitiv- 

 ity analysis to the shape of key prior distributions did 

 not result in any substantial changes for the posterior 



