262 
Fishery Bulletin 116(3-4) 
Abundance of juvenile Chinook salmon (Oncorhynchus tshawytscha) emigrating 
downstream during 2014 by ordinal week, estimated by using 3 competing models 
with data collected at a rotary screw trap for Big Creek, Idaho, in 2014. The 3 mod¬ 
els are the pooled-simple (M PS ), hierarchical within-year (M H wk and hierarchical 
multiyear (M HB ) models. A fourth model, the hierarchical penalized-spline M SPL ine 
model, was unable to run and is not displayed. Gray regions denote estimates pro¬ 
duced for temporal strata with missing data in 2014. 
past record during the spring (Fig. 3). Spring peaks in 
catch varied in both timing and magnitude. Capture 
probabilities were variable but usually <0.10. Most gaps 
in operations start on the descending limb of the catch. 
These features influenced the 2014 estimates for weeks 
17-19 such that the variable peak catches from the past 
record cause broad 95% credible intervals (Fig. 2). There 
were less past data after those strata but catches were 
not as variable; hence 95% credible intervals for strata 
20-24 are not as wide. Note, however, that there were 
some data in every stratum in the past record and the 
M hb model was able to use that information to esti¬ 
mate abundance with a precision measure during strata 
17-24 in 2014. Because the past record is thin during 
these periods, the M HB model estimate incorporates that 
uncertainty in the form of wide credible intervals. 
Discussion 
To our knowledge, this is the first time a hierarchi¬ 
cal Bayesian model has been structured to incorporate 
data on annually recurring species behavior by using 
multiple years of data to improve abundance estimates 
from sparse and missing mark-recapture data. In the 
simulation, the hierarchical multiyear model produced 
comparable estimates to those of the best model when 
data were complete but also produced the most accurate 
and precise estimates when large periods of data were 
missing or reduced. Because the hierarchical multiyear 
model can incorporate information about recurring spe¬ 
cies behavior to improve the accuracy of abundance es¬ 
timation, it can be applied beyond salmonids. Studies 
focused on species that express recurring behavior, par¬ 
ticularly for species with low abundances, could benefit 
from similar hierarchical multiyear models to increase 
the accuracy and precision of abundance estimates ob¬ 
tained from sparse and missing mark-recapture data. 
Current Lincoln-Petersen estimator approaches 
used for juvenile salmonids abundance estimation do 
not provide an effective method for addressing sparse 
or missing data. When strata have low numbers of re¬ 
captured individuals (less than 7), confidence intervals 
obtained from bootstrap approaches are largely unin¬ 
formative (Steinhorst et al., 2004). Manual pooling, 
or the use of software that attempts to optimize the 
balance between satisfying model assumptions while 
maintaining fine-resolution run characteristics (e.g., 
Darroch analysis with rank reduction; Bjorkstedt 4 ), of- 
4 Bjorkstedt, E. P. 2000. DARR (Darroch analysis with 
rank-reduction): a method for analysis of stratified mark- 
