Estimating and predicting collection probability of fish at dams using multistate modeling

Dams can be equipped with a bypass that routes a portion of the fish that enter the turbine intakes away
from the powerhouse into flumes, where they can be counted. Daily passage abundance can be estimated by dividing
the number of fish counted in the bypass by the sampling rate and then dividing the resulting quotient by
the collection probability (i.e., the proportion of the fish population passing the dam that is bypassed). We used
multistate mark–recapture modeling to evaluate six candidate models for predicting the collection probabilities of
radio-tagged subyearling fall Chinook salmon (n = 3,852) as a function of 1–2-d time periods (general model), four
different combinations of outflow (i.e., the total volume of water passing the dam) and turbine allocation (i.e., the
proportion of outflow directed through the turbines), and a null (intercept only) model. The best-fit model was
the additive combination of turbine allocation and outflow, which explained 71% of the null deviance. Cross validation
of the best-fit model accounted for the variation that may arise from different data sets and the ensuing
parameter values on the collection probability estimates and yielded a standard error of 0.613 that can be used to
construct approximate 95% prediction intervals in nonstudy years. Such estimates have been unavailable and will
be useful anywhere estimates of daily passage abundance at dams with bypasses are needed to manage migratory
fishes.