AbstractFrom 1996 to 1999, a Darroch two‐sample stratified capture–recapture experiment was conducted on adult chum salmon Oncorhynchus keta in the Yukon River, Alaska. Annual release of marked chum salmon ranged from 8,527 to 18,631 fish. At the recovery site, 15,581–45,232 fish were examined for marks annually. The percentage of tagged fish among those examined for marks ranged from 2.8% to 6.4%. Abundance estimates ranged from 189,742 (SE = 6,967) to 654,296 (SE = 21,921) fish during the four study years. Weekly estimates ranged from 6,796 to 126,268 fish. The assumption of a closed population was evaluated in light of life history and other data. One closure violation due to sample starting and ending dates was mitigated using migration times. We used statistical analysis and modeling to evaluate assumptions of (1) equal probability of capture, (2) equal probability of movement, and (3) random mixing of fish. Logistic regression suggested that the probability of recapture was not independent of length and gender in 8 of 32 weeks among years. Modeling changes in the probabilities of movement and capture indicated that differences in movement would cause little bias in population estimates relative to differences in capture. However, length‐ and gender‐stratified and unstratified population estimates showed little difference (e.g., in 1997: length‐stratified estimate = 351,138 fish, gender‐stratified estimate = 333,842 fish, unstratified estimate = 333,057 fish). Three explanations were offered for these results: violations were due to changes in the probability of movement, statistical tests were overly sensitive, or the estimator was robust to the assumption violations. Probability of recapture was related to bank orientation (i.e., nonrandom mixing) only in 1999. Annual estimates during the study documented a decline in the spawning population. Weekly estimates indicated that run timing was inconsistent among years. The estimates compared favorably with other assessments of the run. We concluded that the Darroch estimator was reasonably applied to data collected via fish wheels for this population.
Read full abstract