Survival and Selection of Migrating Salmon From Capture–recapture Models With Individual Traits

Abstract

Abstract. Capture-recapture studies are powerful tools for studying animal population dynamics, providing information on population abundance, survival rates, population growth rates, and selection for phenotypic traits. In these studies, the probability of observing a tagged individual reflects both the probability of the individual surviving to the time of recapture and the probability of recapturing an animal, given that it is alive. If both of these probabilities are related to the same phenotypic trait, it can be difficult to distinguish effects on survival probabilities from effects on recapture probabilities. However, when animals are individually tagged and have multiple opportunities for recapture, we can properly partition observed trait-related variability into survival and recapture components. We present an overview of capture-recapture models that incorporate individual variability and develop methods to incorporate results from these models into estimates of population survival and selection for phenotypic traits. We conducted a series of simulations to understand the performance of these estimators and to assess the consequences of ignoring individual variability when it exists. In addition, we analyzed a large data set of Ͼ153 000 juvenile chinook salmon (Oncorhynchus tshawytscha) and steelhead (O. mykiss) of known length that were PIT-tagged during their seaward migration. Both our simulations and the case study indicated that the ability to precisely estimate selection for phenotypic traits was greatly compromised when differential recapture probabilities were ignored. Estimates of population survival, however, were far more robust. In the chinook salmon and steelhead study, we consistently found that smaller fish had a greater probability of recapture. We also uncovered length-related survival relationships in over half of the release group/river segment combinations that we observed, but we found both positive and negative relationships between length and survival probability. These results have important implications for the management of salmonid populations.

Survival of seaward-migrating PIT and acoustic-tagged juvenile Chinook salmon in the Snake and Columbia Rivers: an evaluation of length-specific tagging effects

“…Consistent with previous findings [23,29,[33][34][35], length was positively correlated with survival for nearly all year and reach combinations examined for both PIT and AT fish. One field study failed to detect a relationship between length and survival among acoustic-tagged Chinook salmon smolts emigrating through the Fraser River Basin [36].…”

“…Rather, SURPH.3 applies the overall detection probability of all fish in the cohort to each survival estimate. Indeed, length has been shown to be negatively correlated with the detection probability of PIT-tagged Chinook salmon at Snake and Columbia River dams [35]. However, to confirm results were biased by variable detection probabilities, smaller and larger fish were analyzed separately without a covariate and survival rates were almost entirely dependent on detection rates.…”

Survival of Juvenile Chinook Salmon Passing the Bonneville Dam Spillway in 2007

“…The summer length limitation for tagging restricted our statistical inference to the larger subyearlings in the JMF samples. Length-related detection biases associated with acoustic telemetry have not yet been documented like those for PIT-tag detection systems (Zabel et al 2005), although this bias should not be large, if it exists at all, because detection probabilities like those observed for releases down to the primary array below Bonneville Dam were very high (mean probability = 92.7%). The detection probability to the secondary array averaged 87.5%.…”

Effects of radio-tag characteristics and sample size on estimates of apparent survival

“…Point estimates for groups and/or encounters obtained from small samples may, therefore, deviate substantially from the statistical population. Thus, the second main conclusion is that, where only small samples of individuals (below c. 200 individuals) can be included, inference upon a specific model and on underlying biological mechanisms, should be drawn very carefully, and requires support by additional evidence [7,9]. Even if the difference in QAICc to lower ranked models was large (>2.0), alternative models were difficult to discern due to the uncertainty in estimates for both parameters.…”