Estimation of Survival Probability Through Downstream Dam Passage Using Simulation Models From Experimental Data

ABSTRACT The migration of juvenile anadromous salmonids in the Interior Columbia River Basin requires passing various hydroelectric dams, which pose risks of injury or death. Mark‐recapture studies are used to estimate direct and indirect survival of juvenile salmonids during dam passage. In these studies, data interpretation can be challenging when ideal study parameters are constrained by small sample sizes and inadequate replication as a result of unpredictable events. Even ideal field studies are complicated by fish that cannot be recaptured and whose survival is uncertain. To address cases when field trials yield imperfect or unclear data, or established methods cannot be applied, we developed novel simulation models that incorporated field trials in order to decrease variation around recapture survival estimates for a passage study through McNary Dam on the Columbia River (Washington/Oregon, USA). There were two types of models: one was a randomization model akin to bootstrapping the recapture data to decrease the coefficient of variation; the other type was a Bayesian model that simulates different scenarios to estimate release survival, by incorporating fish with unknown fates. Simulations showed the expected similarity in survival estimates and considerably reduced the variation in those estimates. The Bayesian model showed further reduction in variation around survival compared with the randomization model. It also indicated that concerns around small sample size could be conceptually mitigated by the modeling approach. The model also incorporated individuals that were not recaptured and demonstrated that the modeled worst‐case scenario, where all unrecovered individuals are assumed to be dead, shows higher survival than the mean release survival from the field trials and is comparable to, but slightly lower than, the field recapture survival. All of these findings suggest that the simulation model provides an effective tool widely applicable to field studies of survival when the observed data are imperfect.