Simulating telemetry studies that estimate component mortality rates of imperiled juvenile salmonids

Objective Juvenile salmonids have high mortality during their out-migration into marine environments, and there is disagreement regarding the factors that contribute to low juvenile survival, the relative strength of those factors, and how to develop effective management actions to improve survival. This study provides a planning tool for researchers interested in quantifying component mortality of juvenile salmonids using a multistate mark–recapture model framework for telemetry projects. We discuss approaches to increase study precision and better understand data needs prior to project implementation. Methods We evaluated the accuracy and precision of multistate mark–recapture models by generating simulated data and model code under six model scenarios of increasing complexity and comparing model estimates, mean relative error, and the CV among these scenarios. Model scenario A.1 considers a passive-only acoustic array (low detection probability) and three observation states (alive, predation, and unknown mortality), with predation identified using predation detection acoustic tags. Model scenario A.2 expands model scenario A.1 to include active tracking (high detection probability). Model scenario A.3 expands model scenario A.2 by increasing predation and unknown mortality estimates. Model scenario B expands model scenario A.2 to model four observation states (alive, fish predation, avian predation, and unknown mortality), where avian predation and fish predation are partitioned using predation detection acoustic tags and combined acoustic–radio tags. Model scenario C.1 expands model scenario A.1 by modeling predation and unknown mortality using biological (fork length) and environmental (temperature) covariates. Model scenario C.2 expands model scenario C.1 via an increased detection probability. Results Our results suggest that increasing tagging effort may be more effective than increasing detection probability to improve accuracy and precision, in that lower detection probabilities may produce accurate median estimates but have high variability, especially when including model covariates. Additionally, models with low detection probabilities and low sample sizes (n ≤ 100) did not accurately estimate reach-specific unobserved mortality rates, especially when the simulated mortality estimates were increased. In this study, we did not see a plateau in mean relative error or CV estimates as tagging effort increased up to 1,000 individuals; however, between 150 and 500 individuals, most CVs dropped below 10 and the variability surrounding estimates was substantially reduced. Increased simulation processing time above 150 individuals and the cost of acoustic tags for large tagging effort (n ≥ 150) should be considered when budgeting time for project planning and implementation. Conclusions This study provides realistic model scenarios and code that can be used when designing research projects aimed at identifying component mortality for salmonids or other imperiled juvenile fishes to increase the likelihood of project success while considering the costs of facilitating such research.