Are removal-based abundance models robust to fish behavior?

• Published in: Fisheries research 2017-12, Vol.196, p.160-169
• Main Authors: van Poorten, Brett T., Barrett, Boyd, Walters, Carl J., Ahrens, Robert N.M.
• Format: Article
• Language: English
• Subjects: Abundance; Bias; Depletion estimate; Fish behavior; Mark recapture; Removal estimate
• ISSN: 0165-7836; 1872-6763
• DOI: 10.1016/j.fishres.2017.06.010

[Display omitted] •Removal models are known to be biased; worse if assumptions are not met.•Marking fish may help reduce bias, but only if mark-recapture assumptions are met.•We compare removal model estimates when fish are known to react to fishing.•Density dependent removal model is most robust to uncertainty in behavior.•Marking fish improves performance when assumptions are met; bias is worse if not. Removal methods are some of the most common statistical tools for estimating fish abundance in streams and lakes, yet they are prone to produce biased estimates when the assumption of constant capture probability is violated. In response, numerous authors have modified the classic removal models to control for non-constant capture probability. A variety of fish behaviors can cause capture probability to vary across individuals or over time, such as dominance hierarchies, escaping capture or persistent individual differences in capture probability due to activity or aggression; yet knowing exactly which behaviors may affect capture probability is generally unknown. We assessed the robustness of five removal models (i.e., the Leslie model, three behavior-dependent models and a density dependent capture probability model) and their ability to provide consistently accurate and precise abundance estimates irrespective of the exhibited behavior. We fitted each model to catch data generated from five behavioral models that mimicked a range of animal behaviors in a closed population. Additionally, we evaluated the improvements that can be gained by including marked fish in the removal process and in that case, compared estimation models with a Peterson mark-recapture estimation. Results indicate that no single removal model is robust to non-constant capture probability, however, the density-dependent capture probability model performed moderately better than other models when only removal data were used. We found that the addition of marked fish results in a substantial improvement in accuracy and precision across all removal models when mark-recapture assumptions are met. However, these improvements diminished substantially when mark-recapture assumptions were violated. Due to the difficulties in assessing assumptions, our findings suggest that including marked fish in the removal process may unknowingly reduce accuracy and precision of initial abundance estimate and that this type of experimental design should be avoided in many instances.