Abstract
Redd abundance for steelhead Oncorhynchus mykiss is frequently estimated using design-based redd surveys. Redd abundance estimates may be biased low because redds may not be distinguishable prior to or in between sampling events (i.e., removal bias). Removal bias is typically mitigated by increasing temporal sampling intensity, thereby potentially reducing the level of spatial replication. In this study a model-assisted estimation technique to correct for removal bias was described and evaluated. Parametric survival time models were fit to redd data to estimate redd longevity and parameterize a simulation to evaluate the model-assisted estimation approach. The simulation showed increasing negative bias in redd counts not adjusted for removal bias as redd longevity decreased and as sampling intervals increased. Average bias among simulated datasets was as high as 43% using raw counts when redd longevity was relatively short at a 25-d sampling interval. Average bias was reduced to as low 1.3% using the model-assisted approach. Relative benefits of the model-assisted approach compared with traditional estimates decreased as redd longevity increased, and the accuracy of model-assisted estimates were sensitive to the function used to estimate redd longevity. The results of this study suggest that commonly used sampling intervals can potentially be increased with no appreciable increase in bias when using model-assisted estimation techniques.
Received October 7, 2015; accepted March 21, 2016 Published online July 20, 2016
Acknowledgments
We acknowledge the many people who helped conduct actual surveys. Several people also contributed to earlier ideas for a model-assisted approach, including the following: Christopher Horn, Marika Dobos, Shelley Banks, and Kasey Bliesner. We thank Matthew Falcy, Bryan Stevens, Jeff Falke, Robert Al-Chokhachy, and four anonymous reviewers for providing helpful comments on an earlier version of this manuscript.