COS 314-5 - Using replicated removal sampling to estimate animal abundance in geographically open populations

Abstract: Removal models are important tools for estimating abundance and have seen widespread application to avian, terrestrial, and aquatic taxa. Closure between sampling occasions is a critical assumption of removal models because changes in abundance due to births, deaths, immigration, or emigration causes bias in abundance estimates. Removal models have been extended to open populations by assuming closure between shortly spaced secondary sampling occasions, but not between longer time intervals separating primary sampling occasions. To date, all open removal models require some form of closure between sampling occasions, yet the closure assumption may be difficult or impossible to fulfill in many sampling situations. For example, our analysis was motived by a long-term monitoring program where juvenile Chinook salmon (Oncorhynchus tshawytscha) are sampled with hand-held seines swept through near-shore estuarine habitats. In this sampling situation, closure between successive removal samples (approximately 5 minutes between samples) is violated because individuals are able to move into or out of the sampling area between samples.

To accommodate removal sampling designs applied to geographically open populations, we extended existing open removal models by incorporating a Markovian availability process into the spatially replicated design of N-mixture models. Under this model, the number of animals available for sampling depends on the probability of recruiting new individuals to the sampling area and the probability of remaining in the sampling area given an individual was previously available for sampling. Thus, our model estimates an additional parameter representing recruitment (r) to the superpopulation defined by the total number of unique individuals available for sampling across all sampling occasions.

To test the model, we 1) conducted formal parameter identifiability analysis, 2) applied the model to data simulated by performing removal samples on realizations of a two-dimensional random walk movement model, and 3) applied the model to seining data comprised of geographically open samples and samples within enclosures where closure was enforced. We found that 1) all parameters were identifiable, 2) abundance estimates were unbiased when applied to data generated by realistic animal movements, and 3) the model estimated r > 0 for open seining samples and r = 0 for closed samples, indicating that our model is able to distinguish between open and closed populations. Our removal modeling framework should prove useful for a wide variety of taxa and environments by allowing researchers to formally test closure assumptions, to estimate the degree of closure, and most importantly, to estimate abundance without bias.