Use of an enclosed elk population to assess two non-invasive methods for estimating population size

Abstract

Non-invasive genetic sampling and spatially explicit capture-recapture (SCR) models are used increasingly to estimate abundance of wildlife populations, but have not been adequately tested on gregarious animals such as elk (Cervus canadensis), for which correlated space use and movements violate model assumptions of independence. To evaluate the robustness and accuracy of SCR, and to assess the utility of an alternative non-invasive method for estimating density of gregarious ungulates, we utilized a tule elk (Cervus canadensis nannodes) population of known size within a fenced enclosure on the San Luis National Wildlife Refuge in central California. We evaluated fecal genetic SCR to camera trap-based random encounter model (REM) approaches to density estimation based on comparison to the true abundance. We also subsampled the dataset to explore the effects of varying search effort and elk density on the precision and accuracy of results. We found that SCR outperformed REM methods in the full datasets, and reliably provided accurate (relative bias <10%) and reasonably precise (relative standard error ≤20%) estimates of density at moderately low to high densities (6-17 elk/km2), when the subsampling scenarios yielded a minimum average of 20 recaptures. We also found that the number of samples used to construct detection histories was a reliable predictor of precision, and could be used to establish minimum sampling requirements in future population surveys of elk. Although field-testing in free-ranging populations is needed, our results suggest that non-invasive genetic SCR is a promising tool for future population studies and monitoring of elk and potentially other gregarious ungulates. In contrast, the REM estimate of density was highly inaccurate, imprecise, and highly sensitive to camera parameters.

Robustness of fecal DNA spatial capture‐recapture to clustered space‐use by tule elk

“…This approach was largely successful in allowing us to collect sufficient numbers of samples to obtain acceptably precise estimates in 2 of the 3 populations surveyed. In the CC and LPB populations, our RSE estimates for the hybrid mixture models (10–14%) and single‐sex models (14–21%) compared well with those generated for most other ungulates using an SCR approach, including deer (20–28%; Brazeal et al 2017), giant eland ( Taurotragus derbianus , 25–36%; Jůnek et al 2015), caribou (5–21%; McFarlane et al 2020), and a study that used a captive population of tule elk of known size (12–29%; Brazeal and Sacks 2021). The precision of our estimates at EPR, on the other hand, was poor (30–52%), as expected from our small sample size.…”

“…Previous SCR estimation of density in captive ungulate populations of known size, including elk, also have proven to be relatively unbiased (Jůnek et al 2015, Brazeal and Sacks 2021). An SCR study conducted on an enclosed tule elk population of known size ( N = 79) indicated that relative bias was generally low (<10%) and decreased with increases in sample size (Brazeal and Sacks 2021).…”

“…Previous SCR estimation of density in captive ungulate populations of known size, including elk, also have proven to be relatively unbiased (Jůnek et al 2015, Brazeal and Sacks 2021). An SCR study conducted on an enclosed tule elk population of known size ( N = 79) indicated that relative bias was generally low (<10%) and decreased with increases in sample size (Brazeal and Sacks 2021). Those researchers reported that approximately 70 genotypes (i.e., ~1 genotype on average per individual) were sufficient to estimate abundance (i.e., N = 79) with a relative bias <10% and an RSE <20%.…”

“…To select survey units, we stratified the landscape according to an SDM to ensure that we focused most pellet‐sampling effort in portions of the landscape more likely to be used by elk but also sampled other portions of the landscape sufficiently to avoid biasing estimates (Figure 1; Batter 2020). For each of the 3 populations, we overlaid grids comprised of 4‐km 2 square cells, which we expected to approximate the radius of an average home range of tule elk (Efford and Fewster 2013, Brazeal and Sacks 2021). We dichotomized cells into high‐probability (HP) and low‐probability (LP) of elk occurrence based on the SDM (Batter 2020).…”

“…range of tule elk Fewster 2013, Brazeal andSacks 2021). We dichotomized cells into highprobability (HP) and low-probability (LP) of elk occurrence based on the SDM (Batter 2020).…”