Effect of Undercounting and Model Selection on a Sightability-Adjustment Estimator for Elk

Publisher Website
Google Scholar
Abstract
Aerial surveys of wildlife populations must correct for failure to observe all animals to obtain unbiased population estimates. We captured and radiocollared elk (Cervus elaphus) in northcentral Pennsylvania to develop a linear-logistic model of characteristics of elk groups associated with visibility bias during helicopter surveys. We used this sightability model in a Horvitz-Thompson sightability-adjustment population estimator. The number of elk in a group was positively correlated (P < 0.001), and the percent canopy cover was negatively correlated (P = 0.002) with the probability of observing a group of elk. Observations of elk groups of known size indicated that helicopter crews undercounted elk group sizes, and percentage of elk missed increased as percent canopy cover increased. Probably because of this undercounting, 3 of 6 population estimates from the sightability-adjustment estimator were less than the number of elk known to be alive on the study area. Simulations of elk surveys using our empirical data indicated that our population estimates may have been negatively biased by 20%, because of undercounting. The assumption of complete enumeration of sighted groups of animals should be verified when using this estimator. When all assumptions of the estimator were met during computer simulations, confidence intervals calculated under the assumption of asymptotic normality and log-normality did not provide nominal coverage. We found that a sightability model using only group size as an independent variable, even when the true sightability model also included percent canopy cover, provided population estimates with little negative bias (<1%), shorter confidence intervals, and the lowest mean square error (MSE). Further research is needed on selection of sightability models, and we recommend using a bootstrap technique to calculate confidence intervals. We believe the sightability-adjustment estimator is unsuitable for estimating population size for small populations (less than or equal to 1,000 animals).

This publication has 0 references indexed in Scilit: