Radio-tracking wildlife by triangulation: An evaluation of directional measurement errors and location estimators

Richard Mears Pace, Purdue University

Abstract

Radio signal direction data obtained in wildlife radio-tracking settings, like most measurements, are subject to error. Errors have been given some consideration for large and expensive tracking systems, but evaluations of more commonly used systems are sorely lacking. I evaluated some common wildlife tracking systems in regards to angular error precision. Choosing a representative configuration, I also evaluated estimation procedure performance through simulation. Finally, I examined the problems of using such location data in tests of habitat preference. Angular error data from stationary twin-yagi and single-yagi antennae were adequately fit with von Mises distribution and had circular standard devisions of 2.6$\sp\circ$ and 5.7$\sp\circ$, respectively. Errors in mobile systems also followed von Mises distributions, and twin-yagi antennae were much more precise. Mobile system error variances were considerably larger than those for stationary systems. Establishing truck direction contributed as much or more to overall error variance as did signal detection. In simulation experiments, maximum-likelihood estimates (MLE) exhibited superior performance among estimators examined under homogeneous and certain non-homogeneous error conditions. Least squares estimates (LSE) also performed well under homogeneous azimuth error variances. Under non-homogeneous azimuth error variances, Andrews estimates performed more similarly to MLE and were preferred over LSE. The average of coordinates (AveXY) algorithm examined herein should not be used without some modification. I presented extensions of least squares estimators which allow differing error variances. Simulation results demonstrated superiority of a non-linear weighted least squares (NLWLS) estimation procedure over MLE and Andrews estimators in 4 situations. The NLWLS estimator presented here should be applicable under a large set of conditions. I produced some realistic examples that demonstrated the dependency of specific habitat type misclassification rates on the distribution, juxtaposition and shape of habitat types and the sampling distribution overlaid on a study area. I showed that Type I error rates for Chi-square tests of habitat use versus availability were not stable under these realistic conditions. Attempts to 'correct' misclassified data to improve the power of such tests worsened Type I error rates. I concluded that we need considerable reevaluation of habitat preference studies which employed radio-tracking via triangulation.

Degree

Ph.D.

Advisors

Weeks, Purdue University.

Subject Area

Forestry

COinS