Autocorrelation of Location Estimates and the Analysis of Radiotracking Data

TLDR

Studies of wildlife radiotracking have debated whether autocorrelation in movement data biases home‑range estimates and habitat‑selection tests, yet properly designed sampling that captures representative movement within a defined time frame should mitigate such bias. The authors advise against using location‑based habitat‑selection analyses that treat individual locations as independent samples and instead propose a sampling framework that treats individuals as the sampling unit. They recommend intensive, time‑frame‑specific sampling of each animal’s home range to improve individual estimates, warn that using locations as sample units leads to pseudoreplication, and provide a general sampling design outline.

Abstract

The wildlife literature has been contradictory about the importance of autocorrelation in radiotracking data used for home range estimation and hypothesis tests of habitat selection. By definition, the concept of a home range involves autocorrelated movements, but estimates or hypothesis tests based on sampling designs that predefine a time frame of interest, and that generate representative samples of an animal's movement during this time frame, should not be affected by length of the sampling interval and autocorrelation. Intensive sampling of the individual's home range and habitat use during the time frame of the study leads to improved estimates for the individual, but use of location estimates as the sample unit to compare across animals is pseudoreplication. We therefore recommend against use of habitat selection analysis techniques that use locations instead of individuals as the sample unit. We offer a general outline for sampling designs for radiotracking studies.