Accurate home-range estimates are important for conservation and wildlife management interests, as they inform animal space-use requirements. While home-range estimation is a ubiquitous application of animal tracking data, conventional methods (KDE, MCP, and Brownian bridges) yield downwardly biased home-range area estimates without corresponding confidence intervals (CIs). Advanced statistical techniques are necessary to account for the strong autocorrelation inherent to movement data, the limited number of home-range crossings that one can realistically observe (which set the effective sample size), and temporal irregularities that typify animal tracking data. Reliable confidence intervals are also necessary to make sound statistical inference. Here we describe the development of novel home-range estimators that are uniquely capable of both accounting for autocorrelation and estimating the correct target distribution. Previous methods either assume a lack of autocorrelation (KDE and MCP) or estimate the incorrect target distribution for home-range analysis (Brownian bridges), with both categories of misspecification resulting in downwardly biased home-range area estimates. We cover four new methods that can be used in combination to mitigate these biases: autocorrelated kernel density estimation (AKDE) and its associated CIs, bias-corrected AKDE (AKDEC), optimally weighted AKDE (wAKDE), and perturbative REML (pREML).
Results/Conclusions
Using both real data and data simulated from fitted movement models, we demonstrate the increased statistical efficiency and predictive capabilities of these emerging methods. We show with real data examples that conventional estimates can easily be less than 50% the required area to accurately predict space-use requirements when they ignore autocorrelation, and less than 10% when they estimate the incorrect target distribution. Specifically we show that AKDE can properly accommodate arbitrarily high sampling rates, AKDEC can mitigate the majority small sample size bias given an accurate movement model, wAKDE can completely account for temporal sampling irregularity, and pREML can accurately estimate movement models down to effective sample sizes of 2-3, which, unfortunately are not uncommon in tracking data. Furthermore, and in contrast to previous methods, these newer estimators come equipped with confidence intervals, which we show to be reliable. All of these methods are implemented in the easy to use R package "ctmm", widely available on the CRAN repository.