While advances in animal tracking technology have increased the capacity to collect data to support home range analysis, advances in statistical techniques that can be applied to these data have been lagging. Kernel density estimation (KDE) and polygon based methods are currently the most widely used home range estimators in ecological research, but animal movement data strongly violate the assumption of independence necessary for these to be accurate. We use a new derivation of KDE that accounts for autocorrelation, autocorrelated-KDE (AKDE), to investigate how autocorrelated data influence these methods. We first use simulated data to investigate how the accuracy and precision of commonly used home range estimators (e.g., Href KDE; Hlscv KDE; MCP; LoCoH; and AKDE) vary across a range of sampling frequencies, durations, and movement processes. We then verify these findings using empirical relocation data from a wide range of mammalian, and avian species. Because one of the primary aims of home range estimation is to quantify an animal's lifetime area requirements, we focus particular attention on a cross-validation test of these estimators' abilities to accurately predict an animal's future space use.
Results/Conclusions
We show that conventional estimators are significantly, negatively biased when calculated on autocorrelated data, whereas AKDE was more accurate, and more robust to variation in sampling duration, frequency, and an individual's range crossing time. For coarsely sampled data, although all estimators converged to the true 95% area as the sampling duration increased, conventional methods required substantially more data than AKDE before achieving asymptotic consistency. Crucially, the negative bias from conventional methods worsened as the sampling frequency was increased, whereas AKDE estimates were unaffected. The same pattern of negative bias apparent in simulated data was also found in empirical data, where AKDE estimates were significantly larger than KDE and polygon based estimates. Cross-validation confirmed AKDE range estimates as more appropriate in terms of: i) the number of subsequent relocations included; ii) the statistical overlap; and iii) the mean log-likelihood of the distribution.
Inference from home range estimation can have important implications for ecological theory, and/or species conservation, but only if results are due to biological phenomena and not estimation bias. Due to improved accuracy and robustness, we discourage future use of KDE in favor of the functionally similar, but statistically robust AKDE - now fully implemented in the R package ctmm.