COS 118-6
Rigorous home-range estimation: Rederiving the kernel-density estimator for use with autocorrelated data
Kernel-density estimation (KDE) is unique in its role as a statistically efficient yet non-parametric method for estimating the probability-density function of independent and identically distributed data. Despite its widespread application to home-range estimation, animal tracking data violates the underlying assumption of independence via the presence of autocorrelation. Here, we rederive the kernel-density estimator from first principles, dropping the assumption of independence, and allowing for the presence of autocorrelation. Our new kernel-density estimator is valid for autocorrelated data, which makes it perfectly suited for movement data.
Results/Conclusions
We test our method against both real and simulated data using a sample of Mongolian gazelles and their previously identified autocorrelated movement model. For the individual gazelle with the longest span of data, our autocorrelated KDE predicts a home-range area of 350,000 -- 800,000 km2, while the conventional estimate is only 46,000 -- 53,000 km2. These results are consistent with those from simulated data. As the sampling rate increases and data quality improves, the conventional home-range estimate shrinks with tightening confidence intervals.