Accuracy of home range estimators for homogeneous and inhomogeneous point patterns

Abstract

The only assumption of kernel density estimation, the most popular method of animal home range delineation, is that sample points used to approximate the utilization distribution are independent and identically distributed (i.i.d.), or what is termed ‘homogeneous’ or ‘stationary’. This research evaluated the accuracy of several home range estimators for both homogeneous and inhomogeneous point patterns of animal tracking data. A novel technique was developed to simulate home ranges from an extensive database of observed Florida panther ( Puma concolor coryi) locations. Thirty-five homogeneous and inhomogeneous point distributions, each, were generated and used to delineate known home ranges for individual animals. Random samples of 100, 250, 500, and 1000 points from each simulated distribution were used to evaluate the accuracy of fixed and adaptive kernel density estimation (KDE), local nearest neighbor convex hull ( k-LoCoH), and characteristic hull polygon (CHP) methods for delineating home ranges. Both fixed and adaptive KDE overestimated home ranges by up to 30% for homogeneous samples and underestimated areas by about 50% for inhomogeneous ones. The k-LoCoH method performed relatively better than KDE, tending to produce slight underestimates of home range sizes, but its performance was widely affected by both homogeneity and sample size. The CHP method produced relatively unbiased estimates of home range sizes for both types of samples, with results fairly robust to sample size. These findings demonstrate the influence of inhomogeneity or nonstationarity in animal locational data on the accuracy of home range estimation methods and suggest that ecologists should consider multiple underlying statistical assumptions when selecting a home range estimator.