Linear Home Ranges: Effects of Smoothing, Sample Size, and Autocorrelation on Kernel Estimates

Blundell, G; Maier, J; Debevec, E

Simulations are necessary to assess the performance of home-range estimators because the true distribution of empirical data is unknown, but we must question whether that performance applies to empirical data. Some studies have used empirically based simulations, randomly selecting subsets of data to evaluate estimator performance, but animals do not move randomly within a home range. We created an empirically based simulation using a behavioral model, generated a probability distribution from those data, and randomly selected locations from that distribution in a chronological sequence as the simulated individual moved through its home range. Thus, we examined the influence of temporal patterns of space use and determined the effects of smoothing, number of locations, and autocorrelation on kernel estimates. Additionally, home-range estimators were designed to evaluate species that use space with few restrictions, traveling almost anywhere on the landscape. Many species, however, confine their movements to a geographical feature that conforms to a relatively linear pattern. Consequently, conventional analysis techniques may overestimate home ranges. We used simulations based upon coastal river otters (Lontra canadensis), a species that primarily uses the aquatic-terrestrial interface, to evaluate the efficacy of fixed and adaptive kernel estimates with various smoothing parameters. Measures of shoreline length within contours from fixed kernel analyses and the reference smoothing parameter were best for estimates of 95% home ranges, because smoothing with least squares cross validation (LSCV) often resulted in inconsistent results, excessive fragmentation, and marked underestimates of linear home ranges. Core areas (50% density contours) were best defined with fixed kernel LSCV estimates. Fewer locations underestimated linear home ranges, and there was a subtle positive relation between home-range size and autocorrelation. Generally, as location numbers increased, autocorrelation increased, but differences from the "true" home range decreased. Results were similar for our simulations and empirical data from 13 river otters. Examination of empirical data revealed that data with high positive autocorrelation illustrated seasonal reproductive activities. Because autocorrelation does not negatively influence estimates of linear home ranges, assessment of independence between data points may be more appropriately viewed as a means to identify important behavioral information, rather than as a hindrance.

Linear Home Ranges: Effects of Smoothing, Sample Size, and Autocorrelation on Kernel Estimates

Series