Detecting the influence of environmental covariates on animal movement: a semivariance approach

Movements of organisms are influenced by environmental variation. For example, the movement rates of animals in unsuitable habitats are often different from movement rates of animals foraging in high-quality habitats. Different statistical methods to detect such effects exist, but they often rely on complete and regularly sampled data or require the formulation of an explicit movement model. We propose an extension to a recently introduced semivariance framework to identify the effects of environmental or other kinds of covariates on animal movement. Our extension also applies to sparse and irregularly sampled data, and it does not require the formulation of an explicit movement model. Within this extension of the semivariance framework, the observed movement rates at different time lags are modelled as a linear regression of the environmental covariates. To account for the inherent autocorrelation in semivariance data, we test for the statistical significance of the influence of covariates using a permutation approach. Our approach correctly identified covariates that influenced or did not influence movement rates in a simulation study. In a case study based on tracking data of a single female red deer (Cervus elaphus) individual from southern Austria, an application of the method showed that movement rates peaked during periods of intermediate temperature, but they do not co-vary with altitude and precipitation.