A review of movement models in open population capture-recapture

Understanding rates of survival and recruitment is critical to population management, and capture-recapture methods of estimation are widely used. Spatial models allow for a spatial detection process and can include the movement of activity centres between sampling times. Movement is often treated as a random walk with the step length governed by a probability kernel. However, the movement component of open population spatially explicit capture-recapture models (open SECR) has evolved haphazardly and comparison among studies is difficult. We review published studies, document suitable probability kernels and address the issues of scale and buffer dependence in open SECR by a combination of simulation and case studies on ovenbirds Seiurus aurocapilla and tigers Panthera tigris. Flexible 2-parameter kernels, such as the bivariate t$$ t $$-distribution, fit better than the popular bivariate normal and resulted in higher estimates of survival. We reconcile different parameterizations of the bivariate t$$ t $$-distribution and identify a problem when the kernel is defined in terms of its margins. Movement models failed to separate mortality and emigration in simulated data when the data were a random mixture of long and short movements. Our estimates of ovenbird survival were buffer-dependent, and we interpret this as a sign that the data are inadequate for joint modelling of survival and movement. Estimates of tiger survival were more nearly asymptotic on buffer width. We repeat the warning of earlier authors that movement models are effective for separating mortality and emigration only when the data span the range of movement. We appeal for more complete and consistent reporting of movement models and identify topics for future research.