The importance of sampling design for unbiased estimation of survival using joint live-recapture and live resight models

Abstract

Survival is a key life history parameter that can inform management decisions and life history research. Because true survival is often confounded with permanent and temporary emigration from the study area, many studies must estimate apparent survival (i.e., probability of surviving and remaining inside the study area), which can be much lower than true survival for highly mobile species. One method for estimating true survival is the Barker joint live-recapture/live-resight (JLRLR) model, which combines capture data from a study area (hereafter the capture site) with resighting data from a broader geographic area. This model assumes that live resights occur throughout the entire area where animals can disperse to and this assumption is often not met in practice. Here we use simulation to evaluate survival bias from a JLRLR model under study design scenarios that differ in the site selection for resights: global, random, fixed including the capture site, and fixed excluding the capture site. Simulation results indicate that fixed designs that included the capture site showed negative survival bias, whereas fixed designs that excluded the capture site exhibited positive survival bias. The magnitude of the bias was dependent on movement and survival, where scenarios with high survival and frequent movement had minimal bias. In effort to help minimize bias, we developed a multistate version of the JLRLR and demonstrated reductions in survival bias compared to the single-state version for most designs. Our results suggest minimizing bias can be accomplished by: 1) using a random resight design when feasible and global sampling is not possible, 2) using the multistate JLRLR model when appropriate, 3) including the capture site in the resight sampling frame when possible, and 4) reporting survival as apparent survival if fixed sites are used for resight with the single state JLRLR model.