Evaluation of trap capture in a geographically closed population of brown treesnakes on Guam.
Open population mark-recapture analysis of unbounded populations accommodates some types of closure violations (e.g. emigration, immigration). In contrast, closed population analysis of such populations readily allows estimation of capture heterogeneity and behavioural response, but requires crucial assumptions about closure (e.g. no permanent emigration) that are suspect and rarely tested empirically. In 2003, we erected a double-sided barrier to prevent movement of snakes in or out of a 5-ha semi-forested study site in northern Guam. This geographically closed population of >100 snakes was monitored using a series of transects for visual searches and a 13×13 trapping array, with the aim of marking all snakes within the site. Forty-five marked snakes were also supplemented into the resident population to quantify the efficacy of our sampling methods. We used the program MARK to analyse trap captures (101 occasions), referenced to census data from visual surveys, and quantified heterogeneity, behavioural response, and size bias in trappability. Analytical inclusion of untrapped individuals greatly improved precision in the estimation of some covariate effects. A novel discovery was that trap captures for individual snakes consisted of asynchronous bouts of high capture probability lasting about 7 days (ephemeral behavioural effect). There was modest behavioural response (trap happiness) and significant latent (unexplained) heterogeneity, with small influences on capture success of date, gender, residency status (translocated or not), and body condition. Trapping was shown to be an effective tool for eradicating large brown treesnakes Boiga irregularis (>900 mm snout-vent length, SVL). Synthesis and applications. Mark-recapture modelling is commonly used by ecological managers to estimate populations. However, existing models involve making assumptions about either closure violations or response to capture. Physical closure of our population on a landscape scale allowed us to determine the relative importance of covariates influencing capture probability (body size, trappability periods, and latent heterogeneity). This information was used to develop models in which different segments of the population could be assigned different probabilities of capture, and suggests that modelling of open populations should incorporate easily measured, but potentially overlooked, parameters such as body size or condition.