Evaluation of closed capture-recapture methods to estimate abundance of Hawaiian monk seals

Numerous capture-recapture methods have been developed to estimate abundance, yet the performance of these models is only rarely judged by comparison with true abundance. This study presents a rare opportunity to assess capture-recapture estimates in a free-ranging population with known minimum abundance. Hawaiian monk seal abundance historically has been characterized using a trend index or has been estimated using simple enumeration. Here, I evaluate the applicability of various closed-population capture-recapture models to estimating Hawaiian monk seal abundance and its associated error. I analyzed 12 data sets (two years from each of six subpopulations) representing a wide variety of sampling and logistical scenarios, using models that explored the effects of animal size class (juvenile, subadult, or adult), tag status, and sighting location on initial capture and recapture probabilities. I also explored various models to account for capture heterogeneity among individuals. Size and sex effects always significantly improved model fits, and tag status and location effects were also frequently influential. In most cases, abundance estimated from capture-recapture models was substantially lower than known minimum abundance, suggesting the influence of individual capture heterogeneity. Attributes of individuals known to be alive, but not captured during systematic surveys, did not reveal patterns that explained sources of capture heterogeneity. In some cases, mixture models produced estimates that were less biased but were associated with very large confidence intervals. Among the model types examined, those available in Program CAPTURE performed best; although they are still prone to negative bias, these models nevertheless may prove useful in characterizing population trends in Hawaiian monk seals. This study demonstrates that selection of appropriate closed capture-recapture models can be substantially improved by independent validation.