ESTIMATING POPULATION SIZE FROM DNA‐BASED CLOSED CAPTURE‐RECAPTURE DATA INCORPORATING GENOTYPING ERROR

Animal identification based on DNA samples and microsatellite genotypes is widely used for capture-recapture studies (Woods et al. 1999, Boulanger et al. 2003, Eggert et al. 2003). The method shows promise in field protocols (Woods et al. 1999) and potentially minimal error rates in the DNA analysis (Paetkau 2003). Some studies show much higher error rates in individual identification (Creel et al. 2003). There will be some level of uncertainty, although in some situations the uncertainty level is small, in the identification of individuals from microsatellite genotypes. Closed-population capture-recapture analysis has received substantial attention over the past century. More recently, it has been extended to conditional likelihood parameterizations that allow individual covariates to better estimate capture probability (Huggins 1989, 1991) and mixture models to estimate population size in the presence of individual heterogeneity in capture probability (Norris and Pollock 1996, Pledger 2000). The major focus of research has been developing methods to handle varying capture probability. Any methods developed in the future will also have to account for varying capture probability to obtain robust estimates of population size. While DNA-based capture-recapture studies and standard tagging studies share several common characteristics, they differ in others. In a standard tagging study, the researcher attaches a unique tag to the animal and keeps a list of tags that have been used. In a DNA-based study, the genotype of the individual acts as the tag. Therefore, all individuals are tagged prior to the beginning of the study. Unfortunately, the researcher does not know what genotypes exist in the population and must obtain samples from the animals to extract DNA. In a standard tagging study, if a tag is read that does not match one known to be in the population, the researcher knows that the tag was incorrectly read and then either rereads the tag or ignores the observation. In DNA-based studies, the researcher does not have the luxury of immediately knowing which genotypes may be incorrect. Thus, a new form of sampling uncertainty is introduced. For both standard tagging and DNA-based studies, capture probability is <1.0. This necessitates a way to infer what portion of the population is not captured in order to determine the total population size. For a DNAbased study, capture probability is a combination of the probability of encountering a sample (hair, scat, feather, etc.) and the probability that the sample yields a sufficient quantity and quality DNA to amplify and genotype. Current closed-population capture-recapture analysis for estimating population size assumes an animal's mark is permanent and read correctly when the animal is captured (Otis et al. 1978). The use of genotype based identification can meet these assumptions in some situations, but the cost may be high. The cost comes in 2 pieces that clearly interact: (1) the monetary cost of analyzing the DNA and (2) the information loss when discarding samples that contain some degree of uncertainty in their identification. For example, the protocol described by Paetkau (2003) places a high emphasis on certainty of the genotype of the sample. In doing so, a large number of samples may have to be culled during the analysis. It may be beneficial to allow a small degree of uncertainty in the identification of a sample, perhaps 1-5%, if such a tradeoff would 1 E-mail: plukacs@cnr.colostate.edu