ESTIMATING PREVALENCE OF HYBRIDS IN FREE-RANGING ADMIXED POPULATIONS: A CAPTURE-RECAPTURE MULTIEVENT MODELLING APPROACH

Anthropogenic hybridization is recognized as a major and increasing threat to biodiversity. The estimation of prevalence of hybrids (proportion of hybrids in the total population) is of paramount importance to understand the extent of the phenomenon and consequently inform appropriate management policies. However, estimating hybrids’ prevalence in free-ranging animal populations is far from being trivial due to methodological challenges. Estimates based on the raw percentage of hybrids in the sample are likely to be flawed due to two main sources of bias. First, differences in the detectability of hybrids and parental individuals may cause one of the two categories to be over- or under-represented in the sample. Second, both type-I (erroneously classifying a parental individual as a hybrid) and type-II (erroneously classifying a hybrid as parental) errors may occur due to the limited power of clues currently used to assess hybridization (e.g., molecular markers). We developed a multievent capture-recapture model to estimate hybrids’ prevalence in intermixing populations accounting for both detection probabilities and error rates in hybridization assessment. We defined three states (‘alive and hybrid’, ‘alive and parental’, ‘dead’) and included uncertainty in hybrids identification by allowing for uncertainty in state assignment through the observation process. In addition, we carried out simulation work to assess the precision in prevalence estimates as a function of population size and the level of uncertainty in the identification of hybrids. We illustrated our approach on a wolf x dog hybridization case study in the northern Apennines, Italy. Alleged wolf scats were genotyped and attributed to the category ‘hybrid’ or ‘parental’ according to genetic evidences based on a set of uni and bi-parental genetic markers. As it can be expected that phenotypic cues might reveal the admixed origin of wolves genetically classified as parental, we also estimated type-II errors, and corrected prevalence estimates accordingly, by integrating in the model information from a subsample of individuals that were both phenotypically and genetically assessed. Our approach improves the ‘naive’ estimates of prevalence by acknowledging sources of error in inferring hybridization, and therefore provides a useful tool to more reliably address anthropogenic hybridization in a management context.