Individual identification using hidden Markov models for population monitoring and assessment.

Abstract

Individually distinct acoustic features are present in a wide range of animal species, just as they are in humans, and the widespread success of speaker identification in humans suggests that robust automatic identification of individual animals from their vocalizations is attainable. Despite this, only a few studies to date have yet attempted to use individual distinctiveness to help assess population structure, abundance, and density patterns. Here we present an approach, based on individual identification and clustering using hidden Markov models (HMMs), which enables a more direct mechanism for using individual vocal variability to monitor and assess populations. Current results indicate that the new method is able to give good estimates of local abundance based on vocalization clustering, which can in turn be used in an acoustic mark-recapture framework to estimate population. Limitations to this approach currently include the need for explicit call-type separation prior to individual clustering, which is possible in many species but can create a problem in species with unknown or variable repertoires. Overall, it is hoped that this new technique may lead to a more accurate understanding of population structure and abundance on a larger scale.