Evaluating a Bayesian approach to improve accuracy of individual photographic identification methods using ecological distribution data

Abstract

Photographic identification of individual organisms can be possible from natural body markings. Data from photo-ID can be used to estimate important ecological and conservation metrics such as population sizes, home ranges or territories. However, poor quality photographs or less well-studied individuals can result in a non-unique ID, potentially confounding several similar looking individuals. Here we present a Bayesian approach that uses known data about previous sightings of individuals at specific sites as priors to help assess the problems of obtaining a non-unique ID. Using a simulation of individuals with different confidence of correct ID we evaluate the accuracy of Bayesian modified (posterior) probabilities. However, in most cases, the accuracy of identification decreases. Although this technique is unsuccessful, it does demonstrate the importance of computer simulations in testing such hypotheses in ecology.

Keywords

• Bayesian statistics
• photo-ID
• prior knowledge
• site fidelity