An astronomical pattern‐matching algorithm for computer‐aided identification of whale sharks Rhincodon typus

Abstract

Summary The formulation of conservation policy relies heavily on demographic, biological and ecological knowledge that is often elusive for threatened species. Essential estimates of abundance, survival and life‐history parameters are accessible through mark and recapture studies given a sufficiently large sample. Photographic identification of individuals is an established mark and recapture technique, but its full potential has rarely been exploited because of the unmanageable task of making visual identifications in large data sets. We describe a novel technique for identifying individual whale sharks Rhincodon typus through numerical pattern matching of their natural surface ‘spot’ colourations. Together with scarring and other markers, spot patterns captured in photographs of whale shark flanks have been used, in the past, to make identifications by eye. We have automated this process by adapting a computer algorithm originally developed in astronomy for the comparison of star patterns in images of the night sky. In tests using a set of previously identified shark images, our method correctly matched pairs exhibiting the same pattern in more than 90% of cases. From a larger library of previously unidentified images, it has to date produced more than 100 new matches. Our technique is robust in that the incidence of false positives is low, while failure to match images of the same shark is predominantly attributable to foreshortening in photographs obtained at oblique angles of more than 30°. We describe our implementation of the pattern‐matching algorithm, estimates of its efficacy, its incorporation into the new ECOCEAN Whale Shark Photo‐identification Library, and prospects for its further refinement. We also comment on the biological and conservation implications of the capability of identifying individual sharks across wide geographical and temporal spans. Synthesis and applications. An automated photo‐identification technique has been developed that allows for efficient ‘virtual tagging’ of spotted animals. The pattern‐matching software has been implemented within a Web‐based library created for the management of generic encounter photographs and derived data. The combined capabilities have demonstrated the reliability of whale shark spot patterns for long‐term identifications, and promise new ecological insights. Extension of the technique to other species is anticipated, with attendant benefits to management and conservation through improved understanding of life histories, population trends and migration routes, as well as ecological factors such as exploitation impact and the effectiveness of wildlife reserves.