Visiting Scholar Seminar Series: Andrew Royle, USGS

Capture-recapture methods have a long history in ecology and they provide a wealth of demographic information on animal populations including population size or density, and vital rates such as survival and recruitment. Capture-recapture methods require individual identification of animals in a population which historically required physical marking such as banding, notching, ear tagging, or toe clipping. The field of capture-recapture has been revolutionized by the adoption of new technologies for identification such as noninvasive genetics and photographic identification using remote cameras which do not require physical marking, thus allowing capture-recapture to be applied to many situations where physical marking has been impractical. Recent developments in statistical classification (or machine learning) seek to identify individuals from digital samples such as photographs or audio spectrograms automatically, thus removing or streamlining the step of manual interpretation in the conversion of data from new technologies to individual identities. Turtles are especially amenable to the application of methodologies for individual identification based on features extracted from digital photographs. We are developing approaches to individual identification of turtles from photographs, using a case study of box turtles on the Patuxent Research Refuge. There is a long history of turtle marking on the refuge going back to the 1950s and a large database of known-identity individuals exists. This has been supplemented with photographs from roadside encounters by "citizen scientists" and an on-going effort based on field searches. In my seminar, I will introduce the Patuxent box turtle study and discuss the context of capture-recapture and automated classification.