3 Estimating survival using computer-assisted photographic mark-recapture: The problem of misidentification error

Wednesday, August 5, 2009: 8:40 AM
Taos, Albuquerque Convention Center
Thomas Morrison , Department of Biology, Dartmouth College, Hanover, NH
Jun Yoshizaki , Patuxent Wildlife Research Center,, U.S. Geological Survey,, Laurel, MD
Doug Bolger , Biology, Dartmouth College, Hanover, NH
Jim Nichols , Patuxent Wildlife Research Center, Dartmouth College, Hanover, NH
Recent advances in computer vision technology have expanded the opportunity for using photographic identification within large-scale mark-recapture studies.  Photo-identification relies on variation in natural marking patterns to ‘mark’ and re-identify individuals from photographs.  However, error in the photo matching process, particularly using computer-assisted identification, limits the application of this method within a mark-recapture framework.  False negative errors create an unknown number of ‘ghost’ capture histories in the dataset, which can severely bias survival rate estimates.  The structure and quantity of ghost capture histories depend on the ways in which the photographs are collected, sorted and matched.  Using photographic data from a wild population of migratory wildebeest (Connochaetes taurinus) from the Tarangire Ecosystem, Tanzania and a computer-assisted identification system, we characterize the effect of misidentification error on capture histories and the resulting survival estimates.  We illustrate one approach to minimizing bias in survival estimating by conditioning captures on having been photographed at least twice during the initial sampling period.  


Simulated datasets demonstrate the that bias rapidly increases with increasing misidentification rates.  We estimate survival of both adult female and males.  We summarize the results of other algorithm-based matching programs and emphasize the importance of identifying the source and type of errors when using computer-assisted matching within a capture-recapture framework, particularly in how they violate assumptions of standard models.