Performance of visual vs. software-assisted photo-identification in mark-recapture studies: a case study examining different life stages of the Pacific Horned Frog (Ceratophrys stolzmanni)

The use of individual phenotypic features as non-invasive marks through photo identification (photo-ID) has many advantages for individual-based field studies. However, since error rates vary between species, matching techniques, and image qualities, it is important to assess error rates to evaluate the reliability of photo-ID. We tested the performance of this method in the Pacific Horned Frog, Ceratophrys stolzmanni, a threatened anuran with a variable dorsal coloration. We compared the effectiveness of visual (i.e. human recognition) versus software-assisted recognition techniques (Wild-ID and APHIS), and assessed the potential applicability of these techniques to identify post-metamorphic individuals (froglets). The three techniques differed in their performance. The visual technique was the only one to produce type I errors between pairs of images (i.e. matching images which are not true matching pairs), however it also produced fewer type II errors (i.e. not identifying true matching pairs of images) than the software-assisted techniques. For froglets, we found that similarity scores calculated by photo-ID software did not differ significantly between groups of froglets and older individuals, supporting that colour patterns in this species remain stable after metamorphosis. The lack of type I errors and the substantial reduction in required time achieved through the use of software, make software-assisted techniques (particularly Wild-ID) a valid and convenient matching technique for C. stolzmanni. We propose that photo-ID can also be applied to froglets of this species, since the pattern at this stage appears to be as variable as in adults and maintained through life.