Successful automated photographic identification of larvae of the European Fire Salamander, Salamandra salamandra

Computer-aided individual recognition of animals based on their natural markings has become an indispensa-ble tool in ecology research. However, this is problematic in species with faint patterns. Here we test whether individually reared larvae of the European Fire Salamander (Salamandra salamandra) can be reliably recognized from images of their lateral tail patterns. We used Wild-ID software to (1) estimate the recognition uncertainty resulting from pre-processing the images, (z) quantify how pre-contrasting improves recognition, (3) assess the effect of ontogenetic pattern change on recognition until metamorphosis, and (4) test how recognition performs with larger image libraries. Our results show that discrimination of larvae is highly successful. Pre-processing did not lead to a relevant change in the recognition probability, while pre-contrasting even reduced the recognition probability. The shorter the time interval between two photos, the more readily an individual will be recognized. The overall recognition rate was 99.81%, with false rejection rates (FRR, calculated as the number of falsely rejected images divided by the number of matching attempts) amounting to 4.66, 0.77 and 0.z0% for FRR1 (first image provided by Wild-ID does not match), FRR10 (none of the first ten images provided by Wild-ID match-es) and FRR (none of the first z0 images provided by Wild-ID matches), respectively. These rates are among the lowest ever reported. The inclusion of images of 130 wild-caught larvae did not negatively affect successful individual recognition. Au-tomated photo-identification may therefore be considered a reliable tool for fieldwork on European Fire Salamander larvae.