Hopping from one species to another - Training requirements of frog generalist detection dogs

Wildlife detection dog teams have been widely demonstrated to be a highly sensitive and effective monitoring method. Sourcing suitable samples of target species with which to train detection dogs can be challenging. An alternative is training on similar species (or multiple species), with the aim of the dog generalising to the desired target species or category of species (e.g. 'frog'). If successful, this approach could allow for the detection of species not available for training or not previously discovered, and permit data to be collected on multiple species, if desirable. Our study aims were to determine if dogs could be trained to generalise across novel frog species after training on five species and to record how many individuals within a species would be required for training. Using a scent wheel, we trained six dogs to detect five frog species, sequentially. After reaching training criteria, dogs were assessed during double-blind trials on five novel frog species. 'Training species' belonged to two genera (Litoria sp. and Limnodynastes sp.) and 'novel species' belonged to four genera (Philoria sp., Litoria sp., Mixophyes sp. and Pseudophryne sp.). The number of training sessions required on each frog species before reaching criteria decreased from Species A to Species E. During trained species assessments, the dogs consistently generalised to novel individuals after being trained on 10 individual frogs from each species. When the dogs were assessed on five novel frog species they spontaneously generalised to each novel species on 17/20 instances. The dogs' sensitivity and precision scores across assessments were very high for both the trained species (sensitivity = 91.4 %; precision = 80.8 %) and novel species (sensitivity = 90.0 %; precision = 84.8 %). Our results demonstrate dogs' abilities to spontaneously generalise across multiple frog species, whilst maintaining high target sensitivity and precision. This suggests that the dogs' learned to detect the category of 'frogs' in a controlled setting. This training approach may therefore be highly valuable for training detection dogs on endangered species when animals or samples are challenging, or impossible, to source, or training time is limited. Further research is required to determine whether generalisation across frog species can be demonstrated in live frog detection in field settings.