Should the cocktail training method cause a stir? Training frog generalist detection dogs with combined species samples

Abstract

Wildlife detection dog (WDD) and handler teams can greatly support conservation efforts. Sourcing odourant samples to train these teams is challenging. One option involves training WDDs on common species, from which samples are more accessible, in the hope they will generalise to similar endangered species. Cocktail (mixture) training, where dogs are trained on samples that purposefully contain multiple odourants (e.g. frog species), may be suitable for rapidly training dogs on multiple species. Presently, there is confounding evidence of animals’ abilities to recognise individual components within mixtures. Our study aimed to determine whether training dogs with ‘frog cocktails’ of five different frog species would facilitate generalisation to novel (i.e. previously unencountered) frog species. Seven dogs were trained on five varieties of ‘frog cocktails’. Each cocktail contained skin swabs from four of the five frog species to minimise species odourant overshadowing. Training occurred indoors across 18 sessions on a 12-armed scent wheel. Double-blind assessments were then completed to determine whether dogs recognised each cocktail component (Species A – E). The dogs were assessed on each of the five species (10 trials per species) to measure sensitivity and precision. Lastly, the dogs were assessed on their ability to generalise to novel frog species (Species F – J). When the dogs were tested on cocktail components, they demonstrated poor species recognition, detecting frogs in 16/35 searches. Sensitivity and precision were highly varied between dogs and across frog species, with a mean sensitivity of 60.8 % and precision of 43.6 %. During novel frog species assessments the dogs demonstrated higher generalisation, alerting to 23/30 novel frog species. The majority of dogs scored highly for sensitivity and precision across these novel species assessments, with a mean sensitivity of 81.7 % and precision of 72.9 %. Our results demonstrate the dogs’ relatively poor recognition of the cocktail components following training, presumably because they were likely processing the cocktails as a whole, rather than as a collection of unique odourants. Their ability to process the cocktails elementally may have improved with a longer training period. However, as a result of this cocktail training and a series of assessments on individual cocktail components, the dogs were able to generalise to novel frog species, suggesting the dogs understood the target category of ‘frog’. Our study highlights considerations for employing the cocktail training method, as well as potential benefits of this training for endangered wildlife detection.