Enhancing performance of detection dog field surveys through experimental transects

Abstract

Wildlife detection dogs facilitate scat sample collection for wide-ranging species such as many large carnivores. Performance evaluations are required to effectively compare dog teams. Key metrics for evaluation are the ratio of targets detected (sensitivity) and the ability to discriminate the target scent (precision). These metrics and their predictors are commonly evaluated through experimental assessments; however, the extent to which these can inform field performance has not been extensively explored. We analyzed 4 years of road transect data from field surveys and experimental assessments for 1 dog team searching for cheetah (Acinonyx jubatus) and African wild dog (Lycaon pictus) scat in central-east Namibia. During assessments, we found a sensitivity of 50% for up to 35 m away from the transect line, comparable to other studies. The detection probability for each sample was positively associated with decreasing distance from the samples to the road, favorable wind conditions, the distance covered by the dog during a search, and the length of training the dog had benefited from. We calculated a 32-m effective sweep width (ESW) from assessment transects and found that most field samples were located closer to the transect line than expected based on ESW (78–93% of finds <10 m). We calculated a precision of 82% for field samples, and the probability of correct identification for each sample was weakly positively associated with decreasing humidity. The assessment transects provided valuable information for optimizing field surveys, and we provide practical recommendations for their implementation and recommend integrating them into training routines. We provide evidence for the benefit of searching into the wind and suggest encouraging the dog to search closely around roads for improved efficiency when targeting species with similar ecological traits.