Strong inference from transect sign surveys: combining spatial autocorrelation and misclassification occupancy models to quantify the detectability of a recovering carnivore

Monitoring of species using surveys of ambiguous signs and assuming 100% detectability produces potentially biased occupancy estimates. Novel analytical tools have been developed that correct for bias arising from imperfect detectability, species misidentification and spatial autocorrelation between detection survey replicates that can affect transect surveys. To date they have been applied individually, but their combined value is unclear. The recovery of carnivores such as the European pine marten Martes martes potentially has far reaching, but largely unknown, implications for ecosystem restoration. Analysis of the species' distribution has as yet been crude and hence unsuited for informing management. We aimed to assess the validity of standard scat surveys to provide recommendations to increase inference from future surveys. We employed spatially replicated scat surveys along forest paths in NE Scotland, genetic verification of scat provenance and occupancy modelling techniques to quantify pine marten detectability and variation therein. Detectability for 1 km and 1.5 km transects, comparable to standard protocols, was estimated to be 0.35 and 0.58 respectively, highlighting the importance of accounting for imperfect detectability. Detection probabilities decreased with vegetation cover and increased with path width. Models accounting for spatial autocorrelation between adjacent transect segments suggested that segments of ≥ 400 m could be analysed as spatial replicates with negligible bias. As is the norm, not all scats yielded DNA to genetically verify they were produced by pine marten. This was accounted for through the use of ‘miss-classification occupancy models’ which allowed the use of unverified scats, increasing detection probabilities while accounting for the probability of unverified scats being false positive detections. This study exemplifies that robust inference on species occupancy is achievable through careful consideration of sampling design and the application of readily available analytical techniques. Adopting best-practice need not increase monitoring costs and can even increase cost-efficiency.