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

1. What they did

Pine martens are recovering in Scotland after a long history of persecution, and a lot of money goes into monitoring where they live, mostly by searching paths for their scats. Past surveys assumed that if pine martens were present, one transect would find their scats, and they confirmed the species with DNA because pine marten and red fox scats are easy to confuse. The authors set out to test how reliable these standard scat surveys are, and to show how readily available statistical tools could make them better.

• Study sites. They chose 31 sites in northeast Scotland, each in the largest woodlot of a 100 km² grid square (hectad). Four sites were on game-keepered estates where foxes were controlled, and 27 were picked at random. The design copied earlier national pine marten surveys so the results could be compared.
• Scat survey (June 2013). The lead author walked transects of 2 to 3 km along forest paths, about 100 km in all, and collected every possible pine marten scat, recording each location with Global Positioning System (GPS). For each scat he scored how sure he was (on a 1 to 10 scale) that it came from a pine marten, based on its shape and the characteristic “fruity” smell.
• Path and weather records. About every 50 m he recorded path type, path width, and vegetation height and cover on the path. Rain in the two days before and during each survey was also noted.
• DNA tests. Funding allowed 109 of the 120 scats to be tested in a laboratory to find out whether they came from a pine marten or a fox.
• Analysis. They split each transect into equal segments (0.1 to 1.5 km) and treated the segments as repeated surveys. With occupancy models, they estimated how likely surveyors are to find pine marten scats where pine martens are present, checked whether neighbouring segments were dependent on each other, tested which path and site features affected detection, and tested whether scats without a DNA result could still be used.

2. What they found

Scats collected

• Of 109 scats tested, 47 were confirmed as pine marten, 15 as red fox, and 47 gave no usable DNA. So about a quarter (24%) of the scats that gave DNA were actually from foxes.
• Confirmed pine marten scats were found at 21 of the 31 sites, a simple (“naive”) occupancy of 0.68. Counting the unconfirmed scats as well, five more sites would have been added.

Detection on standard transects

• The chance of finding confirmed pine marten scats where pine martens were present was only 0.35 on a 1 km transect and 0.58 on a 1.5 km transect, the lengths used in earlier national surveys. It rose roughly in a straight line from 0.04 for 100 m to 0.58 for 1.5 km.

Neighbouring segments

• There was no evidence that finding scats in one segment made finding them in the next segment more likely. Segments of 400 m could be safely used as separate repeated surveys, and the authors used 400 m in further analyses as a precaution.

What affected detection

• Taller vegetation and more vegetation cover on the path made scats harder to find. These were the most important features.
• Wider paths gave higher detection when considered together with vegetation height or cover.
• Rain before or during the survey, fox control, path type and the proportion of woodland had no clear effect. The proportion of woodland also did not affect whether sites were occupied.

Field identification versus DNA

• The surveyor’s certainty scores were unrelated to which species had actually produced the scat, so they could not replace DNA testing.
• Models that allow for some misidentified scats let the unconfirmed scats be used. This raised the estimated chance of detecting pine martens, though the difference was not statistically clear at all segment lengths.

3. What they concluded

• Standard pine marten scat surveys miss pine martens at many sites where they are present. Past surveys that assumed perfect detection have underestimated where pine martens occur, and have biased estimates of how fast they are spreading by an unknown amount.
• Occupancy modelling should be used in future surveys of pine martens and many other species. The analysis costs little compared with the fieldwork.
• Field identification of pine marten scats needs DNA confirmation, as also found by more experienced pine marten specialists in other studies. The authors caution that the chance of getting DNA from a scat might itself differ between species.
• Short segments of a transect can be used as repeated surveys without walking longer transects, although spatial autocorrelation models should always be checked in similar studies.
• Earlier surveys could be reanalysed if transect and scat locations were recorded, but the way scats were chosen for DNA testing in those surveys may bias the results. The authors recommend a standard design for choosing scats for DNA tests, testing all scats when possible, or pooling the scats from each segment.
• Models that allow for misidentification make good use of scats without DNA results, and will be especially useful at the edge of the expanding range, where pine martens are scarce.
• Better practice need not cost more, but uptake of these methods is low and needs better training and collaboration.

4. How tracking was or was not used

The study is a sign survey: the lead author searched forest paths for scats and identified possible pine marten scats in the field by their shape and smell. No trailing was reported. Each scat’s species was then checked with DNA, and the surveyor’s own certainty score for each identification was compared with the DNA result.

5. Tracker or observer certification

Not reported. All scats were found and identified by the lead author, and nothing is said about his experience or skill. This matters to the findings: about a quarter of the scats he identified as possible pine marten scats were from foxes, and his certainty scores did not predict the true species. The authors note that other studies by more experienced pine marten specialists found the same.

6. Other technologies, skills or innovations

• Laboratory DNA testing (quantitative PCR) to tell pine marten scats from red fox scats.
• Splitting transects into short segments to use as repeated surveys, with a test for dependence between neighbouring segments.
• Misclassification occupancy models that include scats without DNA results, allowing for the chance that some are misidentified.
• Free software (PRESENCE) for occupancy analysis, and GPS and ArcMap for mapping scats and transects.

7. Implications for conservation (from the paper)

• Monitoring is the basis for managing pine martens, whose spread raises questions about ground-nesting game birds such as the endangered capercaillie and about invasive grey squirrels. Ignoring imperfect detection could lead to misguided management and is a misuse of survey resources.
• Survey planning can use the effects of vegetation and path width to increase detection, though future surveys should re-estimate them.
• Fox control on private estates showed no effect on pine marten detection, which the authors suggest means it did not reduce fox numbers enough to change pine marten scat deposition.