Estimating detection probability for Canada lynx Lynx canadensis using snow-track surveys in the northern Rocky Mountains, Montana, USA

1. What they did

Canada lynx are listed as threatened in the lower 48 United States, so managers need to know where they are and where they are not. Snow-tracking is a low-cost way to find carnivores, but while one track shows a species is present, it is hard to say with confidence that a species is absent. The researchers asked: in areas known to hold resident lynx, how likely are snow-track surveys to find them, and how many repeat surveys would be needed to conclude that lynx are absent?

They worked in two areas of western Montana: the Purcell Mountains and the Seeley Lake area (including the Garnet Range). Radio-collared lynx from other studies confirmed that lynx lived there, but the technicians doing the surveys did not know where the lynx were.

• The grid. They laid a grid of 8 x 8 km squares over occupied lynx habitat. This is roughly the size of an average female lynx home range in southern boreal forests. They surveyed 12 squares in the Purcell Mountains and nine around Seeley Lake.
• The surveys. From January to March in 2006 and 2007, technicians drove snowmobiles at about 30 km per hour along forest roads chosen to pass through as much dense forest as possible. They searched an average of 10 km per square, twice each winter, usually about three days after a fresh snowfall. The same technicians worked in both years.
• Recording. They plotted every route and every lynx track on handheld Global Positioning System (GPS) units. Two tracks counted as separate unless the observers could clearly see, without leaving the route, that both came from the same lynx.
• Analysis. They modelled the number of lynx tracks per square, and the distance between tracks along routes, and used resampling (bootstrapping) to estimate the chance of finding at least one lynx. They then calculated how many repeat surveys would be needed to infer absence. They also tested how much the results would change if some tracks had been misidentified.

2. What they found

Survey effort and detections

• Purcell Mountains: 48 routes, 489 km, 416 lynx track detections. On average one detection every 2.27 km, the first detection after 1.8 km, and 8.5 detections per 10 km.
• Seeley Lake area: 35 routes, 371 km, 130 lynx track detections. On average one detection every 4.29 km, the first detection after 3.5 km, and 3.5 detections per 10 km.
• Results differed between the two areas but not between years, so the two years were analysed together and the two areas separately.

Chance of finding lynx

• After 10 km of searching in a square, the chance of finding at least one lynx track was 0.98 in the Purcell Mountains and 0.91 around Seeley Lake (overall range of the confidence intervals: 0.80 to 0.99).
• Most tracks were less than 1 km apart. After 1 km of survey, the chance of a detection was already 0.79 (Purcell) and 0.63 (Seeley Lake). It levelled off at about 7 km, at 1.0 and 0.97.

Repeat surveys needed

• For 95% confidence: 0.76 surveys in the Purcell Mountains and 1.24 around Seeley Lake.
• For 99% confidence: 1.18 and 1.91 surveys.
• So two surveys would be enough to infer presence or absence of resident lynx with 95% probability.

Misidentified tracks

When up to half the tracks were treated as misidentified, the chance of detection barely changed: about 0.94 (falling to 0.91 at 50%) in the Purcell Mountains, and about 0.80 around Seeley Lake. At most about two surveys (95%) or three surveys (99%) would then be needed.

3. What they concluded

• Snow-track surveys are highly effective for finding resident lynx across large landscapes. The authors suggest the high detection rate comes from lynx being very mobile, which spreads their tracks well across their home ranges.
• Average minimum lynx density in the survey areas was about 2.8 lynx per 100 km², from their collaring data. The results apply to populations of similar density, and could overestimate detection where lynx are scarcer or underestimate it where they are more common.
• Their recommendations to managers: survey all grid squares over forest where lynx are expected (or use adaptive cluster sampling if that is not possible); survey at least 48 hours, and preferably 72 hours, after a track-obliterating snowfall; search at least 7 km per 8 x 8 km square, since more distance adds little; and do at least two repeat surveys.
• Surveying from roads might bias detections, but in an earlier study (577 km of winter backtracking) they found no sign that lynx avoided forest roads, so they think the efficiency of snowmobile surveys is worth it.
• Genetic verification is needed for snow-track surveys to meet evidence standards for species identification. They did not collect DNA here because radio collars confirmed lynx were present. Backtracking a lynx track for at least 3 km would give a good chance of collecting hair or scat for DNA (a probability of 0.86 to 0.91 when combined with a 7 km survey). Collecting genetic samples is essential when surveys are used to document new or unconfirmed lynx populations.
• They have used the same method in Wyoming (2000 to 2002) and Colorado (2010) and found the only lynx known to be there.

4. How tracking was or was not used

Tracking was the survey method. Technicians on snowmobiles identified lynx tracks in snow along forest roads, counted them and recorded each one on GPS. They judged in the field whether two tracks belonged to the same lynx. No trailing was done in this study; backtracking for DNA samples is recommended, based on earlier work.

5. Tracker or observer certification

Reported but not validated. The technicians had several years of experience trapping and snow-tracking lynx and were “confident in their track identifications”, but their skill was not tested and no DNA was collected to confirm the tracks. The same technicians surveyed in both years to reduce observer bias. The authors note that correct identification cannot be assumed in management surveys, especially where staff have little experience of lynx. The sensitivity analysis suggests the detection results would hold even with substantial misidentification.

6. Other technologies, skills or innovations

• Snowmobiles for covering long routes in deep snow and mountain terrain.
• Trimble GeoExplorer 3 data-logging GPS units for routes and track locations, and ArcGIS 9.2 (Geographic Information System, GIS) for mapping.
• Very high frequency (VHF) radio telemetry from concurrent studies to confirm that lynx were present.
• A simple equation (McArdle 1990, Reed 1996) that turns a detection probability into the number of surveys needed to infer absence, chosen as a more accessible option for managers than data-hungry occupancy models.

7. Implications for conservation (from the paper)

• Snow-track surveys are a useful, relatively low-cost way to document lynx populations where their status is uncertain, and to infer absence with high confidence after one or two surveys across large areas such as national forests.
• Relying on anecdotal records (unverifiable sightings or sign) to map ranges of rare species leads to errors and wasted conservation funding.
• Later, more detailed studies of lynx distribution and density may need relatively few repeat visits.