Evaluating Methods for Counting Cryptic Carnivores

1. What they did

Large carnivores are shy, often solitary and nocturnal, range widely and live at low densities, so counting them is hard. Many counting methods exist, but few have been tested against a population whose true size is known. At Phinda Private Game Reserve in South Africa, the researchers had such a population: nearly every leopard on the reserve wore a radio collar.

• Reference density. Between 2002 and 2007 they captured 35 leopards and fitted them with very high frequency (VHF) radio collars; in 2005, 14 collared leopards used the reserve. They tried to locate every leopard once a day, using a handheld Global Positioning System (GPS) receiver or radio triangulation. From the share of locations inside the study area, they calculated a reference density for each three-month quarter and for the year. Safari guides, who searched for leopards twice a day on game drives, radioed in every leopard sighting, which helped show whether any uncollared leopards were present.
• Track counts. They drove four road transects (about 35 km each) in the early morning at 10 to 15 km per hour, with a skilled observer on the front of the vehicle recording fresh (24 hours old or less) leopard tracks and the identity of each leopard. They then turned track density into population density in three ways: with Stander’s (1998) calibration equation from Namibia, with their own “Phinda calibration”, and with a capture and recapture model in which each road count was one “capture” occasion.
• Camera traps. From January to March 2005 they ran 30 camera stations, 2 km apart, each with two film cameras to photograph both flanks. They identified leopards from their coat patterns and estimated population size with capture and recapture models. They tested 10 ways of working out the effectively sampled area: five ways of calculating the buffer width, each added either as a strip around the outer cameras or as a circle around each camera.
• Costs. They compared the money and time each method needed.

2. What they found

Reference density

• The 14 collared leopards gave a mean annual density of 7.33 leopards per 100 km², and 7.07 per 100 km² for January to March, when the surveys ran.
• Only one uncollared leopard was known on the reserve, an adult female who spent little time there. Guides reported 120 leopard sightings in 41,846 hours of searching in 2005.

Track counts

• Over 1,253 km of road, the team counted 76 track sets from 9 collared leopards: one leopard track for every 16.5 km driven.
• Male tracks (47) were found on roads more often than expected from how many males there were.
• Precision improved sharply over the first 200 km driven and levelled off at about 600 km.
• Stander’s calibration gave less than half the reference density (3.19 per 100 km²). The Phinda calibration gave 4.53, still a large underestimate, and the relationship between track density and reference density was not statistically significant.
• Capture and recapture from tracks of known individuals gave 6.45 per 100 km², close to the reference density.

Camera traps

• In 2,400 camera-trap nights, cameras photographed 12 collared leopards on 38 occasions.
• The most accurate estimate (6.97 per 100 km², 1% below the reference) came from a buffer based on how far collared leopards moved outside the camera grid, measured by telemetry, placed around each camera.
• The traditional buffer, half the mean maximum distance moved between camera photographs, came next (6.56 per 100 km²). Other authors had suggested this buffer overestimates density; here it did not.
• Placing the buffer around each camera generally worked better than a strip around the outer cameras.
• Every estimate from every method was lower than the reference density. Camera traps were generally more accurate but less precise than track counts.

Costs (2005)

• Radio telemetry: US$22,312 and 3,497 hours.
• Camera traps: US$4,749 and 174 hours.
• Track counts: US$490 and 216 hours.

3. What they concluded

• Applied correctly, camera-trap surveys give the best balance of rigour and cost for counting cryptic carnivores that can be told apart as individuals.
• The way the effectively sampled area is calculated has a large effect on density estimates. The authors recommend standardising on the individual camera buffer method.
• Track counts are cheap and easy to run over large areas, but their accuracy was poor. They may suit tracking trends in numbers at one site over time, ideally calibrated against at least one other independent estimate.
• Stander’s calibration was developed in semi-arid areas with low carnivore densities and ideal tracking conditions. At higher densities and on mixed substrates it has limited use: habitat, soils, differences between individuals (males used roads more) and road use by protected leopards all affect how many tracks are found.
• Capture and recapture from tracks depends on telling individuals apart from their tracks, which most studies cannot do with certainty.
• Earlier criticism of the traditional camera buffer should be treated with caution, because those studies had few collared animals.

4. How tracking was or was not used

Tracking was one of the two methods being tested. A skilled observer on the front of a vehicle recorded fresh leopard tracks on sandy roads. The team assigned each track set to a known leopard mainly from what telemetry told them: each leopard’s core area, its movements the night before, the nearest collared leopard to the tracks, and, for a few leopards, unique marks in their tracks. To check these identifications, they followed the tracks (trailing) after assigning identity on 23 occasions, and the identity was confirmed every time. Four track sets that could not be confidently identified were left out, as were six track sets believed to be from the uncollared female. The authors note that their method cannot give absolute certainty about identity.

In the discussion, they review other ways of identifying individuals from tracks: highly skilled Ju/’Hoan trackers in Namibia identifying leopards, lions and wild dogs from footprints (Stander et al. 1997), measurements of pugmarks, footprint patterns from track plates, and following tracks until hair or scat could be collected for genetic identification.

5. Tracker or observer certification

Reported but not validated. The track counts were done with a “skilled observer”, who is not named, and no certification or skills test is reported. The safari guides are described as highly trained and skilled in tracking, also without validation. The 23 trailing checks tested the identifications of individual leopards, not the observer’s skill.

6. Other technologies, skills or innovations

• VHF radio collars (about 250 g) and daily radio tracking, with GPS fixes or radio triangulation.
• Camera stations with paired 35 mm film cameras, mounted 20 to 40 cm above the ground and checked every four days.
• Identification of individual leopards from their coat patterns in photographs.
• Program CAPTURE for capture and recapture estimates, and ArcView for working out sampled areas.
• Safari guides reporting sightings by two-way radio as a source of information on uncollared leopards.

7. Implications for conservation (from the paper)

• Reliable population estimates of large carnivores are essential for conservation and management.
• Management agencies often see leopards as too difficult or costly to count, even when setting trophy hunting quotas, so leopard numbers have mostly been guesswork or based on overly simple models. Camera-trap surveys offer a practical, affordable alternative, even in small areas.
• Intensive radio telemetry is the most accurate method but too expensive if density is the only goal.
• The authors hope this first test of camera trapping against a censused population of large cats will encourage wider use of the technique.