Power of Track Surveys to Detect Changes in Cougar Populations

1. What they did

Cougars are active at night, secretive and live at low densities, so changes in their numbers are hard to follow. Radio collaring many animals can give a population estimate, but it is costly and covers only a small area. Managers often use hunter kills, livestock losses or track surveys instead. The authors asked how many track transects would be needed to reliably detect a 30% or 50% change in cougar numbers between two surveys.

• Field data. They used existing data from 28 dry washes in the Aravaipa-Klondyke area of southeastern Arizona, walked in four periods between October 1991 and April 1993. Washes were chosen because tracks could only be seen in dust, sand or mud, and wind and traffic quickly wiped out tracks on dirt roads. A survey started only after 5 days without rain.
• The survey. One trained observer started at first light at the top of each wash and walked downstream for about 4 hours. For each 0.5-km segment, the observer counted cougar scats and track sets (a continuous set of tracks apparently made by one animal) and estimated how many different cougars were present.
• What they analysed. They reduced this to whether each segment had cougar tracks or not, and counted the segments with tracks on an 8-km transect (“track density”). They did not use scat counts, track set counts or estimated numbers of animals, because presence or absence depends less on the observer’s skill, and a high count usually meant a cougar had a kill nearby, not that more cougars were there.
• Simulations. Because track density followed a Poisson distribution, they simulated pairs of surveys and tested how statistical power changed with the number of transects (30 to 270), the accepted risk of a false alarm (α of 0.05, 0.10 or 0.20), the size of the change (30% or 50%), whether numbers rose or fell, and whether the change was even across transects or confined to half of them. They also tested a 10% decrease with 300 to 1,000 transects.

2. What they found

The field surveys

• On average, 7% of the 0.5-km segments had cougar tracks (mean 1.14 segments per 8-km transect).
• Tracks were rarely found in the first kilometre of a wash, where there was no well-developed scour zone with sand. In the lowest parts, cattle tracks tended to wipe out cougar tracks.
• More segments had tracks in the first period (12%) than later; only the drop from period 1 to period 2 (4%) was statistically significant.
• Track density followed a Poisson distribution, not a normal one, and a transect’s count did not predict its next count.

The simulations

• More transects gave more power. To reach 80% power for large changes needed from about 30 transects (a 50% decrease at α = 0.20) to about 190 transects (a 30% increase at α = 0.05).
• Decreases were easier to detect than increases.
• Detecting a 30% decline at α = 0.05 needed about 140 transects; at α = 0.10, about 120; at α = 0.20, about 90. A 50% decline needed only about 50 transects at α = 0.05.
• A small (10%) decrease was out of reach: 400 transects gave only 48% power, and 700 were needed for 80%.
• Most power curves levelled off at 75% to 90% power, so gains beyond that cost many more transects.
• When all the change happened on half the transects, power fell by 2% or less.

3. What they concluded

• About 30 to 190 eight-kilometre transects in dry washes would be needed to detect a 30% to 50% change in a cougar population. Track surveys will usually not detect small yearly changes, but may reveal large changes more efficiently than other methods.
• Even so, the authors believe track surveys are better than the alternatives: unlike hunter kills and livestock losses, they allow statistically valid conclusions, and they are cheap, easy to repeat, easy to standardise between observers, and do not disturb the animals.
• They suggest 80% power as a reasonable goal, and that an α above 0.05 is reasonable in most management situations: 0.10 to 0.20 for threatened populations, and 0.05 to 0.10 for hunted populations.
• Their results hold for any track counts that follow a Poisson distribution, and managers can easily check whether their own data do.
• Transects should be spread through the whole area used by the population, including marginal habitat, rather than only where cougars are expected to be common, and placed where cougars are most likely to travel and their tracks most likely to be visible.

4. How tracking was or was not used

The field data came from foot surveys of dry washes, in which an observer identified cougar tracks and scats in each 0.5-km segment, counted track sets, and estimated the number of individuals. The analysis used only whether each segment had cougar tracks. Track sets that clearly continued from the previous segment were not counted again. The paper also notes how substrate (sand, dust or mud) and cattle trampling affected where tracks could be found.

5. Tracker or observer certification

Reported but not validated. Each transect was walked by “a single trained observer”; the training and the observer’s skill were not described or tested. The authors chose presence-absence data partly because they are less affected by differences in observer skill. A. LeCount, J. Yarchin, R. Thompson, M. Peirce, K. Sergent and T. Waddell are thanked for developing the original field study and for help in the field.

6. Other technologies, skills or innovations

• Computer simulation of survey results with Resampling Stats software, using a resampling test that makes no assumptions about how the data are distributed.

7. Implications for conservation (from the paper)

• Surveys meant to catch a decline need less effort than surveys meant to catch an increase.
• For threatened populations, missing a real decline could mean extinction, so a higher risk of false alarm is acceptable if emergency steps would follow.
• If managers accept low power, the authors question whether the survey has any purpose other than making the public think the manager is “doing something.”