Monitoring Martes Populations in California: Survey Design and Power Analysis

1. What they did

Fishers and American martens have been protected from trapping in California since 1945 and 1954, yet in parts of their historic ranges their numbers are very low, possibly because of timber harvest. Little was known about their status, because the information came from chance sightings and from scattered surveys done for local timber or recreation projects. The authors set out to design a statewide monitoring program that could detect a decline in where these animals occur.

• The detection method. Baited track plates: a sooted aluminium sheet, partly covered with white paper, inside a plywood box. Bait at the back draws the animal across the soot, so they leave clear footprints on the white surface. The tracks of fishers and martens can be told apart.
• The sampling unit. Eight track-plate stations, 0.6 km apart, in a rough 4 by 2 grid. Stations are checked every 2 days, and the survey at a unit stops as soon as the target species is detected at any station, or after eight visits (16 days). Units are placed at least 10 km apart for fishers and 5 km apart for martens, about twice the diameter of an average male home range, so that the same animals are unlikely to be detected at two units.
• Correcting for animals that were missed. They adapted a bias adjustment first developed for Northern Spotted Owls. It uses the average number of visits before the first detection (“latency to first detection”) to estimate how many occupied units were recorded as empty.
• Computer simulation. They built a simulator called POWER that mimics stratified random sampling in two surveys 10 years apart, and tests whether the proportion of units with detections has fallen. They first explored a wide range of possible values, then used their best estimates for fishers: five regional strata, or ten strata when each region was split into late-successional (old-growth) and other conifer forest.
• Precision. They worked out how many units are needed to estimate each stratum’s proportion within a chosen margin of error, and to compare strata.

2. What they found

Correcting for missed animals

• A table of correction factors lets surveyors adjust the proportion of occupied units. For example, with eight visits and a mean of 3.4 visits to first detection, the proportion is multiplied by 1.2016: if 20 of 100 units had detections, the corrected estimate is 24.

Simulations over a range of values

• The more common the animal (the higher the mean proportion of units with detections), the fewer units are needed. About 55 units per stratum would detect a 20% decline with 80% power when the mean proportion is 0.50, but about 300 per stratum would be needed when it is 0.05.
• Differences between strata (how unevenly the animals are spread) mattered much less than the average. With a mean of 0.20 and 100 units per stratum, power to detect a 20% decline dropped only from 0.74 to 0.70 as the variation rose.
• Detecting a 50% decline needed far fewer units than a 20% decline, sometimes less than half as many.

The fisher example

• Estimated proportions of units with detections: 0.30 in the Klamath and Southern Sierra Nevada, 0.20 in the Sierra–Klamath Transition, 0.10 in the Central Sierra Nevada and 0.05 in the Northern Sierra Nevada (where no surveys had detected fishers). Non-old-growth habitats were given half these values.
• Five strata: about 170 units per stratum (850 in total) to detect a 20% decline, and 25 per stratum for a 50% decline.
• Ten strata: about 115 units per stratum (1150 statewide) for a 20% decline, and 17 per stratum for a 50% decline. That is an average of 82 units for each of California’s 14 national forests and national parks.
• Detecting 5% or 10% declines would take over 10,000 or over 5000 units statewide.
• 115 units per stratum would also be enough to tell apart two strata that differ by at least 13% (for example 15% and 30%), with 80% confidence.

3. What they concluded

• A program able to detect even a fairly large (20%) decline in the track-plate index for fishers needs about 1100 or more sampling units across the fisher’s range in California, but this is not prohibitive.
• Requiring 90% power would at least double the effort in most cases, and testing for both increases and decreases would often need 20 to 50% more units.
• Fishers need urgent conservation measures, and a pilot program able to detect 20% declines should start immediately. Detecting a 20% decline with 80% confidence is probably the most liberal standard that should be set for fishers.
• For martens, which appear well distributed and more abundant, a system that only detects catastrophic (50%) declines may do until more resources are available: about 20 units per stratum (200 statewide).
• Sampling only the best habitat would cut effort but could miss declines in marginal habitats that may make up most of the range, so they recommend against it.
• The units should stay in the same places in every survey.
• Other methods were set aside: mark-recapture with cameras would be too variable and too much work statewide; demographic studies are impractical for fishers; track plates are easier and cheaper than cameras at a large scale; snow is too unpredictable in California for snow tracking; and kill-trapping is ruled out because both species are protected.
• The index measures change in occurrence and distribution, not abundance. Its link to population size depends on animals keeping fixed home ranges and on how few transient animals there are. Surveying in spring and summer, before dispersal, may reduce this problem.
• The same stations can also record other carnivores drawn to bait.

4. How tracking was or was not used

The whole design rests on reading footprints left on baited sooted track plates. A unit counts as occupied when the tracks of the target species are recorded at any of its eight stations. The authors cite earlier work showing that fisher and marten tracks on the plates can be easily told apart, and note that no method yet existed to identify individual animals from their tracks. They considered and set aside snow tracking because snow is too unpredictable in California. No new tracking was done for this paper: the numbers come from earlier track-plate and camera surveys.

5. Tracker or observer certification

Not reported. The paper does not say who would identify the tracks on the plates or how their identifications would be checked. It relies on the earlier finding that fisher and marten tracks are easy to distinguish.

6. Other technologies, skills or innovations

• The POWER simulator, written in Minitab, which estimates statistical power for stratified surveys of presence and absence (available on request from H. Stauffer).
• A table of bias adjustments for 4, 6, 8 and 12 visits, adapted from Northern Spotted Owl surveys.
• A table of sampling errors for different proportions and sample sizes, at 80% and 95% confidence.

7. Implications for conservation (from the paper)

• This was the first proposal for statewide monitoring of fishers or martens using non-lethal methods, and is an example of the planning that will be increasingly needed for carnivores that are no longer trapped or hunted.
• A Type I error rate of 20% was chosen on purpose, because missing a real decline is more costly for the animals than a false alarm.
• Monitoring should be planned with power analysis before it starts, because surveys that cannot detect a decline may let a population disappear before rescue efforts begin.