# Using Presence-Absence Data to Build and Test Spatial Habitat Models for the Fisher in the Klamath Region, U.S.A.

> Where do fishers still live in the forests of northern California and southern Oregon? Using footprints on sooted track plates and camera photos from 682 old surveys, plus satellite maps, researchers built a model that correctly predicted fisher detections at almost 80% of new survey stations. Dense forest canopy and large hardwoods stood out, pointing to low- and mid-elevation forests as key to the fisher’s future.

Summary by Kersey Lawrence, Original Wisdom. Updated 9 November 2026. Page: https://www.originalwisdom.com/tracking-literature/1999-carroll-zielinski-presence-absence-data-build/

## Citation and access

- **Authors:** Carlos Carroll, William J. Zielinski and Reed F. Noss
- **Year published:** 1999
- **Journal or source:** Conservation Biology
- **Full citation:** Carroll, C., Zielinski, W. J. and Noss, R. F. 1999. Using presence-absence data to build and test spatial habitat models for the fisher in the Klamath region, U.S.A. Conservation Biology 13(6): 1344–1359.
- **DOI:** 10.1046/j.1523-1739.1999.98364.x
- **Access status:** Hosted (freely available elsewhere)
- **PDF:** https://www.originalwisdom.com/wp-content/uploads/2026/11/1999_Carroll_and_Zielinski_etal_using-presence-absence-data-to-build-and-test-spatial-habitat-models-for-the-fisher.pdf

## Tags

- **Species:** Mammals > Fisher
- **Region:** North America > United States
- **Tracking:** Tracking; Tracks; Tracks and signs identification
- **Behaviour and ecology:** Behaviour; Habitat selection
- **Methods and technology:** Camera traps; Distribution mapping; Geographic Information System (GIS); Logistic regression
- **Data type:** Existing data; Modelling or theoretical; New field data
- **Observer qualification:** Not reported
- **Discipline:** Behaviour and ecology; Wildlife tracking
- **Year published:** 1999

## Detailed keywords

- **Species:** Studied: fisher (Martes pennanti). Present but not studied: black bear (Ursus americanus), which damaged bait stations and made them inoperable on about 15% to 17% of visits. Discussed: Northern Spotted Owl (Strix occidentalis caurina), whose late-successional reserves form one of the land-management categories compared, and American marten (Martes americana), as a species with similar regional patterns. Fisher diet (from the literature, not studied): small to medium-sized mammals, birds and carrion.
- **Location and habitat:** The Klamath region of northwestern California and southwestern Oregon, United States, with adjacent parts of the northern California coast: about 67,000 km², from sea level to 2700 m, with annual rainfall from 3000 mm in the northwest to 500 mm in the southeast. Douglas-fir (Pseudotsuga menziesii) / mixed evergreen-hardwood forest is the most widespread forest type; white fir (Abies concolor) and red fir (Abies magnifica) at higher elevations; pine (Pinus ponderosa and P. sabiniana) and deciduous oak (Quercus garryana and Q. kelloggii) forests to the east; redwood (Sequoia sempervirens) / western hemlock (Tsuga heterophylla) forest with oak woodland to the west. Land categories: wilderness and other protected areas, late-successional reserves, matrix (general forest) lands, redwood parks and private land, including the Hoopa Indian Reservation in the lower Trinity River area (Humboldt County, California). Retrospective survey locations covered about 15,000 km².
- **Topics:** fisher distribution; habitat modelling at several spatial scales (landscape and regional scale versus fine scale); presence-absence (detection and non-detection) survey data; reuse of existing (“retrospective”) agency surveys; model validation with new field surveys; tree canopy closure; large hardwoods (resting and denning cavities, mast); percent conifer; tree size; annual precipitation and regional gradients; spatial autocorrelation; land-management categories and reserve design; source-sink and metapopulation dynamics; habitat linkages; survey design and adaptive monitoring.
- **Statistics or models:** Multiple logistic regression; Spearman rank correlations (univariate screening); generalized additive modelling (GAM) to check curvature; stepwise model fitting (backward and forward); Bayesian (Schwarz) information criterion (BIC) for model comparison; trend surface analysis (polynomial functions of map coordinates); moving-average (moving-window) landscape variables at 1 to 100 km² in a Geographic Information System; standard deviation within the moving window (FOCALSTD) as landscape diversity; Thiessen polygon weighting of unevenly spaced survey points; Cook’s distance for outliers; Clifford-Richardson-Hemon (CRH) correction of effective sample size for spatial autocorrelation; classification tables with an optimal cutpoint; calibration plots. Alpha level 0.10. Small inconsistencies in the paper: the size-class variable is written both “SIZMA” (equation and Table 3) and “SIZEMA” (text); Table 3 prints one p value as “<0:0001”; Table 2’s title says “fishes” for fishers.
- **Data type:** Existing data: 682 survey locations from surveys by land-management agency and tribal staff, 1991 to 1995 (84 surveys shorter than 12 days removed), with satellite (Landsat Thematic Mapper) vegetation layers. New field data: validation surveys at 468 track plate stations in 78 sample units, 1996 to 1997, with ground vegetation plots at each station. Modelling: the habitat model and the predicted distribution map.

## Synopsis

### 1. What they did

The fisher is a forest carnivore, the largest member of the genus Martes. Fishers had disappeared from most of their former range in the western United States, but they were still found in northwestern California. Over most of the region nobody had looked for them, so their distribution was poorly known.

The authors wanted to predict where fishers live across the whole Klamath region, using data that already existed:

- Existing surveys. They gathered results from 682 survey locations, surveyed between 1991 and 1995 by agency or tribal staff, mostly before planned logging or development. These surveys used baited 35 mm cameras, 110 cameras, or sooted track plates (plates coated in soot that record the footprints of visiting animals). Each location was scored as fisher detected or not detected.

- Satellite vegetation maps. In a Geographic Information System (GIS), they measured tree canopy closure, tree size, percent conifer and other features around each site, averaged over “moving windows” of different sizes (1 to 100 km²), as well as precipitation, elevation, roads and map position.

- The model. They built a multiple logistic regression model to predict the chance of detecting a fisher, compared models at different scales, and corrected their tests for spatial autocorrelation (nearby sites being more alike than distant ones).

- Testing it with new surveys (1996 to 1997). They ran systematic surveys at 78 sample units about 10 km apart. Each unit had six enclosed track plate stations, five spaced around a circle of 500 m radius and one in the centre, baited with chicken and checked every other day for eight visits (about 16 days). At each station they also measured the vegetation on the ground.

### 2. What they found

The survey results

- In the existing data, fishers were detected at 174 of 682 locations.

- In the new surveys, fishers were detected at 17.9% of the 468 stations and at 29 of the 78 sample units (37.2%). On average the first detection came after 7.4 days at a sample unit and 9.3 days at a station.

What predicts fishers

- The final model used landscape averages over 10 km² of tree canopy closure, tree size class and percent conifer, plus annual precipitation and position from north to south.

- Fishers were most strongly linked to landscapes with high tree canopy closure. Higher canopy closure had a stronger positive effect in mixed conifer and hardwood landscapes than in purely conifer ones.

- Fisher detection probability was highest in a band centred on the lower Trinity River and decreased to the north and south.

- Larger tree size in the landscape had a generally negative effect in the model, unlike fine-scale studies showing fishers rest in large trees. The authors note this term was hard to interpret because the vegetation variables overlapped.

- Elevation and road density did not add to the model. Models based on the California Wildlife Habitat Relationships (CWHR) habitat ratings for fishers performed poorly at all scales.

How well the model worked

- It classified 80.4% of the original locations correctly, and 78.6% of the new survey stations, using the same cutpoint.

- At the level of whole sample units it was 71.8% correct.

- Sample units with higher predicted probability also had more stations with fisher detections, which the authors had not expected.

- About two-thirds of the variation in fisher distribution remained unexplained.

Ground vegetation at the new stations

- Stations where fishers were detected had larger hardwoods, more conifer and total basal area, and more sheltered (northeast-facing) slopes.

- Adding these plot-level measures did not improve the landscape model. Hardwood diameter was highly significant in both the plot-level and combined models.

Land management

- Among U.S. Forest Service lands, late-successional reserves had a higher average predicted probability (0.128) than wilderness (0.103) or matrix lands (0.087). The redwood parks were high (0.347). Private and other non-public lands were low overall (0.062), with exceptions such as the Hoopa Indian Reservation (0.440).

### 3. What they concluded

- Habitat selection by fishers may be shaped mostly at the home-range scale and above. Regional conservation planning for fishers may therefore be possible without fine-scale data on vegetation or prey.

- Forest cover is the key feature, and it is fairly easy to measure from satellite images. Dense overhead cover may protect fishers from predators, make travel between foraging sites cheaper, give a better microclimate, and support more prey.

- Large hardwoods matter: their cavities are often used for resting and denning, and their mast may support more prey.

- The Douglas-fir / mixed evergreen-hardwood forests of the region may offer fishers an ideal mix: high canopy closure, large wood from conifers, and mast and refuges from hardwoods. These forests recover canopy closure quickly after disturbance, which may help explain why fishers persist here.

- Static habitat models built from presence-absence data are a useful option for species whose population data are very hard to collect. Reusing existing surveys has costs, though: uneven sampling, uneven protocols and errors in the vegetation maps.

### 4. How tracking was or was not used

Fishers were detected by the footprints they left on sooted track plates at baited stations, and by baited cameras. In the existing surveys, the identity of tracks and photographs was verified by co-author W. J. Zielinski. The new validation surveys used enclosed track plate stations only. No trailing was used.

### 5. Tracker or observer certification

Not reported. The paper says that W. J. Zielinski verified the identity of tracks and photographs in the existing survey data, but it gives no qualification or test of track identification skill. It does not say who identified the tracks in the validation surveys; the field team is thanked by name.

### 6. Other technologies, skills or innovations

- Satellite imagery (Landsat Thematic Mapper) and a GIS vegetation layer, with moving-window averages to describe landscapes at several scales.

- The Clifford-Richardson-Hemon (CRH) method to correct significance tests for spatial autocorrelation.

- A systematic-cluster survey design based on the National Forest Inventory hexagonal grid, which also allows landscape and plot-level habitat use to be compared.

- Track plate stations in plastic or plywood enclosures, baited with chicken, with a commercial trapping lure added on the fourth visit at units with no fisher detection yet.

### 7. Implications for conservation (from the paper)

- More protection for biologically productive, low- to mid-elevation forests is important for the long-term survival of fisher populations.

- Late-successional reserves hold more fisher habitat than the strictly protected areas, apart from the redwood parks. Because little low-elevation forest lies in protected areas, fisher conservation may depend on cooperative management across ownerships.

- Western fisher populations are isolated from each other and from populations in Canada and the eastern United States, and there are gaps within the study region, especially on private coastal land with short logging rotations.

- Keeping fishers well distributed may need more canopy closure and the retention of large hardwoods on managed lands, especially in likely habitat linkages. Planning for non-federal lands should prioritize surveys of the potential habitat identified by regional models.

- The model can point future surveys to unsurveyed areas of likely habitat, and the method could be used for many other species with little demographic data.

### 8. Implications for tracking (Original Wisdom's observations)

_These are Original Wisdom's observations, not the authors'._

- The whole model rests on track plate and camera detections. A fisher counts as present only when it leaves identifiable footprints or a photograph, so the accuracy of track identification feeds directly into the habitat model, and the paper reports the verification step but not how it was checked.

- The study shows how many small, existing track-station surveys done for other purposes can be pooled into a regional picture of where a species lives.

## Cite this summary

Lawrence, K. 2026. Summary of: Using Presence-Absence Data to Build and Test Spatial Habitat Models for the Fisher in the Klamath Region, U.S.A. (Carroll and Zielinski et al., 1999). Original Wisdom, Tracking Literature. https://www.originalwisdom.com/tracking-literature/1999-carroll-zielinski-presence-absence-data-build/