# Predicting carnivore occurrence with noninvasive surveys and occupancy modeling

> Can dogs that sniff out scat help map where elusive carnivores live? Across Vermont, detection dog teams, cameras and hair snares surveyed 168 sites for black bears, fishers and bobcats. Bears turned out to follow forest and avoid development, fishers were found almost everywhere but were hard to predict, and bobcats were too rarely detected to model with confidence.

Summary by Kersey Lawrence, Original Wisdom. Updated 9 November 2026. Page: https://www.originalwisdom.com/tracking-literature/2011-long-donovan-predicting-carnivore-occurrence-noninvasive/

## Citation and access

- **Authors:** Robert A. Long, Therese M. Donovan, Paula MacKay, William J. Zielinski and Jeffrey S. Buzas
- **Year published:** 2011
- **Journal or source:** Landscape Ecology
- **Full citation:** Long, R. A., Donovan, T. M., MacKay, P., Zielinski, W. J. and Buzas, J. S. 2011. Predicting carnivore occurrence with noninvasive surveys and occupancy modeling. Landscape Ecology 26(3): 327–340.
- **DOI:** 10.1007/s10980-010-9547-1
- **Access status:** Hosted (freely available elsewhere)
- **PDF:** https://www.originalwisdom.com/wp-content/uploads/2026/11/2011_Long_and_Donovan_etal_predicting-carnivore-occurrence-with-noninvasive-surveys-and-occupancy-modeling.pdf

## Tags

- **Species:** Mammals > American black bear; Mammals > Bobcat; Mammals > Fisher
- **Region:** North America > United States
- **Tracking:** Hair; Scat; Sign surveys; Tracking; Tracks and signs identification
- **Behaviour and ecology:** Behaviour; Habitat selection
- **Methods and technology:** Akaike Information Criterion (AIC); Bootstrapping; Camera traps; Detectability; Distribution mapping; DNA; Geographic Information System (GIS); Hair sampling; Occupancy modelling; Terrain ruggedness index
- **Data type:** New field data
- **Observer qualification:** Not reported
- **Discipline:** Behaviour and ecology; Wildlife tracking
- **Year published:** 2011

## Detailed keywords

- **Species:** Studied: American black bear (Ursus americanus), fisher (Martes pennanti) and bobcat (Lynx rufus). Mentioned as possible bobcat prey: snowshoe hare (Lepus americanus) and squirrels (Sciuridae). Scat detection dogs were the main survey tool.
- **Location and habitat:** The whole state of Vermont (24,963 km²), United States, plus a few sites just west of central Vermont’s border in New York. Elevation 30 m (Lake Champlain shore) to 1,339 m (Mount Mansfield), mean 370 m. Mostly hardwood forest of sugar maple (Acer saccharum), yellow birch (Betula alleghaniensis), paper birch (B. papyrifera) and American beech (Fagus grandifolia); about 79% forested by 1980 after heavy clearing in the 1800s. Human density from 3.7 people per km² (Essex County) to 91 per km² (Chittenden County). Sites on state, U.S. Forest Service, University of Vermont and private land.
- **Topics:** noninvasive surveys of elusive carnivores; scat detection dogs; species distribution and occurrence mapping; imperfect detection (false negatives); habitat and landscape variables (forest cover, human development, roads, wetlands, conserved land, ruggedness, snowfall, latitude); scale (1-km and 5-km radius landscapes); land-use planning; seasonal limits of summer-only surveys.
- **Statistics or models:** Single-season occupancy models (MacKenzie et al. 2002) in program MARK, with detection covariates from earlier work (Long et al. 2007a); univariate screening of 28 to 31 models, then 127 candidate models per species; small-sample Akaike’s Information Criterion (AICc) and Akaike weights; model averaging over the 95% confidence set; goodness of fit by parametric bootstrap (10,000 datasets, Pearson chi-square) in program PRESENCE; receiver operating characteristic (ROC) curves and area under the curve (AUC) for black bear and fisher models; predictive maps in ArcGIS. Inconsistencies noticed: (1) the text says occupancy estimates were 11.5% to 20.7% greater than naive estimates, but for fishers the difference is 21.5 points (83.4% versus 61.9%); (2) the mapping section says each “30-km × 30-km grid cell”, which looks like a typo for 30 m; (3) the Discussion calls the fisher models less informative “than those for black bears and bobcats”, yet bobcat models could not be validated; (4) Table 3 lists “%WETLAND” without a scale for black bears, and Tables 1 and 3 abbreviate the commercial variable differently (%COMMMERCIAL, %COMIND-5K).
- **Data type:** New field data: detection dog, camera and hair snare surveys at 168 sites, May to August 2003 and 2004 (the same surveys reported in Long et al. 2007a and 2007b, here analysed for occurrence and mapped). Land cover, road and other landscape data from existing GIS layers.

## Synopsis

### 1. What they did

Carnivores roam widely and live at low densities, which makes them hard to study. The researchers wanted maps that show where black bears, fishers and bobcats are likely to live across the whole of Vermont, to help wildlife managers and land-use planners. Their aim was good prediction rather than testing ideas about why the animals choose certain habitats.

- Choosing sites. Using a Geographic Information System (GIS), they picked 168 random sites at least 5 km apart (a few closer where access or terrain forced it), covering all major cover types and a wide range of human disturbance, elevation and terrain.

- Surveying with dogs. In May to August of 2003 and 2004, five scat detection dog teams (dog, handler and orienteer) searched a 2-km diamond-shaped route at each site, collecting scats inside the diamond or up to 100 m outside it. Some sites were searched two or three times. DNA analysis confirmed the species for a subset of scats and was used to reduce false detections.

- Cameras and hair snares. At 74 of the sites, they also set baited remote cameras and hair snare rub pads for 14 days. A species counted as detected from an identifiable photo, from large amounts of obvious black bear hair, or from hair confirmed by genetic analysis.

- Modelling. They built occupancy models, which estimate both the chance that a species lives at a site and the chance that the survey finds it if it is there. They tested forest, development, road, wetland and other landscape variables within 1 km and 5 km of each site, averaged the best models, checked their accuracy, and mapped predicted occurrence across the state.

### 2. What they found

Detections

- Detection dog teams made 220 surveys of 168 sites: 135 sites once, 14 twice and 19 three times.

- Black bears were detected at 101 sites (60.1%), fishers at 104 (61.9%) and bobcats at 20 (11.9%).

- After allowing for missed detections, estimated occupancy was 71.6% for black bears, 83.4% for fishers and 32.6% for bobcats.

Black bears

- More forest within 5 km made bears more likely to be present; more developed land within 5 km made them less likely.

- The model was accurate (AUC 0.90). At a cutoff of 0.5, it correctly predicted 95% of sites where bears were found and 69% of sites where they were not.

- Bears were predicted across most of the state, except the Champlain Valley, far western Vermont and areas within 5 to 15 km of large towns.

Fishers

- No single habitat or landscape variable clearly explained where fishers occurred.

- The model was weaker (AUC 0.65): it predicted fisher presence well (97%) but absence poorly (19%), though still better than chance.

- Fishers were predicted across most of Vermont, apart from the Champlain Valley and a pocket of large wetlands in the far northeast.

Bobcats

- No variable stood out, but more mixed forest and, to a lesser extent, more forested wetland within 1 km may be linked to bobcat presence.

- Bobcats were hard to detect (combined detection probability only 0.670, and 0.27 or less per single survey), so the bobcat model could not be checked for accuracy.

- Predicted bobcat areas were patchier, highest in the east-central and northeastern parts of the state.

### 3. What they concluded

- Combining noninvasive surveys with occupancy modelling produced statewide occurrence models, which worked well for black bears.

- Black bears depend on forest and may avoid human activity, which the development variable probably reflected.

- For fishers, the models may have lacked the right variables, fishers may vary in their needs across such a large area, or their needs may be too general to predict well. Modelling at a more local scale may work better.

- Bobcats may choose areas for their prey (such as snowshoe hares and squirrels) more than for land cover. Low detection, and most sites being surveyed only once or twice, made bobcat modelling difficult; at least three surveys per site are recommended.

- The surveys were in summer only, so the maps apply to summer. Bears, fishers and bobcats all change habitat use with the seasons and need special sites for denning and raising young, which this study did not model.

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

Tracks and signs were the core of the survey, but found by dogs rather than people. Scat detection dogs located scats along each route, and these were assigned to species, with DNA analysis confirming the species for a subset and giving a “probability of correct identification”. At hair snares, large amounts of obvious black bear hair were identified by their appearance; other hair was identified genetically. No footprints or trailing were used. Details of how scats were identified are in an earlier paper (Long et al. 2007a), not this one.

### 5. Tracker or observer certification

Not reported. Five detection dog teams did the surveys, but the paper does not describe the handlers’ or orienteers’ experience or how the dogs were trained (it refers to Long et al. 2007a for this). DNA checks on a subset of scats were used to limit wrong identifications.

### 6. Other technologies, skills or innovations

- Scat detection dogs working with a handler and an orienteer along fixed 2-km diamond routes.

- Remote cameras and hair snare rub pads, baited with commercial carnivore lure, a fish-pellet and molasses bag for bears, and (in 2004) raw chicken for fishers.

- Genetic analysis of scats and hair to confirm species.

- Occupancy models combining several survey methods in one detection history, model averaging, and statewide prediction maps from GIS layers.

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

- Occupancy models built from noninvasive survey data can show where carnivores live across whole regions, without capturing or directly watching the animals.

- Land-use planning across large regions, and work on landscape connectivity and barriers to movement, need accurate distribution maps; the models give managers and planners a tool for this in Vermont.

- The maps apply only to summer; seasonal and denning needs require separate study.

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

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

- The study rests entirely on signs (scat and hair) rather than sightings, showing how sign surveys can feed statewide distribution maps when detection is modelled.

- The bobcat results show how much a rarely detected species depends on repeated surveys: one or two visits per site were not enough to model them reliably.

## Cite this summary

Lawrence, K. 2026. Summary of: Predicting carnivore occurrence with noninvasive surveys and occupancy modeling (Long and Donovan et al., 2011). Original Wisdom, Tracking Literature. https://www.originalwisdom.com/tracking-literature/2011-long-donovan-predicting-carnivore-occurrence-noninvasive/