# Where will the cat cross the road? Comparing camera and GPS-based models for identifying wildlife corridors

> Where will a mountain lion cross a highway? On Washington’s Olympic Peninsula, researchers compared corridor maps built from hundreds of trail cameras with maps built from GPS-collared mountain lions. GPS did better at predicting where young dispersing mountain lions travel, but the two approaches agreed most where it matters most, around the busy I-5 highway.

Summary by Kersey Lawrence, Original Wisdom. Updated 5 November 2026. Page: https://www.originalwisdom.com/tracking-literature/2026-barbee-elbroch-where-will-cat-cross/

## Citation and access

- **Authors:** Read Barbee, L. Mark Elbroch, Kimberly Sager-Fradkin, Kristen Phillips, Dylan Bergman, Bethany Ackerman, Shannon Murphie, Andrew Stratton, Caitlin Kupar, Cassandra Sullivan, Glen Kalisz, Sarah N. Sells, Mark Hebblewhite, Hugh Robinson
- **Year published:** 2026
- **Journal or source:** Biological Conservation 322: 112029
- **Full citation:** Barbee, R., Elbroch, L.M., Sager-Fradkin, K., Phillips, K., Bergman, D., Ackerman, B., Murphie, S., Stratton, A., Kupar, C., Sullivan, C., Kalisz, G., Sells, S.N., Hebblewhite, M., and Robinson, H. 2026. Where will the cat cross the road? Comparing camera and GPS-based models for identifying wildlife corridors. Biological Conservation 322: 112029.
- **DOI:** https://doi.org/10.1016/j.biocon.2026.112029
- **Access status:** Open access (CC BY 4.0)
- **PDF:** https://www.originalwisdom.com/wp-content/uploads/2026/11/2026_Barbee_and_Elbroch_etal_where-will-the-cat-cross-the-road.pdf

## Tags

- **Species:** Mammals > Mountain lion
- **Region:** North America > United States
- **Behaviour and ecology:** Behaviour; Connectivity; Dispersal; Habitat selection; Human disturbance; Movement; Road crossings; Wildlife corridors
- **Methods and technology:** Bayesian statistics; Camera traps; Circuitscape; Global Positioning System (GPS) collars; Integrated step selection function; Movement simulation; Occupancy models
- **Data type:** Existing data; Modelling or theoretical
- **Discipline:** Behaviour and ecology
- **Year published:** 2026

## Detailed keywords

- **Species:** Studied: mountain lion (Puma concolor), also called cougar or puma. Camera targets mentioned but not studied: bobcat, coyote, deer, elk, fisher.
- **Location and habitat:** The Olympic Peninsula, the Willapa Hills and adjacent lands west of the Interstate 5 (I-5) highway, Washington State, United States, with predictions extended east to the Cascade Mountains. Temperate rainforest, glaciated mountains and rugged coastline, from 0 to 2,430 m. Includes Olympic National Park, Olympic National Forest, forestry lands, Tribal lands and towns, with dense networks of logging roads and highways.
- **Topics:** landscape connectivity, wildlife corridors, highway crossings, dispersal versus resident movement, habitat fragmentation, human disturbance, comparing camera and GPS methods, cost and invasiveness of methods, multi-species monitoring, Tribal partnerships in research.
- **Statistics or models:** Stacked single-species, single-season occupancy models (Bayesian, R package ubms) with leave-one-out cross-validation. A Circuitscape 5.0 connectivity surface built from occupancy-based resistance. A population-level Poisson integrated step selection function (iSSF) with random slopes for individual animals (R-INLA), used to simulate 200,000 mountain lion paths of about one year each. Validation used Spearman rank correlations over ten equal-area bins. Model agreement was measured with landscape-wide and 1 km moving-window Pearson correlations, and a permutation test. Remote-sensing covariates included the Enhanced Vegetation Index (EVI), Net Primary Productivity (NPP), tree cover, roads (OpenStreetMap), human population and the Human Impact Index.
- **Data type:** Existing long-term data. Camera data from 1,210 sites between 2013 and 2022: the Olympic Cougar Project from 2019, with up to 518 unbaited stations, and the Olympic National Park Fisher Restoration Program from 2013 to 2016, with baited stations. GPS training data: 328,462 locations from 82 mountain lions (2011–2023). Validation data: 193,023 locations from 103 mountain lions (2011–2024). The location data are not public because they include sensitive species locations; the code is on GitHub.

## Synopsis

### 1. What they did

Wildlife corridors are only useful if they sit where animals actually travel. Two common ways to map them are trail-camera surveys and GPS collars. Cameras are cheaper, non-invasive and record many species at once, but give coarser information. GPS collars give detailed movement, but are expensive and require capturing animals. Few studies have compared the two methods using data from the same place and time.

On Washington’s Olympic Peninsula, the authors built two connectivity maps for mountain lions. The first came from camera detections: an occupancy model fed into Circuitscape, which treats animal movement like electrical current flowing across a landscape. The second came from GPS-collared mountain lions: a step selection model was used to simulate 200,000 mountain lions wandering the landscape for about a year each. They then checked both maps against real GPS locations, and compared where the maps agreed. They paid special attention to the I-5 highway corridor, a major barrier.

### 2. What they found

- Both approaches identified the same broad drivers: mountain lions avoided human disturbance (roads, human population, human impact) and favoured areas with more vegetation and prey, plus ambush cover. The step selection model also showed a preference for medium levels of tree cover.

- Both maps predicted real GPS locations well overall (correlations of 1 for the camera-based map and 0.96 for the GPS-based map). The GPS-based map placed more locations in its top half (81% versus 73%).

- The GPS-based map predicted all groups equally well. The camera-based map did well for resident mountain lions but poorly for dispersing young animals (0.37 for females and 0.66 for males), which are more willing to cross poorer habitat.

- Across the whole landscape the two maps agreed only moderately (correlation 0.26). They agreed most in human-dominated areas, including the I-5 corridor, and least in the rugged high mountains.

- Both maps showed similar crossing channels over I-5. These matched two crossings made by GPS-collared dispersing mountain lions during the study, experts’ earlier corridor maps, and priority zones in the state’s connectivity plan.

- One dispersing male swam up to 2.5 km, so islands near the peninsula are within reach of mountain lions.

### 3. What they concluded

- GPS-based models are better when the aim is dispersal, individual movement or fine-scale route choice, and when collaring is feasible.

- Camera-based models may be the better choice for monitoring many species, long-term or large-scale studies, non-invasive sampling and limited budgets, and in heavily developed landscapes where movement options are few.

- Dispersing animals use habitat differently from residents, so models built only on residents may miss dispersal routes. Planners face a choice between supporting residents in safe habitat, supporting dispersers on the move, or both.

### 6. Other technologies, skills or innovations

- GPS collars (Lotek LiteTrack Iridium; Vectronic GlobalStar and VERTEX Lite) recording every 1 to 4 hours

- Browning trail cameras, 46 cm above ground, facing north, unbaited (Olympic Cougar Project); baited stations with hair-collection boxes (Fisher Restoration Program)

- Satellite-derived vegetation and land-cover layers built in Google Earth Engine and QGIS

- Circuitscape run in Julia; R packages ubms, INLA, amt and terra

- Partnership between Panthera, six Tribal nations and the Washington State Department of Transportation

- The authors state they used Claude.ai and ChatGPT to help with the brevity and flow of the writing

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

- Corridor planning depends on which animals it serves: residents, dispersers, or both. Models that ignore dispersers may misplace corridors.

- The two methods agreed most in the developed areas where corridor decisions matter most, such as highway crossings, so camera-based maps can still guide planning there.

- The authors call for more research on how individuals within a species differ in habitat selection, and on building that into connectivity models.

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

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

- Neither method shows exactly where an animal stepped onto a road. Tracks and signs at predicted crossing points, such as road verges, culverts and underpasses, are a low-cost, non-invasive way to check corridor maps on the ground, and like cameras they record many species at once.

## Cite this summary

Lawrence, K. 2026. Summary of: Where will the cat cross the road? Comparing camera and GPS-based models for identifying wildlife corridors (Barbee and Elbroch et al., 2026). Original Wisdom, Tracking Literature. https://www.originalwisdom.com/tracking-literature/2026-barbee-elbroch-where-will-cat-cross/