Cascadia
Analysis Lab

Density Estimation & Spatial Capture-Recapture

Classical mark-recapture estimates detection probability but not density — it requires an assumed study area to convert counts to animals per unit area. Spatial capture-recapture solves this by modeling each individual's activity center and detection probability as a joint function of distance from sampling locations and landscape features. The result: an absolute density estimate in animals per km², free from arbitrary boundary assumptions.

Our Bayesian SCR implementations accommodate individual heterogeneity, sex-specific behavior, seasonal variation, and landscape permeability. We have estimated densities for wolverine, Pacific marten, fisher, mountain lion, wolf, black bear, and a growing list of Pacific Northwest species — combining DNA-based individual ID from the Cascadia Genetics Laboratory with SCR to produce estimates that are often impossible to obtain any other way.

Spatial Ecology & Animal Movement

Where animals go, and why, reveals the functional landscape they inhabit. We analyze GPS telemetry using continuous-time movement models (ctmm), autocorrelated kernel density estimators (AKDE), and integrated step selection functions — characterizing home range dynamics, habitat selection, dispersal, and connectivity at the individual and population level.

Our spatial ecology work integrates LiDAR-derived forest structure data, allowing us to link fine-scale canopy characteristics to wildlife space use at resolutions not possible with conventional remote sensing. This combination is especially powerful for forest-dependent species whose distributions and movements track structural features of old-growth and late-successional habitats.

Species Distribution Models

Where a species occurs, and why, is a different question from how many there are — and often the more consequential one for planning. We build species distribution and occupancy models that link detections to climate, topography, and fine-scale forest structure, then use them to project distributions across landscapes and under changing conditions.

This work draws on unusually deep bench strength. Matt Betts brings decades of forest landscape ecology and long-term bird monitoring, including the habitat-fragmentation and old-growth questions that make distribution modeling difficult. Rebecca Hutchinson works on the statistical and machine learning methods themselves — how to fit these models when detection is imperfect, sampling is opportunistic, and the data were never collected with modeling in mind.

Increasingly we fit joint species distribution models, which estimate many species at once and let information from common species inform inference about rare ones, while separating shared environmental responses from residual co-occurrence. These pair naturally with bulk DNA metabarcoding from the Cascadia Genetics Laboratory: a single sequencing run over environmental or scat samples yields community-wide detections across dozens of taxa at once — exactly the multi-species, presence-heavy data these models are built for.