When the mean is meaningless: Drivers of spatial behavior in a generalist carnivore

Davis, Emily Clare 1 ; Bjornlie, Daniel2; Kindermann, Ryan3; Thompson, Daniel3; Holbrook, Joseph1

Published Sep 19, 2025 on Dryad. https://doi.org/10.5061/dryad.xksn02vvb

Data files

Sep 19, 2025 version files 57.54 MB

README.md

8.08 KB
RSF_input.csv

57.36 MB
RSF_output.csv

165.58 KB

Abstract

Despite the ecological expression and conservation importance of diverse behavioral tactics in animals, there is often friction associated with conventional analytical approaches and inference concerning variation in spatial behavior. Implicitly or explicitly, population-level inferences are often the main objective of studies, but interpretations can be ambiguous in the presence of divergent behavioral tactics across individuals or cohorts, as with generalist species. We pursued a novel analytical approach and assessed the underlying mechanisms driving variation in spatial behaviors of generalist species using the American black bear (Ursus americanus) as our focal species. We quantified individual variation in habitat selection expressed by black bears using individual models for 35-collared bears across four study areas in Wyoming, USA. We modeled how state-dependent factors (age, sex, δ15Nitrogen and body fat) and resource availability influenced behavioral variation in resource selection. We observed vast variation among individuals, demonstrating patterns consistent with a generalist species. Black bear resource selection differed with changes in state-dependence and resource availability. Specifically, traits uniquely important to black bear success, body fat and carnivory, explained variation in selection for forage indexed by NDVI, forests, and riparian areas. Environmental heterogeneity via differences in resource availability magnified behavioral variation in resource selection by black bears. Selection trends for NDVI and deciduous shrubs were explained by resource availability, indicating black bears exhibited functional responses in habitat selection. These insights emerged from our analytical approach; had we implemented a more conventional, population-level assessment we would have simply concluded that black bears displayed behavioral neutrality with respect to forage resources. Acknowledgement of behavioral variation when considering spatial behavior of generalist species provides a more representative understanding of individuals within a population, and our analytical approach offers a solution to uncovering drivers of individual variation in spatial behavior.

Dataset DOI: 10.5061/dryad.xksn02vvb

Description of the data and file structure

The data derive from American black bears (Ursus americanus) fitted with Iridium GPS collars across four study areas in Wyoming, USA. The goal of the study was to quantify individual variation in habitat selection and evaluate how state-dependent traits (age, sex, body fat, δ¹⁵N as an index of carnivory) and resource availability influenced spatial behaviors.

We implemented a novel individual-level analytical approach to resource selection functions (RSFs), allowing us to identify drivers of behavioral variation that would be obscured by conventional, population-level analyses.

Data Products

Two primary data products are included:

Input Data Frames
- Contain the data used to fit individual RSF models for each collared bear (n = 35).
- Include both scaled and unscaled environmental covariates (e.g., NDVI, land cover, forest type, riparian habitat).
- State-dependent variables (age, sex, δ¹⁵N, body fat) are included where applicable.
Output Data Frames
- Contain the results from all individually fitted RSF models.
- Variables include coefficient estimates, standard errors, and p-values.
- Additional intrinsic and extrinsic variables are incorporated to facilitate subsequent models assessing how traits and resource availability shape individual variation in resource selection.

Notes on Use

Input data are formatted as data frames suitable for RSF analysis.
Output data provide model summaries and can be used for meta-analyses, synthesis, or replication of the manuscript’s results.
All spatial covariates were derived from publicly available datasets (see Methods section of the manuscript for sources and processing details).

Files and variables

File: RSF_output.csv

Description: Results from individually fitted resource selection function (RSF) models for collared black bears. Each row represents the parameter estimates for a covariate within an individual-month model, along with associated intrinsic and extrinsic variables.

Variables

Estimate: Coefficient estimate from the RSF model.
Error: Standard error of the estimate.
z: z-value (Estimate ÷ Standard Error).
P: p-value for the z-test of the parameter estimate.
cov: Covariate name corresponding to the estimate (e.g., NDVI, elevation, forest).
aid: Animal ID; unique identifier for each collared black bear.
month: Calendar month of the model (e.g., June, July).
SA: Study area code (e.g., LR = Laramie; other codes correspond to different Wyoming study areas).
weight: Body weight of the individual (kg).
sex: Sex of the individual (M = male, F = female).
est_age: Estimated age at time of capture (not used in analysis).
lab_age: Laboratory-assigned age from tooth analysis (years).
repro_status: Reproductive status of the individual (e.g., with cubs, male; not used in analysis).
bodyfat: Percent body fat at capture (measured via bioelectrical impedance).
weights: Model weights used in meta-analysis across individuals.
hr.avail: Average value of available resources (e.g., forest cover, NDVI, distance to berry patches) across the individual’s home range. (not used)
sa.avail: Average value of available resources across the broader study area.
N: Nitrogen isotope value, used as an index of carnivory.

File: RSF_input.csv

Description: Input dataset used to fit individual resource selection function (RSF) models for collared black bears. Includes both raw and standardized (z-scored) covariates.

Variables

AID: Animal ID; unique identifier for each collared black bear.
TelemDate: Date and time of GPS fix (YYYY-MM-DD HH:MM:SS).
SA: Study area code (e.g., LR = Laramies, others correspond to different Wyoming study areas).
Month: Calendar month of location fix (e.g., June, July).
elev: Elevation at GPS location (meters above sea level).
hli: Heat Load Index; terrain-derived measure of potential solar radiation.
tpi33: Topographic Position Index c
forest: Proportion or probability of forest cover at location (0–1).
may_ndvi: Normalized Difference Vegetation Index (NDVI) for May, representing vegetation greenness/productivity.
june_ndvi: NDVI for June.
july_ndvi: NDVI for July.
aug_ndvi: NDVI for August.
sept_ndvi: NDVI for September.
dist2rip: Distance to nearest riparian habitat (meters).
dist2berry: Distance to mapped berry-producing shrub patches (meters).
used: Resource selection indicator (1 = used bear GPS location; 0 = random available location).

Standardized (z-scored) variables:

TelemDate.z: Standardized time of GPS fix.
elev.z: Standardized elevation.
hli.z: Standardized Heat Load Index.
tpi33.z: Standardized Topographic Position Index.
forest.z: Standardized forest cover.
may_ndvi.z: Standardized May NDVI.
june_ndvi.z: Standardized June NDVI.
july_ndvi.z: Standardized July NDVI.
aug_ndvi.z: Standardized August NDVI.
sept_ndvi.z: Standardized September NDVI.
dist2rip.z: Standardized distance to riparian habitat.
dist2berry.z: Standardized distance to berry patches.

Code/software

All data files can be viewed and analyzed using the free, open-source software R (version ≥ 4.2.0 recommended). The following R packages were used to process, analyze, and visualize the data:

lme4 – for fitting linear and generalized linear mixed-effects models.
afex – for factorial analyses of mixed designs.
performance – for assessing model quality and diagnostics.
tidyverse – for data manipulation and visualization (dplyr, ggplot2, tibble, etc.).
sf – for handling spatial vector data.
mapview – for interactive visualization of spatial data.
rgdal – for reading and writing spatial data formats.
sp – for handling spatial data and coordinate systems.
raster – for working with raster data.
rlist – for flexible list operations in R.

Workflow:

Raw GPS and environmental data were read into R using read.csv() or spatial data functions from sf and raster.
Covariates and availability measures were calculated for individual home ranges and study areas using sf, raster, and dplyr.
Individual-month resource selection function (RSF) models were fit using lme4. Model diagnostics were assessed with performance.

Access information

Other publicly accessible locations of the data:
NA

Data was derived from the following sources:

LANDFIRE Existing Vegetation Type (EVT; LANDFIRE 2020): Used to generate layers of forested habitat, deciduous shrubs (including berry-producing species), and riparian areas, as well as to calculate distances to these features. Available under the LANDFIRE open data license: https://www.landfire.gov/.
Landsat 8 Surface Reflectance Composites (USGS/NASA): Used to calculate monthly and winter Normalized Difference Vegetation Index (NDVI) for Wyoming from 2015–2021. NDVI layers were used as an index of photosynthetic activity and forage availability. Publicly available via Google Earth Engine.

All derived layers (vegetative, topographic, and NDVI) were processed at 30 m resolution and summarized at 500 × 500 m pixel windows to align with black bear spatial ecology. Topographic indices including topographic position index (TPI) and heat load index (HLI) were calculated from elevation data following McCune & Keon (2002) and De Reu et al. (2013).

Analyses including habitat selection functions (HSFs) and mixed-effects models were conducted in R (≥ 4.2.0) using publicly available packages (lme4, afex, performance, tidyverse, sf, mapview, rgdal, sp, raster, rlist).

When the mean is meaningless: Drivers of spatial behavior in a generalist carnivore

Data files

Abstract

README: When the mean is meaningless: Drivers of spatial behavior in a generalist carnivore

Description of the data and file structure

Data Products

Notes on Use

Files and variables

File: RSF_output.csv

Variables

File: RSF_input.csv

Variables

Code/software

Access information