Dataset DOI: 10.5061/dryad.jq2bvq8sj

Repository files

• README.md - overview of the repository, file structure, and instructions for reproducing the analyses.
• 1_fit_reported_models.R - analysis script containing the model specifications and code used to reproduce the reported analyses.
• wolverine_SCR_data_dryad.RData - archived analysis workspace containing the data to reproduce the reported analyses.

This repository contains the analysis data, fitted spatial capture - recapture (SCR) models, and R code used to reproduce the principal analyses reported in the published article:

"Sharing the snow - outdoor recreation and roads reduce density of a rare carnivore."

The archive is intended to allow users to reproduce the reported closed-population SCR model-selection tables and covariate-density relationship plots directly from the supplied fitted models, and to refit the reported closed- and open-population models from the archived analysis data if desired.

Analysis script

1_fit_reported_models.R contains seven sections:

1. Combined-sex closed-population SCR models
2. Female closed-population SCR models
3. Male closed-population SCR models
4. AICc tables for the reported closed-population SCR models
5. Post hoc sensitivity analysis of trap-level recreation and road-density effects on detection
6. Open-population SCR models fitted with openCR
7. Covariate-density relationship plots (Figure 3)

The script documents and can refit the 33 closed-population SCR models appearing in the final model-selection tables: 9 combined-sex, 6 female, and 18 male models. These fitted models are also supplied directly in the .RData archive, so Sections 1 to 3 do not need to be rerun in order to reproduce the published AICc tables. Section 4 can be run immediately after loading the archive.

Sections 5 and 6 provide the code and archived input data required to refit the post hoc detection-sensitivity models and open-population SCR models. These fitted models are not stored in the .RData file.

Section 7 reproduces the marginal covariate-density relationships shown in Figure 3. The spatial objects required to reconstruct the original-unit x-axes are included in the .RData archive, so separate raw TIFF or shapefile inputs are not required.

Required R packages

library(secr)
library(openCR)
library(here)
library(raster)
library(sf)
library(dplyr)
library(ggplot2)
library(patchwork)

Load the archived analysis objects with:

load(here::here("data", "wolverine_SCR_data_dryad.RData"))

Objects stored in wolverine_SCR_data_dryad.RData

The archive contains capture-history objects, habitat masks, spatial objects, and fitted closed-population SCR models. The capture-history and habitat-mask objects are native secr objects rather than conventional rectangular data frames. Variable definitions are given below so that the archived data can be interpreted without prior familiarity with secr.

Capture-history structure and common variables

msCH, msCH_F, msCH_M, and msCH_trapcovs are three-session (2018, 2019, 2020) multisession secr::capthist objects using proximity detectors. A proximity detector records whether an identified individual was detected at a detector during an occasion; individuals are not physically captured or removed.

The three sampling occasions within each annual session are the temporal sampling periods used in the SCR analyses. Detector usage is coded by occasion, with 1 indicating that a detector was operating and 0 indicating that it was not operating. Detector coordinates are projected coordinates in metres in UTM Zone 11N.

The individual identifier is stored as the row name of the individual-covariate table. Identifiers are unique labels assigned to individual wolverines for the analysis.

The individual-level covariate is:

• SEX --- sex of the identified wolverine. F = female; M = male; NA = sex unknown. The sex-specific capture histories contain only the relevant known-sex individuals.

Detector-level covariates common to the capture-history objects are:

• PARK --- binary indicator of whether the detector was located within a provincial (British Columbia: Class A Parks; Alberta: (Wildland) Provincial Parks, Ecological Reserves and Wilderness Areas) or national park: 1 = within a park; 0 = outside a park.
• YEAR --- numeric annual-session index used in construction of the capture histories (1 = 2018, 2 = 2019, 3 = 2020).

The annual detector numbers in msCH are 78 (2018), 152 (2019), and 130 (2020). The combined-sex capture histories contain 61, 128, and 150 detected individuals, respectively; individuals with unknown sex are retained in msCH but excluded from the sex-specific objects.

Closed-population SCR capture histories

• msCH --- combined-sex multisession capture history.
• msCH_F --- female-only multisession capture history.
• msCH_M --- male-only multisession capture history.
• msCH_trapcovs --- copy of the combined-sex capture history containing additional detector-level covariates used in the post hoc detection-sensitivity analysis:
  • COI_01--- detector-level recreation covariate used in the sensitivity analysis. Values are on the processed/model scale used in the analysis rather than raw recreation index units. Derived from the model scale binary COI raster (COI_01_5 below) value at detector locations.
  • ROADDENS --- detector-level road-density covariate. Values are on the processed/model scale used in the analysis rather than raw road-density units. Derived from the model scale road density raster (ROADDENS_5) value at detector locations.

Open-population SCR capture histories

• msCH_openCR_B --- combined-sex capture history used by openCR.

• msCH_openCR_F --- female capture history used by openCR.

• msCH_openCR_M --- male capture history used by openCR.

  All capture histories had two detector-level covariates:

  • PARK --- binary indicator of whether the detector was located within a provincial (British Columbia: Class A Parks; Alberta: (Wildland) Provincial Parks, Ecological Reserves and Wilderness Areas) or national park: 1 = within a park; 0 = outside a park.
  • MR --- mountain-range category of the detector location (C = Columbia Mountains, R = Canadian Rocky Mountains).

These are the capture-history inputs required to refit the open-population analyses in Section 6 of 1_fit_reported_models.R.

Habitat masks

• habmaskFOREST1500_b --- multisession habitat mask used for the closed-population SCR analyses and Figure 3 predictions. Mask points represent possible wolverine activity-centre locations and contain the spatial covariates described below. Mask spacing is 1500 m.
• habmaskALL4000 --- habitat mask used for the open-population SCR analyses. No habitat covariates are stored in this mask. Mask spacing is 4000 m.

For covariates in habmaskFOREST1500_b, suffixes _5 and _10 denote spatial summaries using 5-km and 10-km radii, respectively, except where a number is part of the variable name itself (for example, FOREST_5Y denotes a five-year time period). Continuous covariates used for modelling are stored on the processed/model scale used in the analysis; therefore their values should not generally be interpreted as original measurement units. The archived original-unit spatial objects used for Figure 3 are described separately below.

Covariate processing: Suffixes _5 and _10 indicate covariates summarized using 5-km and 10-km spatial moving window analysis respectively. “Original units” describe covariates before standardization to mean 0 and standard deviation 1, and subsequent moving window analysis. For variables derived from binary source rasters, “Source data” describes the original 0/1 raster; after centering and subsequent moving-window summarization these covariates are all continuous rather than binary.

Recreation covariates

• COI_100MBUFF_5, COI_100MBUFF_10 --- buffered binary Cumulative Outdoor Activity Index (COI; 100-m buffer). Source data: binary indicator (0/1) of recreation presence.
• COI_01_5, COI_01_10 --- binary COI. Source data: binary indicator (0/1) of recreation presence.
• COI_5, COI_10 --- continuous COI intensity. Original units: unitless recreation-intensity index.
• SLED_5, SLED_10 --- snowmobile recreation covariate. Source data: binary indicator (0/1) of snowmobile recreation presence.
• REC_5, REC_10 --- combined binary COI and snowmobile-recreation covariate. Source data: binary indicator (0/1) of recreation presence.

Natural habitat and prey covariates

• MARMOT_K_5, MARMOT_K_10 --- TRIM-landforms derived marmot habitat covariate, including aspects from 135 to 185 degrees. Source data: binary indicator (0/1) of marmot habitat.
• MARMOT_T_5, MARMOT_T_10 --- TRIM-landforms derived marmot habitat covariate, including aspects from 90 to 185 degrees. Source data: binary indicator (0/1) of marmot habitat.
• MARMOT_A_5, MARMOT_A_10 --- TRIM-landforms derived marmot habitat covariate, including all aspects. Source data: binary indicator (0/1) of marmot habitat.
• CARIBOU_5, CARIBOU_10 --- caribou density covariate. Original units: caribou per 1,000 km².
• SNOW_5, SNOW_10 --- persistent spring snow cover. Original units: number of years with persistent spring snow cover out of 17 years (range 0–17).
• MTNRANGE --- mountain sub-area category distinguishing the Columbia Mountains and Canadian Rocky Mountains. Categorical variable: C = Columbia Mountains; R = Canadian Rocky Mountains.

Roads and other human-footprint covariates

• DISTROADS_5, DISTROADS_10 --- distance-to-roads covariate. Original units: metres (m).
• ROADDENS_5, ROADDENS_10 --- road-density covariate. Original units: kilometres of road per square kilometre (km/km²).
• IFL_5, IFL_10 --- intact forest landscape covariate. Source data: binary indicator (0/1) of intact forest landscape.
• NIGHTLIGHT_5, NIGHTLIGHT_10 --- nighttime-light covariate. Original units: unitless nighttime-light index.

Forestry covariates

• FOREST_15Y_5, FOREST_15Y_10 --- clearcuts during the previous 15 years. Source data: binary indicator (0/1) of clearcut presence.
• FOREST_10Y_5, FOREST_10Y_10 --- clearcuts during the previous 10 years. Source data: binary indicator (0/1) of clearcut presence.
• FOREST_5Y_5, FOREST_5Y_10 --- clearcuts during the previous 5 years. Source data: binary indicator (0/1) of clearcut presence.

Trapping covariates

• TRAPPING_10Y_5, TRAPPING_10Y_10 --- trapping during the previous 10 years. Original units: number of wolverines trapped per 1,000 km² of trapline area
• TRAPPING_5Y_5, TRAPPING_5Y_10 --- trapping during the previous 5 years. Original units: number of wolverines trapped per 1,000 km² of trapline area

Covariate names in the habitat mask

Spatial objects used to reproduce Figure 3

These objects retain the spatial information needed to reconstruct the original-unit x-axes used for the marginal covariate--density relationships in Figure 3.

• ROADDENS_raw --- road-density RasterLayer; 250 × 250 m cells, values 0--19.6435 in the archived raster, projected in UTM Zone 11N. Used to reconstruct road density in original units (km/km2).
• SNOW_aligned --- persistent spring-snow RasterLayer aligned to the study-area analysis extent; cell size approximately 392 × 210 m, values 0--17, projected in UTM Zone 11N. The aligned raster is supplied because the original source raster is substantially larger.
• COI_aligned --- aligned binary recreation/COI RasterLayer; 19.1 × 19.1 m cells, values 0 or 1, projected in UTM Zone 11N.
• COI_agg --- spatially aggregated binary COI RasterLayer; approximately 1508.9 × 1508.9 m cells, values 0 or 1, projected in UTM Zone 11N. Used to reconstruct the original-unit recreation x-axis.
• shape_data --- one-feature study-area polygon stored as an sf object in WGS 84 / UTM Zone 11N.

The active Figure 3 workflow uses ROADDENS_raw, SNOW_aligned, and COI_agg and applies the moving-window procedure used in the analysis.

Archived fitted closed-population SCR models

The archive contains 33 fitted secr model objects corresponding to the final manuscript candidate sets: 9 combined-sex models, 6 female models, and 18 male models. These are fitted model objects rather than tabular datasets; their names encode the density covariates included in each candidate model. Exact model formulas are documented in MODEL_REFERENCE.md and 1_fit_reported_models.R.

Combined-sex models

Female models

Male models

Reproducing the main results

Closed-population SCR model-selection tables

Sections 1 to 3 document the full reported model specifications and allow refitting from the archived capture histories and habitat mask if desired. Section 4 reproduces the final AICc tables without refitting the closed-population models:

library(secr)
library(here)

load(here::here("data", "wolverine_SCR_data_dryad.RData"))

# Run Section 4 of 1_fit_reported_models.R

Post hoc detection-sensitivity analyses

Run Section 5. These models use msCH_trapcovs to assess whether the observed associations of recreation or road density with density could instead be explained by spatial heterogeneity in baseline detection probability (g0).

Open-population SCR analyses

Run Section 6. These models are refitted from msCH_openCR_B, msCH_openCR_F, and msCH_openCR_M, using habmaskALL4000.

Figure 3 covariate--density relationships

Run Section 7. This section uses the archived top combined-sex, female, and male density models together with the archived spatial covariates to recreate the marginal density relationships for recreation, road density, and spring snow.

Notes on model fitting

Model fitting is computationally intensive. The closed-population models in Sections 1 to 3 are fitted with list.secr.fit() and can be run in parallel by changing:

ncores <- 4

to an appropriate number of workers for the local machine.

The archive contains the final reported closed-population candidate models rather than every exploratory or preliminary model evaluated during model development. The supplied code therefore reproduces the reported model-selection sets and principal analyses, not the complete exploratory model-screening workflow.

Density surfaces

Spatial density surfaces can be generated from the archived fitted secr models and habmaskFOREST1500_b using standard secr prediction functions. Separate precomputed density-surface rasters are not included because they can be generated directly from the archived models and habitat mask.

File portability

The spatial objects required by the active analysis code are stored directly in wolverine_SCR_data_dryad.RData. Consequently, the active workflow does not depend on machine-specific paths to the original road-density, snow, recreation, or study-area GIS files.

The commented raw-file paths in Section 7 document the original GIS inputs but are not required to reproduce Figure 3 from the Dryad archive.

Citation

Please cite the associated manuscript when using these data or code:

Sharing the snow -outdoor recreation and roads reduce density of a rare carnivore.

The full bibliographic citation and DOI will be added once available.