Abundance-mediated species interactions between coyote, fisher, and marten in Northeastern US

Twining, Joshua1 ; Fuller, Angela1 ; Augustine, Ben2 ; Royle, Andy2

Published Jun 22, 2024 on Dryad. https://doi.org/10.5061/dryad.2bvq83bz2

Data files

Jun 22, 2024 version files 2.40 MB

7dayoccDetectionswhitetaileddeerNZ.csv

37.78 KB
allNY__2013-2021_15kmbuffer_allsitecovs.csv

988.41 KB
allNY_2013_2021_7dayocc_coyote_counts.csv

69.55 KB
allNY_2013-2021_7dayocc_americanmarten_detection_nondetection.csv

128.19 KB
allNY_2013-2021_7dayocc_fisher_counts.csv

85.56 KB
alNY_2013-2021_6kmbuffer_allsitecovs.csv

997.72 KB
juliandays_allNY_2013-2021.csv

79.86 KB
README.md

14.95 KB

Abstract

Ecological theory posits that the strength of interspecific interactions is fundamentally underpinned by the population sizes of the involved species. Nonetheless, contemporary approaches for modelling species interactions predominantly centre around occupancy states. Here, we use simulations to illuminate the inadequacies of modelling species interactions solely as a function of occupancy, as is common practice in ecology. We demonstrate erroneous inference into species interactions due to bias in parameter estimates when considering species occupancy alone. To address this critical issue, we propose, develop, and demonstrate an occupancy-abundance model designed explicitly for modelling abundance-mediated species interactions involving two or more species. When modelling interactions as a function of abundance rather than occupancy, we uncover previously unidentified interactions. Through an empirical case study and comprehensive simulations, we demonstrate the importance of accounting for abundance when modelling species interactions, and we present a statistical framework equipped with MCMC samplers to achieve this paradigm shift in ecological research.

README: Abundance-mediated species interactions

Summary

These are MCMC samplers, data simulators, and processing and run scripts for a occupancy-abundance model for abundance-mediated species interactions with a full example case study. This model framework models detection/non-detection data to estimated occupancy, abundance, and interactions between the species of interest. This model enables the user to apply detection/non-detection data collected over repeat surveys to model interactions as a function of abundance. These samplers are presented in Twining et al. submitted, and are based on adaptations of the Waddle et al. (2010) formulation for modelling species interactions within an occupancy model, but instead of modelling the state model of a subordinate species a function of the occupancy states, it is modelled as a function of abundance (N). We provide a range of MCMC samplers for different ecological scenarios between two or more species including disease- and predator- mediated competition, intraguild predation, mesopredator release, and tri-trophic cascades.

The working directory

Below you fill find descriptions of each folder in this repository and files contained within them.

Models and data simulators (./occupancy_abundance_model/models_and_simulations)

This folder has seven R scripts. Within each script are MCMC samplers and a data simulator for different iterations of the occupancy-abundance model. Model code and implentation is conducted in the nimble package (de Valpine et al. 2022).

1. two_species_occupancy_abundance_data_simulator_and_model.R

This script contains a data simulator and model for simulating a two species occupancy abundance model with spatially varying interaction terms. The code is commented throughout to describe each part of the simulator and script.

2.three_species_occupancy_abundance_abundance_model_and_simulator_pinemarten_greysquirrel_redsquirrel_example.R

This script contains a data simulator and model for simulating a three species occupancy abundance model - modeling the occupancy of the dominant species on abundance of intermediate, and abundance of subordinate, with additional interactions between abundance of intermediae with the subordinate. The code is commented throughout to describe each part of the simulator and script.

3.three_species_abundance_abundance_abundance_model_and_simulator_otter_urchin_kelp_example.R

This script contains a data simulator and model for simulating a three species abundance model - modeling the abundance of the dominant species on abundance of intermediate, and abundnace of subordinate, with additional interactions between abundance of intermediae with the subordinate. The code is commented throughout to describe each part of the simulator and script.

4. three_species_abundance_occupancy_occupancy_model_and_simulator_deer_liverfluke_moose_example.R

This script contains a data simulator and model for simulating a three species abundance model - modeling the abundance of the dominant species on occupancy of intermediate, and occupancy of subordinate, with additional interactions between occupancy of intermediae with the subordinate. The code is commented throughout to describe each part of the simulator and script.

5. occupancy_vs_abundance_mediated_interactions_model_and_simulator.R

This script contains a data simulator and model for simulating a two species occupancy abundance model with and fits models with interactions as a function of occupancy and as a function of abundance. The code is commented throughout to describe each part of the simulator and script.

6. occupancy_abundance_model_three_species_simulator_and_model_variable_detection.R

This script contains a data simulator and model for simulating a three species abundance model - modeling the abundance of the dominant species on abundance of intermediate, and occupancy of subordinate, with additional interactions between abundance of intermediae with the subordinate. The code is commented throughout to describe each part of the simulator and script.

7. two_species_occupancy_abundance_simulator_spatially_varying_interaction_terms

The coyote-fisher-marten case study folder (./occupancy_abundance_model/case_study)

This folder has nine files.

1. alNY_2013-2021_6kmbuffer_allsitecovs.csv

This file contains all of the summarized spatial covariate data at a 6km scale used in the analysis. Each row is a different 6km 2 pixel in New York State, each column is a covariate, and each cell is a value. NAs indicate that data was not avaialble for that grid cell.

The covariates used in the analysis.

Covariate	Description	Source
Deciduous	Proportion of a 6 km2 buffer around the detector made up of deciduous forest	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
Coniferous	Proportion of a 6 km2 buffer around the detector made up of coniferous forest	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
Mixed	Proportion of a 6 km2 buffer around the detector made up of mixed forest	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
Pasture	Proportion of a 6 km2 buffer around the detector made up of pasture	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
Cultivated.Crops	Proportion of a 6 km2 buffer around the detector made up of cultivated crops	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
road_density	Mean number of km of road per km2 in each 6 km2 buffer	calculated from primary and secondary roads raster provided by the NYSDEC, hosted on the github
snow_depth	Mean daily snow depth(m) of the 6 km2 buffer around each detector across the sampling period	National Operational Hydrologic Remote Sensing Centre, 2004. Snow data assimilation system (SNODAS) products (https://doi.org/10.7265/N5TB14TC).
forest_edge	Edge density of combined class of all forest in each 6 km2 buffer around detectors	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
GPP	The amount of carbon captured by plants (kg C MJ-1) in at the detectors	MODIS Land Satellite, 2017 (https://lpdaac.usgs.gov/products/mod17a2hv006/)
Camera	The camera trap model used at each site (Bushnell, Recoynx, Browning)	Fieldwork datasheets

2.allNY__2013-2021_15kmbuffer_allsitecovs.csv

This file contains all of the summarized spatial covariate data at a 15km scale used in the analysis. Each row is a different 15km 2 pixel in New York State, each column is a covariate, and each cell is a value. NAs indicate that data was not avaialble for that grid cell.

The covariates used in the analysis.

Covariate	Description	Source
Deciduous	Proportion of a 15 km2 buffer around the detector made up of deciduous forest	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
Coniferous	Proportion of a 15 km2 buffer around the detector made up of coniferous forest	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
Mixed	Proportion of a 15 km2 buffer around the detector made up of mixed forest	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
Pasture	Proportion of a 15 km2 buffer around the detector made up of pasture	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
Cultivated.Crops	Proportion of a 15 km2 buffer around the detector made up of cultivated crops	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
road_density	Mean number of km of road per km2 in each 15 km2 buffer	calculated from primary and secondary roads raster provided by the NYSDEC, hosted on the githu
snow_depth	Mean daily snow depth(m) of the 15 km2 buffer around each detector across the sampling period	National Operational Hydrologic Remote Sensing Centre, 2004. Snow data assimilation system (SNODAS) products (https://doi.org/10.7265/N5TB14TC).
forest_edge	Edge density of combined class of all forest in each 15 km2 buffer around detectors	NLCD, 2019 (https://www.mrlc.gov/data/nlcd-2019-land-cover-conus)
GPP	The amount of carbon captured by plants (kg C MJ-1) in at the detectors	MODIS Land Satellite, 2017 (https://lpdaac.usgs.gov/products/mod17a2hv006/)
Deer	The probability of occupancy (ψ) of white-tailed deer in a 15 km2 buffer around each detector	Calculated from detection/non-detection data and covariate hosted in the repository (see below)
Camera	The camera trap model used at each site (Bushnell, Recoynx, Browning)	Fieldwork datasheets

3. allNY_2013-2021_7dayocc_americanmarten_detection_nondetection.csv

This file contains detection/non-detection data for American marten between the years of 2013-2021 in New York state. Each row is a site, each column is an occasion, each cell is a detection/non-detection record. NAs indicate that the site was not sampled on that occasion.

4. allNY_2013_2021_7dayocc_coyote_counts.csv

This file contains the count data for coyote between the years of 2013-2021 in New York state. Each row is a site, each column is an occasion, each cell is a detection/non-detection record. NAs indicate that the site was not sampled on that occasion.

5. allNY_2013-2021_7dayocc_fisher_counts.csv

This file contains the count data for fisher between the years of 2013-2021 in New York state. Each row is a site, each column is an occasion, each cell is a detection/non-detection record. NAs indicate that the site was not sampled on that occasion.

6. 7dayoccDetectionswhitetaileddeerNZ.csv

This file contains the detection/non-detection data for white-tailed deer between the years of 2016-2018 in the northern zone of New York state. Each row is a site, each column is an occasion, each cell is a detection/non-detection record. NAs indicate that the site was not sampled on that occasion.

7. juliandays_allNY_2013-2021.csv

This file contains the sampling dates (in ordinal format) for each sampling occasion for the winter surveys from 2013-2021. Each row is a site, each column is an occasion. NAs indicate that the site was not sampled on that occasion.

8. 'coyote_fisher_marten_abu_ocu_binomialobsmodel_nimble_model.R'

This is the nimble occupancy-abundance model that is fit to the coyote-fisher-marten data files above. The code is commented out to describe each part of the model.

9. nimble_occu_abu_coyote_fisher_marten_binomialversion_processingandruncode.R

This is the data loading, formatting, and run script for the case study three species occupancy-abundance model. The code is commented out to describe each stage of the process.

Methods

In the case study using empirical data, we examine the intraguild interactions between three carnivores, a top mesopredator in the system, the coyote, an intermediate mesopredator, the fisher, and a small carnivore, the American marten. There is a long history of examining intraguild interactions between fisher and marten through harvest (e.g., Hardy, 1907; Krohn, Zielinski, & Boone, 1997). Recent harvest-based evidence was used to infer negative interactions between all three species, with fishers being limited through intraguild killing by coyotes, and martens being limited by both fisher and coyotes (Jensen & Humphries, 2019). Nonetheless, the three species co-occur over much of the marten’s limited range in New York State and recent analysis using Rota et al. (2016) co-occurrence models was inconsistent with previous hypotheses. This analysis found fisher occupancy was higher conditional on coyote presence, and marten occurred independently from both other species (Twining et al. In Press). Nonetheless, as explored in Simulation study I, a focus on occupancy states (and ignoring the abundance of species) to infer interactions may limit inference. As an example, we fit the occupancy-abundance model to the landscape scale camera trapping dataset on these species previously analysed using co-occurrence models in Twining et al. (In Press). Detection frequency data on coyotes and fisher, and detection-non detection data on American marten was obtained from camera traps deployed to monitor occurrence of the target species in northeastern New York State. Sampling was conducted from January-March 2016-2018 throughout the Adirondack and Tug Hill regions of northern New York State. This sampling used a stratified random sampling design to select 195 15 km² sample units across the region of interest. A standardized methodology was used across all surveys. At each site, a camera trap was deployed randomly within the 15 km²grid. Camera traps were secured to trees approximately 1.0-1.5 m above ground. A bait station was placed on a tree opposite the camera trap and secured to the tree using wire mesh. At all sites skunk-based call lures were applied. Cameras were deployed for 3 weeks (21 days) at each location after which cameras were retrieved. Cameras and bait were checked halfway through 3 weeks of sampling with batteries and bait replaced and replenished as necessary. We used a weekly occasion length. For the detection/non-detection data only one detection of the subordinate species was possible per weekly period. For the count data, we allowed a single detection each 24-hr period over each week and summed the days with detections into weekly counts for the species. The same sample units were sampled in each of the three sampling years (except for 13 sites that were not sampled in 2017).