PosterPDF Available

Resource selection function of elk in northeastern Nevada - Prelim. Results

Resource selection function of elk in northeastern Nevada
Introduction
Meghan P. Keating1, J.C. Brockman1, M.G. Lohman1, and C.J. McKee2
1University of Nevada, Reno, Dept. of Natural Resources and Environmental Science, 1664 N. Virginia St., Reno, NV, 89512.
2Nevada Department of Wildlife, 6980 Sierra Center Pkwy #120, Reno, NV, 89511.
Data and Study Area
Collected from February 2014 to July 2019
GPS points were taken twice a day using Vectronic brand GPS radio collars
Split data into summer and winter seasons (nsummer = 3,840, nwinter = 9,861)
Data were broken into model training and validation datasets
500000 550000 600000 650000 700000
4500000 4550000 4600000 4650000 4700000 4750000 4800000
Winter Intensity of Use
Easting
Northing
0.3
0.4
0.5
0.6
0.7
0.8
0.9
500000 550000 600000 650000 700000
4500000 4550000 4600000 4650000 4700000 4750000 4800000
Summer Intensity of Use
Easting
Northing
0.2
0.4
0.6
0.8
Model
Estimated the effects of selected predictor variables in a hierarchical Bayesian framework
Presence data (pai,s) – Bernoulli trial centered around a site- and individual-specific mean (µi,s)
!"#$% &Bern '#$%
Mean was calculated using a generalized mixed effects model
logit '#$% ()*%+ , -#
bvalues were assumed to be sampled from a prior distribution centered around zero
+ & Normal .$ /..0
Methods Conclusion
Results
Covariate Selection
Random forest estimate of importance
Calculated relative importance
R package Boruta [6]
Goodness of Fit/Cross Validation
Boyce Index
R package ecospat [2]
AUC-ROC curve using validation dataset
R package rms [3] and ROCR [5]
Model Uncertainties and Future Directions
Random forest estimates determined a large proportion of
environmental variables
Results of autocorrelation in data partitioning
High Boyce index indicates model overfitting
In the future:
Sample grid in “checkerboard” pattern to generate
training and validation datasets
Replace random forest with information criterion
Include local and proper scoring method
Bier score
Discussion
Summer model:
Positive correlation with higher elevation
Winter model:
Positive correlation with higher elevation and slope
Negative correlation with roads
Background
Rocky Mountain Elk (Cervus elaphus nelsoni) are an important source
of funding for the Nevada Department of Wildlife (NDOW)
Estimated 17,500 elk in Nevada [1]
Range between northeastern Nevada and southern
Idaho
Disjoint herds in southern Nevada
Habitat use by elk
Influenced by an attempt to either maximize energy
gained or minimize energy used to obtain resources [4]
Elk movement
Influenced by both biological (food availability, predator
presence,) and topographic factors (elevation,
ruggedness) [7]
Aim
Create an index of habitat use and preference of elk within the state
of Nevada
Inform managers of elk habitat selection
Indicate potentially suitable habitat within unoccupied
areas of the state
Literature Cited: [1] Beck, J.L. (2003) Elk summer range habitat, nutritional ecology, and carrying capacity in the Jarbidge Mountains, Nevada. Ph.D. thesis, University of Idaho.; [2] Di Cola, V. et al. (2017) ecospat: an r package to support spatial analyses and modeling of species niches and distributions. Ecography 40, 774–787; [3] Harrell Jr. et al. (2019)
Package rms’. Vanderbilt University 229; [4] Kie, J.G. (1999) Optimal foraging and risk of predation: effects on behavior and social structure in ungulates. Journal of Mammalogy 80, 1114–1129; [5] Kursa, M.B. et al. (2010) Feature selection with the Boruta package. J Stat Softw 36, 1–13. [6] Sing, T. et al. (2005) Rocr: visualizing classifier performance in R.
Bioinformatics 21, 3940–3941; [7] Thomas, J. et al (1982) Elk of North America: ecology and management. Stackpole Books.
Funding: Wildlife Restoration
Act
Agencies: Nevada
Department of Wildlife
Biologists: Kari Huebner,
Matt Jeffress
UNR Faculty: Dr. Kevin T.
Shoemaker,
Dr. Kelley Stewart
Covariate Selection
Random forest variable importance estimates for summer and winter models. Importance, on the y-axis, was
measured as the z-standardized mean decrease in accuracy for each variable, shown on the x-axis.
Goodness of Fit/Cross-Validation
Estimates of bvalues for the
influence of environmental
variables on elk presence.
Boyce indices (left) indicate good model fit with the training datasets. Cross-validation AUC-ROC curves (right) show excellent
model performance for the summer model and poor model performance in the winter model.
Acknowledgments
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.