Density Dependent Habitat Selection by Michigan Wolves

Authors: Shawn T. O'Neil, Joseph K. Bump of Michigan Technological University and Dean E Beyer, Jr. of Michigan Department of Natural Resources. Presentation given at the 2015 Midwest Wolf Stewards Conference at Northland College. April 2015

Transcript

1. Density Dependent Habitat
   Selection by Michigan Wolves
   Modeling wildlife-habitat relationships
   over space and time
   Shawn T. O’Neil1, Joseph K. Bump1, and Dean E. Beyer, Jr2. 2015.
   1Michigan Technological University
   2Michigan Department of Natural Resources
   

   View Slide

2. Outline
   • Why study habitat selection?
   • Habitat selection by wolves: what we know
   • The importance of density dependence in habitat and
   niche modeling
   • The density dependent habitat selection model for
   Michigan wolves
   • Results
   • Conclusion & implications for management
   • Next steps
   

   View Slide

3. Why study habitat selection?
   • A link between the animal and its environment
   o Resource Selection Function
   o Species Distribution Model
   • Identifying habitat, or the ecological niche
   “the combination of resources and environmental conditions necessary to allow
   occupancy, survival, and reproduction of individuals (Morrison et al. 1992)” – Franklin et
   al. 2000
   o Predicting species distributions
   o Protecting wildlife habitat
   o Conserving populations
   

   View Slide

4. Habitat selection by wolves: what
   we know
   • Common predictors
   (resources vs. risk)
   o Prey densities
   o Human densities
   o Roads
   o Agriculture
   • Context-dependent
   predictors
   o Elevation, slope, ruggedness
   o Streams
   o Land cover
   • Forested, open, edges
   o Snow
   Probability of wolf occurrence based on road and deer
   densities (upper, Potvin et al. 2005),
   and road densities (lower, Mladenoff et al. 1995)
   

   View Slide

5. • Newer developments
   • Context dependent selection of
   o Road networks
   o Logging roads
   o Human activity
   A) Selection during the day B) Selection at night
   Wolf pack selection of the Ya Ha Tinda Ranch near Banff National Park during day vs. night hours
   Adapted from Hebblewhite & Merrill (2008)
   

   View Slide

6. Habitat selection by wolves: newer
   developments
   • Importance of temporal variability
   o Interannual (Uboni et al. 2015a)
   • Pack differences (learned behavior)
   • Pack size
   • Precipitation
   • But not prey distribution!
   o Seasonal (Uboni et al. 2015b)
   • Roads
   • Elevation
   • Openness
   • Habitat type
   • A shifting habitat mosaic
   o Weather
   o Human land use (i.e. logging)
   o Prey distributions
   Uboni et al. 2015b
   

   View Slide

7. The importance of density
   dependence in habitat and niche
   modeling
   • Density dependence likely contributes to temporal
   variability in habitat selection
   o missing from most habitat selection models
   • May be a key variable when population is increasing (or
   decreasing)
   • Effect on habitat availability
   o Territorial species
   o Per capita availability of important habitat decreases as density increases
   o Competition for high quality sites
   • Losers forced into marginal habitat
   

   View Slide

8. Mean (A1
   ) = 1.17
   Available
   Mean (A1
   )= 1.17
   Mean (U1
   ) = 0.93
   U1
   / A1
   = 0.79
   Mean (A2
   )= 1.52
   Mean (U1
   ) = 0.93
   U1
   / A1
   = 0.61
   Mean (A1
   ) = 1.17
   Mean (U1
   ) = 0.93
   Used
   Available
   Mean (U1
   ) = 0.93
   Mean (A2
   ) = 1.52
   Available 2
   Used
   The used habitat distribution is
   a weighted version of the
   available distribution
   The habitat selection function
   is proportional to the ratio
   between habitat use and
   availability (Aarts et al. 2013)
   Consider the habitat selection function:
   Any change in the availability distribution of a resource can
   cause changes in the habitat selection function!
   - The Generalized Functional Response (GFR) in resource
   selection (Matthiopoulos et al. 2011, Aarts et al. 2013)
   

   View Slide

9. • If density influences per-capita availability:
   o Availability declines as density increases
   o The use-availability ratio is unlikely to be static (i.e. selection or avoidance
   of a resource may change); we expect a “functional response” relating
   to density
   o Limiting features are likely to have greatest selection at low densities
   o Selection at high densities may not represent greatest habitat quality
   • Habitat selection is likely density dependent
   From McLoughlin et al. (2010)
   

   View Slide

10. Indications of density dependent
    MI population growth
    Number of wolves
    Estimated available habitat (km2) Home range size (km2)
    2000 2005 2010
    

    View Slide

11. Mills 2008, Fig. 6.4, p. 123
    

    View Slide

12. The density dependent habitat
    selection model for Michigan
    wolves
    • We modeled habitat selection by Michigan wolves
    during recolonization (1994-2013)
    • Telemetry locations were used to generate
    seasonal home ranges
    o Each individual with ≥ 14 locations per summer/winter
    o 962 seasonal ranges
    • 258 individuals, 111 unique packs
    

    View Slide

13. Model covariates
    Prey indices
    Buck harvest
    % Deer wintering complex
    Distance deer wintering complex
    Mean annual snow depth
    Prey
    

    View Slide

14. Model covariates
    Natural features
    Elevation
    CV (Elevation)
    Forest-Open Edge Density
    Stream density
    % Open
    % Water/Wetlands
    Slope
    Prey
    Natural
    

    View Slide

15. Model covariates
    Minor road density
    Human Influence
    Prey
    Natural
    % Impervious surface
    % Agriculture
    Distance to major road
    Human
    

    View Slide

16. • Circular assessment window
    o ¼ of mean wolf home range area (50.75 km2)
    • Characterizing availability
    o Restricted to areas not already occupied by existing
    packs
    • Known pack territories from track surveys (Beyer
    et al., Michigan DNR)
    • Used vs. Available sampling design
    o For seasonal range within year:
    • 5 locations per seasonal range (Used distribution
    = )
    • 50 locations per seasonal range (Available
    distribution = )
    Prey
    Natural
    Human
    Cells included for processing
    Processing cell
    Focal statistics with moving neighborhood illustration
    http://resources.arcgis.com/en/help/main/10.1/index.html
    
    1995
    2010
    

    View Slide

17. • Habitat selection coefficients can be generated
    using logistic regression (McDonald 2013)
    • Additional structure was needed for longitudinal
    data (time series)
    o Repeated sampling of the same packs over time
    o Random spatial variation due to sampling design
    • Uneven trapping, monitoring effort
    • Generalized Linear Mixed Models (GLMMs)
    o Binomial distributed response (Used = 1, Available = 0)
    o Random effects:
    • PACK, BIOLOGICAL YEAR, SURVEY UNIT
    o R 3.1.1, Package ‘lme4’
    o Habitat selection function = = exp() = exp(′ ) =
    exp( 1
    1
    +. . . +
    
    )
    …2001 2002 2003 2004…
    ..Pack 1 Pack 1 Pack 1
    Pack 2 Pack 2
    Pack 3 Pack 3 Pack 3..
    

    View Slide

18. • Effect of wolf density
    o Wolf density per survey unit
    o MI DNR abundance estimation tracks wolf densities spatially and
    temporally (e.g. Potvin et al. 2005)
    o 3 classes:
    • Low (25th percentile) = < 5 wolves / 1000 km2
    • Medium (25th – 75th percentile) = 5 – 10 wolves / 1000 km2
    • High (75th – 100th percentile) = > 10 wolves / 1000 km2
    o Interaction terms
    • Wolf Density * , where = 1
    , … ,
    (environmental covariates)
    • Model Selection
    o Full model sets for each category (prey, natural features, human
    influence)
    o Sequentially reduced each model by removing insignificant effects
    o Combined results into full model
    

    View Slide

19. Results
    • Final habitat selection model
    o Most of original predictors significantly influenced wolf selection
    o Many also varied significantly with wolf density
    • Cross Validation
    o Leave-one (year) – out, refit model to predict on the omitted data
    • Observed vs. expected, # locations per resource selection class
    o Spearman’s rank correlation (
    ) = 0.94 (averaged for all years)
    o R2 = 0.92
    

    View Slide

20. Density-
    dependent
    predictors of
    habitat selection
    • Strength of
    selection/avoidance
    declines with increasing
    wolf density:
    • Bucks harvested (deer
    density)
    • % Impervious surface
    • Elevation
    • CV(elevation)
    • Stream density
    • Strength increases with
    increasing density
    • % Agriculture
    

    View Slide

21. Predicting over space
    and time
    • Accounting for
    • Changes in wolf density
    • Longitudinal data
    • Random process variation (sampling &
    monitoring design)
    • Results in
    • High predictive ability
    • A more informative model
    

    View Slide

22. Conclusion & Implications
    • Wolf habitat selection varies depending on density
    • Selection for some important predictors is strongest
    at low wolf densities
    • Selection for other (perhaps essential) predictors
    remains constant
    o Deer wintering complex
    • Density dependent habitat selection in territorial
    species suggests site-dependent regulation
    o Potential for source-sink dynamics (Mosser et al. 2009)
    o Need to consider interstate policies in management plans
    

    View Slide

23. Next steps
    • Does habitat selection correlate with fitness?
    o Population growth rate
    o Adult survival rate
    o Event-specific mortality rate
    • Does a source-sink dynamic exist between higher
    and lower quality habitats?
    • Does this regulate population growth?
    o Driven by resources vs. risk?
    

    View Slide

24. Acknowledgments:
    Dean Beyer, Jr., Erin Largent, Bob Doepker, Steve Carson,
    Brian Roell, Chris Webster, Michigan Department of
    Natural Resources
    • List of references:
    • Aarts, G., J. Fieberg, S. Brasseur, and J. Matthiopoulos. 2013. Quantifying the effect of habitat availability on species
    distributions. Journal of Animal Ecology 82:1135-1145.
    • Franklin, A. B., D. R. Anderson, R. J. Gutiérrez, and K. P. Burnham. 2000. Climate, habitat quality, and fitness in Northern Spotted
    Owl populations in northwestern California. Ecological Monographs 70:539-590.
    • Hebblewhite, M., and E. Merrill. 2008. Modelling wildlife-human relationships for social species with mixed-effects resource
    selection models. Journal of Applied Ecology 45:834-844.
    • Houle, M., D. Fortin, C. Dussault, R. Courtois, and J. P. Ouellet. 2010. Cumulative effects of forestry on habitat use by gray wolf
    (Canis lupus) in the boreal forest. Landscape Ecology 25:419-433.
    • Lesmerises, F., C. Dussault, and M. H. St-Laurent. 2012. Wolf habitat selection is shaped by human activities in a highly
    managed boreal forest. Forest Ecology and Management 276:125-131.
    • Matthiopoulos, J., M. Hebblewhite, G. Aarts, and J. Fieberg. 2011. Generalized functional responses for species distributions.
    Ecology 92:583-589.
    • McDonald, L., B. Manly, F. Huettmann, and W. Thogmartin. 2013. Location-only and use-availability data: analysis methods
    converge. Journal of Animal Ecology 82:1120-1124.
    • McLoughlin, P. D., D. W. Morris, D. Fortin, E. Vander Wal, and A. L. Contasti. 2010. Considering ecological dynamics in resource
    selection functions. Journal of Animal Ecology 79:4-12.
    • Mills, L. S. 2012. Conservation of wildlife populations: demography, genetics, and management. John Wiley & Sons.
    • Mladenoff, D. J., T. A. Sickley, R. G. Haight, and A. P. Wydeven. 1995. A regional landscape analysis and prediction of
    favorable gray wolf habitat in the Northern Great Lakes Region. Conservation Biology 9:279-294.
    • Mosser, A., J. M. Fryxell, L. Eberly, and C. Packer. 2009. Serengeti real estate: density vs. fitness-based indicators of lion habitat
    quality. Ecology Letters 12:1050-1060.
    • Potvin, M. J., T. D. Drummer, J. A. Vucetich, D. E. Beyer, R. O. Peterson, and J. H. Hammill. 2005. Monitoring and habitat
    analysis for wolves in upper Michigan. Journal of Wildlife Management 69:1660-1669.
    • Uboni, A., D. W. Smith, J. S. Mao, D. R. Stahler, and J. A. Vucetich. 2015a. Long- and short-term temporal variability in habitat
    selection of a top predator. Ecosphere 6:art51.
    • Uboni, A., J. A. Vucetich, D. R. Stahler, and D. W. Smith. 2015b. Interannual variability: a crucial component of space use at
    the territory level. Ecology 96:62-70.
    

    View Slide

25. Thank you! Questions?
    

    View Slide

26. Supplemental
    

    View Slide

27. Category Predictor (mean within moving window) Abbreviation Range (95% CI)
    Prey Availability Buck harvest (bucks / km2) BUCK 0.3 – 2.3
    % Deer wintering complex PDWC 0.0 – 96.7
    Distance to deer wintering complex (km) DDWC 0.0 – 11.8
    Human % Agriculture AG 0.0 – 29.9
    Distance to highway (km) HWY 1.1 – 20.1
    % Developed impervious IMP 0.0 – 4.6
    Minor road density (km / km2) RDENS 0.3 – 2.5
    Natural Elevation (m) ELEV 187 – 521
    Coefficient of variation of elevation CVEL 0.0 – 0.19
    Slope (°) SLO 0.4 – 5.3
    Open:Forested edge density (km / km2) EDGE 0.1 – 3.5
    % Open OPEN 0.1 – 38.1
    % Water and wetlands WET 0.0 – 83.6
    Stream density (km / km2) SDENS 0.0 – 5.4
    Interacting Variables Winter snow depth (cm) SNOW 83.7 – 502.2
    Season (summer/winter) SEAS
    

    View Slide

28. Model Fixed Effects Structure BIC δBIC
    Prey
    (Full)
    + + × + + × +
    + ×
    22664.3 36.9
    − :
    Prey (Reduced
    1)
    + + × + × + ( + 22653.4 26.1
    − :
    Prey (Reduced
    2)
    + + × + × + × 22642.9 15.6
    − :
    Prey (Reduced
    3)
    + × + × + × + 22627.3
    ℎ
    

    View Slide

29. Category Predictor
    
    SE ( ) Z P (Z)
    Prey Buck Harvest (BUCK) 5.051 0.518 9.755 < 0.001
    % Deer Wintering Complex (PDWC) 0.891 0.106 8.422 < 0.001
    Distance to DWC (DDWC) -3.188 0.288 -11.088 < 0.001
    Human % Impervious Surface (IMP) -27.924 5.113 -5.462 < 0.001
    % Agriculture (AG) 1.599 0.832 1.922 0.055
    Distance to Major Highway (HWY) -1.037 0.375 -2.762 0.006
    Natural Forest:Open Edge Density (EDGE) 3.818 0.196 19.479 < 0.001
    Stream Density (SDENS) 3.248 0.547 5.939 < 0.001
    % Water or Wetland (WET) -1.285 0.166 -7.730 < 0.001
    Elevation (ELEV) 7.839 0.486 16.114 < 0.001
    ELEV2 -3.943 0.957 -4.119 < 0.001
    CV Of Elevation (CVEL) 4.475 0.457 9.794 < 0.001
    Slope (SLO) -4.592 0.264 -17.405 < 0.001
    Annual Snow Depth (SNOW) -1.289 0.256 -5.040 < 0.001
    Seasonal Season (SEAS) -0.114 0.041 -2.799 0.005
    PDWC × SEAS (winter) 0.891 0.106 8.422 < 0.001
    PDWC × SNOW -2.185 0.375 -5.828 < 0.001
    Density
    Dependent
    BUCK | Wolf Density (WOLF) = Medium -1.090 0.525 -2.076 0.038
    IMP | WOLF = Medium 19.688 5.242 3.755 < 0.001
    AG | WOLF = Medium -3.034 0.901 -3.366 0.001
    HWY | WOLF = Medium -0.317 0.456 -0.696 0.487
    ELEV | WOLF = Medium -3.570 0.377 -9.468 < 0.001
    CVEL | WOLF = Medium -2.485 0.493 -5.040 < 0.001
    SDENS | WOLF = Medium 0.339 0.604 0.562 0.574
    WOLF = Medium 1.137 0.134 8.470 < 0.001
    BUCK | WOLF = High -4.215 0.575 -7.324 < 0.001
    IMP | WOLF = High 20.328 5.215 3.898 < 0.001
    AG | WOLF = High -6.669 0.927 -7.192 < 0.001
    HWY | WOLF = High 2.302 0.474 4.856 < 0.001
    ELEV | WOLF = High -3.189 0.392 -8.135 < 0.001
    CVEL | WOLF = High -3.545 0.458 -7.735 < 0.001
    SDENS | WOLF = High -2.252 0.596 -3.777 < 0.001
    WOLF = High 1.522 0.144 10.578 < 0.001
    

    View Slide

30. Year
    R2
    R2
    Marginal
    HSF
    Conditional
    HSF
    1995-1998 0.91 0.42 0.94 0.65
    1999 0.95 0.98 0.96 0.99
    2000 0.96 0.8 0.92 0.92
    2001 0.99 0.91 0.99 0.99
    2002 0.92 0.82 0.98 0.99
    2003 0.99 0.96 0.99 0.99
    2004 0.99 0.94 0.97 0.96
    2005 0.94 0.83 0.99 0.99
    2006 0.99 0.91 0.99 0.98
    2007 0.95 0.6 0.95 0.9
    2008 0.94 0.5 0.98 0.93
    2009 0.88 0.9 0.95 0.91
    2010 0.93 0.83 0.82 0.98
    2011 0.9 0.77 0.85 0.92
    2012 0.98 0.89 0.92 0.8
    2013 0.92 0.87 0.87 0.86
    Yearly Avg. 0.95 0.81 0.94 0.92
    

    View Slide

31. • Consider the habitat selection function for a single
    variable, minor road density
    o = the resource selection function (RSF) =
    o = exp() = exp(′ ) = exp( 1
    1
    +. . . +
    
    )
    • 1
    , … ,
    are environmental covariates
    • 1
    , … ,
    are coefficient estimates (+ = selection, - = avoidance)
    o Typically from logistic regression
    o =
    
    • is the used habitat distribution
    • is the available habitat distribution
    • Thus, the used habitat distribution is a weighted version of the
    available distribution
    • is proportional to the ratio between habitat use and availability
    (Aarts et al. 2013)
    • Use proportional to availability
    o Importance of the availability distribution in the use-availability ratio
    o Any change in can cause changes in , the estimated habitat
    selection function
    

    View Slide

32. Carrying capacity?
    • Mladenoff 1999:
    o 581 – 1,357
    o But,
    • Potvin et al. model indicates less available habitat constrained by
    deer densities
    • Our results indicate habitat suitability is dynamic
    o K will vary from year to year
    o Good year for deer vs. bad year?
    • Bad year might mean more vulnerable prey
    • Good year might mean more deer…
    o Will wolves start preying on moose if deer stay low?
    o Habitat suitability should be measured by what promotes population
    growth
    • Reproduction & survival
    • Importance of getting good deer numbers
    

    View Slide