1 FINAL REPORT TO FLORIDA FISH AND WILDLIFE CONSERVATION COMMISSION ON CONTRACT R112219563 WITH THE UNIVERSITY OF TENNESSEE Black Bear Population Size and Density in Apalachicola, Big Cypress, Eglin, Ocala/St. Johns, and Osceola Study Areas, Florida 18 August 2016 JACOB HUMM, Department of Forestry, Wildlife and Fisheries, University of Tennessee, 274 Ellington Plant Sciences Building, Knoxville, TN 37996, USA. J. WALTER McCOWN, Fish & Wildlife Research Institute, Florida Fish & Wildlife Conservation Commission, 1105 S.W. Williston Rd., Gainesville, FL 32601-9044, USA. BRIAN K. SCHEICK, Fish & Wildlife Research Institute, Florida Fish & Wildlife Conservation Commission, 1105 S.W. Williston Rd., Gainesville, FL 32601-9044, USA. JOSEPH D. CLARK, Principal Investigator, U.S. Geological Survey, Southern Appalachian Research Branch, University of Tennessee, 274 Ellington Plant Sciences, Knoxville, TN 37996, USA ABSTRACT: We performed a statewide population assessment for Florida black bears (Ursus americanus floridanus) based on spatially explicit capture-mark-recapture modeling (SCR) using DNA collected at barbed-wire hair sampling sites during 2014 and 2015. We used SCR to estimate density and abundance of the 5 major bear populations in Florida. We used a 3 x 3 sampling cluster array spaced over a combined 38,960 km 2 to estimate parameters for the Eglin, Apalachicola, Osceola, Ocala/St. Johns, and Big Cypress bear populations. Several landscape variables helped refine density estimates for the 5 populations we sampled. Detection probabilities were affected by site-specific behavioral responses coupled with sex effects. Model-averaged bear population estimates ranged from 102.0 (95% CI = 55.7 – 212.0) bears or 0.021 bears/km 2 (95% CI = 0.012 – 0.44) for the Eglin population to 1,192.6 bears (95% CI = 950.8 – 1,519.5) or 0.127 bears/km 2 (95% CI = 0.101 – 0.161) for the Ocala/St. Johns
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
FINAL REPORT TO FLORIDA FISH AND WILDLIFE CONSERVATION
COMMISSION ON CONTRACT R112219563 WITH THE UNIVERSITY OF
TENNESSEE
Black Bear Population Size and Density in Apalachicola, Big Cypress, Eglin,
Ocala/St. Johns, and Osceola Study Areas, Florida
18 August 2016
JACOB HUMM, Department of Forestry, Wildlife and Fisheries, University of Tennessee, 274
Ellington Plant Sciences Building, Knoxville, TN 37996, USA.
J. WALTER McCOWN, Fish & Wildlife Research Institute, Florida Fish & Wildlife
Conservation Commission, 1105 S.W. Williston Rd., Gainesville, FL 32601-9044, USA.
BRIAN K. SCHEICK, Fish & Wildlife Research Institute, Florida Fish & Wildlife Conservation
Commission, 1105 S.W. Williston Rd., Gainesville, FL 32601-9044, USA.
JOSEPH D. CLARK, Principal Investigator, U.S. Geological Survey, Southern Appalachian
Research Branch, University of Tennessee, 274 Ellington Plant Sciences, Knoxville, TN
37996, USA
ABSTRACT: We performed a statewide population assessment for Florida black bears (Ursus
americanus floridanus) based on spatially explicit capture-mark-recapture modeling (SCR) using
DNA collected at barbed-wire hair sampling sites during 2014 and 2015. We used SCR to
estimate density and abundance of the 5 major bear populations in Florida. We used a 3 x 3
sampling cluster array spaced over a combined 38,960 km2 to estimate parameters for the Eglin,
Apalachicola, Osceola, Ocala/St. Johns, and Big Cypress bear populations. Several landscape
variables helped refine density estimates for the 5 populations we sampled. Detection
probabilities were affected by site-specific behavioral responses coupled with sex effects.
Model-averaged bear population estimates ranged from 102.0 (95% CI = 55.7 – 212.0) bears or
0.021 bears/km2 (95% CI = 0.012 – 0.44) for the Eglin population to 1,192.6 bears (95% CI =
950.8 – 1,519.5) or 0.127 bears/km2 (95% CI = 0.101 – 0.161) for the Ocala/St. Johns
2
population. The total population estimate for our 5 study areas was 3,900 bears (95% CI =
2,919.7 – 5,373.5).
INTRODUCTION
The Florida black bear (Ursus americanus floridanus) historically occurred throughout the state
but was reduced to an estimated 300–500 bears by the 1970s due to loss of habitat and
unregulated killing (Brady and Maehr 1985). As a result, Florida classified the black bear as a
State Threatened Species throughout most counties in 1974. Today, the Florida black bear is
comprised of 7 distinct subpopulations within the state (i.e., Apalachicola, Eglin, Osceola,
Ocala/St. Johns, Chassahowitzka, Highlands/Glades, and Big Cypress; Dixon et al. 2007). Some
of these populations are small (e.g., Chassahowitzka) and all are impacted by habitat
fragmentation, which restricts movements and genetic interchange among subpopulations (Dixon
et al. 2006, Dixon et al. 2007). Additionally, a large number of bears are killed on Florida
highways each year (Florida Fish and Wildlife Conservation Commission [FWC], unpublished
data).
Simek et al. (2005) estimated the size of the Apalachicola, Big Cypress, Eglin, Osceola,
Ocala, and St. Johns bear subpopulations from 2001 to 2003 using mark-recapture techniques
based on DNA extracted from bear hair (Paetkau et al. 1995). Hair samples were collected from
barbed wire sampling sites and genotyped to individual animals; these genetic data were treated
as marks. The advantages of this technique compared with traditional live-capture are that it
minimizes capture biases and is relatively cost effective. Simek et al. (2005) placed baited hair-
sampling sites (hair traps) within a smaller portion of occupied bear range in each of Florida’s 6
major subpopulations so that about 4 hair traps would be present within the estimated summer
home range of each female (Otis et al. 1978). Sites were constructed by enclosing 4–6 trees with
3
2 strands of barbed wire, 25 cm and 50 cm high. Baits consisting of corn and pastries were hung
within each enclosure. Sites were checked after 2 occasions of 6–8 days each, allowed to remain
unbaited for 6–8 days, and the process was repeated 3 more times for a total of 8 weekly
sampling occasions. Capture probabilities were high (p = 0.28) during each 12- to 16-day
session, even after considerable subsampling of the hair collected. Because only a portion of the
area occupied by each subpopulation was sampled, population and density estimates were
extrapolated to the entire occupied range, assuming homogeneous and equivalent densities across
the broader area. Abundances ranged from 63–101 at Eglin to 729–1,056 at Ocala.
Simek et al. (2005) used Program CAPTURE (Otis et al. 1978, Rexstad and Burnham
1991) to estimate within-year population parameters. Program CAPTURE may not always
properly select among competing models or detect capture heterogeneity when it is present
(Menkens and Anderson 1988, Stanley and Burnham 1998, Boulanger et al. 2002) and options
for modeling heterogeneous capture probabilities are limited to non-parametric estimators (i.e.,
Jackknife [Otis et al. 1978] and Chao methods [Chao 1989]). Likelihood-based estimators have
since been developed to estimate capture heterogeneity (Huggins 1989, 1991; Pledger 2000),
thus permitting comparisons among all models using information-theoretic methods (Burnham
and Anderson 2002). Information-theoretic procedures are considered superior to the model
selection method in Program CAPTURE (Stanley and Burnham 1998) and also allow model
averaging of parameter estimates, which helps account for model selection error and improves
inference (Luikart et al. 2010).
A number of other refinements in mark-recapture methodology have been developed
since the Simek et al. (2005) study, including spatially explicit capture-mark-recapture (SCR)
and cluster sampling. SCR incorporates trap location data into the estimation process (Borchers
4
and Efford 2008) and is most commonly used to estimate population density (D). However,
estimation of population abundance (N) is also possible and may be more robust to biases caused
by spatial heterogeneity in capture probabilities, which is common for species with large home
ranges like black bears (Royle et al. 2014). Efford and Fewster (2013) found that spatially
explicit models for estimating N were robust to gaps between sampling sites and heterogeneous
animal distributions, thus allowing for efficient cluster sampling designs that can be used to
sample a wider and more representative geographic area. Spatial covariates can also be used to
estimate density in the areas between clusters not sampled. These advances are important
because non-spatial mark-recapture analyses are based on the assumption that all animals have
an equal probability of capture with respect to their location in the sampling grid and,
consequently, traps have to be closely spaced to avoid gaps in the sampling pattern. Such dense
trap spacing meant that Simek et al. (2005) could only sample a portion of the 6 study areas
without hundreds of sites per subpopulation. Extrapolations to the larger study areas were based
on the unrealistic but unavoidable assumption that bear population characteristics in the sampled
area were similar throughout the general study area. Cluster sampling using SCR has since been
evaluated and found to be a reliable method for estimating bear abundance and density across
extensive areas given appropriate trap spacing (Sollmann et al. 2012, Efford and Fewster 2013,
Sun et al. 2014).
Under strict statewide protection and management, Florida black bear numbers were
thought to have increased and the subspecies was removed from the State Threatened Species
List in 2012 (Telesco 2012). That removal was contingent upon the formulation of a
management plan that would maintain viable populations of black bears in suitable habitat. Our
objectives were to use spatially explicit methods to estimate bear population abundance and
5
density for the same major bear populations in the state surveyed by Simek et al. (2005).
Because the Ocala and St. Johns study areas listed by Simek et al. (2005) included contiguous
bear subpopulations that are genetically indistinguishable (Dixon et al. 2007) and now
administered as one (FWC 2012), we combined them to create a single Ocala/St. Johns study
area.
METHODS
Study Areas
Our study focused on 5 subpopulations of the Florida black bear, the extents of which ranged
from the Florida panhandle region to the southern tip of the peninsula (Fig. 1). The Eglin study
area was located in the western panhandle and was comprised of areas in and around Eglin Air
Force Base. The Apalachicola study area was located in the eastern panhandle region and was
comprised of habitat in and around Apalachicola National Forest. The Osceola study area was in
the northern border of the Florida Peninsula and was comprised of habitat in and around Osceola
National Forest. The Ocala/St.Johns study area was located in north-central Florida and was
comprised of habitat in and around Ocala National Forest as well as Flagler and Volusia counties
east of the St. Johns River. Finally, the Big Cypress study area was in the southern portion of the
Florida peninsula and was comprised of habitat in and around Big Cypress National Preserve.
The combined total area sampled was 38,960 km2.
Before establishing hair traps, we evaluated a number of cluster configurations to assess
bias and optimize efficiency. First, we obtained the Simek et al. (2005) trap and capture data for
2003. We used only 1 year of data because each of our population estimates were to be made
from 1 year of sampling. We estimated 2003 bear densities, capture probabilities (g0) and a
home range parameter (σ) for each of the study areas using secr, which is an R-based (R Core
6
Team 2015) SCR routine based on maximum likelihood estimation methods (Efford 2004).
Given those estimates, we then conducted simulations of various trap configurations and cluster
sizes in secr for each study area to assess bias and precision. We evaluated 2 x 2, 3 x 3, and 4
x 4 trap clusters, traps within clusters spaced 500, 1,000, 1,500, 2,000, 2,500, 3,000, 3,500, and
4,000 m apart, spacings between clusters (center to center) of 10,000, 12,000, 14,000, 16,000,
18,000, 20,000, 25,000, and 30,000 m, and sampling periods of 4, 6, or 8 weeks. The 3 x 3 trap
cluster configuration with traps 2,000 m apart, clusters spaced 16,000 m between cluster centers,
and conducted over a 6-week sampling period performed well for all study areas, resulting in low
bias and reasonable confidence intervals (J. Clark, U.S. Geological Survey, unpublished data).
Based on this trap cluster configuration, we mapped proposed hair traps and field
personnel were instructed to find sites with suitably spaced trees within 250 m of the assigned
trap coordinates for constructing the hair traps. The areas to which our cluster sampling was
applied were loosely based upon a map of primary and secondary bear range in Florida
developed by Scheick and McCown (2014). However, when unable to strictly adhere to site
locations due to human development, impenetrable habitat, property access, or road access, we
selected an alternate site within 600 m. If no suitable site was within 600 m of the proposed
location was available, we dropped that hair trap from the cluster. We constructed and checked
hair traps on Osceola and Ocala/St. Johns during 2014 and on Eglin, Apalachicola, and Big
Cypress during 2015.
Sample Collection
Hair traps for all study areas consisted of enclosures comprised of 2 strands of barbed wire
stretched around 3–5 trees. We positioned the strands 35–40 and 65–70 cm above the ground
and blocked variations in the terrain (e.g., small gullies, mounds) with vegetation and woody
7
debris to prevent bears from crossing over or under the wires. We hung bait (bakery products)
from a line that spanned the enclosure. We also used commercial bear lure (Code Blue Bear
Magnet Raspberry Donut Attractant, Code Blue, Calera, Alabama, USA, or Bait Station Bear
Bait, Evolved, New Roads, Louisiana, USA) as a long-distance scent attractant. We placed hair
samples in coin envelopes and stored them at room temperature prior to analysis. We used
lighters and propane torches to burn any remaining hair off the barbs after each hair collection
occasion. We checked and rebaited all hair traps weekly for 6 consecutive weeks, beginning in
June of each year.
Genetic Analysis
Hair samples were shipped to Wildlife Genetics International (WGI; Nelson, British Columbia,
Canada) for genotyping. Due to a high volume of samples, subsampling routines were
implemented for both years (Laufenberg et al. 2016). Because Augustine et al. (2014) identified
potential problems arising from subsampling in conjunction with a potential behavioral response
to traps, we genotyped all samples from week 1 to evaluate the potential for introduced
behavioral bias from subsequent recaptures (i.e., “trap-happy” bears) during 2014. One sample
per site per week was selected for genotyping for weeks 2–6. Technicians at WGI randomized
the samples within each site-week and selected the first sample encountered containing >30
underfur or 5 guard hair roots. If none of the samples at a site-week met this quality threshold,
technicians chose the best available sample from the site, using a minimum quality threshold of 1
guard hair root or 5 underfur hairs. If none of the samples met this more lenient threshold, the
site was left out of the analysis for that sampling event. Analyses subsequent to the Augustine et
al. (2014) paper indicated that subsampling bias is not significant with SCR methods when a
consistent percentage of the hair samples are subsampled from week to week (B. Augustine,
8
University of Kentucky, unpublished data). Thus, in 2015 we subsampled for all weeks but
selected 2 samples at random per visited hair trap per week for genotyping (ensuring the two
samples were from different sides of the hair trap) to maximize the success rate for all sampling
weeks while reducing the number of duplicate samples.
Following standard protocols (Woods et al. 1999, Paetkau 2003, Roon et al. 2005), DNA
was extracted using QIAGEN DNeasy Blood and Tissue spin columns. The number of markers
required to correctly identify individuals depends on the subpopulation’s size and genetic
structure. WGI had analyzed the samples collected by Simek et al. (2005) and used their
knowledge of each subpopulation’s genetic structure in their marker recommendations (WGI,
unpublished data 2014). Thus, the analysis of individual identity was based on 8 markers
comprised of a gender marker and 7 microsatellites, except 9 markers (8 microsatellites and 1
gender marker) were used for samples from Big Cypress. Samples that match at all but one or
two markers may be different individuals (often siblings) or they may be the same individual
misidentified by genotyping errors (Paetkau 2003). To find and correct such misidentifications,
all 1- or 2-mismatched markers were reanalyzed, a process that effectively ensured that the
number of individuals identified in the dataset had not been affected by undetected genotyping
error (Kendall et al. 2009).
Population Analysis
We used ver. 2.10.2 of the R package ‘secr’ to estimate population parameters (Efford 2004,
Efford et al. 2004, Borchers and Efford 2008, Efford 2012, R Core Team 2015) within an
information theoretic model selection framework based on Akaike’s Information Criterion
adjusted for small sample size (AICc, Burnham and Anderson 1998). We evaluated models
whereby heterogeneity in detection probability (g0) or a home range parameter (σ) was explained
9
by a sex covariate (h2, Pledger 2000). For example, male home ranges are generally larger than
those for females so the probability that a site would be found by a bear (g0) and the distance
from the activity center that a bear would likely be detected (σ) could differ by sex. We also
evaluated models whereby heterogeneity in detection probabilities was explained by a site-
specific behavioral response (bk) to trap encounter (i.e., “trap-happy” or “trap-shy”). We also
modeled potential differences in g0 during week 1 versus weeks 2–6 to reflect our subsampling
scheme during 2014.
We evaluated a number of land cover variables and other landscape metrics as covariates
for bear density in each study area. We used state-level land use/land cover (LULC) data at 10-
m spatial resolution (i.e., cell size) from FWC and Florida Natural Areas Inventory Cooperative
Land Cover Map v3.1 (FWC and Florida Natural Areas Inventory 2015). We also used
TIGER/Line® roads data (U.S. Bureau of Census 2015); both were processed with ArcMap
(ArcGIS 10.2.2 for Desktop, c 1999-2013 ESRI Inc., www.esri.com).
Because the number of land cover classes in the LULC database was large, we grouped
individual classes into categories that we deemed to be potentially important to bears (e.g., mast-
producing cover). First, we created a “forest” layer consisting of the following classes from the
Florida Land Cover Classification System (Kawula 2014): Upland Hardwood Forest (1110),