deer and browsing in pilot mountain: an evaluation of ...€¦ · 2017 for white-tailed deer...
TRANSCRIPT
DEER AND BROWSING IN PILOT MOUNTAIN: AN EVALUATION OF
POPULATION SURVEY METHODS AND BROWSING PRESSURE IN A NORTH
CAROLINA STATE PARK
BY
ROBERT W. BALDWIN
A Thesis Submitted to the Graduate Faculty of
WAKE FOREST UNIVERSITY GRADUATE SCHOOL OF ARTS AND SCIENCES
in Partial Fulfillment of the Requirements
for the Degree of
MASTER OF SCIENCE
Biology
May 2019
Winston-Salem, NC
Approved by:
T. Michael Anderson, Ph.D., Advisor
Miles R. Silman, Ph.D., Chair
Robert Erhardt, Ph.D.
ii
ACKNOWLEDGEMENTS
I’d first like to thank my advisor, Dr. Michael Anderson for his guidance in experimental
design, analysis methods, thoughtful edits, and selecting the correct brand of Craftsman
pliers. His help has been invaluable for all aspects of my graduate experience. I’d also
like to thank my committee members Dr. Miles Silman and Dr. Robert Erhardt for
providing me with assistance, edits, and enthusiasm. I owe a huge thanks to Dr. Jared
Beaver for his guidance throughout my graduate and undergraduate research. I wouldn’t
be where I am today without Jared and I’m eternally grateful he agreed to show me the
ropes three years.
I’m permanently indebted to the many lab-mates, graduate students, undergraduate
assistants, and technicians that braved the slopes of Pilot Mountain with me; it takes a
village to do any sort of fieldwork, even when the field site is only twenty minutes from
campus. Thank you for helping collect data, providing feedback, and most importantly
keeping a smile on my face when faced with seemingly insurmountable tasks. Lastly, I’d
like to thank my loving family that has supported me every step of my journey in
graduate education.
iii
TABLE OF CONTENTS
List of Illustrations and Tables ........................................................................................v
List of Abbreviations ....................................................................................................... ix
Abstract ..............................................................................................................................x
Introduction: ................................................................................................................... xi
Chapter 1: Evaluating the use of drones equipped with thermal imaging technology
as an effective method for estimating wildlife .................................................................... 1
Abstract ............................................................................................................................2
Introduction ......................................................................................................................4
Study Area .................................................................................................................................. 8
Methods ...........................................................................................................................9
Results ............................................................................................................................13
Discussion ......................................................................................................................14
Management Implications ..............................................................................................19
Acknowledgements ........................................................................................................20
Literature Cited ..............................................................................................................21
Figures and Tables .................................................................................................................. 30
Chapter 2: N-mixture models provide reliable population density estimates of a
large, free-ranging ungulate from camera trap data ........................................................ 35
Abstract ..........................................................................................................................36
Introduction ....................................................................................................................38
Methods .........................................................................................................................42
Results ...........................................................................................................................46
Discussion .....................................................................................................................48
Literature Cited ..............................................................................................................53
Figures and Tables .........................................................................................................63
Supplemental Methods ...................................................................................................68
iv
Chapter 3: Sustained browsing pressure structures vegetation communities in a
North Carolina State Park ...................................................................................................... 70
Abstract ..........................................................................................................................70
Introduction ...................................................................................................................71
Methods .........................................................................................................................74
Results ...........................................................................................................................78
Discussion .....................................................................................................................80
Literature Cited ..............................................................................................................83
Figures and Tables ........................................................................................................90
Supplementary Figures ..................................................................................................96
Curriculum Vitae ............................................................................................................97
v
ILLUSTRATIONS AND TABLES
Chapter 1
Figure 1. Flight path (white lines) and transects flown (highlighted in yellow and
labelled) and at Auburn University deer research facility, Alabama USA, 16–17 March
2017 for white-tailed deer thermal drone surveys.
Figure 2. Screenshot illustrating ESRI’s Full Motion Video add-in tool used to identify
and georeference individuals (or group of animals) detected during aerial thermal drone
flights, within the Auburn University deer research facility in Alabama, USA, 16–17
March 2017. Full Motion Video allowed us to observe the flight footage and current
location of the aircraft along our flight path when marking locations of each individual
and for comparison among observers. Yellow lines indicate programed aircraft flight path
and transects and purple lines indicate actual aircraft flight path. Colored dots indicate
marked white-tailed deer observations.
Figure 3. Thermal white-tailed deer population estimates by flight and observer for
Auburn University deer research facility, Alabama USA, 16–17 March 2017. Estimates
were calculated using 2 separate approaches: transect-specific density and total count.
Boxes represent interquartile range of transect-specific population estimates, midline
estimate, and dashed lines represent the range. Dotted lines indicate the total count
estimates. Slash line indicated known deer population within the high-fenced research
facility.
Figure 4. Kernel-smoothed probability density distribution of recorded white-tailed deer
observation times for surveys conducted at Auburn University deer research facility,
vi
Alabama USA, 16–17 March 2017. Identically shaped curves indicate consistency of
observation between observers.
Figure 5. Thermal images of white-tailed deer observations in both open vegetative
cover and mixed-deciduous forest from an altitude of 100 m above ground level during
both morning (AM; sunrise) and evening (PM; sunset) flights conducted within the
Auburn University deer research facility in Alabama, USA, 16–17 March 2017.
Mornings flights occurred within 0619–0719 and evening flights within 1821–1921.
Confident thermal signatures of animals are marked with solid line circles and dashed
line circles marked unclear signatures. Time (AM vs PM) and cover type for each image:
(a) AM, open; (b) AM, forested; (c) PM, open; (d) PM, forested.
Chapter 2
Figure 1. Locations of camera sites in Pilot Mountain State Park mountain section, North
Carolina USA.
Figure 2. Mean deer population density estimated by N-mixture modelling (NMM),
mark-recapture (MR), and un-manned aerial FLIR (UAI) methods. Error bars represent
95% confidence intervals of estimates.
Figure 3. NMM estimated abundance by topographic aspect of camera site. Boxes
represent interquartile range of abundances, midline represents median, and points
represent outliers.
vii
Figure 4. The NMM estimated mean deer population density of PMSP with increased
survey length. The solid line represents LOESS best fit line to visualize trends. The
shaded region represents the 95% confidence interval of UAI validation. Error bars
represent 95% credible intervals.
Figure 5. Temporal analysis of PMSP deer population dynamics in relation to seasonal
vegetational changes. Three-month bins were used to evaluate seasonal trends in local
populations and resource availability. Green points represent mean NDVI value by
season, black points represent mean deer density. Lines represent LOESS best fit line to
visualize trends. Error bars represent 95% CI on mean best unbiased predictor site
abundance.
Chapter 3
Figure 1. Herbivore exclosure sites at Pilot Mountain State Park, NC USA. Each point
represents a paired fenced and un-fenced plot.
Figure 2. NMDS ordinated using Bray-Curtis dissimilarities between sites. Low stress
(<0.2) indicates the validity of the ordination. Boxplots represent interquartile ranges of
NMDS axis values by treatment. Midlines of boxes represent median values, dashed lines
represent extent and circles represent outliers.
Figure 3. Topographic aspect effects on site composition. Aspect values represent a
gradient from due south (-1) to due north (+1). Aspect was the sole significant predictor
viii
of site composition (β = 38.78, χ2 = 8.81, df = 1, P = 0.003). There was no evidence of a
treatment effect on site composition.
Figure 4. The effects of elevation on site diversity. Elevation values were scaled to
prevent difficulties in model convergence due to large absolute differences in predictors.
Elevation was the sole significant predictor of site diversity (β = 0.81, χ2 = 13.95, df = 1,
P = 0.00019).
Figure 5. Effects of elevation on species evenness. High evenness values represent lower
levels of species dominance. Although elevation was the best fitting predictor of
evenness, this relationship was not significant (β = 0.08, χ2 = 1.24, df = 1, P = 0.27).
Figure 6. Effects of local deer abundance on vegetation structure. AGB was square-root
transformed to conform to the assumptions of normality. AGB significantly decreased
with increased deer abundance (β = -0.39, χ2 = 9.51, df = 1, P = 0.0025).
Figure S1. Caging treatment effect on species evenness. Boxes represent interquartile
range of abundances, midline represents median, and points represent outliers
ix
ABBREVIATIONS
AGB: Aboveground biomass
AGL: Above ground level
AIC: Aikake’s information criteria
AICc: Aikake’s information criteria corrected for small sample size
ANOVA: Analysis of variance
FLIR: Forward-looking infrared
FMV: Full motion video
GSD: Ground sample distance
MODIS: Moderate Resolution Imaging Spectroradiometer
MR: Mark-resight
NDVI: Normalized difference vegetation index
NMDS: Non-metric multidimensional scaling
NMM: N-mixture modelling
PMSP: Pilot Mountain State Park, NC USA
RGB: Red-green-blue
UAI: Unmanned aerial infrared
VLOS: Visual line of sight
x
ABSTRACT
Eastern hardwood forests are becoming increasingly homogenous largely due to the
persistent browse effects of white-tailed deer. Deer browse selectively and place heavy
herbivory pressure on understory vegetation communities when populations are high. In
order to adequately control deer populations, however, managers must first be able to
accurately and reliably estimate populations. Even the best-informed management actions
may fail to mitigate community effects of browsing in areas throughout the eastern
United States due to potential browse legacy effects. The broader objectives of this thesis
were to develop a validation technique for free-ranging populations of deer, compare
different methods of quantifying deer populations using remote photography surveys, and
explore potential legacy effects of chronic herbivory using large herbivore exclosures.
Using drones equipped with thermal imagers, we were able to quantify captive
populations up to 97% accurately, allowing us to effectively validate other methods. All
of the remote photography survey methods investigated generated statistically
indistinguishable results with our drone validations, giving us confidence in all methods.
Lastly, we saw no significant vegetative regeneration in our exclosures, indicating that
historic browse pressure in our study area likely is exhibiting legacy effects on its
vegetation community.
xi
INTRODUCTION
The floral and faunal diversity of eastern hardwood forests (EHFs) are declining
at an alarming rate (Webster et al., 2018). Both understory and overstory communities are
becoming increasingly homogenous across space and do not necessarily reflect their
ecological legacy (Rooney et al., 2004). The continued degradation of these habitats
results in the replacement of endemics adapted to localized pressures with more
generalist species and invasive aliens (Hobbs and Mooney, 1998). In order for natural
resource managers to better protect the biodiversity and endemism of EHFs, a more
comprehensive understanding of the factors contributing to homogenization is needed.
White-tailed deer (Odocoileus virginianus; hereafter deer) are keystone
herbivores in the eastern United States (Waller and Alverson, 1997) and have profound
homogenizing effects on understory composition and structure due to selective browsing
pressure (Rossell et al., 2007; Rossell Jr. et al., 2005). Over-abundance of deer limits
vegetative cover and food availability to many wildlife species found in forested areas
(deCalesta, 1994; Côte et al., 2004) and ultimately reduces both floral and faunal
diversity (Hester et al., 2000; Webster et al., 2005; Rossell et al., 2007). Despite the
extensive literature documenting understory damage caused by deer browse, vegetation
and animal communities are managed independently, with little regard to their
connectivity and trophic relationships (Wisdom et al., 2006). This disconnect between
population management and natural ecological processes is particularly concerning given
the disproportionate role of humans in deer population management.
In order to adequately manage deer populations in a landscape without large
predators, land managers need better understanding of reliability and accuracy of their
xii
baseline population data. Mid-Atlantic EHFs historically were home to a diverse array of
large predators before widespread range contractions from habitat encroachment and
active extirpation by humans greatly limited local predator populations (Laliberte and
Ripple, 2004). Today, gray wolves (Canis lupus), red wolves (Canis rufus) and eastern
cougars (Puma concolor couguar) have been completely eradicated from the area,
whereas black bears (Ursus americanus) only exist in fragmented patches. As a result,
prey populations, particularly deer, have dramatically increased due to reduced predation
pressure (Côte et al., 2004), despite attempted top down control by humans. Hunting is
the most common form of deer population reduction, but low hunter participation,
restrictive bag limits, and short hunting seasons limit hunters from exercising the same
top-down control over deer populations as do natural predators (Brown et al., 2000)
Current population management prioritizes hunter satisfaction (i.e equitable distribution
of antlered harvest and hunter success) over the comprehensive management of browse
effects in many areas (Riley et al., 2003). Yet, even those management efforts targeting
browse effects are further limited by a lack of understanding of the biases associated with
common techniques of population estimation; managers need reliable and accurate
methods to quantify populations in order to establish adequate management goals
(Engeman, 2005).
The broad aim of this thesis was to foster holistic management of both population
management and understory regeneration by evaluating and exploring the population
survey methods and the legacy effects of historically over-browsed areas. My specific
objectives were to 1) establish a validation method for population surveys using
unmanned aerial vehicles, 2) evaluate the efficacy of N-mixture modelling as a reliable
xiii
replacement for traditional camera survey methods, and 3) evaluate the compositional
and structural vegetation changes elicited by sustained browsing pressure at PMSP.
1
CHAPTER 1: EVALUATING THE USE OF DRONES EQUIPPED WITH
THERMAL IMAGING TECHNOLOGY AS AN EFFECTIVE METHOD FOR
ESTIMATING WILDLIFE
JARED T. BEAVER1, Department of Biology and Center for Energy, Environment, and
Sustainability, Wake Forest University, Winston-Salem, NC 27109, USA
ROBERT W. BALDWIN1, Department of Biology and Center for Energy, Environment,
and Sustainability, Wake Forest University, Winston-Salem, NC 27109, USA
MAX MESSINGER, Department of Biology and Center for Energy, Environment, and
Sustainability, Wake Forest University, Winston-Salem, NC 27109, USA
CHAD NEWBOLT, School of Forestry and Wildlife Sciences, Auburn University,
Auburn, AL 36849, USA
STEPHEN S. DITCHKOFF, School of Forestry and Wildlife Sciences, Auburn
University, Auburn, AL 36849, USA
MILES R. SILMAN, Department of Biology and Center for Energy, Environment, and
Sustainability, Wake Forest University, Winston-Salem, NC 27109, USA
1These authors contributed equally to this work
Note: This chapter is formatted according to the guidelines of the Journal of Wildlife
Management. RWB curated data, analyzed data, created figures, and contributed to the
drafting and editing of the manuscript.
2
ABSTRACT Drones equipped with thermal sensors have shown ability to overcome
some of the limitations often associated with traditional human-occupied aerial surveys
(e.g., low detection, high operational cost, human safety risk). However, their accuracy
and reliability as a valid population technique has not been adequately tested. We tested
the effectiveness of using a miniaturized thermal sensor equipped to a drone (thermal
drone) for surveying white-tailed deer (Odocoileus virginianus) populations using a
captive deer population of known abundance at Auburn University’s deer research
facility, Alabama, USA, 16–17 March 2017. We flew 3 flights beginning 30 minutes
prior to sunrise and sunset (1 morning and 2 evening) consisting of 15 non-overlapping
parallel transects (18.8 km) using a small fixed-wing aircraft equipped with a non-
radiometric thermal infrared imager. Deer were identified by 2 separate observers by
their contrast against background thermal radiation and body shape. Our average thermal
drone density estimate (69.8 deer/km2, 95% CI: 52.2–87.6), was 78% of the mean known
value of 90.2 deer/km2, exceeding most sighting probabilities observed with thermal
surveys conducted using human-occupied aircraft. However, during evening flights
thermal contrast was improved and our drone-based density estimate (89.9 deer/km2) was
nearly identical (99.7%) to the mean known value, indicating that time of flight, in
conjunction with local vegetation types, determines video contrast and influences ability
to distinguish deer. Human observers had consistency in the time of observations and
resulting deer density estimates differed only 5% between observers. The method
provides the ability to perform accurate and reliable population surveys in a safe and
cost-effective manner compared to traditional aerial surveys and are only expected to
continue to improve as sensor technology and machine learning analytics continue to
3
advance. Furthermore, the replicability of autonomous flight results in methodology with
superior precision that increases statistical power to detect population trends across
surveys.
KEY WORDS aerial, deer, density estimation, drones, Odocoileus virginianus,
population methods, thermal imaging, wildlife surveys.
4
INTRODUCTION
One of the primary tenets of population management is that abundance and monitoring
information accurately describe the current population state or trend (Collier et al. 2013).
The testing and adoption of methods that provide rigorous, accurate estimates of
population size is not only relevant, but requisite for informed population management of
many wildlife species (Williams et al. 2002, Collier et al. 2013). However, true
population size for most wild populations is unknown (Yoccoz et al. 2001, Hodgson et al.
2016), and unless a survey method has been tested on a population of known size it is not
possible to directly assess the accuracy of any count method.
No single species has had more work focused on population size estimation and
methodological evaluation than the white-tailed deer (Odocoileus virginianus; hereafter
deer; Gill et al. 1997, Lancia et al. 2005, Collier et al. 2013). A variety of survey
techniques have been developed for estimating deer population size: browse surveys
(Aldous 1944, Tremblay et al. 2005), harvest data reconstruction (Roseberry and Woolf
1991, Millspaugh et al. 2009), pellet counts (Eberhardt and Van Etten 1956, Van Etten
and Bennett 1965, Goode et al. 2014), ground-based infrared and thermal imaging
surveys (Wiggers and Beckerman 1993, Gill et al. 1997, Collier et al. 2007), spotlight
surveys (McCullough 1982, Mitchell 1986, DeYoung 2011), and camera surveys
(Jacobson et al. 1997, Koerth and Kroll 2000) among others. However, aerial surveys
remain the best option for counting large mammals, especially over large areas (Caughley
1974, Jachmann 1991, Linchant et al. 2015). Aerial surveys provide more reliable
estimates of population size than ground-based techniques (Naugle et al. 1996, Beaver et
al. 2014) because of high detection rates (Bernatas and Nelson 2004, Millette et al. 2011)
5
and their ability to randomly sample across the landscape, avoiding biases inherent to
road-based sampling (Diefenbach 2005, Drake et al. 2005, Kissell and Nimmo 2011,
Beaver et al. 2014).
Aerial surveys conducted using human-occupied aircraft have been primarily
based on human visual detection (Caughley 1974, Poole et al. 2013, Chrétien et al. 2016),
are logistically difficult to implement, costly (Watts et al. 2010, Linchant et al. 2015), and
pose a health risk for operators (Jones et al. 2006, Watts et al. 2010). Aviation accidents
are the most common cause of work-related death for wildlife biologists in the United
States (Sasse 2003). The risks and costs associated with use of human-occupied aircraft
for aerial surveys makes drones (also known as unmanned aerial vehicles/systems,
remotely piloted aircraft) particularly attractive for wildlife population assessments. Early
use of drones has shown a vast array of diverse ecological applications (e.g. Vermeulen et
al. 2013, Christiansen et al. 2016, Evans et al. 2016, Wich et al. 2016) and the ability to
reduce cost and risk to humans (Watts et al. 2010, Seymour et al. 2017). Additionally,
they provide easier logistics and manipulation than manned aircraft (Linchant et al. 2015)
and avoid disturbances associated with ground surveys (Hodgson et al. 2018). These
benefits have led many practitioners to label drones as a powerful tool for wildlife
ecology (Chabot and Bird 2012, Linchant et al. 2015, Christie et al. 2016, Seymour et al.
2017).
An emerging area in drone research is miniaturized sensors allowing data to be
collected at extremely fine spatial and temporal resolutions highly suited to ecological
applications (Watts et al. 2012, Anderson and Gaston 2013, Messinger et al. 2016,
Hodgson et al. 2018). One of the more popular is the use of commercially available
6
thermal infrared sensors (hereafter thermal; Kissell and Nimmo 2011), providing a high-
contrast method of discriminating endotherms from their surroundings (Stark et al. 2014,
Burke et al. 2019) and allowing for detection of animals at night and in low light
conditions, which is an advantage over conventional multispectral (red-green-blue; RGB)
cameras (Israel et al. 2011, Witczuk et al. 2018). However, previous approaches using
thermal have experienced visibility limitations similar to those found in traditional
photographic methods, with the landscape obscuring animals not only through occlusion,
but also through isothermality (Witczuk et al. 2018).
Early results indicate the use of drones equipped with RGB and/or thermal
cameras in wildlife monitoring may be a viable alternative to typical field methods (e.g.
Christie et al. 2016, Seymour et al. 2017, Hodgson et al. 2018, Linchant et al. 2018,
Witczuk et al. 2018). Furthermore, as drone platforms, sensors and computer vision
techniques develop, the accuracy and cost-effectiveness of drone-based approaches will
also likely improve (Hodgson et al. 2018). Yet, important issues must be resolved prior to
wider drone application, including short flight endurance, optimizing resolution with area
coverage, and regulatory hurdles such as the requirement in many jurisdictions to
maintain visual light of sight (VLOS) with the aircraft at all times (Linchant et al. 2018).
Additionally, methods for estimating wildlife populations remain poorly understood
because most drone-based surveys to date have focused on the plausibility of wildlife
monitoring applications rather than their effectiveness as a viable improvement upon
current survey methods (Linchant et al. 2015).
The primary objective of this study was to determine the effectiveness of using
drones equipped with thermal for surveying wildlife populations by estimating deer
7
density and abundance for a captive deer population of known abundances. We predicted
this method would provide improved detection rates and more reliable population
estimates than current aerial survey methods while also providing biologists and wildlife
managers with a benchmark by which to compare and validate their existing survey
methods. We also identify strengths and weaknesses of the drone-based thermal approach
as a real-time survey tool and suggest paths for improvements and future directions.
8
STUDY AREA
We conducted our study at Auburn University’s deer research facility (Fig. 1) located in
Lee County which lies in the Piedmont region of east-central Alabama, USA. The facility
was constructed in October 2007 and consisted of 174 ha enclosed by 2.5-m steel fence.
The enclosed population was comprised of wild animals captured during the construction
of the facility and their descendants. Vegetation was approximately 40% open fields
maintained for hay production, 13% bottomland hardwoods (various oak [Quercus spp.]),
26% mature, naturally regenerated mixed hardwoods (various oak and hickory [Carya
spp.]) and loblolly pine (Pinus taeda), 11% early regenerated thicket areas consisting
primarily of Rubus spp., sweetgum (Liquidambar styraciflua), eastern red cedar
(Juniperus virginina), and Chinese privet (Ligustrun sinense), and 10% 10–15 year old
loblolly pine. A second order creek bisected the property and provided a stable source of
water year-round. For a more complete description of the facility, see Neuman et al.
(2016) and Newbolt et al. (2017).
9
METHODS
Data Collection
We conducted all surveys over the study area using the Ritewing Drak aircraft, a small
flying-wing style fixed-wing aircraft with a 1.5-m wingspan and maximum takeoff
weight of 4.3 kg. The aircraft was controlled by a Cube Autopilot (Hex Technology Ltd,
Hong Kong, CN) running the ArduPlane software stack (ArduPilot,
http://firmware.ardupilot.org/Plane/, accessed 10 March 2017). Flight paths were
programmed using Mission Planner (Mission Planner Version 1.3,
http://ardupilot.org/planner/, accessed 10 March 2017) using the integral terrain-
following feature to allow for precise, repeatable flights at a fixed altitude above ground
level (AGL). The drone was hand-flown during take-off and landing using the remote-
control system while all other phases of flight were under autopilot control and monitored
using Mission Planner.
The aircraft was equipped with the 640 × 480-pixel non-radiometric thermal
infrared imager (FLIR Vue Pro 640, 13-mm lens, 45° horizontal FOV, 30 hz). The
camera was mounted looking vertically downward to obtain the nadir view, and the
horizontal axis of the camera was aligned perpendicular to the flight path. This sensor
was self-calibrating, with a precision of 0.05 °C (< 50 mk).
Three flights were made in total, with 1 morning flight on 16 March 2017 and 2
evening flights on 16 and 17 March 2017, all of which were initiated 30 minutes prior to
sunrise and sunset. Mean temperature during the morning flight was -2° C while the
mean temperatures during the evening flights were 13° C on 16 March and 21° C on 17
10
March (Auburn Airport weather station, Auburn, AL). Weather was calm (winds <6
km/hr) and clear on both days (Auburn Airport weather station, Auburn, AL).
Flights were performed at 100 m AGL, providing an 84-m horizontal field of
view and a 13-cm ground sample distance. The aircraft flew at a constant speed of
approximately 21 m/s (75 km/hr) in an east-west oriented transect pattern, with 15
parallel transects, totaling 18.8 km in length, spaced evenly 100 m apart across the study
area (Fig. 2). Video and telemetry data were collected continuously in flight. After flight,
telemetry logs from the autopilot were trimmed to correspond with video clips,
reformatted for ingestion into ArcGIS Full Motion Video (FMV; Environmental Systems
Research Institute, Inc., Redlands, CA), and the two were multiplexed in FMV to create
georeferenced video.
Video Analysis
Georeferenced video footage from each of the 3 flights was analyzed at half speed by 2
separate observers to compare the effects of observer performance and flight conditions
on resulting density estimates. Deer were identified by their contrast against background
thermal radiation and body shape. When a thermal signature of an animal (or group of
animals) was detected, the location and time of each observation were marked and
recorded using FMV to mark locations of each individual and ensure density estimates
were generated from observations of the same animals (Fig. 2). We evaluated observation
agreement among observers by binning individual observations within 5 second windows
to control for small differences in observer recognition time of animals and compared
between observers using the kernel-smoothed probability density of observation time
11
using package sm (Bowman and Azzalin 2018) in program R (R Version 3.4.0, www.r-
project.org, accessed 2 Feb 2019).
Population Estimation
Data analysis was conducted in program R using package car (Fox and Weisberg 2011).
We used 2 methods to assess deer abundance from thermal footage. The first used
transects to provide a sample-based density estimate (deer/km2). Deer observations were
summed by flight transect and divided by their respective areas (transect length x
estimated strip width [field of view]) to calculate transect-specific densities per observer
(Naugle et al. 1996). Transect densities were then subsetted into 3 groups of every third
transect so that each used transect was 300 m apart to avoid the possibility of multiple
observations of animals across consecutive transects (i.e. transects for group 1: 1, 4, 7,
10, 13; group 2: 2, 5, 8, 11, 14; group 3: 3, 6, 9, 12, 15). However, it is worth noting that
multiple detection of the same animals on different transects does not introduce bias if
animal movement is random relative to the transect lines (Buckland et al. 2001). We then
calculated a mean density per flight by averaging the 3 mean group densities. This
process was repeated per observer.
The second method to assess abundance used the total number of deer counted per
complete flight. Total count values were adjusted by flight area to account for incomplete
area coverage (88%) of flights. Confidence intervals (95%, CI) of calculated abundances
and densities were generated from a 10,000-iteration bootstrapped distribution of transect
densities. Densities were compared between observers using a two-sample, two-tailed t-
test. Densities were compared by flight, and the interaction of flight and observer using
analysis of variance (ANOVA) with type two sum of squares. Both estimates were
12
conducted without prior knowledge of the known population size.
Accuracy Assessment
The deer population at Auburn University’s deer research facility was intensively
monitored, with capturing and camera surveys occurring every year to track the
population. Individuals within the facility had previously been captured and assigned a
unique number, freeze branded, and tagged in both ears with cattle ear tags (see Neuman
et al. [2016] and Newbolt et al. [2017] for specific capture and handling techniques).
Capture techniques followed American Society of Mammalogists’ guidelines (Sikes and
Gannon 2011) and were in accordance with Auburn University’s Institutional Animal
Care and Use Committee (PRNs 2008-1417, 2008-1421, 2010-1785, 2011-1971, 2013-
2372, 2016-2964, 2016-2985). A number (10–15 individuals) of young-of-the-year deer
were captured and released outside the facility at ~6 months of age each year to control
deer density and maintain appropriate numbers of individuals across age classes. Natural
mortalities further mediated the population density of the facility. The deer within the
facility had no movement restrictions outside of the high fence boundary. Occasionally, a
deer was missed during annual tagging, and the vegetative characteristics and large size
of the facility did not allow for continuous monitoring of all individuals over the course
of the study. As such, we used a combination of methods to estimate our known deer
abundance as was used in Keever et al. (2017). No hunting occurred within the research
facility.
13
RESULTS
At the time of our study, the Auburn deer facility contained approximately 151–163 deer
(86.8–93.7 deer/km2), with a mean value of 157 deer (90.2 deer/km2; Fig. 3). The average
density estimate from the thermal camera equipped drone (hereafter, thermal drone) was
69.8 deer/km2 (95% CI: 52.3–87.40; Fig. 3), resulting in an average sighting probability
of 78% (56–100%) of the known estimate or a correction factor of 1.28. Total abundance
from thermal drone counts, adjusted for incomplete coverage (88%) of the study area,
was 120 or (95% CI: 90.9–142.2; Fig. 3), 56–94% of the known estimate.
Estimated densities differed only 5% between observers and were not
significantly different (P = 0.68). The probability density distributions of observations
differed between observers in the morning flight but the shape and structure were nearly
identical for evening flights (Fig. 4), indicating consistent identification of heat signatures
by separate observers.
Video analysis indicated differing video contrast between morning and evening
flights (Fig. 5). The mean estimated population density was significantly less in the
morning flight (44.4 deer/km2, 95% CI: 28.9–52.4) than in evening flights (89.9
deer/km2, 95% CI 67.2–97.8, F = 3.70, P = 0.05; Fig. 3) and evening flight density and
total abundance estimates (142 deer) were nearly identical (99.7%, 90.4%) to the mean
known density and total abundance values (90.2 deer/km2, 157 deer; Fig. 3). Despite the
differing contrast between timing of the flights, there was no significant difference
between observers (t = -0.47, P = 0.68), or the interaction of flight and observer (F =
0.89, P = 0.43; Fig. 3).
14
DISCUSSION
Performance Assessment
Thermal drone surveys provided an accurate, time efficient, and highly repeatable
method of quantifying deer populations resulting in density estimates and total counts
(89.9 deer/km2, 142 deer) for our high-contrast evening flights nearly identical to known
values (90.2 deer/km2, 157 deer). Averaged thermal drone density estimates had a
sighting probability of 78% of present individuals, exceeding those often reported from
thermal surveys (<75%) conducted using human-occupied aircraft (Parker and Driscoll
1972, Caughley 1974, Bartmann et al. 1986, Beasom et al. 1986), and the high-contrast
evening flights achieved a sighting probability of nearly 100%. Several studies have
found that the detection of deer with thermal imagery from human-occupied aerial flights
offered a varied performance ranging from 37% to 98% (Croon et al. 1968, Wiggers and
Beckerman 1993, Naugle et al. 1996, Potvin and Breton 2005). In fact, detection rates
reported specifically for deer are usually 60–80% under optimal conditions, and below
50% when detection conditions are less favorable (Bartmann et al. 1986, Beasom et al.
1986, DeYoung et al. 1989, Potvin et al. 1992, Chrétien et al. 2016). Recently, some
attempts utilizing drone platforms equipped with either thermal or RGB sensors have
been made to overcome these limitations influencing sighting probability.
The sighting probabilities obtained in this study lead to correction factors ranging
from 0 for the evening flights to 1.28 overall. Linchant et al. (2018) used drones equipped
with RGB for counting hippopotamuses (Hippopotamus amphibious) over a homogenous
landscape and found an almost identical correction factor (1.22) to ours that averaged
both morning and evening flights. However, one advantage of thermal imagers over
15
conventional RGB cameras is the ability to distinguish animals based on their body heat
and increased detection of animals at night and during low light conditions (Burke et al.
2019). Chrétien et al. (2016) found that detection of thermal objects was more accurate
than RGB. Witczuk et al. (2018) showed that terrestrial ungulates could be identified in
both leafless deciduous forests and coniferous forests using thermal cameras. We also
detected deer in all major vegetation types (deciduous/mixed/and coniferous forest, open
areas). However, heat emission is not constant throughout the day (Felton et al. 2010) and
can fluctuate depending on a variety of factors such as weather condition (e.g., ambient
temperature, cloud cover; Burke et al. 2019) and environment (e.g., land cover, aspect;
Franke et al. 2012, Chrétien et al. 2016, Witczuk et al. 2018, Burke et al. 2019), making
timing of surveys a critical study component.
Timing of surveys and topographical complexity can result in unequal warming of
areas (thermal cluttering) mediated by exposure to sunlight (i.e. background thermal
radiation varies across the landscape). This is especially true in areas with objects with
thermal inertia such as boulders and wet ground or water (Burke et al. 2019). Witczuk et
al. (2018) indicated that in forested areas, the contrast between trees and ground can
influence our ability to detect animals, and found that during morning flights, some sun-
exposed tree trunks and branches were just beginning to heat up, resulting in thermal
cluttering and low contrast. We observed that the residual heat captured during the day by
vegetation allowed better distinction of foliage in forested habitats during evening flights,
causing the ground signature to be relatively cooler. Thus, evening images were clearer
and had a homogenous background, such that animal signatures stood out with enough
contrast for relatively easy detection (Witczuk et al. 2018). The accuracy of our data
16
collected during evening flights suggests a correction factor may not be needed when
flight conditions are optimized for a particular area. Conversely, during morning flights,
vegetation was cold due to thermal radiation exchange with the clear sky at night while
the ground remained relatively warm, deer became difficult to distinguish from the
background during this period due to decreased contrast.
In this study we used a non-radiometric sensor that did not allow temperature
thresholds to be set. Thermal exchange between animals and their environment is well
understood (Porter and Gates 1969), and using a radiometric sensor thermal threshold
that can be tailored to the calculated radiative properties of the organism(s) of interest can
greatly increase the thermal contrast between an animal and the terrain and likely
improve accuracy of this technique.
Human Observers
The huge amount of data and the need for multiple observers to minimize potential
observer bias are often described as an important drawback for using drones for wildlife
surveys (Linchant et al. 2015). However, Linchant et al. (2018) noted that experienced
observers needed smaller correction factors to estimate the total population of deer and
that estimates from separate observers quickly converge with small amounts of
experience. Our results support this conclusion as our experienced human observers had
consistency in both the time of deer observations and resulting density estimates of their
respective video analyses. Observer-specific probability density curves of observation
time were nearly identically shaped for high-contrast evening flights, indicating observers
generally counted the same individual animals during the study. However, there was no
difference between flight and observer observations overall, despite the low-contrast
17
morning flight. Meaning during the morning flight that while both observers identified
approximately the same number of deer in lower quality footage, they did so at different
times, suggesting increased false positive observations were made due to the difficulty in
separating the thermal image of the deer from the thermal clutter in the landscape.
Thermal and RGB drone surveys can collect detailed information rapidly
(Linchant 2015), potentially creating scenarios where the time and cost of manual
classification of imagery from aerial surveys becomes exorbitant. Our study area was
small (each flight provided near complete coverage of our study area and the complete
video footage for each survey was approximately 20 minutes only) and our observers
only had 64 total minutes of video footage to analyze. Larger study areas and more
complex data sets might necessitate automated image classification, which would save
time and costs associated with human-observer analysis (Chabot and Francis 2016,
Seymour et al. 2017). However, the development of these technologies is still in its
infancy, and these approaches are of limited use if models must be re-trained for new
datasets or require extensive set up (Seymour et al. 2017). Additionally, Hodgson et al.
(2018) illustrated that while semi-automated counts did streamline the process after
algorithm training, manual counts performed equally well. However, when the number of
subjects is high or repeat counts are required at different time points, labor investment
needed for manual counting can be substantial and the incorporation of machine-learning
capabilities will be more necessary (Chabot and Francis 2016, Burke et al. 2019).
Due to the enclosed nature of our study area and the sole dedication of the facility
to research with deer, we were unable to evaluate the ability to distinguish between
multiple species, a potential limitation of the method in areas with multiple species of
18
large mammals. Further research on characteristics of thermal signatures for other large
land-dwelling mammal species is needed, and future studies should assess and integrate a
correction method that considers the sighting probability of deer for different cover types
(e.g., deciduous forest, evergreen forest, open/field.; Chrétien et al. 2016). With further
development of drone platforms, sensors, and computer vision techniques, drone-based
approaches to wildlife estimation are also expected to continue to improve and become
more widespread (Longmore et al. 2017, Seymour et al. 2017, Hodgson et al. 2018).
19
MANAGEMENT IMPLICATIONS
Autonomous drone-based surveys using thermal sensors allow precisely replicated
surveys and population estimates that equal or exceed traditional airborne techniques.
This in turn gives wildlife managers greater ability to detect relatively small changes in
populations, thereby improving the ability to apply and evaluate management efforts
while simultaneously allowing surveys of difficult terrain and drastic reductions in risk to
personnel. However, for thermal aerial surveys to be successfully applied to wildlife
management and conservation issues, certain aspects of these surveys must be
considered. First, observer bias and experience must be accounted for during data
analysis. The impact of training on observer performance is poorly understood, but
common sense dictates that performance will benefit from training. Second, data quality
will be a function of vegetative cover, topography, and thermal contrast, and the
interactions of these variables will vary with season, time of day, and region. It will be
imperative to have an adequate understanding of the thermal landscape of a study area
and how these variables influence thermal video contrast before attempting thermal drone
surveys. Finally, study area size will influence the successful collection of data because
of technological limitations with drones. Currently, small drones are limited to flights of
several minutes to a few hours depending on the model and payload, forcing managers to
balance coverage area with the minimum resolution for animal identification. However,
expansion of the survey coverage to larger areas can be expected in the near future as
regulations for civilian use of drones are constantly changing. As drone surveys are
expanded to larger areas, data processing and analysis workflow will become of
increasing importance.
20
ACKNOWLEDGMENTS
We thank Wake Forest University’s Center for Energy, Environment and Sustainability,
the Wake Forest Unmanned Systems Laboratory, and the Department of Biology for
financial support and other contributions to this project. Thank you to Patrick Fox for his
assistance conducting flights and ensuring proper and safe flight operations and to Kyle
Blackburn for his assistance with data organization and image selection.
21
LITERATURE CITED
Aldous, S. E. 1944. A deer browse survey method. Journal of Mammalogy 25:130–136.
Anderson, K., and K. J. Gaston. 2013. Lightweight unmanned aerial vehicles will
revolutionize spatial ecology. Frontiers in Ecology and the Environment 11:138–
146.
Bartmann, R. M., L. H. Carpenter, R. A. Garrott, and D. C. Bowden. 1986. Accuracy of
helicopter counts of mule deer in pinyon-juniper woodland. Wildlife Society
Bulletin 14:356–363.
Beasom, S. L., F. G. Leon III, and D. R. Synatzske. 1986. Accuracy and precision of
counting white-tailed deer with helicopters at different sampling intensities.
Wildlife Society Bulletin 14:364–368.
Beaver, J. T., C. A. Harper, R. E. Kissell, L. I. Muller, P. S. Basinger, M. J. Goode, F. T.
Van Manen, W. Winton, and M. L. Kennedy. 2014. Aerial vertical-looking
infrared imagery to evaluate bias of distance sampling techniques for white-tailed
deer. Wildlife Society Bulletin 38:419–427.
Bernatas, S., and L. Nelson. 2004. Sightability model for California bighorn sheep in
canyonlands using forward-looking infrared (FLIR). Wildlife Society Bulletin
32:638–647.
Bowman, A. W., and A. Azzalini. 2018. R package 'sm': nonparametric smoothing
methods. version 2.2-5.6 URL: http://www.stats.gla.ac.uk/~adrian/sm.
Buckland, S. T., D. R. Anderson, K. P. Burnham, J. L. Laake, D. L. Borchers, and L.
Thomas. 2001. Introduction to distance sampling: Estimating abundance of
biological populations. Oxford University Press, Oxford, United Kingdom.
22
Burke, C., M. Rashman, S. Wich, A. Symons, C. Theron, and S. Longmore. 2019.
Optimizing observing strategies for monitoring animals using drone-mounted
thermal infrared cameras. International Journal of Remote Sensing 40:439–467.
Caughley, G. 1974. Bias in aerial survey. The Journal of Wildlife Management 38:921–
933.
Chabot, D., and D. M. Bird. 2012. Evaluation of an off-the-shelf unmanned aircraft
system for surveying flocks of geese. Waterbirds 35:170–174.
Chabot, D., and C. M. Francis. 2016. Computer-automated bird detection and counts in
high- resolution aerial images: A review. Journal of Field Ornithology 87:343–
359.
Chrétien, L.-P., J. Théau, and P. Ménard. 2016. Visible and thermal infrared remote
sensing for the detection of white-tailed deer using an unmanned aerial system.
Wildlife Society Bulletin 40:181–191.
Christiansen, F., A. M. Dujon, K. R. Sprogis, J. P. Y. Arnould, and L. Bejder. 2016.
Noninvasive unmanned aerial vehicle provides estimates of the energetic cost of
reproduction in humpback whales. Ecosphere 7:e01468.
Christie, K. S., S. L. Gilbert, C. L. Brown, M. Hatfield, and L. Hanson. 2016. Unmanned
aircraft systems in wildlife research: Current and future applications of a
transformative technology. Frontiers in Ecology and the Environment 14:241–
251.
Collier, B. A., S. S. Ditchkoff, J. B. Raglin, and J. M. Smith. 2007. Detection probability
and sources of variation in white-tailed deer spotlight surveys. Journal of Wildlife
Management 71:277–281.
23
Collier, B. A., S. S. Ditchkoff, C. R. Ruth, and J. B. Raglin. 2013. Spotlight surveys for
white-tailed deer: Monitoring panacea or exercise in futility? The Journal of
Wildlife Management 77:165–171.
Croon, G. W., D. R. McCullough, C. E. Olsen, and L. M. Queal. 1968. Infrared scanning
techniques for big game censusing. Journal of Wildlife Management 32:751–759.
DeYoung, C. A. 2011. Population dynamics. Pages 147–180 in D. G. Hewitt, editor.
Biology and management of white-tailed deer. CRC Press, Boca Raton, Florida,
USA.
DeYoung, C. A., F. S. Guthery, S. L. Beasom, S. P. Coughlin, and J. R. Heffelfinger.
1989. Improving estimates of white-tailed deer abundance from helicopter
surveys. Wildlife Society Bulletin 17:275–279.
Diefenbach, D. R., 2005. The ability of aerial surveys using thermal infrared imagery to
detect changes in abundance of white-tailed deer on Pennsylvania state forest.
Commonwealth of Pennsylvania, Department of Conservation and Natural
Resources, Bureau of Forestry, Harrisburg, USA.
Drake D., C. Aquila, and G. Huntington. 2005. Counting a suburban deer population
using forward-looking infrared radar and road counts. Wildlife Society Bulletin
33:656–661.
Eberhardt, L., and R. C. Van Etten. 1956. Evaluation of the pellet group count as a deer
census method. The Journal of Wildlife Management 20:70–74.
Evans, L. J., T. H. Jones, K. Pang, S. Saimin, and B. Goossens. 2016. Spatial ecology of
estuarine crocodile (Crocodylus porosus) nesting in a fragmented landscape.
Sensors, 16.
24
Felton, M., K. P. Gurton, J. L. Pezzaniti, D. B. Chenault, and L. E. Roth. 2010. Measured
comparison of the crossover periods for mid- and long- wave IR (MWIR and
LWIR) polarimetric and conventional thermal imagery. Optics Express
18:15704–15713.
Fox, J., and S. Weisberg. 2011. An {R} Companion to Applied Regression. Second
Edition. Thousand Oaks CA: Sage. URL:
http://socserv.socsci.mcmaster.ca/jfox/Books/Companion.
Franke, U., B. Goll, U. Hohmann, and M. Heurich. 2012. Aerial ungulate surveys with a
combination of infrared and high-resolution natural colour images. Animal
Biodiversity and Conservation 35:285–293.
Gill, R. M. A., M. L. Thomas, and D. Stocker. 1997. The use of portable thermal imaging
for estimating deer population density in forest habitats. Journal of Applied
Ecology 34:1273–1286.
Goode, M. J., J. T. Beaver, L. I. Muller, J. D. Clark, F. T. van Manen, C. A. Harper, and
P. S. Basinger. 2014. Capture-recapture of white-tailed deer using DNA from
fecal pellet groups. Wildlife Biology 20:270–278.
Hebblewhite, M., and D. T. Haydon. 2010. Distinguishing technology from biology: A
critical review of the use of GPS telemetry data in ecology. Philosophical
Transactions of the Royal Society of London. Series B: Biological Sciences
365:2303–2312.
Hodgson, J. C., R. Mott, S. M. Baylis, T. T. Pham, S. Wotherspoon, A. D. Kilpatrick, R.
Raja Segaran, I. Reid, A. Terauds, and L. P. Koh. 2018. Drones count wildlife
more accurately and precisely than humans. Methods in Ecology and Evolution
25
9:1160–1167.
Hodgson, J. C., S. M. Baylis, R. Mott, A. Herrod, and R. H. Clarke. 2016. Precision
wildlife monitoring using unmanned aerial vehicles. Scientific Reports 6:1–7.
Israel, M. 2011. A UAV-based roe deer fawn detection system. International Archives of
the Photogrammetry. Remote Sensing and Spatial Information Sciences
38(1/C22):51–55.
Jachmann, H. 1991. Evaluation of four survey methods for estimating elephant densities.
African Journal of Ecology 29:188–195.
Jacobson, H. A., J. C. Kroll, R. W. Browning, B. H. Koerth, and M. H. Conway. 1997.
Infrared-triggered cameras for censusing white-tailed deer. Wildlife Society
Bulletin 25:547–556.
Jones, G. P., L. G. Pearlstine, and H. F. Percival. 2006. An assessment of small
unmanned aerial vehicles for wildlife research. Wildlife Society Bulletin 34:750–
758.
Keever, A. C., C. P. Mcgowan, S. S. Ditchkoff, P. K. Acker, J. B. Grand, and C. H.
Newbolt. 2017. Efficacy of N-mixture models for surveying and monitoring
white-tailed deer populations. Mammal Research 62:413–422.
Kissell, Jr., R. E., and S. K. Nimmo. 2011. A technique to estimate white-tailed deer
density using vertical-looking infrared imagery. Wildlife Biology 17:85–92.
Koerth, B. H., and J. C. Kroll. 2000. Bait type and timing for deer counts using cameras
triggered by infrared monitors. Wildlife Society Bulletin 28:630–635.
Lancia, R. A., W. L. Kendall, K. H. Pollock, and J. D. Nichols. 2005. Estimating the
number of animals in wildlife populations. Pages 106–133 in C. E. Braun, editor.
26
Techniques for wildlife investigations and management, sixth edition. The
Wildlife Society, Bethesda, Maryland, USA.
Linchant, J., J. Lisein, J. Semeki, P. Lejeune, and C. Vermeulen. 2015. Are unmanned
aircraft systems (UASs) the future of wildlife monitoring? A review of
accomplishments and challenges. Mammal Review 45:239–252.
Linchant, J., S. Lhoest, S. Quevauvillers, P. Lejeune, C. Vermeulen, J. Semeki
Ngabinzeke, B. Luse Belanganayi, W. Delvingt, and P. Bouché. 2018. UAS
imagery reveals new survey opportunities for counting hippos. PLoS ONE
13:e0206413.
Longmore, S. N., R. P. Collins, S. Pfeifer, S. E. Fox, M. Mulero-Pázmány, F. Bezombes,
A. Goodwin, M. De Juan Ovelar, J. H. Knapen, and S. A. Wich. 2017. Adapting
astronomical source detection software to help detect animals in thermal images
obtained by unmanned aerial systems. International Journal of Remote Sensing
38:2623–2638.
McCullough, D. R. 1982. Evaluation of night spotlighting as a deer study technique.
Journal of Wildlife Management 46:963–973.
Messinger, M., G. P. Asner, and M. Silman. 2016. Rapid assessments of Amazon forest
structure and biomass using small unmanned aerial systems. Remote Sensing
8(8):615.
Millette, T. L., D. Slaymaker, E. Marcano, C. Alexander, and L. Richardson. 2011.
AIMS-Thermal - a thermal and high resolution color camera system integrated
with GIS for aerial moose and deer census in northeastern Vermont. Alces 47:27–
37.
27
Millspaugh, J. J., J. R. Skalski, R. L. Townsend, D. R. Diefenbach, M. S. Boyce, L. P.
Hansen, and K. Kammermeyer. 2009. An evaluation of sex-age-kill (SAK) model
performance. Journal of Wildlife Management 73:442–451.
Mitchell, W. A. 1986. Deer spotlight census: section 6.4.3, U.S. Army Corp of Engineers
wildlife resources management manual. U.S. Army Engineer Waterways
Experiment Station Technical Report EL-86-53, Vicksburg, Mississippi, USA.
Neuman, T.J., C. H. Newbolt, S. S. Ditchkoff, and T. D. Steury. 2016. Microsatellites
reveal plasticity in reproductive success of white-tailed deer. Journal of
Mammalogy 97:1441–1450.
Newbolt, C. H., P. K. Acker, T. J. Neuman, S. I. Hoffman, S. S. Ditchkoff, ans T. D.
Steury. 2017. Factors influencing reproductive success in male white-tailed deer.
Journal of Wildlife Management 81:206–217.
Naugle, D. E., J. A. Jenks, and B. J. Kernohan. 1996. Use of thermal infrared censing to
estimate density of white-tailed deer. Wildlife Society Bulletin 24:37–43.
Parker, Jr., H. D., and R. S. Driscoll. 1972. An experiment in deer detection by thermal
scanning. Journal of Range Management 25:480–481.
Poole, K. G., C. Cuyler, and J. Nymand. 2013. Evaluation of caribou Rangifer tarandus
groenlandicus survey methodology in West Greenland. Wildlife Biology 19:225–
239.
Porter, W. P., and D. M. Gates. 1969. Thermodynamic equilibria of animals with
environment. Ecological Monographs 39:245–270.
Potvin, F., and L. Breton. 2005. From the field: testing 2 aerial survey techniques on deer
in fenced enclosures—visual double-counts and thermal infrared sensing. Wildlife
28
Society Bulletin 33:317–325.
Potvin, F., L. Breton, L. P. Rivest, and A. Gingras. 1992. Application of a double-count
aerial survey technique for white-tailed deer, Odocoileus virginianus, on Anticosti
Island, Quebec. Canadian Field-Naturalist 106:435–442.
Roseberry, J. L., and A. Woolf. 1991. A comparative evaluation of techniques for
analyzing white-tailed deer harvest data. Wildlife Monographs 3–59.
Sasse, D. B. 2003. Job-related mortality of wildlife workers in the United States, 1937–
2000. Wildlife Society Bulletin 31:1015–1020.
Seymour, A. C., J. Dale, M. Hammill, P. N. Halpin, and D. W. Johnston. 2017.
Automated detection and enumeration of marine wildlife using unmanned aircraft
systems (UAS) and thermal imagery. Scientific Reports 7:1–10.
Sikes R. S., and W. L. Gannon. 2011. Guidelines of the American Society of
Mammalogists for the use of wild mammals in research. Journal of Mammalogy
92:235–253.
Stark, B., B. Smith, and Y. Chen. 2014. Survey of thermal infrared remote sensing for
Unmanned Aerial Systems. IEEE 1294–1299.
Tremblay, J.-P., I. Thibault, C. Dussault, J. Huot, and S. D. Côté. 2005. Long-term
decline in white-tailed deer browse supply: can lichens and litterfall act as
alternative food sources that preclude density-dependent feedbacks. Canadian
Journal of Zoology 83:1087–1096.
Van Etten, R. C., and C. L. Bennett. 1965. Some sources of error in using pellet-group
counts for censusing deer. The Journal of Wildlife Management 29:723–729.
Vermeulen C., P. Lejeune, J. Lisein, P. Sawadogo, P. Bouché. 2013. Unmanned aerial
29
survey of elephants. PLoS ONE 8:e54700.
Watts, A. C., V. G. Ambrosia, and E. A. Hinkley. Unmanned aircraft systems in remote
sensing and scientific research: Classification and considerations of use. Remote
Sensing 4:1671–1692.
Watts, A. C., J. H. Perry, S. E. Smith, M. A. Burgess, B. E. Wilkinson, Z. Szantoi, P. G.
Ifju, and H. F. Percival. 2010. Small unmanned aircraft systems for low-altitude
aerial surveys. Journal of Wildlife Management 74:1614–1619.
Wich, S., and L.P. Koh. 2018. Conservation drones, mapping and monitoring
biodiversity. Oxford University Press, Oxford, United Kingdom.
Wiggers, E. P., and S. F. Beckerman. 1993. Use of thermal infrared sensing to survey
white-tailed deer populations. Wildlife Society Bulletin 21:263–268.
Williams, B. K., J. D. Nichols, and M. J. Conroy. 2002. Analysis and management of
animal populations. Academic Press, San Diego, California, USA.
Witczuk, J., S. Pagacz, A. Zmarz, and M. Cypel. 2018. Exploring the feasibility of
unmanned aerial vehicles and thermal imaging for ungulate surveys in forests -
preliminary results. International Journal of Remote Sensing 39:5504–5521.
Yoccoz, N. G., J. D. Nichols, and T. Boulinier. 2001. Monitoring of biological diversity
in space and time. Trends in Ecology & Evolution 16:446–453.
30
FIGURES AND TABLES
Figure 1. Flight path (white lines) and transects flown (highlighted in yellow and
labelled) and at Auburn University deer research facility, Alabama USA, 16–17 March
2017 for white-tailed deer thermal drone surveys.
31
Figure 2. Screenshot illustrating ESRI’s Full Motion Video add-in tool used to identify
and georeference individuals (or group of animals) detected during aerial thermal drone
flights, within the Auburn University deer research facility in Alabama, USA, 16–17
March 2017. Full Motion Video allowed us to observe the flight footage and current
location of the aircraft along our flight path when marking locations of each individual
and for comparison among observers. Yellow lines indicate programed aircraft flight path
and transects and purple lines indicate actual aircraft flight path. Colored dots indicate
marked white-tailed deer observations.
32
Figure 3. Thermal white-tailed deer population estimates by flight and observer for
Auburn University deer research facility, Alabama USA, 16–17 March 2017. Estimates
were calculated using 2 separate approaches: transect-specific density and total count.
Boxes represent interquartile range of transect-specific population estimates, midline
estimate, and dashed lines represent the range. Dotted lines indicate the total count
estimates. Slash line indicates known deer population within the high-fenced research
facility.
33
Figure 4. Kernel-smoothed probability density distribution of observation time for
surveys conducted at Auburn University deer research facility, Alabama
34
USA, 16–17 March 2017. Identically shaped curves indicate consistency of observation
between observers.
Figure 5. Thermal images of deer observations in both open vegetative cover and mixed-
deciduous forest from an altitude of 100 m above ground level during both morning (AM;
sunrise) and evening (PM; sunset) flights conducted within the Auburn University deer
research facility in Alabama, USA, 16–17 March 2017. Sunrise flights occurred within
0619–0719 and sunset flights within 1821–1921. Confident thermal signatures of animals
are marked with solid line circles and dashed line circles marked unclear signatures. Time
(AM vs PM) and cover type for each image: (a) AM, open; (b) AM, forested; (c) PM,
open; (d) PM, forested.
35
CHAPTER 2: N-MIXTURE MODELS PROVIDE RELIABLE POPULATION
DENSITY ESTIMATES OF A LARGE, FREE-RANGING UNGULATE FROM
CAMERA TRAP DATA
ROBERT W. BALDWIN*, Department of Biology and Center for Energy, Environment,
and Sustainability, Wake Forest University, Winston-Salem, NC 27109
JARED T. BEAVER, Department of Biology and Center for Energy, Environment, and
Sustainability, Wake Forest University, Winston-Salem, NC 27109
MATT WINDSOR, Pilot Mountain State Park, North Carolina State Parks, 1792 Pilot
Knob Park Rd, Pinnacle, NC 27043
MAX MESSINGER, Department of Biology and Center for Energy, Environment, and
Sustainability, Wake Forest University, Winston-Salem, NC 27109
MILES R. SILMAN, Department of Biology and Center for Energy, Environment, and
Sustainability, Wake Forest University, Winston-Salem, NC 27109
T. MICHAEL ANDERSON, Department of Biology and Center for Energy,
Environment, and Sustainability, Wake Forest University, Winston-Salem, NC 27109
*Corresponding author email: [email protected]
Note: This chapter is formatted according to the guidelines of the Journal of Applied
Ecology. RWB helped conceive study, conducted fieldwork, analyzed data, and wrote the
manuscript.
36
ABSTRACT
1. Camera traps are an increasingly popular species monitoring tool yet remain
controversial for estimating populations due to inherent data interpretation
difficulties. The objectives of our study were to compare two commonly used
population estimation analysis methods using camera trap data, N-mixture
modelling and mark-resight, and validate their accuracy with an independent
dataset.
2. We evaluated these methods using observations of white-tailed deer (Odocoileus
virgnianus) from a camera-trapping study at Pilot Mountain State Park, NC USA.
Deer densities derived from unmanned aerial thermal imagery were used to
independently validate both methods. Environmental covariates were included in
N-mixture models to explore spatial abundance relationships and seasonal
abundance models were fit to 18 months of camera observations to explore
temporal dynamics in local populations.
3. N-mixture modelling, mark-resight, and unmanned aerial surveys converged on
similar density estimates, indicating the accuracy of both camera survey methods.
Our N-mixture modelling estimates were robust over a large range of temporal
survey lengths (from 1 – 168 h sample units). Mark-resight methods required less
samples and provided demographic information valuable in management but
required more survey effort to identify individuals and suffered from a lack of
measurable precision.
4. Our N-mixture modelling approach additionally allowed us to robustly measure
spatial and temporal dynamics within our study area. Deer in our study area
37
preferred southern aspects, likely due to the greater amount of browse in these
areas. Temporal patterns showed two-fold seasonal fluctuations of deer density
and suggest our study area acts as a hunting season refuge, presumably due to
tradeoffs between forage availability and mortality risk.
5. Synthesis and applications. Both N-mixture modelling and mark-resight analysis
provide reliable deer population estimates from camera trap data. N-mixture
modelling allows for robust population estimates of commonly occurring species
and allows for the exploration of spatial and temporal hypotheses.
KEYWORDS
Camera trapping, N-mixture modelling, density estimates, population, mark-resight,
North Carolina, spatial occupancy models, white-tailed deer, demographic methods,
imperfect detection
38
INTRODUCTION
Population surveys provide fundamental baseline information for a wide spectrum of
ecological processes, including population dynamics (Freckleton et al., 2005), species
interactions (Vásquez et al., 2007), disease prevalence (Hochachka and Dhondt, 2000),
and biodiversity (Gotelli and Colwell, 2001). However, the accuracy of these estimates
often remains unexplored due to the high cost and effort of validating data collected at
large, population-level, scales.
Remote photography surveys have surged in popularity since the development of
affordable commercially available infrared-triggered camera traps (Cutler and Swann,
1999; Koerth and Kroll, 2000; Rowcliffe and Carbone, 2008). Established cameras serve
as permanent observers, operating continuously in a wide range of environmental and
climatic conditions with minimal human attention (Larrucea et al., 2007). As a result,
camera traps are less invasive, less labor intensive, and more cost effective than other
ecological monitoring techniques (Silveira et al., 2003; Heilbrun et al., 2006; Rowcliffe et
al., 2008). Despite their widespread use, however, generating accurate population
estimates from camera trap data remains a challenge due to inherent sampling error
(Meek et al., 2015). For example, camera traps capture a limited portion of their assumed
sampling area, confounding true absences of the target species with their non-detection
(Burton et al., 2015, Kéry & Royle 2016). Estimates not accounting for ‘imperfect
detection’ will misrepresent target populations and lead to poor management decisions.
Although several analytical methods account for imperfect detection of camera traps
(Rowcliffe et al., 2008; Howe et al., 2017), modified mark and recapture surveys are
those most commonly used by managers.
39
Mark and re-sight surveys (MR) are commonly used in coordination with camera
traps for estimating populations but are limited by the need to identify individuals. The
method applies traditional mark and recapture analysis to camera trap data by using the
photographic capture rates of individual animals to estimate abundance (Burton et al.,
2015). Traditional MR methods operate under two assumptions: that individuals are
correctly identified and have equal probabilities of detection (Pollock et al., 1990). The
necessity of individual identification also restricts the application of MR to species with
visual markings such as large, patterned felids (Karanth & Nichols, 1998) or ungulates
with conspicuous antler patterns (Jacobson et al. 1997). Recent improvements of MR
methods account for heterogenous detection of individuals, space use of individuals, and
imperfect detection (Royle & Young, 2008), but the necessity of individual identification
and fine-scaled movement data prevents their application to many wildlife species and
warrants the exploration of other methods.
Another approach, N-mixture modelling (NMM), uses spatially and temporally
replicated point counts to estimate abundance without the need for identification of
individuals (Royle, 2004; Mackenzie & Royle, 2005). NMM treats the probability of
observing individuals (i.e., ‘detection’) and true occupancy separately by using variation
in site-specific, temporally replicated point counts to estimate the probability that
individuals are detected (p). NMM assumes that variation in point counts are due to p
alone (e.g., Royle, 2004), that populations are closed, there are no false positive species
identifications, that detections are independent, and there is a constant, homogeneous
detection probability for all individuals (Kéry & Royle, 2016). NMM allows for the
incorporation of environmental covariates into both detection and occupancy models and
40
distinguishes non-detection from true absences (Dénes et al., 2015). In doing so, NMMs
allow biologists to investigate temporal and spatial relationships with fine-scale
population fluctuations, allowing for precisely targeted management actions. Although
NMMs have been used to estimate population characteristics of birds, amphibians, and
tropical mammals (Kéry et al. 2005; Ahumada et al., 2013; Ficetola et al. 2018), their
accuracy remains questionable because true population estimates are typically
unavailable for many free-ranging species.
Despite their popularity, NMMs have been recently criticized due to their reliance
on unrealistic assumptions listed above (Barker et al., 2017). Double counting of
individuals, population fluctuations, and heterogenous detection probability constitute
major violations of the assumptions of the NMM approach, yet are unavoidable, to some
degree, within field surveys (Link et al., 2018). Recent simulation studies show that these
violations only marginally affect model fit, but radically change resulting population
estimates (Duarte et al., 2018; Link et al. 2018). Previous validations of NMM have
relied upon simulated data conforming to the underlying assumptions of the model
(Royle, 2004) and do not necessarily validate the model in situations where modelling
assumptions may not be realistic (i.e. free-ranging populations of animals). Further
validation of the NMM approach using alternative techniques is needed in this field.
The broad goals of this study were to compare survey methods using a free-
ranging population of white-tailed deer (Odocoileus virginianus; hereafter deer) and use
resulting estimates to connect spatial and temporal population fluctuations with
ecological processes within a moderately-sized (10 km2) protected area. Deer are
considered keystone herbivores in the eastern United States (Waller & Alverson 1997)
41
and at elevated densities can have profound homogenizing effects on understory
composition and structure due to selective browsing pressure (Rossell et al., 2007;
Rossell Jr. et al., 2005). Male deer, identifiable by their antler characteristics (Jacobson et
al., 1997), serve as an ecologically relevant model for the assessment of both MR and
NMM population methods. MR surveys have been used widely to study deer (Jacobson et
al., 1997; McCoy et al., 2011), but NMMs, although widely used for other species, have
been sparsely applied to deer in a management context due to their potential to violate
modelling assumptions (Keever et al., 2017).
Our approach is novel in that we apply a third method, unmanned aerial vehicles
equipped with infrared imagers (UAI), as a means of validating the MR and NMM
camera survey estimates. The UAI method has been validated using captive populations
of deer in a natural setting (Baldwin et. al, in review), thus providing an ideal benchmark
for comparison. Our study had three specific aims: (1) validate camera survey estimates
with UAI analysis, (2) compare the estimates obtained by N-mixture modelling with
those of MR estimates and (3) use resulting estimates to connect population fluctuations
to spatial and temporal ecological processes at a local scale.
42
MATERIALS AND METHODS
Study Area
This study was conducted on the ~10 km2 (1024 ha) mountain section of Pilot Mountain
State Park (PMSP), NC USA, 36.3425° N, 80.4768° W (Figure 1). The mountain section
of the park consists of a quartzite monadnock that rises 738 meters above sea level and
427 meters above the surrounding hillside. The differing levels of soil moisture and
exposure amongst Pilot Mountain’s south/west and north/east slopes drive vegetational
differences along these aspects. Pine-oak/heath dominate on the south/west facing slopes
as a result of xeric conditions from prevailing southwestern winds and intense sun
exposure. Dominant species include Table Mountain pine (Pinus pungens) and pitch pine
(P. rigida), with an understory of mountain laurel (Kalmia latifolia), and purple
rhododendron (Rhododendron catawbiense). Scrub oak (Quercus ilicifoli), blackjack oak
(Q. marilandica), and chestnut oak (Q. montana) are also present, with an herb layer
comprised mainly of beetleweed (Galax urceolata) and trailing arbutus (also known as
mayflower; Epigaea repens). On more mesic east- and north-facing slopes dominance
shift from pines (P. pungens, P. rigida) to oaks (predominantly Q. montana), with a
poorly-developed herbaceous layer of primarily beetleweed (Galax urceolata; Randall et
al. 1995).
Survey design
Twenty-two camera sites were established throughout the mountain section of
PMSP, creating a camera density of approximately one camera per 0.41 km2 (40.5
hectares). Original site locations were chosen using a systematic design on a randomly
generated grid, with minor adjustments made to minimize slope and ease of camera
43
access (Jacobson et al., 1997). Cuddeback E3 model cameras (Non Typical, Inc., Green
Bay, WI) were used and set with “high” trigger sensitivity and “optimal” detection range.
Camera sites were established in June of 2016 and maintained continuously thereafter;
memory cards were replaced monthly and batteries weekly. Cameras were mounted
facing north to reduce backlighting at sunrise and sunset. Obstructing vegetation was
cleared from the camera view to reduce the likelihood of empty captures; plastic
identification tags were mounted opposite cameras to facilitate site recognition. Photos
from January 1 – March 31, 2018 (90 days) were used to compare models to maintain
assumptions of closed populations and equal detectability of animals with the minimal
canopy coverage necessary for UAI flights. Photos from June 2016 – November 2017
were consolidated and used to investigate temporal trends in NMM estimated population
density.
Image analyses
All photos were initially classified by number of adult males, adult females, fawns, and
individuals of indeterminable demographic characteristics. Subsequently, we applied the
method of Jacobson et al. (1997): the ratio of unique adult males, identified by antler
characteristics, to total adult male captures was applied to the total captures of adult
females and fawns, providing an estimate of unique number of individuals in both
classes. Population density was calculated by dividing the resulting estimates by the total
study area.
Model creation
Captures were tallied in photos collected at least 10 minutes apart to avoid double
counting and aggregated by hour over the course of the 90-day study period (j = 2,160;
44
Royle, 2004). Aggregations of increasing length (1 - 168 h) were used to analyze the
sensitivity of null models to changes in temporal discretization of continuous camera
detections. Elevation data were extracted from the USGS National Elevation Dataset and
used to calculate site-specific slope and aspect (U.S. Geological Survey, 2013). Distance
to edge, distance to road, and distance to trail were calculated for each site. The
hierarchical model used was in the form:
1. 𝑁𝑖 ~ 𝑃𝑜𝑖𝑠𝑠𝑜𝑛(𝜆𝑖); with log(𝜆𝑖) = 𝛽0 + 𝛽1
2. 𝐶𝑖𝑗|𝑁𝑖 ~ 𝐵𝑖𝑛𝑜𝑚𝑖𝑎𝑙(𝑁𝑖, 𝑝𝑖𝑗); with log(𝑝𝑖𝑗) = 𝛼0 + 𝛼1
Where 𝑁 is the latent abundance of site 𝑖 , 𝐶 is the point count at site i and time 𝑗, 𝜆 is
mean abundance and 𝑝 is the detection probability of individuals at site i and time 𝑗.
Covariates (𝛽, 𝛼) are incorporated into the observation and occupancy portions of the
model using log-link functions.
Candidate models were created using environmental covariates in both the
observation and occupancy processes and top model was selected using Aikake’s
information criterion (AIC). An empirical Bayes best unbiased predictor of mean site
abundance was calculated from a simulated posterior distribution of the top performing
model along with 95% credible intervals. Population density was calculated by summing
mean abundance by site and dividing by the area of PMSP. Upon identification and
validation of top-performing model, we used modelling parameters to estimate deer
abundances from September 2016 – December 2017. Observation data were grouped into
three-month windows for modelling in order to control for seasonal differences in
detection probability and animal behavior. Normalized difference vegetation index
(NDVI) values were extracted from 16 day, 250m resolution MODIS data for each
45
camera site and averaged by season to portray vegetation dynamics over time. Data
analysis was performed using packages unmarked (Fiske & Chandler, 2011),
AICcmodavg (Mazerolle, 2017), raster (Hijmans, 2019), and MODIS (Mattiuzzi &
Detsch, 2019) in R version 3.4.0 (R Core Team, 2018).
UAI validation
We conducted five flights using the Linn Aerospace Hummingbird quadcopter equipped
with a 640 x 480-pixel non-radiometric thermal infrared imager (FLIR Vue Pro 640; 13-
mm lens, 45° horizontal FOV, 30 hz). Three morning and two evening flights were
conducted within one hour of sunrise or sunset on February 8 – 13, 2018. All flights were
performed at 120 m above ground level (AGL), providing a 100 m horizontal field of
view. The aircraft flew in an east-west oriented transect pattern, with 12 parallel
transections, totaling 13.75 km in length, spaced evenly 300 m apart across the study area
to avoid possible double counts. Flight coverage area (transect length x width of view)
accounted for 13.5% (1.38 km2ha) of the study area. Video was collected continuously in
flight and was processed and analyzed according to the methods of Baldwin et al. 2019
(SI Methods). Densities were averaged between transects and flights to provide validation
density estimates. An analysis of variance (ANOVA) using type 2 sum of squares was
used to test for significant differences between flight estimates to ensure the reliability of
the benchmark.
46
RESULTS
We obtained 10,234 deer photos over our total 21-month study period and 1695 photos
during our three-month validation period.
Both camera survey methods generated overlapping density estimates with our
UAI validation. The UAI method estimated the deer population density of PMSP to be
31.0 deer / km2 (95% CI: 23.7 – 40.0). There was no significant difference in estimates
between flight-specific estimates (F4,25 = 1.98, P = 0.13), giving us confidence in the
reliability of UAI as a validation benchmark. NMM estimated the population density at
34.6 deer / km2 (95% CI: 27.7 – 42.3), which differed only 26% from the MR estimate of
25.3 deer / km2 . We determined all estimates to be statistically indistinguishable due to
the inclusion of all means within the 95% CIs of the UAI estimate (Figure 2).
Our top performing NMM contained significant predictors within both
observation and occupancy portions of the model. The detection model contained an
interaction between slope and aspect (β = 0.365, SE = 0.128, P = 0.004), which was
expected due to variability in topography and vegetation cover which modified the
sampling area of the camera traps. Aspect was also a significant (β = 0.757, SE = 0.237,
P = 0.0005) driver of site abundance, with higher deer abundances strongly associated
with the southern aspects of the study area (Figure 3).
By grouping the camera trap captures into time periods of differing lengths, we
were able to conduct a sensitivity analysis on how survey length influenced our estimates
of deer density. Different discretization lengths of surveys had no significant effect on
resulting estimates. Increasing length of survey decreased NMM density estimates by up
to 5% from surveys of one hour and approached an asymptote of 25 deer / km2 (Figure
47
4). There were no significant differences amongst mean density estimates and all
discretizations generated density estimates overlapping our UAI validation (Figure 4).
The precision of these estimates, however, increased with survey length.
Our NMM temporal analysis revealed consistent seasonal fluctuations in local
deer density over the course of our 18-month study period. PMSP deer density increased
steadily in summer months, peaked in autumn months, and slowly decreased throughout
the winter and spring – varying inversely with seasonal NDVI patterns (Figure 5).
Although the amplitude of change differed across years, the general trends remained
consistent between years. While camera detection was likely lower during periods of
abundant understory growth and standing biomass (i.e., May – September), our higher
estimates of deer density demonstrate that this did not likely introduce bias into our
results.
48
DISCUSSION
Both camera survey methods used in this study generated population density estimates
that were comparable to our UAI validation, signaling that either method may be
considered a viable option for estimating deer populations. The small differences between
our estimates were likely due to the various assumptions of the respective methods. MR
deer surveys rely on photographic capture of all individual male animals and only
account for imperfect detection by inflating estimates 10% as determined by baited
surveys of captive populations (Jacobson et al.,1997). Proper correction factors have not
been explored for un-baited MR surveys; failure to capture or correctly identify all
individual males present within our study area likely resulted in lower density estimates
than other methods (Figure 3). Likewise, although correction factors improve the
accuracy of resulting parameter estimates, they also conflate observation and site
occupancy, which limits the possible application of parameter estimates to ecological
processes. NMM accounts for imperfect detection by modelling detection probability and
abundance separately (Royle, 2004), which results in more accurate density estimates
given the assumptions of the model are upheld. A recent study validated extensions of
NMMs on large, free-ranging ungulates using telemetry data to estimate immigration and
emigration rates (Ketz et al., 2018), but individual movement data are time intensive,
costly, and rarely available to stakeholders. Our findings demonstrate that basic NMM
(i.e., using model estimation procedures in freely available, open source software) can
provide accurate abundance estimates for free-ranging ungulates and can be used to test
hypotheses of abundance relationships with spatial and temporal predictors.
49
Within our study area topographic aspect influenced deer abundance, thus
highlighting the importance of abundance-space relationships. Topographic aspect
directly impacts soil nutrient and organic matter content due to differing exposure and
moisture regimes (Yimer et al., 2006), resulting in contrasting vegetation community
composition between aspects (Bennie et al., 2006). Deer preferentially select areas of
higher relative nutritive quality when the costs of selection of other available areas are
equivalent (Pauley, 1993). Within our study area, individuals occupied southern slopes
more heavily than northern slopes (Figure 3), which is likely due to higher amounts of
available browse within the mixed forest overstory area (Williams & Oosting, 1944). The
southern slopes of our study area also have easier access to surrounding agricultural
lands, whereas the northern slopes are bounded by more suburban areas. Although habitat
preferences of deer are well-documented and easily explained, our results demonstrate
that NMM can illuminate spatial relationships of abundance in a robust manner to test
hypotheses of space-abundance relationships. These benefits are compounded by the ease
of expanding NMM throughout time to investigate temporal process of local population
dynamics.
Our results suggest that PMSP experiences seasonal fluctuations in deer
population density due to the interplay of risk and resource availability. Ungulates trade-
off mortality risk and forage quality when selecting habitat (Hebblewhite & Merrill,
2009), valuing safety over forage when risks of predation are especially high (Lone et al.,
2015). In the resource-rich agricultural landscapes surrounding our study area, human
recreational hunting pressure poses a substantial risk to deer during the autumn hunting
season. Public lands shelter disproportionate numbers of ungulates to their size in the
50
western united states during hunting seasons due to hunting restriction and prohibition
(Proffitt et al., 2013; DeVoe et al. 2019); our temporal analysis suggests that PMSP
serves as a similar mortality refuge for local deer. Conversely, during the months of the
early growing season, the deer population density of PMSP gradually decreases as
surrounding agricultural lands with higher forage availability become more favorable.
These seasonal fluctuations in population density demonstrate the importance of
considering behavior and spatial configurations of the landscape when interpreting
population survey data. Survey methods and designs must be carefully selected based on
behavioral characteristics of focal species and management goals of estimates (Burton et
al., 2015).
Temporal and spatial replication are necessary for NMM but represent a tradeoff
in practitioner effort (Shannon et al., 2014). Greater spatial replication facilitates the
detection of rare or cryptic species, whereas greater temporal replication allows for more
precise estimates of common species (Mackenzie & Royle, 2005). The similarity of our
estimates results, in part, from detailed knowledge of deer space use; home ranges and
core area sizes of deer are extensively detailed (Stewart et al., 2011). In Appalachian
forests, deer have an average core area of use of 0.24 km2 and home range size of 10 km2
during winter (Campbell et al., 2010), but home ranges and space use are also known to
vary with habitat quality (Marchinton & Hirth, 1984) and season (Nicholson et al., 1997).
Our camera density of 1 camera per 4 km2 minimized the possibility of detecting animals
at more than one camera site within our study area while maintaining adequate coverage
for the capture of individual animals necessary for MR analysis. However, as we cannot
ensure complete closure between our camera sites, our interpretation of abundance
51
changes from absolute abundance to one of ‘use’ (Latif et al., 2015). Site specific
abundances better represent the mean number of animals using a specific area since
double counting of some animals across sites likely resulted in greater NMM estimates.
Although the validation of our NMM approach suggests that minimal double-counting
occurred, the possibility cannot be completely discounted.
Our results demonstrate the robust nature of NMM by showing that different
discretization of survey lengths only marginally affected our resulting estimates. Camera
traps provide a continuous record of detection, but camera surveys are often discretized
for computational and analytical ease (Guillera-Aroita, 2011). Practitioners are advised to
select the smallest possible time unit of ecological inference to prevent unnecessary data
reduction, but the effects of changing discretization periods are poorly understood
(Mackenzie & Royle, 2005). Previous studies show that lengthening discretization
periods increase estimated detection probability while occupancy estimates remain
constant (Steenweg et al., 2016). Our results suggest that increasing the length of our
effective survey periods had only a marginal effect on the resulting abundance estimates
and all survey lengths generated population estimates within the range of our UAI
validation (Figure 4). Hence, NMM presents a flexible option for the population
estimation of commonly occurring species from camera data; practitioners may discretize
camera surveys as determined by data availability without fear of manipulation of
resulting estimates.
The goals of management action should be the determining factor for choosing
camera trap survey methods. MR surveys allow for easier calculation of demographic
information, such as population age and sex structure, compared to NMM due to lower
52
sampling requirements for analysis. MR surveys can be imperative for population
management decisions which oftentimes necessitate demographic targeting (Gordon et
al., 2004). However, calculating precision for MR surveys can be difficult; due to the
logistics involved, replicate survey data are often unavailable. Managers need the ability
to compare their results across space and time in order to evaluate the effects of past
decisions and plan for future action (Engeman, 2005). NMM provide robust estimates of
precision and allow for better comparison of estimates across space and time. A
combination of approaches may be necessary to achieve all management goals of a given
area; our findings suggest that managers may choose either MR and NMM methods of
analysis with confidence that their results accurately reflect the population of their study
area.
53
LITERATURE CITED
Ahumada, J. A., Hurtado, J., & Lizcano, D. (2013). Monitoring the Status and Trends of
Tropical Forest Terrestrial Vertebrate Communities from Camera Trap Data:
A Tool for Conservation. PLOS ONE, 8(9), e73707.
doi:10.1371/journal.pone.0073707
Barker, R. J., Schofield, M. R., Link, W. A., & Sauer, J. R. (2018). On the reliability of
N-mixture models for count data. Biometrics, 74(1), 369–377. doi:
10.1111/biom.12734
Bennie, J., Hill, M. O., Baxter, R., & Huntley, B. (2006). Influence of slope and aspect on
long-term vegetation change in British chalk grasslands. Journal of Ecology,
94(2), 355–368. doi: 10.1111/j.1365-2745.2006.01104.x
Burton, A. C., Neilson, E., Moreira, D., Ladle, A., Steenweg, R., Fisher, J. T., … Boutin,
S. (2015). REVIEW: Wildlife camera trapping: a review and recommendations
for linking surveys to ecological processes. Journal of Applied Ecology, 52(3),
675–685. doi: 10.1111/1365-2664.12432
Campbell, T. A., Laseter, B. R., Ford, W. M., & Miller, K. V. (2004). Feasibility of
localized management to control white-tailed deer in forest regeneration areas.
Wildlife Society Bulletin, 32(4), 1124–1131. doi: 10.2193/0091-
7648(2004)032[1124:FOLMTC]2.0.CO;2
Christie K. S, Gilbert S. L, Brown C. L, Hatfield M., & Hanson L. (2016). Unmanned
aircraft systems in wildlife research: current and future applications of a
transformative technology. Frontiers in Ecology and the Environment, 14(5),
241–251. doi: 10.1002/fee.1281
54
Côté, S. D., Rooney, T. P., Tremblay, J.-P., Dussault, C., & Waller, D. M. (2004).
Ecological Impacts of Deer Overabundance. Annual Review of Ecology,
Evolution, and Systematics, 35(1), 113–147. doi:
10.1146/annurev.ecolsys.35.021103.105725
Cutler, T. L., & Swann, D. E. (1999). Using Remote Photography in Wildlife Ecology: A
Review. Wildlife Society Bulletin (1973-2006), 27(3), 571–581.
deCalesta, D. S. (1994). Effect of White-Tailed Deer on Songbirds within Managed
Forests in Pennsylvania. The Journal of Wildlife Management, 58(4), 711–718.
doi: 10.2307/3809685
Dénes, F. V., Silveira, L. F., & Beissinger, S. R. (2015). Estimating abundance of
unmarked animal populations: accounting for imperfect detection and other
sources of zero inflation. Methods in Ecology and Evolution, 6(5), 543–556. doi:
10.1111/2041-210X.12333
Diefenbach, D. R., Finley, J. C., Luloff, A. E., Stedman, R., Swope, C. B., Zinn, H. C., &
Julian, G. J. S. (2005). Bear and Deer Hunter Density and Distribution on Public
Land in Pennsylvania. Human Dimensions of Wildlife, 10(3), 201–212. doi:
10.1080/10871200591003445
Duarte, A., Adams, M. J., & Peterson, J. T. (2018). Fitting N-mixture models to count
data with unmodeled heterogeneity: Bias, diagnostics, and alternative approaches.
Ecological Modelling, 374, 51–59. doi: 10.1016/j.ecolmodel.2018.02.007
Engeman, R. M. (2005). Indexing principles and a widely applicable paradigm for
indexing animal populations. Wildlife Research, 32(3), 203–210. doi:
10.1071/WR03120
55
Ficetola, G. F., Barzaghi, B., Melotto, A., Muraro, M., Lunghi, E., Canedoli, C., …
Manenti, R. (2018). N -mixture models reliably estimate the abundance of small
vertebrates. Scientific Reports, 8(1), 10357. doi: 10.1038/s41598-018-28432-8
Freckleton, R. P., Gill, J. A., Noble, D., & Watkinson, A. R. (2005). Large-scale
population dynamics, abundance–occupancy relationships and the scaling from
local to regional population size. Journal of Animal Ecology, 74(2), 353–364. doi:
10.1111/j.1365-2656.2005.00931.x
Gordon, I. J., Hester, A. J., & Festa‐Bianchet, M. (2004). REVIEW: The management of
wild large herbivores to meet economic, conservation and environmental
objectives. Journal of Applied Ecology, 41(6), 1021–1031. doi: 10.1111/j.0021-
8901.2004.00985.x
Gotelli, N. J., & Colwell, R. K. (2001). Quantifying biodiversity: procedures and pitfalls
in the measurement and comparison of species richness. Ecology Letters, 4(4),
379–391. doi: 10.1046/j.1461-0248.2001.00230.x
Guillera-Arroita, G., Morgan, B. J. T., Ridout, M. S., & Linkie, M. (2011). Species
Occupancy Modeling for Detection Data Collected Along a Transect. Journal of
Agricultural, Biological, and Environmental Statistics, 16(3), 301–317. doi:
10.1007/s13253-010-0053-3
Halls, L. K. (1984). White-tailed Deer: Ecology and Management. Stackpole Books.
Hebblewhite, M., & Merrill, E. H. (2009). Trade-offs between predation risk and forage
differ between migrant strategies in a migratory ungulate. Ecology, 90(12), 3445–
3454. doi: 10.1890/08-2090.1
56
Heilbrun, R. D., Silvy, N. J., Peterson, M. J., & Tewes, M. E. (2006). Estimating Bobcat
Abundance Using Automatically Triggered Cameras. Wildlife Society Bulletin,
34(1), 69–73. doi: 10.2193/0091-7648(2006)34[69:EBAUAT]2.0.CO;2
Hester, A. J., Edenius, L., Buttenschøn, R. M., & Kuiters, A. T. (2000). Interactions
between forests and herbivores: the role of controlled grazing experiments.
Forestry: An International Journal of Forest Research, 73(4), 381–391. doi:
10.1093/forestry/73.4.381
Hijmans, Robert J. (2019). raster: Geographic Data Analysis and Modeling.
R package version 2.8-19. https://CRAN.R-project.org/package=raster
Hochachka, W. M., & Dhondt, A. A. (2000). Density-dependent decline of host
abundance resulting from a new infectious disease. Proceedings of the National
Academy of Sciences, 97(10), 5303–5306. doi: 10.1073/pnas.080551197
Hodgson, J. C., Mott, R., Baylis, S. M., Pham, T. T., Wotherspoon, S., Kilpatrick, A. D.,
… Koh, L. P. (2018). Drones count wildlife more accurately and precisely than
humans. Methods in Ecology and Evolution, 9(5), 1160–1167. doi: 10.1111/2041-
210X.12974
Howe, E. J., Buckland, S. T., Després‐Einspenner, M.-L., & Kühl, H. S. (2017). Distance
sampling with camera traps. Methods in Ecology and Evolution, 8(11), 1558–
1565. doi: 10.1111/2041-210X.12790
Jacobson, H. A., Kroll, J. C., Browning, R. W., Koerth, B. H., & Conway, M. H. (1997).
Infrared-Triggered Cameras for Censusing White-Tailed Deer. Wildlife Society
Bulletin (1973-2006), 25(2), 547–556.
57
Johnson, A. S., Hale, P. E., Ford, W. M., Wentworth, J. M., French, J. R., Anderson, O.
F., & Pullen, G. B. (1995). White-Tailed Deer Foraging in Relation to
Successional Stage, Overstory Type and Management of Southern Appalachian
Forests. The American Midland Naturalist, 133(1), 18–35. doi: 10.2307/2426344
Karanth, K. U., & Nichols, J. D. (1998). Estimation of Tiger Densities in India Using
Photographic Captures and Recaptures. Ecology, 79(8), 2852–2862. doi:
10.1890/0012-9658(1998)079[2852:EOTDII]2.0.CO;2
Kéry, M., & Royle, J. A. (2016). Applied Hierarchical Modeling in Ecology (Vol. 1).
Elsevier.
Kéry, M., Royle, J. A., & Schmid, H. (2005). Modeling Avian Abundance from
Replicated Counts Using Binomial Mixture Models. Ecological Applications,
15(4), 1450–1461. doi: 10.1890/04-1120
Ketz, A. C., Johnson, T. L., Monello, R. J., Mack, J. A., George, J. L., Kraft, B. R., …
Hobbs, N. T. (2018). Estimating abundance of an open population with an N-
mixture model using auxiliary data on animal movements. Ecological
Applications, 28(3), 816–825. doi: 10.1002/eap.1692
Kissell, R. E., & Nimmo, S. K. (2011). A technique to estimate white-tailed deer
Odocoileus virginianus density using vertical-looking infrared imagery. Wildlife
Biology, 17(1), 85–92. doi: 10.2981/10-040
Koerth, B. H., & Kroll, J. C. (2000). Bait Type and Timing for Deer Counts Using
Cameras Triggered by Infrared Monitors. Wildlife Society Bulletin (1973-2006),
28(3), 630–635.
58
Larrucea, E. S., Brussard, P. F., Jaeger, M. M., & Barrett, R. H. (2007). Cameras,
Coyotes, and the Assumption of Equal Detectability. Journal of Wildlife
Management, 71(5), 1682–1689. doi: 10.2193/2006-407
Latif, Q. S., Ellis, M. M., & Amundson, C. L. (2016). A broader definition of occupancy:
Comment on Hayes and Monfils. The Journal of Wildlife Management, 80(2),
192–194. doi: 10.1002/jwmg.1022
Linchant, J., Lisein, J., Semeki, J., Lejeune, P., & Vermeulen, C. (2015). Are unmanned
aircraft systems (UASs) the future of wildlife monitoring? A review of
accomplishments and challenges. Mammal Review, 45(4), 239–252. doi:
10.1111/mam.12046
Link W. A., Schofield M. R., Barker R. J., & Sauer J. R. (2018). On the Robustness of N‐
mixture models. Ecology, 0(ja). doi: 10.1002/ecy.2362
Lone, K., Loe, L. E., Meisingset, E. L., Stamnes, I., & Mysterud, A. (2015). An adaptive
behavioural response to hunting: surviving male red deer shift habitat at the onset
of the hunting season. Animal Behaviour, 102, 127–138. doi:
10.1016/j.anbehav.2015.01.012
Mackenzie, D. I., & Royle, J. A. (2005). Designing occupancy studies: general advice
and allocating survey effort. Journal of Applied Ecology, 42(6), 1105–1114. doi:
10.1111/j.1365-2664.2005.01098.x
Mattiuzzi, M. & Detsch, F. (2019). MODIS: Acquisition and Processing of
MODIS Products. R package version 1.1.5.
Mazerolle, M. J. (2017) AICcmodavg: Model selection and multimodel inference
basedon (Q)AIC(c). R package version 2.1-1.
59
Mccoy, J. C., Ditchkoff, S. S., & Steury, T. D. (2011). Bias associated with baited camera
sites for assessing population characteristics of deer. The Journal of Wildlife
Management, 75(2), 472–477. doi: 10.1002/jwmg.54
Meek, P. D., Ballard, G., Claridge, A., Kays, R., Moseby, K., O’Brien, T., … Townsend,
S. (2014). Recommended guiding principles for reporting on camera trapping
research. Biodiversity and Conservation, 23(9), 2321–2343. doi: 10.1007/s10531-
014-0712-8
Meek, Paul D., Ballard, G.-A., & Fleming, P. J. S. (2015). The pitfalls of wildlife camera
trapping as a survey tool in Australia. Australian Mammalogy, 37(1), 13–22. doi:
10.1071/AM14023
Nicholson, M. C., Bowyer, R. T., & Kie, J. G. (1997). Habitat Selection and Survival of
Mule Deer: Tradeoffs Associated with Migration. Journal of Mammalogy, 78(2),
483–504. doi: 10.2307/1382900
ORNL DAAC. (2018). MODIS and VIIRS Land Products Global Subsetting and
Visualization Tool. ORNL DAAC, Oak Ridge, Tennessee, USA. Accessed 04 04,
2019. Subset obtained for product at 36.3425° N, 80.4768° W, time period: 01
June 2016 to 16 January 2017, and subset size: 4 x 4 km. doi:
10.3334/ORNLDAAC/1379
Parsons, A. W., Simons, T. R., Pollock, K. H., Stoskopf, M. K., Stocking, J. J., &
O’connell, A. F. (2015). Camera traps and mark-resight models: The value of
ancillary data for evaluating assumptions. The Journal of Wildlife Management,
79(7), 1163–1172. doi: 10.1002/jwmg.931
60
Pauley, G. R., Peek, J. M., & Zager, P. (1993). Predicting White-Tailed Deer Habitat Use
in Northern Idaho. The Journal of Wildlife Management, 57(4), 904–913. doi:
10.2307/3809096
Pollock, K. H., Nichols, J. D., Brownie, C., & Hines, J. E. (1990). Statistical Inference for
Capture-Recapture Experiments. Wildlife Monographs, (107), 3–97.
Proffitt, K. M., Gude, J. A., Hamlin, K. L., & Messer, M. A. (2013). Effects of hunter
access and habitat security on elk habitat selection in landscapes with a public and
private land matrix. The Journal of Wildlife Management, 77(3), 514–524. doi:
10.1002/jwmg.491
Rossell, C. R., Patch, S., & Salmons, S. (2007). Effects of Deer Browsing on Native and
Non-native Vegetation in a Mixed Oak-Beech Forest on the Atlantic Coastal
Plain. Northeastern Naturalist, 14(1), 61–72. doi: 10.1656/1092-
6194(2007)14[61:EODBON]2.0.CO;2
Rossell Jr., C. R., Gorsira, B., & Patch, S. (2005). Effects of white-tailed deer on
vegetation structure and woody seedling composition in three forest types on the
Piedmont Plateau. Forest Ecology and Management, 210(1–3), 415–424. doi:
10.1016/j.foreco.2005.02.035
Rowcliffe, J. M., Field, J., Turvey, S. T., & Carbone, C. (2008). Estimating animal
density using camera traps without the need for individual recognition. Journal of
Applied Ecology, 45(4), 1228–1236. doi: 10.1111/j.1365-2664.2008.01473.x
Royle, J. A., & Young, K. V. (2008). A Hierarchical Model for Spatial Capture–
Recapture Data. Ecology, 89(8), 2281–2289. doi: 10.1890/07-0601.1
61
Royle J. A. (2004). N‐Mixture Models for Estimating Population Size from Spatially
Replicated Counts. Biometrics, 60(1), 108–115. doi: 10.1111/j.0006-
341X.2004.00142.x
Shannon, G., Lewis, J. S., & Gerber, B. D. (2014). Recommended survey designs for
occupancy modelling using motion-activated cameras: insights from empirical
wildlife data. PeerJ, 2. doi: 10.7717/peerj.532
Silveira, L., Jácomo, A. T. A., & Diniz-Filho, J. A. F. (2003). Camera trap, line transect
census and track surveys: a comparative evaluation. Biological Conservation,
114(3), 351–355. doi: 10.1016/S0006-3207(03)00063-6
Steenweg, R., Whittington, J., Hebblewhite, M., Forshner, A., Johnston, B., Petersen, D.,
… Lukacs, P. M. (2016). Camera-based occupancy monitoring at large scales:
Power to detect trends in grizzly bears across the Canadian Rockies. Biological
Conservation, 201, 192–200. doi: 10.1016/j.biocon.2016.06.020
Vázquez, D. P., Melián, C. J., Williams, N. M., Blüthgen, N., Krasnov, B. R., & Poulin,
R. (2007). Species abundance and asymmetric interaction strength in ecological
networks. Oikos, 116(7), 1120–1127. doi: 10.1111/j.0030-1299.2007.15828.x
Waller, D. M., & Alverson, W. S. (1997). The White-Tailed Deer: A Keystone
Herbivore. Wildlife Society Bulletin (1973-2006), 25(2), 217–226.
Webster, C. R., Jenkins, M. A., & Rock, J. H. (2005). Long-term response of spring flora
to chronic herbivory and deer exclusion in Great Smoky Mountains National
Park, USA. Biological Conservation, 125(3), 297–307. doi:
10.1016/j.biocon.2005.03.027
62
Williams, R. M., & Oosting, H. J. (1944). The Vegetation of Pilot Mountain, North
Carolina: A Community Analysis. Bulletin of the Torrey Botanical Club, 71(1),
23–45. doi: 10.2307/2481485
Yimer, F., Ledin, S., & Abdelkadir, A. (2006). Soil organic carbon and total nitrogen
stocks as affected by topographic aspect and vegetation in the Bale Mountains,
Ethiopia. Geoderma, 135, 335–344. doi: 10.1016/j.geoderma.2006.01.005
63
FIGURES AND TABLES
Figure 1. Locations of camera sites in Pilot Mountain State Park mountain section, North
Carolina USA.
64
Figure 2. Mean deer population density estimated by N-mixture modelling (NMM),
mark-recapture (MR), and un-manned aerial infrared (UAI) methods. Error bars represent
95% confidence intervals of estimates.
65
Figure 3. NMM estimated abundance by topographic aspect of camera site. Boxes
represent interquartile range of abundances, midline represents median, and points
represent outliers.
66
Figure 4. The NMM estimated mean deer population density of PMSP with increased
survey length. The solid line represents LOESS best fit line to visualize trends. The
shaded region represents the 95% confidence interval of UAI validation. Error bars
represent 95% credible intervals.
67
Figure 5. Temporal analysis of PMSP deer population dynamics in relation to seasonal
vegetational changes. Three-month bins were used to evaluate seasonal trends in local
populations and resource availability. Green points represent mean NDVI value by
season, black points represent mean deer density. Lines represent LOESS best fit line to
visualize trends. Error bars represent 95% CI on mean best unbiased predictor site
abundance.
68
SUPPLEMENTARY METHODS
Flight parameters
Our Linn Aerospace Hummingbird quadcopter was controlled by a Cube Autopilot (Hex
Technology Ltd, Hong Kong, CN) running the Arduplane software stack. Flight paths
were programmed using Mission Planner (Michael Oborne, 2019). This flight control
system provides for fully autonomous flight and terrain following, allowing for precise,
repeatable flight paths. The drone was hand-flown during take-off and landing using the
Pixhawk flight controller while all other phases of flight were under autopilot control and
monitored using Mission Planner. The aircraft was equipped with a 640 x 480-pixel non-
radiometric thermal infrared imager (FLIR Vue Pro 640; 13-mm lens, 45° horizontal
FOV, 30 hz). The camera was mounted looking vertically downward to obtain the nadir
view and the camera horizontal axis was aligned perpendicular to the flight path. This
sensor is self-calibrating, with a marketed precision of 0.05 °C (<50 mk). Comparisons
with ground-based temperature measurements indicate that it is accurate within +/- 5 °C.
Data processing
Video was collected continuously in flight and after flight, data flash logs from the
autopilot were trimmed to correspond with video clips, reformatted for ingestion into
ArcGIS Full Motion Video (ESRI 2017), and the two were multiplexed in Full Motion
Video (FMV) to create georeferenced video.
Georeferenced video footage from each of the five flights was analyzed at half
speed by three separate observers. Deer were identified by their contrast against
background thermal radiation and body shape. The location and time of each observation
were marked and recorded using the Full Motion Video plugin for ArcGIS to geotag each
69
individual and ensure all observations remained within the known study area. Deer
observations of each observer were summed by flight transect and divided by their
respective areas to calculate transect-specific densities (Baldwin et al., in review).
70
CHAPTER 3: SUSTAINED BROWSING PRESSURE STRUCTURES
VEGETATION COMMUNITIES IN A NORTH CAROLINA STATE PARK
ABSTRACT
White-tailed deer (Odocoileus virginianus) dramatically alter forests through selective
browsing pressure, reducing species diversity, structural complexity, and shifting
dominance towards shade and browse tolerant invasive species. Previous studies suggest
that prolonged, intense browsing may exert considerable legacy effects on vegetative
communities for decades after the alleviation of browsing and may be the cause of
alternative vegetative stable states within eastern forests. This study investigated the
legacy of intensive browsing in a North Carolina State Park using paired herbivore
exclosures. Relationships of ordination axis values, Simpson’s dominance (D2),
Simpson’s evenness, and aboveground biomass with environmental covariates and
exclosure treatments were evaluated using linear mixed models. There were no
significant differences in composition, diversity, or structure between exclosure
treatments, indicating possible browse legacy effects within our study area. Our study
provides baseline data for longitudinal exploration of the vegetative regeneration in
heavily browsed areas.
KEYWORDS
Deer, browsing, species diversity, community composition, vegetative structure,
exclosure, North Carolina
71
INTRODUCTION
Large populations of deer create more simplified and less diverse vegetative communities
through sustained browsing pressure (Nuzzo et al. 2017). In areas of high deer
population, palatable species are rapidly defoliated, resulting in low probabilities of
survival and reproduction (Augustine and Frelich 1998, Augustine and Decalesta 2003,
Côte et al. 2004). Intense browsing shifts community composition away from native
palatable species; long-lived herbaceous perennials are often dispersal-limited and are
quickly lost from heavily browsed communities (Webster et al. 2018), allowing for the
invasion of alien species and reductions of floral diversity (Vavra et al. 2007). As such,
intensive deer browsing comprises a substantial threat to present-day biodiversity
(Rooney & Waller, 2003) These ecological effects of sustained browsing pressure are
compounded by the direct effects of browsing on vegetational structure.
Chronic herbivory of high deer populations reduces structural complexity of
understory vegetation, affecting the faunal composition of the area. High populations of
deer drastically reduce standing palatable vegetation biomass (Webster et al., 2018),
shrinking surviving adult plants in stature and size variance (Knight et al., 2009). Heavy
browsing often suppresses nitrogen-fixing plants, which dramatically alters nutrient
cycling of the surrounding ecosystem and reduces the overall primary productivity of a
given area (Hobbs 1996, Schoenecker et al. 2004). Small mammals rely on vegetational
structural complexity for cover from predators (Rossell & Rossell, 1999; Cote et al.
2004) and bird populations fluctuate in direct accordance with changes in structure
(McShea & Rappolle, 2001; Martin et al., 2011). As such, reductions in vegetation
biomass ultimately reduce the faunal diversity of the area and ultimately result in
72
cascading trophic effects (Nuttle et al. 2014). However, the reduction of vegetative
structural complexity is not a mere product of current browse pressure; areas exhibit
legacy effects of historic browsing pressure decades after its alleviation.
Sustained intensive browsing can create a vegetative legacy of browsing that
extends far past the alleviation of browsing pressure. Browse-tolerant species often
dominate heavily browsed areas and may competitively exclude heavily browsed species
even decades after the alleviation of acute browsing pressure (Carson et al. 2005,
Tanentzap et al. 2012, Nuttle et al. 2014). Reduced light availability, increased soil
compaction, and increased thickness of soil E horizon elicited by long-term browsing
further prevent the re-establishment of understory communities (Sabo et al. 2017).
Additionally, chronic herbivory pressure can also result in the depletion of palatable
native species from the seed bank due to the mortality of adult plants (Beauchamp et al.
2013, Bertiller 1996). Invasion of invasive species may further exacerbate compositional
changes, while remaining masked in quantitative investigations of diversity due to their
replacement of natives (Habeck and Schultz 2015). These compounding effects of
sustained browse pressure can affect successional processes (Stromayer and Warren,
1997) and may ultimately result in alternative vegetative stable states (Augustine et al.,
1998). While the lack of longitudinal data precludes the investigation of successional
states in this initial study, the exclusion of herbivores from certain areas may be used to
investigate both immediate browse effects and the potential for long term regeneration.
Herbivore exclosures are amongst the most popular methods of quantifying white-
tailed deer browse effects on vegetation structure and composition (Horsley et al. 2003).
The technique spatially excludes herbivores from areas using fencing in order to compare
73
vegetation between areas subject to browse and areas alleviated from browsing pressure.
Although herbivore exclosure studies have a long history in the research of vegetation-
herbivore dynamics (eg. Hough, 1949; Anderson & Katz, 1993; Kain et al. 2011), recent
studies have questioned the technique due to its reduction of the gradient of browsing
pressure to a binary response and difficulty in interpreting non-linear relationships
(Rooney & Waller, 2003). Although previous studies have experimentally manipulated
deer populations to investigate browse effects across deer density gradients (Tremblay &
Potvin 2006), a better understanding of the relationship between deer herbivory and
vegetation dynamics is needed across a heterogeneous landscape of herbivory.
The objectives of this study were to assess the legacy effects of sustained deer
browse on understory vegetation of PMSP by comparing the community composition,
diversity, and structure of aboveground plants in herbivore exclosures to those within
deer-accessible plots. Our approach was novel in that we were able to estimate site-
specific deer abundances using an N-mixture modelling, allowing us to explore potential
relationships of browse pressure across a gradient of sustained browsing pressure. We
expected that fenced and un-fenced plots would show no significant differences in
composition and structure due to the short-term nature of our study and the legacy of
sustained intensive browse within our study area.
74
METHODS
Study Area
This study was conducted on the ~10 km2 mountain section of Pilot Mountain State Park
(PMSP), NC USA, 36.3425° N, 80.4768° W. The mountain section of the park consists
of a quartzite monadnock that rises 738 meters above sea level and 427 meters above the
surrounding hillside.
Pilot Mountain’s south/west and north/east slopes exhibit easily observable
vegetational differences. Pine-oak/heath dominate on the south/west facing slopes likely
as a result of xeric conditions from prevailing southwestern winds and intense sun
exposure. Dominant species include table mountain pine (Pinus pungens) and pitch pine
(P. rigida), with an understory of mountain laurel (Kalmia latifolia), and purple
rhododendron (Rhododendron catawbiense). Scrub oak (Quercus ilicifoli), blackjack oak
(Q. marilandica), and chestnut oak (Q. montana) are also present, with an herb layer
comprised mainly of beetleweed (Galax urceolata) and trailing arbutus (also known as
mayflower; Epigaea repens). On more mesic east- and north-facing slopes dominance
shift from pines (P. pungens, P. rigida) to oaks (predominantly Q. montana), with a
poorly-developed herbaceous layer of primarily beetleweed (Galax urceolata).
Prior to the advent of the prescribed burning plan in 2009, PMSP had been fire-
suppressed since the 1940s (M. Windsor, personal communication). Fire scars dating to
1859 reveal an average historical fire return interval of ~ 4 years (D. Cuppinger, personal
communication) before the suppression of fire within the property to preserve logging
interests. The current prescribed burning plan replicates the historic burn interval to
maintain a low fuel load on the property and facilitate the recovery of the understory
75
community. Implementation of burning, however, has failed to regenerate the understory,
except in those areas of the park naturally excluded from deer herbivory by geological
characteristics.
Recent deer population surveys using an N-mixture modelling approach (Royle,
2004) indicate that PMSP supports a mean population density of 35 deer / km2, although
the distribution of that population is not uniform across the landscape (Baldwin et al., in
review). Southern slopes contain relatively higher deer populations than northern slopes
likely due to increased amounts of available forage.
Survey design
Ten pairs of 10m x 10m vegetation plots separated by 2m were established June-July
2018 in a random stratified design (Figure 1). Plots were randomly assigned a fenced or
unfenced treatment. Fenced plots were surrounded by 2m high wire fencing and uncaged
plots were marked by fluorescently painted rebar posts. Each vegetation plot was divided
into four 5m x 5m subplots for sampling.
Vegetation samples were collected in September of 2018. Each subplot was
sampled once at a random location determined by a blind bean-bag toss. An additional
sample was taken at a random location within the plot for a total of five samples per plot.
At each sampling location, community composition was evaluated using a modified
Daubenmire method of coverage estimation (Daubenmire 1953, USDA 1997). Aerial
percent cover for each species present was visually estimated to the nearest 0.5% within a
50 x 50 cm frame. Cardboard cutouts representing 1 – 25% of the total frame area were
used to assist with visual estimation. Note that total cover may exceed 100% because
76
species cover is estimated independently for each species. Aboveground biomass was
sampled at each randomly selected location by clipping all individual plants rooted within
two 0.1 m2 strips bordering the sampling frame. Samples were dried overnight in
convection drying oven and dry mass was measured and classified by plant functional
type for each site.
Data analysis
Environmental covariates (slope, aspect, elevation) were calculated from USGS National
Elevation (USGS, 2019) dataset for inclusion in statistical analyses. All covariates were
scaled for analysis to reduce difficulties in model convergence due to drastic differences
in raw values. Topographic aspect was cosine transformed in order to control for
circularity of measurements; cosine-transformed aspect values represent a gradient from
due south to due north (-1 - +1). Time since last burn was collated by site from prescribed
burn records. Site-specific deer abundances were extracted from the nearest available
survey site as an index of relative browsing pressure. Non-metric multidimensional
scaling (NMDS) was used to quantify community composition along environmental
gradients using Bray-Curtis dissimilarities. This analysis was performed in Program R
version 3.4.0 (R Core Team 2018) using the package vegan (Oksanen et al. 2019). Site
diversity was calculated using Simpsons dominance index (D2) due to its ability to
evaluate changes in diversity in areas with high levels of species dominance (Morris et al.
2014). Simpson’s evenness was also calculated for each plot to test for shifts in
dominance relationships, with lower values representing higher levels of species
dominance in plots.
77
Community composition (NMDS axis values), diversity, and evenness were
modelled as a function of environmental covariates, browsing pressure, fire frequency,
fencing treatment, and biologically relevant interactions using linear mixed models,
treating site as a random effect to control for multiple observations. AGB was square-root
transformed in order to adhere to assumptions of normality. Mixed effect models were
created using the package lme4 (Bates et al. 2015). Top models were chosen from an a
priori determined suite of candidate models using Aikake’s information criteria corrected
for small sample size (AICc) and significance of terms were assessed using Wald χ2
analysis of deviance with type two sum of squares. Models were created, selected and
evaluated for statistical significance using the packages car (Fox & Weisberg 2011) and
AICcmodavg (Mazerolle 2017) in program R version 3.4.0 (R Core Team, 2018).
78
RESULTS
Thirty-two total species were identified from eight of the paired vegetation plots; sites
two and six were completely devoid of vegetation both inside and outside of exclosures,
indicative of the extreme browsing pressure of the area. Plots ranged in species richness
from 2-10 species present. Red maple (Acer rubrum), small black blueberry (Vacinium
tenellum), and mountain laurel were present at the highest number of sites. Japanese
stiltgrass (Microstegium vimineum), an invasive grass, dominated where present.
Our ordination and composition modelling indicated that site composition was
largely driven by topographic aspect. NMDS ordination demonstrated high fit (non-
metric R2 = 0.87) and low stress (0.11), demonstrating its reliability as a quantification of
community composition (Figure S1). Our top model included aspect as the sole
significant predictor of NMDS axis values (β = 38.78, χ2 = 8.81, df = 1, P = 0.003). There
was no evidence of a treatment effect on composition (Figure 3).
Similarly, plot diversity and evenness were best predicted by elevation. Site
diversity significantly increased with elevation (β = 0.81, χ2 = 13.95, df = 1, P = 0.00019)
(Figure 4). There were no other significant predictors of diversity, despite AICc support
for a model containing aspect and elevation as additive fixed effects. Although our top
fitting model for species evenness contained elevation as a covariate (Figure 5), species
evenness did not show a significant relationship with any predictors (β = 0.08, χ2 = 1.24,
df = 1, P = 0.27). Trends from our evenness analysis support increased levels of
dominance with un-fenced treatment (Figure S2) and increased browsing pressure, but
these relationships were not significant.
79
Local browsing pressure drove plot AGB regardless of treatment. Our top model
included deer abundance as the sole predictor of AGB. Biomass significantly decreased
with increased browsing pressure (β = -0.39, χ2 = 9.51, df = 1, P = 0.0025) and this
relationship explained high amounts of variance (Figure 6). Again, there was no evidence
of a treatment effect; there were no significant differences in biomass between fenced and
unfenced plots (β = 0.15, χ2 = 0.064, df = 1, P = 0.80).
80
DISCUSSION
Compositional changes between plots in our study area were only significantly associated
with topographic aspect. Aspect influences the amount of solar radiation an area receives,
resulting in micro-climatic differences in air temperature (Desta et al., 2004), soil
moisture (Bennie et al. 2008), and soil nutrient content (Hicks & Frank, 1984). Southern-
facing slopes receive more direct sunlight than northern facing slopes, resulting in higher
light exposure and lower soil moisture conditions than northern facing slopes. Within our
study area, these moisture and sunlight regimes drive compositional difference between
our plots (Figure 2) and is a more significant predictor than browsing pressure,
management regime, elevation or caging treatment. This demonstrates potential legacy
effects of chronic browsing pressure within the area; sustained browsing pressure at
PMSP has resulted in the homogenous defoliation of palatable species across the
landscape (Rossell et al., 2005). The remaining vegetative communities within our study
area contain species of little nutritional value for deer and hence are structured by
environmental gradients instead of browsing pressure.
Diversity analyses revealed that elevation is the sole significant predictor of both
D2. Site diversity significantly increased at higher elevations (Figure 3), demonstrating
trends consistent with previous studies detailing elevational-diversity gradients. Plant
species diversity has been shown to have a hump-shaped relationship with elevation in
ecosystems around the world; intermediate elevations associated with transitional zones
contain the highest amounts of biodiversity (Lomolino, 2008). Although most studies
focusing on elevational gradients explore diversity trends scaling thousands of meters,
rather than hundreds, subtle differences in elevation can still influence diversity patterns
81
due to micro-climatic and soil chemistry changes (Noss et al., 2014). In our study area,
higher elevations are associated with community shifts from predominantly heath-oak
forest to open canopy pine communities (Williams and Oosting, 1944), due likely to
changes in temperature and soil moisture content with elevation.
Simpson’s evenness follows a similar general pattern as D2, with higher elevation
sites showing lower amounts of species dominance. Although elevation was the best
fitting predictor of evenness, it was not a significant relationship (Figure 4). There were
no significant predictors of plot evenness amongst our environmental covariates and
caging treatments. Unfenced plots did show higher levels of species dominance than plots
inside fencing treatments (Figure S1), potentially indicating the beginning of a treatment
effect. Deer are known to facilitate the dominance of shade tolerant, unpalatable species
and ferns by defoliating palatable species (Webster et al. 2018). In the lower elevations of
our study area, browse-tolerant invasive Japanese stiltgrass outcompetes heavily-browed
species and thrives in areas of high shade and soil moisture. Unless browsing pressure is
alleviated from these areas, it is likely that stiltgrass and other invasive species will
increase their dominance, particularly within the vegetative biomass of the area.
Site-specific deer abundance showed the strongest relationship with AGB,
demonstrating the propensity of browsing pressure to elicit structural vegetation changes.
Higher browsing pressures significantly decreased the AGB of plots due to consistent
removal of palatable forage (Figure 6). In areas with high resident deer populations,
much of the standing biomass <2m high is depleted resulting in a clear browse line
designating the physiological limits of deer browse (Bressette et al., 2012). Heavily
browsed areas, such as our study area, may be primed to maintain high deer populations
82
at the expense of the system due to complex aboveground and belowground trophic
cascades resulting from original reductions of biomass (Bressette et al., 2012). As a
result, areas with chronic herbivory pressure can experience similar legacy effects of
browse decades after the alleviation of browsing (White, 2012).
The lack of a treatment effect within any of our responses indicate that PMSP
may either be subject to legacy effects from long-term deer overpopulation or that the
timing of our fence installation did not sufficiently protect plants during the growing
season. As we expected, we saw no fencing treatment effects in any of our measured
response variables, which may indicate that the legacy of sustained browse at our study
area will hinder vegetative recovery despite active management efforts (Banta et al.,
2005). Given the clear browse line at our study area, and its function as a hunting refuge
during autumn months (Baldwin et al., chapter 2), it likely that the vegetative
communities of PMSP will exhibit substantial legacy effects for the near future.
However, our herbivore exclosures were established only three months prior to
vegetation sampling and were not installed before the growing season. Early vegetative
growth provides highly nutritious forage (Blair et al. 1977) and can be depleted at the
seedling life stages in areas of particularly high browsing pressure (Cote et al. 2004). In
order to evaluate these hypotheses and determine the regenerative potential of PMSP, this
study should serve as community baseline for comparison to annual collections of
vegetation data.
83
LITERATURE CITED
Anderson, R. C., & Katz, A. J. (1993). Recovery of browse-sensitive tree species
following release from white-tailed deer Odocoileus virginianus zimmerman
browsing pressure. Biological Conservation, 63(3), 203–208. doi: 10.1016/0006-
3207(93)90713-B
Augustine, D. J., & Decalesta, D. (2003). Defining deer overabundance and threats to
forest communities: From individual plants to landscape structure. Écoscience,
10(4), 472–486. doi: 10.1080/11956860.2003.11682795
Augustine, D. J., & Frelich, L. E. (n.d.). Effects of White-Tailed Deer on Populations of
an Understory Forb in Fragmented Deciduous Forests. Conservation Biology, 12(5),
995–1004. doi: 10.1046/j.1523-1739.1998.97248.x
Banta, J. A., Royo, A. A., Kirschbaum, C., & Carson, W. P. (2005). Plant communities
growing on boulders in the Allegheny National Forest: Evidence for boulders as
refugia from deer and as a bioassay of overbrowsing. Natural Areas Journal. 25(1):
10-18., 25(1), 10–18.
Bennie, J., Huntley, B., Wiltshire, A., Hill, M. O., & Baxter, R. (2008). Slope, aspect and
climate: Spatially explicit and implicit models of topographic microclimate in chalk
grassland. Ecological Modelling, 216(1), 47–59. doi:
10.1016/j.ecolmodel.2008.04.010
Bergström, R., & Edenius, L. (2003). From twigs to landscapes – methods for studying
ecological effects of forest ungulates. Journal for Nature Conservation, 10(4), 203–
211. doi: 10.1078/1617-1381-00020
84
Blair, R. M., Short, H. L., & Epps, E. A. (1977). Seasonal Nutrient Yield and
Digestibility of Deer Forage from a Young Pine Plantation. The Journal of Wildlife
Management, 41(4), 667–676. doi: 10.2307/3799988
Bressette, J. W., Beck, H., & Beauchamp, V. B. (2012). Beyond the browse line:
complex cascade effects mediated by white-tailed deer. Oikos, 121(11), 1749–1760.
doi: 10.1111/j.1600-0706.2011.20305.x
Côté, S. D., Rooney, T. P., Tremblay, J.-P., Dussault, C., & Waller, D. M. (2004).
Ecological Impacts of Deer Overabundance. Annual Review of Ecology, Evolution,
and Systematics, 35(1), 113–147. doi: 10.1146/annurev.ecolsys.35.021103.105725
Daubenmire, R.F. (1959). A canopy cover method of vegetational analysis. Northwest
Science. 33: 43-46
Desta, F., Colbert, J. J., Rentch, J. S., & Gottschalk, K. W. (2004). Aspect Induced
Differences in Vegetation, Soil, and Microclimatic Characteristics of an
Appalachian Watershed. Castanea, 69(2), 92–108. doi: 10.2179/0008-
7475(2004)069<0092:AIDIVS>2.0.CO;2
Habeck, C. W., & Schultz, A. K. (2015). Community-level impacts of white-tailed deer
on understorey plants in North American forests: a meta-analysis. AoB PLANTS, 7.
doi: 10.1093/aobpla/plv119
Hicks, R. R., & Frank, P. S. (1984). Relationship of aspect to soil nutrients, species
importance and biomass in a forested watershed in West Virginia. Forest Ecology
and Management, 8(3), 281–291. doi: 10.1016/0378-1127(84)90060-4
Hobbs, N. T. (1996). Modification of Ecosystems by Ungulates. The Journal of Wildlife
Management, 60(4), 695–713. doi: 10.2307/3802368
85
Horsley, S. B., Stout, S. L., & deCalesta, D. S. (2003). White-Tailed Deer Impact on the
Vegetation Dynamics of a Northern Hardwood Forest. Ecological Applications,
13(1), 98–118. doi: 10.1890/1051-0761(2003)013[0098:WTDIOT]2.0.CO;2
Hough, A. F. (1949). Deer and Rabbit Browsing and Available Winter Forage in
Allegheny Hardwood Forests. The Journal of Wildlife Management, 13(1), 135–141.
doi: 10.2307/3796131
Kain, M., Battaglia, L., Royo, A., & Carson, W. P. (2011). Over-browsing in
Pennsylvania creates a depauperate forest dominated by an understory tree: Results
from a 60-year-old deer exclosure. The Journal of the Torrey Botanical Society,
138(3), 322–326. Retrieved from JSTOR.
Knight, T. M., Caswell, H., & Kalisz, S. (2009). Population growth rate of a common
understory herb decreases non-linearly across a gradient of deer herbivory. Forest
Ecology and Management, 257(3), 1095–1103. doi: 10.1016/j.foreco.2008.11.018
Lomolino, M. V. (2001). Elevation gradients of species-density: historical and
prospective views. Global Ecology and Biogeography, 10(1), 3–13. doi:
10.1046/j.1466-822x.2001.00229.x
Martin, T. G., Arcese, P., & Scheerder, N. (2011). Browsing down our natural heritage:
Deer impacts on vegetation structure and songbird populations across an island
archipelago. Biological Conservation, 144(1), 459–469. doi:
10.1016/j.biocon.2010.09.033
Mazerolle, M. J. (2017) AICcmodavg: Model selection and multimodel inference
basedon (Q)AIC(c). R package version 2.1-1.
86
McShea, W. J., & Rappole, J. H. (2000). Managing the Abundance and Diversity of
Breeding Bird Populations through Manipulation of Deer Populations. Conservation
Biology, 14(4), 1161–1170. doi: 10.1046/j.1523-1739.2000.99210.x
Morris, E. K., Caruso, T., Buscot, F., Fischer, M., Hancock, C., Maier, T. S., … Rillig,
M. C. (n.d.). Choosing and using diversity indices: insights for ecological
applications from the German Biodiversity Exploratories. Ecology and Evolution,
4(18), 3514–3524. doi: 10.1002/ece3.1155
Noss, R. F., Platt, W. J., Sorrie, B. A., Weakley, A. S., Means, D. B., Costanza, J., &
Peet, R. K. (2015). How global biodiversity hotspots may go unrecognized: lessons
from the North American Coastal Plain. Diversity and Distributions, 21(2), 236–
244. doi: 10.1111/ddi.12278
Nuttle, T., Ristau, T. E., & Royo, A. A. (2014). Long-term biological legacies of
herbivore density in a landscape-scale experiment: forest understoreys reflect past
deer density treatments for at least 20 years. Journal of Ecology, 102(1), 221–228.
doi: 10.1111/1365-2745.12175
Nuzzo, V., Dávalos, A., & Blossey, B. (2017). Assessing plant community composition
fails to capture impacts of white-tailed deer on native and invasive plant species.
AoB PLANTS, 9(4). doi: 10.1093/aobpla/plx026
Rooney, T. P., & Waller, D. M. (2003). Direct and indirect effects of white-tailed deer in
forest ecosystems. Forest Ecology and Management, 181(1), 165–176. doi:
10.1016/S0378-1127(03)00130-0
Rossell, C. Reed, Gorsira, B., & Patch, S. (2005). Effects of white-tailed deer on
vegetation structure and woody seedling composition in three forest types on the
87
Piedmont Plateau. Forest Ecology and Management, 210(1), 415–424. doi:
10.1016/j.foreco.2005.02.035
Rossell, C.R., & Rossell, I. M. (1999). Microhabitat selection by small mammals in a
southern Appalachian fen in the USA. Wetlands Ecology and Management, 7(4),
219–224. doi: 10.1023/A:1008473029564
Royle J. Andrew. (2004). N‐Mixture Models for Estimating Population Size from
Spatially Replicated Counts. Biometrics, 60(1), 108–115. doi: 10.1111/j.0006-
341X.2004.00142.x
Royo Alejandro A., Collins Rachel, Adams Mary Beth, Kirschbaum Chad, & Carson
Walter P. (2010). Pervasive interactions between ungulate browsers and disturbance
regimes promote temperate forest herbaceous diversity. Ecology, 91(1), 93–105. doi:
10.1890/08-1680.1
Sabo, A. E., Frerker, K. L., Waller, D. M., & Kruger, E. L. (2017). Deer‐mediated
changes in environment compound the direct impacts of herbivory on understorey
plant communities. Journal of Ecology, 105(5), 1386–1398. doi: 10.1111/1365-
2745.12748
Schoenecker, K. A., Singer, F. J., Zeigenfuss, L. C., Binkley, D., Menezes, R. S. C., &
Krausman. (2004). Effects of elk herbivory on vegetation and nitrogen processes.
Journal of Wildlife Management, 68(4), 837–849. doi: 10.2193/0022-
541X(2004)068[0837:EOEHOV]2.0.CO;2
Stromayer, K. A. K., & Warren, R. J. (1997). Are Overabundant Deer Herds in the
Eastern United States Creating Alternate Stable States in Forest Plant Communities?
Wildlife Society Bulletin (1973-2006), 25(2), 227–234.
88
Tanentzap, A. J., Kirby, K. J., & Goldberg, E. (2012). Slow responses of ecosystems to
reductions in deer (Cervidae) populations and strategies for achieving recovery.
Forest Ecology and Management, 264, 159–166. doi: 10.1016/j.foreco.2011.10.005
Thébault, E., & Fontaine, C. (2010). Stability of Ecological Communities and the
Architecture of Mutualistic and Trophic Networks. Science, 329(5993), 853–856.
doi: 10.1126/science.1188321
Tremblay, J.-P., Huot, J., & Potvin, F. (2006). Divergent nonlinear responses of the
boreal forest field layer along an experimental gradient of deer densities. Oecologia,
150(1), 78–88. doi: 10.1007/s00442-006-0504-2
Vavra, M., Parks, C. G., & Wisdom, M. J. (2007). Biodiversity, exotic plant species, and
herbivory: The good, the bad, and the ungulate. Forest Ecology and Management,
246(1), 66–72. doi: 10.1016/j.foreco.2007.03.051
Webster, C. R., Dickinson, Y. L., Burton, J. I., Frelich, L. E., Jenkins, M. A., Kern, C. C.,
… Willis, J. L. (2018a). Promoting and maintaining diversity in contemporary
hardwood forests: Confronting contemporary drivers of change and the loss of
ecological memory. Forest Ecology and Management, 421, 98–108. doi:
10.1016/j.foreco.2018.01.010
Webster, C. R., Dickinson, Y. L., Burton, J. I., Frelich, L. E., Jenkins, M. A., Kern, C. C.,
… Willis, J. L. (2018b). Promoting and maintaining diversity in contemporary
hardwood forests: Confronting contemporary drivers of change and the loss of
ecological memory. Forest Ecology and Management, 421, 98–108. doi:
10.1016/j.foreco.2018.01.010
89
White, M. A. (2012). Long-term effects of deer browsing: Composition, structure and
productivity in a northeastern Minnesota old-growth forest. Forest Ecology and
Management, 269, 222–228. doi: 10.1016/j.foreco.2011.12.043
Williams, R. M., & Oosting, H. J. (1944). The Vegetation of Pilot Mountain, North
Carolina: A Community Analysis. Bulletin of the Torrey Botanical Club, 71(1), 23–
45. doi: 10.2307/2481485
Wisdom, M. J., Vavra, M., Boyd, J. M., Hemstrom, M. A., Ager, A. A., & Johnson, B. K.
(2006). Understanding Ungulate Herbivory–Episodic Disturbance Effects on
Vegetation Dynamics: Knowledge Gaps and Management Needs. Wildlife Society
Bulletin, 34(2), 283–292. doi: 10.2193/0091-7648(2006)34[283:UUHDEO]2.0.CO;2
90
FIGURES AND TABLES
Figure 1. Herbivore exclosure sites at Pilot Mountain State Park, NC USA. Each point
represents a paired fenced and un-fenced plot.
91
Figure 2. NMDS ordinated using Bray-Curtis dissimilarities between sites. Low stress
(<0.2) indicates the validity of the ordination. Boxplots represent interquartile ranges of
NMDS axis values by treatment. Midlines of boxes represent median values, dashed lines
represent extent and circles represent outliers.
92
Figure 3. Topographic aspect effects on site composition. Aspect values represent a
gradient from due south (-1) to due north (+1). Aspect was the sole significant predictor
of site composition (β = 38.78, χ2 = 8.81, df = 1, P = 0.003). There was no evidence of a
treatment effect on site composition.
93
Figure 4. The effects of elevation on site diversity. Elevation values were scaled to
prevent difficulties in model convergence due to large absolute differences in predictors.
Elevation was the sole significant predictor of site diversity (β = 0.81, χ2 = 13.95, df = 1,
P = 0.00019).
94
Figure 5. Effects of elevation on species evenness. High evenness values represent lower
levels of species dominance. Although elevation was the best fitting predictor of
evenness, this relationship was not significant (β = 0.08, χ2 = 1.24, df = 1, P = 0.27).
95
Figure 6. Effects of local deer abundance on vegetation structure. AGB was square-root
transformed to conform to the assumptions of normality. AGB significantly decreased
with increased deer abundance (β = -0.39, χ2 = 9.51, df = 1, P = 0.0025).
96
SUPPLEMENTARY FIGURES
Figure S1. Caging treatment effect on species evenness. Boxes represent interquartile
range of abundances, midline represents median, and points represent outliers.
97
CURRICULUM VITAE
Robert W. Baldwin
Winston-Salem, NC | [email protected] | 312.316.7050
Education
2019 M.S. Biology, Wake Forest University, Winston-Salem, NC
Thesis: Deer and browsing in Pilot Mountain: an exploration of population
estimation and browse effects of white-tailed deer at a North Carolina state park.
Advisor: Dr. T. Michael Anderson
PA: 3.9
2017 B.S. Biology, Wake Forest University, Winston-Salem, NC, May 2017
Bachelor of Science in Biology, minors in Economics and Environmental Science
Thesis: An evaluation of camera survey methods for white-tailed deer
(Odocoileus virginianus).
Advisor: Dr. Miles R. Silman
GPA: 3.5
Honors in Biology, Cum laude
Research Interests
My research pursues a combination of applied and basic science questions about
the population and spatial ecology of large ungulates. Specifically, I’ve evaluated
different methods of quantifying populations and investigated vegetative
community browse effects in overpopulated areas.
98
Research Experience
2017 – 2019 M.S Research, Pilot Mountain State Park, NC, USA
2016 – 2017 Undergraduate Research Assistant, Pilot Mountain State Park, NC,
USA
Selected Honors, Awards, and Fellowships
2018 Elton C. Cocke Travel Award
2017 Carolina Biological Supply Award for Best Undergraduate Research
Conference Presentations and Posters
2018 Southeastern Deer Study Group, Nashville, TN
2017 Undergraduate Honors and Awards, Winston-Salem, NC
2017 Poster, Southeastern Deer Study Group, St. Louis, MO
Teaching Experience
2017 – 2019 Teaching Assistant, Introduction to Evolutionary Biology and Ecology,
Department of Biology, Wake Forest University
2016 – 2017 Econometrics Tutor, Wake Forest Learning Assistance Center, Wake
Forest University
Extramural Professional Activities
Peer reviewer for Journal of Wildlife Management