Designing landscapes for bird conservation in the Southeastern United States
by
Allison Theresa Moody
A dissertation submitted to the Graduate Faculty of
Auburn University
in partial ful llment of the
requirements for the Degree of
Doctor of Philosophy
Auburn, Alabama
August 4, 2012
Keywords: conservation priorization, birds, climate change
Copyright 2012 by Allison Theresa Moody
Approved by
James B. Grand, Professor of Forestry and Wildlife Sciences
Jaime A. Collazo, Professor of Biology and Forestry
Craig Guyer, Professor, Biological Sciences
Geo rey E. Hill, Professor of Biology
Michael S. Mitchell, Adjunct Research Professor of Wildlife Conservation
Abstract
The Southeastern United States has high conservation importance because of the re-
gion?s habitat and species diversity, ecological processes, and evolutionary potential; however,
it also warrants strong concern because of historical habitat loss, future threats, and inad-
equate protection. Based on data from United States Geological Survey, we estimate that
only 12% of Southern Coastal Plain ecoregion is under permanent protection. Conserva-
tion planning for this large and varied ecoregion is complicated by variable data availability
across regions, habitats, and species. As part of a larger project to ensure adequate cur-
rent and future habitat for bird populations, I de ned a suite of focal species; developed
a method to determine conservation priorities; and integrated future land cover conditions
into conservation priorities.
For my  rst chapter, I elicited expert knowledge of specieshabitat associations in order
to de ne a suite of focal species for specieshabitat modeling. I wanted to use multiple focal
species to reduce the risk of missing endemic or range-restricted species, to include species
with substantial public interest or conservation resources, and to represent all habitat types
in the study region. Fifty-three experts attended elicitation meetings and were asked to iden-
tify and score the habitat characteristics required for each potential focal species. I used two
selection methods to develop focal species lists based on expert knowledge. The Lambeck
method systematically selected species based on their threat category and the structured de-
cision making process based on species with non-overlapping habitat associations. I assessed
the overall list composed of species on both lists using an online survey. From online re-
sponses, I added 11 species to the focal species list which we then used to model conservation
priorities in the Southeastern US.
ii
In order to prioritize large areas for conservation, I developed a process that integrates
spatial reserve design principles including prioritizing vegetation patches that are large, round
and close to other patches. I compare the results of this prioritization process using three
di erent conservation proxies: vegetation types, focal species, and focal species values (fsv)
derived from online expert elicitation. Three binary grids were used to develop priority
surfaces based on vegetation type suitability, conservation lands, and urban. The other two
prioritization methods used focal species to identify priority areas by using additional species-
speci c datasets potential habitat and putative source populations. We used the density
of each binary grid, calculated by a two-dimensional kernel density estimator, to calculate
conservation priority for each location in a regular 200m grid across the entire SAMBI
area. Using only vegetation type density to create conservation priority maps resulted in
more high conservation priority areas compared to focal species prioritization except for the
most restricted vegetation types, such as those that were maritime-associated. Conservation
priority surfaces created using focal species and fsv were very similar. Using vegetation type
alone to create priority surfaces required fewer data and the data are more readily available
(all sourced from publically available datasets), but it did not re ect species habitat use
making it problematic for conservation e orts targeted at species.
Finally, in order to provide a tool to enable stakeholders to conserve species and habi-
tats that are currently present and to integrate future habitat conditions to allow species to
respond to climate change, I designed conservation priority areas for two habitats, open pine
and maritime forest, that are expected to respond to di erent aspects of climate change, in-
creased fragmentation and sea level rise, respectively. Land cover projections were developed
for years 2000 to 2100 at 10-year time intervals for three global climate change models. We
included  ve binary spatio-temporal grids to prepare habitat priority maps: (1) potential
habitat and (2) putative source population distributions for each of the focal species; (3)
suitability models for each habitat; (4) conservation lands; and (5) urban areas. Overall pri-
ority surfaces were created by combining priority surfaces from each time interval. For both
iii
habitats, di erences between priority surfaces created with discounted or summed future
conditions a ected how valuable areas were to conservation but not where those areas were
within the region, and surfaces did not di er signi cantly between climate scenarios. Similar-
ities among alternatives of future conditions may be a result of scale because climate change
may have a strong local, but weak regional e ect. Having six similar alternatives suggests a
set of consistent conservation priorities that can be relied upon to conserve bird populations
in the study region. As additional information is gathered relating to climate-change-driven
land cover changes, alternatives may diverge which makes repeating the modeling process
very important.
iv
Acknowledgments
First, I would like to thank my supervisor Barry Grand for giving a seabird lab biologist
a chance to work on conservation planning for land birds. Im really happy he gave me the
chance because I?ve really enjoyed my time at Auburn and the project. Id also like to thank
his wife, Kathy, for hosting so many lab parties and especially for introducing me to Bu alo
chicken dip and cream cheese with Pick-a-Pepper sauce.
Thank you to my committee, Geo Hill, Craig Guyer, Jaime Collazo, and Mike Mitchell,
who challenged and encouraged me through this process. Your help was invaluable.
I couldnt have  nished this degree without Patti Staudenmaier, the School of Forestry
and Wildlife Sciences, the Auburn University Graduate School and the Auburn University
of International Students. Thank you for putting up with my absentmindedness.
To all my friends at Auburn, thanks for letting me vent, answering my questions, and
sharing a drink or two. To my labmates, Dalinda and Phil Damm, Carrie Johnson, Bruce
Hitch, Rob (and Lauren) Allgood, Brad Alexander, Catherine Rideout, Tyler Kreps, and
Kevin Kleiner. Special thanks to Amy Silvano, Joie Cannon, Alyson Webber, Christina
Schmidt (and Hector), and Jenn Dietlo .
Id like to acknowledge some of my previous teachers who have helped get me to this
point. My undergrad thesis supervisor, Anne Storey, and my MSc supervisor, Keith Hobson,
who introduced me to research and, because of their enthusiasm, made being a biologist look
like the best job in the world.
Finally, to my family, Bill, Marg, Elaine, (Jeremy, can?t wait to meet you in person!),
John, Katie and Dominic, thanks for talking to me about normal things. And last but not
at all least, thanks Brian. I cant wait until were living in the same time zone and not on
v
opposite sleep schedules. Also, thanks for all the editing. All remaining mistakes are my
own.
vi
Table of Contents
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Incorporating expert knowledge in decision-support models for avian conservation 3
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Case study context . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3 Focal species approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Elicitation of focal species . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4.1 Two approaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4.2 Expert selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4.3 Focal species identi cation meetings . . . . . . . . . . . . . . . . . . . 9
2.4.4 Analysis of the elicited data using the Lambeck method . . . . . . . . 12
2.4.5 Analysis of the elicited data using the SDM method . . . . . . . . . . 13
2.4.6 Results: a list of focal species to support conservation planning in the
SAMBI region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4.7 Validation through online surveys . . . . . . . . . . . . . . . . . . . . 15
2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5.1 Quantitative vs. qualitative data . . . . . . . . . . . . . . . . . . . . 18
2.5.2 Visualizing the data . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.5.3 Online surveys . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.5.4 Implementing the results . . . . . . . . . . . . . . . . . . . . . . . . . 19
vii
3 Focal species as method to plan spatially explicit conservation priorities . . . . . 26
3.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3.1 Study site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3.2 Focal species selection and weights . . . . . . . . . . . . . . . . . . . 29
3.3.3 Spatial data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3.4 Density estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3.5 Modeling priorities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.3.6 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.1 Focal species selection . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.2 Distribution of priority scores . . . . . . . . . . . . . . . . . . . . . . 35
3.4.3 Priority maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.5.1 Proxies for conservation . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.5.2 Comparison with other conservation prioritization methods . . . . . . 41
3.5.3 Complications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4 Conservation priorities in an uncertain future . . . . . . . . . . . . . . . . . . . 67
4.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.3.1 Study site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.3.2 Future land cover . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3.3 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.3.4 Density estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.3.5 Modeling priorities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
viii
4.3.6 Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.4.1 Projected priority surfaces . . . . . . . . . . . . . . . . . . . . . . . . 77
4.4.2 Habitat priorities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
ix
List of Figures
2.1 Study region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.2 Process used to create focal species list . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Details of two methods to select focal species . . . . . . . . . . . . . . . . . . . 22
3.1 First step to create priority surfaces . . . . . . . . . . . . . . . . . . . . . . . . 44
3.2 Second step to create priority surfaces . . . . . . . . . . . . . . . . . . . . . . . 44
3.3 Skew for conservation priority values of vegetation types in SAMBI . . . . . . . 45
3.4 Kurtosis for conservation priority values of vegetation types in SAMBI . . . . . 45
3.5 Alluvial forested wetland conservation priority surfaces . . . . . . . . . . . . . . 46
3.6 Beach conservation priority surfaces . . . . . . . . . . . . . . . . . . . . . . . . 47
3.7 Estuary conservation priority surfaces . . . . . . . . . . . . . . . . . . . . . . . 48
3.8 Grassland conservation priority surfaces . . . . . . . . . . . . . . . . . . . . . . 49
3.9 Longleaf conservation priority surfaces . . . . . . . . . . . . . . . . . . . . . . . 50
3.10 Maritime forest conservation priority surfaces . . . . . . . . . . . . . . . . . . . 51
3.11 Non-alluvial forested wetland conservation priority surfaces . . . . . . . . . . . . 52
3.12 Open pine conservation priority surfaces . . . . . . . . . . . . . . . . . . . . . . 53
x
3.13 Shrub-scrub conservation priority surfaces . . . . . . . . . . . . . . . . . . . . . 54
3.14 Slope forest conservation priority surfaces . . . . . . . . . . . . . . . . . . . . . 55
3.15 Upland forest conservation priority surfaces . . . . . . . . . . . . . . . . . . . . 56
3.16 Wetland conservation priority surfaces . . . . . . . . . . . . . . . . . . . . . . . 57
3.17 Di erences between habitat priority surfaces . . . . . . . . . . . . . . . . . . . . 58
3.18 Di erences between priority surfaces of alluvial forested wetland, beach, estuary,
and grassland . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.19 Di erences between priority surfaces of longleaf, maritime forest, non-alluvial
forested wetland, and open pine . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.20 Di erences between priority surfaces of shrub-scrub, slope forest, upland forest,
and wetland . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.1 Study region . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.2 Conservation priority surfaces for open pine forest in the Southeastern US, 2000
and 2100 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.3 Conservation priority surfaces for maritime forest in the Southeastern US, 2000
and 2100 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.4 Open pine forest  nal conservation priority surface, summed and discounted . . 86
4.5 Maritime forest  nal conservation priority surface, summed and discounted . . . 87
4.6 Di erence between initial conservation priorities and those of 2100 . . . . . . . . 88
4.7 Di erence among priority surfaces for di erent climate scenarios . . . . . . . . . 89
xi
4.8 Distribution of priority values for habitats in 2000, summed across 100 years, and
with a 4% annual discount . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
4.9 Skew and kurtosis by year of land cover projection . . . . . . . . . . . . . . . . 91
xii
List of Tables
2.1 Habitat characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2 Example of a table used to select the focal species using structured decision-making 24
2.3 Example of the tabular summaries generated from our online survey to validate
the selection of focal species . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.1 Landforms associated with general vegetation types . . . . . . . . . . . . . . . . 62
3.2 Territory size and dispersal distance for focal bird species . . . . . . . . . . . . . 63
4.1 Habitat association, selection method, territory size, and dispersal distance for
focal bird species for open pine and maritime forests . . . . . . . . . . . . . . . 92
xiii
Chapter 1
Introduction
Conservation areas have been designed with a range of goals in mind including preserving
biodiversity [114]; species abundance [116]; functional ecosystems [105] [121] or individual
species [99] [107]. At a broad spatial scale, these conservation areas can work together to
form a reserve system of complimentary sites that provides protection for a more varied set of
organisms or ecosystems. The spatial arrangement of conservation areas within a reserve set
has been the subject of much debate, although some basic principles have been agreed upon
[133] [89]. Individual sites should be as large as possible to facilitate species persistence [22]
and as round as possible to reduce the e ect of edges and to maximize the amount of core
habitat [89]. For a reserve system, several sites in close proximity or directly connected are
of higher conservation value than a series of more distant sites because it facilitates dispersal
among sites [37]. Finally, conservation areas should be systematically placed to protect
vulnerable species, or ecosystems or biodiversity rather than being placed where economic
or political forces permit protection [57].
The Southeastern United States has high conservation importance because of the re-
gion?s habitat and species diversity, ecological processes, and evolutionary potential; however,
it also warrants strong concern because of historical habitat loss, future threats, and inad-
equate protection [81]. Less than 12% of the land in the Southeast is under government
protection or easement and resources are limited for management of these lands or for fur-
ther land acquisitions. Conservation planning over this large and varied extent is further
complicated by numerous stakeholders that have a vested interest in conservation e orts.
State and federal governments, non-governmental organizations and private landowners are
1
all involved with conservation of bird species throughout the Southeastern US and coordina-
tion amongst these groups is importance to ensure e orts are not unnecessarily duplicated.
Therefore, it is important to develop tools that enable conservationists to prioritize regions
for their conservation value.
I worked with the Atlantic Coast Joint Venture to develop a decision support tool to
enable stakeholders to plan and coordinate conservation e orts for birds across the Coastal
Plain of the Southeastern US. In order to provide the most useful decision support tool, I
needed to address several signi cant sources of uncertainty. Each of my chapters addresses
one or more of these sources of uncertainty. In addition, providing stakeholders with infor-
mation about how uncertainty was addressed and how a decision tool was developed allows
them to make more informed decisions [36].
To begin, I used two processes to create focal species lists based on expert derived data
(Chapter 1). Conservation planning for this large region is complicated by inconsistent data
availability across regions, habitats, and species. When empirical data about species and
habitats are lacking, experts may be the best source of information [54]. Secondly, using
the suite of focal species, I considered how using focal species to determine conservation
priorities di ered from using vegetation types (Chapter 2). Much discussion has taken place
over how well focal species work to conserve biodiversity [104] [102], but working at a broad
spatial extent, it is not possible to model every species. Finally, the very causes of concern
to species persistence, climate change and urbanization, introduce further complexity to
conservation planning because of their uncertainty. I developed a series of conservation
priority surfaces under three di erent climate scenarios which gives stakeholders a sense of
the possible variation in future conditions (Chapter 3).
2
Chapter 2
Incorporating expert knowledge in decision-support models for avian conservation
2.1 Introduction
Bird abundance in the United States has been declining for more than half a century,
likely as a result of habitat changes [125] [78]. In the Southeastern United States, habitat
and management changes including deforestation, reforestation, urban growth, and  re sup-
pression have reduced the availability of high-quality habitats and have increased habitat
fragmentation [130] [30] [126]. Short-term projections suggest that urbanization will con-
tinue to reduce forest areas and increase their fragmentation [129]. In the long term, climate
change will alter precipitation and temperature patterns, and rising sea levels will reduce
coastal habitat [43]. It is therefore important to conserve what is currently available (species,
habitats, and ecosystems) and plan for future conservation. To e ectively protect or increase
bird populations in this context, conservation must maintain or increase habitat quality and
quantity. However, given limited resources, it?s important to focus e orts where they have
the greatest bene t rather than where land is economically unimportant [95]. Furthermore,
complex systems with multiple species and habitats may require trade-o s among con icting
conservation objectives.
The Southeastern United States has high conservation importance because of the re-
gion?s habitat and species diversity, ecological processes, and evolutionary potential; however,
it also warrants strong concern because of historical habitat loss, future threats, and inade-
quate protection [81]. Based on data from USGS [98], we estimate that only 12% of Southern
Coastal Plain ecoregion is under permanent protection. Conservation planning for this large
and varied ecoregion is complicated by variable data availability across regions, habitats,
3
and species. Therefore, to support conservation planning, experts may be the best source of
information [54].
In conservation planning, experts are often used to evaluate potential threats [18] [122],
select high-priority areas [80] [15] [56], de ne initial values for Bayesian modeling [54] [49],
and propose conservation targets [40] [23] [2]. Experts can provide critical insights when
there are multiple con icting objectives, when empirical data about species and habitats
is lacking, threats are uncertain, and it?s necessary to focus on a few key species. This is
common when developing large-scale, long-term plans. However, when expert knowledge
supports conservation decisions, little information may be provided or recorded about the
experts or how their knowledge was collected and used [49].
To support avian conservation in the Southeastern United States by ensuring adequate
current and future habitat, we elicited expert knowledge of specieshabitat associations, habi-
tat management needs, and threats. This chapter highlights one aspect of our project: the
use of expert knowledge to de ne a suite of focal species for specieshabitat modeling that
would subsequently support the development of a decision-support tool. These species rep-
resent the present and future habitat needs of other species that cannot be modeled given
time and resource constraints. The  nal tool will be a series of spatially explicit landscape
models based on the habitat needs of focal species that indicate where to focus conservation
e orts.
2.2 Case study context
The elicitation exercises reported here formed the foundation for a much broader study
that had three major objectives: to assess the current ability of habitats to sustain avian
populations; to model future conditions based on projected urban growth, conservation
programs, and climate change and predict the response of avian populations; and to enhance
coordination among stakeholders during all planning stages. Stakeholders provide access to
4
information that may be unavailable elsewhere, and help us to address the concerns of those
who will enact conservation actions, thereby leading to better outcomes [97].
The project covered the South Atlantic Migratory Bird Initiative (SAMBI) area [128]
(Fig 2.1). The area extends from the Atlantic Coast in the east to the boundary between
the Coastal Plain and the western Piedmont. Historically, this area was dominated by  re-
maintained longleaf pine (Pinus palustris) savanna [83], but only 2% of this habitat remains
after conversion to agriculture, pine plantations, and urban areas [126]. Frequent  res created
high biodiversity [126], including a high proportion (40%) of endemic plant species [127] and
30 threatened or endangered vertebrates [126]. Other important habitats include bottomland
hardwood forest dominated by  ood-tolerant species such as cypress (Taxodium distichum)
and tupelo (Nyssa aquatica) [41]. Unique non-alluvial forested wetlands include rainfall-
driven pocosins, Carolina bays, and pitcher plant (Sarracenia spp.) bogs [100]. SAMBI?s
coastal area has extensive barrier islands and highly productive estuarine wetlands [21].
2.3 Focal species approach
When ecosystem management targets focal species, the goal is to protect many other
species [57]. In contrast to conservation based on ecosystems or ecosystem functions, focal
species indicate the quantity and arrangement of conservation areas and allow planning at
a  ner scale [102]. Among focal species, sub-categories include indicator, keystone,  ag-
ship, umbrella, and landscape species [13]. Indicator species re ect ecosystem health or
biodiversity [51]. We did not explicitly select biodiversity indicators because the low spa-
tial resolution of remote-sensing data leads to the apparent co-occurrence of many species
[26]. In comparison, keystone species are more in uential than their abundance suggests
[92]; in the SAMBI area, they include gopher tortoises (Gopherus polyphemus), which ex-
cavate burrows used by many other species [31]. Flagship species are species that attract
public support and may promote conservation of associated species, even though this may
not be an explicit conservation goal in  agship species management [111]. Umbrella species
5
require large habitat patches, so their conservation explicitly protects many other species
in those large areas [79]. Landscape species resemble umbrella species in requiring large
areas, but also require a speci c habitat composition [106]. Because the SAMBI area has
so many di erent conservation goals (e.g., restoring rare species, increasing populations of
hunted species, preserving common species), we did not want to restrict experts to any one
type of focal species. Using multiple focal species reduces the risk of missing endemic or
range-restricted species when planning reserves [50] [40] and explicitly includes species with
substantial public interest or conservation resources.
Using focal species to guide conservation e orts has been criticized. Andelman and
Fagan [3] showed that selecting focal species using a range of criteria did not improve pro-
tection of the greatest number of species at a minimum number of sites than randomly
chosen species. The e ectiveness of focal species also varies with the taxa that are selected
[102]. For instance, basing conservation areas on birds did not protect butter ies [28], nor
did protecting large mammals protect smaller mammals [12] However, focal species can be
e ective in more limited situations; for example, protecting focal butter ies protected other
butter ies and protecting focal birds protected other birds [28]. Since our objective was to
use avian focal species to represent other birds, rather than overall biodiversity, the focal
species approach was appropriate for our purposes.
Although focal species are commonly used for conservation planning, selecting them
based on expert knowledge is less common. In the Bolivian Andes and the Republic of
the Congo, Coppolillo et al. [18] selected landscape species for conservation planning. They
selected four to six large vertebrate species at each site to represent the habitat requirements,
threat sensitivity, and ecological function of other species, and their importance to humans.
At both locations, experts identi ed potential focal species, scored each species using the
above mentioned criteria, and selected the  nal suite of focal species. These experts were
 eld biologists, managers, and people who knew the species or area; they scored species
using (in order) published and unpublished literature and their own knowledge. Although
6
Coppolillo et al. used experts to select these species, they reported insu cient detail to guide
other researchers interested in using experts to support conservation e orts. For example,
they didn?t discuss the extent to which the experts resorted to non-literature information
sources nor did they detail how they elicited information from the experts.
2.4 Elicitation of focal species
For our purposes, we wanted a species suite that would represent all habitat types
de ned by the SAMBI Plan [128], including species with large area requirements and species
requiring management. Our initial list of potential focal species comprised 65 key species
identi ed in the SAMBI Plan. We subsequently used the two processes described in Section
4.1 to develop lists based on expert knowledge using two selection methods. Finally, we
validated the two subsets of the overall list against the original list of 65 species.
2.4.1 Two approaches
To select focal species, we used Lambeck?s selection process [50] and a method rooted in
structured decision-making (SDM). The former method has been used to select focal species
(e.g., [102]); the latter was a modi cation of Gregory and Keeny?s [29] decision-making
process. We designed our elicitation process to work with the SDM method, but added the
Lambeck method because it re nes the species selection by focusing on landscape design and
management rather than expert elicitation.
Lambeck modi ed the umbrella species concept by systematically selecting species based
on their threat category [50], with an emphasis on protecting the most sensitive species [102].
For example, connectivity should support species with restricted dispersal ability, and patch
size should sustain species with large area requirements. This method used empirical data
from published literature and  eld research rather than expert opinion. Rather than eliciting
quantitative data from experts during the Lambeck analysis, we modi ed the method to
accept qualitative expert data. We believed this would help experts reach a consensus more
7
quickly and maximize participation by experts who lacked con dence in their ability to
provide precise data.
Gregory and Keeney [29] have broad experience in decision analysis and have used their
SDM methods to de ne and solve resource management issues. SDM, unlike the Lambeck
method, helps stakeholders to make decisions, and we modi ed the process to use expert
opinion. SDM comprises a  ve-step procedure for solving problems: state the problem,
establish objectives that can be evaluated, design alternative solutions, evaluate each alter-
native?s consequences, and assess tradeo s before reaching a decision. Although SDM can
help individuals to reach a decision, it is especially useful for groups.
In our modi ed SDM process, we used habitat characteristics as objectives and potential
focal species as alternatives based on their association with each habitat characteristic. Ex-
perts prepared an alternatives table that rated each species according to the strength of its
linkage with each characteristic. As experts characterized habitat needs, similarities emerged
among species. High similarity between the habitat requirements of two species justi ed re-
moval of the species of lower conservation or management concern from the species list. The
species list was reduced using criteria that will be used to manage the conservation system
[132]; in the SAMBI project, this will be through habitat acquisition (coarse-scale) and en-
hancement (medium-scale), so we emphasized similarities among coarse- and medium-scale
habitat characteristics. The level of spatial detail is an important aspect of the present exer-
cise, since habitat planning and management will be based on remote-sensing data (satellite
photos used to provide land-use and vegetation type data) stored in the geographical infor-
mation system software that will be used in a subsequent stage of this project to develop
landscape models.
2.4.2 Expert selection
The Atlantic Coast Joint Venture, a partnership of governmental and non-governmental
organizations that strives to provide healthy ecosystems to support healthy avian populations
8
across jurisdictions, organized the SAMBI project. Working through the Joint Venture gave
us access to many experts. We limited participation to experts associated with SAMBI but
did not limit their number. We wanted the largest group possible because no individual
understands all potential focal species [122] and broad participation reduces the bias caused
by extreme views [54]. We invited all SAMBI members, including biologists and managers,
from the Joint Venture team. Of 278 invitees, 53 attended elicitation meetings. During
follow-up surveys, we again invited all SAMBI members; of those who attended the elicitation
meetings, 16 participated in a conference call and 15 completed at least part of the surveys.
Experts included representatives from state and federal government agencies in Virginia,
North Carolina, South Carolina, Georgia, and Florida; non-governmental organizations in-
cluded The Nature Conservancy, Ducks Unlimited, Audubon Society state chapters, the Tall
Timbers Research Station, the North Carolina Museum of Natural Sciences, the University
of Florida, and the University of Georgia.
Our initial list included 65 potential focal species identi ed in the SAMBI Conservation
Plan [128]. We wanted experts to consider the species associated with particular habitat
characteristics (Table 2.1). Large scale specieshabitat associations were found in the liter-
ature [34], but medium-scale details of habitat preferences were di cult to determine. We
felt that experts who study or work with a species would know this information, even if
they did not publish it. Our preliminary work with the experts suggested that certain habi-
tat characteristics extended across habitats and could be considered apart from the larger
habitat types. For instance, bare ground is found in both grasslands and wetlands, and
closed canopies are found in both deciduous and mixed forests. We presented the species list
alphabetically to avoid biasing expert responses.
2.4.3 Focal species identi cation meetings
From August to November 2008 we held 2-day meetings in each state. The  rst day
introduced the SDM process and summarized the project; during the afternoon, we began
9
species selection. During the selection process, we divided experts into four groups based
on their stated area of knowledge or comfort: waterfowl (e.g., ducks, geese, and swans),
land birds, waterbirds (e.g., herons, rails, gulls, and terns), and shorebirds (e.g., sandpipers
and plovers). We generally had more waterfowl experts than other types, but we also had
several land bird experts. We usually combined shorebird and waterbird experts because so
few were present. At the Georgia meeting, only one individual had shorebird and waterbird
expertise, but North Carolina and Florida had numerous experts in this category.
On the second day, we reviewed the previous day?s work and discussed landscape de-
sign issues. To encourage discussion, we started with simple examples. For example, we
picked a bird with well-known, well-de ned habitat preferences and asked experts to review
that example. We knew some experts personally and could direct questions to an appro-
priate expert. Facilitators answered questions and clari ed characteristics during meetings
to reduce bias due to imprecise language [49]. To elicit information, we asked experts to
identify important habitat characteristics for the SAMBI priority species [128]. At the  rst
meeting, we did not initially present the specieshabitat association tables because creating
an alternatives table without preconceptions is a key step in the SDM process [29]. How-
ever, this made the process unworkably slow because experts wanted to assign the species to
habitats rather than to habitat characteristics, so we subsequently presented our prepared
tables, and were much more successful at focusing experts on the process. For subsequent
meetings, we started with matrices of potential focal species and habitat characteristics. We
asked each expert to identify and score the habitat characteristics required for each potential
focal species in their group (Table 2.2). Their scoring choices ranged from 1 (bene cial or
preferred) to 5 (detrimental or avoided). Experts could also report insu cient information
or that the relationship was neutral by not scoring the species. We also let experts answer in
more detail, for example, to describe a relationship where the species preferred a moderate
level of a habitat characteristic but avoided either extreme. For each species, we also asked
10
the experts to note whether the species were umbrella,  agship, biodiversity indicator, key-
stone, and habitat or dietary specialist or generalist. We did not provide access to published
data (e.g.,  eld guides, species accounts, Internet searches), so they answered based on their
own knowledge or experience.
Experts were comfortable with the scoring system except when we did not clarify the
direction of the scoring. For example, the "depth of water" characteristic was confusing
because we did not specify whether this meant shallow or deep water. When opinions di ered
about species preferences for deep versus shallow water, the results were ambiguous. When
experts revealed this problem, we asked them to add a brief description after their score
to indicate how they interpreted the scale so we understood their intent when we compiled
our data. Subsequently, we provided de nitions so that all experts used the same scoring
criteria.
During our elicitation meetings, experts were given equal weight and group members
worked to achieve consensus. Because we held meetings in each state, experts tended to know
each other. This made it possible that professional relationships in uenced their answers
[29], such as when someone deferred to a superior in their organization, so the answers may
have been biased towards the opinions of the most senior experts. We did not address this
source of bias because we assumed that the most senior experts had the most experience
and knowledge and that this therefore provided an acceptable, if unmeasured, weighting.
After experts completed the exercise, we compiled their answers and presented them to
the whole group the next day. We did not prevent them from commenting on the results
from other groups. During this stage of the process, we did not need a high level of individual
participation because the smaller groups had already reached a consensus before sharing their
results with the full group. A few experts sometimes monopolized the discussion, which may
have introduced bias [49]. When this happened, we asked the original group to con rm
whether their results should be modi ed. When group members had di erent opinions, we
recorded all answers rather than forcing an arti cial consensus. On the second day, we also
11
asked experts to de ne key characteristics that ensured the functionality of each habitat, such
as  re in an open pine ecosystem. The habitat characteristics were similar to those used for
focal species, and we framed our problem by asking whether each habitat characteristic was
important to the habitat?s functioning. Experts then scored the characteristics as limiting
(i.e., restricted the habitat?s value) or compensatory (i.e., important, but allowed tradeo s
among characteristics). This was done as a group, and experts discussed each habitat until
they reached a consensus.
The initial set of meetings provided a framework for selecting potential focal species and
modeling the habitat conuration. For both selection methods (Lambeck and SDM), we used
the same data tables, but we used di erent processes to create the focal species lists (Fig
2.2). For each species and each corresponding habitat characteristic, we created an overall
score by combining the scores from all states. We used the majority score unless there was
a disagreement (a characteristic was said to be both avoided and preferred by a species), in
which case we kept the range of scores.
2.4.4 Analysis of the elicited data using the Lambeck method
We used Lambeck?s process [50] to create a list of the potential focal species (Fig 2.2).
To begin the Lambeck process, we reduced the length of the expert-elicited list by excluding
species that had secure populations, abundant game species, and species identi ed as being
of moderate concern (thus, low priority) in the SAMBI Plan. If there was any uncertainty, we
retained a species. In the next step, we subdivided the remaining species based on di erences
in pattern and process; that is, we distinguished species that required habitat reconstruction
from those that could live in existing habitat with appropriate management. Reconstruction-
limited species required changes to the landscape pattern, such as creating additional habitat
patches, improving connectivity between patches, or creating larger habitat patches[50].
Management-limited species are sensitive to the rate or intensity of landscape processes,
such as  re frequency or grazing intensity [50]. We did not make these categories mutually
12
exclusive because some species may currently lack adequate habitat and their habitat may
require management once it has been established.
Among the reconstruction-limited species, we de ned three subcategories based on the
expert scores: Area-limited species required a large patch size, and resource-limited species
had preferred or required habitat characteristics that could not be detected by remote sens-
ing. For example, several duck species rely on submerged aquatic vegetation and other species
require dead standing trees for nesting, but currently these habitat characteristic cannot be
mapped using remote sensing. The third subdivision was dispersal-limited species. However,
experts concurred that this was not a limiting factor for the priority species in the SAMBI
area.
We de ned management-limited species as any species that scored "bene cial" or "pre-
ferred" for any of the disturbance categories, as well as any species that required human-based
management, including several duck species that relied on managed wetlands for their winter
habitat.
In the  nal step, Lambeck suggests that for each habitat, one should select the most
limited species for each pattern and process, and design the landscape based on the needs
of those species. For example, the species with the largest area requirement would de ne
the minimum patch size, the species with the shortest dispersal distance would de ne the
maximum distance between patches, and the species most sensitive to disturbance would
de ne the management protocol. We chose not to take this step because we wanted to retain
the largest possible list for the experts to evaluate.
2.4.5 Analysis of the elicited data using the SDM method
To select the focal species using SDM, we used  ve habitat characteristics scored by the
experts that could be estimated from landscape-level data: proximity to coast, water type,
water depth, forest type, and canopy (Table 2.2; Fig 2.3). For example, some species prefer
coastal areas, some avoid coastal areas, and some have no preference. Among those that
13
preferred coastal areas, we further subdivided species according to their preferred habitat
types such as intertidal beach, coastal marshes, and shallow areas. We subdivided birds that
avoided coastal areas, but which still required open water or wetland habitats, into birds
that preferred shallow water and those that had no preference regarding water depth. Birds
exhibiting no preference in their use of coastal and non-coastal areas were subdivided into
those that used riparian areas, avoided riparian areas, used emergent marshes, or preferred
shallow water. Forest-associated species were divided by the type of forest they preferred
and then into those that preferred closed and open canopy. Finally, some birds preferred
open habitats.
After grouping birds based on these associations, we selected one species as the represen-
tative focal species. We generally picked species with the most complete habitat associations.
For example, we selected the American Oystercatcher as a focal species associated with shal-
low water along beaches. Species with similar requirements included the Piping Plover, the
Red Knot, the Whimbrel, the Least Tern, and the Black Skimmer.
2.4.6 Results: a list of focal species to support conservation planning in the
SAMBI region
The focal species selected using the SDM and Lambeck methods included 35 of the initial
65 species, with 11 species common to both lists (Fig. 2.2). The SDM method selected 10
species that were not chosen using the Lambeck method, and the Lambeck method selected
14 species that were not chosen using SDM. The column labeled "Online survey" represents
information that we used to validate our results.
To create a list of species that would be validated and used to develop the decision-
support tool, we used species common to both lists, all selected land birds that appeared
in both lists, and waterbirds, waterfowl, and shorebirds that appeared in the SDM list. We
excluded the Lambeck list from the latter group because the experts agreed that waterfowl,
waterbird, and shorebird habitat tended to overlap at the level of the data we used, and the
14
SDM method let us assess where habitat overlaps were likely to occur; it was therefore a
better list for our purposes. We retained all land birds because we had no reason to prefer
either selection method.
2.4.7 Validation through online surveys
To validate the list of focal species, we created a follow-up survey using the online survey
software SurveyMonkey [118]. The online survey included supporting documentation and
was introduced to respondents during a conference call. We asked the experts to review
and rank the selected focal species and to add or remove species as necessary. We provided
criteria for evaluating whether a species was a suitable focal species. The focal species could
meet more than one of the following criteria: representative of other species, well-known bi-
ology, easily sampled or observed, sensitive to disturbance, umbrella species,  agship species,
habitat specialist, dietary specialist, or keystone species [13] Our questionnaire listed species
associated with each habitat type in the SAMBI Plan [128], with focal species highlighted,
although we did not state the selection method used to select them. The participants scored
the suitability of each species as a focal species using ranks ranging from 1 (very poorly) to
5 (very well); they could also respond that they had insu cient personal knowledge to rank
the species.
The scoring process let us create a "focal species value" and a measure of uncertainty
that we used to assign species weights in the landscape model (Table 2.3). The mean score
provided a measure of the relative value of each focal species and the variation in scores
provided us with a measure of uncertainty. For example, if all participants assigned a score
of 4 to a species, we were con dent that the species was a good focal species. In contrast, we
had less con dence if participants assigned an equal number of 3s, 4s, and 5s. For example,
experts di ered in their opinions of the Black-Throated Green Warbler as a focal species for
alluvial forested wetlands: 4 of 12 experts thought it was a good or very good focal species.
In contrast, 12 of 13 experts scored the Prothonotary Warbler, which was not included in our
15
focal species list, as a good or very good focal species; the other expert declared insu cient
personal knowledge. We did not remove any species from the focal species list based on the
online validation, but we did add 11 species (Table 2.2) to our landscape model based on
the expert scores.
2.5 Discussion
Neither selection method produced a list that we considered entirely suitable for con-
servation planning. Each method selected at least one species per habitat included in the
SAMBI Plan, but the online validation survey included several species that were not included
by either method and several that were not suitable focal species. For example, experts gave
Bachman?s Sparrow, the Cerulean Warbler, the Redhead, the Canvasback, and the Sandhill
Crane an average focal species value less than 2 ("poor"). However, the Redhead and the
Canvasback were added to the initial focal species list because they were resource-limited
species according to Lambeck?s de nition [50]. When re-evaluating the species, experts may
have reduced the value of these species because we did not indicate that resource limita-
tion was a criterion. Species values may also have decreased if they were uncommon in the
study region, such as the Cerulean Warbler [35] and the Sandhill Crane [120], or if they only
overwintered in the region, such as the Redhead [135] and the Canvasback [74]. Bachman?s
Sparrow had a low value in only one habitat (early successional and shrub-scrub) of the three
in which it occurs; it had a high value in the other two habitats (longleaf pine - slash pine
(Pinus elliotti)  atwoods and mature open pine). We began our online surveys by informing
the experts that our list required revision, and engaging the experts in this way let them
criticize more freely.
In our list of potential focal species, we only used species in the SAMBI plan [128]
that were associated with particular habitats, although experts could add species during the
meetings. This gave us 65 species, out of a total of 172 species rated as being of highest,
high, and moderate concern (see Table 1 in Watson and Malloy [128]). It was important
16
that the habitats of our focal species represent the full suite of habitats used by all species
identi ed in the SAMBI Plan, and we believe we accomplished this because the focal species
we chose cover all habitats in the SAMBI Plan.
There may be concerns about the repeatability of our selection process because we asked
experts to score birdhabitat associations without referring to published materials. We made
this choice rather than using references to complete the tables ourselves because elicitation
of knowledge not found in the published literature was a key goal of the process [86] [97]. A
di erent set of experts may provide di erent knowledge, thereby limiting the repeatability
of the results. However, using a large group of experts and limiting answers to a discrete
qualitative scale improved the reliability of the process. Using a simple scoring process
likely also improved the ability to reach consensus. For example, asking experts to quantify
canopy heterogeneity would produce a wide array of values, but similar focal species would
be selected as long as there was general agreement on the direction and strength of the
relationship between habitat quality and factors such as canopy heterogeneity. Insisting on
consensus can eliminate potentially important di erences of opinion among experts, but it
was appropriate for our project. Grouping the experts (e.g., land vs. water birds) probably
increased the repeatability of our results by eliminating outlier answers that would arise when
experts speculated about specieshabitat combinations they were not truly familiar with.
Although neither the Lambeck method nor the SDM method was ideal for selecting a
suite of focal species, combining expert opinion with these processes had bene ts for selecting
focal species. Both methods provided an initial list of species we could subsequently ask the
experts to validate. Many expert knowledge studies have not included detailed information
about their process (e.g., [18]). We hope that our experience will help others who are
considering a focal species approach based on expert elicitation. To improve such a process,
we suggest the following:
17
2.5.1 Quantitative vs. qualitative data
Qualitative data is easier to explain to experts and does not require extensive analytical
knowledge [54]. Requesting qualitative rather than quantitative data probably increased our
response rate because more experts would have felt su ciently con dent to participate, and
this also decreased the time it took us to collect and review the data. Eliciting quantitative
data would have provided more detailed data, but our project did not require such detailed
information. However, care should be taken to ensure that questions are well de ned. Pre-
validation of the survey in a practice session with quali ed people who will not be part of
the  nal expert group is recommended.
2.5.2 Visualizing the data
Flow diagrams [50] and in uence diagrams [61] are commonly used to visualize data.
However, it would have been di cult and time-consuming to identify focal species by devel-
oping such tools during the meetings. Asking experts to complete tables of species - habitat
associations provided information about a large number of species (n = 65) and habitats in
a short period of time (2 days). Without this approach, gathering the expert data would
have taken much longer. We don?t believe that in uence diagrams would have been useful,
since they are typically used to characterize beliefs based on the relationships among system
states and objectives, and we lacked su cient information to characterize all those relation-
ships. By focusing experts on entering data in tables, we reduced variability and increased
consensus. Although we wanted consensus answers, that may not be appropriate for projects
with di erent goals.
Although we could have used in uence diagrams or  owcharts developed prior to the
meetings, we wanted the experts to guide the process rather than reacting to tools that we
presented. Using unfamiliar visualization tools would have required the experts to under-
stand our process for diagramming the important relationships.
18
2.5.3 Online surveys
When time or money is limited, online surveys can rapidly and inexpensively collect data
from experts. However, if reaching a consensus among experts is an objective, as it was for
us, this would be di cult to accomplish using an online survey. The individual, anonymous
nature of online surveys facilitates gathering of independent ideas and avoids groupthink,
which results from inappropriate group cohesion [48], but eliminates the dialogue required
to seek consensus. Online surveys facilitate quantifying values and their uncertainty even
with qualitative scoring systems, but require relatively large numbers of participants.
We found the online survey program SurveyMonkey economical, easy to use and suf-
 ciently  exible to structure our questions e ectively, but it seemed designed for simpler
surveys and smaller groups of respondents. If online surveys will be used to gather data
from experts, their design should be modi ed so they will be more suitable for this type of
research.
2.5.4 Implementing the results
Using experts in our planning process  lled data gaps in the published literature, ensured
that we had appropriately de ned the problems and objectives (e.g., population goals versus
speci c management actions), and will increase user con dence in our  nal products [20]
[137]. The list of focal species that we developed will be used to prioritize areas for bird
conservation in the SAMBI area. However, the Southeastern United States is home to many
other at-risk species, including amphibians, reptiles, and mammals [126], that were not
included in our planning process. The selection process described in this chapter can be
extended to include these species, and expert opinion may be even more valuable because
so little published information is available about some of these species.
19
Figure 2.1: The study region in the Southeastern United States included coastal plain regions
of Virginia, North Carolina, South Carolina, Georgia, and Florida.
20
Figure 2.2: The process we used to create the list of focal species based on an elicitation
of expert knowledge using two selection methods. Initially, we met the experts in person
to learn their opinion of how species use di erent habitat characteristics. We used this
information in the two selection methods to create lists of potential focal species. We then
combined the two lists into a single list, which we asked the experts to review in an online
meeting. We will use the results of this review to create a spatially explicit model to support
bird conservation e orts.
21
Figure 2.3: We used two di erent processes to select focal species using the results of our
elicitation of expert opinion. (a) A process based on that of Lambeck [50]. (b) A process
based on structured decision-making (SDM). Species common names are those designated
in American Ornithologists? Union[1]
22
Table
2.1:
Characteristics
of
the
habitats
used
to
inform
the
selection
of
focal
sp
ecies
and
to
de ne
key
functional
characteristics
Habitat
class
Characteristic
Commen
ts
Hydrological
coastal
use
areas
adjacen
tto
coast,
not
necessarily
marine
habitat
water
typ
e
water
depth
salinit
y
presence
of
subme
rged
aquatic
vegetation
aquatic
macroin
vertebra
tes
turbidit
y
 o
oding
includes
both
seasonal
and
tidal
 o
oding
high-energy
shore
low-energy
shore
Disturbance
an
y
high
 re
frequ
ency
ev
ery
3to
5y
ears
gro
wing
season
 res
Vegetation
canop
yco
ver
mid-storyunderstorylow
basal
area
less
than
50
sq
ft
per
acre
old
or
mature
trees
required
for
nesting
or
foraging
mature
forest
bare
ground
Other
patc
hsize
so
cial
aggregation
individuals
or
pairs
asso
ciate
with
others
with
or
without
ov
erlap
large
forest
patc
h
requires
alarge
patc
hof
con
tiguous
forest
elev
ation
urban
av
oidance
edges
bet
ween
habitat
typ
es
or
bet
ween
land
and
water
large
home
ran
ge
inv
asiv
esp
ecies
23
Table
2.2:
Exa
mple
of
atable
used
to
select
the
focal
sp
ecies
using
struct
ured
decision-mak
ing
in
whic
hexp
erts
rank
ed
sp
ecie
spreference
for
or
av
oidance
of
sp
eci c
habitat
characteristics.
Thirte
en
additional
habitat
typ
es
and
42
asso
ciated
sp
ecies
were
similarly
assessed.
Scores
ranged
from
1(prefer)
to
5(a
void).
If
exp
erts
disagreed,
the
range
of
answ
ers
wa
srep
orted.
Sp
ecies
were
selected
as
focal
sp
ecies
selected
on
Lam
bec
k?s
metho
d[5
0]
(L),
using
SDM
(SDM),
using
both
metho
ds
(b
oth),
or
using
an
online
surv
ey
(on
line).
Other
sp
ecies
were
not
selected
as
focal
sp
ecies
(NS).
Common
name
sconform
to
those
in
the
American
Ornitholog
ists?
Union
standard
[1]
Habitat
Sp
ecies
Metho
d
Hydrology
Structure
Coastal
W
ater
typ
e
Depth
Forest
typ
e
Canop
y
Beac
ha
American
Oystercatc
her
both
1
1-o
cean
5
Piping
Plo
ve
L
1
1-o
cean
5
Wilson?s
Plo
ver
L
1
1-o
cean
5
Whim
brel
L
1
1-o
cean
5
Least
Tern
NS
2
1-o
cean
4
Blac
kSkimmer
NS
2
1-o
cean
4
Ro
yal
Tern
NS
1
1-o
cean
Beac
hb
Red
Knot
both
Deep-w
ater
coastal
c
Redhead
both
Blac
kScoter
L
1
1-o
cean
1
Can
vasbac
k
L
1-4
1-2
op
en
Lesser
Scaup
L
1
1-op
en
1-medium
Inland
d
Sandhill
Crane
SDM
4
American
W
oo
dco
ck
NS
4
Riparian
d
Louisiana
W
aterthru
sh
SDM
1-riparian
5
deciduous
2
Acadian
Flycatc
her
online
1-riparian
deciduous
1
Cerulean
W
arble
r
L
1-riparian
deciduou
s
1
Non-riparian
d
Blac
k-throated
Green
W
arble
r
SDM
2-not
riv
ers
mixed
1
Prothonotary
W
arbler
online
2
deciduous
abreeding
and
non-breeding
habitat
bmigration
habitat
cnon-breeding
habitat
dbreeding
habitat
24
Table
2.3:
Example
of
the
tabular
summaries
generated
from
our
online
surv
ey
to
validate
the
selection
of
focal
sp
ecies.
Exp
erts
were
ask
ed
"Ho
w
well
did
the
sp
ecies
work
as
afo
cal
sp
ecies
within
alluvial
forested
wetlan
ds?"
Sp
ecies
groups
for
11
additional
habitat
typ
es
were
assessed
and
similarly
summarized.
Num
bers
represen
tthe
num
ber
of
exp
erts
who
chose
a
giv
en
score.
Totals
represen
tthe
sum
of
eac
h
score
multiplied
by
the
num
ber
of
resp
onden
ts
who
chose
that
score.
The
weigh
ted
score
equals
the
total
score
divided
by
the
num
ber
of
exp
erts
who
felt
they
had
su cien
tkno
wledge
to
assig
n
ascore.
Common
names
conform
to
those
in
th
eAmerican
Ornithologists?
Union
standa
rd
[1]
Sp
ecies
Verypo
orly
Po
orly
Neutral
W
ell
Very
well
Insu cien
t
kno
wledge
Total
score
W
eigh
ted
score
1
2
3
4
5
Prothonotary
W
arbler
0
0
0
1
10
1
53.64
3.3
Sw
ainson?s
W
arbler
0
1
1
2
7
1
47.76
3.0
Yello
w-throated
W
arbler
0
1
0
5
5
1
47.28
2.9
W
oo
dDuc
k
0
1
2
1
7
1
46.68
2.9
Sw
allo
w-tailed
Kite
0
1
2
1
5
3
37.08
2.6
Cerulean
W
arble
r
2
1
3
4
1
1
33.60
2.1
Blac
k-throated
Green
W
arbl
er
2
1
4
3
1
1
32.64
2.0
Mallard
2
1
3
3
0
3
24.96
1.8
25
Chapter 3
Focal species as method to plan spatially explicit conservation priorities
3.1 Abstract
Here we present a novel method for prioritizing large areas for conservation which in-
tegrates spatial reserve design principles including prioritizing vegetation patches that are
large, round and close to other patches. We compare the results of this prioritization process
using three di erent conservation proxies because over large areas, it is impractical to model
all species. For proxy conservation targets for avian conservation, we used vegetation type
based on publically available data; focal species, which required additional habitat mod-
els; and expert-derived focal species values (fsv) were used to weight species based on how
well species functioned as focal species. Three binary grids were used to develop priority
surfaces based on vegetation type suitability, conservation lands, and urban. The other
two prioritization methods used focal species to identify priority areas by using additional
species-speci c datasets potential habitat and putative source populations. We used the
density of each binary grid, calculated by a two-dimensional kernel density estimator, to cal-
culate conservation priority for each location in a regular 200 m grid across the entire SAMBI
area. Density grids were combined for all associated focal species, suitability, conservation
land, and urban data where appropriate. Our method produced rounded conservation areas
because kernel density was calculated over an elliptical window. This was especially evident
for long linear sites like those of maritime forest which showed increased conservation priority
in areas that made the sites less linear and more round. Kernel density estimation also al-
lowed increases in priority value of sites within the dispersal distance of birds from potential
source populations. Using only vegetation type density to create conservation priority maps
resulted in more high conservation priority areas compared to focal species prioritization
26
except for the most restricted vegetation types, such as those that were maritime-associated.
Conservation priority surfaces created using focal species and fsv were very similar. The
use of fsv should be determined by resources available to researchers because gathering the
data can be done using an online survey. Using vegetation type alone to create priority
surfaces required fewer data and the data are more readily available (all sourced from publi-
cally available datasets), but it did not re ect species habitat use making it problematic for
conservation e orts targeted at species.
3.2 Introduction
To be e ective at protecting or increasing populations or individuals, conservation e orts
must maintain or increase habitat quality and quantity. However, given limited conserva-
tion resources, it is important to ensure conservation e orts are focused where they have
the greatest bene t rather than concentrated where land is unimportant economically [95].
Two major decisions that must be made when conservation planning are the targets of the
planning and what proxies should be used. The two broad categories of targets are species
or taxa, and conservation features such as vegetation type. If the target is a single species,
conservation planning can be done directly from data from the species, but when problems
get more complex and involve numerous targets, it becomes more imperative to select appro-
priate proxies for the planning. Here we consider conservation planning over a large area and
targeting avian species of concern using three di erent proxies: vegetation types associated
with avian habitat, a suite of focal species, and focal species weighted by how well they
function as focal species.
The use of focal species in conservation planning continues to be controversial. Focal
species are selected for their sensitivity to conservation actions and when management actions
are focused on these species, it is intended to provide protection for many other species in the
ecosystem [57]. In contrast to conservation planning based on vegetation types or landforms,
focal species can provide information on quantity and arrangement of conservation areas as
27
well as allowing planning on a  ner scale [102]. However, using focal species in conservation
e orts has been criticized. Based on computer simulation, Andelman and Fagan [3] found
focal species selected according to a number of schemes (e.g. selecting all big carnivores),
did not perform any better than randomly chosen suites of species. The e ectiveness of
focal species has also been found to vary when the taxa of focal species di ered from that
of target species for conservation action [102]. For instance, selecting conservation areas
based on bird species did not protect butter y species [28], nor did protecting habitat for
large mammals protect smaller mammals [12]. However, using focal species in more limited
situations, such as within a realm (terrestrial, marine, or freshwater) or using vertebrate
species as conservation surrogates for nonvertebrates, does appear somewhat e ective [104].
Using features like habitat or vegetation types to design conservation priorities is an
alternative to using focal species. In this method, areas where vegetation types suitable for
avian conservation are common, high quality, and close together are prioritized over less
appropriate sites [70]. Using features like vegetation types or abiotic data like geology or
climate to conserve biodiversity may not be as e ective as using focal species [104]. However,
data about these features are easily available for large areas whereas data for individual
species may be lacking or may vary with location. Our objective was to compare focal
species-based and vegetation type-based conservation prioritization using a novel method
based on the density estimation of these proxies.
Conservation prioritization gives conservation practitioners a way of distinguishing among
areas based on their conservation potential or value. Here we present a novel method for
prioritizing large areas for conservation that integrates spatial reserve design principles in-
cluding prioritizing vegetation patches that are large, round and close to other patches ??.
Current prioritization methods are limited with respect to their spatial explicitness and re-
quire extensive time and computational resources. Our method should be straightforward
to calculate and yet provide detailed information about where to focus resources for conser-
vation e orts including land acquisition, vegetation restoration and management.
28
3.3 Methods
3.3.1 Study site
Our study area was the Coastal Plain of the southeastern United States including the
eastern portions of North Carolina, South Carolina, Georgia as well as southern Virginia
and the panhandle of Florida (Fig.4.1). Based on the South Atlantic Migratory Bird Ini-
tiative (SAMBI) Plan [128], the habitats were grouped into general vegetation type cate-
gories: grasslands and associated habitats (grassland); managed and palustrine emergent
wetlands (freshwater wetlands); early successional and shrub-scrub (scrub); alluvial forested
wetlands; non-alluvial forested wetlands (including pocosins, and Carolina Bays); maritime
forest and shrub-scrub; estuarine emergent wetlands (estuary); beaches and dunes (beach);
longleaf/slash pine  atwoods and savannas (longleaf); mature open pine; hardwood/pine
mixed forest (upland forest); and riparian/mixed mesic forest (slope forest).
3.3.2 Focal species selection and weights
We worked with experts to select a suite of focal species to represent the larger assem-
blage of avian species in the SAMBI area [72]. These experts consisted of representatives
from state and federal government agencies and non-governmental conservation organiza-
tions. The list of potential focal species was based on the species of concern associated
with habitats in SAMBI [128]. Experts were asked to identify and score vegetation char-
acteristics required for each potential focal species. We created a focal species list based
on expert responses including species of management concern, species that were sensitive to
the landscape arrangement, and within a habitat, species with non-overlapping vegetation
requirements [72].
To assess whether species should be added or removed from the focal species list and to
determine focal species values, we ranked species using an online tool [118]. The tool included
supporting documentation and was introduced to respondents during a conference call. We
29
asked experts to review and rank all species of concern including those we selected as focal
species. We provided criteria for evaluating whether a species was a suitable focal species.
The focal species could meet one or more of the following criteria: representative of other
species, well-known biology, easily sampled or observed, sensitive to disturbance, umbrella
species,  agship species, habitat specialist, dietary specialist; or keystone species [13]. Our
questionnaire listed species associated with each habitat category and we highlighted focal
species that were included on our focal species list. The participants scored the suitability of
each species as a focal species using values from 1 (would perform very poorly) to 5 (would
perform very well); they could also respond that they had insu cient personal knowledge to
evaluate the species. For each species, an overall focal species value (fsv) was calculated by
averaging expert scores.
3.3.3 Spatial data
There were up to  ve potential binary grids (matrices) developed for identifying priority
areas for conserving each vegetation type but these di ered with conservation target, which
were vegetation type, focal species, and focal species values. We used ArcGIS Version 9.3
(The Math Works, Natick, Massachusetts) and MATLAB Version 7.10.0.499 (Environmen-
tal Systems Research Institute, Redlands, California) to create and manipulate spatial data.
The SAMBI area was divided into a regular 200 m grid (Fig. 3.1) and each grid cell was
scored as a 0 (absent) or 1 (present) depending on whether speci c resources or landscape
characteristics were present. Three binary grids were used to develop priority surfaces based
just on vegetation type (1) suitability, (2) conservation lands, and (3) urban. Both priority
calculation methods using focal species used  ve binary grids: (1) suitability, (2) conserva-
tion land and (3) urban, as described above; and two species-speci c grids created for each
focal species that bred in the SAMBI region, (4) potential habitat and (5) putative source
populations. The grid for potential habitat was based on range maps, land cover, ancillary
data (landform, stream location, etc.), and minimum patch size requirements combined with
30
extensive literature review and expert opinion as described by McKerrow et al. [65]. Simi-
larly, the putative source population grid for each focal species was mapped by identifying
patches of potential habitat that were large enough to support at least 200 territories [123].
Territory size was determined from literature review (Table 3.2).
We developed binary grids of suitability for the priority vegetation types using land-
scape characteristics and landform (Table 3.1). These data were used to exclude areas
where it was considered impossible for a speci c vegetation type to occur. For all vegetation
types, we used landform data derived from National Elevation Dataset for Southeastern Gap
Analysis Project [65]. For maritime-associated vegetation types (maritime forest, estuary,
beach), their inland extent was restricted by a manually digitized boundary based on the
maximum extent of existing maritime-associated habitats. Conservation lands for SAMBI
were extracted from the protected areas database for the United States [98] which includes
federal, state, non-governmental, and land trust lands set aside for conservation. We inter-
sected suitability data with the PAD-US data for each vegetation type so that only suitable
conservation lands were included in the prioritization for each respective vegetation type.
Grassland, longleaf and mature open pine could all be managed with  re so we excluded
urban areas from their conservation priority surfaces. For the urban layer we included areas
categorized as developed open space, and low-, medium-, and high-intensity developed.
3.3.4 Density estimation
After we developed the binary grids for all the landscape characteristics, we calculated
the density of each binary grid to calculate conservation priority for each grid cell. Areas
where a given characteristic were clustered were given a high density value (Fig. 3.1). We
mapped density using a two-dimensional kernel density estimator [110] in the Kernel Density
Estimation Toolbox (kdtools[7]) using MATLAB. The estimator is used to calculate density
based on probability of occurrence across a gridded space. For each observation, weight is
assigned to each grid cell as a function of the distance from the observation to the center of
31
the cell. We used a bivariate normal kernel with a  xed bandwidth which assigns weights that
decline rapidly in a sigmoid fashion with distance from each observation based on the spatial
distribution of all observations. For potential habitat, suitability, and conservation lands we
use the normal scale rule, and bandwidth, the parameter that determines the diameter of
the kernel (kernel size, h), based on the equation:
h = 1:0592qn 0:2 (3.1)
where h is the bandwidth, n is the number of observations, and
q = min(std(x);R=1:34) (3.2)
where x is the grid point(s), std is the standard deviation of x, and R is the interquartile
range of x.
To emulate colonization potential of putative source populations for each species, we
used a kernel size equal to the estimated dispersal distance. We used the larger of natal or
breeding dispersal distance based on published sources (see Table 3.2 for list of references),
or for species without known dispersal distances, we used an allometric equation based on
size and diet classi cation (Table 3.1) [119]. For urban areas, we used kernel size of 1200
m, which represents a distance of minimal impediment to the use of prescribed  re (Grand
unpubl.). Each density dataset was scaled to a range of 0-1 by:
w = w=max(w) (3.3)
where w is the weight assigned to each grid cell, before being used in further analysis
to avoid unequal weighting in the prioritization of surfaces with di erent numbers of density
grids.
32
3.3.5 Modeling priorities
We calculated priority by combining the grids of densities for all associated focal species,
suitability, conservation land, and urban data where appropriate (Fig. 3.2). During meetings
with experts, we asked them what data layers were essential for each vegetation type [72]
and we used this information to create models for vegetation types.
To calculate priority surfaces for vegetation types where experts determined  re was a
limiting characteristic (Table 3.1), we used the equation:
Pri = (1  Ui)SiCi (3.4)
where for vegetation type i, Pri is the grid of priority scores (priority surface); and Si,
and Ci are the density grids for suitability and conservation lands, and U is the density grid
of urban areas.
For all other vegetation types that did not include the urban layer, we used the equation:
Pri = SiCi (3.5)
Pri was also scaled to range of 0-1 by:
Pri = Pri=max(Pri) (3.6)
Priority surfaces were calculated using focal species and where  re was a limiting char-
acteristic, we used:
Pri = (1  Ui)Si
0
@Ci +
miX
j=1
EHij +
miX
j=1
SPij
1
A (3.7)
where for vegetation type i, Pr is the grid (matrix) of priority scores; and S and C
are the weights for suitability and conservation lands, 1  U is the weight of the non-urban
33
areas and for the m focal species, EH and SP are the potential habitat and putative source
population weights focal species j.
For all other vegetation types that did not include the urban layer, we used the equation:
Pri = Si
0
@Ci +
miX
j=1
EHij +
miX
j=1
SPij
1
A (3.8)
Probability surfaces were scaled to a range of 0-1 by:
Pri = Pri=max(Pri) (3.9)
To weight priority surfaces by focal species value, we used the above equations with the
addition of focal species value (fsv). For example, including the urban layer:
Pri = (1  Ui)Si
0
@Ci +
miX
j 1
fsv(EHij +SPij)
1
A (3.10)
3.3.6 Analysis
We made spatially-explicit comparisons of priority surfaces by subtracting one priority
surface from another. Because surfaces were normed to one, areas that were not di erent
received a score of 0, and di erences scored between -1 and 1. We de ned no di erence
between surfaces as less than 95% of pixels di ering. We also compared surfaces by calcu-
lating skew and kurtosis of priority value distributions. Skewness ( ) measures distribution
symmetry as follows:
 =
nX
i=1
 
qi   q3
 
=(n 1)std3 (3.11)
where q is the priority score of an individual pixel and  q the mean score of all pixels for
a surface, n is the number of pixels and std is the standard deviation of the priority scores.
A normal distribution has a value of 0, negative values indicate left-skewed distribution, i.e.
a long left tail, and positive values indicated right-skew.
34
Kurtosis ( ) which measures peakedness is:
 =
 nX
i=1
 
qi   q4
 
=(n 1)std4
!
 3 (3.12)
where q is the priority score of an individual pixel and  q the mean score of all pixels
for a surface, n is the number of pixels and std is the standard deviation of the priority
scores. A normal distribution has a kurtosis value of 0 (more accurately known as excesses
kurtosis), more peaked distributions (leptokurtotic) having positive values and less peaked
distributions, negative values.
3.4 Results
3.4.1 Focal species selection
There were 53 experts who participated in the elicitation meetings and 15 who provided
our species rankings. We did not remove any species from the focal species list based on the
ranks, but we added 11 species to our model based on the expert scores. As the initial focal
species list was 31 species, our prioritization process used a total of 42 species (Table 3.1)
[72].
3.4.2 Distribution of priority scores
For priority surfaces created using vegetation types, the maritime-associated vegetation
types (beach, estuary, and maritime forest) and the non-alluvial forested wetlands had pos-
itive values for skew ( > 1; 3.3) indicating that a small number of sites were given high
priority (Pr > 0:75). Grassland was the only vegetation type to show a left skew ( > 1)
and the rest of the vegetation types were normally distributed (1 < < 1). All vegetation
types had positive kurtosis values indicating distributions more peaked than normal, with
beach, maritime forest and estuary having the most highly peaked distributions ( > 10;
Fig. 3.4).
35
The distribution of priority scores was highly skewed ( > 4) and leptokurtotic for
narrowly distributed vegetation types (beach, estuary, and maritime forest) for priority sur-
faces of both weighted and unweighted focal species. For vegetation types that were less
restricted, slope forest, upland forest, grassland, and wetlands distributions demonstrated
moderate skewness (1 < < 4) and less leptokurtosis than narrowly distributed vegetation
types, and for vegetation types with a relatively wide distribution across the SAMBI, skew-
ness was low ( < 1) and they were mesokurtotic. Overall for all vegetation types except
beach, priority surfaces created using focal species values were less symmetrical and more
peaked than surfaces created without fsv.
3.4.3 Priority maps
Generally, using only vegetation type density to create the conservation priority maps
resulted in more high priority areas except for the most restricted vegetation types, those
that were maritime-associated. Beach (Fig. 3.6), estuary (Fig. 3.7), and maritime forest
(Fig. 3.10) showed very similar distributions regardless of the method used to create the
priority surfaces (Fig. 3.18 and 3.19), although there were slightly more mid-priority areas
when only vegetation type density was used. The area of highest priority occurred along
the coast from Brunswick, Georgia to Charleston, South Carolina. High priority areas for
wetlands (Fig. 3.16 and 3.20) were also found in this area when focal species densities were
used to create the priority surfaces. In contrast, when vegetation type density alone was
used, the priority of the interior of SAMBI increased signi cantly and the highest priority
area occurred around the Osceola National Forest and the Okefenokee National Wildlife
Refuge on the Florida-Georgia border.
The highest priority areas of alluvial forested wetland for all methods were concentrated
in 6 regions along the lower Roanoake River; Waccamaw, Little, and Great PeeDee drainage;
Cooper River; con uence of the Congaree and Wateree Rivers; middle and lower Savannah
River; and the lower Altamaha River (Fig. 3.5). Only a small percentage (< 2%) of the
36
SAMBI was of high conservation priority (Pr>75th percentile) for alluvial forested wetland
bird habitat using focal species; however using just vegetation type, there was a much greater
percentage of the area which was high priority. High priority areas for nonalluvial forested
wetlands (Fig. 3.11), slope (Fig. 3.14) and upland forest (Fig. 3.15) systems occurred
in association with alluvial forested wetland vegetation type when priority surfaces were
created using focal species. Conversely, using only vegetation type density to create the
priority surfaces for these vegetation types resulted in much more broadly distributed and
more numerous high priority areas.
In comparison, moderate priority (0:25 < Pr < 0:75) grassland conservation areas
were common (> 97%) and widely distributed (Fig. 3.8). This was due to the relatively
widespread occurrence of suitable areas for grassland conservation across the SAMBI. How-
ever, using focal species compared with only vegetation type density produced very di erent
priority surfaces. Using focal species, the highest priority regions (Pr > 0:75) occurred in
interior portions of northeast Florida from south and east of Valdosta, Georgia to north-
west of Gainesville, Florida. Using only vegetation type, most of SAMBI region including
northern areas which were of low priority using focal species, was high priority with only
urban areas standing out as unsuitable. The distribution of priority areas for scrub was also
widely distributed and compared with grassland, extended farther north in SAMBI using
focal species (Fig. 3.13) and was lower priority using only vegetation type.
The highest priority (Pr > 0:75) areas for longleaf conservation were concentrated in
northern Florida and along the Georgia-Florida border south of Thomasville, Georgia when
the priority surface was created using focal species (Fig. 3.9). There were more high priority
areas for longleaf when only vegetation type density was used to create the priority surface
and the highest priority areas were situated around Apalachicola National Forest in Florida
and in Francis Marion National Forest in South Carolina with more moderate priorities
throughout North and South Carolina.
37
Moderate priority areas for conservation of open pine systems were widespread because
suitable areas were widely available. Using focal species, the highest priority areas also fell
along the Georgia-Florida border south of Thomasville, Georgia (Fig. 3.12). When only
vegetation type density was used, the highest priority areas were much farther inland along
the fall line that separates the coastal plain and Appalachian ecoregions. There were also high
priority open pine conservation areas around the national forests in Florida (Apalachicola
and Osceola), and South Carolina (Francis Marion) and the Okefenokee National Wildlife
Refuge.
Priority surfaces created using focal species values were similar to those without except
priority values were generally lower (Fig. 3.17, 3.18, 3.19, and 3.20).
3.5 Discussion
One of our objectives was to integrate principles of conservation We used kernel density
estimation to provide a single approach to address conservation area size, shape, and connec-
tivity which allowed us to integrate principles of conservation biology into a spatially explicit
conservation planning process. Kernel density estimation prioritized large patches of appro-
priate sites because it considered the neighborhood around each pixel so areas with a high
density of appropriate sites had the highest priority. Larger, denser patches minimize edge
and maximize core area, and both are desirable characteristics for reserve design [22] [57].
Our method produces rounded shapes because kernel density is calculated over an elliptical
or circular window, which smoothed the resulting surface. This was especially evident for
long linear sites like those of maritime forest which showed increased conservation priority in
areas that made the sites less linear and more round, although they still remained restricted
to areas adjacent to the ocean. Kernel density estimation also allowed us to increase the pri-
ority value of sites within the distance a bird could move through varying the kernel size. We
used the dispersal distance of birds as the distance over which connection between patches
was important. Birds have relatively large dispersal distance, especially the focal species we
38
considered (all greater than 11,000 m except Red-cockaded woodpecker; Table 3.2) but our
method could be modi ed for other species for which connectivity is a more limiting factor
in designing conservation reserves by setting kernel size at a smaller distance.
3.5.1 Proxies for conservation
Of the three prioritization methods, conservation priority surfaces created using focal
species without fsv provided the best re ection of species habitat requirements. Vegetation
type densities alone did provide somewhat comparable priority surfaces for vegetation types
that were least constrained with respect to suitable sites, for example non-alluvial forested
wetlands. The priority surfaces calculated with fsv were very similar to those created with-
out, suggesting gathering information to calculate focal species values may not be an e ective
use of resources. However, the focal species selection process [72] could be easily combined
with the online ranking tools we used to gauge focal species values. Our process split the
expert elicitation process into a resource-intensive two day set of in person meetings and a
follow-up online survey that took less than an hour to complete. The online ranking process
could be used alone to delineate a focal species suite by selecting the top ranked species for
further modeling, as well as being used to calculate fsv. Having the entire process online
would require much less time and money although the di erence in participation rates be-
tween the in person and online parts of our process suggest there may be a participation
problem with a solely online survey.
The di erence between using focal species or vegetation type as the proxy in conser-
vation prioritization was much greater than between using fsv or not. This di erence was
most pronounced in wide-spread habitats that were not restricted in their site requirements
and in habitats which had species with restricted ranges, for example grasslands, open pine
and slope and upland forests. The maritime-associated vegetation types were restricted in
suitable sites that conservation surfaces based on focal species vegetation type density were
in the same locations; however, skew for beach and estuary showed much higher values when
39
focal species were included. For vegetation types associated with rivers (alluvial forested
wetlands, slope and upland forests), there were substantial di erences between priority sur-
faces created using vegetation type and focal species proxies. Most noticeably focal species
based priority surfaces had small areas of high priority for these habitats along rivers, while
the surfaces created with vegetation types had generally moderate priority over the entire
region. For these vegetation types, the priority maps with the focal species are more re-
 ective of species habitat requirements and therefore species conservation, than the priority
maps without the focal species. Similarly, priority surfaces for grasslands and open pine
habitat had lower conservation values in northern regions using focal species, In this case,
both focal species for both vegetation types were at their northern range limits within the
SAMBI extent and areas north of their ranges were of less value than areas within their
range.
Which priority surface stakeholders should use in their decisions is best determined by
examining the stakeholder objectives. If stakeholder goals are to conserve wildlife, it is im-
portant to use focal species in the conservation prioritization process because the resulting
surface better represents species habitat requirements by integrating species-speci c land-
scape preferences. On the other hand, if a more general objective of vegetation or habitat
conservation is preferred, priority surfaces based on vegetation type density could be used.
Although there were di erences between priority surfaces calculated using focal species or
vegetation types, the areas of highest conservation priority are similar between the methods
for most habitats. The exception was riparian-associated habitats for which using vegetation
type density only provided a poor model of conservation priorities because high priorities
were not associated with rivers. Although using vegetation type alone to create priority
surfaces requires less data and the data are more readily available (all sourced from publi-
cally available datasets), it does not necessarily re ect species habitat use well enough. This
problem only occurs with some vegetation types, but unless it is known ahead of time which
40
vegetation types would be a ected, we suggest conservation planners use focal species to
create priority surfaces.
3.5.2 Comparison with other conservation prioritization methods
Previous work on conservation prioritization methods can be broken down into two
broad categories: optimization and heuristic processes. Optimization method considers all
possible arrangements of sites and selects the solution that best meets the objective [33].
Common objectives include every species represented at a minimum of number of sites or
conserve the most species in protected sites with the least cost. Heuristic methods vary but
their general process is to select a site that best ful lls the objective, then select another site
that compliments that initial selection, and continue to select sites until the objective has
been met [69]. So for example, at the  rst step the site with the rarest species is selected; that
site is then taken out of consideration and the site with the next rarest species is selected,
etc. MARXAN (http://www.uq.edu.au/marxan/) is a popular heuristic algorithm that is
used to select the minimum number of sites or minimum total area needed to represent all
biodiversity for a speci c cost. Heuristic methods often overestimate the number of sites
compared to the optimal solution but optimization is di cult or impossible to apply to large
or complex problems [96] [133].
Our prioritization method is signi cantly di erent from optimization and heuristic pri-
oritization methods because it remains in a spatial context and is able to integrate several
principles of conservation reserve design. It has proved di cult to integrate spatial criteria
into either optimization or heuristic methods [133] [33]. Rules can be set to minimize dis-
tance between sites [82]; maximizing the number of adjacent sites selected [75] or minimize
perimeter length to increase compactness of selected sites [63], but applying multiple rules
increases the complexity of the calculations for the solution. The output of optimization and
heuristic methods also do not give the conservation value of sites that were not selected for
41
conservation. Our output evaluates all sites on the landscape, like optimization methods,
but gives a conservation priority value for all sites, unlike optimization.
3.5.3 Complications
Our priority surfaces may overestimate conservation value to species because priority
areas are often a mixture of land cover types, although the ratio of appropriate to inappro-
priate habitat is correlated with the priority value, i.e. areas with high conservation priority
have higher proportion of suitable habitat than low priority areas. This is because kernel
density estimation does not require all pixels of similar priority levels to be one vegetation
type. However, some vegetation types can be considered complimentary habitats rather
than incompatible habitats. For example, in our project grassland is distinct from open pine
but the undergrowth species use may be contiguous between these vegetation types, with
grassland lacking the scattered pines of open pine habitat. Therefore, grassland mixed with
open pine habitat should still have high conservation value for open pine species. Moreover,
our prioritization method is based not just on prioritization of current habitat (i.e. existing
habitat grids) but also on the potential to restore habitat in areas where it does not occur
but the site is suitable for restoration (i.e. suitability grids).This apparent overestimation of
conservation value is actually a re ection of the potential of a site rather than just current
conditions.
Our process places higher priority on sites that are within or immediately adjacent to
existing patches of suitable habitat, unless the areas under consideration in low priority areas
are large enough to be of higher total value. We did not attempt to estimate this relationship,
but, sensitivity analysis could determine the threshold sizes at which parity between levels
of conservation priority occur. Further research should focus on the e ects of patch size as
they relate to species requirements and priorities.
Over 85% of the land in the Southeastern US is privately owned [17] so stakeholders,
which in our project did not include private landowners, are restricted to acquiring land
42
that is available for purchase or easement and for which they have su cient resources to
buy. Restoration and management of existing patches of habitat can be done on both
publically owned and privately owned lands through cooperation with landowners, but is
again limited by time and monetary resources. All of these factors limit the actions available
to our stakeholders with respect to conservation. For this reason our prioritization method
is particularly useful since it compares the full extent of the region so comparisons can be
made even among moderate or low priority areas. We did not delineate polygons of highest
conservation priority to create a set of ideal reserve sites but a set of conservation areas could
be drawn by arbitrarily selecting a priority level or patch.
43
Figure 3.1: Priority surfaces start with (a) gridded space in which (b) resources are found
and (c) kernel density estimation is used to calculate the density of these resources
Figure 3.2: Density maps of (a) focal species and (b) suitable sites are combined and (c)
urban areas are subtracted in order to create a priority surface for a  re-dependent vegetation
type
44
Figure 3.3: Skew for conservation priority values of vegetation types in SAMBI
Figure 3.4: Kurtosis for conservation priority values of vegetation types in SAMBI
45
Figure
3.5
:Alluvial
forested
wetland
conserv
ation
priorit
ysurfaces
(w
armer
colors
indicate
higher
priorit
y)
for
eac
hv
ege
tation
typ
eu
sing
(a)
onl
ydensit
yof
vegetation
typ
es;
(b)
densities
of
veg
etation
typ
es
and
focal
sp
ecies;
(c)
de
nsities
of
vegetation
typ
es
and
focal
sp
ecies
using
focal
sp
ecies
value
s.
Inset
histograms
are
distribut
ion
of
priorit
yv
alues
for
eac
hsurf
ace
46
Figure
3.6:
Beac
hconserv
ation
priorit
ysurfaces
(w
armer
colors
indicate
higher
priorit
y)
for
eac
hv
egetation
typ
eusing
(a)
only
densit
yof
vegetation
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
de
nsities
of
vegetation
typ
es
and
focal
sp
ecies
using
focal
sp
ecies
values.
Inset
histograms
are
distribution
of
priorit
yv
alues
for
eac
hsurface
47
Figure
3.7:
Estuary
conser
vation
prio
rit
ysurfaces
(w
armer
color
sindicate
high
er
priorit
y)
for
eac
h
vegetation
typ
eusing
(a)
only
densit
yof
vegetation
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
densities
of
vegetation
typ
es
and
focal
sp
ecies
usin
gfo
cal
sp
ecies
values.
Inset
histograms
are
distribution
of
priorit
yv
alues
for
eac
hsurface
48
Figure
3.8:
Grasslan
dconserv
ation
priorit
ysurfaces
(w
armer
colors
indicate
higher
priorit
y)
for
eac
hv
egetation
typ
eusing
(a)
only
densit
yof
vegetation
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
densities
of
vegetation
typ
es
and
focal
sp
ecies
usin
gfo
cal
sp
ecies
values.
Inset
histograms
are
distribution
of
priorit
yv
alues
for
eac
hsurface
49
Figure
3.9:
Longleaf
conserv
ation
priorit
ysurfaces
(w
armer
colors
indicate
higher
pr
iorit
y)
for
eac
h
vegetation
typ
eusing
(a)
only
densit
yof
vegetation
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
densities
of
vegetation
typ
es
and
focal
sp
ecies
usin
gfo
cal
sp
ecies
values.
Inset
histograms
are
distribution
of
priorit
yv
alues
for
eac
hsurface
50
Figure
3.10:
Maritime
forest
conserv
ation
priorit
y
surfaces
(w
armer
colors
indicate
higher
priorit
y)
for
eac
h
vegetation
typ
e
using
(a)
only
densit
yof
vegetat
ion
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
densities
of
vegetation
typ
es
and
focal
sp
ecies
using
focal
sp
ecies
values.
Inset
histograms
are
distribution
of
priorit
yv
alues
for
eac
hsurface
51
Figure
3.11:
Non-alluvial
forested
wetland
conserv
ation
priorit
y
surfaces
(w
armer
colors
indicate
higher
priorit
y)
for
eac
h
vegetation
typ
eusing
(a)
only
densit
yof
vegetation
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
densities
of
vegetation
typ
es
and
focal
sp
ecies
using
focal
sp
ecies
values.
Inset
histograms
are
distribution
of
priorit
yv
alues
for
eac
hsurface
52
Figure
3.12:
Op
en
pine
conserv
ation
priorit
ysurfaces
(w
armer
colors
indicate
hig
her
priorit
y)
for
eac
hv
egetation
typ
eusing
(a)
only
densit
yof
vegetation
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
densities
of
vegetation
typ
es
and
focal
sp
ecies
usin
gfo
cal
sp
ecies
values.
Inset
histograms
are
distribution
of
priorit
yv
alues
for
eac
hsurface
53
Figure
3.13:
Shrub-scr
ub
conserv
ation
priorit
ysurfaces
(w
armer
colors
indicate
higher
priorit
y)
for
eac
hv
egetation
typ
eusing
(a)
only
densit
yof
vegetation
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
densities
of
vegetation
typ
es
and
focal
sp
ecies
using
focal
sp
ecies
values.
Inset
histograms
are
distribution
of
pr
iorit
yv
alues
for
eac
hsurface
54
Figure
3.14:
Slop
eforest
conserv
ation
priorit
ysurfaces
(w
armer
colors
indicate
higher
priorit
y)
for
eac
hv
egetation
typ
eusing
(a)
only
densit
yof
vegetation
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
densities
of
vegetation
typ
es
and
focal
sp
ecies
using
focal
sp
ecies
values.
Inset
histograms
are
distribution
of
pr
iorit
yv
alues
for
eac
hsurface
55
Figure
3.15:
Upland
forest
conserv
ation
priorit
ysurfaces
(w
armer
colors
indicate
higher
priorit
y)
for
eac
hv
egetation
typ
eusing
(a)
only
densit
yof
vegetation
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
densities
of
vegetation
typ
es
and
focal
sp
ecies
using
focal
sp
ecies
values.
Inset
histograms
are
distribution
of
pr
iorit
yv
alues
for
eac
hsurface
56
Figure
3.16:
W
etland
conserv
ation
prior
ity
surfaces
(w
armer
colors
indicate
higher
priorit
y)
for
eac
hv
egetation
typ
eusing
(a)
only
densit
yof
vegetation
typ
es;
(b)
densities
of
vegetation
typ
es
and
focal
sp
ecies;
(c)
densities
of
vegetation
typ
es
and
focal
sp
ecies
usin
gfo
cal
sp
ecies
values.
Inset
histograms
are
distribution
of
priorit
yv
alues
for
eac
hsurface
57
Figure 3.17: Di erence between priority surfaces created using vegetation only and focal
species (black bars with standard deviation); and focal species and fsv (gray bars with
standard deviation). Dashed line is 0.05 di erence and bars below this line we have de ned
as not signi cantly di erent.
58
Figure 3.18: For (a) alluvial forested wetland, (b) beach, (c) estuary, and (d) grassland,
di erence between priority surfaces created using (left) vegetation only and focal species;
and (right) focal species and fsv. Warm colors are where conservation priority was lower
for (left) focal species compared with vegetation only and (right) fsv compared with focal
species. Cool colors are where priorities were higher.59
Figure 3.19: For (e) longleaf, (f) maritime forest, (g) non-alluvial forest wetland, and (h)
open pine, di erence between priority surfaces created using (left) vegetation only and focal
species; and (right) focal species and fsv. Warm colors are where conservation priority was
lower for (left) focal species compared with vegetation only and (right) fsv compared with
focal species. 60
Figure 3.20: For (i)shrub-scrub, (j) slope forest, (k) upland forest, and (l) wetland, di erence
between priority surfaces created using (left) vegetation only and focal species; and (right)
focal species and fsv. Warm colors are where conservation priority was lower for (left) focal
species compared with vegetation only and (right) fsv compared with focal species.
61
Table
3.1:
Landforms
asso
ciated
with
general
vegetation
typ
es
of
the
Southeastern
United
States
(slop
eand
relativ
eelev
ation)
and
whether
 re
was
imp
ortan
tto
managemen
tof
the
vegetation
typ
e.
All
vegetation
typ
es
also
excluded
op
en
water
(streams,
lak
es,
and
oceans).
Y
=
included
in
mo
del.
Steepslop
e
Slop
e
crest
Upp
er
slop
e
Flatsummit
Plateau at
Sideslop
e
Ra
vine
Dry at
Moist at
Slop
e
bottom
Fire
Slop
e
>
25
o
6 
25
o
6 
25
o
<
6o
<
6o
<
6o
6 
25
o
<
6o
<
6o
<
6o
Elev
ation
highest
high
highest
mid
low
mid
mid
low
Vegetation
typ
e
Alluvial
forested
wetlands
Y
Y
Y
Y
Y
Beac
h
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Estuary
Y
Grassland
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Longleaf
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Maritime
forest
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Non-alluvialforested
wetland
Y
Y
Y
Y
Y
Y
Y
Op
en
pine
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Shrub-scrub
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Slop
eforest
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
Upland
forest
Y
Y
Y
Y
Y
Y
Y
Y
Y
W
etland
Y
Y
Y
62
Table
3.2:
Territory
size
and
disp
ersal
distance
for
focal
bird
sp
ecies.
Disp
ersal
distance
was
based
on
the
maxim
um
dista
nce
of
published
adult
(A)
or
natal
(N)
dis
persals
or
calcu
lated
(C)
using
an
allometric
equation
[119].
W
edid
not
use
either
territo
ry
size
or
disp
ersal
distance
for
win
tering
sp
ecies.
Fo
cal
sp
ecies
value
(f
sv
)w
as
deriv
ed
fro
m
exp
ert
input
Sp
ecies
Territorysize
Disp
ersal
distance
Typ
e
disp
ersal
Reference
fsv
Habitat
(m
2)
(m)
Acadian
 ycatc
her
16,300
1,000
A
[131]
2.5
upland
forest
Empidonax
vir
esc
ens
3.1
slop
eforest
American
blac
kduc
k
30,000
300
A
[53]
2.8
wetland
Anas
rubrip
es
2.5
estuary
American
kestrel
1,000,000
38,800
N
[112]
2.6
longleaf
Falc
osp
arverius
American
oystercatc
her
1,000
20,000
A
[76]
2.1
beac
h
Haematopus
pal
liatus
Bac
hman?s
sparro
w
25,000
21,000
C
[24]
0.9
scrub
Peuc
aea
aestivalis
2.0
longleaf
1.7
op
en
pine
Blac
k-throated
green
warbler
14,000
13,000
C
[73]
1.3
allufor
Dendr
oic
avir
ens
1.7
nonallufor
Bro
wn-headed
nuthatc
h
28,000
19,000
C
[134]
1.6
nonallufor
Sitta
pusil
la
2.1
longleaf
2.0
op
en
pine
Cerulean
warbler
10,400
1,000
A
[35]
1.4
allufor
Dendr
oic
ac
erule
a
1.2
upland
forest
1.2
slop
eforest
Ch
uc
k-will?s-wido
w
70,000
54,000
C
[115]
1.4
nonallufor
Caprimulgus
car
olinensis
Common
ground
do
ve
4,200
a
100,000
N
[9]
1.4
maritime
forest
Columbina
passerina
Field
sparro
w
7,600
20,000
C
[11]
1.9
scrub
Spizel
la
pusil
la
63
Sp
ecies
Territorysize
Disp
ersal
distance
Typ
e
disp
ersal
Reference
fsv
Habitat
(m
2)
(m)
Henslo
w?s
sparro
w
1,800
20,000
C
[39]
1.6
grassland
Ammo
dramus
henslowii
1.3
scrub
1.4
longleaf
Ho
oded
warbler
6,250
14,000
C
[14]
1.9
slop
eforest
Wilsonia
citrina
Ken
tuc
ky
warbler
22,100
500
A
[62]
1.3
upland
forest
Op
orornis
formosus
King
rail
4,000
b
31,000
C
[87]
1.9
wetland
Ral
lus
ele
gans
Least
bittern
97,000
44,000
C
[88]
1.7
wetland
Ixobrychus
exilis
Least
tern
3
80,000
A
[124]
1.8
beac
h
Sternula
antil
larum
Loggerhead
shrik
e
80
,000
70,000
N
[136]
1.5
grassland
Lanius
ludovicianus
1.9
scrub
1.3
longleaf
Louisiana
waterthrush
16,000
4,000
N
[60]
1.3
upland
forest
Parkesia
motacil
la
1.8
slop
eforest
Nelson?s
sharp-tailed
sparro
w
22,000
c
20,000
A
[108]
1.9
estuary
Ammo
dramus
nelson
Northern
bob
white
16,187
100,000
N
[10]
1.6
grassland
Colinus
vir
ginianus
2.0
scrub
1.8
longleaf
1.8
op
en
pine
Northern
parula
4,000
11,000
C
[71]
1.5
nonallufor
Parula
americ
ana
1.5
maritime
forest
Northern
pin
tail
win
tering
1.7
wetland
Anas
acuta
64
Sp
ecies
Territorysize
Disp
ersal
distance
Typ
e
disp
ersal
Reference
fsv
Habitat
(m
2)
(m)
Pain
ted
bun
ting
20,000
8,000
A
[55]
2.0
maritime
forest
Passerina
ciris
Piping
plo
ver
280
d
600,000
A
[?]
2.2
beac
h
Char
adrius
melo
dus
Prairie
warbler
15,000
3,400
A
[77]
1.8
scrub
Dendr
oic
adisc
olor
1.3
maritime
for
est
1.6
longleaf
1.4
op
en
pine
Prothonotary
warbler
10,000
3,900
N
[85]
2.2
allufor
Pr
otonotaria
citr
ea
1.6
nonallufor
Red
knot
win
tering
1.9
beac
h
Calidris
canutus
Red-co
ckaded
wo
odp
eck
er
80
0,000
5,400
N
[46]
1.3
nonallufor
Pic
oides
bor
ealis
2.0
longleaf
1.9
op
en
pine
Red-headed
wo
odp
eck
er
30,000
1,040
A
[113]
1.4
nonallufor
Melanerp
es
erythr
ocephalus
1.6
longleaf
Redhead
win
tering
1.0
estuary
Aythya
americ
ana
Saltmarsh
sharp-tailed
sparro
w
win
tering
1.9
estuary
Ammo
dramus
caudacutus
Sandhill
Crane
920,000e
45,000
C
[120]
1.0
grassland
Grus
canadensis
Seaside
sparro
w
5,100
1,600
A
[91]
1.9
estuary
Ammo
dramus
maritimus
Summer
tanager
100,000
22,000
C
[103]
n/a
Pir
anga
rubr
a
Sw
ainson?s
warbler
14,000
15,400
N
[4]
1.9
allufor
Limnothlypis
swainsonii
1.3
upland
forest
65
Sp
ecies
Territorysize
Disp
ersal
distance
Typ
e
disp
ersal
Reference
fsv
Habitat
(m
2)
(m)
1.8
slop
eforest
Sw
allo
w-tailed
kite
12
,000,000
140,000
C
[66]
1.6
allufor
Elanoides
for c
atus
W
oo
dduc
k
125,000
3,800
N
[38]
1.9
allufor
Aix
sponsa
1.7
nonallufor
W
oo
dstork
3
340,000
C
[19]
1.5
wetland
Mycteria
americ
ana
Yello
w-throated
warbler
22,800
13,000
C
[64]
1.9
allufor
Dendr
oic
adominic
a
1.3
maritime
for
est
a[101]b[47]c[109]d[16]e[8]
66
Chapter 4
Conservation priorities in an uncertain future
4.1 Abstract
We targeted avian conservation in the Southeastern United States in partnership with
the Atlantic Coast Joint Venture, because bird abundance in the United States has been de-
clining for over half a century. This is likely a result of habitat changes due to urbanization
and forest management. Global climate change is expected to lead to warming temperatures
and changes in precipitation that should further a ect bird habitat. We wanted to provide a
tool to enable stakeholders to conserve species and habitats that are currently present and to
integrate future habitat conditions to allow species to respond to climate change. Our con-
servation prioritization method was developed for use over a large geographic area extending
from southern Virginia through northern Florida. It also allowed us to abide by principles
of conservation biology and conservation reserve design, including prioritizing larger areas
over smaller ones, and preferentially selecting areas in close proximity to one another. We
designed conservation priority areas for two habitats, open pine and maritime forest, that
are expected to respond to di erent aspects of climate change, increased fragmentation and
sea level rise, respectively. Land cover projections were developed for years 2000 to 2100 at
10-year time intervals for three global climate change models, speci cally the a2, a1b, and b1
climate scenarios. We included  ve binary spatio-temporal grids to prepare habitat priority
maps: (1) potential habitat and (2) putative source population distributions for each of the
focal species; (3) suitability models for each habitat; (4) conservation lands; and (5) urban
areas. For both open pine and maritime forest habitats, di erences between priority surfaces
created with discounted or summed future conditions a ected how valuable areas were to
conservation but not where those areas were within the region, and surfaces did not di er
67
signi cantly between climate scenarios. Similarities among alternatives of future conditions
may be a result of scale because changes due to climate change may have a strong local,
but weak regional e ect. Having six similar alternatives is helpful because it suggests a set
of consistent conservation priorities that can be relied upon to conserve bird populations in
the South Atlantic Migratory Bird Initiative region. As additional information is gathered
relating to climate-change-driven land cover changes, the alternatives may diverge which
makes repeating the modeling process very important.
4.2 Introduction
Adding new protected areas or extending current protected areas is key to enable species
to adapt to climate change [32]. Modeling future land use/land cover given various cli-
mate change scenarios and integrating this information into conservation prioritization plans
should increase population persistence as species respond to climate change. Without in-
tegrating these future conditions, suitable habitat may shift out of protected areas where
species may become vulnerable to land use changes. For example, if a habitat is expected to
shift northward, prioritizing habitat conservation towards the northern edge of the habitat?s
range should allow species to shift their range as the habitat shifts.
A major obstacle to implementing conservation goals occurs when researchers design
solutions and practitioners do not understand the solution, interpret the solution as too
prescriptive or researchers simply fail to address problems useful to practitioners [52] [93].
This is particularly the case with dynamic problems that are mathematically complex and
di cult to implement [90] such as the e ect of climate change on species and ecosystem per-
sistence. Alternatively, simple target-based approaches with management actions applied
in a highly controlled system are problematic because these models are too simple and do
not address real world complexity [27]. The trade-o then is between conservation plan-
ning using complex mathematical solutions that are di cult to interpret and implement,
and oversimplifying the problem which reduces e ectiveness. We propose a new method of
68
conservation prioritization that is simple to use compared with traditional optimization or
heuristic methods and integrates a number of biological and spatial objectives in a multi-
species context.
It is important to both conserve what is currently present, in terms of species, habitats
and ecosystems, and to plan for coming conservation challenges. We targeted avian con-
servation in the Southeastern United States in partnership with the Atlantic Coast Joint
Venture, because bird abundance in the United States has been declining for over half a cen-
tury, likely a result of habitat changes [125] [78]. In the Southeastern U.S., urban areas are
growing at the expense of forests [30] and short-term projections suggest urbanization will
continue to reduce forest area and increase fragmentation [129]. In the long term, climate
change is expected to lead to changes in precipitation and temperature patterns [44]. In the
Southeastern US, future precipitation trends are uncertain but temperatures are expected to
warm by 2-3C degrees in the next 100 years [44]. We predicted a warmer climate would lead
to land cover changes that would shift terrestrial birds northward and would shift coastal
species inland with rising sea levels. The conservation prioritization models based on result-
ing species distributions would therefore also shift northward and inland. In order to allow
species to respond to climate change, we must integrate future conservation priorities with
current priorities.
Conservation prioritization should combine information about species and habitat re-
quirements over large geographic extents based on principles of conservation biology, in-
cluding prioritizing larger areas over smaller ones, and preferentially selecting areas in close
proximity [22]. Prioritization should integrate future land use/land cover conditions under
di erent climate scenarios to provide conservationists with tools for determining how impor-
tant various areas are to conservation now and in the future. Finally, such tools should be
developed in collaboration with stakeholders to ensure the result is a tool that will be useful.
69
We examine two avian habitats that are expected to be a ected by climate change:
open pine and maritime forests under three climate scenarios. Open pine forest is an impor-
tant habitat for several endemic bird species in the Southeastern US including Red-cockaded
Woodpecker (Picoides borealis), Bachman?s Sparrow (Peucaea aestivalis), and Brown-headed
Nuthatch (Sitta pusilla). Open pine forests are similar to longleaf pine (Pinus palustris)
forests in that they have an open canopy and a grassy ground layer from which shrubs are
excluded by  re. These open canopy pine forests historically dominated the Southeastern
US but have been much reduced since European settlement. In open pine forests, climate
change is expected to increase  re and insect outbreak frequency leading to more fragmen-
tation. Maritime forest is a coastal habitat that is important for migratory neotropical
songbirds such as Bicknell?s Thrush (Catharus bicknelli), as well as providing breeding habi-
tat for species of concern including Painted Bunting (Passerina ciris), Common Ground
Dove (Columbina passerina), and Prairie Warbler (Setophaga discolor). Maritime forest is
already much reduced from its historic extent due to human development along the coast
and climate change will add additional stress to the habitat through increased sea levels.
Because these two habitats will respond di erently to climate change, we can compare how
conservation priorities may shift over time due to di erent processes.
4.3 Methods
4.3.1 Study site
The study area was the Coastal Plain of the Southeastern United States including
the eastern portions of North Carolina, South Carolina, Georgia as well as a small part of
southern Virginia and part of the panhandle of Florida (Fig. 4.1). Our project focused on
conservation of sustainable bird populations in this region in maritime forest and mature
open pine forest. Using species at risk associated with these two habitats, a suite of focal
species were selected by experts (Table 4.1) [72].
70
4.3.2 Future land cover
We developed data from land cover projections for 2000-2100 at 10 year time intervals
for three di erent climate change scenarios [65]. The three scenarios, a2, a1b and b1, dif-
fered in regards to energy sources, global population size and economic growth [43]. The
a2 scenario was the worst case scenario of the three we considered. It simulated a continu-
ously increasing global population, a slow switch to more e cient technology, and continued
reliance on fossil fuels [43]. The best case scenario was b1 under which the global popu-
lation peaks mid-century and the economy switches to clean energy sources and there is a
general focus on sustainability. The other scenario, a1b, predicts some development of al-
ternative sources of energy, a population peaking at mid-century and a balance between the
use of fossil and non-fossil fuels. We use land cover projections computed by collaborators
at North Carolina State University. They used SLAMM (http://www.slammview.org/) to
model sea level rise, SLEUTH (http://www.essc.psu.edu/SLEUTH/) to model urbanization
and VDDT (http://essa.com/tools/vddt/downloadvddt/) to model land change probabili-
ties. Downscaled climate change e ects were used to determine how much sea levels would
rise each decade and to modify the insect outbreak and  re-related disturbance frequency in
the land cover modeling.
Our objective was to develop a method for prioritizing habitat conservation across very
large landscapes incorporating current landscape conditions and predictions of future land-
scapes based on the presumption that the con guration, size, and proximity of conserved
areas a ects the sustainability of targeted populations. We included up to  ve binary spatio-
temporal datasets (grids) used to prepare habitat priority maps: (1) potential habitat and
(2) putative source population distributions for each of the focal species; (3) suitability mod-
els for each habitat; (4) conservation lands; and (5) urban areas. The SAMBI region was
divided into a regular 200 m grid and grid squares were evaluated for each dataset descrip-
tion. For example, if protected areas occurred in a grid square, that square would get a value
71
of one in the conservation grid. We used ArcGIS Version 9.3 (Environmental Systems Re-
search Institute, Redlands, California) and MATLAB Version 7.10.0.499 (The Math Works,
Natick, Massachusetts) to create and manipulate spatial data.
4.3.3 Datasets
In our analysis, we used three binary grids based on habitat (1) suitability, (2) conserva-
tion land, and (3) urban area. We developed binary maps of suitability of locations for open
pine and maritime forest habitats using land cover projections, landscape characteristics,
and landform. These data were used to exclude areas where it was considered impossible
for a speci c habitat type to occur. For open pine, we used time-invariant landform data
derived from National Elevation Dataset (NED) for Southeast Gap Analysis Project [65].
However, we developed suitability maps for each time step for maritime forest based on
projected land cover because of expected changes in sea level and ensuing coast lines and
the low thematic resolution of NED. Maritime forest was restricted to its inland extent by a
manually digitized boundary at each time step based on the maximum extent of any existing
maritime-associated habitat which also included estuary and beach habitats.
Two binary grids were created for each focal species that bred in the South Atlantic
Migratory Bird Initiative (SAMBI) [128] region at each time step. Maps of potential habitat
were based on range maps, land cover, ancillary data (stream location, elevation, etc.),
and minimum patch size requirements determined by extensive literature review and expert
opinion as described by McKerrow [65]. Once future land covers were created, potential
habitat models for species were applied at each time step. Putative source populations of
each focal species were mapped by identifying patches of potential habitat that were large
enough to support at least 200 territories [123] (Table 4.1).
Conservation lands for SAMBI were extracted from the protected areas database for
the United States (PAD-US) [98], which includes federal, state, non-governmental, and land
trust lands set aside for conservation. We intersected suitability data with the PAD-US data
72
for each habitat so that only suitable conservation lands were included in the prioritization
for each respective habitat. This grid was calculated once for all time steps.
Because land cover projections incorporated urban growth, urban areas were extracted
from the land use projections for each time step. We excluded areas mapped as developed
open space, low- medium- and high-intensity developed lands.
4.3.4 Density estimation
We mapped density of each binary layer using a two-dimensional kernel density esti-
mator [110] in the Kernel Density Estimation Toolbox (kdtools [7]) using MATLAB. The
estimator was used to calculate density based on probability of occurrence across a grid-
ded space. For each observation, weight was assigned to each grid cell as a function of the
distance from the observation to the center of the cell. We used a bivariate normal kernel
with a  xed bandwidth which assigns weights that decline rapidly in a sigmoid fashion with
distance from each observation. The weights were summed across all observations resulting
in a smoothed surface that re ects the two-dimensional density of resources. For potential
habitat, suitability, and conservation land grids we used the normal scale rule, and band-
width, the parameter that determines the diameter of the kernel (kernel size, h), based on
the equation:
h = 1:0592qn 0:2 (4.1)
where h is the bandwidth, n is the number of observations, and
q = min(std(x);R=1:34) (4.2)
where x is the grid point(s), std is the standard deviation of x, and R is the interquartile
range of x. We divided the grid of weights by the density of a binary grid indicating the
extent of terrestrial areas in order to avoid underestimating density on the edges of our
73
study area and large waterbodies [68]. To emulate colonization potential of putative source
populations for each species, we used a kernel size equal to the estimated dispersal distance
(Table 1). We used the larger of natal or breeding dispersal distance, or for species without
known dispersal distances, an allometric equation based on size and diet classi cation [119].
For urban areas, we used kernel size of 1200 m, which represents a distance of maximal
impediment to the use of prescribed  re (Grand unpubl.). Each density grid was scaled to
range of 0-1 by:
w = w=max(w) (4.3)
where w is the weight assigned to each grid cell before being used in further analysis to
avoid unequal weighting in the prioritization.
4.3.5 Modeling priorities
We calculated priorities for each habitat in each time step by combining the grids of
weights for all associated focal species, suitability and conservation lands data, and urban
data where appropriate. During meetings with experts from state and federal government
and non-governmental organizations, we asked which data were essential for habitat function
[72] and we used this information to create models for both habitats. For mature open pine,
where experts determined  re was a limiting characteristic, we used the equation:
Prit = (1  Uit)Sit
0
@Ci +
miX
j=1
EHijt +
miX
j=1
SPijt
1
A (4.4)
where for habitat i in time t, Pri is the grid of priority scores (priority surface); and
Si, and Ci are the density grid for suitability and conservation lands, U is the density grid
of the urban areas and for the mi focal species, EHi and SPi are the potential habitat and
putative source population weights focal species j.
Maritime forest did not include the urban grid and was thus:
74
Prit = Sit
0
@Ci +
miX
j=1
EHijt +
miX
j=1
SPijt
1
A (4.5)
Prit was also scaled to range of 0-1 by:
Prit = Prit=max(Prit) (4.6)
To incorporate the e ects of climate and urban growth as expressed in the land cover
projections, we summed the priority surfaces for each time step in an undiscounted fashion:
Prit =
TX
t=1
Prit (4.7)
or we applied a 4% per year discounting rate:
Prit =
TX
t=1
(Prit(1 + 0:04)10) (4.8)
This process reduced the value of future land cover conditions and give them less in u-
ence in our conservation plans. The discounting rate was based on the 30-year United States
government treasury bond rate [6].
4.3.6 Analysis
To evaluate the di erences between priority surfaces, we subtracted one grid from an-
other. Because surfaces were normed to one, areas that were not di erent received a score
of 0, and di erences scored between -1 and 1.
We also compared distributions of the priority values of the surfaces using skew ( ):
 =
nX
i=1
 
qi   q3
 
=(n 1)std3 (4.9)
where q is the priority score of an individual pixel and  q the mean score of all pixels for
a surface, n is the number of pixels and std is the standard deviation of the priority scores.
75
A normal distribution has a value of 0, negative values indicate left-skewed distribution, i.e.
a long left tail, and positive values indicated right-skew.
To look at the shape of the distribution curves of the priority surfaces , we also used
kurtosis ( ):
 =
 nX
i=1
 
qi   q4
 
=(n 1)std4
!
 3 (4.10)
where q is the priority score of an individual pixel and  q the mean score of all pixels for
a surface, n is the number of pixels and std is the standard deviation of the priority scores.
Skewness measured symmetry of the distribution where a normal distribution equaled 0,
negative values indicated left skewed distribution, i.e. a long left tail, and positive values
indicated right skew. A left skewed distribution would indicate less high priority habitat
than with a normal distribution. Kurtosis measured peakedness with normal distribution
having a value of 0, more peaked distributions (leptokurtotic) having positive values and
less peaked distributions, negative values. A highly peaked distribution would suggest more
areas with mid-value priority for conservation.
4.4 Results
We produced a total of 363 priority maps for 11 species every 10 years from 2000 through
2100 for three di erent climate scenarios. In all climate scenarios (Fig. 4.2), high priority
open pine habitat occurred near large patches of existing forest: in northern Florida in the
region of Apalachicola National Forest, near Oconee National Forest in Georgia, along the
coast of South Carolina near Francis Marion National Forest and inland near the Coastal
Plain/Piedmont boundary in South and North Carolina in the area of Uwharrie, Pisgah
and Sumter National Forests. The highest priority areas for maritime forest occurred along
the coast of South Carolina, around Hilton Head, and Georgia, around Cumberland Island
National Seashore and Jekyll Island State Park (Fig. 4.3).
76
4.4.1 Projected priority surfaces
We produced 6 priority surfaces for each habitat by combining the projected priorities
for each climate scenario by both discounting and summing future conditions (Fig. 4.4 and
4.5). For open pine, there was more area of moderate conservation priority when future
priorities were integrated compared with the initial priority surface, with the discounted
priority surface having slightly more high priority areas than the summed surface (Fig. 4.8).
Maritime forest, on the other hand, was more restricted in extent and the di erence between
the discounted and undiscounted priority surfaces were minimal as was the di erence between
the initial and summed or discounted surfaces. The di erences between climate scenarios
for both habitats were minimal (Fig. 4.7).
Spatially, the largest change in priority values between the initial and  nal time step
priority surfaces for open pine was in northern Florida with the western area decreasing in
priority while the eastern regions increased (Fig. 4.6). Over time, maritime forest increased
in priority along the Gulf Coast of Florida and in coastal areas near the Virginia/North
Carolina border. The coastal region of the Carolinas showed large decreases in priority likely
due to sea level rise.
4.4.2 Habitat priorities
The distribution of priority scores for maritime forest at each time step was highly right-
skewed (  3:2) and leptokurtotic (  11:1) indicating a small number of sites were given
high (Pr> 0:75) priority (Fig. 4.9). For open pine, which had a relatively wide distribution
across the SAMBI region, skewness was low (  1:0) and the distribution was similar to
normal distribution ( 0:5    1:3). For both habitats, there was minimal di erence in
skewness or kurtosis between climate scenarios.
77
4.5 Discussion
Our results illustrate the di culties associated with prioritizing areas for conservation
based on predictions of future conditions in the SAMBI region. Priority areas were shifted,
more so for open pine habitat than maritime forest. However, the overall e ect of climate
change on the distribution of priority areas in both habitats was not very di erent compared
to priorities based on 2000 (recent) land cover. This was particularly true when we discounted
to allow for uncertainty in the prediction of future landscapes.
Modeling the response of ecosystems in the southeastern US to global climate change
is di cult for several reasons. Future climate trends in this area are not as clear as those in
other areas of the US, making down-scaling the global climate models di cult. The IPCC
predicts slightly warming temperatures over the next 30 to 40 years [43], but models do not
agree on the direction of precipitation trends. This disagreement among the climate predic-
tions is problematic to models that are based on the e ects of temperature and precipitation
patterns on mortality and growth rates of vegetation [84]. However, the vegetation models
that were used to create our predicted land cover data were based on pest outbreaks and  re
frequency, not changes due directly to precipitation even though the rates of tree pest and
disease outbreaks are expected to increase due to changes in temperature or precipitation
[84].
Additionally, the time frame over which we are considering land cover changes may be
too short for Southeastern ecosystems to show major changes. Loblolly pine trees have an
average longevity of over 150 years [117] so our 100 year timeframe represents less than one
generation of unharvested trees. Although it is unlikely major range shifts due to climate
change could occur quickly, catastrophic events like introduced pest species or hurricanes
could result in major mortality and succession could be signi cantly a ected by climate
changes [43]. The process used by our colleagues at NCSU to model future conditions does
rely on these catastrophic, although highly unpredictable events, to a ect forest conditions.
78
Birds, unlike vegetation, could be more likely to show range shifts in response to climate
change within the next 100 years and many of the focal species we used in our modeling are
at or near their northern range limits in the SAMBI region [24] [9] [55] [134] [46]. Bachmans
Sparrow, Brown-headed Nuthatch, Red-cockaded Woodpecker, Common Ground Dove and
Painted Buntings should therefore be expected to increase their density or extend their range
northward if the resources they depend on are a ected by temperature [58] [59]. Nevertheless,
these species will probably not respond to climate change solely through temperature; it is
much more likely their responses will be through interactions between species including
predator/prey dynamics, phenology, pest and disease susceptibility [84]. Because we only
used models based on predicted patterns of potential habitat on the landscape, we cannot
address all processes that may have in uence species abundance or distribution.
The process of creating prioritization surfaces can help stakeholders better de ne their
objectives [29] [89]. With maritime forest prioritization, stakeholders realized priority sur-
faces created for maritime forest are insu cient to plan conservation areas at a small scale.
Areas of high conservation priority occur along the coast of Georgia and South Carolina,
but  ner scale prioritization is needed to identify speci c parts of the coast. For example,
it may turn out that sheltered bays have the most amount of maritime forest and other
maritime-associated habitats of high conservation priority, but that requires smaller scale
modeling of sea level rise than our project could address. The process of prioritization for
open pine habitat has also led stakeholders to question the e ects of conservation lands on
the prioritization values. Conservation lands are included directly in our modeling processes
as a grid, as well as indirectly as the location of large tracts of existing habitat and putative
source populations. Stakeholders worried conservation priorities were limited to areas adja-
cent to already established conservation areas and were required to better de ne the role of
the prioritization surfaces is it important to acknowledge how important current protected
areas are for conserving species or should existing protected areas be excluded from the
prioritization surface?
79
There is often pressure to produce the best single answer to conservation problems but
providing stakeholders with a range of answers illustrates the range of uncertainty in the
system. We produced a series of possible alternatives to the question of where conservation
priorities occur under di erent climate change scenarios. The IPCC [45] considers all cli-
mate scenarios equally sound and discourages averaging among scenarios. The three climate
scenarios we modeled, therefore, gave us three alternatives for future conditions. We also
modeled summary priority surfaces using two di erent methods, either discounting future
condition or summing without discounting. This gives stakeholders six alternatives for con-
servation planning. Surprisingly however, the di erences among these alternatives were not
pronounced. The di erences between priority surfaces created using discounted or summed
future conditions a ected how valuable areas were to conservation but not where those areas
were within the SAMBI region, and our priority surfaces did not di er signi cantly between
climate scenarios. This similarity may be a result of scale because changes due to climate
change, speci cally sea level rise and  re or insect outbreaks, may have a strong local, but
weak regional e ects.
Adaptive resource management is a cyclical process [42] and allows managers to make
decisions in situations with a high degree of uncertainty, such as global climate change [5].
Our project will be used in an adaptive management context and therefore, we do not want
to suggest which alternative priority surface is best, only that the decision support tools we
produced can help in the next stage of the process. When additional information becomes
available about the e ect of climate change or about stakeholder objectives, this information
can be integrated into our process and new decision support tools can be developed. Our
priority surfaces are forecasts of future conditions in the form of tools to allow stakeholders
to make better decisions [67].
While there are many sources of uncertainty in our models of conservation priority areas
in the Southeastern US, it is important to avoid using uncertainty to prevent conservation
80
actions [32]. Our results showed conservation priority surfaces changed in intensity so val-
ues of areas changed over the time, but at large scales, areas of high conservation priority
remained of high priority regardless of their exact conservation value. The consistency of
relative priorities gives con dence the areas we highlight as high priority will remain vital
for conservation into the future.
"If we wait decades for certain knowledge of climate-change e ects, land-use
change will have already dictated the conservation landscape, and the scope
for adapting to climate change will be minimal. Conversely, if we act now, we
will have to act in the face of considerable uncertainty. Dealing intelligently
with uncertainty in a landscape with considerable space to make choices seems
our best option. Acting intelligently will therefore require taking some risks and
convincing society and policy makers that risks are worth taking. The alternative
is letting uncertainty become an excuse for inaction." [32]
81
Figure 4.1: The study region in the Southeastern United States included coastal plain regions
of Virginia, North Carolina, South Carolina, Georgia, and Florida.
82
Figure
4.2:
Conserv
ation
priorit
y
surfaces
for
op
en
pine
forest
in
the
Southeastern
US
sho
wing
di eren
ty
ears
and
climate
scenarios:
(a)
conserv
ation
prio
rities
for
initial
conditions
(2000);
(b)
conserv
ation
priorities
pro
jected
to
2100
under
climate
scenarios
a1b,
a2
and
b1.
Inset
 gures
sho
w
distribution
of
conserv
ation
priorit
yv
alues.
W
arm
co
lors
indicate
areas
of
high
conserv
ation
priorit
yand
co
ol
colors
indicate
low
er
priorit
y.
83
Figure
4.3:
Conserv
ation
priorit
y
surfaces
for
maritim
eforest
in
the
Southeastern
US
sho
wing
di eren
ty
ears
and
climate
scenarios:
(a)
conserv
ation
prio
rities
for
initial
conditions
(2000);
(b)
conserv
ation
priorities
pro
jected
to
2100
under
climate
scenarios
a1b,
a2
and
b1.
Inset
 gures
sho
w
distribution
of
conserv
ation
priorit
yv
alues.
W
arm
co
lors
indicate
areas
of
high
conserv
ation
priorit
yand
co
ol
colors
indicate
low
er
priorit
y.
84
85
Figure 4.4: Open pine  nal conservation priority surface including land cover from all years
(2000-2100), for climate scenarios a1b, a2 and b1; (a) summed and (b) discounted by 4%
per year. Warm colors indicate areas of high conservation priority and cool colors indicate
lower priority.
86
Figure 4.5: Maritime forest  nal conservation priority surface including land cover from all
years (2000-2100), for climate scenarios a1b, a2 and b1; (a) summed and (b) discounted
by 4% per year. Warm colors indicate areas of high conservation priority and cool colors
indicate lower priority
87
Figure 4.6: Di erence between initial conservation priorities and those of 2100 for (a) open
pine and (b) maritime forest for climate scenario a2. Warm colors indicate where conservation
priority increased over time and cool colors where priorities decreased. Green designates no
change in priority values.
88
Figure 4.7: Di erence among open pine and maritime forest priority surfaces for climate
scenarios, a2 (worst case scenario), b1 (best case scenario), and a1b (middle way) Reference
line indicates less that 5% di erence between the surfaces which we have de ned as no
signi cant di erence.
89
Figure 4.8: Distribution of priority values for open pine (top) and maritime forest (bot-
tom) habitats in 2000 (Initial), summed across 100 years (Summed), and with a 4% annual
discount (Discounted) under a2 climate scenario.
90
Figure 4.9: Skew (top) and kurtosis (bottom) of priority scores for open pine (yellow, blue,
and pink curves) and maritime forest (black, red, and green curves) by year of land cover
projection under a2, a1b, and b1 climate scenarios. Both habitats show positive values
of skew which indicates long right tails and little area of high conservation priority, but
maritime forest had higher skew values. Comparatively, while maritime forest has large
positive kurtosis values which indicate highly peaked distribution of priority values, open
pine has kurtosis values around zero which indicates a normal distribution. There was little
variation among climate scenarios so curves largely overlap.
91
Table
4.1:
Habitat
asso
ciation,
selection
metho
d,
territor
ysize,
and
disp
ersal
dis
tance
for
focal
bird
sp
ecies
for
spatially-explicit
conserv
ation
prioritization
pro
cess
in
th
eSoutheastern
United
States.
Fo
cal
sp
ecies
were
selected
using
structure
decision
making
(SDM),
Lam
bec
ks
pro
cess
(Lam
bec
k),
both
selection
metho
ds
(b
oth)
or
using
an
online
su
rvey
(on
line).
Territo
ry
size
was
based
on
published
literature.
Disp
ersal
distance
was
based
on
the
maxim
um
distance
of
published
adul
t(A)
or
natal
(N)
disp
ersals
or
calculated
(C)
using
an
allometric
equation
[119]
Habitat
Sp
ecies
Latin
name
Selectionmetho
d
Territorysize
Disp
ersal
distance
Typ
e
of disp
ersal
Reference
Op
en
pine
Bac
hman?s
sparro
w
Peuc
aea
aestivalis
Lam
bec
k
25,000
21,00
0
C
[24]
Bro
wn-headed
nuthatc
h
Sitta
pusil
la
SDM
28,000
19,000
C
[134]
Field
sparro
w
Spizel
la
pusil
la
online
7,600
20,000
C
[11]
Northern
bob
white
Colinus
vir
ginianus
Lam
bec
k
16,187
100,0
00
N
[10]
Red-co
ckaded
wo
odp
eck
er
Pic
oides
bor
ealis
both
800,000
5,400
N
[46]
Maritime
forest
Common
ground
do
ve
Columbina
passerina
both
4,200
100,000
N
[9]
Northern
parula
Parula
americ
ana
online
4,000
11,000
C
[71]
Pain
ted
bun
ting
Passerina
ciris
Lam
bec
k
20,000
8,000
A
[55]
Yello
w-throated
warbler
Dendr
oic
adominic
a
online
22,800
13,000
C
[64]
Both
Prairie
warbler
Dendr
oic
adisc
olor
Lam
bec
k
15,000
3,400
A
[77]
92
Bibliography
[1] American Ornithologists Union (AOU). 1998. Check-list of North American
birds, 7th ed. American Ornithologists Union, Washington, DC.
[2] Amici V, F Geri, and C Battisti. 2010. An integrated method to create
habitat suitability models for fragmented landscapes. Journal for Nature
Conservation 18:215223.
[3] Andelman SJ, and WF Fagan. 2000. Umbrellas and  agships: e cient
conservation surrogates or expensive mistakes? Proceedings of the National
Academy of Sciences USA 97:59545959.
[4] Anich, NM, TJ Benson, JD Brown, C Roa, JC Bednarz, RE Brown
and JG Dickson. 2010. Swainson?s Warbler (Limnothlypis swainsonii),
The Birds of North America Online (A Poole, ed.). Ithaca: Cornell
Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/126
[5] Arvai, J., G Bridge, N Dolsak, R Franzese, T Koontz, A Luginbuhl, P
Robbins, K Richards, K Smith Korfmacher, B Sohngen, J Tansey, and
A Thompson. 2006. Adaptive management of the global climate problem:
bridging the gap between climate research and climate policy. Climatic
Change 78: 217-225.
[6] Bazelon, C, and K Smetters. 1999. Discounting inside the Washington D.C.
Beltway. The Journal of Economic Perspectives 13: 213-228.
[7] Beardah, CC, and MJ Baxter. 1996. MATLAB Routines for Kernel Den-
sity Estimation and the Graphical Presentation of Archaeological Data. In:
Interfacing the Past, Computer Applications and Quantitative Methods in
Archaeology, Analecta Prehistorica Leidensia 28, (H Kammermans and K
Fennema, eds.). University of Leiden, Leiden, pp. 179-184.
[8] Bennett, AJ. 1989. Movements and home ranges of Florida sandhill cranes.
Journal of Wildlife Management 53: 830-836.
[9] Bowman, R. 2002. Common Ground-Dove (Columbina passerina), The
Birds of North America Online (A Poole, ed.). Ithaca: Cornell Lab
of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/645
93
[10] Brennan, LA. 1999. Northern Bobwhite (Colinus virginianus), The
Birds of North America Online (A Poole, ed.). Ithaca: Cornell Lab
of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/397
[11] Carey, M, M Carey, DE Burhans and DA Nelson. 2008. Field Sparrow
(Spizella pusilla), The Birds of North America Online (A Poole, ed.).
Ithaca: Cornell Lab of Ornithology; Retrieved from the Birds of North
America Online: http://bna.birds.cornell.edu/bna/species/103
[12] Caro TM. 2001. Species richness and abundance of small mammals inside
and outside an African national park. Biological Conservation 98:251257.
[13] Caro TM and G O?Doherty. 1999. On the use of surrogate species in con-
servation biology. Conservation Biology 13:805814.
[14] Chiver, I, LJ Ogden and BJ Stutchbury. 2011. Hooded Warbler (Wilsonia
citrina), The Birds of North America Online (A Poole, ed.). Ithaca: Cornell
Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/110
[15] Cipollini KA, AL Maruyama, and CL Zimmerman. 2005. Planning for
restoration: a decision analysis approach to prioritization. Restoration
Ecology 13:460470.
[16] Cohen, JB, EH Wunker, JD Fraser. 2008. Substrate and vegetation selec-
tion by nesting piping plovers. Wilson Journal of Ornithology 120: 404-407.
[17] Connor, RC and A.J. Hartsell. 2002. Chapter 16: Forest area and con-
ditions. In: Southern forest resource assessment. USDA Forest Service,
Southern Research Station, General Technical Report SRS-53 (DN Wear
and JG Greis, eds.). USDA Forest Service, Asheville, NC, pp. 357-402.
[18] Coppolillo P, H Gomez, F Maisels, and R Wallace. 2004. Selection crite-
ria for suites of landscape species as a basis for site-based conservation.
Biological Conservation 115:419430.
[19] Coulter, MC, JA Rodgers, JC Ogden and FC Depkin. 1999. Wood Stork
(Mycteria americana), The Birds of North America Online (A Poole, ed.).
Ithaca: Cornell Lab of Ornithology; Retrieved from the Birds of North
America Online: http://bna.birds.cornell.edu/bna/species/409
[20] Cowling RM and RL Pressey RL 2003 Introduction to systematic con-
servation planning in the Cape Floristic Region. Biological Conservation
112:113.
[21] Dame R, M Alber, D Allen, M Mallin, C Montague, A Lewitus, A
Chalmers, R Gardner, C Gilman, B Kjerfve, J Pinckney, and N Smith.
94
2000. Estuaries of the South Atlantic Coast of North America: their geo-
graphical signatures. Estuaries 23:793819.
[22] Diamond, JM.1975. The island dilemma: lessons of modern biogeographic
studies for the design of natural reserves. Biological Conservation 7: 129-
146.
[23] Didier KA, D Wilkie, I Douglas-Hamilton, L Frank, N Georgiadis, M Gra-
ham, F Ihwagi, A King, A Cotterill, D Rubenstein, and R Woodro e. 2009.
Conservation planning on a budget: a "resource light" method for mapping
priorities at a landscape scale? Biodiversity Conservation 18:19792000.
[24] Dunning, JB. 2006. Bachman?s Sparrow (Peucaea aestivalis), The
Birds of North America Online (A Poole, ed.). Ithaca: Cornell Lab
of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/038
[25] Elliott-Smith, E and SM Haig. 2004. Piping Plover (Charadrius melo-
dus), The Birds of North America Online (A Poole, ed.). Ithaca: Cornell
Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/002
[26] Favreau JM, CA Drew, GR Hess, MJ Rubino, FH Koch, and KA Eschel-
bach. 2006. Recommendations for assessing the e ectiveness of surrogate
species approaches. Biodiversity Conservation 15:39493969.
[27] Ferrier, S, DP Faith, A Arponen, and M Drielsma. 2009. Community-level
approaches to spatial conservation prioritization. In: Spatial conservation
prioritization: quantitative methods and computational tools. (A Moilanen,
KA Wilson, HP Possingham, eds.). Oxford University Press, New York,
NY, pp. 94-109.
[28] Fleishman E, RB Blair, and DD Murphy. 2001. Empirical validation of a
method for umbrella species selection. Ecological Applications 11:14891501.
[29] Gregory RS and RL Keeney. 2002. Making smarter environmental man-
agement decisions. Journal of the American Water Resources Association
38:16011612.
[30] Gri th JA, SV Stehman, and TR Loveland. 2003. Landscape trends in
mid-Atlantic and southeastern United States ecoregions. Environmental
Management 32:572588.
[31] Guyer C and MA Bailey. 1993. Amphibians and reptiles of longleaf pine
communities. In: Hermann SM ed The longleaf pine ecosystem: ecology,
restoration and management. Proc. 18th Tall Timbers Fire Ecology Conf.
Tall Timbers Research, Inc., Tallahassee, FL, pp 139158.
95
[32] Hannah, L. 2011. Climate change, connectivity, and conservation success.
Conservation Biology 25: 1139-1142.
[33] Haight, RG and SA Snyder. 2009. Integer programming methods for reserve
selection and design. In: Spatial conservation prioritization: quantitative
methods and computational tools. (A Moilanen, KA Wilson, HP Possing-
ham, eds.). Oxford University Press, New York, NY, pp. 43-57.
[34] Hamel, PB. 1992. Land manager?s guide to the birds of the South. The
Nature Conservancy, Chapel Hill, NC.
[35] Hamel, PB. 2000. Cerulean Warbler (Dendroica cerulea), The Birds
of North America Online (A Poole, ed.). Ithaca: Cornell Lab of
Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/511
[36] Hammond, JS, RL Keeney and H Rai a. 1999. Smart choices. Broadway
Books, New York, NY.
[37] Hanski, I. 1999. Metapopulation ecology. Oxford University Press, New
York, NY.
[38] Hepp, GR and FC Bellrose. 1995. Wood Duck (Aix sponsa), The
Birds of North America Online (A Poole, ed.). Ithaca: Cornell Lab
of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/169
[39] Herkert, JR, PD Vickery and DE Kroodsma. 2002. Henslow?s Sparrow (Am-
modramus henslowii), The Birds of North America Online (A Poole, ed.).
Ithaca: Cornell Lab of Ornithology; Retrieved from the Birds of North
America Online: http://bna.birds.cornell.edu/bna/species/672
[40] Hess, GR and TJ King. 2002. Planning open spaces for wildlife. I. Select-
ing focal species using a Delphi survey approach. Landscape and Urban
Planning 58:2540.
[41] Hodges, JD. 1997. Development and ecology of bottomland hardwood sites.
Forest Ecology and Management 90:117125.
[42] Holling CS. 1978. Adaptive environmental assessment and management.
John Wiley and Sons, New York, NY.
[43] Intergovernmental Panel on Climate Change (IPCC). 2007a. Climate
change 2007: synthesis report. Contribution of Working Groups to the
Fourth Assessment Report of the Intergovernmental Panel on Climate
Change. Cambridge University Press, Cambridge.
96
[44] Intergovernmental Panel on Climate Change (IPCC). 2007b: Summary
for Policymakers. In: Climate Change 2007: The Physical Science Basis.
Contribution of Working Group I to the Fourth Assessment Report of the
Intergovernmental Panel on Climate Change. Cambridge University Press,
Cambridge, United Kingdom and New York, NY, USA.
[45] Intergovernmental Panel on Climate Change (IPCC). 2000. Summary for
Policymakers: IPCC special report, emissions scenarios. A special report of
IPCC Working Group III. Cambridge University Press, Cambridge, United
Kingdom and New York, NY, USA.
[46] Jackson, JA. 1994. Red-cockaded Woodpecker (Picoides borealis), The
Birds of North America Online (A Poole, ed.). Ithaca: Cornell Lab
of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/085
[47] James, RD. 1985. Habitat management guidelines for birds of Ontario wet-
lands: including marshes, swamps and fens or bogs or various types. On-
tario Ministry of Natural Resources, Peterborough, ON.
[48] Janis, IL. 1972. Victims of groupthink. Houghton Mi in Company, Boston,
MA.
[49] Kuhnert, PM, Martin TG, and SP Gri ths. 2010. A guide to eliciting and
using expert knowledge in Bayesian ecological models. Ecological Letters
13:900914.
[50] Lambeck, RJ. 1997. Focal species: a multi-species umbrella for nature con-
servation. Conservation Biology 11:849856.
[51] Landres, PB, J Verner, and JW Thomas JW. 1988. Ecological uses of
vertebrate indicator species: a critique. Conservation Biology 2:316327.
[52] Lauber, TB, RC Stedman, DJ Decker, and BA Knuth. 2011. Linking knowl-
edge to action in collaborative conservation. Conservation Biology 25: 1186-
1194.
[53] Longcore, JR, DG Mcauley, GR Hepp and JM Rhymer. 2000. American
Black Duck (Anas rubripes), The Birds of North America Online (A Poole,
ed.). Ithaca: Cornell Lab of Ornithology; Retrieved from the Birds of North
America Online: http://bna.birds.cornell.edu/bna/species/481
[54] Low-Choy S, R O?Leary, and K Mengersen. 2009. Elicitation by design
in ecology: using expert opinion to inform priors for Bayesian statistical
models. Ecology 90:265277.
[55] Lowther, PE, SM Lanyon and CW Thompson. 1999. Painted Bunting
(Passerina ciris), The Birds of North America Online (A Poole, ed.).
97
Ithaca: Cornell Lab of Ornithology; Retrieved from the Birds of North
America Online: http://bna.birds.cornell.edu/bna/species/398
[56] MacNally, R, E Fleishman, JR Thomson, and DS Dobkin. 2008. Use of
guilds for modelling avian responses to vegetation in the Intermountain
West USA. Global Ecology Biogeography 17:758769.
[57] Margules CR, and RL Pressey. 2000. Systematic conservation planning.
Nature 405:243253.
[58] Matthews, SN, LR Iverson, AM Prasad, and MP Peters. 2007-ongoing.
A climate change atlas for 147 bird species of the eastern United States
database. Northern Research Station, USDA Forest Service, Delaware,
Ohio. Available from: http://www.nrs.fs.fed.us/atlas/bird (accessed May
2012)
[59] Matthews, SN, LR Iverson, AM Prasad, and MP Peters. 2011. Changes in
potential habitat of 147 North American breeding bird species in response
to redistribution of trees and climate following predicted climate change.
Ecography 34: 933-945.
[60] Mattsson, BJ, TL Master, RS Mulvihill and WD Robinson. 2009. Louisiana
Waterthrush (Parkesia motacilla), The Birds of North America Online (A
Poole, ed.). Ithaca: Cornell Lab of Ornithology; Retrieved from the Birds
of North America Online: http://bna.birds.cornell.edu/bna/species/151
[61] McCann RK, BG Marcot, and R Ellis. 2006. Bayesian belief networks: ap-
plications in ecology and natural resource management. Canadian Journal
of Forest Research 36:3053-3062.
[62] McDonald, MV. 1998. Kentucky Warbler (Oporornis formosus), The
Birds of North America Online (A Poole, ed.). Ithaca: Cornell Lab
of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/324
[63] McDonnell, MD, HP Possingham, IR Ball, and EA Cousins. 2002. Mathe-
matical methods for spatially cohesive reserve design. Environmental Mod-
eling and Assessment 7: 107-114.
[64] McKay, B, and GA Hall. 2012. Yellow-throated Warbler (Dendroica do-
minica), The Birds of North America Online (A Poole, ed.). Ithaca: Cor-
nell Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/223
[65] McKerrow AJ, SG Williams, AL Silvano, EA Kramer, KJ Kleiner, TS
Earnhardt, JW Lee, MJ Rubino, M Pyne, KW Samples, CM Belyea, AE
Ernst, JB Grand, MD MacKenzie, and JA Collazo. Southeast Gap Anal-
ysis Project. U.S. Geological Survey, Gap Analysis Program, Moscow, ID.
Available from: http://www.basic.ncsu.edu/segap/ (accessed May 2012)
98
[66] Meyer, KD. 1995. Swallow-tailed Kite (Elanoides for catus), The Birds
of North America Online (A Poole, ed.). Ithaca: Cornell Lab of
Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/138
[67] Millar, CI, NL Stephenson, and SL Stephens. 2007. Climate change and
forests of the future: managing in the face of uncertainty. Ecological Ap-
plications 17: 2145-2151.
[68] Miller, DA. 2004. Factors related to temporal and spatial variation in nest
survival for Dusky Canada Geese on the Copper River Delta, Alaska. MSc
thesis. Auburn University, Auburn, AL.
[69] Moilanen, A, and I Ball. 2009. Heuristic and approximate optimization
methods for spatial conservation prioritization. In: Spatial conservation
prioritization: quantitative methods and computational tools. (A Moilanen,
KA Wilson, HP Possingham, eds.). Oxford University Press, New York,
NY, pp. 58-69.
[70] Moilanen, A, AMA Franco, RI Early, R Fox, B Wintle, and CD Thomas.
2005. Prioritizing multiple-use landscapes for conservation: methods for
large multi-species planning problems. Proceedings of the Royal Society,
Series B, 272: 1885-1891.
[71] Moldenhauer, RR and DJ Regelski. 2012. Northern Parula (Setophaga
americana), The Birds of North America Online (A Poole, ed.). Ithaca:
Cornell Lab of Ornithology; Retrieved from the Birds of North America
Online: http://bna.birds.cornell.edu/bna/species/215
[72] Moody, AT and JB Grand. 2012. Incorporating expert knowledge in deci-
sion support models for bird conservation. In: Expert Knowledge and Its
Application in Landscape Ecology. (AH Perera, CA Drew, CJ Johnson,
eds.). Springer, New York, pp. 109-129.
[73] Morse, DH. and AF Poole. 2005. Black-throated Green Warbler (Dendroica
virens), The Birds of North America Online (A Poole, ed.). Ithaca: Cornell
Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/055
[74] Mowbray ,TB. 2002. Canvasback (Aythya valisineria), The Birds
of North America Online (A Poole, ed.). Ithaca: Cornell Lab of
Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/659
[75] Nalle, DJ, JL Arthur, and J Sessons. 2002. Designing compact and con-
tiguous reserve networks with a hybrid heuristic algorithm. Forest Science
48: 59-68.
99
[76] Nol, E and RC Humphrey. 1994. American Oystercatcher (Haematopus pal-
liatus), The Birds of North America Online (A Poole, ed.). Ithaca: Cornell
Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/082
[77] Nolan Jr., V, ED Ketterson and C A Buerkle. 1999. Prairie Warbler (Den-
droica discolor), The Birds of North America Online (A Poole, ed.). Ithaca:
Cornell Lab of Ornithology; Retrieved from the Birds of North America
Online: http://bna.birds.cornell.edu/bna/species/455
[78] North American Bird Conservation Initiative (NABCI), U.S. Committee.
2009. The state of the birds, United States of America, 2009. U.S. Depart-
ment of the Interior, Washington, DC.
[79] Noss, RF. 1990. Indicators for monitoring biodiversity: a hierarchical ap-
proach. Conservation Biology 4:355364.
[80] Noss, RF, C Carroll, K Vance-Borland, and G Wuerthner. 2002. A multi-
criteria assessment of the irreplaceability and vulnerability of sites in the
Greater Yellowstone ecosystem. Conservation Biology 16:895908.
[81] Olson, DM and E Dinerstein. 1998. The global 200: a representation ap-
proach to conserving the Earth?s most biologically valuable ecoregions.
Conservation Biology 12:502515.
[82]  Onal, H, and RA Briers. 2002. Incorporating spatial criteria in optimum re-
serve network selection. Proceedings of the Royal Society of London, Series
B, 269: 2437-2441.
[83] Outcalt, KW and RM She eld. 1996. The longleaf pine forest: trends
and current conditions. USDA Forest Service, Southern Research Station,
Asheville, Resource Bulletin SRS-9.
[84] Parmesan, C. 2006. Ecological and evolutionary responses to recent climate
changes. Annual Review of Ecology, Evolution, and Systematics 37: 637-
669.
[85] Petit, LJ. 1999. Prothonotary Warbler (Protonotaria citrea), The Birds
of North America Online (A Poole, ed.). Ithaca: Cornell Lab of
Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/408
[86] Pierce SM, RM Cowling, AT Knight, AT Lombard, M Rouget, and T Wolf.
2005. Systematic conservation planning products for land-use planning: in-
terpretations for implementation. Biological Conservation 125:441458.
[87] Poole, AF, LR Bevier, CA Marantz and B Meanley. 2005. King Rail (Rallus
elegans), The Birds of North America Online (A Poole, ed.). Ithaca: Cornell
100
Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/003
[88] Poole, AF, P Lowther, JP Gibbs, FA Reid and SM Melvin. 2009. Least
Bittern (Ixobrychus exilis), The Birds of North America Online (A Poole,
ed.). Ithaca: Cornell Lab of Ornithology; Retrieved from the Birds of North
America Online: http://bna.birds.cornell.edu/bna/species/017
[89] Possingham, HP, SJ Andelman, BR Noon, S Trombulak, and HR Pul-
liam. 2001. Making smart conservation decisions. In: Conservation biology
research priorities for the next decade (ME Soul and GH Orians, Eds).
Island Press, Washington, DC, pp. 225-244.
[90] Possingham, HP, A Moilanen, and KA Wilson. 2009. Accounting for habitat
dynamics in conservation planning. In: Spatial conservation prioritization:
quantitative methods and computational tools. (A Moilanen, KA Wilson,
HP Possingham, eds.). Oxford University Press, New York, NY, pp. 58-69.
[91] Post, W, W Post and JS Greenlaw. 2009. Seaside Sparrow (Ammodramus
maritimus), The Birds of North America Online (A Poole, ed.). Ithaca:
Cornell Lab of Ornithology; Retrieved from the Birds of North America
Online: http://bna.birds.cornell.edu/bna/species/127
[92] Power ME, D Tilman, JA Estes, BA Menge, WJ Bond, LS Mills, G Daily,
JC Castilla, J Lubchenco, and RT Paine. 1996. Challenges in the quest for
keystones. Bioscience 46:609620.
[93] Prendergast, JR, RM Quinn, and JH Lawton. 1998. The gaps between
theory and practice in selecting nature reserves. Conservation Biology 13:
484-492.
[94] Pressey, RL, HP Possingham, and CR Margules. 1996a. Optimality in re-
serve selection algorithms: when does it matter how much? Biological
Conservation 76: 259-267.
[95] Pressey, RL, S Ferrier, TC Hager, CA Woods, SL Tully, and KM Wein-
man. 1996b. How well protected are the forests of north-eastern New South
Wales? Analyses of forest environments in relation to formal protection
measures, land tenure, and vulnerability to clearing. Forest Ecology and
Management 85:311333.
[96] Pressey, RL, HP Possingham, and JR Day. 1997. E ectiveness of alternative
heuristic algorithms for identifying indicative minimum requirements for
conservation reserves. Biological Conservation 80: 207-219.
[97] Pressey RL and MC Bottrill. 2009. Approaches to landscape- and seascape-
scale conservation planning: convergence, contrasts and challenges. Oryx
43:464475.
101
[98] Protected Areas Database of the United States (PAD-US) 2010.
National Gap Analysis Program. Moscow ID. Available from:
http://gapanalysis.usgs.gov/padus/ (accessed December 2010)
[99] Ranganathan, J, KMA Chan, KU Karanth, and JLD Smith. 2008. Where
can tigers persist in the future? A landscape-scale density-based population
model for the Indian subcontinent. Biological Conservation 141: 67-77.
[100] Richardson, CJ. 2003. Pocosins: hydrologically isolated or integrated wet-
lands on the landscape? Wetlands 23:563576.
[101] Riveka-Milan, FE. 1996. Nest density and success of Columbids in Puerto
Rico. Condor 98:100-113.
[102] Roberge JM and P Angelstam. 2004. Usefulness of the umbrella species
concept as a conservation tool. Conservation Biology 18:7685.
[103] Robinson, WD. 1996. Summer Tanager (Piranga rubra), The Birds
of North America Online (A Poole, ed.). Ithaca: Cornell Lab of
Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/248.
[104] Rodrigues, ASL and TM Brooks. 2007. Shortcuts for biodiversity conser-
vation planning: the e ectiveness of surrogates. Annual Review of Ecology,
Evolution, and Systematics 38: 13-37.
[105] Roux, DJ, JL Nel, PJ Ashton, AR Deacon, FC deMoor, D Hardwick, L Hill,
CJ Kleynhans, GA Maree, J Moolman, and RJ Scholes. 2008. Designing
protected areas to conserve riverine biodiversity: lessons from a hypothet-
ical redesign of the Kruger National Park. Biological Conservation 141:
100-117.
[106] Sanderson EW, KH Redford, A Vedder, PB Coppolillo, and SE Ward. 2002
A conceptual model for conservation planning based on landscape species
requirements. Landscape and Urban Planning 58:4156.
[107] Shen, G, C Feng, Z Xie, Z Ouyang, J Li, and M Pascal. 2008. Proposed
conservation landscape for giant pandas in the Minshan Mountains, China.
Conservation Biology 22: 1144-1153.
[108] Shriver, WG, TP Hodgman and AR Hanson. 2011. Nelson?s Sparrow (Am-
modramus nelsoni), The Birds of North America Online (A Poole, ed.).
Ithaca: Cornell Lab of Ornithology; Retrieved from the Birds of North
America Online: http://bna.birds.cornell.edu/bna/species/719
[109] Shriver, WG, TP Hodgman, JP Gibbs, PD Vickery. 2010. Home range
sizes and habitat use of Nelsons and saltmarsh sparrows. Wilson Journal
of Ornithology 122: 340-345.
102
[110] Silverman, BW. 1986. Density estimation for statistics and data analysis.
Chapman and Hall, New York, NY.
[111] Simberlo , D. 1998. Flagships, umbrellas, and keystones: is single species
management pass in the landscape era? Biological Conservation 83:247257.
[112] Smallwood, JA and DM Bird. 2002. American Kestrel (Falco sparverius),
The Birds of North America Online (A Poole, ed.). Ithaca: Cornell
Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/602
[113] Smith, KG, JH Withgott and PG Rodewald. 2000. Red-headed Woodpecker
(Melanerpes erythrocephalus), The Birds of North America Online (A
Poole, ed.). Ithaca: Cornell Lab of Ornithology; Retrieved from the Birds
of North America Online: http://bna.birds.cornell.edu/bna/species/518
[114] Smith, RJ, J Easton, BA Nhancale, AJ Armstrong, J Culverwell, SD
Dlamini, PS Goodman, L Lo er, WS Matthews, A Monadjem, CM
Mulqueeny, P Ngwenya, CP Ntumi, B Soto, and N Leader-Williams. 2008.
Designing a transfrontier conservation landscape for the Maputaland cen-
tre of endemism using biodiversity, economic and threat data. Biological
Conservation 141: 2127-2138.
[115] Straight, CA. and RJ Cooper. 2000. Chuck-will?s-widow (Caprimulgus car-
olinensis), The Birds of North America Online (A Poole, ed.). Ithaca: Cor-
nell Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/499
[116] Stralberg, D, DL Applegate, SJ Phillips, MP Herzog, N Nur and N
Warnock. 2009. Optimizing wetland restoration and management for avian
communities using a mixed integer programming approach. Biological Con-
servation 142: 94-109.
[117] Strauss, SH and FT Ledig. 1985. Seedling architecture and life history
evolution in pines. American Naturalist 125: 702-715.
[118] SurveyMonkey. Palo Alto, CA. Available from:
http://www.surveymonkey.com/ (accessed September 2010)
[119] Sutherland, GD, AS Harestad, K Price, and KP Lertzman. 2000. Scaling
of natal dispersal distances in terrestrial birds and mammals. Conservation
Ecology 4: 16.
[120] Tacha, TC, SA Nesbitt and PA Vohs. 1992. Sandhill Crane (Grus canaden-
sis), The Birds of North America Online (A Poole, ed.). Ithaca: Cornell
Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/031
103
[121] Tallis, H, Z Ferdana and E Gray. 2008. Linking terrestrial and marine
conservation planning and threats analysis. Conservation Biology 22: 120-
130.
[122] Teck, SJ, BS Halpern, CV Kappel, F Micheli, KA Selkoe, CM Crain, R
Martone, C Shearer, J Arvai, B Fischho , G Murray, R Neslo, and R Cooke.
2010. Using expert judgment to estimate marine ecosystem vulnerability
in the California Current. Ecological Applications 20:14021416.
[123] Thomas, CD. 1990. What do real population dynamics tell us about mini-
mum viable populations sizes? Conservation Biology 4: 324-327.
[124] Thompson, BC, JA Jackson, J Burger, LA Hill, EM Kirsch and
JL Atwood. 1997. Least Tern (Sternula antillarum), The Birds of
North America Online (A Poole, ed.). Ithaca: Cornell Lab of
Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/290
[125] Valiela, I, and P Martinetto. 2007. Changes in bird abundance in eastern
North America: urban sprawl and global footprint? Bioscience 57:360370.
[126] Van Lear, DH, WD Carrol, PR Kapeluck, and R Johnson. 2005. History
and restoration of the longleaf pine-grassland ecosystems: implications for
species at risk. Forest Ecology and Management 211:150165.
[127] Walker, J. 1998. Ground layer vegetation in longleaf pine landscapes: an
overview for restoration management. In: Kush JS ed Proceedings of the
longleaf pine ecosystem restoration symposium. Longleaf Alliance, Fort
Lauderdale, Report no. 3, pp 213.
[128] Watson, C, and K Malloy. 2006. The South Atlantic Migratory Bird Initia-
tive implementation plan: an integrated approach to conservation of "all
birds across all habitats". Atlantic Coast Joint Venture Report, Charleston,
SC.
[129] Wear, D, J Pye and K Ritters. 2004. De ning conservation priorities using
fragmentation forecasts. Ecology and Society 9:4.
[130] Wear, DN. 2002. Chapter 6: land use. In: Wear DN, Greis JG Eds South-
ern forest resource assessment. USDA Forest Service, Southern Research
Station, Asheville, Gen. Tech. Rep. SRS-53, pp 153173.
[131] Whitehead, DR and T Taylor. 2002. Acadian Flycatcher (Empidonax
virescens), The Birds of North America Online (A Poole, ed.). Ithaca:
Cornell Lab of Ornithology; Retrieved from the Birds of North America
Online: http://bna.birds.cornell.edu/bna/species/614
104
[132] Wiens, JA, Hayward GD, Holthausen RS, Wisdom MJ. 2008. Using surro-
gate species and groups for conservation planning and management. Bio-
science 58:241252.
[133] Williams, JC, CS ReVelle, and SA Levin. 2004. Using mathematical op-
timization models to design nature reserves. Frontiers in Ecology and the
Environment 2: 98-105.
[134] Withgott, JH and KG Smith. 1998. Brown-headed Nuthatch (Sitta pusilla),
The Birds of North America Online (A Poole, ed.). Ithaca: Cornell
Lab of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/349
[135] Woodin, MC, Michot TC. 2002. Redhead (Aythya americana), The
Birds of North America Online (A Poole, ed.). Ithaca: Cornell Lab
of Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/695
[136] Yosef, R. 1996. Loggerhead Shrike (Lanius ludovicianus), The Birds
of North America Online (A Poole, ed.). Ithaca: Cornell Lab of
Ornithology; Retrieved from the Birds of North America Online:
http://bna.birds.cornell.edu/bna/species/231
[137] Younge, A, and S Fowkes. 2003. The Cape action plan for the environ-
ment: overview of an ecoregional planning process. Biological Conservation
112:1528.
105