Habitat Quality in Conservation’S Neglected Geography

Home , Ecotope, Iolana, Iolana iolas, Nectarivore

MAPPING HABITAT QUALITY IN CONSERVATION’S NEGLECTED GEOGRAPHY

Ian Breckheimer

A thesis submitted to the faculty of the University of North Carolina at Chapel Hill in partial fulfillment of the requirements for the degree of Masters of Science in the Curriculum for the Environment and Ecology.

Chapel Hill 2012

Approved by:

Aaron Moody

Conghe Song

Peter White

ABSTRACT

IAN BRECKHEIMER: Mapping Habitat Quality in Conservation’s Neglected Geography (Under the direction of Aaron Moody)

This thesis describes conceptual and methodological work that aims to advance the science of modeling and mapping wildlife habitat in human-modified landscapes. First, I review how researchers have defined and measured the quality of wildlife habitat over the past four decades. I then demonstrate a new approach to quantifying habitat quality by modeling habitat for the federally endangered Red-cockaded Woodpecker (Picoides borealis,

RCW) across the Onslow Bight, a one million hectare region of North Carolina’s coastal plain. Next, I describe the development and operation of a GIS toolbox for ArcGIS 9.3, called “Connect”, designed to help conservation practitioners incorporate habitat connectivity considerations into land management and land-use planning. In two stakeholder-driven case studies, I use Connect to prioritize private land parcels for connectivity conservation in fragmented habitats around Fort Bragg, NC, and evaluate the effectiveness of a proposed corridor in promoting dispersal for RCW in the face of urban development.

iii

ACKNOWLEDGEMENTS

The author would like to thank John Lay, Alexa J. McKerrow, Cecil Frost, Matt

Simon, Jennifer Costanza, Jeff Walters, Anne Trainor, Will Fields, Nick Haddad, R. Todd

Jobe, Janet Pearson, Ryan Elting, Adam Terrando, Curtis Beleya, the staff of the US Fish and Wildlife Service, the staff of Marine Corps Base Camp Lejeune, the staff of the U.S.

Forest Service and the staff of US Army Fort Bragg for providing data that contributed to this effort.

Matt Simon, Jennifer Costanza, Anne Trainor, Austin Milt, Naomi Schwartz,

Amanda Chunco, Douglas J. Bruggeman, and Aaron Moody all provided invaluable support and feedback on the manuscript at various stages. Austin Milt provided programming support for the development of the software described in this thesis.

Much of this work was financially supported by grants from the Strategic

Environmental Research and Development Program (SI-1656 and RC-1471).

TABLE OF CONTENTS

LIST OF TABLES ...... vii

LIST OF FIGURES ...... viii

CHAPTER I—MEASURING HABITAT QUALITY IN THE NEGLECTED GEOGRAPHY ...... 2

The Neglected Geography of Conservation ...... 2

Habitat Quality - A Conceptual Overview ...... 5

Theory Meets Practice ...... 12

Defining Habitat Quality for the Neglected Geography ...... 19

Works Cited in Chapter I ...... 26

CHAPTER II—MODELING RED-COCKADED WOODPECKER (PICOIDES BOREALIS) HABITAT QUALITY AT HIGH RESOLUTION AND LARGE EXTENTS USING LIDAR ...... 34

Introduction ...... 30

Methods ...... 39

Study Area ...... 39

Field Data ...... 40

Remotely-sensed Data ...... 41

Habitat Standards ...... 43

Modeling prevalence ...... 44

Modeling fitness ...... 46

Results ...... 50

LiDAR data validation...... 50

Prevalence model ...... 51

Fitness model ...... 53

Discussion ...... 57

Works Cited In Chapter II ...... 62

CHAPTER III—THE CONNECT TOOLBOX: GIS TOOLS SUPPORTING LANDSCAPE CONNECTIVITY FOR WILDLIFE ...... 67

Introduction ...... 67

Quantifying connectivity ...... 68

Toolbox Description ...... 71

Overall Structure ...... 71

Create Connectivity Model ...... 72

Prioritize Landscape Features ...... 74

Generate Landscape Network ...... 75

Case Study Application ...... 77

Study Area: Spring Lake, NC ...... 77

Focal species ...... 79

Management Question 1: Where is the best connectivity bang for the buck? ...... 80

Management Question 2: Will a corridor be enough? ...... 86

Conclusion ...... 93

Works Cited in Chapter III ...... 96

APPENDIX A—EMPIRICAL PAPERS REVIEWED IN TABLE 1 ...... 100

LIST OF TABLES

1. Metrics of habitat quality used in empirical papers on ‘habitat quality’ or ‘patch quality’ published between 1983 and 2010 ...... 13

2. Vegetation structural criteria for the USFWS “recovery standard” and “managed stability standard” from USFWS (2003) along with corresponding conditions used to select forested plots on Camp Lejeune...... 44

3. Parameters incorporated into regressions of RCW group size ...... 48

4. Habitat types on conserved and non-conserved lands in the Onslow Bight derived from remotely sensed data ...... 51

5. Percentage of habitat types within 800m of active RCW cluster centers on Camp Lejeune (MCBCL), Croatan National Forest (CNF) and Holly Shelter Game Land (HSGL) in 2001, as determined by remote sensing data...... 53

6. Best performing regression models explaining RCW group size at 105 clusters over four years of observation, 1998-2001...... 54

7. Variables incorporated into optimal regression models of RCW groupsize, and multi-model weighted parameter estimates from all models within 2.3 ΔAICc of the optimum...... 55

8. Resistance values of different landcover types used to create dispersal models for each species ...... 84

vii

LIST OF FIGURES

1. A. The traditional geographic focus of wildlife conservation. B. Conservation's neglected geography ...... 3

2. The number of papers published between 1983 and 2010 with a title or keywords containing "Habitat Quality" or "Patch Quality" ...... 5

3. The ideal-free and ideal-despotic distributions. A.The ideal-free distribution. B. The ideal-despotic distribution ...... 6

4. Common habitat types and their positions with respect to the three "dimensions" of habitat quality ...... 23

5. The Onslow Bight region of North Carolina, our study area...... 40

6. Examples of LiDAR and field data used to build the RCW habitat model. (A) Heights of raw LiDAR returns in a 900 ha region of Camp Lejeune. (B) Forestry plots used to train the habitat model. (C) Fifty- meter wide transect of LiDAR data from the southeastern portion of Figure 6a ...... 42

7. Spline correlogram showing spatial autocorrelation of mean RCW group size from 1998 -2001 in the Onslow Bight ...... 47

8. LiDAR-derived forest structure metrics compared against field data metrics collected at Camp Lejeune for (A) tree height and (B) understory density ...... 50

9. Habitat model for the Red Cockaded Woodpecker in 2001 derived from LiDAR and other remotely-sensed data corresponding to the USFWS recovery standard ...... 52

10. Histogram of mean RCW group size in Croatan National Forest and Camp Lejeune from 1998-2001 ...... 54

11. Three strategies for modeling landscape connectivity. A. Graph-based approach based on a threshold of geographic distance. B. Least-cost path model. C. A circuit-based approach...... 70

12. The overall workflow of the Connect Toolbox ...... 72

13. Major lands managed for wildlife in the study area circa 2010 ...... 78

14. Focal species. Not to scale. Photos modified from originals by the US Fish and Wildlife Service ...... 80

viii

15. A. RGB composite of predicted relative frequency of dispersal habitat use for target species on private lands that have no permanent legal protection for wildlife. B. Tax value of privately owned parcels in the study area in 2008. C. Per-parcel mean predicted probability of urban development in 2100 from the SLEUTH-3r urban growth model. D. Connectivity conservation rank for parcels greater than 1ha ...... 86

16. Fitted exponential variogram model for the RCW resistance surface in urban areas ...... 89

17. RCW resistance surfaces for the baseline scenario (A), and one of five replicate resistance surfaces generated for the developent scenario (B) and the development and restoration scenario (C) ...... 89

18. A. Relative frequency of RCW dispersal habitat use for the baseline scenario. B. Average change from the baseline scenario with 2100 urban development. C. Average change from the baseline scenario with 2100 urban development and aggressive habitat restoration in the draft Carver’s Creek State Park proposed boundary ...... 92

19. Summary statistics representing changes in overall landscape connectivity for RCWs in the two landscape scenarios relative to the baseline ...... 93

ix CHAPTER I MEASURING HABITAT QUALITY IN THE NEGLECTED GEOGRAPHY The Neglected Geography of Conservation

Terrestrial and freshwater environments stand to lose up to one in five species over the next century as a consequence of habitat destruction, invasive species, and climate change (Millennium Ecosystem Assessment 2009). The magnitude of the problem and the limited resources available to address it has lead conservation biologists and policy-makers to perform a sort of "landscape triage" (Hobbs & Kristjanson 2003) where environments that are relatively unmodified by human activity are prioritized for conservation and management at the expense of lands and waters that are "degraded" by human use. The primary result of this triage is evident in the targeting of protected-area networks to areas with relatively little recent human impact. These ‘natural’ areas shelter a large proportion of species that are rare or at-risk (Rodrigues et al. 2004). This approach was pioneered in North America and

Western Europe (Sellars 1999), but has expanded to become the primary species conservation strategy worldwide over the past four decades (Brandon et al. 1998).

As governments and non-governmental organizations incorporate a greater proportion of "wild" or "natural" lands into protected areas, the focus of species- conservationists is beginning to shift to more highly human-modified landscapes. These regions are often a patchy mosaic of lands with different types of human use. Farms, working forests, residential communities are interspersed with patches of less-disturbed

habitat. In the United States, these lands are mostly in private ownership, and although they are unlikely to support the full suite of species that they did historically, management of these human-modified lands is key to the survival of many species of conservation concern, both in the US (Groves et al. 2000; Robles et al. 2008) and worldwide (Gallo et al. 2009).

Species conservation efforts on these lands are being approached with an increasing amount of resources, and an increasing diversity of policy tools, including economic subsidies, conservation easements (Cheever 2001), and safe-harbor agreements (Wilcove & Lee 2004).

Despite their importance to many species, our knowledge of the ecology of these human-modified private lands is generally poor relative to wild lands in the public trust

(Knight 1999). This is the case for two major reasons. First, it is difficult to obtain physical access to private lands that are owned and managed by many different private parties (Hilty

& Merenlender 2003). Second, researchers have historically eschewed lands under intense human use in their attempt to understand ecological patterns and processes in their

“pristine” state. Responding to this research vacuum, the last decade has seen the emergence of two fledgling fields, Countryside Biogeography (Daily 2000), and Urban Ecology (Adams

2005) that attempt to predict and explain the spatial pattern of species and ecological processes in human-dominated landscapes. Despite their rise in prominence, work in these two fields is poorly integrated into the existing body of knowledge in Conservation Biology more generally (Goddard et al. 2010). Both because of the importance of these regions to conservation and because of their historical oversight, I refer to privately owned lands under significant human use, both urban and rural, as being part of Conservation Biology's larger

“neglected geography” (Knight 1999, Figure 1).

Figure 1: A. The traditional geographic focus of wildlife conservation (unshaded areas). In this landscape, these are patches of fire-maintained longleaf pine forest in the vicinity of Ft. Bragg NC. B. Conservation's neglected geography (unshaded areas): Farms, pastures, residences, and managed forests adjacent to more “pristine” environments.

When managing threatened species in the neglected geography, the triage process still remains of central importance: how do we allocate scarce conservation resources to the protection and management of species in human-modified lands? Pervasive human influence makes triage strategies based on assessing the "naturalness" or "ecological integrity" of these lands potentially problematic. Few people would judge artillery impact craters or golf courses to be “natural”, for example, but they harbor important habitats for threatened amphibians

(Fields & Simon 2009) and birds (US Fish & Wildlife Service 2003), respectively. Performing effective triage in the neglected geography requires us to understand the importance of particular parts of the landscape to the maintenance of particular populations of imperiled organisms. In short, as a first step to prioritizing parts of the neglected geography for conservation or management, stakeholders must first identify which parts of the landscape represent “good quality habitat” for target species. But what is the most appropriate way of defining and measuring habitat quality?

The concept of habitat quality has a long history in wildlife ecology, but despite its deep roots and increasing use in the academic literature (Figure 2), researchers are not in agreement about how habitat quality should be defined and measured. Although Van Horne

(1983), and others (Johnson 2007) define habitat quality as the direct contribution of a habitat patch to regional population growth, other researchers (Hall et al. 1997) define habitat quality more broadly, as the ability of a habitat patch to promote population persistence. Moreover, as I will show, the metrics of habitat quality in widespread use have limited applicability in the neglected geography because they do not incorporate the influence of landscape configuration and dispersal on population persistence.

In this chapter I will review historical and contemporary approaches to quantifying habitat quality for animals in terrestrial and riverine environments, with particular attention to approaches that can be applied to the neglected geography. I will then outline a new conceptual framework for quantifying habitat quality that stresses the importance of quantifying the potential prevalence, fitness, and connectivity of wildlife populations across the landscape. In the following chapters I describe advances in estimating habitat quality using this general framework. First, I describe an approach to fusing information on prevalence, fitness, and connectivity to estimate habitat quality for the red-cockaded woodpecker (Picoides borealis) in the fragmented landscape of the Onslow Bight in coastal

North Carolina. I then describe the structure and application of new a software toolbox, called Connect designed to facilitate the measurement of landscape connectivity for wildlife.

I demonstrate the usefulness of these tools by answering a set of management-relevant questions regarding landscape connectivity for multiple threatened species in the vicinity of

Ft. Bragg in the North Carolina Sandhills.

Figure 2: The number of papers published between 1983 and 2010 with a title or keywords containing "Habitat Quality" or "Patch Quality". Source: ISI Web of Science.

Habitat Quality - A Conceptual Overview

The notion that different types of environments are more favorable for some species than others predates Darwin (von Humboldt 1805), but our modern conception of habitat quality owes a great debt to foundational habitat-selection theory developed by Fretwell and

Lucas (Fretwell & Lucas 1969; Fretwell 1972). Fretwell and Lucas reasoned that if (1) organisms selected habitat so as to maximize their individual fitness in a landscape of discrete patches of varying potential fitness, (2) fitness was a decreasing function of population density, and (3) organisms were free to move between habitat patches, they would arrange themselves into a spatial distribution in which organisms had equal actual fitness, but varying population densities between habitat patches. Fretwell and Lucas called this equilibrium condition the “ideal-free distribution” (Fretwell & Lucas 1969).

The distribution is “free” in that it assumes that resident individuals have no fitness advantage relative to colonizing individuals. Because this condition often does not hold for

territorial organisms or those with strong dominance hierarchies (Kaufmann 1983), the authors also examine an alternative case, which they call the “ideal-despotic distribution”

(Fretwell 1972). In this case, prospective colonizers perceive their potential fitness in a habitat as being lower than the actual fitness of the residents. This leads to an equilibrium condition where both population density and fitness are positively correlated across habitat patches. Patches of habitat with a higher potential fitness (absent the influence of conspecifics) should support higher population densities and should be inhabited by individuals with a higher fitness (Figure 3).

Figure 3: The ideal-free and ideal-despotic distributions in patches of varying quality. In both figures, habitat quality increases linearly with patch size A. In the ideal-free distribution, organisms arrange themselves among habitat patches so that individuals have equal fitness but exist at different densities in patches of varying quality. B. In the ideal-despotic distribution, social interactions lead to organisms settling in sub-optimal habitat with reduced fitness.

Fretwell and Lucas's predictions about the relationship between territoriality, fitness, and population density were influential. Regardless of whether a population was closer to the ideal-free or ideal-despotic distributions, population density could always be expected to be positively correlated with an organism's potential fitness absent the influence of conspecifics.

This quantity, which Fretwell and Lucas call “baseline suitability” represents the first rigorous definition of habitat quality to appear in the literature. It also provides a theoretical justification for using population density as a metric or indicator of habitat quality because it is much more easily estimated than potential fitness. Population density became a popular indicator of habitat quality for habitat assessments spearheaded by the US Fish and Wildlife

Service in the 1970s and early 1980s (Carey 1980; Maurer et al. 1980).

Despite its widespread adoption by managers, the use of population density as a measure of habitat quality was not universally accepted. In 1983, Beatrice Van Horne's influential paper “Density as a Misleading Indicator of Habitat Quality,” questioned the idea that the most favorable environments support relatively high population densities. She pointed to several empirical examples (Krebs 1971; States 1976) where social interactions produced a negative relationship between fitness and density, seemingly contrary to the predictions of Fretwell and Lucas. Van Horne proposed an alternative definition of habitat quality: the mean fitness of individuals per unit area of habitat. “High quality” environments, in Van Horne's conception, are the ones that make the greatest demographic contribution to an organism's future population. This definition emphasizes the importance of quantifying not just population density, but also the density of breeding individuals as well as their fecundity and reproductive success. Van Horne's paper discouraged a generation of researchers from using population density as a metric of habitat quality (Perot & Villard

2009). As of July 2011, Van Horne's paper had been cited more than 1500 times.

Van Horne's definition of habitat quality focuses on patch-specific demographic rates, largely ignoring the landscape context of those patches. Although several authors

(Tamarin 1978; Waser 1985) noted that some patches of habitat were net exporters of individuals to the regional population (population “sources”), while others could be

considered net importers (population “sinks”), population sinks were thought to be rare and transient in nature because organisms should evolve to avoid relatively low-fitness environments (Svärdson 1949) and populations in sink habitats should quickly go locally extinct. It wasn't until 1988 that L. Ron Pulliam demonstrated that population sinks could be ecologically and evolutionarily stable under conditions similar to Fretwell and Lucas's “ideal- despotic distribution” if the number of immigrants to a patch was great enough to counter the annual population deficits caused by low fecundity or high mortality (Pulliam &

Danielson 1991; Pulliam 1988). In Pulliam's view, Van Horne's definition of habitat quality should be amended to track the fate of immigrants and emigrants from a habitat patch.

High-fecundity, low-mortality environments could contribute little to the regional population if “excess” individuals cannot successfully emigrate to other habitats. Put another way, to determine the quality of a habitat patch, one must understand both demographic processes occurring within the patch and dispersal processes that connect patches together in the larger landscape. This perspective is necessary if we wish to map habitat quality in the neglected geography, where habitat fragmentation becomes a threat to population persistence (Fahrig 2002).

The landscape perspective on habitat quality became more prominent in the mid-

1990s with the development of modern metapopulation models and spatially explicit population models. Metapopulation models use discrete habitat patches as their fundamental unit of organization. Patches can either be occupied or unoccupied for a particular species, and patches are subject to stochastic extinction and colonization events depending on patch attributes. In the first modern metapopulation models (such as Hanski's 1994 “incidence function model”), colonization and extinction probabilities were simple functions of patch area and geographic isolation. More recently, researchers have integrated other patch

attributes contributing to habitat quality, such as food resources (Fleishman et al. 2002;

Ozgul et al. 2006) and non-random dispersal (Rabasa et al. 2007). A key advantage of these models is that their parameters can be readily estimated from presence/absence surveys using likelihood-based (Moilanen 1999) or Bayesian (O’Hara et al. 2002; ter Braak & Etienne

2003) methods. These techniques allowed researchers to test hypotheses regarding how changes in patch area, isolation, and factors influence population persistence in both real

(Kallimanis et al. 2005) and hypothetical (Brito & Fernandez 2002; Vuilleumier et al. 2007) landscapes. One of the most important general conclusions to emerge from these metapopulation models is that metapopulation extinction risk rises precipitously at certain thresholds of habitat fragmentation (Bascompte & Sole 1996; Fahrig 2002; Ovaskainen &

Hanski 2003).

Metapopulation models also allow us to quantify the importance of particular habitat patches for overall population persistence. For example, Ovaskainen and Hanski

(Ovaskainen & Hanski 2003) define a metric of patch quality that represents the proportion of all colonization events contributed by a particular patch when the metapopulation is in equilibrium, e.g. overall rates of patch colonization are equal to rates of extinction. Similar metrics can be computed for other, non-equilibrium situations by removing patches from a model and tracking changes in the size of the metapopulation or its estimated time to extinction. These metrics represent an important advance in measuring habitat quality, because they integrate patch-level and landscape-level influences on the ability of a patch to promote population persistence.

In contrast to the relative simplicity of most metapopulation models, spatially- explicit population models (SEPMs) simulate the births, deaths, and dispersals of large numbers of individual organisms in virtual geographic space. The first of these models were

developed in the mid-1980s (Urban & Shugart 1984). Like metapopulation models, spatially explicit population models have been used to address a number of research questions about the relationship between demographic rates, habitat configuration, dispersal, and population persistence, both in general terms (Fahrig 1997) and with respect to particular species of conservation concern (Letcher et al. 1998; Heinrichs et al. 2010). Unlike metapopulation models, however, SEPMs can be applied to situations where habitat patches are poorly defined, or when local extinction and colonization events are rare. Habitat characteristics are often incorporated into spatially explicit population models by altering demographic rates within habitat patches or by adjusting the density dependence of particular demographic parameters (Minor et al. 2008). Similar to metapopulation models, SEPMs can examine the importance of habitat patches by running model experiments with alternative landscapes in which patches have been added or removed (Heinrichs et al. 2010). SEPMs can incorporate detailed species-specific information on demography, dispersal, and habitat configuration, but their complexity makes parameterization, interpretation, and validation of these models difficult (Letcher et al. 1998; Grimm et al. 2005).

The complexity and high data requirements of SEPMs has caused several researchers to examine whether simpler methods could be found to approximate the estimates of habitat quality derived from these models. For example, Minor and Urban (Minor & Urban 2007) examined the ability of graph-theoretical connectivity metrics (Urban & Keitt 2001) to serve as a suitable proxy, finding that several simple connectivity indices were correlated with a composite metric of patch “goodness” derived from a SEPM parameterized for the Wood

Thrush (Hylocichla mustelina). Minor's study is part of a growing body of literature that aims to quantify landscape connectivity, or the degree to which a landscape facilitates the movement of organisms between resource patches (Taylor et al. 1993). The importance of

landscape connectivity to the persistence of species has been demonstrated in a wide variety of environments; however the explosion of literature on landscape connectivity has mostly neglected the influence of within-patch demography on dispersal, or the interactions between habitat area, dispersal, and population persistence highlighted by the metapopulation literature.

The theoretical developments of Fretwell, Van Horne, Pulliam, and Hanski have substantially clarified our picture of what conditions are important for maintaining population persistence of a wide variety of organisms in many different types of landscapes.

In concert, the ways that researchers in the theoretical literature define habitat quality, and the primary means that they suggest for measuring it, have shifted substantially in the past four decades. While early developments emphasized the importance of measuring population density as an indicator of potential fitness, later workers found that population density alone was a poor indicator of habitat quality, and instead stressed the importance of jointly estimating population density and fitness within habitat patches. The source-sink concept put habitat quality in a landscape context, and in the 1990s, the development of powerful metapopulation and spatially explicit population models provided us with tools to estimate the importance of individual patches of habitat to overall population persistence.

Advances in the past decade have brought us new ways to estimate the degree to which the landscape around habitat patches influences dispersal between them. These conceptual and methodological advances make it possible, for the first time, to estimate the importance of individual parts of the landscape for the persistence of regional populations. I would argue that this represents the most appropriate definition of habitat quality if we wish to allocate scarce resources to effectively conserve threatened species in the neglected geography.

Theory Meets Practice

The theoretical literature surrounding habitat quality makes it clear that the prevalence and fitness of organisms in habitat patches and the connectivity between those patches all contribute to their ability to support population persistence, but how have these concepts been put into practice? To answer this question I reviewed 194 empirical studies published between 1983 (the year of Van Horne’s seminal paper) and 2010 with the phrases

“habitat quality” or “patch quality” in the title (Appendix A). I restricted my search to papers that had been cited at least twice in ISI Web of Science, and to papers dealing with organisms in terrestrial or lotic freshwater environments. Additionally, I restricted my search to papers that treat habitat quality as taxon-specific. The journals with the greatest representation in my review were Oecologia, Ecology, Biological Conservation, Journal of Animal

Ecology, Conservation Biology, and Oikos. The taxonomic scope of my review is broad, but reflects the overall bias of the literature towards vertebrates, especially birds and small mammals (Clark 2002). Despite the increasing breadth and sophistication of habitat quality theory, most of the empirical literature treats habitat quality informally, and often the term is not explicitly defined. The empirical studies fall into two general categories: those treating habitat quality as an independent variable (or an explanatory variable, Dodge et al. 2006), and those treating habitat quality as a dependent variable (or a response variable, Dodge et al.

2006).

Most commonly (123 out of 194 studies, Table 1), researchers use habitat quality as an independent variable. In experimental studies of this type, manipulating habitat quality often means manipulating the abundance (Schilman & Roces 2003), availability (Boivin et al.

2004), or nutritional value (Dixon & Kundu 1998; Stanko-Mishic et al. 1999) of food resources. For example, Armstrong and Stamp (2003) manipulate habitat quality for parasitic wasps (Polistes dominulus) by varying the ratio of palatable to unpalatable prey items near wasp colonies in order to understand their foraging strategies. Kohlmann and Risenhoover (1996) perform a similar supplementation study with Bobwhite Quail (Colinus virginianus). In addition to manipulative experiments, many observational studies use habitat quality as an independent variable as well. In these studies, “habitat quality” is often used as a general term to describe differences in the habitat structure (Prenda et al. 2001), vegetative composition (Pettorelli et al. 2005), or food resources (Winnie et al. 2008) between study sites. The frequency that habitat quality is a dependent variable (a quantity to be estimated) varies considerably by taxon. Studies focusing on butterflies (11 of 16 studies) or other insects (19 of 21 studies) much more frequently considered habitat quality as an independent variable than studies focused on birds (39 of 75), or fish (4 of 11). Because my general aim is to improve methods for estimating habitat quality, I will review studies that use “habitat quality” as a dependent variable in more detail.

Empirical studies that consider habitat quality as a dependent variable use a variety of different habitat quality indicators. Using the health or condition of individuals in a particular habitat as an indicator is particularly common across a variety of taxa. For example, Senar et al. (2002) assess habitat quality for the Citril Finch (Serinus citrinella) by comparing body mass and a morphometric fat index of adults between two study sites in Spain. Sinsch et al. (2007) compare age at maturity in Greed Toads (Bufo viridis) in a set of sites differing in the amount

of surrounding human land use in northern Germany. Researchers studying birds have also used plumage characteristics to assess habitat quality (Ferns & Hinsley 2008), arguing that high quality habitats should provide more precursors to feather pigments. Blood serum chemistry has been used to predict habitat quality in Black Bears (Ursus americanus, Hellgren et al. 1993), and the blood serum concentration of the stress hormone corticosterone has been successfully used to predict habitat quality in overwintering songbirds (Marra &

Holberton 1998) as well as amphibians (Homan et al. 2003). Only a few of these studies directly relate measures of the condition of organisms to demographic rates. Studies of songbirds provide the best example. Johnston et al. (2006) demonstrate that the decline in winter body mass of American Redstarts (Setophaga ruticilla) was highly correlated with rates of survival across a variety of habitats in Jamaica.

Within studies that use individual characteristics as measures of habitat quality, researchers often focus on components of fitness such as reproductive performance or survival. In many bird studies, clutch size (Powell & Powell 1986), fledgling number (Beyer et al. 1996), nesting success (Wightman & Germaine 2006; Frey et al. 2008), or laying date

(Wilkin et al. 2007) is compared between different sites or habitat types. Researchers sometimes use multiple measures of reproductive performance to assess habitat quality.

Weinberg and Roth (1998), for example, use the number of fledglings per female, the nesting success rate, and the number of fledglings per unit area as indicators of habitat quality for the Wood Thrush (Hylocichla mustelina) nesting in small forest fragments and a continuous forest in Delaware. Rather than focusing on reproductive performance, other studies examine the other major component of fitness, survival. Paradis and Croset (1995) for example, compare mortality rates of Mediterranean Pine Vole (Microtus duodecimcostatus) between meadows and apple orchards in southern France, while Weiss et al. (1988) assess

the relationship between topographic position and habitat quality for Euphydryas butterflies using larval mortality rates.

Despite Van Horne's warning that the most favorable environments do not always harbor the highest densities of organisms, many researchers continue to use population density (Lavers & HainesYoung 1996; Ellis 2003), occupancy (Webb et al. 2007; Anadon et al. 2007) or other measures of prevalence as the primary indicators of habitat quality. For example, Perot and Villard (2009) found a strong positive relationship between territory density and productivity for Ovenbirds (Seiurus aurocapilla) nesting in New Brunswick,

Canada. On that basis they conclude that territory density is the most appropriate indicator of habitat quality in their study system. The case for using population density as an indicator of habitat quality was bolstered by a meta-analysis by Bock & Jones (2004) which found that areas with higher population density also had higher reproductive output in the majority of bird populations monitored in North America and Europe. The authors noted, however, that negative relationships between population density and reproductive output were more common in human-modified environments. This result suggests that using population density alone as an indicator of habitat quality may be inappropriate for populations of threatened organisms in human-modified landscapes that characterize the neglected geography. I will expand on this argument in the next section.

Interestingly, although a wide variety of studies assess habitat quality by examining population density, or individual fitness components, relatively few studies (10 out of 194) use metrics of habitat quality similar to the one advocated by Van Horne and others (mean fitness per unit area, Johnson 2007). This is perhaps related to the difficulty of arriving at simultaneous measures of fecundity, mortality and population density in cryptic organisms or those with low site fidelity. Again, studies of birds provide the best examples. The landmark

study of habitat quality in Northern Spotted Owls (Strix occidentalis caurina) by Franklin et al.

(2000) tests hypotheses about the relationships between climate, habitat characteristics, and habitat quality by estimating relationships between climate and habitat predictors and individual fitness components. The authors then combine these components to evaluate the potential population growth rate of owls at each study site. Pidgeon et al. (2006) uses a somewhat similar approach to evaluate habitat quality indicators for the Black-throated

Sparrow (Amphispiza bilineata) in the northern Chihuahuan Desert. They compare estimates of fitness components in different habitats to results from a simple population model that indicates the degree to which each habitat serves as a population source. Pidgeon et al. found that information about nesting success and nesting density were required to approximate habitat quality rankings derived from the population model, and using individual fitness components such as fecundity to estimate habitat quality was insufficient to assess the importance of particular habitats to maintaining the regional population.

Where population density or demographic information is not available, researchers have proposed a variety of other habitat quality indicators. Individual turnover rates within different habitats, for example, have been proposed as a habitat quality indicator under the assumption that organisms in close-to-optimal habitat will have higher site-fidelity than individuals in sub-optimal habitat (Winker et al. 1995; Belanger & Rodriguez 2002). Lyons

(2005) also suggests that individual foraging effort should be inversely related to habitat quality if near-optimal habitats provide a greater abundance or availability of food resources.

Booker et al. (2004) estimate habitat quality for juvenile trout and salmon in a UK stream by comparing a bioenergetic model of net foraging energy intake to observed counts of foraging and resting fish.

When testing hypotheses regarding the influence of habitat attributes on habitat quality, researchers often focus on the vegetative structure or composition within habitat patches. Care must be taken, however, to ensure that the chosen predictors are measured at the spatial grain and extent that is most relevant to the target organisms (Johnson 2007).

These optimal scales can be difficult to define a-priori (Chalfoun & Martin 2007). Moreover, it is only in the past decade that wildlife biologists have begun to consider the influence of the larger landscape on habitat quality. In my survey of the literature, less than half of the papers that treat habitat quality as a dependent variable (27 of 71) incorporate information on the spatial configuration of habitat as potential predictors. Of these, 20 were published after the year 2000.

It appears that the gap between habitat quality theory and the practice of assessing it is wide indeed. Although researchers most commonly use indicators of habitat quality relating to the prevalence or fitness of organisms, few empirical studies provide enough information to assess the ability of particular habitat patches to support population persistence (“habitat quality” as defined by Hall et al. 1997). Instead, most of these studies seek general “rules of thumb” regarding habitat conditions that are favorable to a species in order to guide habitat management. For example, Hunt (1996) demonstrates that American

Redstarts (Setophaga ruticilla) breeding in early successional forests have higher population densities, higher rates of mating success, and smaller territories than those breeding in mature hardwood forests. On this basis, he suggests that increasing the amount of early successional habitat will benefit this species. These rules of thumb based on metrics of population performance are useful for guiding habitat management, but they are often not sufficient for assessing habitat quality in the neglected geography for reasons I explain in the next section.

Defining Habitat Quality for the Neglected Geography

The neglected geography, which harbors habitat for a large proportion of the world's imperiled species, is a particularly challenging place to define and measure habitat quality.

Many imperiled wildlife populations in the neglected geography share particular features that make the traditional ways of measuring habitat quality problematic, as I will discuss below.

Moreover, the fractured ownership regimes and dynamics of shifting land-use that characterize the neglected geography make collection of data to support any kind of assessment of habitat quality a challenging task. In this section I will discuss the major features of the neglected geography that pose problems for traditional ways of defining and measuring habitat quality. I will then outline a new conceptual framework for measuring habitat quality in the neglected geography that I believe strikes an appropriate balance between practicality and rigor. Because of the special challenges posed by ecological traps, habitat fragmentation, and rapid landscape change, I believe that assessments of the ability of habitat to promote population growth or persistence in the neglected geography should generally include indicators related to the prevalence, fitness, and connectivity of populations in habitat patches. I contend that these three attributes can usefully be thought of as the three “dimensions” of habitat quality.

As we saw in the previous section, habitat quality has typically been assessed by relating habitat and landscape attributes either to measures of an organism's prevalence or fitness. Relatively few studies evaluate both prevalence and fitness simultaneously (but see

Vierling 1999; Knutson et al. 2006; Kroll & Haufler 2007). Although these simultaneous measurements would be redundant for assessing habitat quality under the ideal-free or ideal- despotic distributions, wildlife populations in the neglected geography are likely to violate the assumptions of both ideal models. First, both models assume that habitat selection is

adaptive: organisms will arrange themselves in space so as to maximize their individual fitness. As argued by Robertson and Hutto (2006), it is especially perilous to assume that habitat selection is adaptive in human-modified landscapes because they often differ substantially from the environments in which these organisms evolved. Cues that organisms use to identify high-quality habitat may be misleading, leading them into “ecological traps”

(sensu Dwernychuk & Boag 1972). The importance of ecological traps in the neglected geography is supported by Bock and Jones' (2004) finding that bird population density was more likely to be negatively correlated with reproductive output in human-modified landscapes than in landscapes with little human activity. The possibility that the social structure of organisms causes the accumulation of subdominant individuals in marginal habitat (Van Horne 1983), as well as the possibility that the landscape contains ecological traps, underlines the importance of estimating both the prevalence and fitness of organisms to assess habitat quality in the neglected geography.

Further, using only prevalence and fitness to assess habitat quality implicitly assumes that organisms have freedom of movement between all patches of habitat. Wildlife populations in the neglected geography, particularly those with limited dispersal ability, may not readily disperse between breeding habitats or other resource patches in human-modified landscapes. A failure to account for the configuration of the surrounding landscape can bias estimates of the relationships between other habitat features and fitness or prevalence

(Mortelliti et al. 2010). Moreover, isolation can also interact with prevalence and fitness to produce widely divergent outcomes. For example, isolation of small populations can lead to inbreeding depression (Wright et al. 2008), but isolation of relatively large populations can promote adaptation to the local environment (Verhoeven et al. 2011). For organisms with patchy populations, patch connectivity should often be positively correlated with prevalence

because connectivity serves to buffer demographic fluctuations within habitat patches and allows patches to be recolonized following local extinction (Proctor et al. 2005; Wright et al.

2008). Indeed numerous metapopulation studies have demonstrated a positive link between patch connectivity and occupancy in patchy landscapes (Moilanen & Hanski 1998;

Fleishman et al. 2002; Voegeli et al. 2010). Hodgson et al. (2009), however, demonstrate that ecological succession and rapidly shifting patterns of land-use can serve to weaken the relationship between connectivity and occupancy. Nonetheless, over management-relevant timescales, parts of the landscape that promote dispersal (i.e. dispersal corridors; Haddad et al. 2003) are critical to maintaining viable populations in fragmented landscapes. These dispersal habitats often include areas where organisms breed, but also include areas where organisms are found only during dispersal. If these dispersal habitats are critical to population persistence, I argue that they should also be considered “high quality” habitat.

The special challenges posed by the neglected geography make defining habitat quality in strict demographic terms (as advocated by Johnson 2007) problematic. Those parts of the neglected geography that are most important for maintaining regional population persistence are not necessarily those with the most favorable demographic rates. In the neglected geography, I believe it is useful to define habitat quality broadly, as the ability of habitat patches to support regionally persistent populations. Metapopulation models and spatially explicit population models give us two different frameworks to integrate information on prevalence, fitness, and connectivity to evaluate the importance of particular habitat patches for population persistence quantitatively. For example, Heinrichs et al. (2010) use an individual-based, spatially explicit population model to rank patches of Ord’s

Kangaroo Rat (Dipodomys ordii) breeding habitat according to their contribution to population persistence in southern Alberta. Similarly, Horne et al. (2011) use a metapopulation model to

evaluate the influence of the partial destruction of individual habitat patches on overall population persistence in Hooded Warblers (Dendroica chrysoparia) in the vicinity of Ft. Hood,

Texas. Both groups of researchers found that the importance of a habitat patch for population persistence could not be reduced to a simple function of prevalence, fitness, or connectivity alone.

Metapopulation models and spatially explicit population models are powerful tools for evaluating threats to, and management strategies for, imperiled organisms. Nearly two decades after the development of these models, however, their use is still primarily restricted to well-studied species. Very little life-history data exists for the vast majority of imperiled organisms (Millennium Ecosystem Assessment 2009). If we don't have enough information to build a metapopulation model or a spatially explicit population model for species at risk in the neglected geography, how should we define and evaluate habitat quality for those species? I argue that, in the neglected geography, in the absence of a spatially explicit model that links prevalence, fitness, and connectivity to regional population growth or persistence, it is useful to define habitat quality as a set of three attributes of habitat patches which I call the three dimensions of habitat quality: (1) the ability of a patch to support occupancy or prevalence, (2) the ability of a patch to support high fitness (3) the ability of a habitat patch to promote landscape connectivity through dispersal to other habitat patches (Figure 4).

Unlike approaches that focus only on within-patch demography, defining habitat quality with respect to prevalence, fitness, and connectivity captures the key features of habitat patches that influence their ability to promote population persistence in human-dominated landscapes. We cannot assume that parts of the landscape where prevalence is high are the most suitable for a species (Heinrichs et al. 2010). Conversely, we cannot assume that areas with high individual fitness make the greatest contribution to overall population growth.

Finally, even non-breeding habitats with low prevalence can be critical to regional population persistence if they promote dispersal.

Figure 4: Common habitat types and their positions with respect to the three "dimensions" of habitat quality.

Defining habitat quality this way forces us to answer three related questions about the characteristics of habitat itself. First, we must ask: Which features of the environment promote the ability of habitat to support a high prevalence of organisms? The most appropriate metric of prevalence depends on the biology of organisms and on the resources available to measure them. For conspicuous species or those with high site-fidelity, population density can often be estimated directly, although with considerable research effort. For mobile organisms, other metrics of prevalence, such as probability of occupancy

(MacKenzie 2006) may be more appropriate. For still others, we will be forced to infer prevalence from geolocated museum records or other presence-only data (Elith et al. 2011).

With respect to fitness, we must ask: Given that an organism is present and breeding, what features of the environment promote high fecundity, low mortality, or both? As we saw in the previous section, because of the difficulty of collecting demographic data, most assessments of habitat quality using fitness-related metrics use indirect indicators such as body condition or fitness components such as adult mortality in place of fitness itself. The most appropriate fitness-related metric, again, depends on both the biology of organisms and the resources that are available to support study. With respect to connectivity we must ask: How do features of the environment promote or prevent the dispersal of organisms between habitat patches? In other words, what is the “resistance” of landscape features to dispersal? Unlike prevalence or fitness, resistance must be inferred indirectly from mark- recapture (Ricketts 2001) or movement data (Trainor et al. in preparation).

Considering all three dimensions of habitat quality may not be appropriate in some circumstances, even in the neglected geography. For example, connectivity may not be an important issue for organisms that disperse readily in human-dominated landscapes, or for species with habitat that is spatially contiguous (Hanski et al. 2004). For these organisms, defining habitat quality in demographic terms (such as recommended by Johnson 2007) makes more sense. Considering only fitness may not be appropriate when evaluating habitat quality for migratory species in non-breeding habitat, such as overwintering songbirds. In these cases, mortality may be a better metric for evaluating habitat quality (Johnson 2006).

Finally, it may be redundant to evaluate prevalence, fitness, and connectivity separately if they have a consistent, strong, positive relationship. Although I have argued that these are likely to be decoupled in the neglected geography, prevalence, fitness, and connectivity may be strongly correlated for many species in large blocks of continuous habitat (Bock and

Jones 2004).

Defining habitat quality with respect to prevalence, fitness, and connectivity provides managers with information that they can use to make decisions about where to concentrate different types of management effort in the neglected geography. To return to a previous example, Hunt (1996) argues that increasing the amount of early-successional habitat will benefit American Redstarts because birds that breed in those areas have greater demographic performance. This “rule of thumb” provides useful guidance for managing habitat for this species, however defining and measuring habitat quality with respect to prevalence, fitness, and connectivity would provide information that could allow the spatial targeting of different types of management action. Areas with high prevalence and fitness are appropriate targets for conservation, while areas with high occupancy but relatively low fitness might be appropriate targets for restoration. Finally, areas with high connectivity, but low potential fitness and prevalence could be managed to provide conditions that promote dispersal. The relationships between the three “dimensions” of habitat quality may be complex, but by measuring all three in concert, we may be able to manage habitat for wildlife more effectively and arrive at a better understanding of the conditions required for population persistence in human-modified landscapes.

Works Cited in Chapter I

Adams, L.W., 2005. Urban wildlife ecology and conservation: A brief history of the discipline. Urban Ecosystems, 8, pp.139-156.

Anadon, J. D., Gimenez, A., Martinez, M., Palazon, J. A., & Esteve, M. A. 2007. Assessing changes in habitat quality due to land use changes in the spur-thighed tortoise Testudo graeca using hierarchical predictive habitat models. Diversity and Distributions, 13(3), pp.324-331.

Armstrong, T. & Stamp, N., 2003. Effects of prey quantity on predatory wasps (Polistes dominulus) when patch quality differs. Behavioral Ecology and Sociobiology, 54(3), pp.310- 319.

Bascompte, J. & Sole, R.V., 1996. Habitat fragmentation and extinction thresholds in spatially explicit models. Journal of Animal Ecology, pp.465–473.

Belanger, G. & Rodriguez, M., 2002. Local movement as a measure of habitat quality in stream salmonids. Environmental Biology of Fishes, 64(1-3), pp.155-164.

Beyer, D., Costa, R., Hooper, R., & Hess, C., 1996. Habitat quality and reproduction of red- cockaded woodpecker groups in Florida. Journal of Wildlife Management, 60(4), pp.826- 835.

Bock, C.E. & Jones, Z.F., 2004. Avian habitat evaluation: should counting birds count? Frontiers in Ecology and the Environment, 2, pp.403-410.

Boivin, G., Fauvergue, X. & Wajnberg, E., 2004. Optimal patch residence time in egg parasitoids: innate versus learned estimate of patch quality. Oecologia, 138(4), pp.640- 647.

Booker, D., Dunbar, M. & Ibbotson, A., 2004. Predicting juvenile salmonid drift-feeding habitat quality using a three-dimensional hydraulic-bioenergetic model. Ecological Modelling, 177(1-2), pp.157-177. ter Braak, C.J.F. & Etienne, R.S., 2003. Improved Bayesian analysis of metapopulation data with an application to a tree frog metapopulation. Ecology, 84(1), pp.231–241.

Brandon, K., Redford, K.H. & Sanderson, S.E., 1998. Parks in peril: people, politics, and protected areas, Cambridge Univ Press.

Brito, D. & Fernandez, F.A.S., 2002. Patch relative importance to metapopulation viability: the Neotropical marsupial Micoureus demerarae as a case study. Animal Conservation, 5(1), pp.45–51.

Carey, A.B., 1980. Multivariate analysis of niche, habitat, and ecotope. The use of multivariate statistics in studies of wildlife habitat. US Dep. Agric., For. Serv. Gen. Tech. Rep. RM-87, pp.104–113.

Chalfoun, A.D. & Martin, T.E., 2007. Assessments of habitat preferences and quality depend on spatial scale and metrics of fitness. Journal of Applied Ecology, 44(5), pp.983-992.

Cheever, F., 2001. Property Rights and the Maintenance of Wildlife Habitat: The Case for Conservation Land Transactions. Idaho L. Rev., 38, p.431.

Clark, J. A., & May, R. M. 2002. Taxonomic bias in conservation research. Science, 297(5579), pp.191-192.

Daily, G.C., 2000. Countryside biogeography and the provision of ecosystem services. In Nature and human society: the quest for a sustainable world: proceedings of the 1997 Forum on Biodiversity.

Dixon, A. & Kundu, R., 1998. Resource tracking in aphids: programmed reproductive strategies anticipate seasonal trends in habitat quality. Oecologia, 114(1), pp.73-78.

Dodge, Y., Cox, D. & Commenges, D., 2006. The Oxford dictionary of statistical terms, Oxford University Press, USA.

Dwernychuk, L. & Boag, D., 1972. Ducks nesting in association with gulls-an ecological trap? Canadian Journal of Zoology, 50(5), pp.559–563.

Elith, J., Phillips, S. J., Hastie, T., Dudík, M., Chee, Y. E., & Yates, C. J., 2011. A statistical explanation of MaxEnt for ecologists. Diversity and Distributions, 17(1), pp.43-57.

Ellis, S., 2003. Habitat quality and management for the northern brown argus butterfly Aricia artaxerxes (Lepidoptera: Lycaenidae) in North East England. Biological Conservation, 113(2), pp.285-294.

Fahrig, L., 2002. Effect of Habitat Fragmentation on the Extinction Threshold: A Synthesis. Ecological Applications, 12(2), pp.346–353.

Fahrig, L., 1997. Relative effects of habitat loss and fragmentation on population extinction. The Journal of wildlife management, pp.603–610.

Ferns, P.N. & Hinsley, S.A., 2008. Carotenoid plumage hue and chroma signal different aspects of individual and habitat quality in tits. Ibis, 150(1), pp.152-159.

Fields, W. & Simon, M.C., 2009. COS 28-9: Predicting the locations of breeding pools to assess landscape connectivity for a rare amphibian. In Proceedings of the The 94th ESA Annual Meeting.

Fleishman, E., Ray, C., Sjogren-Gulve, P., Boggs, C. & Murphy, D., 2002. Assessing the roles of patch quality, area, and isolation in predicting metapopulation dynamics. Conservation Biology, 16(3), pp.706-716.

Franklin, A. B., Anderson, D. R., Gutiérrez, R., & Burnham, K. P., 2000. Climate, habitat quality, and fitness in northern spotted owl populations in northwestern California. Ecological Monographs, 70(4), pp.539–590.

Fretwell, S.D., 1972. Populations in a seasonal environment, Princeton Univ Pr.

Fretwell, Stephen Dewitt & Lucas, H.L., 1969. On territorial behavior and other factors influencing habitat distribution in birds. Acta Biotheoretica, 19, pp.16-36.

Frey, C.M., Jensen, W.E. & With, K.A., 2008. Topographic patterns of nest placement and habitat quality for grassland birds in tallgrass prairie. American Midland Naturalist, 160(1), pp.220-234.

Gallo, J. A., Pasquini, L., Reyers, B., & Cowling, R. M., 2009. The role of private conservation areas in biodiversity representation and target achievement within the Little Karoo region, South Africa. Biological Conservation, 142(2), pp.446–454.

Goddard, M.A., Dougill, A.J. & Benton, T.G., 2010. Scaling up from gardens: biodiversity conservation in urban environments. Trends in Ecology & Evolution, 25(2), pp.90-98.

Grimm, V., Revilla, E., Berger, U., Jeltsch, F., Mooij, W. M., Railsback, S. F., Thulke, H. H., et al., 2005. Pattern-oriented modeling of agent-based complex systems: lessons from ecology. Science, 310(5750), pp.987–991

Groves, C. R., Kutner, L. S., Stoms, D. M., Murray, M. P., Scott, J. M., Schafale, M., Weakley, A. S., Pressey, R.L., 2000. Owning up to our responsibilities. Precious heritage: the status of biodiversity in the United States, pp.275–300.

Haddad, N. M., Bowne, D. R., Cunningham, A., Danielson, B. J., Levey, D. J., Sargent, S., & Spira, T., 2003. Corridor use by diverse taxa. Ecology, 84(3), pp.609–615.

Hall, L.S., Krausman, P.R. & Morrison, M.L., 1997. The Habitat Concept and a Plea for Standard Terminology. Wildlife Society Bulletin, 25(1), pp.173-182.

Hanski, I., 1994. A practical model of metapopulation dynamics. Journal of Animal Ecology, 63(1), pp.151–162.

Hanski, I., Gaggiotti, O.E. 2004. Ecology, genetics, and evolution of metapopulations, Elsevier Amsterdam, The Netherlands.

Heinrichs, J. A., Bender, D. J., Gummer, D. L., & Schumaker, N. H., 2010. Assessing critical habitat: Evaluating the relative contribution of habitats to population persistence. Biological Conservation, 143(9), pp.2229-2237.

Hellgren, E., Rogers, L. & Seal, U., 1993. Serum Chemistry And Hematology of Black Bears - Physiological Indexes of Habitat Quality or Seasonal Patterns. Journal Of Mammalogy, 74(2), Pp.304-315.

Hilty, J. & Merenlender, A.M., 2003. Studying biodiversity on private lands. Conservation Biology, 17(1), pp.132–137.

Hobbs, R.J. & Kristjanson, L.J., 2003. Triage: How do we prioritize health care for landscapes? Ecological Management & Restoration, 4(s1), pp.39-45.

Hodgson, J.A., Moilanen, Atte & Thomas, C.D., 2009. Metapopulation responses to patch connectivity and quality are masked by successional habitat dynamics. Ecology, 90, pp.1608-1619.

Homan, R., Regosin, J., Rodrigues, D., Reed, J., Windmiller, B., & Romero, L., 2003. Impacts of varying habitat quality on the physiological stress of spotted salamanders (Ambystoma maculatum). Animal Conservation, 6(1), pp.11-18.

Horne, J.S., Strickler, K.M. & Alldredge, M., 2011. Quantifying the importance of patch- specific changes in habitat to metapopulation viability of an endangered songbird. Ecological Applications, 21, pp.2478-2486. von Humboldt, A., 1805. Essay on the Geography of Plants. In Foundations of biogeography: classic papers with commentaries. National Center for Ecological Analysis and Synthesis.

Hunt, P., 1996. Habitat selection by American redstarts along a successional gradient in northern hardwoods forest: Evaluation of habitat quality. Auk, 113(4), pp.875-888.

Johnson, M.D., 2007. Measuring habitat quality: A review. Condor, 109(3), pp.489-504.

Johnson, M. D., Sherry, T. W., Holmes, R. T., & Marra, P. P., 2006. Assessing habitat quality for a migratory songbird wintering in natural and agricultural habitats. Conservation Biology, 20(5), pp.1433-1444.

Kallimanis, a. S., Kunin, W. E., Halley, J. M., & Sgardelis, S. P., 2005. Metapopulation extinction risk under Spatially autocorrelated disturbance. Conservation Biology, 19(2), pp.534-546.

Kaufmann, J.H., 1983. On the definitions and functions of dominance and territoriality. Biological Reviews, 58(1), pp.1–20.

Knight, R.L., 1999. Private lands: the neglected geography. Conservation Biology, 13(2), pp.223– 224.

Knutson, M., Powell, L., Hines, R., Friberg, M., & Niemi, G., 2006. An assessment of bird habitat quality using population growth rates. Condor, 108(2), pp.301-314.

Kohlmann, S. & Risenhoover, K., 1996. Using artificial food patches to evaluate habitat quality for granivorous birds: An application of foraging theory. Condor, 98(4), pp.854-857.

Krebs, J.R., 1971. Territory and Breeding Density in the Great Tit, Parus Major L. Ecology, 52(1), pp.3-22.

Kroll, A.J. & Haufler, J.B., 2007. Evaluating Habitat Quality for the Dusky Flycatcher. Journal of Wildlife Management, 71, pp.14-22.

Lavers, C., & HainesYoung, R. 1996. Using models of bird abundance to predict the impact of current land-use and conservation policies in the flow country of Caithness and Sutherland, northern Scotland. Biological Conservation, 75(1), pp.71–77.

Letcher, B. H., Priddy, J. A., Walters, J. R., & Crowder, L. B., 1998. An individual-based, spatially-explicit simulation model of the population dynamics of the endangered red-cockaded woodpecker, Picoides borealis. Biological Conservation, 86(1), pp.1-14.

Lyons, J., 2005. Habitat specific foraging of Prothonotary warblers: Deducing habitat quality. Condor, 107(1), pp.41-49.

MacKenzie, D.I., 2006. Occupancy estimation and modeling: inferring patterns and dynamics of species occurrence, Academic Press.

Marra, P. & Holberton, R., 1998. Corticosterone levels as indicators of habitat quality: effects of habitat segregation in a migratory bird during the non-breeding season. Oecologia, 116(1-2), pp.284-292.

Maurer, B.A., McArthur, L.B. & Whitmore, R.C., 1980. Habitat associations of birds breeding in clearcut deciduous forests in West Virginia. The use of multivariate statistics in studies of wildlife habitat. US Dep. Agric., For Serv. Gen. Tech. Rep. RM-87, pp.167–172.

Millennium Ecosystem Assessment, 2009. Ecosystems and human well-being: biodiversity synthesis. Summary for Decision-makers. Washington, DC.: World Resources Institute. The full range of reports is available on the Millennium Ecosystem Assessment web site. Retrieved on, pp.03–10.

Minor, E.S. & Urban, D.L., 2007. Graph theory as a proxy for spatially explicit population models in conservation planning. Ecological Applications, 17(6), pp.1771–1782.

Minor, E. S., McDonald, R. I., Treml, E. A., & Urban, D. L., 2008. Uncertainty in spatially explicit population models. Biological Conservation, 141(4), pp.956-970.

Moilanen, A., 1999. Patch occupancy models of metapopulation dynamics: efficient parameter estimation using implicit statistical inference. Ecology, 80(3), pp.1031–1043.

Moilanen, A & Hanski, I, 1998. Metapopulation dynamics: Effects of habitat quality and landscape structure. Ecology, 79(7), pp.2503-2515.

Mortelliti, A., Amori, G. & Boitani, L., 2010. The role of habitat quality in fragmented landscapes: a conceptual overview and prospectus for future research. Oecologia, 163, pp.535-547.

O’Hara, R. B., Arjas, E., Toivonen, H., & Hanski, I., 2002. Bayesian analysis of metapopulation data. Ecology, 83(9), pp.2408–2415.

Ovaskainen, O. & Hanski, I., 2003. How much does an individual habitat fragment contribute to metapopulation dynamics and persistence? Theoretical Population Biology, 64(4), pp.481-495.

Ozgul, A., Armitage, K. B., Blumstein, D. T., Vanvuren, D. H., & Oli, M. K., 2006. Effects of patch quality and network structure on patch occupancy dynamics of a yellow- bellied marmot metapopulation. Journal of Animal Ecology, 75, pp.191-202.

Paradis, E. & Croset, H., 1995. Assessment of Habitat Quality In The Mediterranean Pine Vole (Microtus-duodecimcostatus) by the Study of Survival Rates. Canadian Journal Of Zoology-Revue Canadienne De Zoologie, 73(8), pp.1511-1518.

Perot, A. & Villard, M.-A., 2009. Putting Density Back into the Habitat-Quality Equation: Case Study of an Open-Nesting Forest Bird. Conservation Biology, 23, pp.1550-1557.

Pettorelli, N., Gaillard, J.-M., Yoccoz, N. G., Duncan, P., Maillard, D., Delorme, D., Van Laere, G., Toigo, C., 2005. The response of fawn survival to changes in habitat quality varies according to cohort quality and spatial scale. Journal of Animal Ecology, 74(5), pp.972-981.

Pidgeon, A.M., Radeloff, V.C. & Mathews, N.E., 2006. Contrasting measures of fitness to classify habitat quality for the black-throated sparrow (Amphispiza bilineata). Biological Conservation, 132(2), pp.199–210.

Powell, G. & Powell, A., 1986. Reproduction by Great White Herons Ardea-Herodias in Florida Bay as an Indicator of Habitat Quality. Biological Conservation, 36(2), pp.101- 113.

Prenda, J., Lopez-Nieves, P. & Bravo, R., 2001. Conservation of otter (Lutra lutra) in a Mediterranean area: the importance of habitat quality and temporal variation in water availability. Aquatic Conservation-Marine and Freshwater Ecosystems, 11(5), pp.343-355.

Proctor, M. F., McLellan, B. N., Strobeck, C., & Barclay, R. M., 2005. Genetic analysis reveals demographic fragmentation of grizzly bears yielding vulnerably small populations. Proceedings of the Royal Society B: Biological Sciences, 272(1579), pp.2409 - 2416.

Pulliam, H. Ronald & Danielson, Brent J., 1991. Sources, Sinks, and Habitat Selection: A Landscape Perspective on Population Dynamics. The American Naturalist, 137, p.S50- S66.

Pulliam, H.R., 1988. Sources, sinks, and population regulation. American Naturalist, pp.652– 661.

Rabasa, S.G., Gutierrez, D. & Escudero, A., 2007. Metapopulation structure and habitat quality in modelling dispersal in the butterfly Iolana iolas. Oikos, 116(5), pp.793-806.

Ricketts, T.H., 2001. The matrix matters: effective isolation in fragmented landscapes. The American Naturalist, 158(1), pp.87–99.

Robertson, B.A. & Hutto, R.L., 2006. A framework for understanding ecological traps and an evaluation of existing evidence. Ecology, 87(5), pp.1075–1085.

Robles, M. D., Flather, C. H., Stein, S. M., Nelson, M. D., & Cutko, A., 2008. The geography of private forests that support at-risk species in the conterminous United States. Frontiers in Ecology and the Environment, 6(6), pp.301-307.

Rodrigues, A. S. L., Andelman, S. J., Bakarr, M. I., Boitani, L., Brooks, T. M., Cowling, R. M., Fishpool, L. D. C., et al., 2004. Effectiveness of the global protected area network in representing species diversity. Nature, 428(6983), pp.640-643.

Schilman, P. & Roces, F., 2003. Assessment of nectar flow rate and memory for patch quality in the ant Camponotus rufipes. Animal Behaviour, 66(Part 4), pp.687-693.

Sellars, R.W., 1999. Preserving nature in the national parks: a history, Yale University Press.

Senar, J., Conroy, M. & Borras, A., 2002. Asymmetric exchange between populations differing in habitat quality: a metapopulation study on the citril finch. Journal of Applied Statistics, 29(1-4), pp.425-441.

Sinsch, U., Leskovar, C., Drobig, A., Koenig, A., & Grosse, W.-R., 2007. Life-history traits in green toad (Bufo viridis) populations: indicators of habitat quality. Canadian Journal of Zoology-Revue Canadienne de Zoologie, 85(5), pp.665-673.

Stanko-Mishic, S. & others, 1999. Manipulation of habitat quality: effects on chironomid life history traits. Freshwater Biology, 41(4), pp.719–727.

States, J.B., 1976. Local Adaptations in Chipmunk (Eutamias amoenus) Populations and Evolutionary Potential at Species’ Borders. Ecological Monographs, 46(3), pp.221-256.

Svärdson, G., 1949. Competition and habitat selection in birds. Oikos, 1(2), pp.157–174.

Tamarin, R. H. 1978. Dispersal, population regulation, and K-Selection in field mice. American Naturalist, 112(985), pp.545–555.

Taylor, P. D., Fahrig, L., Henein, K., & Merriam, G., 1993. Connectivity is a vital element of landscape structure. Oikos, 68(3), pp.571–573.

Urban, D. & Keitt, T., 2001. Landscape connectivity: a graph-theoretic perspective. Ecology, 82(5), pp.1205–1218.

Urban, DL & Shugart Jr, H., 1984. Avian demography in mosaic landscapes: modeling paradigm and preliminary results, Oak Ridge National Lab., TN (USA).

US Fish and Wildlife Service, 2003. Recovery Plan for the Red-cockaded Woodpecker (Picoides Borealis)., US Fish and Wildlife Service, Southeast Region.

Van Horne, B. 1983. Density as a misleading indicator of habitat quality. Journal of Wildlife Management, 47(4), pp.893–901.

Verhoeven, K. J. F., Macel, M., Wolfe, L. M., & Biere, A., 2011. Population admixture, biological invasions and the balance between local adaptation and inbreeding depression. Proceedings of the Royal Society B: Biological Sciences, 278(1702), p.2.

Vierling, K.T., 1999. Habitat quality, population density and habitat-specific productivity of red-winged blackbirds (Agelaius phoeniceus) in Boulder County, Colorado. The American Midland Naturalist, 142, pp.401-409.

Voegeli, M., Serrano, D., Pacios, F., & Tella, J. L., 2010. The relative importance of patch habitat quality and landscape attributes on a declining steppe-bird metapopulation. Biological Conservation, 143(5), pp.1057-1067.

Vuilleumier, S. et al., 2007. How patch configuration affects the impact of disturbances on metapopulation persistence. Theoretical Population Biology, 72(1), pp.77–85.

Waser, P. M. 1985. Does competition drive dispersal? Ecology, 66(4), pp.1170–1175.

Webb, T.J., Noble, D. & Freckleton, R.P., 2007. Abundance–occupancy dynamics in a human dominated environment: linking interspecific and intraspecific trends in British farmland and woodland birds. Ecology, 76, pp.123–134.

Weinberg, H.J. & Roth, R.R., 1998. Forest area and habitat quality for nesting Wood Thrushes. The Auk, pp.879–889.

Weiss, S., Murphy, D. & White, R., 1988. Sun, Slope, and Butterflies - Topographic Determinants Of Habitat Quality For Euphydryas-Editha. Ecology, 69(5), Pp.1486- 1496.

Wightman, C.S. & Germaine, S.S., 2006. Forest stand characteristics altered by restoration affect Western Bluebird habitat quality. Restoration Ecology, 14(4), pp.653-661.

Wilcove, D.S. & Lee, J., 2004. Using Economic and Regulatory Incentives to Restore Endangered Species: Lessons Learned from Three New Programs. Conservation Biology, 18(3), pp.639-645.

Wilkin, T.A., Perrins, C.M. & Sheldon, B.C., 2007. The use of GIS in estimating spatial variation in habitat quality: a case study of lay-date in the Great Tit Parus major. Ibis, 149(2), pp.110-118.

Winker, K., Rappole, J.H. & Ramos, M.A., 1995. The use of movement data as an assay of habitat quality. Oecologia, 101(2), pp.211–216.

Winnie, Jr., J.A., Cross, P. & Getz, W., 2008. Habitat quality and heterogeneity influence distribution and behavior in African buffalo (Syncerus caffer). Ecology, 89(5), pp.1457- 1468.

Wright, L.I., Tregenza, T. & Hosken, D.J., 2008. Inbreeding, inbreeding depression and extinction. Conservation Genetics, 9(4), pp.833–843.

CHAPTER II MODELING RED-COCKADED WOODPECKER (PICOIDES BOREALIS) HABITAT QUALITY AT HIGH RESOLUTION AND LARGE EXTENTS USING LIDAR Introduction

Ecologists, conservation biologists, and land managers cannot effectively plan for the recovery of imperiled species without knowing the extent and spatial distribution of suitable habitat (Morrison 2006), but how can we accurately map habitat over large spatial extents?

Embedded in this problem are two separate but interrelated challenges: how best to represent the complex relationships between organisms and their environment, and how to collect high-resolution habitat data on the large scales relevant to regional conservation planning. Many imperiled species respond non-linearly to variation in habitat characteristics which are difficult to map at large spatial scales (Guisan et al. 2002), and habitat data are often poor or absent on privately-owned lands that could contribute to the recovery of many species (Robles et al. 2008; Knight 1999). Moreover, as they are currently implemented, most regional-scale habitat models do not explicitly integrate information on how habitat characteristics and landscape configuration jointly influence the fitness of target species

(Haynes et al. 2007; Cozzi et al. 2008). Here we address these challenges using forest structural data derived from airborne laser altimetry (LiDAR), and model habitat quality for the federally endangered Red-cockaded Woodpecker (Picoides borealis, RCW) across nearly a million hectares of North Carolina’s Coastal Plain.

At regional scales (>100,000 ha), ecologists often rely on correlative ecological niche models (CENMs) to predict habitat suitability for species based on climate, topography, land-cover, or other remotely-sensed data (Elith & Leathwick 2009). Regional-scale CENMs typically use spatial records of the presence of a species as a dependent variable (Brotons et al. 2004) and relate these records to environmental data using either regression-based methods such as generalized additive models (Guisan et al. 2002), or machine-learning algorithms such as neural networks (Lek et al. 1996) or Maxent (Dudik & Schapire 2004).

At smaller scales wildlife biologists often model habitat using indicators of fitness such as body mass (Pettorelli et al. 2002), productivity (Pidgeon et al. 2006), or social structure

(Atwood 2006) as dependent variables. These indicators are then often related to habitat data collected in the field to reveal local habitat factors that affect the fitness of target species

(Walters et al. 2002). Both methods provide valuable information on the ecological relationships that influence the distribution and abundance of species, but so far there has been little integration of these two complimentary approaches (Aldridge & Boyce 2007).

Such integration would allow investigators to better prioritize critical habitat patches for conservation at large extents, and assess the relative contribution of local factors describing patch quality and regional-scale factors such as habitat connectivity in contributing to the presence and abundance of target species.

A major barrier to the integration of presence-based CENMs with fitness-based habitat models has been the scale mismatch between coarse scale environmental data sets typically used to train CENMs, and fine-scale field-based habitat information that contributes to models of wildlife species fitness. In the past decade, a growing number of studies have used LiDAR data to characterize habitat structure at high resolution or relatively large spatial extents, particularly for birds (Bradbury et al. 2005). Forest canopy parameters

such as height, density, and heterogeneity can be derived from LiDAR data, enabling predictive models of bird species distributions based on known life history characteristics or habitat affinities (Goetz et al. 2007; Graf et al. 2009; Hill et al. 2004; Seavy et al. 2009). For example, Bradbury et al. (2005) showed that canopy height and roughness metrics derived from LiDAR helped predict breeding success for Blue Tits and Great Tits in a woodland preserve in Great Britain. Most of these studies use waveform LiDAR (Lefsky et al. 2002) and high-density (i. e. >1 pulse/m2) discrete-return LiDAR (Maltamo et al. 2005) which is preferable for precise measurements of forest structure. However LiDAR data collected specifically for forestry and wildlife applications generally cover areas from a few (Riaño et al. 2003) to several thousand hectares (Falkowski et al. 2009) because of cost and data- management considerations. Yet its now-ubiquitous use for digital terrain mapping has made low-density (<1 pulse/m2), discrete-return LiDAR data sets publically available at statewide extents. In one of the most extensive studies to date, Hawbaker et al. (2009) used airborne, low-density (0.4 returns/m2) LiDAR to map vertical canopy structure across

53,000ha of forest in Wisconsin. This extent is still small compared to the range of most bird species, however, and the utility of low-density discrete return LiDAR for mapping wildlife habitat at regional extents ( > 100,000 ha) has been suggested (Vierling 2008) but not yet demonstrated.

Habitat models are developed under the assumption that the habitat characteristics important to the species of concern are well-known. One imperiled species for which these characteristics are well understood is the Red-cockaded Woodpecker (RCW). RCWs are endemic to isolated fragments of fire-maintained pine forest in the southeastern Coastal

Plain of North America, and are the only woodpecker species known to excavate cavities in living pine trees (Conner et al. 2001). RCWs prefer to nest in longleaf pines (Pinus palustris)

that are more than 60 years old when available, but when restricted they will use loblolly

(>70 yr) or shortleaf (>75 yr) pine, and have also been known to inhabit slash, pond, pitch, and Virginia pines (Conner et al. 2001; James et al. 2001). There is evidence that they prefer forests or woodlands with little pine or hardwood understory or midstory for both nesting

(Conner et al. 2001) and foraging (Walters et al. 2002). According to Walters et al., the selection of foraging habitat is largely dependent on what is available. In old-growth longleaf pine woodlands such as the Wade Tract in Georgia, RCWs forage on pines >50cm dbh, whereas in landscapes dominated by younger forests, they may accept pines as small as 25cm dbh (Conner 2001). In agreement with James et al. (2001) and others, Walters et al. (2002) describes high quality habitat (both nesting and foraging) for RCWs to be "open woodlands with little or no hardwood or pine midstory and intermediate densities of large old pines, including some old-growth pines." These recommendations have been incorporated into the land management guidelines of the most recent version of the US Fish and Wildlife Service's recovery plan for RCWs (U.S. Fish & Wildlife Service 2003).

A social group of RCWs consists of a single breeding pair and one to several helpers that share in defending a common territory, constructing nest cavities, incubating eggs, and feeding both nestlings and fledglings (Lennartz et al. 1987). The presence of helpers has been shown to dramatically decrease the mortality of breeding birds (Khan et al. 2001), and enhance nestling survival (Conner et al. 2004). The size of RCW groups has long been used as a measure of long-term fitness in this species (Conner & Rudolph 1991; Walters et al.

2002), and here we use it as our key indicator of habitat quality. RCW habitat-fitness relationships have typically been evaluated using field-based surveys of forest structure in the immediate vicinity of clusters (Walters et al. 2002). More recently, work has begun to focus on the possible effects of landscape structure (the arrangement of habitat patches) on

breeding success and population trends in RCWs. Schiegg et al. (2002) and Bruggeman et al.

(2009) have both used individual-based models to explore the effects of spatial arrangement of habitat on RCW population trends. The models show that the spatial arrangement of

RCW groups and their habitat could influence social structure and population persistence in this species, but there is significant uncertainty about how these results apply to RCW populations in real landscapes.

We used a regional (one million ha), low-density (0.27 returns/m2) discrete-return

LiDAR data set and field-based data to develop a habitat model for RCWs across the

Onslow Bight region of coastal North Carolina. To build the model we leveraged two very different but complimentary modeling tools. To relate the spatial locations of known areas of good quality habitat to LiDAR-based forest structural data we used Maxent (Phillips &

Dudik 2008). To estimate the major habitat and landscape factors that influence the size of

RCW groups, we applied multi-model inference to linear mixed models (Pinhero & Bates

2000; Burnham & Anderson 2002). The two complementary modeling approaches were used to predict the distribution of suitable habitat on the landscape and to estimate the number of birds that a particular location might support. We hypothesized that 1) LiDAR- derived forest structure metrics would be strongly correlated with field-based measures of vegetation structure in ways that allow us to distinguish good RCW habitat at a high resolution over large spatial extents, and 2) both forest structural features in the immediate vicinity of RCW clusters, and the position of those clusters in the larger landscape would influence the size, and presumably the long-term success, of RCW groups. We addressed the second hypothesis by combining the coarse-scale CENM and fitness-based approaches.

Methods

Study Area

The Onslow Bight region (Figure 5) covers approximately one million hectares, from the inner coastal plain to the barrier islands and is home to North Carolina’s second largest

RCW population. Prior to European settlement, an estimated 48% of the region was longleaf pine or mixed pine habitat, (C. Frost & J. Costanza, unpublished data), much of which was subject to high frequency, low-intensity ground fires which limited the encroachment of shrubs and hardwood trees (van Lear et al. 2005). Currently, the major vegetation types in the landscape are managed pine plantations, fire-suppressed unmanaged pine forest and pocosin wetlands dominated by dense evergreen shrubs. Natural features in the region are threatened by rapid development. Human populations in the Onslow Bight’s

11 counties are expected to grow 29% by 2020 and the population of Pender County is projected to increase by 74% in that period (NC Office of State Budget and Management

2008). Major public landholdings in the Onslow Bight include US Marine Corps Base Camp

Lejeune (US Department of Defense, henceforth Camp Lejeune), Croatan National Forest

(US Forest Service, henceforth CNF), Cedar Island National Wildlife Refuge (US Fish and

Wildlife Service), and several Game Lands (NC Wildlife Resources Commission).

Collectively, these comprise 15% of the landscape. Private agencies such as The Nature

Conservancy (TNC, 1%) and the North Carolina Forestry Foundation (Hofmann Forest,

3%) also manage land in the Onslow Bight. The largest extant populations of RCWs exist on Croatan National Forest, Camp Lejeune, and Holly Shelter Game Land.

Figure 5: The Onslow Bight region of North Carolina, our study area. “Managed Lands” incorporate wildlife conservation as a land-management goal.

Field Data

To train the habitat model, we used field-based forestry data collected on Camp

Lejeune, the second-largest public landholding on the Onslow Bight. RCW foraging habitat was characterized across the entire base from 1998-2000 (Geotechnical and Environmental

Consultants 2000). In this three-year period 5,670 circular plots with a radius of 11.3m were established in a regular grid with a spacing of 100.6m north-south by 241.4m east-west. Plot centers were established using differentially corrected GPS. The numbers of pine and hardwood stems in 5cm dbh classes were collected at each plot, along with visual estimates of understory density (in 4 classes) and height (5 classes). A representative tree was selected

in each plot and its height was estimated with a clinometer. In alternating plots the age of the representative tree was estimated using an increment borer (Geotechnical and

Environmental Consultants 2000).

To relate forest structural variables to RCW social structure we assembled a database of social group size for RCW clusters on Camp Lejeune and Croatan National Forest for the years 1998-2001, corresponding to the time period of the forest plot data. A large-scale and long-term banding program was in place at this time, and social group size was recorded from multiple point counts at each cluster location according to monitoring guidelines established by the US Fish and Wildlife Service (U.S. Fish and Wildlife Service 2003). A complete four-year group size data set was available for a total of 105 family groups, including 48 groups in Camp Lejeune and 57 groups in CNF.

Remotely-sensed Data

Between 2001 and 2006, the State of North Carolina and the Federal Emergency

Management Administration partnered to fund the collection of LiDAR data for the purpose of building high resolution (6.1m) digital elevation models for the entire state. This data was collected with sensors that yield a lower density of laser returns than is typical for forestry applications, nonetheless LiDAR data for the Onslow Bight contains 2.2 billion georeferenced three-dimensional points representing places where the laser pulses reflected off of vegetation, buildings, and the ground surface (Figure 6a, 6c). The LiDAR data set was collected by three contractors from January to March 2001, using three different sensors.

The sensor that covered the majority (78%) of the study region was a Leica Geosystems

Aeroscan system which had the lowest return-density of the three sensors, and thus the greatest amount of uncertainty regarding its usefulness for deriving forest structure. This

sensor had a nominal post spacing of 3m and an average return density of 0.27 returns/m2.

The sensor was mounted on an aircraft flown at an altitude between 3658m and 2438m, and scanned an angle of 50 degrees, yielding overlapping swaths of data 2274m to 3411m wide.

The raw data was collected in proprietary data formats and converted to industry-standard

LAS format by the NC Floodplain Mapping Program. The conversion process stripped all information from the points except for their spatial location.

Figure 6: Examples of LiDAR and field data used to build the RCW habitat model. (A) Heights of raw LiDAR returns in a 900 ha region of Camp Lejeune. (B) Forestry plots used to train the habitat model. (C) Fifty-meter wide transect of LiDAR data from the southeastern portion of Figure 6a (gray box) showing different characteristic patterns of returns in different habitat types.

We used the software package Fusion/LDV (McGaughey 2009) to process the raw laser returns (Figure 6a, 6c) into forest structural information. We built a canopy height model which subtracts the return elevations from a high-resolution (6.1m) digital elevation model and fit a smoothed spline surface to the tallest return in each 15m raster cell.

Previous studies have shown that RCWs are particularly sensitive to mid-story structure between 2.14m and the bottom of the forest canopy (Walters et al. 2002), so in addition to canopy height, we also created estimates of percent horizontal vegetation cover, canopy cover, midstory cover, and understory cover at a 30m resolution. Total vegetative cover was estimated as the proportion of total returns greater than 0.61m in height. Canopy cover was computed as the proportion of returns above 6.1m in height. We defined midstory cover as the number of returns between 2.14m and 6.10m in height divided by the total number of returns below 6.10m. Similarly, we defined understory cover as the number of returns between 0.61m and 2.14m in height divided by the number of returns below 2.14m.

Habitat Standards

The US Fish and Wildlife Service synthesized the body of literature on RCW habitat requirements into a set of vegetation structural criteria used to guide land managers (Table

2). This “recovery standard” describes vegetation conditions that are believed to be adequate for long-term population increase in this species (U.S. Fish and Wildlife Service

2003). We used the criteria that could be derived from the forestry plots collected on Camp

Lejeune to select areas with structural characteristics that met this recovery standard (Figure

6b). This subset of plots became locations used to train our habitat model based on remotely-sensed data. Surprisingly, out of 5670 forested plots collected on Camp Lejeune, only 44 met all of the structural criteria in the “recovery standard”. We suspect that this is an artifact of the sampling methodology: the habitat standards were intended to be evaluated at the scale of entire forest stands (U.S. Fish and Wildlife Service 2003), while the plot data was collected at a smaller scale. We are confident that the selected plots do meet the habitat standard, but have no confidence that the excluded plots do not meet the standard. In

short, this field data set gives us good information about the presence of RCW habitat, but says little about its absence. This restricts the field of techniques that we could use to build the habitat model to those that use presence-only data, such as Maxent, described below.

Table 2: Vegetation structural criteria for the USFWS “recovery standard” and “managed stability standard” from USFWS (2003) along with corresponding conditions used to select forested plots on Camp Lejeune.

Recovery Standard Criteria Queried vegetation plots There are 45 or more stems/ha of pines that are ≥ 60 years in age and > 35 cm dbh. Large old pines; ≥ 60 years old AND > 35 Minimum basal area for these pines is 4.6 m2/ha. cm dbh AND ≥ 4.6 m2/ha basal area (BA)

Basal area of pines 25.4-35 cm dbh is between 0 and 9.2 m2/ha Medium pines BA ≤ 9.2 m2/ha

Basal area of pines < 25.4 cm dbh is below 2.3 m2/ha and below 50 stems/ha Small pines BA < 2.3 m2/ha

Basal area of all pines ≥ 25.4 cm dbh is at least 9.2 m2/ha. That is, the minimum basal area Medium (b) and large pines (a) BA ≥ 9.2 for pines in categories (a) and (b) above is 9.2 m2/ha. m2/ha

Groundcovers of native bunchgrass and/or other native, fire-dependent herbs ≥ 40% of ground and midstory plants and are dense enough to carry growing season fire at least once NOT INCLUDED every 5 years.

No hardwood midstory exists, or if a hardwood midstory is present it is sparse and < 2.1m Understory density “none” OR “light” in height AND understory height < 3.4 m

Canopy hardwoods are absent or < 10% of the number of canopy trees in longleaf forests NOT INCLUDED and < 30% of the number of canopy trees in loblolly and shortleaf forests.

All of this habitat is within 0.8 km of the center of the cluster, and preferably, 50% or more NOT INCLUDED is within 0.4 km of the cluster center. Foraging habitat is not separated by more than 61m or non-foraging areas. Non-foraging areas include (1) any predominantly hardwood forest, (2) pine stands less than 30 years old, NOT INCLUDED (3) cleared land such as agricultural lands or recently clearcut areas, (4) paved roadways, (5) utility rights of way, and (6) bodies of water

Modeling prevalence

We used the field data plots that met the recovery standard (44 plots), to train a habitat suitability model using Maxent 3.2.1 (Phillips & Dudik 2008). Maxent is a machine- learning algorithm that finds the probability distribution of maximum entropy (that which is closest to uniform), subject to the constraints that the expected values of a set of environmental variables under that estimated distribution matches their empirical distribution (Dudik & Schapire 2004; Phillips et al. 2006). As environmental predictors, we

used canopy height, total cover, canopy cover, midstory cover, and understory cover estimates derived from LiDAR, along with satellite-derived landcover information from

(Southeast Gap Analysis Project 2008), and simplified SSURGO soils data. Inclusion of soils data allowed us to better differentiate longleaf pine woodlands from pine plantations, which are commonly located on peatland pocosin soils in this landscape.

Maxent’s logistic outputs are continuous estimates of habitat suitability bounded between zero and one. To convert this continuous output to a binary habitat classification, we developed a numerical optimization procedure that determines which threshold minimizes the total proportion of the landscape that is misclassified. Comparing the results of the habitat model to high-resolution infrared aerial photographs, we determined that the model was overpredicting habitat on some mature loblolly pine (Pinus taeda) plantations.

These forest stands are unlikely to meet the recovery standard because they are typically harvested well-before the trees reach 60 years of age (pers. obs.), often considered the minimum age to support RCW cavities (U.S. Fish and Wildlife Service 2003). In order to find the logistic threshold that eliminates these plantations from the model while capturing high quality RCW habitat, we heads-up digitized both pine plantations on private lands which were selected by Maxent, and longleaf pine woodlands within 200m of active RCW clusters where we had records of prescribed fire in the period 1998-2001 (unpublished data).

We considered these longleaf pine woodlands to be good quality habitat. Our optimization routine minimized the proportion of pine plantations captured by the model plus the proportion of good habitat absent from the model (Equation 1).

(1)

In this equation, Pc is the area of pine plantation captured by the model, Pt is the total area of pine plantation digitized, Gn is the area of good habitat not captured by the model, and Gt is the total area of recovery standard habitat digitized.

For each of the LiDAR-derived habitat models, we applied a mask based on the

Southeast GAP analysis project’s land cover (Southeast Gap Analysis Project 2008) that excluded areas not dominated by evergreen trees. RCWs require pine-dominated forests, and because LiDAR data was collected from December-March, it does not accurately characterize the density, cover, or height of deciduous vegetation.

Modeling fitness

To further refine our habitat model and test our hypothesis that both site-level forest structural characteristics and landscape position influence RCW group size , we regressed group size against three pools of predictors: forest structural characteristics within

800m of cluster centers, landscape attributes within 2-8km of clusters, and a pool of predictors that incorporated both forest structural characteristics and landscape attributes.

To generate predictors for the regression analysis, we used five of the LiDAR-derived metrics that were related to forest structure characteristics previously shown to affect RCWs: canopy height, canopy cover, midstory cover, understory cover, and total cover. We then computed focal means for each variable in 100-, 200-, 400-, and 800-m circular windows around each cluster center. To examine how habitat heterogeneity influences group size, we also computed focal standard deviations for these variables at the same scales. In order to examine the influence of habitat availability on group size, we computed the area that was captured by our model of the recovery standard within 400, 800, 2000, and 4000m of cluster centers (Table 3). An analysis of spatial autocorrelation in the data revealed that social group

size is spatially autocorrelated (Global Moran’s I=0.34, p<0.001) at lag distances of 500 to

6,000m (Figure 7). To account for this large-scale variation in our regression analysis, we computed several landscape metrics, including the kernel density of active RCW clusters at

2,000, 4,000, and 8,000m, the distance from a vacant cluster, and the betweenness centrality and node degree (Minor & Urban 2007) of each cluster in a graph of all active clusters using the program NetworkX (Hagberg et al. 2009).

Figure 7: Spline correlogram showing spatial autocorrelation of mean RCW group size from 1998 -2001 in the Onslow Bight, computed according to Bjørnstad and Falck (2001).

Table 3: Parameters incorporated into regressions of RCW group size. Variable Source Habitat Structure Variables Numeric year of groups size observation starting in 1998 Field Data Categorical variable indicating if the cluster center was burned from 1998- Field Data Area of good quality foraging habitat within 800m of cluster (hectares). LiDAR 2001. Area of pine forest within 800m of cluster center (hectares) LiDAR Understory cover in 30m raster cell containing cluster center. LiDAR Mean understory cover within 100m of cluster center. LiDAR Standard deviation of understory cover within 100m of cluster center. LiDAR Standard deviation of understory cover within 200m of cluster center. LiDAR Mean understory cover within 800m of cluster center. LiDAR Midstory cover in 30m raster cell containing cluster center. LiDAR Standard deviation of midstory cover within 200m of cluster center. LiDAR Standard deviation of midstory cover within 400m of cluster center. LiDAR Mean midstory cover within 800m of cluster center. LiDAR Canopy cover in 30m raster cell containing cluster center. LiDAR Mean canopy cover within 100m of cluster center. LiDAR Standard deviation of canopy cover within 400m of cluster center. LiDAR Vegetation cover in raster cell containing cluster center. LiDAR Standard deviation of vegetation cover within 400m of a cluster center. LiDAR Mean vegetation cover within 400m of cluster center. LiDAR Canopy height in 15m raster cell containing cluster center. LiDAR Mean Canopy height within 100m of cluster center. LiDAR Standard deviation of canopy height within 200m of a cluster center. LiDAR Mean Canopy Height within 400m of cluster. LiDAR Landscape Variables Kernel-computed density of RCW clusters within 4 kilometers of a cluster Field Data Area of good quality habitat within 2000m of a cluster center. Maxent center. Straight line distance from the nearest center of high cluster density (m). Field Data Area of pine forest within 2000m of a cluster center. GAP Betweenness centrality of a cluster in a graph containing all active clusters. Networkx Straight line distance from the nearest cluster that was inactive in 2001. Field Data

Count data thought to often follow a Poisson distribution (Clark 2007), but simple

linear models with both the sum of group size from 1998-2001, the mean of group size in

the same period, and the count of group size in 2001 showed that models using a Gaussian

distribution function and identity link had a lower AICc. Because of the large number of

possible parameters, we carefully selected those parameters which were strongly correlated with group size, and were related to habitat features previously considered important for

RCWs (Table 3). We then built generalized linear mixed models of all possible combinations of predictors incorporating five parameters or less and selected those models that were within 2.3 ΔAIC of the optimal model (Burnham & Anderson 2002). Because group size in a particular cluster is highly correlated between years, all models incorporated cluster number as a random effect. We further screened this pool of credible models by eliminating those in which any of the terms exhibited strong multicollinearity among predictors (Pearson correlation coefficient > 0.75). To examine the influence of forest structural characteristics on group size independent of the landscape context we constructed a second pool of models that incorporated only forest structure metrics. This model pool was subjected to the same screening process. To examine the influence of landscape structure on group size independent of habitat factors, we constructed a third pool of models that contained only landscape characteristics as predictors. Each pool of highly credible models was then averaged according to each model’s Akaike weight to generate multi-model parameter estimates and confidence intervals (Burnham & Anderson 2002). Multi-model parameter estimates account for uncertainty in the data as well as uncertainty in model selection among well-performing regression models. Parameters with significant multi-model slopes are highly significant across the pool of top-performing models.

All statistical analyses were performed with the software package R (2.9.0, R Core

Development Team 2009), using the contributed packages “nlme” (Pinheiro 2000), “spdep”

(Bivand 2008), and “ncf” (Bjørnstad & Falck 2001)

Results

LiDAR data validation

LiDAR-derived forest structure metrics were moderately to strongly correlated with field-based measurements of vegetation structure in the 1801 forest plots dominated by pines on Camp Lejeune (Figure 8a). LiDAR-derived canopy height estimates were strongly correlated with ground-based estimates of the height of a representative tree within each

0.04-ha circular plot (R2=0.57, t=48.6, p<0.001). Root mean squared deviation between ground-based and LiDAR-based canopy height estimates was 3.65 m. Other metrics derived from LiDAR did not have an exact field-based analog. For example, field data included only a categorical estimation of understory density; however this was clearly related to our

LiDAR-derived estimate of understory horizontal cover (Figure 8b).

Figure 8: LiDAR-derived forest structure metrics compared against field data metrics collected at Camp Lejeune for (A) tree height and (B) understory density. In (B), random jitter has been added to the x-axis for clarity.

Prevalence model

Diagnostics indicated that the recovery standard model performed well (Regularized training gain 3.497, AUC=0.995, p<0.001). Using the numerical optimization procedure we chose a logistic threshold of 0.102, which resulted in a final binary habitat model for the recovery standard that captured 91% of digitized high quality longleaf pine woodlands, while capturing just 11% of the loblolly pine plantations that were initially given high suitability values by Maxent. Although the habitat model was not trained with the locations of RCW clusters, it captured the spatial locations of the centers of 121 out of 163 clusters known to be active from 1998-2001 (74%). The model also captured 69% of the core habitat area within 50m of active cluster centers.

Based on our Maxent-derived habitat model, we estimate that approximately 3.2%

(34,200 ha) of the Onslow Bight landscape currently meets the USFWS recovery standard for RCWs (Figure 9). Larger areas of suitable habitat are concentrated on public lands near existing populations of birds. Numerous small, highly fragmented areas of suitable habitat exist on private lands, comprising approximately 55% of the total good quality habitat on the landscape (Table 4).

Table 4: Habitat types on conserved and non-conserved lands in the Onslow Bight derived from remotely sensed data. We considered lands as “conserved” if they are subject to a conservation easement or owned by an entity that explicitly incorporates wildlife conservation as a land-management goal. Habitat Type Non-protected ha (%) Protected ha (%) Total ha (% of landscape) RCW good quality habitat 19,056 (56%) 15,138 (44%) 34,194 (3%) Other forested 380,695 (81%) 87,231 (19%) 467,925 (46%) Scrub 42,769 (83%) 84,62 (17%) 51,231 (5%) Pocosin 61,807 (43%) 82,705 (57%) 144,512 (14%) Open 229,882 (91%) 21,687 (9%) 251,569 (25%) Herbaceous Wetland 44,600 (69%) 20,105 (31.07%) 64,705 (6%)

Figure 9: Habitat model for the Red Cockaded Woodpecker in 2001 derived from LiDAR and other remotely-sensed data corresponding to the USFWS recovery standard (Red). Models were created using Maxent and trained with points located on Marine Corps Base Camp Lejeune.

Clusters located on Camp Lejeune had considerably more available good quality habitat within 800m of cluster centers than those located on Croatan National Forest or

Holly Shelter Game Land. Pocosin wetlands, not included in the habitat model, are more common in the landscape near clusters located in Croatan National Forest and Holly Shelter

Game Land than in Camp Lejeune (Table 4).

Based on our Maxent-derived habitat model, we estimate that approximately 3.2%

exist on private lands, comprising approximately 55% of the total good quality habitat on the landscape (Table 4).

Clusters located on Camp Lejeune had considerably more available good quality habitat within 800m of cluster centers than those located on Croatan National Forest or

Holly Shelter Game Land. Pocosin wetlands, not included in the habitat model, are more common in the landscape near clusters located in Croatan National Forest and Holly Shelter

Game Land than in Camp Lejeune (Table 5).

Table 5: Percentage of habitat types within 800m of active RCW cluster centers on Camp Lejeune (MCBCL), Croatan National Forest (CNF) and Holly Shelter Game Land (HSGL) in 2001, as determined by remote sensing data. Habitat Type All CNF HSGL MCBC Recovery Standard 29% 25% 12% 44% L Pocosin 27% 24% 60% 10% Habitat Other Forest 34% 42% 21% 31% Scrub 3% 3% 4% 3% Open/Developed 6% 5% 2% 9% Non-forested 1% 1% 1% 2%

Wetlands Fitness model

Average group sizes from 1998-2001 ranged from 0.50 to 5.75 birds with a mean of

2.61 (Figure 10). Clusters located on Camp Lejeune were significantly larger (mean 3.08) than those located in Croatan National Forest (mean 2.22, t=5.72, p<0.001). Group size also varied over time. The average within-cluster standard deviation was 0.62, and group sizes increased slightly from 1998-2001 (R2=0.02, t=3.12, p=0.002).

Figure 10: Histogram of mean RCW group size in Croatan National Forest and Camp Lejeune from 1998-2001.

Table 6: Best performing regression models explaining RCW group size at 105 clusters over four years of observation, 1998-2001. All models were linear mixed effects models with cluster identifier as a random effect. Models were ranked according to sample-size corrected Akaiki’s Information Criterion (AICc).

Model # Params AICc Δ AICc AICc Wt. Log Lik. Landscape + Structure 8 1036.084 0.000 0.985 -509.292 Landscape Only 5 1044.844 8.759 0.012 -517.119 Structure Only 7 1047.811 11.726 0.003 -516.328

Table 7: Variables incorporated into optimal regression models of RCW groupsize, and multi-model weighted parameter estimates from all models within 2.3 ΔAICc of the optimum. Asterisk indicates the parameter is significant (p<0.05). Multi-model standard errors and tests of significance are conditional on a pool of optimal models. Pearson Corr. Multi-model Multi- Parameter (Group Size) Estimate model S.E. Models with Habitat Structural Characteristics (29 models) Intercept n/a 0.948 0.625* Number of years after 1998 n/a 0.145 0.030* Mean canopy height <400m from cluster -0.316 -0.028 0.007* Std. dev. of canopy cover <400m from cluster -0.079 -0.779 0.360* Std. dev. of midstory cover <200m from cluster -0.421 -0.468 0.140 Understory cover at cluster center (30m raster cell) -0.158 -0.049 0.033 Mean understory cover <100m from cluster -0.263 -0.092 0.070 Mean canopy cover <100m from cluster -0.138 0.246 0.213 Mean understory cover <200m from cluster -0.341 -0.107 0.108 Mean midstory cover <800m from cluster -0.435 -0.123 0.155 Mean canopy height <100m from cluster -0.315 0.009 0.012 Std. dev. of understory cover <100m from cluster -0.199 -0.073 0.093 Canopy height at cluster center (30m raster cell) -0.301 0.006 0.008 Area of Good Quality Habitat <800m from cluster 0.364 0.002 0.002 Std dev. of understory (ha)cover <200m from cluster -0.29 -0.123 0.171 Mean midstory cover <200m from cluster -0.332 0.163 0.229 Mean vegetation cover <400m from cluster -0.41 0.148 0.258 Vegetation cover at cluster center (30m raster cell) -0.086 -0.036 0.148 Was the cluster center burned from 1998-2001? n/a -0.025 0.174 Area of pine forest <800m from cluster center. 0.245 0.000 0.003 Std dev. of canopy height <200m from cluster -0.135 -0.002 0.017 Midstory cover at cluster center (30m raster cell) 0.133 0.004 0.042 Canopy cover at cluster center (30m raster cell) -0.111 -0.006 0.133 Std. dev. of vegetation cover <400m from cluster -0.215 -0.005 0.373 Models with Landscape Attributes (11 models) Intercept n/a 1.394 0.248* Number of years after 1998 n/a 0.145 0.030* Density of active clusters within 4km (clusters/ha) 0.563 249.768 49.779* Area of Good Quality Habitat within 2km of cluster 0.396 0.001 0.001 Distance from center(ha) of sub-population (m) -0.361 0.000 0.000 Area of pine forest within 2km of cluster (ha) 0.122 0.032 0.032 Betweenness centrality of cluster 0.071 0.000 0.000 Distance from nearest non-active cluster 0.31 0.000 0.000 Density of active clusters within 2km (clusters/ha) 0.473 -26.079 52.460 Models with Habitat Characteristics and Landscape Attributes (4 models) Intercept n/a 0.959 0.467* Years after 1998 n/a 0.145 0.030* Density of active clusters within 4km (clusters/ha) 0.563 209.207 58.304* Std. dev. of midstory cover <200m from cluster -0.421 -0.363 0.148* Mean canopy height <400m from cluster -0.316 -0.011 0.007 Std. dev. of canopy cover <400m from cluster -0.079 -0.497 0.375

Optimal linear mixed models incorporating only forest structural variables explained approximately 47% of the observed variation in group size (Table 6). Optimal models incorporating only landscape variables explained approximately 46% of the observed variation. When optimal models were constructed with the full pool of predictors, the best models incorporated both forest structural variables and landscape variables (pseudo-R2 =

0.58). Several forest structural characteristics and landscape attributes were moderately to strongly correlated with RCW group size (Table 7). For each of the three pools of models

(structural variables, landscape variables, and all variables), several credible models were very close to the optimal model as measured by AICc. Because these models were biologically credible and indistinguishable from an information-theoretic perspective, we used model averaging based on Akaike weights (Burnham & Anderson 2002) to improve the robustness of our inference about the factors influencing RCW group size.

Parameters and standard errors for the major factors affecting RCW group size were estimated based on multi-model inference (Table 7). Parameter estimates and standard errors account for uncertainty in model selection among best-performing models as well as uncertainty in the data. Our analysis shows that although many factors are correlated with

RCW group size, the strongest and most robust predictors were the density of woodpecker clusters within four kilometers of cluster centers, and the standard deviation of midstory cover within 200m of cluster centers (Table 7). When only habitat structural characteristics within 800m of a cluster center were considered, canopy height and the standard deviation of canopy cover within 400m were both negatively related to group size.

Discussion

LiDAR data was useful for bridging the gap in scale and analytical approach between

CENMs and fitness-based habitat models to accurately map habitat over large spatial extents. Although the LiDAR data set was collected for the purpose of digital terrain mapping and used low-density sampling typically applied to that purpose, we show that the three-dimensional pattern of laser returns was strongly correlated with ground-based measurements of vegetation structure. This regional-scale, high-resolution vegetation information allows us to build models of RCW habitat quality that combine the strengths of

Maxent, namely its ability to incorporate presence-only data and nonlinear relationships, with the strengths of regression-based approaches which permit us to make robust statistical and biological inference about the factors that affect RCW groups size, a strong indicator of long-term fitness.

Our study represents, to our knowledge, the only regional-scale assessment of habitat quality for this endangered species. According to our model only a small proportion of the landscape met forest structural criteria for good quality RCW habitat in 2001. This probably represents a greater than 90% reduction in habitat compared to conditions before European settlement (C. Frost and J. Costanza, unpublished data). Somewhat surprisingly, we found a relatively large proportion of good quality habitat (55%) on lands that currently do not have any measure of legal protection. The vast majority of these areas are small and isolated from established populations of RCWs, however, indicating that they may not be valuable habitat for RCW unless the connectivity between habitat patches is improved. A large proportion of the landscape that is not currently good habitat (≈440,000 ha or ≈50% of the landscape) has mineral soils that could potentially support pine-dominated, fire maintained vegetation communities (Frost and Costanza, unpublished), and slightly less than half of that area (46%,

220,000 ha) currently supports some type of pine-dominated forest (Southeast Gap Analysis

Project 2008). Key to the restoration of these areas for RCW habitat would be the reintroduction of periodic low-intensity fires (Van Lear et al. 2005, Brockway & Lewis 1997).

Our habitat model indicates that, in 2001, RCW clusters on Camp Lejeune had access to substantially more good quality habitat and had significantly larger social groups than those clusters located in Croatan National Forest. This may be somewhat explained by the relative abundance of peatland pocosin habitat and the relative scarcity of longleaf-pine promoting mineral soils in Croatan National Forest compared to Camp Lejeune. We also estimate that Camp Lejeune contains the largest area of potentially restorable publicly owned land on the Onslow Bight.

Our remote-sensing derived model does not directly measure the basal area and size distribution of pines or the amount of herbaceous groundcover present in forest stands— two criteria present in the USFWS recovery standard for RCWs. For this reason our model may have a limited ability to differentiate between mature longleaf pine woodlands, often considered the “gold standard” of RCW habitat, and mature loblolly pine plantations that have been recently thinned. These two forest types have similar vegetative structure, but the value of pine plantations as RCW habitat is unclear (U.S. Fish and Wildlife Service 2003).

The variation between LiDAR-derived and field measured canopy heights (RMSD 3.14m) is greater than that of many LiDAR-based habitat studies (Seavy et al. 2009), but still gives valuable information on canopy height over a broad spatial extent. Although the LiDAR data we used was collected in winter, we still were able to derive meaningful estimates of vegetation understory and midstory cover. Lower strata of pine forests in this region typically consist of a mixture of evergreen and deciduous broadleaf species, many of which retain senescent leaves until late winter (pers. obs.).

Our initial hypothesis, that both habitat structural characteristics and landscape attributes influence RCW group size, was supported by our data analysis. Regression results indicate that RCW social group size is strongly related to both vegetation structural characteristics near clusters, and their position in the larger landscape. These results are broadly consistent with the previous findings of Conner and Rudolph (1991), James et al.

(2001), and Walters et al. (2002) with a few key differences. In an analysis of habitat factors affecting group size for 47 RCW clusters in Apalachicola National Forest, James et al. (2001) found that mean group size was strongly and positively related to the density of large pines, wiregrass cover, and density of old-growth pines, and negatively correlated with the amount of woody vegetation within the "core stand” or forestry compartment containing the RCW cluster center. All relationships that were measured at a smaller scale (within 800m of the cluster center) were not strong, and regression results were not reported by James et al.

(2001). In a study of 30 RCW clusters near Fort Bragg in the North Carolina Sandhills,

Walters et al. (2002) found that mean group size was positively correlated with density of old

“flat-top” longleaf pines, and negatively correlated with the height of the hardwood midstory and the density of medium-sized pines 25.4-35.6cm dbh. These variables were measured within a large number of vegetation plots within the home range of a cluster, which had a kernel estimated mean area of 84 ha.

In contrast to these past studies, our analysis suggests that both habitat structural characteristics and landscape attributes influence the size of RCW social groups. The most robust predictor of group size was the density of groups on the landscape within four kilometers of a given cluster. Cluster density could influence group size via several possible mechanisms. First, a high density of clusters may increase the probability that a breeding female will be quickly replaced by a dispersing young female if she dies. This mechanism

was favored by Conner and Rudolph (1991) as an explanation for the influence of forest fragmentation on group size. Second, a high density of clusters in the surrounding landscape could contribute to the arrival of “floater” males that sometimes become helpers in territories other than their natal territory, increasing group size. Third, a high density of clusters may influence the dispersal behavior of fledglings. Birds that are fledged in landscapes crowded with existing clusters may be more likely to stay and become helpers in their natal territory compared to clusters in landscapes with few other active RCW territories.

This has previously been demonstrated for RCW populations in North Carolina by Pasinelli and Walters (2002).

Group size was also strongly related to the spatial variability of midstory cover within

200m of cluster centers. This relationship was correlated with, but stronger than, the relationship between group size and the absolute value of midstory cover. To our knowledge, no one has previously examined the influence of the spatial pattern of habitat structure on the success of RCWs, and our results suggest that these relationships merit further study. We also found a significant negative relationship between the average height of the forest canopy and the size of RCW groups. We suggest that this relationship can be explained by the fact that mature loblolly pine forest, often considered sub-optimal for

RCWs, is substantially taller than mature stands of longleaf pine, at least for trees on Camp

Lejeune for which height and age data are available. Collectively, our results are consistent with, and extend the findings of other investigators that have examined habitat requirements of RCWs: large RCW groups tend to be located in areas of high cluster density, and in pine forests with a uniformly open midstory and a low density of understory shrubs.

Although our analysis provides important information on the areas of the landscape that are most likely to support additional large groups of RCWs, further information is

needed in order to identify actions that are most likely to bring about recovery of RCW populations over the long term. Group size is an emergent property of RCW social organization that depends on rates of fecundity, dispersal, and mortality not only at the level of the individual cluster, but of the surrounding clusters as well. Spatially explicit population models (SEPMs, Dunning et al. 1995; Wiegand et al. 2004) that link habitat quality to a species’ vital rates and dispersal processes can best account for these interactions and show particular promise for guiding habitat conservation and restoration actions for RCWs

(Schiegg et al. 2002; Letcher et al. 1998; Bruggeman & Jones 2008).

Our approach of fusing remotely-sensed forest structure data and existing habitat standards with field-based indicators of a target-species’ is broadly applicable to modeling habitat for RCW throughout its range, as well as for other imperiled taxa that are sensitive to vegetation structural characteristics and habitat fragmentation such as the Spotted Owl (Strix occidentalis) and Greater Sage-grouse (Centrocercus urophasianus). Our general approach may become even more applicable as high-quality remote-sensing data, particularly LiDAR and hyperspectral imagery, becomes more widely available at lower cost. An increasing number of US states are using low-density, discrete-return LiDAR to produce terrain models over large spatial extents. Such data sets are (or soon will be) available for North Carolina,

Florida, Louisiana, Ohio, Maryland and New Jersey. Our analysis provides one strategy by which this and other data could be used to improve knowledge of habitat requirements for imperiled species and to prioritize areas of habitat for conservation and restoration.

Works Cited In Chapter II

Aldridge, C. L., & Boyce, M. S. 2007. Linking occurrence and fitness to persistence: habitat- based approach for endangered greater sage-grouse. Ecological Applications, 17(2), pp.508-526.

Atwood, T. C. 2006. The influence of habitat patch attributes on coyote group size and interaction in a fragmented landscape. Canadian Journal of Zoology 84:80–87.

Bivand, R. 2008. Applied spatial data analysis with R. Springer, New York.

Bjørnstad, O. N. & Falck, W. 2001. Nonparametric spatial covariance functions: estimation and testing. Environmental and Ecological Statistics, 8(1), pp.53-70.

Bradbury, R. B., Hill, R. A., Mason, D. C, Hinsley, S. A., Wilson, J. D., Balzter, H., Anderson, G. Q. A., Whittingham, M. J., Davenport, I. J. & Bellamy, P. E. 2005. Modeling relationships between birds and vegetation structure using airborne LiDAR data: a review with case studies from agricultural and woodland environments. Ibis, 147(3), pp.443-452.

Brockway, D. G. & Lewis, C. E. 1997. Long-term effects of dormant-season prescribed fire on plant community diversity, structure and productivity in a longleaf pine wiregrass ecosystem. Forest Ecology and Management, 96(1-2), pp.167-183.

Brotons, L., Thuiller, W., Araújo, M. B. & Hirzel, A. H. 2004. Presence-absence versus presence-only modelling methods for predicting bird habitat suitability. Ecography, 27(4), pp.437-448.

Bruggeman, D. J. & Jones, M. L. 2008. Should Habitat Trading Be Based on Mitigation Ratios Derived from Landscape Indices? A Model-Based Analysis of Compensatory Restoration Options for the Red-Cockaded Woodpecker. Environmental Management, 42(4), pp.591-602.

Burnham, K. P. & Anderson, D. R. 2002. Model selection and multimodel inference: a practical information-theoretic approach. Springer Verlag.

Clark, J. S. 2007. Models for ecological data: an introduction. Princeton University Press, Princeton.

Conner, R. N. & Rudolph, D. C. 1991. Forest Habitat Loss, Fragmentation, and Red- Cockaded Woodpecker Populations. The Wilson Bulletin 103(3), pp. 446-457.

Conner, R. N., Rudolph, D. C., & Walters, J. R.. 2001. The red-cockaded woodpecker: surviving in a fire-maintained ecosystem. University of Texas Press, Austin.

Conner, R. N., Saenz, D., Schaefer, R. R., McCormick, J. R., Rudolph, D. C. & Burt D. B. 2004. Group Size and Nest Success in Red-Cockaded Woodpeckers in the West Gulf Coastal Plain: Helpers Make a Difference. Journal of Field Ornithology, 75(1), pp.74-78.

Cozzi, G. G., Mueller, C. B. & Krauss, J. 2008. How do local habitat management and landscape structure at different spatial scales affect fritillary butterfly distribution on fragmented wetlands? Landscape Ecology, 23(3), pp.269-283.

Dudik, M., Phillips, S. J. & Schapire, R. E. 2009. Maximum Entropy Modeling of Species Geographic Distributions. Ecological Modelling, 190(1-2), pp.231-259.

Dudik, M. & Schapire, R. E. 2004. A maximum entropy approach to species distribution modeling. ACM International Conference Proceeding Series; Vol. 69. Proceedings of the twenty- first international conference on Machine learning, pp.655-662.

Dunning, J. B. J., Stewart, D. J., Danielson, B. J., Noon, B. R., Root, T. L., Lamberson, R. H. & Stevens, E. E. 1995. Spatially explicit population models: current forms and future uses. Ecological Applications, 5(1), pp. 3-11.

Elith, J. & Leathwick, J. R. 2009. Species distribution models: ecological explanation and prediction across space and time. Annual Review of Ecology, Evolution, and Systematics, 40, pp.677-697.

Falkowski, M. J., Evans, J. S., Martinuzzi, S., Gessler, P. E. & Hudak, A. T. 2009. Characterizing forest succession with lidar data: An evaluation for the Inland Northwest, USA. Remote Sensing of Environment, 113(5), pp. 946-956.

Geotechnical and Environmental Consultants. 2000. Forest Inventory Report, Red-Cockaded Woodpecker Foraging Habitat Inventory, Marine Corps Base Camp Lejeune, North Carolina. Unpublished report to Marine Corps Base Camp Lejeune; 7 p.

Goetz, S., Steinberg, D., Dubayah, R. & Blair, B. 2007. Laser remote sensing of canopy habitat heterogeneity as a predictor of bird species richness in an eastern temperate forest, USA. Remote Sensing of Environment, 108, pp.254-263.

Graf, R. F., Mathys, L. & Bollmann, K. 2009. Habitat assessment for forest dwelling species using LiDAR remote sensing: Capercaillie in the Alps. Forest Ecology and Management, 257(1), pp.160-167.

Guisan, A., Edwards, T. C. & Hastie, T. 2002. Generalized linear and generalized additive models in studies of species distributions: setting the scene. Ecological Modelling, 157(2- 3), pp.89-100.

Hagberg, A., Schult, D. & Swart, P. 2009. NetworkX, High Productivity Software for Complex Networks. https://networkx.lanl.gov.

Hawbaker, T. J., Keuler, N. S., Lesak, A. A., Gobakken, T., Contrucci, K. & Radeloff, V. C. 2009. Improved estimates of forest vegetation structure and biomass with a LiDAR- optimized sampling design. Journal of Geophysical Research, 114(3), G00E04.

Haynes, K., Dillemuth, F., Anderson, B., Hakes, A., Jackson, H., Jackson, S. E. & Cronin J. 2007. Landscape context outweighs local habitat quality in its effects on herbivore dispersal and distribution. Oecologia, 151(3), pp.431-441.

Hill, R. A., Hinsley, S. A., Gaveau, D. L. A. & Bellamy, P. E. 2004. Predicting habitat quality for Great Tits (Parus major) with airborne laser scanning data. International Journal of Remote Sensing, 25(22), pp.4851-4855.

James, F. C., Hess, C. A., Kicklighter, B. C. & Thum, R. A. 2001. Ecosystem management and the niche gestalt of the red-cockaded woodpecker in longleaf pine forests. Ecological Applications, 11(3), pp.854-870.

Khan, M. Z., McNabb, F. M. A., Walters, J. R. & Sharp, P. J. 2001. Patterns of Testosterone and Prolactin Concentrations and Reproductive Behavior of Helpers and Breeders in the Cooperatively Breeding Red-Cockaded Woodpecker (Picoides borealis). Hormones and Behavior,40, pp.1-13.

Knight, R. L. 1999. Private lands: the neglected geography. Conservation Biology 13(2), pp.223- 224.

Lear, D. H. V., Carroll, W. D., Kapeluck, P. R. & Johnson, R. 2005. History and restoration of the longleaf pine-grassland ecosystem: Implications for species at risk. Forest Ecology and Management, 211(1-2), pp.150-165.

Lefsky, M. A., Cohen, W. B., Parker, G. G. & Harding, D. J. 2002. Lidar Remote Sensing for Ecosystem Studies. Bioscience, 52(1), pp.19-30.

Lek, S., Delacoste, M., Baran, P., Dimopoulos, I., Lauga J. & Aulagnier, S.. 1996. Application of neural networks to modelling nonlinear relationships in ecology. Ecological Modelling 90(1), pp.39-52.

Lennartz, M., Hooper, R. & Harlow, R. 1987. Sociality and cooperative breeding of red- cockaded woodpeckers, Picoides borealis. Behavioral Ecology and Sociobiology, 20(2), pp.77- 88.

Letcher, B. H., Priddy, J. A., Walters, J. R. & Crowder, L. B. 1998. An individual-based, spatially-explicit simulation model of the population dynamics of the endangered red-cockaded woodpecker, Picoides borealis. Biological Conservation, 86(1), pp.1-14.

Maltamo, M., Packalén, P., Yu, X., Eerikäinen, K., Hyyppä, J. & Pitkänen, J. 2005. Identifying and quantifying structural characteristics of heterogeneous boreal forests using laser scanner data. Forest Ecology and Management 216 (1-3), pp.41-50.

McGaughey, R. J. 2009. FUSION/LDV: Sofware for LiDAR Data Analysis and Visualization. 2.70. http://www.fs.fed.us/eng/rsac/fusion/

Minor, E., D. Urban. 2007. Graph theory as a proxy for spatially explicit population models in conservation planning. Ecological Applications, 17(6), pp.1771-1782.

Morrison, M. L. 2006. Wildlife-habitat relationships: concepts and applications. Island Press, Washington.

NC Office of State Budget and Management. 2008. NC County/State Population Projections. December, 2008.

North Carolina Onslow Bight Conservation Forum. 2003. Memorandum of Understanding (unpublished).

Pasinelli, G. & Walters, J. R. 2002. Social and environmental factors affect natal dispersal and philopatry of male red-cockaded woodpeckers. Ecology, 83(8), pp.2229-2239.

Pettorelli, N., Gaillard, J., Laere, G. V. Duncan, P., Kjellander, P. Liberg, O., Delorme, D. & Maillard, D. 2002. Variations in adult body mass in roe deer: the effects of population density at birth and of habitat quality. Proceedings of the Royal Society B: Biological Sciences, 269(1492), pp.747-753.

Phillips, S. J., R. P. Anderson, & R. E. Schapire. 2006. Maximum entropy modeling of species geographic distributions. Ecological Modelling, 190(3-4), pp.231-259.

Phillips, S. J. & Dudik, M. 2008. Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography, 31(2), pp.161-175.

Pidgeon, A. M., Radeloff, V. C. & Mathews, N. E. 2006. Contrasting measures of fitness to classify habitat quality for the black-throated sparrow (Amphispiza bilineata). Biological Conservation, 132(2), pp.199-210.

Pinheiro, J. C. 2000. Mixed-effects models in S and S-PLUS. Springer, New York.

R Core Development Team. 2009. R: A language and environment for statistical computing. R Foundation for Statistical Computing: Vienna, Austria. http://www.r-project.org/

Riaño, D., Meier, E., Allgöwer, B., Chuvieco, E. & Ustin, S. L. 2003. Modeling airborne laser scanning data for the spatial generation of critical forest parameters in fire behavior modeling. Remote Sensing of Environment, 86, pp.177-186.

Robles, M. D., Flather, C. H., Stein, S. M., Nelson, M. D. & Cutko, A. 2008. The geography of private forests that support at-risk species in the conterminous United States. Frontiers in Ecology and the Environment, 6(6), pp.301-307.

Schiegg, K., Walters, J. R. & Priddy, J. A. 2002. The consequences of disrupted dispersal in fragmented red-cockaded woodpecker Picoides borealis populations. Journal of Animal Ecology, 71(4), pp.710-721.

Seavy, N. E., Viers, J. H. & Wood, J. K. 2009. Riparian bird response to vegetation structure: a multiscale analysis using LiDAR measurements of canopy height. Ecological Applications, 19(7), pp.1848-1857.

Southeast Gap Analysis Project. 2008. Southeast GAP Regional Land Cover 2001 (NC Subsection).

U.S. Fish and Wildlife Service. Recovery plan for the red-cockaded woodpecker (Picoides borealis): second revision. Atlanta, GA: U.S. Fish and Wildlife Service; 2003.

Van Lear, D. H., Carroll, W. D., Kapeluck, P. R., & Johnson, R., 2005. History and restoration of the longleaf pine-grassland ecosystem: Implications for species at risk. Forest Ecology and Management, 211(1-2), pp.150–165.

Vierling, K. T., Vierling, L. A., Gould, W. A., Martinuzzi, S., & Clawges, R. M., 2008. Lidar: shedding new light on habitat characterization and modeling. Frontiers in Ecology and the Environment, 6(2), pp.90–98.

Walters, J. R., Daniels, S. J. , Carter III, J. H., & Doerr, P. D., 2002. Defining quality of red- cockaded woodpecker foraging habitat based on habitat use and fitness. Journal of Wildlife Management, 66(4), pp.1064-1082.

Wiegand, T., Revilla, E. & Knauer, F. 2004. Dealing with uncertainty in spatially explicit population models. Biodiversity and Conservation, 13(1), pp.53-78.

CHAPTER III THE CONNECT TOOLBOX: GIS TOOLS SUPPORTING LANDSCAPE CONNECTIVITY FOR WILDLIFE Introduction

Movement between habitat patches promotes the persistence of imperiled wildlife species by buffering demographic fluctuations within habitat patches (Fahrig & Merriam

1985; Franken & Hik 2004; Proctor et al. 2005), allowing organisms to avoid inbreeding depression (Frankham 1998; Wright et al. 2007), and possibly by allowing species to track their environment in the face of climate change (Opdam & Wascher 2004; Heller & Zavaleta

2009; Knowlton & Graham 2010). The degree to which the landscape facilitates or impedes this movement is called landscape connectivity (Taylor et al. 1993). Over the past decade, landscape connectivity has escaped from the confines of the landscape ecology literature, and the promotion of connectivity is becoming a central principle in the design of nature reserves, and even municipal green spaces (Ewan et al. 2004). Landscape connectivity considerations are typically integrated into reserve designs by preserving or restoring corridors of natural vegetation between larger habitat patches (Haddad et al. 2003), or ensuring that habitat patches are spatially aggregated (Fraterrigo et al. 2009; Leidner &

Haddad 2011).

To effectively manage landscapes for connectivity, we must integrate connectivity considerations with the other important “dimensions” of habitat quality:

prevalence and fitness (Chapter 1). Moreover, we must also identify the features of landscapes that promote dispersal of multiple species of concern, and evaluate the effects of landscape change on connectivity. Here we describe a new set of GIS-based conservation planning tools that aim to facilitate the incorporation of landscape connectivity considerations into spatial conservation planning in a way that addresses the latter two concerns. Our tools allow users to leverage limited biological data to map dispersal corridors, measure the impact of landscape changes on connectivity, and prioritize parts of the landscape for connectivity conservation across multiple species. Our tools, which we call the

Connect Toolbox (http://www.unc.edu/depts/geog/lbe/Connect/) build off of existing free and open-source software, and are written as a geoprocessing toolbox for ArcGIS 9.3

(ESRI). First, we describe our conceptual approach to quantifying connectivity. Next we describe the overall workflow of the Connect Toolbox and the operation of the component tools. Finally we present two brief case studies that demonstrate how the tools can be used to answer important connectivity management questions for imperiled species.

Quantifying connectivity

Many different methods exist for quantifying landscape connectivity, but all of these methods are based on some model of the dispersal behavior of organisms, whether explicit or implicit. Individual-based simulations (DeAngelis & Mooij 2005) treat dispersal explicitly, representing the movement of large numbers of dispersing virtual organisms across simulated landscapes. In these models, the predicted flux of dispersing organisms provides a straightforward metric of connectivity. Individual-based models, however, are complex, computationally intensive, and have high data requirements (Grimm et al. 2005). The complexity of individual-based simulations has led to the development of simpler

approaches to quantifying connectivity where dispersal behavior is treated implicitly. The most popular approach of this type is to classify habitat into discrete patches, and then use network routing algorithms from graph theory (Urban & Keitt 2001; Urban et al. 2009) to identify the degree to which a particular patch of habitat facilitates dispersal within the larger network of habitat patches. Graph-based approaches are simple and computationally efficient; however, graph-theoretic metrics of connectivity are sensitive to the way that connections between habitat patches are defined (Minor & Urban 2008; Zetterberg et al.

2010).

The simplest approach to defining connections considers patches as either connected or disconnected, depending on a threshold of geographic distance (Figure 11a). This approach essentially assumes that dispersal is a simple step function of distance (Moilanen

2011). Features of the landscape that may present barriers to dispersal are not accounted for.

To overcome this limitation, other researchers (Chardon et al. 2003; Pinto & Keitt 2008;

Wang et al. 2009) define movement “costs” for different parts of the intervening landscape, and calculate the strength of connections between habitat patches by the distance of the path with the least cost. These least-cost path (LCP) models (Figure 11b) allow us to measure the strength of connections between habitat patches in ways that account for the influence of the intervening matrix on dispersal, however they also assume that organisms have complete knowledge of the landscape, and will always choose the path with the least cost (Sawyer et al.

2011). In LCP models, changes to the intervening matrix will not influence the cost of movement unless those changes alter the least-cost path (Figure 11b).

The inherent complexity of individual-based models, and the limitations of LCP models, leads us to adopt a different strategy for quantifying connectivity in the Connect

Toolbox. Our approach is based on circuit theory (McRae 2006; McRae et al. 2008; Figure

11c), in which dispersing organisms are taken to be analogous to electrical current flowing through a network of resistors. The resistors are represented by a raster dataset with different values depending on the resistance that landscape features pose to animal movement. This “resistance surface” is equivalent to the “cost surface” of LCP models, but instead of assuming that all organisms will follow the least-cost path, circuit-based models assume that animals will follow all of the paths available to them in proportion to the total resistance of those paths.

Figure 11: Three strategies for modeling landscape connectivity. A. Graph-based approach based on a threshold of geographic distance. Habitat patches are connected (black line) if their boundary is closer than a fixed threshold distance and disconnected if they are further away. B. Least-cost path model. The strength of connections between habitat patches varies with the resistance-weighted distance of the least-cost path between habitat patches. C. A circuit-based approach. Every patch is connected to every other patch with a connection that varies in strength according to the effective resistance between patches.

Circuit-based connectivity models provide two important ways of quantifying connectivity. First, because the density of electrical current in the network of resistors can be considered proportional to the flux of dispersing organisms assuming a random-walk (Figure

11c), circuit-based dispersal models allow us to quantify the connectivity value of arbitrary portions of the landscape, not just habitat patches themselves. Second, these models provide a way to calculate the effective resistance between pairs of habitat patches (McRae 2006).

Effective resistance, or the electrical resistance across the entire network of resistors, provides a metric of ecological distance that integrates the geographic distance between habitat patches, the resistance of the intervening landscape, and the number of alternative dispersal paths available. Network algorithms from graph theory can then be used to calculate summary statistics that represent how efficiently the overall network routes organisms between habitat patches. Because the strength of connections between habitat patches in circuit-based models depends on the structure of the entire intervening landscape, our strategy allows us to measure the influence of virtually any landscape change on connectivity, as long as it can be represented by changes in the “resistance” of the landscape.

Details of how circuit-base dispersal models are implemented in the Connect Toolbox are described in Section 3.

Toolbox Description

Overall Structure

The Connect Toolbox is a set of geoprocessing tools written in Python for ArcGIS

9.3., and provides the interface, modelbuilder, and scripting functionality of standard

ArcGIS toolboxes. The Connect Toolbox provides users with the ability to (i) create simple circuit theory-based models of animal dispersal using the Create Connectivity Model tool, (ii) combine dispersal models from multiple species in order to rank parts of the landscape for connectivity conservation using the Prioritize Landscape Features tool, and (iii) the ability to generate network-based summary statistics that represent measures of overall landscape connectivity using the Generate Landscape Network tool. (Figure 12).

Figure 12: The overall workflow of the Connect Toolbox. The Create Connectivity Model tool allows users to create simple models of wildlife dispersal from occurrence data, and “resistance maps” which provide information on how landscape features influence dispersal. The Prioritize Landscape Features tool allows users to combine these models to prioritize parts of the landscape according to their connectivity value across multiple species, optionally incorporating information on threats to and costs of conserving those landscape features. The Generate Landscape Network tool allows users to calculate overall measures of connectivity in order to compare land-use scenarios.

Create Connectivity Model

Most analyses using the Connect Toolbox will begin by creating models of animal dispersal using the Create Connectivity Model tool. This tool models animal movements using circuit theory, implemented by the Python package Circuitscape

(http://www.circuitscape.org). As mentioned in the previous section, circuit-based dispersal models assume that dispersing organisms are analogous to electrical current flowing over a landscape composed of conductors with various amounts of resistance, represented by a raster dataset. Finding appropriate resistance values for these landscape features is an active area of research and little consensus has emerged regarding the most appropriate way to do

this (Spear et al. 2010). In one of the first quantitative approaches, Ricketts (2001) used a likelihood-based approach to estimate resistance values based on a large butterfly mark- recapture dataset. More recently Trainor et al. (in preparation) used a combination of Maxent

(Phillips & Dudík 2008) and a discrete-choice model to estimate resistance for dispersing

Red-cockaded Woodpeckers. Where data is lacking, researchers have also used less rigorous approaches. For example Eycott et al. (2011) describe a process for synthesizing expert opinion into resistance values using a process called Delphi analysis.

Before the tool is run, users define “nodes” which represent the sources and destinations of dispersing organisms. To generate maps of the relative frequency of dispersal habitat use, the Create Connectivity Model tool iteratively connects one node to an arbitrary one-volt current, and connects all of the other nodes to the ground. Current flows from the focal node to all of the other nodes in proportion to the effective resistance between the node pairs according to Ohm's law. The density of current flowing across each resistor (a raster pixel) is recorded for each iteration of the model, and those current density maps are summed across all of the iterations of the model to produce a final current density map. The

Create Connectivity Model tool can also be used to calculate effective resistance between each pair of nodes. This information can be passed to the Generate Landscape Features tool for further analysis. When used in this way, the tool connects one node to the current source, and the other to the ground for each pair of nodes. Optionally, instead of connecting every node to every other node, users can define a minimum and maximum distance at which nodes should be connected.

Prioritize Landscape Features

The Prioritize Landscape Features tool combines outputs from dispersal models for multiple species into a single landscape prioritization that ranks pixels or parcels according to their value across all species using the Zonation algorithm (Moilanen et al. 2005). Because of the structure of circuit-based dispersal models, the maps of current density generated by the

Create Connectivity Model tool are essentially on an arbitrary scale determined by the number of nodes, the size of the landscape, and the voltage applied to each node. This poses a challenge if we wish to combine current density maps from different species together to find areas that are important for multi-species dispersal. The Zonation algorithm bypasses this limitation because it considers, not absolute values, but rather the proportion of the total distribution of dispersal habitat that is present in user-defined parts of the landscape.

The Zonation algorithm starts by ranking all pixels (or user-defined patches) at the edge of the map according to their “value.” Although the definition of value in Zonation is flexible, we define the “value” of cell i across all species j as

(2)

where Qj(S) is the proportion of the distribution of species j remaining in the set of all of the unranked patches S, wj is the weight given to species j, ti is the degree of threat

(probability of extirpation) assigned to patch i, and ci is the economic cost of protecting patch i. The algorithm then removes the least-valuable patch (the patch with the lowest delta) and re-evaluates all patches along the edges of the remaining cells, cells where one of its neighbors has already been removed.

Zonation provides an efficient way to find the set of conservation reserves that will maximize coverage of the distribution of a large set of organisms, a well-known reserve

design optimization problem (Cabeza & Moilanen 2001). Zonation also provides us with the opportunity to incorporate information on conservation costs, development threats, and uncertainty in the underlying dispersal maps. While information on conservation costs and development threats are incorporated directly into the assessment of patch “value” (delta in equation 1 above), uncertainty is incorporated by penalizing the inputs according to their degree of uncertainty (e.g. taking the lower confidence interval of the species-specific input maps). This “distribution discounting” approach (Moilanen et al. 2006) has been shown to produce robust-optimal results in the face of uncertainty in maps of species distributions.

The output from the Prioritize Landscape Features tool is a raster dataset with parts of the landscape ranked from highest to lowest by their overall value for multi-species connectivity. By default, the algorithm ranks every raster cell for which there is species data, but users can also input a map that aggregates cells into arbitrary patches. This feature is useful if users wish to rank management units, land parcels or other arbitrary parts of the landscape

Generate Landscape Network

The Generate Landscape Network tool uses network routing algorithms from graph theory (Minor & Urban 2008), implemented in the Python package NetworkX (Hagberg et al. 2009), to measure the overall connectedness of individual habitat patches and evaluate the connectivity of landscapes for individual species under alternative scenarios. The tool connects each habitat patch to every other patch with a connection (called an edge) that is weighted according to the effective resistance between those two nodes (estimated using the

Create Connectivity Model tool). The tool then calculates the path of least resistance that connects all of the nodes together. This path, called the Minimum Spanning Tree, represents

potentially important connections between habitat patches, the hypothetical “backbone” of the habitat network (Minor & Urban 2008).

For each habitat patch in the network, the tool calculates several measures of the connectedness of that patch. By default, the tool creates a “complete graph” which connects every habitat patch to every other patch. This poses a challenge for traditional methods of calculating the centrality of nodes, because the path of least resistance between two patches never passes through another patch. To overcome this limitation, our tool calculates the betweenness centrality of nodes along the minimum spanning tree. This metric represents how central each patch is in the backbone of the network. We also calculate the current flow centrality of each node (Brandes & Fleischer 2005) which represents the proportion of an arbitrary total current that flows through each node.

The tool can also calculate summary statistics for the entire network that can be used to compare habitat management or development scenarios. Although a wide variety of summary statistics are available, we chose to incorporate two of them into the output of the

Generate Landscape Network tool. The first, the total resistance of the minimum spanning tree, represents the overall traversability of the network's backbone. The second, the average resistance of all of the connections in the network (often called the Characteristic Path

Length) is a more sensitive measure of connectivity because it weights central and peripheral connections equally. It is important to note that the total resistance of the minimum spanning tree is only comparable between two networks if the number of patches and configuration of connections is identical. If habitat patches are added or removed between scenarios, then comparing this metric between the two networks is not meaningful. This is less of a concern for the Characteristic Path Length.

Case Study Application

We developed the Connect Toolbox in close partnership with the North Carolina

Sandhills Conservation Partnership (NCSCP, http://www.ncscp.org), a group of regional stakeholders that aim to conserve the longleaf pine ecosystem, a mosaic of fire-maintained communities endemic to the coastal plain of the Southeastern US (Van Lear et al. 2005).

Working with our NCSCP partners, we identified two management questions that were regionally relevant and also provide a good demonstration of the range of types of management questions we have designed the Connect Toolbox to answer. In this section we provide a brief description of these two management questions, the process that we used to answer them, and their regional implications for conserving connectivity for our target wildlife species.

Study Area: Spring Lake, NC

The town of Spring Lake, NC (Figure 13), is a rapidly urbanizing community of approximately 8000 people situated on the eastern edge of the Sandhills physiographic province. Much of the surrounding lands are part of Fort Bragg and Pope Air Force Base.

Military training activities, and more recently, prescribed burning, have maintained a frequent, low-intensity fire regime on much of Fort Bragg. Fire-maintained ecological communities in the region, including longleaf pine (Pinus palustris) savannahs and flatwoods, and associated herbaceous wetlands, support numerous animal species of conservation concern. Expansion of military infrastructure and related urban growth in Spring Lake and the northern margin of Fayetteville poses a potential threat to these species by decreasing the connectivity between existing populations, reducing opportunities for population expansion into newly restored areas, and limiting the ability of land managers to use prescribed fire as a

management tool near the wildland-urban interface (Costanza & Moody 2011). These factors have led the Department of Defense to partner with conservation organizations and the

State of North Carolina to establish a network of protected areas near existing military lands.

In 2005 the State of North Carolina authorized the creation of the new Carvers Creek State

Park, which sought to connect the former Rockefeller estate at Long Valley Farm to existing protected lands along the Cape Fear River. Planning and land-acquisition for the new park is an ongoing partnership between the State of North Carolina, The Nature Conservancy, and the Department of Defense. The creation of the new park has the potential to enhance the recreational value of the landscape and to serve as an important dispersal corridor for multiple imperiled wildlife species.

Figure 13: Major lands managed for wildlife in the study area circa 2010. Data: Department of Defense, North Carolina Department of Transportation, ESRI. 78

Focal species

For this case-study, we focus on three taxa of particular conservation concern

(Figure 14). The federally endangered Red-cockaded Woodpecker (Picoides borealis, RCW) is a cooperatively breeding, cavity-nesting bird endemic to mature fire-maintained pine woodlands in the southeastern US. This species requires open-canopy pine-dominated environments for breeding, foraging, and juvenile dispersal (Trainor et al., in preparation).

The NC Sandhills population of RCWs is the largest in the nation (US Fish and Wildlife

Service 2003). Fledgling RCWs in this population have been observed to disperse up to 20 km, although the majority select breeding territories less than 6km from where they were born. A radio-tracking study performed from 2006-2007 revealed that dispersing fledglings avoid open areas and fire-suppressed forests with a high density of hardwoods and understory trees (Trainor et al. in preparation) Another target species, the Saint-Francis Saytr

(Neonympha mitchellii ssp. franciscii, SFS), is one of the world's most narrowly distributed butterflies. Also a Federally Endangered species, the known distribution of SFS is restricted to herbaceous wetlands on Fort Bragg. SFS is suspected to have a metapopulation structure

(Kuefler et al. 2010), as current subpopulations are found near the margins of temporary ponds maintained by beaver (Castor canadensis). Dispersal between local subpopulations is thought to be rare, and release experiments performed with a surrogate species, the

Appalachian Brown Saytr (Moody et al. 2011) suggest that open fields and upland forests present relatively impermeable barriers to dispersing individuals. Our third target species, the

Eastern Tiger Salamander (Ambystoma tigrinum), breeds in depressional wetlands and other water bodies free of fish such as borrow-pits (Madison & Farrand III 1998). Adults are typically found in forested upland environments, and have been found as far as three

kilometers from their natal pond (W. Fields, personal communication). Our three target organisms have substantially different life-histories and dispersal behaviors, and are found in very different habitats within the broader longleaf-pine ecosystem. This makes the connectivity of landscapes for these three species, when taken together, potentially a suitable proxy for the connectivity of other organisms dependent on the longleaf pine ecosystem in this landscape.

Figure 14: Focal species. Not to scale. Photos modified from originals by the US Fish and Wildlife Service.

Management Question 1: Where is the best connectivity bang for the buck? Our NCSCP partners wanted to identify private lands that contribute the most to landscape connectivity for multiple wildlife species in longleaf pine ecosystems surrounding

Fort Bragg. These lands could be targeted for acquisition for future nature reserves such as

Carvers Creek State Park, conservation easements, or other voluntary habitat conservation programs such as safe-harbor (Wilcove & Lee 2004). To identify priority parcels for the conservation of landscape connectivity, we used the Create Connectivity Model tool to create circuit-based dispersal models for each of our target species, the Red-cockaded

Woodpecker, Saint Francis Satyr, and Eastern Tiger Salamander. We then used the reserve- design algorithm in the Prioritize Landscape Features tool to combine these maps of the relative frequency of dispersal habitat use with data on potential acquisition costs properties, and estimates of the probability of urban development derived from an urban growth simulation to find the set of private lands that could do the most to conserve connectivity for the least cost. We chose to incorporate data on economic costs and development threats into our assessment because both have recently emerged as key considerations in systematic conservation planning theory. When land acquisition or management costs differ drastically between different parts of the landscape the most cost-effective portfolio of protected lands can be dramatically less costly than a similar portfolio that considers only the biological value of the landscape (Naidoo et al. 2006). Similarly, because rates of land conversion, the greatest threat to connectivity, also differ drastically along the urban-to-rural gradient, conservation actions should target the parcels where both the connectivity value and degree of threat is high (Bode et al. 2008).

To identify private lands with high connectivity value across all of our target species, we first created a map of current and potential habitat patches for each species. In the dispersal models, these represented the potential sources and destinations of dispersing organisms. For tiger salamanders and SFS, breeding habitats are small enough relative to the size of the landscape (<100ha) that they could be effectively represented as point features.

Breeding areas for RCW family groups are also small (less than ~60ha) and we represented

the sources and destinations of dispersing birds as points located at the geographic center of the breeding territories. For each species, we mapped both existing habitats and places thought to be highly suitable for the species, even if their occupancy status was negative or unknown. For RCW, we included all mapped territories, including both active and inactive, that were located between 1988 and 2008, including areas on Ft. Bragg identified as potential sites for additional territories. For SFS, and Tiger Salamander, we included all known or historic occurrences of the species that could be located with high accuracy (less than +-

50m), and also environmentally suitable habitats initially identified using a species distribution model (Maxent) and then confirmed by field visits or by interpretation of high- resolution aerial photography (Fields & Simon 2009). Potential habitat was included in the dispersal models because of uncertainty surrounding the occupancy status of habitats (in the case of Tiger Salamander), the shifting temporary nature of habitats (in SFS) and year-to-year turnover in the occupancy status of territories (in RCW).

To create resistance maps, we used a different approach for each species depending on the data that was available. For SFS and Tiger Salamander, we estimated the resistance of four broad habitat types (herbaceous wetland, forested wetland, upland forest, and unforested) by synthesizing the results of mark-recapture, radiotracking, and release experiments (described in detail in Moody et al. 2011). Observed entry probabilities, distributions of jump lengths, and turn angle distributions were first incorporated into an individually-based, biased random walk model for each species. We then calculated habitat- specific displacement rates (Kareiva & Shigesada 1983; Kuefler et al. 2010) and multiplied these by the observed probability of entering each habitat to generate an estimate of the resistance of each habitat type. More specifically, the resistance, θ, of each habitat type j was taken to be

(3)

where pj is the average probability of entry into a particular habitat and and dj is the mean squared displacement rate of the organism dispersing in habitat j. These estimates were

then rescaled so that the most resistant habitat had a resistance of 100. We then used landcover information from the 2006 National Landcover Database (NLCD, Xian et al. 2009) to

create a map of these broad habitat types for our study area at a 30m pixel resolution.

For RCWs, we estimated landscape resistance by training a species distribution model on points representing the locations of dispersing female fledgling RCWs (n=33) that were radio-tracked on Ft. Bragg in 2006 and 2007 (detailed in Trainor et al. in preparation).

The model was trained with forest structural information (including canopy height, and forest density) from a statewide LiDAR dataset from the NC Floodplain Mapping Program.

Results from the species distribution model were then rescaled by a variety of exponents and the best model was selected by regressing the resulting functional distances against records of actual dispersals from a long-term banding dataset using a discrete-choice model (Cooper

& Millspaugh 1999, Trainor et al. in preparation). Table 8 shows the relative resistances used in the dispersal models for each species summarized by NLCD landcover type. After assembling the inputs, we created dispersal models for each species using the Create

Connectivity Model tool (Figure 15a). To incorporate possible low-frequency, long-distance dispersal, the models were created by connecting each potential habitat patch to every other habitat patch.

Potential land acquisition costs were estimated by calculating the per-hectare tax value of privately-owned properties in the study region, which spans parts of Cumberland,

Harnett, Moore, and Hoke Counties (n=2,273, Figure 15b). Where available, estimates included the value of existing structures. Tax values ranged from less than $2,000 to greater 83

than $600,000 per hectare with a median value of $11,390. Tax values are an imperfect metric of market values for real property and probably substantially underestimate market rates, however tax values do reflect overall patterns in land-use along the urban-rural gradient in the region.

Table 8: Resistance values of different landcover types used to create dispersal models for each species. RCW: Red-cockaded Woodpecker, SFS: Saint Francis Saytr, TS: Tiger Salamander. Relative Resistance

NLCD Cover Description RCW SFS TS Class 11 Open Water 70.8885 100 100 21 Developed, Open Space 16.7197 100 100 22 Developed, Low Intensity 18.899 100 100 23 Developed, Medium Intensity 47.7798 100 100 24 Developed, High Intensity 57.61 100 100 31 Barren Land 75.5272 100 100 41 Deciduous Forest 23.3684 6.1296 13.2252 42 Evergreen Forest 9.7849 6.1296 13.2252 43 Mixed Forest 23.7445 6.1296 13.2252 52 Scrub / Shrub 41.2455 100 100 71 Grassland / Herbaceous 23.9802 100 100 81 Pasture / Hay 57.3331 100 100 82 Row Crops 74.7855 100 100 90 Woody Wetlands 35.2782 16.0298 26.4219 95 Emergent Herbaceous Wetland 39.9124 9.6511 26.4219

We estimated development threats using a cellular automaton urban-growth simulation (SLEUTH, Jantz et al. 2010) developed as part of the USGS Southeast Regional

Assessment Project (Terrando et al., in preparation). The SLEUTH model was trained on a time-series of spatial road-density data from 1988 - 2009. Spatial spread parameters were estimated for the US Census Bureau Combined Statistical Area encompassing the study region. The model output represents a 60m resolution, grid-based probability of urban development derived from 100 monte-carlo iterations of the model. To incorporate this data into the parcel-level analysis, we computed the average probability of urban development for

each parcel for the model year 2100 (Figure 15c). We took the resulting value to represent the overall probability that the parcel would be converted to urban land-use, which is relatively impermeable to the movement of all three of our study taxa.

Finally, we used the Prioritize Landscape Features tool to rank parcels according to their cost and threat-weighted connectivity value across all species. In this run of the tool, each species was weighted equally, and lands currently managed for conservation were not removed until the end of the run, which biases the rankings slightly in favor of lands that are adjacent to currently protected areas. Our results indicate that a variety of private lands could contribute to the conservation of landscape connectivity in the study area. Parcels given the highest ranks were areas that harbor potential dispersal habitat for St. Francis Saytr to the

West of the Overhills area along Buffalo and Duncan creeks, and areas that potentially connect the northeast section of Ft. Bragg to the Long Valley Farm section of Carver's

Creek State Park. Although tax values in this part of the landscape are relatively high, these parcels are among the most important on the landscape for conserving connectivity for

RCW. A combination of dispersal value for RCW, relatively low tax values, and high development threats, contribute to high ranks for parcels to the north of Long Valley Farm and in the vicinity of Buffalo Lakes. Overall, only a few areas are important for conserving connectivity of more than one species (Figure 15a).

Figure 15: A. RGB composite of predicted relative frequency of dispersal habitat use for SFS (Red), RCW (Green), and Tiger Salamander (Blue) on private lands that have no permanent legal protection for wildlife. B. Tax value of privately owned parcels in the study area in 2008. C. Per-parcel mean predicted probability of urban development in 2100 from the SLEUTH-3r urban growth model. D. Connectivity conservation rank for parcels greater than 1ha in size from the Prioritize Landscape Features tool incorporating information from maps A – C.

Management Question 2: Will a corridor be enough?

Our partners were also interested in using the Connect Toolbox to measure the connectivity consequences of planned habitat conservation and restoration efforts. Chief among these efforts is the ongoing development of Carver's Creek State Park, established by the NC Legislature in 2005 with the prospect of the donation of Long Valley Farm and the

Sandhills Properties from the Nature Conservancy . The park is currently ~1600 ha in size, but future land acquisitions could boost that to more than 3400 ha. The NC Division of

Parks and Recreation released a draft plan for public comment in July of 2011 (N.C.

Division of Parks and Recreation 2011). Park managers and other local stakeholders were interested in whether habitat conservation and restoration in the Park could improve connectivity for longleaf pine-associated wildlife. In particular, ongoing and planned longleaf pine restoration in the park has the potential to increase dispersal between existing RCW sub-populations in the Overhills and Northeast sections of Ft. Bragg, which are currently relatively isolated (Trainor et al., in preparation).

The park will develop in the context of potentially rapid land-use change due to the expansion of military facilities and associated civilian infrastructure. A 2008 land-use study

(Dougherty 2008) estimated that municipalities adjacent to Ft. Bragg would be absorbing an additional 3,300 people per year in the period from 2005 - 2013, and population growth in greater Fayetteville has been among the most rapid in North Carolina over the past decade

(an increase of 66% from 2000 - 2010). Along with our partners, we wanted answer two questions regarding the long-term development of Carver's Creek State Park in the context of urbanization: (1) If the park develops according to the Draft Master Plan, will it serve as an effective dispersal corridor for RCWs? and (2) Could the increase in habitat connectivity from longleaf pine restoration in the Park offset losses in connectivity anticipated with future urban development?

To answer these two questions using the Connect Toolbox, we first created a baseline model of RCW dispersal in the current landscape using the Create Connectivity

Model Tool. This is the same model we used for RCW dispersal in Management Question 1.

Next, we modified this baseline model to represent two different landscape scenarios, circa the year 2100. In one scenario, we created additional dispersal barriers in areas that are anticipated to experience urban development. In another scenario, we augmented the urban growth with aggressive longleaf pine restoration in suitable areas that are targeted for

incorporation into Carvers Creek State Park. For each scenario we used the Generate

Landscape Network Tool to calculate overall metrics of landscape connectivity for dispersing RCWs, using the resistance surface developed in the first case study. By comparing connectivity metrics for the two landscape change scenarios to metrics from the baseline scenario, we were able to measure the relative influence of possible future urban growth and habitat restoration on landscape connectivity for RCWs.

To build plausible urban development scenarios, we built off of parcel-level probabilities of urban development from the SLEUTH-3r urban growth model previously described in Management Question 1. In order to use this information to modify the resistance surface, we treated the process of urban development for each property parcel in the landscape as a single Bernoulli trial (equivalent to a weighted coin-flip) with a probability taken from the SLEUTH-3r model. We used this procedure to create a series of discrete realizations of the model that differ in the parcel-by-parcel spatial pattern of urban development, but have the overall amount and general distribution of development predicted by SLEUTH-3r. For those parcels that are “developed” in each realization of the model, we then replaced currently forested pixels in the resistance surface with pixels from a random surface that replicates the distributional properties (mean and range) as well as the pattern of spatial dependence (Figure 16) of the resistance surface in areas that are urban in the baseline scenario. To generate the random values based on the variogram model in in

Figure 16, we used the GRASS GIS program r.random.surface (Ehlschlaeger & Goodchild

1994).

To build the development and restoration scenarios, we modified the resistance maps from the urban development scenario by replacing the resistance values in upland forested areas within the draft Master Plan boundary for Carvers Creek State Park with

random values that approximate the mean and distribution of good quality habitat within active RCW breeding territories on Ft. Bragg, which have a median resistance of 2.77.

Figure 16: Fitted exponential variogram model for the RCW resistance surface in urban areas. Nugget: 0.197 Exponent: 64.311 Sill: 1.272 Bandwidth: 20m.

Figure 17: RCW resistance surfaces for the baseline scenario (A), and one of five replicate resistance surfaces generated for the development scenario (B) and the development and restoration scenario (C).

In this scenario, resistance values for planned park facilities such as visitors centers, parking areas, and maintained open areas, as well as resistance values for wetlands and lowland hardwood forests, were not modified (Figure 17).

For all three scenarios, we used the centers of both active and inactive territories as well as habitat partitions identified as possible sites for RCW reintroduction on Ft. Bragg as the sources and destinations of dispersing organisms. Because we are not incorporating population dynamics into our model, there is significant uncertainty about the future occupancy status of RCW territories, and we wanted to account for possible dispersal between all potential habitats in our long-term projection of connectivity under urban growth. Further, the graph-based connectivity metrics that we compute for each scenario are sensitive to changes in the topology of the habitat patches on the landscape. We therefore used the same map of RCW territories for each scenario in order to make the model scenarios comparable.

Our results indicate that urban growth will substantially reduce dispersal of RCWs through private lands in the study area over the next decades. Current density, which is proportional to the density of visitation by dispersing organisms, decreases by up to 96% on private lands impacted by urban growth (Figure 18). Without additional conservation or restoration of dispersal habitat, we predict that the largest absolute decreases in dispersal will occur to the north of the current Carvers Creek Study area, and north of the Overhills area on Ft. Bragg. The model also predicts that dispersal will become more concentrated on lands currently managed for conservation, especially in the vicinity of the Green Belt around the

Cantonment Area on Ft. Bragg, although increases in dispersal habitat use on managed lands are small. The development and restoration scenario indicates that dispersing juvenile RCWs are likely to move through the future Carver's Creek State Park if upland habitats are

restored to longleaf pine habitat of comparable quality to RCW breeding habitat on Ft.

Bragg (Figure 18). Because urban growth makes much of the surrounding landscape relatively impermeable to dispersal, our model results indicate that dispersing birds are

“funneled” through the Carver's Creek corridor in the development and restoration scenario.

Indeed, the magnitude of current density through the modeled corridor in this scenario is comparable to the most frequently-used dispersal habitats on currently managed lands.

When we calculated summary statistics for each model scenario using the Generate

Landscape Network tool, we found that anticipated urban growth would substantially decrease overall landscape connectivity for RCWs. Two key measures of the overall connectivity of the landscape, the resistance of the minimum spanning tree, and the characteristic path length, showed substantial decreases in connectivity with urban growth

(Figure 19). This conclusion appeared to be robust to our assumptions about the exact pattern of urban development. We also found that the creation of a dispersal corridor in the

Carvers Creek State Park master plan area would substantially, but not totally, ameliorate the negative effects of urban development on landscape connectivity for RCWs over the next century. Although the creation of a corridor lowers the effective resistance of the landscape between RCW territories near either end of the corridor to below that of the baseline scenario, this route represents only a small set of the possible paths for dispersing birds across the landscape. Because our connectivity model assumes that dispersing organisms travel along all possible paths in proportion to their resistance (McRae et al. 2008), widespread urban development causes the landscape to become less traversable overall regardless of the existence of the corridor.

Figure 18: A. Relative frequency of RCW dispersal habitat use for the baseline scenario. B. Average change from the baseline scenario with 2100 urban development. C. Average change from the baseline scenario with 2100 urban development and aggressive habitat restoration in the draft Carver’s Creek State Park boundary. Maps B and C represent the average change from five replicate runs with different patterns of urban development.

Overall, our assessment is that Carvers Creek State Park could serve as an effective dispersal corridor for RCWs, and that longleaf pine restoration in the park could substantially contribute to the recovery of this endangered species. Our assessment, however, hinges on several strong assumptions. The first is that the future shape and extent of the park will reflect the draft Master Plan. The actual extent of the new park is constrained by negotiations with individual landowners and the appropriation of funds for acquisition and management of the new lands. Moreover, longleaf pine restoration practices that incorporate prescribed fire are difficult to implement in areas near residential

development along the wildland-urban interface (Costanza et al. 2011). Although our simulated restoration scenario avoided areas less than 50m from existing structures and property boundaries without firebreaks, some additional areas may be impractical to manage with prescribed fire if development encroaches further, or guidelines for smoke management become more restrictive. Our analysis also relies on the common assumption that urbanization processes and aggregate trends many decades in the future will resemble those experienced in the recent past, an assumption that is almost certainly incorrect.

Figure 19: Summary statistics representing changes in overall landscape connectivity for RCWs in the two landscape scenarios relative to the baseline. Individual dots represent results from five replicate runs with different reasonable patterns of urban development.

Conclusion

Landscape connectivity's recent rise in the academic literature has given birth to a large number of approaches to model it, measure it and map it. In light of the large number of approaches available, we believe that there is a need for relatively simple tools that allow people to incorporate landscape connectivity considerations into land-use and infrastructure planning. This is especially true on and around public lands, where the managing agencies are often required to incorporate assessments of potential impacts to wildlife on any major change in the use of public lands and waters. These assessments frequently evaluate changes

in habitat area, but relatively few consider impacts to the connectivity between habitat patches (Marcot et al. 2001). We demonstrate that the Connect Toolbox can be used to assess the connectivity consequences of landscape change, including alternative development and habitat restoration scenarios. Our case studies also demonstrate that the Connect

Toolbox can utilize readily available types of biological information to assess which parts of landscapes can best contribute to conserving landscape connectivity for multiple species, and identify those areas that can be conserved or managed at the least cost. The Connect

Toolbox provides conservation stakeholders with a platform with which to make data- informed decisions about where to focus management efforts, and where (not) to site infrastructure in order to conserve landscape connectivity.

Although we anticipate that the Connect Toolbox can be applied to a wide variety of problems at a variety of spatial scales, our approach has limitations that make it more applicable to particular types of problems, and less applicable to others. Currently, the most important limitation of the Connect Toolbox is that the dispersal modeling framework that we use, based on circuit theory, cannot incorporate directional connectivity. This is because the effective resistance between a current source and the ground over a network of resistors is always the same if the source and ground are reversed (McRae et al. 2008). This constraint limits the ability of the Connect Toolbox to adequately represent dispersal in highly directional systems, such as stream networks, and possibly its ability to represent directional migration of species under changing climates.

The Connect Toolbox is currently written as a set of geoprocessing tools for ArcGIS

9.3, and works with the version of python (2.5.x) that is “hard-wired” to this version of

ArcGIS. Because a 64-bit variant of this Python version is not available, memory currently limits the Create Connectivity Model tool to landscapes less than approximately 2-million

cells. This limitation is not present in the Python bundled with AcGis 10.x (2.6.x), and future releases of the Connect Toolbox will allow users to model connectivity on larger landscapes.

For large landscapes with many species, several of the tools may have long running times.

The running time of the Create Connectivity Model tool, in particular, is sensitive to the size of the landscape and the number of habitat nodes considered for each species. When calculating pairwise resistances between each node, the number of computations c increases with the number of nodes n according to c = n(n - 1) / 2 as long as a traversable path exists from every node to every other node. When mapping corridors, computation time increases non-linearly with the number of nodes, as each computation takes substantially longer to complete. We have run analyses with 1000 nodes on landscapes up to 2 million cells, but this calculation took approximately three days to complete on a modern desktop computer.

Reducing the area of analysis, increasing the cell size, and aggregating nodes into larger habitat patches surrounded by Short-Circuit regions are all legitimate ways to increase the speed of the calculations. The Prioritize Landscape Features tool can run on grids substantially larger than those permitted by the Create Connectivity Model tool, however the calculations may still be slow for large landscapes if a large number of species are considered.

Works Cited in Chapter III

Bode, M. et al., 2008. Cost-effective global conservation spending is robust to taxonomic group. Proceedings of the National Academy of Sciences, 105(17), p.6498.

Brandes, U. & Fleischer, D., 2005. Centrality measures based on current flow. STACS 2005, pp.533–544.

Cabeza, M. & Moilanen, A., 2001. Design of reserve networks and the persistence of biodiversity. Trends in Ecology & Evolution, 16(5), pp.242–248.

Chardon, J.P., Adriaensen, F. & Matthysen, E., 2003. Incorporating landscape elements into a connectivity measure: a case study for the Speckled wood butterfly (Pararge aegeria L.). Landscape Ecology, 18(6), pp.561–573.

Cooper, A.B. & Millspaugh, J.J., 1999. The application of discrete choice models to wildlife resource selection studies. Ecology, 80(2), pp.566–575.

Costanza, J.K. & Moody, A., 2011. Deciding where to burn: Stakeholder priorities for prescribed burning of a fire-dependent ecosystem. Ecology and Society, 16(1), p.14.

DeAngelis, D.L. & Mooij, W.M., 2005. Individual-Based Modeling of Ecological and Evolutionary Processes. Annual Review of Ecology, Evolution, and Systematics, 36, pp.147- 168.

Dougherty, J., 2008. Ft. Bragg / Pope Air Force Base Joint Land-use Study Update. Report from the Regional Land Use Advisory Commission. http://www.rluac.com/images/Projects/2008%20JLUS%20Update.pdf

Ehlschlaeger, C.R. & Goodchild, M.F., 1994. Dealing with uncertainty in categorical coverage maps: defining, visualizing, and managing errors. In Proceedings, second Associations for Computing Machinery Workshop on Advances in Geographic Information Systems. National Institute of Standards and Technology, Gaithersburg, MD, December. pp. 101–106.

Ewan, J., Fish Ewan, R. & Burke, J., 2004. Building ecology into the planning continuum: case study of desert land preservation in Phoenix, Arizona (USA). Landscape and urban planning, 68(1), pp.53–75.

Eycott, A.E., Marzano, M. & Watts, K., 2011. Filling evidence gaps with expert opinion: The use of Delphi analysis in least-cost modelling of functional connectivity. Landscape and Urban Planning, 103(3-4), pp.400-409.

Fahrig, L. & Merriam, G., 1985. Habitat patch connectivity and population survival. Ecology, pp.1762–1768.

Fields, W. & Simon, M.C., 2009. COS 28-9: Predicting the locations of breeding pools to assess landscape connectivity for a rare amphibian. In The 94th ESA Annual Meeting.

Franken, R. & Hik, D., 2004. Influence of habitat quality, patch size and connectivity on colonization and extinction dynamics of collared pikas Ochotona collaris. Journal of Animal Ecology, 73(5), pp.889-896.

Frankham, R., 1998. Inbreeding and extinction: island populations. Conservation Biology, 12(3), pp.665–675.

Fraterrigo, J.M., Pearson, S.M. & Turner, M.G., 2009. Joint effects of habitat configuration and temporal stochasticity on population dynamics. Landscape Ecology, 24(7), pp.863– 877.

Grimm, V. et al., 2005. Pattern-Oriented Modeling of Agent-Based Complex Systems: Lessons from Ecology. Science, 310(5750), pp.987-991.

Haddad, N.M. et al., 2003. Corridor use by diverse taxa. Ecology, 84(3), pp.609–615.

Hagberg, A., Schult, D. & Swart, P., NetworkX, High Productivity Software for Complex Networks. https://networkx.lanl.gov.

Heller, N.E. & Zavaleta, E.S., 2009. Biodiversity management in the face of climate change: a review of 22 years of recommendations. Biological conservation, 142(1), pp.14–32.

Jantz, C.A. et al., 2010. Designing and implementing a regional urban modeling system using the SLEUTH cellular urban model. Computers, Environment and Urban Systems, 34(1), pp.1–16.

Kareiva, P. M., & Shigesada, N. 1983. Analyzing insect movement as a correlated random walk. Oecologia, 56(2), pp.234–238.

Knowlton, J.L. & Graham, C.H., 2010. Using behavioral landscape ecology to predict species’ responses to land-use and climate change. Biological Conservation, 143(6), pp.1342-1354.

Kuefler, D. et al., 2010. The conflicting role of matrix habitats as conduits and barriers for dispersal. Ecology, 91, pp.944-950.

Leidner, A.K. & Haddad, Nick M, 2011. Combining Measures of Dispersal to Identify Conservation Strategies in Fragmented Landscapes. Conservation Biology, 25(5), pp.1022-1031.

Madison, D.M. & Farrand III, L., 1998. Habitat use during breeding and emigration in radio- implanted tiger salamanders, Ambystoma tigrinum. Copeia, pp.402–410.

Marcot, B.G. et al., 2001. Using Bayesian belief networks to evaluate fish and wildlife population viability under land management alternatives from an environmental impact statement. Forest Ecology and Management, 153(1-3), pp.29-42.

McRae, B. H, 2006. Isolation by resistance. Evolution, pp.1551–1561.

McRae, B. H, Dickson, B. G, et al., 2008. Using circuit theory to model connectivity in ecology, evolution, and conservation. Ecology, 89(10), pp.2712–2724.

Minor, E. S & Urban, D. L, 2008. A graph-theory framework for evaluating landscape connectivity and conservation planning. Conservation Biology, 22(2), pp.297–307.

Moilanen, Atte, 2011. On the limitations of graph-theoretic connectivity in spatial ecology and conservation. Journal of Applied Ecology, 48(6), pp.1543-1547.

Moody, A. et al., 2011. Final Report for Project RC-1471: Mapping Habitat Connectivity for Multiple Rare, Threatened, and Endangered Species on and Around Military Installations. Strategic Environmental Research and Development Program.

Naidoo, R. et al., 2006. Integrating economic costs into conservation planning. Trends in Ecology & Evolution, 21(12), pp.681–687.

N.C. Division of Parks and Recreation. 2011. Draft Master Plan for Carver's Creek State Park. http://www.ncparks.gov/Visit/parks/cacr/master_plan.php

Opdam, P. & Wascher, D., 2004. Climate change meets habitat fragmentation: linking landscape and biogeographical scale levels in research and conservation. Biological conservation, 117(3), pp.285–297.

Phillips, S.J. & Dudík, M., 2008. Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography, 31, pp.161-175.

Pinto, N. & Keitt, Timothy H., 2008. Beyond the least-cost path: evaluating corridor redundancy using a graph-theoretic approach. Landscape Ecology, 24, pp.253-266.

Proctor, M.F. et al., 2005. Genetic analysis reveals demographic fragmentation of grizzly bears yielding vulnerably small populations. Proceedings of the Royal Society B: Biological Sciences, 272(1579), pp.2409 -2416.

Ricketts, T.H., 2001. The matrix matters: effective isolation in fragmented landscapes. The American Naturalist, 158(1), pp.87–99.

Sawyer, S.C., Epps, C.W. & Brashares, J.S., 2011. Placing linkages among fragmented habitats: do least-cost models reflect how animals use landscapes? Journal of Applied Ecology, 48(3), pp.668-678.

Spear, S.F. et al., 2010. Use of resistance surfaces for landscape genetic studies: considerations for parameterization and analysis. Molecular Ecology, 19(17), pp.3576- 3591.

Taylor, P.D. et al., 1993. Connectivity is a vital element of landscape structure. Oikos, 68(3), pp.571–573.

Urban, D. & Keitt, T., 2001. Landscape connectivity: a graph-theoretic perspective. Ecology, 82(5), pp.1205–1218.

Urban, Dean L. et al., 2009. Graph models of habitat mosaics. Ecology Letters, 12(3), pp.260- 273.

U.S. Fish and Wildlife Service. Recovery plan for the red-cockaded woodpecker (Picoides borealis): second revision. Atlanta, GA: U.S. Fish and Wildlife Service; 2003.

Van Lear, D.H. et al., 2005. History and restoration of the longleaf pine-grassland ecosystem: Implications for species at risk. Forest Ecology and Management, 211(1-2), pp.150-165.

Wang, I.J., Savage, W.K. & Bradley Shaffer, H., 2009. Landscape genetics and least-cost path analysis reveal unexpected dispersal routes in the California tiger salamander (Ambystoma californiense). Molecular Ecology, 18(7), pp.1365-1374.

Wilcove, D.S. & Lee, J., 2004. Using Economic and Regulatory Incentives to Restore Endangered Species: Lessons Learned from Three New Programs. Conservation Biology, 18(3), pp.639-645.

Wright, L.I., Tregenza, T. & Hosken, D.J., 2007. Inbreeding, inbreeding depression and extinction. Conservation Genetics, 9, pp.833-843.

Xian, G., Homer, C. & Fry, J., 2009. Updating the 2001 National Land Cover Database land cover classification to 2006 by using Landsat imagery change detection methods. Remote Sensing of Environment, 113(6), pp.1133-1147.

Zetterberg, A., Mörtberg, U.M. & Balfors, B., 2010. Making graph theory operational for landscape ecological assessments, planning, and design. Landscape and Urban Planning, 95(4), pp.181-191.

APPENDIX A: EMPIRICAL PAPERS REVIEWED IN TABLE 1

Anadón, J.D. et al., 2007. Assessing changes in habitat quality due to land use changes in the spur-thighed tortoise Testudo graeca using hierarchical predictive habitat models. Diversity and Distributions, 13, pp.324-331.

Armstrong, T. & Stamp, N., 2003. Effects of prey quantity on predatory wasps (Polistes dominulus) when patch quality differs. Behavioral Ecology and Sociobiology, 54(3), pp.310- 319.

Arnhem, E. et al., 2008. Selective logging, habitat quality and home range use by sympatric gorillas and chimpanzees: a case study from an active logging concession in southeast Cameroon. Folia Primatologica, 79, pp.1-14.

Arroyo-Rodríguez, V. & Mandujano, S., 2006. Forest fragmentation modifies habitat quality for Alouatta palliata. International Journal of Primatology, 27, pp.1079-1096.

Barnes, T.G. et al., 1995. An assessment of habitat quality for northern bobwhite in tall fescue-dominated fields. Wildlife Society Bulletin, pp.231–237.

Bauerfeind, S.S., Theisen, A. & Fischer, K., 2008. Patch occupancy in the endangered butterfly Lycaena helle in a fragmented landscape: effects of habitat quality, patch size and isolation. Journal of Insect Conservation, 13, pp.271-277.

Belanger, G. & Rodriguez, M., 2002. Local movement as a measure of habitat quality in stream salmonids. Environmental Biology of Fishes, 64(1-3), pp.155-164.

Bengtson, S. & Bloch, D., 1983. Island land bird population-densities in relation to island size and habitat quality on the Faroe Islands. Oikos, 41(3), pp.507-522.

Bergeron, J. & Jodoin, L., 1993. Intense grazing by voles (Microtus pennsylvanicus) and its effect on habitat quality. Canadian Journal of Zoology-Revue Canadienne de Zoologie, 71(9), pp.1823-1830.

Betts, M.G. et al., 2008. Dynamic occupancy models reveal within-breeding season movement up a habitat quality gradient by a migratory songbird. Ecography, 31, pp.592-600.

Beyer Jr, E. et al., 1996. Habitat quality and reproduction of red-cockaded woodpecker groups in Florida. The Journal of wildlife management, pp.826–835.

Binzenhöfer, B. et al., 2007. Connectivity compensates for low habitat quality and small patch size in the butterfly Cupido minimus. Ecological Research, 23, pp.259-269.

Boivin, G., Fauvergue, X. & Wajnberg, E., 2004. Optimal patch residence time in egg parasitoids: innate versus learned estimate of patch quality. Oecologia, 138, pp.640-647.

Bonte, D. et al., 2003. Patch quality and connectivity influence spatial dynamics in a dune wolfspider. Oecologia, 135(2), pp.227-233.

100

Breininger, D.R., Burgman, M.A. & Stith, B.M., 1999. Influence of habitat quality, catastrophes, and population size on extinction risk of the Florida scrub-jay. Wildlife Society Bulletin, pp.810–822.

Briner, T., Nentwig, W. & Airoldi, J., 2005. Habitat quality of wildflower strips for common voles (Microtus arvalis) and its relevance for agriculture. Agriculture Ecosystems & Environment, 105(1-2), pp.173-179.

Brooks, R. & Dodge, W., 1986. Estimation of habitat quality and summer population-density for muskrats on a watershed basis. Journal of Wildlife Management, 50(2), pp.269-273.

Brooks, R.P. & Dodge, W.E., 1986. Estimation of habitat quality and summer population density for muskrats on a watershed basis. The Journal of wildlife management, pp.269– 273.

Chapman, C. et al., 2005. Assessing dietary protein of colobus monkeys through faecal sample analysis: a tool to evaluate habitat quality. African Journal of Ecology, 43(3), pp.276-278.

Chauvenet, A.L.M. et al., 2010. Optimal allocation of conservation effort among subpopulations of a threatened species: How important is patch quality? Ecological Applications, 20(3), pp.789–797.

Ciani, A.C. et al., 2001. Effects of water availability and habitat quality on bark-stripping behavior in Barbary macaques. Conservation biology, 15(1), pp.259–265.

Clark, J.A. & May, R.M., 2002. Taxonomic bias in conservation research. Science, 297(5579), p.191.

Corbalan, V., Tabeni, S. & Ojeda, R.A., 2006. Assessment of habitat quality for four small mammal species of the Monte Desert, Argentina. Mammalian Biology, 71(4), pp.227- 237.

Denoel, M. & Lehmann, A., 2006. Multi-scale effect of landscape processes and habitat quality on newt abundance: Implications for conservation. Biological Conservation, 130(4), pp.495-504.

Dhondt, A.A., 2010. Effects of competition on great and blue tit reproduction: intensity and importance in relation to habitat quality. Journal of Animal Ecology, 79(1), pp.257–265.

Dixon, A. & Kundu, R., 1998. Resource tracking in aphids: programmed reproductive strategies anticipate seasonal trends in habitat quality. Oecologia, 114(1), pp.73-78.

Doligez, B. et al., 2008. Spatial scale of local breeding habitat quality and adjustment of breeding decisions. Ecology, 89(5), pp.1436–1444.

101

Donald, P.F. et al., 2010. Rapid declines in habitat quality and population size of the Liben (Sidamo) Lark Heteromirafra sidamoensis necessitate immediate conservation action. Bird Conservation International, 20(01), pp.1–12.

Dunbar, R., 1987. Habitat quality, population-dynamics, and group composition in colobus monkeys (Colobus guereza). International Journal of Primatology, 8(4), pp.299-329.

Ellis, S., 2003. Habitat quality and management for the northern brown argus butterfly Aricia artaxerxes (Lepidoptera: Lycaenidae) in North East England. Biological Conservation, 113(2), pp.285-294.

Erbilgin, N. et al., 2002. Population dynamics of Ips pini and Ips grandicollis in red pine plantations in Wisconsin: within- and between-year associations with predators, competitors, and habitat quality. Environmental Entomology, 31, pp.1043-1051.

Falcucci, A. et al., 2009. Assessing habitat quality for conservation using an integrated occurrence-mortality model. Journal of Applied Ecology, 46(3), pp.600–609.

Ferns, P.N. & Hinsley, Shelley A., 2008. Carotenoid plumage hue and chroma signal different aspects of individual and habitat quality in tits. Ibis, 150(1), pp.152-159.

Fleishman, E. et al., 2002. Assessing the roles of patch quality, area, and isolation in predicting metapopulation dynamics. Conservation Biology, 16(3), pp.706-716.

Focardi, S. et al., 2006. Inter-specific competition from fallow deer Dama dama reduces habitat quality for the Italian roe deer Capreolus capreolus italicus. Ecography, 29(3), pp.407-417.

Forsman, J.T. et al., 2008. Competitor density cues for habitat quality facilitating habitat selection and investment decisions. Behavioral Ecology, 19(3), p.539.

Fowles, A.P. & Smith, R.G., 2006. Mapping the habitat quality of patch networks for the marsh fritillary Euphydryas aurinia (Rottemburg, 1775) (Lepidoptera, Nymphalidae) in Wales. Journal of Insect Conservation, 10, pp.161-177.

Franklin, A.B. et al., 2000. Climate, habitat quality, and fitness in northern spotted owl populations in northwestern California. Ecological Monographs, 70(4), pp.539–590.

Franzen, M. & Nilsson, S.G., 2010. Both population size and patch quality affect local extinctions and colonizations. Proceedings of the Royal Society B-Biological Sciences, 277(1678), pp.79-85.

Fraser, D.F. et al., 1999. Habitat quality in a hostile river corridor. Ecology, 80(2), pp.597–607.

102

Frey, C.M., Jensen, W.E. & With, K.A., 2008. Topographic patterns of nest placement and habitat quality for grassland birds in tallgrass prairie. The American Midland Naturalist, 160(1), pp.220–234.

Gallant, D. et al., 2004. An extensive study of the foraging ecology of beavers (Castor canadensis) in relation to habitat quality. Canadian Journal of Zoology-Revue Canadienne de Zoologie, 82(6), pp.922-933.

Garcia, J. et al., 2007. Spatial analysis of habitat quality in a fragmented population of little bustard (Tetrax tetrax): Implications for conservation. Biological Conservation, 137(1), pp.45-56.

Goubault, M. et al., 2004. Selection strategies of parasitized hosts in a generalist parasitoid depend on patch quality but also on host size. Journal of Insect Behavior, 17(1), pp.99- 113.

Greenfield, K.C. et al., 2002. Vegetation management practices on Conservation Reserve Program fields to improve northern bobwhite habitat quality. Wildlife Society Bulletin, pp.527–538.

Greenfield, K. C. et al., 2003. Effects of Burning and Discing Conservation Reserve Program Fields to Improve Habitat Quality for Northern Bobwhite (Colinus virginianus). The American Midland Naturalist, 149, pp.344-353.

Griffen, B.D. & Drake, J.M., 2008. Effects of habitat quality and size on extinction in experimental populations. Proceedings of the Royal Society B: Biological Sciences, 275, pp.2251-2256.

Gunnarsson, T.G. et al., 2005. Seasonal matching of habitat quality and fitness in a migratory bird. Proceedings of the Royal Society B: Biological Sciences, 272, pp.2319-2323.

Gunness, M.A., Clark, R.G. & Weatherhead, P.J., 2001. Counterintuitive parental investment by female dabbling ducks in response to variable habitat quality. Ecology, 82(4), pp.1151–1158.

Hall, L.S. & Morrison, M.L., 1998. Responses of mice to fluctuating habitat quality II. A supplementation experiment. The Southwestern Naturalist, pp.137–146.

Hanrahan, T.P., Geist, D.R. & Arntzen, E.V., 2005. Habitat quality of historic Snake River fall Chinook salmon spawning locations and implications for incubation survival. Part 1: substrate quality. River Research and Applications, 21, pp.455-467.

Hart, P. & Jackson, P., 1986. The influence of sex, patch quality, and travel time on foraging decisions by young-adult Homo sapiens l. Ethology and Sociobiology, 7(2), pp.71-89.

Haynes, K.J. et al., 2006. Landscape context outweighs local habitat quality in its effects on herbivore dispersal and distribution. Oecologia, 151, pp.431-441.

103

Hellgren, E.C., Rogers, L. & Seal, U., 1993. Serum chemistry and hematology of black bears: physiological indices of habitat quality or seasonal patterns? Journal of Mammalogy, pp.304–315.

Hill, R. et al., 2004. Predicting habitat quality for Great Tits (Parus major) with airborne laser scanning data. International Journal of Remote Sensing, 25(22), pp.4851-4855.

Hinsley, S. et al., 2002. Quantifying woodland structure and habitat quality for birds using airborne laser scanning. Functional Ecology, 16(6), pp.851-857.

Hinsley, S.A. et al., 2006. The application of lidar in woodland bird ecology: climate, canopy structure, and habitat quality. Photogrammetric Engineering & Remote Sensing, 72(12), pp.1399–1406.

Hunt, P.D., 1996. Habitat selection by American Redstarts along a successional gradient in northern hardwoods forest: evaluation of habitat quality. The Auk, pp.875–888.

Hunter, M. & Willmer, P., 1989. The potential for interspecific competition between 2 abundant defoliators on oak - leaf damage and habitat quality. Ecological Entomology, 14(3), pp.267-277.

Iwamoto, T. & Dunbar, R., 1983. Thermoregulation, habitat quality and the behavioral ecology of Gelada baboons. Journal of Animal Ecology, 52(2), pp.357-366.

Jaquiéry, J. et al., 2008. Habitat-quality effects on metapopulation dynamics in greater white- toothed shrews, Crocidura russula. Ecology, 89(10), pp.2777–2785.

Jenkins, A., 2000. Hunting mode and success of African peregrines Falco peregrinus minor: does nesting habitat quality affect foraging efficiency? Ibis, 142(2), pp.235-246.

Jessup, B., 1998. A strategy for simulating brown trout population dynamics and habitat quality in an urbanizing watershed. Ecological Modelling, 112(2-3), pp.151-167.

Johnson, M.D. et al., 2006. Assessing habitat quality for a migratory songbird wintering in natural and agricultural habitats. Conservation Biology, 20, pp.1433-1444.

Jorgensen, J.C. et al., 2009. Linking landscape-level change to habitat quality: an evaluation of restoration actions on the freshwater habitat of spring-run Chinook salmon. Freshwater Biology, 54(7), pp.1560–1575.

Kamil, A.C., Misthal, R.L. & Stephens, D.W., 1993. Failure of simple optimal foraging models to predict residence time when patch quality is uncertain. Behavioral Ecology, 4(4), p.350.

Kawecki, T., 1997. Habitat quality ranking depends on habitat-independent environmental factors: A model and results from Callosobruchus maculatus. Functional Ecology, 11(2), pp.247-254.

104

Kleijn, D. et al., 2010. Adverse effects of agricultural intensification and climate change on breeding habitat quality of Black-tailed Godwits Limosa l. limosa in the Netherlands. Ibis, 152(3), pp.475–486.

Knutson, M.G. et al., 2006. An assessment of bird habitat quality using population growth rates. The Condor, 108, p.301.

Kohlmann, S. & Risenhoover, K., 1996. Using artificial food patches to evaluate habitat quality for granivorous birds: An application of foraging theory. Condor, 98(4), pp.854-857.

Koops, M. & Abrahams, M., 1999. Assessing the ideal free distribution: Do guppies use aggression as public information about patch quality? Ethology, 105(9), pp.737-746.

Kostrzewa, A., 1996. A comparative study of nest-site occupancy and breeding performance as indicators for nestinghabitat quality in three European raptor species. Ethology Ecology & Evolution, 8, pp.1-18.

Krauss, J. et al., 2005. Relative importance of resource quantity, isolation and habitat quality for landscape distribution of a monophagous butterfly. Ecography, 28(4), pp.465-474.

Kroll, A.J. & Haufler, J.B., 2007. Evaluating habitat quality for the dusky flycatcher. Journal of Wildlife Management, 71, pp.14-22.

Kruitbos, L. et al., 2010. The influence of habitat quality on the foraging strategies of the entomopathogenic nematodes Steinernema carpocapsae and Heterorhabditis megidis. Parasitology, 137(2), pp.303–309.

Lambrechts, M.M. et al., 2004. Habitat quality as a predictor of spatial variation in blue tit reproductive performance: a multi-plot analysis in a heterogeneous landscape. Oecologia, 141, pp.555-561.

Lanyon, S. & Thompson, C., 1986. Site fidelity and habitat quality as determinants of settlement pattern in male painted buntings. Condor, 88(2), pp.206-210.

Lavers, C., HainesYoung, R. & Avery, M., 1996. The habitat associations of dunlin (Calidris alpina) in the Flow Country of northern Scotland and an improved model for predicting habitat quality. Journal of Applied Ecology, 33(2), pp.279-290.

Leather, S., Wellings, P. & Dixon, A., 1983. Habitat quality and the reproductive strategies of the migratory morphs of the bird cherry-oat aphid, Rhopalosiphum padi (l), colonizing secondary host plants. Oecologia, 59(2-3), pp.302-306.

Li, Zhaoyuan & Rogers, M.E., 2005. Habitat quality and range use of white-headed langurs in Fusui, China. Folia Primatologica, 76, pp.185-195.

Li, Z.Y. & Rogers, E., 2004. Habitat quality and activity budgets of white-headed langurs in Fusui, China. International Journal of Primatology, 25(1), pp.41-54.

105

Lin, Y. & Batzli, G., 2001. The influence of habitat quality on dispersal demography, and population dynamics of voles. Ecological Monographs, 71(2), pp.245-275.

Lin, Y.K. et al., 2006. Effects of patch quality on dispersal and social organization of prairie voles: An experimental approach. Journal of Mammalogy, 87(3), pp.446-453.

Lobon-Cervia, J., 2008. Habitat quality enhances spatial variation in the self-thinning patterns of stream-resident brown trout (Salmo trutta). Canadian Journal of Fisheries and Aquatic Sciences, 65(9), pp.2006-2015.

Lohmus, A. & Vali, U., 2004. The effects of habitat quality and female size on the productivity of the lesser spotted eagle Aquila pomarina in the light of the alternative prey hypothesis. Journal of Avian Biology, 35(5), pp.455-464.

Luck, G., 2002. Determining habitat quality for the cooperatively breeding Rufous Treecreeper, Climacteris rufa. Austral Ecology, 27(2), pp.229-237.

Lurz, P., Garson, P. & Wauters, L., 1997. Effects of temporal and spatial variation in habitat quality on red squirrel dispersal behaviour. Animal Behaviour, 54(Part 2), pp.427-435.

Lyons, J.E., 2005. Habitat-specific foraging of prothonotary warblers: deducing habitat quality. The Condor, 107, p.41.

Mallory, M.L., McNicol, D.K. & Weatherhead, P.J., 1994. Habitat quality and reproductive effort of Common Goldeneyes nesting near Sudbury, Canada. The Journal of Wildlife Management, 58(3), pp.552–560.

Martin, J. & Joron, M., 2003. Nest predation in forest birds: influence of predator type and predator’s habitat quality. Oikos, 102(3), pp.641-653.

Martinez, J.-J.I., Mokady, O. & Wool, D., 2005. Patch Size and Patch Quality of Gall- inducing Aphids in a Mosaic Landscape in Israel. Landscape Ecology, 20, pp.1013-1024.

Mathieu, J. et al., 2010. Habitat quality, conspecific density, and habitat pre-use affect the dispersal behaviour of two earthworm species, Aporrectodea icterica and Dendrobaena veneta, in a mesocosm experiment. Soil Biology and Biochemistry, 42(2), pp.203–209.

Matter, S. & Roland, J., 2002. An experimental examination of the effects of habitat quality on the dispersal and local abundance of the butterfly Parnassius smintheus. Ecological Entomology, 27(3), pp.308-316.

Matter, S.F. et al., 2009. Interactions between habitat quality and connectivity affect immigration but not abundance or population growth of the butterfly, Parnassius smintheus. Oikos, 118(10), pp.1461–1470.

106

Maurer, B.A., 1986. Predicting habitat quality for grassland birds using density-habitat correlations. The Journal of Wildlife Management, pp.556–566.

McCorquodale, S.M., 1991. Energetic considerations and habitat quality for elk in arid grasslands and coniferous forests. Journal of Wildlife Management, pp.237–242.

McLoughlin, P., Ferguson, S. & Messier, F., 2000. Intraspecific variation in home range overlap with habitat quality: A comparison among brown bear populations. Evolutionary Ecology, 14(1), pp.39-60.

Moilanen, A. & Hanski, I., 1998. Metapopulation dynamics: effects of habitat quality and landscape structure. Ecology, 79(7), pp.2503–2515.

Moorhouse, T.P., Gelling, M. & Macdonald, D.W., 2009. Effects of habitat quality upon reintroduction success in water voles: Evidence from a replicated experiment. Biological Conservation, 142(1), pp.53-60.

Morrison, M. & Hall, L., 1998. Responses of mice to fluctuating habitat quality I. Patterns from a long-term observational study. Southwestern Naturalist, 43(2), pp.123-136.

Muller, K.L. et al., 1997. The effects of conspecific attraction and habitat quality on habitat selection in territorial birds (Troglodytes Aedon). The American Naturalist, 150, pp.650- 661.

Muratori, F., Boivin, G. & Hance, T., 2008. The impact of patch encounter rate on patch residence time of female parasitoids increases with patch quality. Ecological Entomology, 33(3), pp.422–427.

Nakagawa, N., 1989. Feeding strategies of japanese monkeys against deterioration of habitat quality. Primates, 30(1), pp.1-16.

Nakashima, Y., Teshiba, M. & Hirose, Y., 2002. Flexible use of patch marks in an insect predator: effect of sex, hunger state, and patch quality. Ecological Entomology, 27(5), pp.581-587.

Newcomb Homan, R. et al., 2003. Impacts of varying habitat quality on the physiological stress of spotted salamanders (Ambystoma maculatum). Animal Conservation, 6, pp.11- 18.

Norris, D., 2005. Carry-over effects and habitat quality in migratory populations. Oikos, 109(1), pp.178-186.

Nussey, D.H. et al., 2007. The relationship between tooth wear, habitat quality and late-life reproduction in a wild red deer population. Journal of Animal Ecology, 76, pp.402-412.

Nystrand, M., 2006. Influence of age, kinship, and large-scale habitat quality on local foraging choices of Siberian jays. Behavioral Ecology, 17(3), p.503.

107

Oliva-Paterna, F., Minano, P. & Torralva, M., 2003. Habitat quality affects the condition of Barbus sclateri in Mediterranean semi-arid streams. Environmental Biology of Fishes, 67(1), pp.13-22.

Oliva-Paterna, F., Vila-Gispert, A. & Torralva, M., 2003. Condition of Barbus sclateri from semi-arid aquatic systems: effects of habitat quality disturbances. Journal of Fish Biology, 63(3), pp.699-709.

Olson, K.A. et al., 2009. Short Communication A mega-herd of more than 200,000 Mongolian gazelles Procapra gutturosa: a consequence of habitat quality. Oryx, 43(1), pp.149-153.

Olsson, O., Brown, J. & Smith, H., 2002. Long- and short-term state-dependent foraging under predation risk: an indication of habitat quality. Animal Behaviour, 63(Part 5), pp.981-989. van Oort, H. et al., 2006. Habitat Quality, Social Dominance and Dawn Chorus Song Output in Black-Capped Chickadees. Ethology, 112, pp.772-778.

Ortega, Y.K. & Capen, D.E., 1999. Effects of forest roads on habitat quality for ovenbirds in a forested landscape. The Auk, pp.937–946.

Ozgul, A. et al., 2006. Effects of patch quality and network structure on patch occupancy dynamics of a yellow-bellied marmot metapopulation. Journal of Animal Ecology, 75, pp.191-202.

Panek, M., 1997. Density-dependent brood production in the Grey Partridge Perdix perdix in relation to habitat quality. Bird Study, 44(Part 2), pp.235-238.

Paradis, E., 1995. Survival, immigration and habitat quality in the Mediterranean pine vole. Journal of Animal Ecology, pp.579–591.

Paradis, E. & Croset, H., 1995. Assessment of habitat quality in the Mediterranean pine vole (Microtus duodecimcostatus) by the study of survival rates. Canadian Journal of Zoology- Revue Canadienne de Zoologie, 73(8), pp.1511-1518.

Partridge, L.W., Britton, N.F. & Franks, N.R., 1996. Army Ant Population Dynamics: The Effects of Habitat Quality and Reserve Size on Population Size and Time to Extinction. Proceedings of the Royal Society B: Biological Sciences, 263(1371), pp.735 -741.

Pausas, J., Braithwaite, L. & Austin, M., 1995. Modeling habitat quality for arboreal marsupials in the south coastal forests of New-South-Wales, Australia. Forest Ecology and Management, 78(1-3), pp.39-49.

Pederson, J.C., Farentinos, R. & Littlefield, V.M., 1987. Effects of logging on habitat quality and feeding patterns of Abert squirrels. Western North American Naturalist, 47(2), pp.252–258.

108

Perot, A. & Villard, M.-A., 2009. Putting Density Back into the Habitat-Quality Equation: Case Study of an Open-Nesting Forest Bird. Conservation Biology, 23, pp.1550-1557.

Pettorelli, N et al., 2003. Age and density modify the effects of habitat quality on survival and movements of roe deer. Ecology, 84(12), pp.3307-3316.

Pettorelli, N. et al., 2002. Variations in adult body mass in roe deer: the effects of population density at birth and of habitat quality. Proceedings of the Royal Society B: Biological Sciences, 269, pp.747-753.

Pettorelli, N. et al., 2001. Population density and small-scale variation in habitat quality affect phenotypic quality in roe deer. Oecologia, 128, pp.400-405.

Pettorelli, N. et al., 2005. The response of fawn survival to changes in habitat quality varies according to cohort quality and spatial scale. Journal of Animal Ecology, 74, pp.972-981.

Possingham, H., 1992. Habitat selection by 2 species of nectarivore - habitat quality isolines. Ecology, 73(5), pp.1903-1912.

Powell, G. & Powell, A., 1986. Reproduction by great white herons Ardea-Herodias in Florida Bay as an indicator of habitat quality. Biological Conservation, 36(2), pp.101-113.

Power, M.E., 1984. Habitat quality and the distribution of algae-grazing catfish in a Panamanian stream. The Journal of Animal Ecology, pp.357–374.

Prenda, J., Lopez-Nieves, P. & Bravo, R., 2001. Conservation of otter (Lutra lutra) in a Mediterranean area: the importance of habitat quality and temporal variation in water availability. Aquatic Conservation: Marine and Freshwater Ecosystems, 11, pp.343-355.

Prudic, K.L., Oliver, J.C. & Bowers, M.D., 2005. Soil nutrient effects on oviposition preference, larval performance, and chemical defense of a specialist insect herbivore. Oecologia, 143, pp.578-587.

Pywell, R. et al., 2004. Assessing habitat quality for butterflies on intensively managed arable farmland. Biological Conservation, 118(3), pp.313-325.

Rabasa, S.G., Gutierrez, D. & Escudero, Adrian, 2007. Metapopulation structure and habitat quality in modelling dispersal in the butterfly Iolana iolas. Oikos, 116(5), pp.793-806.

Rabasa, S.G., Gutiérrez, D. & Escudero, Adrián, 2008. Relative importance of host plant patch geometry and habitat quality on the patterns of occupancy, extinction and density of the monophagous butterfly Iolana iolas. Oecologia, 156, pp.491-503.

109

Rayor, l., 1985. Effects of habitat quality on growth, age of first reproduction, and dispersal in Gunnison prairie dogs (Cynomys-gunnisoni). Canadian Journal of Zoology-Revue Canadienne de Zoologie, 63(12), pp.2835-2840.

Reijnen, R. & Foppen, R., 1994. The effects of car traffic on breeding bird populations in woodland .1. evidence of reduced habitat quality for willow warblers (Phylloscopus trochilus) breeding close to a highway. Journal of Applied Ecology, 31(1), pp.85-94.

Reiskind, M.H. & Wilson, M.L., 2004. Culex restuans (Diptera: Culicidae) oviposition behavior determined by larval habitat quality and quantity in southeastern Michigan. Journal of Medical Entomology, 41, pp.179-186.

Reynolds-Hogland, M.J. & Mitchell, M.S., 2007. Effects of roads on habitat quality for bears in the southern Appalachians: A long-term study. Journal of Mammalogy, 88(4), pp.1050-1061.

Robakiewicz, P. & Daigle, W., 2004. Patch quality and foraging time in the crab spider Misumenops asperatus Hentz (Araneae: Thomisidae). Northeastern Naturalist, 11, pp.23- 32.

Romero-Calcerrada, R. & Luque, S., 2006. Habitat quality assessment using weights-of- evidence based GIS modelling: The case of Picoides tridactylus as species indicator of the biodiversity value of the Finnish forest. Ecological Modelling, 196(1-2), pp.62-76.

Root, K., 1998. Evaluating the effects of habitat quality, connectivity, and catastrophes on a threatened species. Ecological Applications, 8(3), pp.854-865.

Rudolf, V.H.W. & Rodel, M.-O., 2004. Oviposition site selection in a complex and variable environment: the role of habitat quality and conspecific cues. Oecologia, 142, pp.316- 325.

Russell, R. et al., 1994. The impact of variation in stopover habitat quality on migrant rufous hummingbirds. Conservation Biology, 8(2), pp.483-490.

Santos, T. et al., 2008. Habitat quality predicts the distribution of a lizard in fragmented woodlands better than habitat fragmentation. Animal Conservation, 11(1), pp.46-56.

Sawchik, J., Dufrene, M. & Lebrun, P., 2003. Estimation of habitat quality based on plant community, and effects of isolation in a network of butterfly habitat patches. Acta Oecologica-International Journal of Ecology, 24(1), pp.25-33.

Schilman, P. & Roces, F., 2003. Assessment of nectar flow rate and memory for patch quality in the ant Camponotus rufipes. Animal Behaviour, 66(Part 4), pp.687-693.

Schooley, R.L. & Branch, L.C., 2007. Spatial heterogeneity in habitat quality and cross-scale interactions in metapopulations. Ecosystems, 10, pp.846-853.

110

Sergio, F., Pedrini, P. & Marchesi, L., 2003. Spatio-temporal shifts in gradients of habitat quality for an opportunistic avian predator. Ecography, 26(2), pp.243-255.

Shahabuddin, G. et al., 2000. Persistence of a frugivorous butterfly species in Venezuelan forest fragments: the role of movement and habitat quality. Biodiversity and Conservation, 9(12), pp.1623-1641.

Sherry, T.W. & Holmes, R.T., 1996. Winter habitat quality, population limitation, and conservation of Neotropical-Nearctic migrant birds. Ecology, 77(1), pp.36–48.

Sinsch, U. et al., 2007. Life-history traits in green toad (Bufo viridis) populations: indicators of habitat quality. Canadian Journal of Zoology-Revue Canadienne de Zoologie, 85(5), pp.665- 673.

Sjoberg, K. et al., 2000. Response of Mallard ducklings to variation in habitat quality: An experiment of food limitation. Ecology, 81(2), pp.329-335.

Smart, J. et al., 2008. Changing land management of lowland wet grasslands of the UK: impacts on snipe abundance and habitat quality. Animal Conservation, 11(4), pp.339- 351.

Smith, J., Ahearn, S. & McDougal, C., 1998. Landscape analysis of tiger distribution and habitat quality in Nepal. Conservation Biology, 12(6), pp.1338-1346.

Smith, J.A.M., Reitsma, L.R. & Marra, P.P., 2010. Moisture as a determinant of habitat quality for a nonbreeding Neotropical migratory songbird. Ecology, 91(10), pp.2874– 2882.

Stanko-Mishic, S. & others, 1999. Manipulation of habitat quality: effects on chironomid life history traits. Freshwater Biology, 41(4), pp.719–727.

Stasek, D.J., Bean, C. & Crist, T.O., 2008. Butterfly Abundance and Movements Among Prairie Patches: The Roles of Habitat Quality, Edge, and Forest Matrix Permeability. Environmental Entomology, 37, pp.897-906.

Stauss, M. et al., 2005. Sex ratio of Parus major and P-caeruleus broods depends on parental condition and habitat quality. Oikos, 109(2), pp.367-373.

Stevens, V.M. & Baguette, M., 2008. Importance of Habitat Quality and Landscape Connectivity for the Persistence of Endangered Natterjack Toads. Conservation Biology, 22, pp.1194-1204.

Sullivan, S.M.P., Watzin, M.C. & Hession, W.C., 2006. Differences in the Reproductive Ecology of Belted Kingfishers (Ceryle alcyon) Across Streams with Varying Geomorphology and Habitat Quality. Waterbirds, 29, pp.258-270.

111

Sutherland, W., 1998. The effect of local change in habitat quality on populations of migratory species. Journal of Applied Ecology, 35(3), pp.418-421.

Tilgar, V., Mänd, R. & Leivits, A., 1999. Effect of calcium availability and habitat quality on reproduction in pied flycatcher Ficedula hypoleuca and great tit Parus major. Journal of avian biology, pp.383–391.

Tremblay, I. et al., 2005. The effect of habitat quality on foraging patterns, provisioning rate and nestling growth in Corsican blue tits Parus caeruleus. Ibis, 147(1), pp.17-24.

Valkama, J. & Korpimaki, E., 1999. Nestbox characteristics, habitat quality and reproductive success of Eurasian kestrels. Bird Study, 46(Part 1), pp.81-88.

Van Horne, B., 1983. Density as a misleading indicator of habitat quality. Journal of Wildlife Management, 47(4), pp.893-901.

Vanpé, C. et al., 2009. Access to mates in a territorial ungulate is determined by the size of a male’s territory, but not by its habitat quality. Journal of Animal Ecology, 78(1), pp.42– 51.

Vickery, P.D., Hunter Jr, M.L. & Wells, J.V., 1992. Use of a new reproductive index to evaluate relationship between habitat quality and breeding success. The Auk, pp.697– 705.

Virgos, E., 2001. Role of isolation and habitat quality in shaping species abundance: a test with badgers (Meles meles L.) in a gradient of forest fragmentation. Journal of Biogeography, 28(3), pp.381-389.

Virkkala, R., 1990. Ecology of the Siberian tit Parus cinctus in relation to habitat quality: effects of forest management. Ornis Scandinavica, pp.139–146.

Visconti, P. & Elkin, C., 2009. Using connectivity metrics in conservation planning - when does habitat quality matter? Diversity and Distributions, 15, pp.602-612.

Vögeli, M. et al., 2010. The relative importance of patch habitat quality and landscape attributes on a declining steppe-bird metapopulation. Biological Conservation, 143(5), pp.1057–1067.

Wagner, J. & Wise, D., 1997. Influence of prey availability and conspecifics on patch quality for a cannibalistic forager: Laboratory experiments with the wolf spider Schizocosa. Oecologia, 109(3), pp.474-482.

Walters, K., Brough, C. & Dixon, A., 1988. Habitat quality and reproductive investment in aphids. Ecological Entomology, 13(3), pp.337-345.

112

Webb, J.K., Shine, R. & Pringle, R.M., 2005. Canopy Removal Restores Habitat Quality for an Endangered Snake in a Fire Suppressed Landscape M. J. Lannoo, ed. Copeia, 2005, pp.894-900.

Weinberg, H.J. & Roth, R.R., 1998. Forest area and habitat quality for nesting wood thrushes. The Auk, pp.879–889.

Weiss, S.B., Murphy, D.D. & White, R.R., 1988. Sun, slope, and butterflies: topographic determinants of habitat quality for Euphydryas editha. Ecology, pp.1486–1496.

Wheatley, M., Larsen, K.W. & Boutin, S., 2002. Does density reflect habitat quality for North American red squirrels during a spruce-cone failure? Journal of Mammalogy, 83, pp.716-727.

Whitten, P., 1988. Effects of patch quality and feeding subgroup size on feeding success in vervet monkeys (Cercopithecus aethiops). Behaviour, 105, 1(2), pp.35–52.

Widen, P., 1994. Habitat quality for raptors - a field experiment. Journal of Avian Biology, 25(3), pp.219-223.

Wiegand, T. et al., 2008. Animal habitat quality and ecosystem functioning: exploring seasonal patterns using NDVI. Ecological Monographs, 78(1), pp.87–103.

Wightman, C.S. & Germaine, S.S., 2006. Forest stand characteristics altered by restoration affect western bluebird habitat quality. Restoration Ecology, 14(4), pp.653-661.

Wilkin, T.A., King, L.E. & Sheldon, B.C., 2009. Habitat quality, nestling diet, and provisioning behaviour in great tits Parus major. Journal of Avian Biology, 40, pp.135- 145.

Wilkin, T.A., Perrins, C.M. & Sheldon, B.C., 2007. The use of GIS in estimating spatial variation in habitat quality: a case study of lay-date in the great tit Parus major. Ibis, 149(2), pp.110-118.

Winker, K., Rappole, J.H. & Ramos, M.A., 1995. The use of movement data as an assay of habitat quality. Oecologia, 101(2), pp.211–216.

Winnie Jr, J.A., Cross, P. & Getz, W., 2008. Habitat quality and heterogeneity influence distribution and behavior in African buffalo (Syncerus caffer). Ecology, 89(5), pp.1457– 1468.

Winterrowd, M. & Devenport, L., 2004. Balancing variable patch quality with predation risk. Behavioural Processes, 67(1), pp.39-46.

Wong, S.N.P., Saj, T.L. & Sicotte, P., 2006. Comparison of habitat quality and diet of Colobus vellerosus in forest fragments in Ghana. Primates, 47, pp.365-373.

113

Yamanaka, T. et al., 2009. Evaluating the relative importance of patch quality and connectivity in a damselfly metapopulation from a one-season survey. Oikos, 118(1), pp.67–76.

Zanette, L., 2001. Indicators of habitat quality and the reproductive output of a forest songbird in small and large fragments. Journal of Avian Biology, 32(1), pp.38-46.

114