Abstract Mozelewski, Tina Grace

ABSTRACT

MOZELEWSKI, TINA GRACE. Forecasting for Intended Consequences in Conservation Decision-Making. (Under the direction of Dr. Robert Scheller).

Global conservation efforts are facing mounting challenges to halt biodiversity loss under accelerating anthropogenic change. Climate change is triggering changes in species phenology, shifting species distributions, disaggregating ecological communities, and altering habitat availability. Land use and land cover changes are driving habitat conversion, fragmenting intact habitat, and disrupting ecological service provision. Conservation practitioners must reconcile limited resources with escalating demands, and do so under deep uncertainty as landscapes are transformed and existing management paradigms are challenged. This places a premium on the capacity to select effective strategies. In this regard, simulation models are powerful tools to forecast and evaluate the potential outcomes of conservation decisions. They can help bound the conservation decision-making process, provide critical information about costs and benefits, and illuminate trends in performance that are essential to make robust conservation decisions under global change. Yet such forecasting has typically tested the static implementation of one-time conservation actions or lacked consideration of landscape dynamics. Conservation science needs forecasts that include the systematic, continuous, and dynamic implementation multiple of strategies to better support conservation planning under uncertainty.

The four research objectives in this dissertation were driven by this need. My first chapter is a review of the use of simulation modeling to inform conservation planning for intended consequences. This chapter serves to introduce a framework for integrating simulation modeling into conservation decision-making as global change threatens established management approaches and warrants the need for new conservation advances. Subsequent chapters serve as case studies for this framework. My second chapter lays the ground work for conservation strategy implementation in a spatially explicit landscape change model and demonstrates the proactive assessment of conservation strategy performance under uncertainty. Additionally, I introduced a new metric to evaluate how quickly different conservation strategies move the landscape towards, or away from, the intended target called the ‘conservation velocity’. I found that conservation velocity strongly varied both by restoration target and by the conservation strategy employed, with the highest variance occurring when longleaf pine restoration was targeted. My third chapter evaluates how the four conservation strategies implemented in

Chapter Two facilitated landscape connectivity, or the ability of a landscape to support species movement between resource patches, even when the same on the ground management actions were conducted. I found significant differences in both the degree to which strategies facilitated landscape connectivity and the ways in which connectivity was achieved, especially for

specialist species. Finally, for my fourth chapter, I evaluated the ability of conservation to

facilitate connectivity under accelerating climate and land use change. I found that in the most

extreme climate and land use change scenario evaluated, even aggressive conservation failed to

substantially improve landscape connectivity. The goal of this dissertation is to introduce a

framework to forecast landscape-level responses to conservation prior to on the ground

deployment and to use this framework to explore potential trade-offs between a suite of

conservation strategies.

by Tina Grace Mozelewski

A dissertation submitted to the Graduate Faculty of North Carolina State University in partial fulfillment of the requirements for the degree of Doctor of Philosophy

Fisheries, Wildlife, and Conservation Biology

Raleigh, North Carolina 2021

APPROVED BY:

______Robert Scheller Jodi Forrester Committee Chair

______Barry Goldfarb Adam Terando

______Martha Reiskind

DEDICATION

Let everything that I have, and everything that I am, praise the Lord.

BIOGRAPHY

Tina was born and raised in Hazelwood, Missouri. From a young age, she absolutely loved animals and could often be found giving well-researched presentations to her parents in an attempt to convince them to let her get a new pet. Tina attended undergrad at Saint Louis

University where she completed dual Bachelors of Science degrees in Biology and

Environmental Science. While at St. Louis U., Tina discovered a love of conservation biology and knew that it was the career path for her. She received her master’s from the University of

Missouri-St. Louis in Ecology, Evolution, and Systematics before moving to the desert southwest to work as a wildlife biologist. This time in the desert was impactful and inspired Tina to pursue her PhD studying how to incorporate global change dynamics into conservation planning for more sustainable biodiversity conservation. She began her doctoral degree program in 2016 under the direction of Dr. Robert Scheller, where she pursued her interests in integrating conservation decision-making with landscape change modeling to support conservation robust to global change.

iii

ACKNOWLEDGMENTS

I’d like to thank my friends, my family, and Aaron for their love, support, and occasionally, commiseration as I’ve followed this crazy conservation biology dream of mine. I could not have come this far without them cheering me on, spurring me forward, and picking me up when I needed it. I’m especially grateful for my three nieces, who bring me so much joy and for whom I hope I can demonstrate that following your dreams will always be worth it. I count myself incredibly blessed to have found amazing community every place I’ve lived on this journey to become a conservation biology professor. From Yuma to Las Vegas to Portland to Raleigh, words can’t express my gratitude for the people who have welcomed me into their lives, around their kitchen tables, and into their circles of friends. From the bottom of my heart, thank you. My fellow FER graduate students, especially Melinda Martinez and Elly Gay, have been the source of so much laughter and support. They have truly made my graduate experience so much more fun, and well, survivable. Much inspiration, comradery, and assistance have come from past and present colleagues in the Dynamic Ecosystems and Landscapes Lab. I’m particularly thankful for

Zachary Robbins, without whom I may still not have a working model. I also sincerely appreciate the time and effort put in by my committee to help with manuscript revisions, references, and career advice. Additionally, I want to thank George Hess letting me learn from his teaching brilliance. Finally, I am impossibly grateful for my advisor, Robert Scheller. I can’t imagine pursuing a PhD without his guidance, unwavering support, and friendship.

TABLE OF CONTENTS

LIST OF TABLES ...... vii LIST OF FIGURES ...... viii Chapter 1: Forecasting for Intended Consequences...... 1 1.1 Abstract ...... 1 1.2 Introduction ...... 1 1.3 The Role of Forecasting ...... 2 1.4 Current Applications of Forecasting for Conservation Innovation...... 5 1.4.1 Assessing Conservation Innovations in Silico ...... 5 1.4.2 Facilitating the Decision-Making Process ...... 6 1.4.3 Business-as-Usual Management ...... 7 1.5 Case Studies ...... 7 1.5.1 Forecasting for Restoration Planning...... 7 1.5.2 Protecting Threatened or Endangered Populations ...... 9 1.5.3 Consequences of Reintroductions and Translocations ...... 10 1.6 Emerging Opportunities to Forecast Intended Consequences ...... 12 1.7 Caveats and Room for Improvement ...... 16 1.8 Summary ...... 21 1.9 References ...... 23

Chapter 2: Conservation Velocity ...... 34 2.1 Abstract ...... 34 2.2 Introduction ...... 35 2.3 Methods...... 37 2.3.1 Study Area ...... 37 2.3.2 Model Description and Parameterization...... 39 2.3.3 Conservation Scenarios ...... 42 2.3.3.1 Economic ...... 43 2.3.3.2 Cluster ...... 44 2.3.3.3 Geodiversity ...... 45 2.3.3.4 Opportunistic...... 47 2.3.4 Habitat Targets ...... 48 2.3.5 Conservation Velocity ...... 49 2.4 Results ...... 50 2.4.1 Habitat Quality Score ...... 50 2.4.2 Conservation Velocity ...... 52 2.5 Discussion ...... 54 2.5.1 Limitations and Future Research Directions...... 57 2.6 References ...... 59

Chapter 3: Forecasting Conservation Strategy Selection Influences on Landscape Connectivity ...... 67 3.1 Abstract ...... 67 3.2 Introduction ...... 68 3.3 Methods...... 70

3.3.1 Study Area ...... 70 3.3.2 Conservation Strategies ...... 72 3.3.3 Species and Replicates ...... 74 3.3.4 Landscape Modeling ...... 74 3.3.5 Connectivity ...... 77 3.4 Results ...... 81 3.4.1 Nodes and Links ...... 81 3.4.2 Equivalent Connected Area ...... 83 3.4.3 ECAintra, ECAdirect, and ECAstep ...... 87 3.5 Discussion ...... 89 3.5.1 Limitations and Future Work ...... 92 3.5.2 Conclusions ...... 94 3.6 References ...... 96

Chapter 4: Conservation Contributions to Landscape Connectivity under Global Change ...... 103 4.1 Abstract ...... 103 4.2 Introduction ...... 103 4.3 Methods...... 106 4.3.1 Study Area ...... 106 4.3.2 Landscape Change Modeling ...... 107 4.3.3 Conservation Scenarios ...... 110 4.3.4 Connectivity Analysis ...... 111 4.4 Results ...... 114 4.4.1 Nodes and Links ...... 114 4.4.2 Equivalent Connected Area ...... 117 4.4.3 ECAintra, ECAdirect, and ECAstep ...... 120 4.5 Discussion ...... 123 4.5.1 Limitations and Future Work ...... 126 4.5.2 Conclusion ...... 127 4.6 References ...... 129

Appendices ...... 135 Appendix A ...... 136 Appendix B ...... 150 Appendix C ...... 152 Appendix D ...... 153

LIST OF TABLES

Table 1.1 Risks that can hinder the utility and feasibility of simulation modeling for conservation and restoration innovation, and how to deal with those risks for better modeling outcomes ...... 18

Table 2.1 Harvest and restoration prescriptions simulated on the study landscape in central North Carolina ...... 41

Table 2.2 Relative percent of each conservation velocity to the fastest conservation velocity (% max Vcons) for each restoration target ...... 53

Table 3.1 Harvest and restoration prescriptions simulated on the study landscape in central North Carolina ...... 76

Table 3.2 Changes to Equivalent Connected Area (ECA) over time for each conservation strategy (cluster, economic, geodiversity, and opportunistic) and species guild (habitat specialists and generalists with 1000m and 5000m dispersal distances) .... 85

Table 4.1 Harvest and restoration prescriptions simulated on the study landscape in central North Carolina ...... 109

Table 4.2 Changes to Equivalent Connected Area (ECA) over time for under RCP 4.5 with and without land use change (LUC) and conservation ...... 118

Table 4.3 Changes to Equivalent Connected Area (ECA) over time for under RCP 8.5 with and without land use change (LUC) and conservation ...... 120

Table A.1 Comparison of our forest stand sizes to private land owner stand data collected by North Carolina State University’s Forestry Extension specialists ...... 144

Table A.2 Detailed harvest and restoration prescriptions ...... 145

Table B.2 Resistance values used to characterize landscape resistance to movement for specialist and generalist species guilds ...... 150

Table C.1 Probability of selection weights used for forest stand selection under the geodiversity strategy ...... 152

Table C.2 Probability of selection weights used for forest stand selection under the cluster strategy ...... 152

vii

LIST OF FIGURES

Figure 1.1 Workflow describing the use of forecasting to assess the intended consequences of conservation and restoration actions and foster conservation innovation...... 4

Figure 2.1 Study area in central North Carolina for assessing conservation strategies using a spatially explicit modeling framework, land cover types derive from NLCD 2016 data...... 39

Figure 2.2 Workflow of conservation actions integrated into a spatially explicit landscape change model parameterized for the study area ...... 42

Figure 2.3 Cost per hectare of land across the study area, used in the selection of forest stands for conservation under the economic strategy ...... 44

Figure 2.4 Forest stands that fell within 100m, 500m, 1000m, and 2500m buffers of an already established protected area, used in the selection of forest stands for conservation under the cluster strategy ...... 45

Figure 2.5 Geodiversity scores across the study area as described in TNC’s Resilient and Connected Landscapes map, used in the selection of forest stands for conservation under the geodiversity strategy ...... 47

Figure 2.6 Comparison of changes in average pre- and post-restoration habitat quality scores across conservation strategies and restoration targets ...... 51

Figure 2.7 The conservation velocity of each conservation strategy and restoration target, averaged across model replicates ...... 53

Figure 2.8 Comparison of the conservation velocities of each conservation strategy for all three restoration targets, plotted with a fixed intercept. Targets were: a) longleaf restoration b) pine mix restoration and c) hardwood mix restoration ...... 54

Figure 3.1 Study area in Central North Carolina for assessing the influence of conservation strategies on landscape connectivity, land cover types derived from NLCD 2016 data ...... 71

Figure 3.2 Workflow of conservation actions integrated into a spatially explicit landscape change model parameterized for the study area, with model outputs of future forest habitat projections used to conduct a graph theory network analysis using Conefor ...... 80

Figure 3.3 Habitat networks from 2020-2100 showing the number of habitat nodes (panels a through d) and links (panels e through h) for four conservation strategies: cluster (a, d), economic (b, f), geodiversity (c, g), and opportunistic (d, h). Metrics are shown in blue for specialist species and orange for generalist species. Darker

viii

colors in the bottom row of graphs refer to short-distance (1000m) dispersal organisms and lighter colors refer to long-distance (5000m) dispersers ...... 82

Figure 3.4 Equivalent connected area metrics and ECA components for each species guild and conservation strategy from 2020 to 2100. Metrics are shown in blue for specialist species and orange for generalist species. Darker colors refer to short- distance (1000m) dispersal organisms and lighter colors refer to long-distance (5000m) dispersers ...... 88

Figure 3.5 Differences in habitat networks between the cluster and economic strategies for long dispersal habitat specialists ...... 90

Figure 3.6 Differences in habitat networks between the cluster and economic strategies for long dispersal habitat specialists ...... 91

Figure 4.1 The study area in central North Carolina, shown with abbreviated NLCD 2016 land use types ...... 107

Figure 4.2 Habitat networks from 2020-2100 showing the number of habitat nodes (panels a and b) and links between nodes (panels c and d) for four scenarios of global change (RCP 4.5 and 8.5 with and without land use change) with and without conservation (conservation strategies included clustering conservation around established conservation cores and prioritizing geodiverse lands for conservation ...... 116

Figure 4.3 Equivalent connected area indices (panels a and b) and ECA components (panels c through h) for global change scenarios and conservation strategies from 2020-2100 ...... 122

Figure 4.4 Habitat network changes under RCP 8.5 with and without conservation for both no land use change and land use change scenarios ...... 124

Figure A.1 Difference in biomass between LANDIS-II initial communities map and Wilson et al. (2013) maps of live tree species basal area for the study landscape ...... 137

Figure A.2 Top 20 species by biomass in FIA plots in North Carolina, South Carolina, and Virginia ...... 138

Figure A.3 Single cell, single species calibration of simulated aboveground biomass for red maple (AcerRubr) and white oak (QuerAlba), validated against the top 25% of FIA plots by aboveground carbon per age to represent ideal growing conditions ...... 140

Figure A.4 Calibration of leaf area index (LAI) for red maple and white oak single cells, validated against oak/gum/cypress forest values from He et al. 2012 ...... 140

Figure A.5 National Atmospheric Deposition Program nitrogen deposition data for the past 20 years plotted as a function of annual precipitation from U.S. Climate Data to parameterize atmospheric nitrogen slope intercept in the model ...... 141

Figure A.6 Management units without conservation strategies added to landscape ...... 143

Figure D.1 Conceptual depiction of habitat node identification by grouping contiguous habitat cells that co-occurred in time steps t and t-1 discretely from those that only occurred in time step t ...... 153

Chapter 1: Forecasting for Intended Consequences

1.1 Abstract

Restoration and conservation innovations face numerous challenges that often limit widespread adoption, including uncertainty of outcomes, risk averse or status quo biased management, and unknown trade-offs. These barriers often result in cautious conservation that does not consider the true cost of impeding innovation, and overemphasizes the risks of unintended consequences versus the opportunities presented by proactive and innovative conservation, the intended consequences. Simulation models are powerful tools for forecasting and evaluating the potential outcomes of restoration or conservation innovations prior to on-the- ground deployment. These forecasts provide information about the potential trade-offs among the risks and benefits of candidate management actions, elucidating the likelihood that an innovation will achieve its intended consequences and at what cost. They can also highlight when and where business-as-usual management may incur larger costs than alternative management approaches over the long-term. Forecasts inform the decision-making process prior to the implementation of emergent, proactive practices at broad scales, lending support for management decisions and reducing the barriers to innovation. Here we review the science, motivations, and challenges of forecasting for restoration and conservation innovations.

1.2 Introduction

There are barriers to any conservation action, including cost of implementation, environmental regulations, local acceptance (or lack thereof) of intervention, and the lack of monitoring to inform conservation action selection and document system response (Knight et al.

2006). Innovative or novel conservation (e.g., genomic translocations, restoration silviculture, floodplain engineering, soil amendments) faces additional challenges that may limit widespread

adoption, including the uncertainty of outcomes and unknown trade-offs compared to established management approaches. Risk averse policy and management may avoid novel solutions when the trade-offs among risks and benefits are unclear and the uncertainty of consequences is high.

These additional barriers often result in conservation that fails to deviate from past management approaches, a status quo bias (Samuelson & Zeckhauser 1988), particularly when the perceived costs and risks of innovation are high relative to established strategies. We have personally witnessed how the same conservation and restoration practices are used repeatedly across agencies because it is easier and more acceptable to use techniques that have a proven track record of at least modest success rather than to risk the funding and social capital on a new idea.

Traditional approaches may, however, be detrimental under global change, when past management interventions may not be sufficient to maintain (or adapt) ecosystems under novel conditions (Lindenmayer et al. 2008; Hobbs et al. 2009) and an ever-changing future (Hodgson et al. 2009; Polasky et al. 2011; Kujala et al. 2013). Here we highlight how forecasting can be used to evaluate the consequences (intended and unintended) of conservation decisions and provide examples of how forecasting has been used to support conservation decision-making and implementation. We further suggest opportunities where forecasting can inform promising new conservation applications, and finally, describe the risks of forecasting and where we see opportunities for improvement to fully realize the potential of forecasting for intended consequences of conservation innovation.

1.3 The Role of Forecasting

Simulation (or 'computational’) models are powerful, well-established tools for forecasting and evaluating conservation actions, allowing managers to address some of the barriers to conservation innovation (DeAngelis et al 1998, Twilley et al. 1999, Larson et al.

2004, Nilsson et al. 2005, Ravenscroft et al. 2010, Brum et al. 2019, Tulloch et al. 2020).

Simulation models and their associated visualizations inform many critical policy and decision-

making endeavors today (Borner et al. 2018), including global climate change (Neilson et al.

2005; Lempert & Groves 2010), pandemic spread and response (Hall et al. 2007; Giordano et al.

2020), transportation planning and operation (Robinson 2012), homeland security risk

assessment (Ezell 2012), and business process efficiency and performance (Diaz et al. 2012)

among others. Such simulation models can be used for ecological forecasting, projecting changes

to ecosystems and their components in response to one or more environmental drivers.

Forecasting has rapidly developed over the past decade thanks to improvements in software and

hardware (Scheller 2018) and, combined with the rise in availability of large ecological data sets

that expand the scales of observation, is being increasingly used to examine pressing ecological

problems (Cheruvelil & Soranno 2018). Forecasts now have the capacity to project ecosystem and landscape changes in response to environmental drivers, such as climate variability or disturbance events, or in response to human actions, such as land use change or management

(NOAA 2016).

Simulation modeling for conservation decision-making is typically operated under a

‘scenario’ framework (Peterson et al. 2003; Soares-Filho et al. 2006) whereby multiple competing conservation strategies are compared using consistent inputs that represent a variety of possible futures of the modeled system. Such information can subsequently inform ongoing discussions about the proper pace and scale of management actions and can inform the design of

robust management strategies under global change (Lempert et al. 2007). The formation of

scenarios that encapsulate a range of possible futures allows forecasts of social-ecological

systems (Schlueter et al. 2012; Thompson et al. 2012) to explicitly incorporate many of the key

uncertainties of the system into the decision-making process, e.g. C02 emissions scenarios (van

Vuuren et al. 2011), land use change futures (Thompson et al. 2011), sea level rise (LaFever et al. 2007), and shifting wildfire and disturbance regimes (Borchers 2005). These scenarios are then combined with proposed management actions to capture the uncertainty that will influence management response (e.g., Scheller et al. 2011; Maxwell et al. 2020) (Figure 1.1). By directly incorporating uncertainty, land managers gain a ‘bigger picture’ focus that allows them to conceptualize a broader range of consequences, so that they become better equipped to proactively manage for the future.

Figure 1.1. Workflow describing the use of forecasting to assess the intended consequences of conservation and restoration actions and foster conservation innovation.

1.4 Current Applications of Forecasting for Conservation Innovation

1.4.1 Assessing Conservation Innovations in Silico

Forecasts of the consequences of conservation are particularly valuable for assessing emergent or experimental practices and for identifying the associated trade-offs between effort and efficacy of new strategies. Field-testing of many innovative practices is inherently expensive, and depending on the management action, can take years or decades to conduct; true replication is generally not feasible. Forecasting allows for the replicable assessment of these innovations at a pace and cost much more amenable to management timelines and budgets

(Torrubia et al. 2014). Additionally, forecasts can represent the implementation of innovative practices across an entire landscape, where the full costs and benefits of management are typically revealed. They can validate whether a novel strategy achieves its target goals (e.g. its intended consequences), how long it takes to achieve those goals, and the cost to achieve intended goals, all before deployment. In doing so, forecasting allows managers to evaluate the potential outcomes from a variety of restoration and conservation strategies on their landscapes, ranging from the familiar to the highly innovative (Perring et al. 2015). It also plays a critical role in reframing which sources of uncertainty are most important, especially for landscapes with a high potential for unprecedented change (Martin et al. 2011). As anthropogenic change accelerates and many landscape and management precedents cease to apply (Lindenmayer et al.

2008), forecasting can help managers to confront novel ecosystems and new management challenges and assess the ability of conservation innovations to achieve their intended consequences.

1.4.2 Facilitating the Decision-Making Process

Forecasting facilitates large-scale conservation and restoration planning across agency and ownership boundaries, supporting collaborative efforts, pooling of resources and knowledge, and enabling co-management among decision-makers, scientists, and stakeholders (Berkes

2009). It can support structured decision making (SDM) whereby multiple parties engaged in the decision-making process explicitly identify objectives and performance metrics, assess alternatives, and make decisions based on stakeholder values and system uncertainties (Gregory

& Long 2009). Once objectives are specified, forecasting can play a critical role in the SDM process, representing system behavior and projecting system responses to candidate management actions. Because forecasting provides information about the trade-offs among actions – both business-as-usual and more proactive approaches - and outcomes (Martin et al. 2009; Spies et al.

2017), it enables a more collaborative decision analytical framework and facilitates co- production of knowledge. Stakeholders can become full partners in the process of forecasting by sharing data and local knowledge, assessing model outcomes, and ensuring that outputs meet their needs (Armitage et al. 2011). This can be combined with participatory modeling, in which stakeholders formulate management problems and help to develop and test feasible solutions, to allow managers to collaboratively identify possible futures across a range of likelihoods and assess potential management responses (Voinov et al. 2018; Vukomanovic et al. 2019). By encouraging managers and stakeholders to consider a broader future through participating in forecasting efforts, they are given a greater license to imagine more intensive interventions, bridging the gap between risk averse and risk inclined managers.

1.4.3 Business-As-Usual Management

Finally, forecasting can also highlight when business-as-usual management will not be

sufficient and when more innovative practices will be required to achieve the intended

consequences. Projecting the approximate magnitude of intervention that could be required from

the outset might help managers to make more realistic decisions (Hobbs 2007; Brudvig 2011). It

will also highlight when managers need to deploy new practices to achieve a goal. This is

especially pertinent when the goal is to recreate a historic reference point despite anthropogenic

change, or after some threshold of irreversibility has been crossed (Aronson et al. 1993; Suding

et al. 2004; Hobbs et al. 2009; Jackson & Hobbs 2009).

1.5 Case Studies

Here we provide three applications where forecasting has been or is being used to support

restoration and conservation planning, particularly in regards to the consequences of and trade-

offs among potential actions that drive the decision-making process.

1.5.1 Forecasting for Restoration Planning

Forecasting has demonstrated utility for projecting the potential for ecological restoration on degraded landscapes (Cantarello et al. 2011) and for comparing the costs and benefits of various restoration strategies to inform restoration planning (Birch et al. 2010; Keane et al.

2017). Identifying where and how to restore a landscape after anthropogenic degradation plays a critical role in determining project success or failure. These decisions are difficult and complex, and include (but are not limited to) contending with multiple land owners, budgetary constraints, and the numerous potential restoration strategies available. Uncertainty associated with global change compounds these challenges as climate change, land use change, altered disturbance

regimes, and unknown interactions between agents of change can all influence restoration project

outcomes in unanticipated ways.

Cantarello and others (2011) used simulation modeling to examine the feasibility of

passive restoration on a tropical dryland forest landscape undergoing multiple interacting

anthropogenic disturbances. Using a spatially explicit model of forest dynamics, the authors

simulated passive restoration- allowing the natural regeneration, dispersal, and colonization of

vegetation after the removal of the causes of ecological degradation- on two landscapes in

Mexico and assessed this restoration approach under multiple disturbance regimes. This allowed for the anticipation of restoration outcomes and revealed the dynamic impacts of interacting disturbances on restoration success. Results showed that passive landscape-scale restoration was an ecologically viable option for tropical dryland forest recovery but that the combined effects of livestock grazing and fire reduced forest cover and lessened the efficacy of restoration efforts.

In another example, Birch et al. (2010) used forecasting to evaluate the ability of passive and active restoration to combat environmental degradation in Latin America. The authors compared the cost effectiveness of three restoration strategies (passive, passive with fencing and fire suppression to protect the area, and active [tree planting, fencing, and fire suppression]) by estimating the net value of ecosystem services provided under the different forest restoration scenarios and weighing this against the cost of implementation. Using a simulation model to construct these restoration scenarios, they performed a cost-benefit analysis to compare the cost of each restoration approach to the monetized estimates of a suite of ecosystem services derived from model outputs, including carbon sequestration, timber production, and tourism. These modeling efforts revealed that ecosystem service benefits varied not only by the restoration strategy used but also between study areas, though passive restoration was demonstrated to be

the most cost effective approach overall. The incorporation of forecasts enabled not only the

assessment of restoration action cost and benefit trade-offs across a suite of regionally important

ecosystem services, but also informed the selection of site-specific restoration strategies.

1.5.2 Protecting Threatened or Endangered Populations

Conservation forecasting has been extensively deployed in the context of protecting

threatened and endangered (T&E) species (e.g. Akcakaya et al. 2004; Scheller et al. 2011). The

task of recovering T&E species has grown more complicated as the active management for these

populations may conflict with other landscape goals, including restoration goals. For example, in

the western United States, the locally-distinct population of the fisher (Pekania pennanti), which

requires old-growth habitat for nesting and denning, has been historically reduced by logging and

trapping (Zieinski et al. 1995). Although fishers persist, their habitat is now fragmented and at

risk due to large and intense wildfires. This threat is anticipated to grow due to climate change

and a more active wildfire regime. Simultaneously, forest managers want to restore a fire regime

characterized by more frequent and less intense wildfires (Klimaszewski-Patterson et al. 2018) in

order to increase forest resilience to climate change (North et al. 2015). Doing so will require the

extensive use of fuel treatments, including forest thinning. Although fuel treatments will reduce

the long-term risk of fires that may be catastrophic to fisher persistence, their immediate effect

will be a reduction in local fisher habitat quality. As a result of the uncertainty of the outcomes,

fuel treatments have been widely contested by many environmental groups, who have promoted

the precautionary principle whereby interventions should be avoided unless positive outcomes

are certain.

Scheller and others (2011) used forecasting to assess the trade-offs between short-term reductions in habitat quality and long-term reductions in wildfire risk to fisher populations,

particularly given the uncertainty introduced by climate change. To do this, they coupled a

landscape-change model that simulated succession, wildfires, and fuel treatments (Syphard et al.

2011) with a metapopulation model of fisher dispersal, births, and mortality (Spencer et al.

2011). The authors forecasted and compared the direct (reduced habitat quality) and the indirect

(reduced probability of habitat loss from wildfire) effects of fuel treatments on fisher

metapopulation dynamics over time. They concluded that fuel treatments were more effective at

restoring high frequency, low severity fire regimes under climate change due to a greater

probability of intersection between fire and treatment areas (Syphard et al. 2011). And fuel

treatments reduced the loss of fisher habitat to wildfire, conferring significant benefit to fisher

populations despite the large variability and uncertainty generated by climate change (Scheller et

al. 2011). The research also highlighted the risk to non-contiguous fisher habitat (Spencer et al.

2011) and the need for reintroductions to formerly occupied areas to reduce the risk of local

extirpation. Their research demonstrates the use of forecasting to assess trade-offs between controversial management actions and conflicting management objectives before actions are undertaken.

1.5.3 Consequences of Reintroductions and Translocations

American chestnut (Castanea dentata) was decimated by fungal blight nearly a century ago and is functionally extinct today. Genetically engineered blight resistant chestnuts are actively being developed (Westbrook et al. 2020b) with the intention of releasing them onto their native landscapes. Although chestnut restoration in general is not widely opposed, some groups have opposed the use of genetic engineering to achieve this aim. Aside from the novelty of genetic manipulation to confer resistance to a wild population, a landscape-scale (extending across much of the eastern United States) restoration of this magnitude has not been previously

attempted. Prior to investment in such a significant and contested restoration effort, forecasting

can inform critical questions about restoration cost and feasibility including: How much effort

(area per year) would be required to restore chestnut to the landscape, and at what expense

(Westbrook et al. 2020a)? And what are the consequences for ecosystem structure and

functioning (Jacobs et al. 2013)? To address these considerations, Gustafson et al. (2017; 2018)

simulated the reintroduction of blight-resistant chestnut using a suite of increasingly aggressive

restoration approaches and forecasted its long-term effects on forest composition and carbon

storage in Maryland, USA. They found that restoration would need to be extensive and

committed to succeed within a reasonable time frame (50-100 years). Furthermore, they found that the reintroduced chestnut would not extirpate any existing tree species nor would it substantially alter carbon carrying capacity, providing support for this introduction of genetically engineered trees.

Climate change threatens forest health (Vose et al. 2016; Seidl et al. 2017), and facilitated migration has been proposed as a technique to ensure long-term functioning (Millar et al. 2007;

Duveneck & Scheller 2015). Similar to the landscape-scale introduction of a novel chestnut trait,

facilitated migration would need to occur at an unprecedented scale to be effective. In this case,

entirely novel communities would be created that have never existed anywhere. Facilitated

migration is not restoration but is instead the intentional creation of novel communities for the

purpose of ensuring ecosystem functionality and service provision under climate change. Ideally,

these communities would naturally perpetuate themselves as intentionally introduced species

become established so as to eliminate their long-term management overhead. Duveneck and

Scheller (2015) used a simulation framework to assess the potential for facilitated migration (aka

‘climate suitable planting’ of trees) to increase forest carbon storage and improve functional

diversity in the Midwest, USA. Although simulations of facilitated migration resulted in self- sustaining forest communities (the intentionally introduced species successfully established and reproduced) and increased forest carbon, the effects on functional diversity were projected to be negligible due to the loss of existing species on the landscape due to climate change. Lucash et al. (2017) similarly compared climate suitable planting against more traditional silvicultural approaches in the Midwest and found that climate suitable planting had the capacity to substantially increase forest resilience under climate change. In both studies, the model simulation results indicated that extensive and committed management would be required to achieve the stated goals of self-sustaining climate suitable planting.

1.6 Emerging Opportunities to Forecast Intended Consequences

The advancement of many technologies and the progression of conservation and restoration theory has created opportunities to substantially expand the utility of forecasting.

Here we outline near-term needs and opportunities where forecasting is well-positioned to improve the decision-making process, particularly given the conservation actions that are being proposed (Seddon et al. 2014).

The ability of scientists to develop forecasts that allow for the assessment and vetting of innovative practices before widespread deployment has particular resonance today because of the variety and scope of emergent conservation innovations. These innovations are energizing and exciting but also fraught with challenges due to their associated uncertainty: any true innovation is relatively untested and therefore inherently uncertain as its risks, benefits, and odds of success are unknown. Forecasting scenarios can elucidate worst-case outcomes and landscape responses to innovations. Forecasts can also help to distinguish innovations that are both practical and effective given the magnitude of forthcoming global changes from those that may not be the best

use of resources (Perring et al. 2015). Conservationists are starting to envision the ‘Accretocene’,

an era when we will begin to recover species and ecosystem functionality faster than our current

rate of loss (Wright 2020). Simulation modeling will be a valuable asset to test big ideas that will

allow us to intentionally shape the ecological trajectory of landscapes and usher in the

Accretocene.

The incorporation of large geographic areas and long time horizons combined with the

freedom to explore intensive and creative ideas at relatively low cost allows managers to take a

big picture approach to stewarding the future of their landscapes. Managers are empowered to

think about how their decisions can shape the course of their landscapes, not just in terms of

what can be accomplished in one funding cycle or even one 50-year plan. One such idea is the

concept of rewilding, which aims to increase biodiversity while reducing past and present human

impacts by restoring species and ecological processes to the landscape (Donlan 2005, Lorimer et

al. 2015, Svenning et al. 2016). A central focus of the rewilding movement is to use species

introductions, reintroductions, or de-extinctions to address trophic cascades (Fuhlendorf et al.

2009; Lorimer et al. 2015). De-extinction (the resurrection of an extinct species, Shapiro 2017)

and novel species introduction (Schlaepfer et al. 2011) are exciting but potentially risky concepts

that would benefit from forecasting, which could be used to simulate the addition of a species to

a new system. Managers and decision-makers could then assess whether de-extinction is feasible/practical, or the potential consequences from the reintroduction of an extinct species or the introduction of a species previously not found in that ecosystem (i.e. a functional trait-based approach to restoration, Funk et al. 2008; Laughlin 2014). For species for which we have little data (e.g., an extinct species), forecasts can use functional information, validated with information from closest living relatives, to estimate plausible reintroduction outcomes. If de-

extinction efforts fail across all scenarios, or only succeed with heavy, continuous management intervention, this may be cause for reconsideration (Nogués-Bravo et al. 2016). Likewise, if the introduction of a novel species fails to fill the intended ecological niche, managers may be more inclined to consider a different course of action. In the chestnut case study above, we highlight how forecasting has been used to inform the management required in tandem with functional de- extinction (a species that has been functionally removed from a landscape, even if some survivors persist in a diminished role).

Similarly, conservationists should consider forecasting the adaptive capacity of new, genetically engineered or modified species or genomic translocations and their effects at landscape scales prior to widespread use. Such innovations have been proposed to accelerate population and landscape responses to rapid change, but the outcomes of largescale implementation are largely unknown (Rice & Emery 2003). Considering the consequences of genomic interventions at the landscape scale is currently outside the scope of most simulation models. Modest but essential new investment is necessary to sufficiently capture the spread of the intervention, including potential hybridization with the existing population. The scaffolding already exists in most simulation models, although modifications are required. As an example, if simulating genomic intervention of plants, a new plant trait would need to be added that indicates whether the genome has been altered and its effects on other traits already represented or to contrast the various possibilities of an unknown trait. If simulating the genomic intervention of an animal population, it would be critical to capture its effect on population and metapopulation dynamics.

Forecasting could also prove useful for the assessment of landscape and ecosystem responses to aggressive, intensive, and largescale management efforts like those required for the

removal of invasive species or the remediation of ecological degradation. In these scenarios, the

management investments required to achieve targeted outcomes are time and money intensive.

By simulating a spectrum of removal approaches and forecasting inter-specific interactions and

species responses to management and environmental conditions, managers can estimate what

species will re-colonize and dominate after invasive plants or animals are removed; how much

management effort will be required for eradication; and which management strategies resulted in

the strongest and most long-lasting outcomes. Critically, it may also be able to inform the

unintended consequences and secondary effects of invasive species removal in more detail than

qualitative evaluation, especially in instances of multiple invaders (Zavaleta et al. 2001). Using

forecasting to answer these questions can provide important information for deciding whether

invasive species removal is practicable and/or the best use of resources. On highly degraded

areas, land managers can also use forecasting to assess what is feasible and the amendments or

interventions that may be necessary to enable restoration. For example, soil amendments have

been proposed and tested on small scales to restore ecological function (e.g., Silver et al. 2018;

Hemes et al. 2019). Forecasting could assess large-scale rollouts of different soil amendments or seed enhancement technologies (e.g. Clemente et al. 2004; Madsen et al. 2016) in different proportions, assessing vegetation growth and recruitment responses, to optimize restoration outcomes and vet the cost of implementation. Similarly, forecasting could be used to assess the

benefits of intermediate planting that might be required for soil remediation before restoration

planting, evaluating restoration project response with and without remediation efforts (Stanturf et

al. 2014). For hydrologic engineering, forecasting could include climate change forecasts to

assess long-term efficacy before substantial investments in infrastructure are made (Day Jr. et al.

2005; Poff et al. 2016).

1.7 Caveats and Room for Improvement

Despite their promise, simulation models are not always the appropriate tool and their deployment requires a nontrivial investment of resources that in some instances may be better allocated. Forecasts are not appropriate when the conservation situation is dire and/or immediate.

When the ecological consequences of doing nothing are very large and the problem is urgent, the risks of action are clearly justified and forecasting would only delay a response. For example, removing rodents from small oceanic islands has immediate, large, and proven benefits for biodiversity (Howald et al. 2007; Keitt et al. 2011). The value of information provided by forecasting would, in this case, be small relative to the cost of delaying action. Though often less expensive and less time-intensive than full-scale, on-the-ground experimentation, developing forecasts still requires considerable time invested in data collection, parameterization, and validation (testing against empirical data) that would slow conservation response to an emergency situation. The information to be gained from forecasting should be weighed against the seriousness and immediacy of the conservation need when deciding its usefulness. Under such circumstances, alternatives such as expert opinion should be considered.

Additionally, forecasting comes with its own set of risks and uncertainties that can constrain the use and usefulness of modeling efforts (Table 1.1). Insufficient data and software bugs can curtail the use of simulation modeling altogether. Models operated on the wrong scale or across scenarios that are too timid can reduce the value of information they provide to managers and fail to contribute meaningfully to conservation and restoration decision-making.

Uncertainty surrounding future drivers of ecological change can instill a false sense of confidence that the model is capturing a range of representative possible futures. Fundamental misunderstandings of the system being modeled can produce model outputs that fail to reflect

current and future trajectories of the system, leading to inappropriate or ineffective management decisions. Still, there are potential solutions to reduce these risks and improve model performance to the benefit of future conservation and restoration planning and innovation.

Table 1.1. Risks that can hinder the utility and feasibility of simulation modeling for conservation and restoration innovation, and how to deal with those risks for better modeling outcomes. Forecasting risk Consequences Potential solutions If the decision-makers do not agree on the value An incremental approach can build confidence in the of forecasting or on the correct approach, the forecasting approach by iteratively testing larger landscapes, Insufficient buy- results will be contested and resources will be more conservation alternatives, and/or including more future in from key wasted. Too frequently, forecasters overwhelm drivers of change. Frequent and early sharing of forecast stakeholders decision-makers with complexity before they results allows decision-makers to learn about the merits and are comfortable with the process. limitations of the knowledge produced. Without sufficient data, simulation model Ongoing monitoring and data collection can fill in data gaps parameterization is either infeasible or results in and allow simulation models to better capture system Insufficient data models that fail to replicate system dynamics dynamics. The rise in availability of large-scale ecological closely enough that model outputs are useful for datasets is also shrinking prospective data limitations. making decisions about that system. Global change is laden with uncertainty because Scenarios of global change should contain a broad range of long-term actions and policies are uncertain. In plausible futures therefore providing a ‘stress test’ of Future drivers of addition, forecasts cannot accommodate all proposed conservation actions. These scenarios do not change are uncertainties; there are many unknown typically include all possible futures as represented by black unpredictable unknowns (‘black swans’) that cannot be swans. Decision-makers must be provided the proper context: anticipated or incorporated, into simulation All decisions are made with a large degree of uncertainty models (Taleb 2007). about the future. Modelers should assemble a diverse group of stakeholders to When scenarios are too timid, they often tell co-design a spectrum of scenarios that range from Scenarios are too modelers and stakeholders what they already conservative to highly innovative (McBride et al. 2017). timid knew rather than allowing them to learn from Facilitators are often required to ensure that decision-makers potential ecological changes in the future. are considering options beyond what is immediately possible given resource constraints. If forecasting informs critical decisions or resource Increasing model allocations, conservation organization should seek out open- Software bugs in the model can produce complexity source models with a rigorous design and proven testing spurious results or can result in unnecessary raises the risk of (Scheller et al. 2010). In addition, open-access software delays if the bugs cannot be quickly resolved. software bugs repositories provide transparency and ready reporting of software errors.

Table 1.1 (continued).

The utility of forecasts and their appropriate Selecting a model Select a model that matches the scale of the question(s) applications are scale-dependent. Every that was designed asked. Avoid model selection based on external pressures or forecasting tool has an optimal scale for which for use at a favoritism. Large ecological questions that cross multiple it was designed and at which its expected bias different scale scales may require the integration of multiple models. is minimized (Obeysekera & Rutchey 1997). Decision-makers may develop a false sense of Forecasters need to work closely with decision-makers to Decision-makers accuracy, a belief that the model is capturing ensure that they are aware of all sources of uncertainty (e.g., do not understand the real world. Conversely, decision-makers Dietze 2017) before forecasts are produced. Multiple forecasting may reject model outcomes if they do not iterations allow decision-makers to provide feedback uncertainty understand the uncertainties in the data or throughout the process. future drivers of change. When managers and decision-makers are The use of a decision theory framework can help managers unclear about management objectives, Unclear set explicit objectives prior to any modeling work, enabling forecasting is often used to ask/inform the management forecasting that asks questions useful for informing wrong questions that ultimately do not objectives management decisions and supporting management contribute meaningfully to management objectives (Milner-Gulland & Shea 2017). decisions. Fundamental misunderstandings about the Model validation using either a subset of reserved data or system being modeled can lead to erroneous historical trajectories is essential to determine if model Misunderstanding assumptions about model selection and outputs are an acceptable/appropriate representation of the system dynamics parameterization, yielding model results that system’s current conditions. If forecasts fail to reflect system do not reflect the future trajectory of the dynamics, further observation of the system is needed to system. inform model development.

We also do not want to imply that forecasting is capable of addressing all management

needs, particularly as new innovations are imagined, developed, and delivered at an ever

increasing pace. Many improvements are necessary to fully realize the potential of forecasting

for conservation and restoration innovation. As an example, the integration of forecasting could

streamline the adaptive management process. Adaptive management is a continuous loop of

management strategy design, implementation, and monitoring used to systematically test

assumptions, learn from past efforts, and adapt management actions to improve outcomes

(Salafsky & Margoluis 2001). While adaptively managed conservation is the gold standard, it has proven difficult to effectively implement due to issues of multiple scales and stakeholders

(Gregory et al. 2006) and funding and time limitations (Westgate et al. 2013). Incorporating

forecasting into the adaptive management framework can reduce time and funding constraints by

virtually assessing potential adaptive management strategies prior to implementation to inform

management action selection. However, doing so will require forecasts that are embedded within

the conservation process and the minimization of many of their barriers: cumbersome interfaces,

high programming skill requirements, data availability, and data management, among others.

Additionally, integrating remote sensing and artificial intelligence-powered monitoring into

forecasting would allow managers to proactively manage landscapes across larger scales. We see

these as substantial and much needed improvements for forecasting.

Forecasting outcomes of conservation interventions would also benefit from a more fully

realized integration of social-ecological systems into simulation modeling, including positive and

negative feedback loops (Jacobs et al. 2013; Levin et al. 2013). The science of ‘ecological

forecasting’ (Clark et al. 2001) has made tremendous progress. The science of ‘social-ecological

forecasting’ is in its relative infancy. Most social-ecological forecasting is ‘top-down’ whereby

the social system influences the ecological system but not vice-versa. Recent advances in social- ecological forecasting suggest that a shift is on the horizon (Sotnik 2018) although these advances haven’t reached the maturity necessary to be incorporated into restoration and conservation planning (Bolte et al. 2007). A greater incorporation of complex human decision- making and its associated uncertainties, and the inclusion of more explicit, detailed feedbacks between social and ecological systems are needed to truly capture the influences of management on social-ecological systems (Schlueter et al. 2012).

Finally, particular attention needs to be focused on how the community of managers engages with simulation models and their outputs to ensure that forecasts are easily accessible, interpretable, and useful for decision-making. Recent advances in visualization offer a way to make findings more approachable to managers, including the use of tangible landscapes and virtual reality to translate model outputs into observable landscape elements (Lewis & Sheppard

2006; Tabrizian et al. 2016, Huang et al. 2019). They can also more tangibly connect managers with the realities and challenges of the novel landscapes they may be managing in the future

(Rubio-Tamayo et al. 2017). However, these advances in immersive visualization are not yet widely utilized, in part because their application to ecology and environmental science is in its relative infancy (Huang et al. 2020). Providing platforms for enhanced visualization of model outputs and broadening the use of immersive experiences holds great promise for improved accessibility and more effective communication with managers and decision-makers.

1.8 Summary

Global change is adding substantial uncertainty to the management of landscapes for biodiversity conservation, making it harder for managers to anticipate which conservation and restoration approaches are most likely to achieve their intended consequences. This is especially

true for emergent or novel management interventions, the consequences of which are often

unknown. Forecasting allows managers to test candidate management actions across a range of

possible futures and uncertainties and receive feedback on performance, often in less time and at

less expense than on-the-ground experimentation. In this way, forecasting can facilitate the adoption of innovative new management strategies by equipping managers with information about potential trade-offs and response trends, crucial considerations when deciding how to invest limited conservation resources. As landscape change accelerates, forecasting can help both

risk averse and risk inclined managers to visualize and evaluate potential outcomes of

conservation and restoration innovations, supporting management advances under global change.

1.9 REFERENCES

Akçakaya HR, Radeloff VC, Mladenoff DJ, Hong HS. 2004. Integrating landscape and metapopulation modeling approaches: Viability of the sharp‐tailed grouse in a dynamic landscape. Conservation Biology 18:526-537.

Armitage D, Berkes F, Dale A, Kocho-Schellenberg E, Patton E. 2011. Co-management and the co-production of knowledge: Learning to adapt in Canada's Arctic. Global Environmental Change 21:995-1004.

Aronson J, Floret C, Le Floch E, Ovalle C, Pontanier R. 1993. Restoration and rehabilitation of degraded ecosystems in arid and semi-arid lands. I. A view from the South. Restoration Ecology 1:8-17.

Berkes F. 2009. Evolution of co-management: Role of knowledge generation, bridging organizations and social learning. Journal of Environmental Management 90:1692-1702.

Birch JC, Newton AC, Aquino CA, Cantarello E, Echeverría C, Kitzberger T, Schiappacasse I, Garavito NT. 2010. Cost-effectiveness of dryland forest restoration evaluated by spatial analysis of ecosystem services. PNAS 107:21925-21930.

Bolte JP, Hulse DW, Gregory S, Smith C. 2007. Modeling biocomplexity–actors, landscapes and alternative futures. Environmental Modelling & Software 22:570-579.

Borchers JG. 2005. Accepting uncertainty, assessing risk: Decision quality in managing wildfire, forest resource values, and new technology. Forest Ecology and Management 211:36-46.

Börner K, Rouse WB, Trunfio P, Stanley HE. 2018. Forecasting innovations in science, technology, and education. PNAS 115:12573-12581.

Brudvig LA. 2011. The restoration of biodiversity: Where has research been and where does it need to go? American Journal of Botany 98:549–558.

Brum FT, Pressey RL, Bini LM, Loyola R. 2019. Forecasting conservation impact to pinpoint spatial priorities in the Brazilian Cerrado. Biological Conservation 240:108283.

Cantarello E, Newtona AC, Hill RA, Tejedor-Garavito N, Williams-Linera G, López-Barrera F, Manson RH, Golicher DJ. 2011. Simulating the potential for ecological restoration of dryland forests in Mexico under different disturbance regimes. Ecological Modeling 222:1112-1128.

Cheruvelil KS, Soranno PA. 2018. Data-Intensive Ecological Research Is Catalyzed by Open Science and Team Science. BioScience 68:813–822.

Clark JS, Carpenter SR, Barber M, Collins S, Dobson A, Foley JA, Lodge DM, Pascual M, Pielke R, Pizer W, Pringle C. 2001. Ecological forecasts: an emerging imperative. Science 293:657-660.

Clemente AS, Werner C, Maguas C, Cabral MS, Martins-Loucao MA, Correia O. 2004. Restoration of a limestone quarry: Effect of soil amendments on the establishment of native Mediterranean sclerophyllous shrubs. Restoration Ecology 12:20-28.

Day Jr. JW, Barras J, Clairain E, Johnston J, Justic D, Kemp GP, Ko JY, Lane R., Mitsch WJ, Steyer G, Templet P, Yañez-Aarancibia A. 2005. Implications of global climatic change and energy cost and availability for the restoration of the Mississippi delta. Ecological Engineering 24:253-365.

DeAngelis DL, Gross LJ, Huston MA, Wolff WF, Fleming DM, Comiskey EJ, Sylvester SM. 1998. Landscape modeling for everglades ecosystem restoration. Ecosystems 1998:64- 75.

Diaz R, Behr JG, Tulpule M. 2012. Business Process Modeling. In: (Handbook of Real-World Applications in Modeling and Simulation), {[eds.] [Sokolowski JA, Banks CM.]}. John Wiley & Sons, Hoboken, 206-260.

Dietze MC. 2017. Ecological forecasting. Princeton University Press.

Donlan J. 2005. Re-wilding North America. Nature 436:913-914.

Duveneck MJ, Scheller RM. 2015. Climate‐suitable planting as a strategy for maintaining forest productivity and functional diversity. Ecological Applications 25:1653-1668.

Ezell BC. 2012. Homeland security risk modeling. Business Process Modeling. In: (Handbook of Real-World Applications in Modeling and Simulation), {[eds.] [Sokolowski JA, Banks CM.]}. John Wiley & Sons, Hoboken, 129-164.

Fischer J, Peterson GD, Gardner TA, Gordon LJ, Fazey I, Elmqvist T, Felton A, Folke C, Dovers S. 2009. Integrating social-ecological resilience and optimisation for conservation. Trends in Ecology and Evolution 24:549–554.

Fuhlendorf SD, Engle DM, Kerby J, Hamilton R. 2009. Pyric Herbivory: Rewilding Landscapes through the Recoupling of Fire and Grazing. Conservation Biology 23:588-598. 24

Funk JL, Cleland, EE, Suding KN, Zavaleta ES. 2008. Restoration through reassembly: plant traits and invasion resistance. Trends in Ecology and Evolution 23:695-703.

Giordano G, Blanchini F, Bruno R, Colaneri P, Di Filippo A, Di Matteo A, Colaneri M. 2020. Modelling the COVID-19 epidemic and implementation of population-wide interventions in Italy. Nature Medicine 26:855-860.

Gregory R, Long G. 2009. Using structured decision making to help implement a precautionary approach to endangered species management. Risk Analysis 29:518–32.

Gregory R, Ohlson D, Arvai J. 2006. Deconstructing adaptive management: Criteria for applications to environmental management. Ecological Applications 16:2411-2425.

Gustafson EJ, de Bruijn A, Lichti N, Jacobs DF, Sturtevant BR, Foster J, Miranda BR, Dalgleish HJ. 2017. The implications of American chestnut reintroduction on landscape dynamics and carbon storage. Ecosphere 8, https://doi.org/10.1002/ecs2.1773.

Gustafson EJ, Sturtevant BR, de Bruijn AM, Lichti N, Jacobs DF, Kashian DM, Miranda BR, Townsend PA. 2018. Forecasting effects of tree species reintroduction strategies on carbon stocks in a future without historical analog. Global Change Biology 24:5500- 5517.

Hall IM, Gani R, Hughes HE, Leach S. 2007. Real-time epidemic forecasting for pandemic influenza. Epidemiology & Infection 135:372-385.

Hemes KS, Chamberlain SD, Eichelmann E, Anthony T, Valach A, Kasak K, Szutu D, Verfaillie J, Silver WL, Baldocchi DD. 2019. Assessing the carbon and climate benefit of restoring degraded agricultural peat soils to managed wetlands. Agricultural and Forest Meteorology 268:202-214.

Hobbs RJ. 2007. Setting effective and realistic restoration goals: key directions for research. Restoration Ecology 15:354–357.

Hobbs RJ, Higgs E, Harris JA. 2009. Novel ecosystems: implications for conservation and restoration. Trends in Ecology and Evolution 24:599-605.

Hodgson JA, Thomas CD, Wintle BA, Moilanen A. 2009. Climate change, connectivity and conservation decision making: back to basics. Journal of Applied Ecology 46:964-969.

Howald G, Donlan CJ, Galván JP, Russell JC, Parkes J, Samaniego A, Wang Y, Veitch D, Genovesi P, Pascal M, Saunders A, & Tershy B. 2007. Invasive rodent eradication on islands. Conservation Biology 21:1258–1268.

Huang J, Lucash MS, Scheller RM, Klippel A. 2020. Walking through forests of the future: using data-driven virtual reality to visualize forests under climate change. International Journal of Geographical Information Science, doi: 10.1080/13658816.2020.1830997.

Huang J, Lucash MS, Simpson MB, Helgeson C, Klippel A. 2019. Visualizing natural environments from data in virtual reality: Combining realism and uncertainty. In: 2019 IEEE Conference on Virtual Reality and 3D User Interfaces (VR), 23–27 March 2019 Osaka. Piscataway, NJ: IEEE, 1485–1488.

Jacobs DF, Dalgleish HJ, Nelson CD. 2013. A conceptual framework for restoration of threatened plants: the effective model of American chestnut (Castanea dentata) reintroduction. New Phytologist 197:378–393.

Jackson ST, Hobbs RJ. 2009. Ecological restoration in the light of ecological history. Science 325:567-569.

Keane RE, Holsinger LM, Mahalovich MF, Tomback DF. 2017. Evaluating future success of whitebark pine ecosystem restoration under climate change using simulation modeling. Restoration Ecology 25:220-233.

Keitt B, Campbell K, Saunders A, Clout M, Wang Y, Heinz R, Newton KM, Tershy B. 2011.The global islands invasive vertebrate eradication database: A tool to improve and facilitate restoration of island ecosystems. In: (Island invasives: Eradication and management), {[eds.] [Veitch CR, Clout MN, Towns DR.]}. IUCN, Gland, 74-77.

Klimaszewski-Patterson A, Weisberg PJ, Mensing SA, Scheller RM. 2018. Using paleolandscape modeling to investigate the impact of Native American-set fires on Pre- Columbian forests in the southern Sierra Nevada, California, USA. Annals of the American Association of Geographers 108:1635-1654.

Knight AT, Cowling RM, Campbell BM. 2006. An operational model form implementing conservation action. Conservation Biology 20:408-419.

Kujala H, Burgman MA, Moilanen A. 2013. Treatment of uncertainty in conservation under climate change. Conservation Letters 6:73-85.

LaFever DH, Lopez RR, Feagin RA, Silvy NJ. 2007. Predicting the impacts of future sea-level rise on an endangered lagomorph. Environmental Management 40:430-437.

Larson MA, Thompson III FR, Millspaugh JJ, Dijak WD, Shifley SR. 2004. Linking population viability, habitat suitability, and landscape simulation models for conservation planning. Ecological Modeling 180:103-118.

Laughlin DC. 2014. The intrinsic dimensionality of plant traits and its relevance to community assembly. Ecology 102:186-193.

Lempert RJ, Groves DG. 2010. Identifying and evaluating robust adaptive policy responses to climate change for water management agencies in the American west. Technological Forecasting & Social Change 77:960-974.

Lempert RJ, Groves DG, Popper SW, Bankes SC. 2007. A general, analytic method for generating robust strategies and narrative scenarios. Management Science 52:514-528.

Levin S, Xepapadeas T, Crépin AS, Norberg J, de Zeeuw A, Folke C, Hughes T, Arrow K, Barrett S, Daily G, Ehrlich P, Kautsky N, Mäler KG, Polasky S, Troell M, Vincent JR, Walker B. 2013. Social-ecological systems as complex adaptive systems: modeling and policy implications. Environment and Development Economics 18:111– 132.

Lewis JL, Sheppard SRJ. 2006. Culture and communication: Can landscape visualization improve forest management consultation with indigenous communities? Landscape and Urban Planning 77:291-313.

Lindenmayer DB, Fischer J, Felton A, Crane M, Michael D, Macgregor C, Montague‐Drake R, Manning A, Hobbs RJ. 2008. Novel ecosystems resulting from landscape transformation create dilemmas for modern conservation practice. Conservation Letters 1:129–135.

Lorimer J, Sandom C, Jepson P, Doughty C, Barua M, Kirby KJ. 2015. Rewilding: Science, Practice, and Politics. Annual Review of Environment and Resources 40:39-62.

Lucash MS, Scheller RM, Gustafson EJ, Sturtevant BR. 2017. Spatial resilience of forested landscapes under climate change and management. Landscape Ecology 32:953-969.

Madsen MD, Davies KW, Boyd CS, Kerby JD, Svejcar TJ. 2016. Emerging seed enhancement technologies for overcoming barriers to restoration. Restoration Ecology 24:S77–S84.

Martin J, Runge MC, Nichols JD, Lubow BC, Kendall WL. 2009. Structured decision making as a conceptual framework to identify thresholds for conservation and management. Ecological Applications 19:1079-1090.

Martin J, Fackler PL, Nichols JD, Lubow BC, Eaton MJ, Runge MC, Stith BM, Langtimm CA. 2011. Structured decision making as a proactive approach to dealing with sea level rise in Florida. Climate Change 2011:185-202.

Maxwell CJ, Serra-Diaz JM, Scheller RM, Thompson JR. 2020. Co-designed management scenarios shape the responses of seasonally dry forests to changing climate and fire regimes. Journal of Applied Ecology 57:1328-1340.

McBride MF, Lambert KF, Huff ES, Theoharides KA, Field P, Thompson JR. 2017. Increasing the effectiveness of participatory scenario development through codesign. Ecology and Society 22:16. https://doi.org/10.5751/ES-09386-220316.

Millar CI, Stephenson NL, Stephens SL. 2007. Climate change and forests of the future: managing in the face of uncertainty. Ecological applications 17:2145-2151.

Milner-Gulland EJ, Shea K. 2017. Embracing uncertainty in applied ecology. Journal of Applied Ecology 54:2063-2068.

National Oceanic and Atmospheric Administration. 2016. NOAA ecological forecasting: Protecting human health and coastal economies with early warning signs. Available at: [https://oceanservice.noaa.gov/ecoforecasting/noaa.html#:~:text=An%20ecological%20f orecast%20predicts%20changes,and%20communities%20may%20be%20affected]. Last accessed 02 June 2020.

Neilson RP, Pitelka LF, Solomon AM, Nathan R, Midgley GF, Fragoso JMV, Lischke H, Thompson K. 2005. Forecasting Regional to Global Plant Migration in Response to Climate Change. BioScience 55:749-759.

Nilsson C, Lepori F, Malmqvist B, Törnlund E, Hjerdt N, Helfield JM, Palm D, Östergren J, Jansson R, Brännäs E, Lundqvist H. 2005. Forecasting Environmental Responses to Restoration of Rivers Used as Log Floatways: An Interdisciplinary Challenge. Ecosystems 8:779–800.

Nogués-Bravo D, Simberloff D, Rahbek C, Sanders NJ. 2016. Rewilding is the new Pandora's Box in conservation. Current Biology 26:R87-R91.

North MP, Stephens SL, Collins BM, Agee JK, Aplet G, Franklin JF, Fule PZ. 2015. Reform forest fire management. Science 349:1280-1281.

Obeysekera J, Rutchey K. 1997. Selection of scale for Everglades landscape models. Landscape Ecology 12:7-18.

Perring MP, Standish RJ, Price JN, Craig MD, Erickson TE, Ruthrof KX, Whiteley AS, Valentine LE, Hobbs RJ. 2015. Advances in restoration ecology: rising to the challenges of the coming decades. Ecosphere 6:1-25.

Peterson GD, Cumming GS, Carpenter SR. 2003. Scenario Planning: A tool for conservation in an uncertain world. Conservation Biology 17:358-366.

Polasky S, Carpenter SR, Folke C, Keeler B. 2011. Decision-making under great uncertainty: environmental management in an era of global change. Trends in Ecology and Evolution 26: 398-404.

Poff NL, Brown CM, Grantham TE, Matthews JH, Palmer MA, Spence CM, Wilby RL, Haasnoot M, Mendoza GF, Dominique KC, Baeza A. 2016. Sustainable water management under future uncertainty with eco-engineering decision scaling. Nature Climate Change 6:25-34.

Ravenscroft C, Scheller RM, Mladenoff DJ, White MA. 2010. Forest restoration in a mixed- ownership landscape under climate change. Ecological Applications 20:327-346.

Rice KJ, Emery NC. 2003. Managing microevolution: restoration in the face of global change. Frontiers in Ecology and the Environment 1:469-478.

Robinson RM. 2012. Transportation. Business Process Modeling. In: (Handbook of Real-World Applications in Modeling and Simulation), {[eds.] [Sokolowski JA, Banks CM.]}. John Wiley & Sons, Hoboken, 93-128.

Rubio-Tamayo J, Barrio MG, García FG. 2017. Immersive Environments and Virtual Reality: Systematic Review and Advances in Communication, Interaction and Simulation. Multimodal Technologies and Interaction 1:21.

Salafsky N, Margoluis R. 2001. Breaking the cycle: developing general and yet non‐trivial guiding principles for using protected area conservation strategies. In: (Making parks work). {[eds.] [Van Schaik CP, Terborgh J, Davenport L, Rao M.]}. Island Press, Washington, D.C., 420– 434.

Samuelson W, Zeckhauser R. 1988. Status quo bias in decision making. Journal of Risk Uncertainty 1:7-59.

Scheller RM. 2018. The challenges of forest modeling given climate change. Landscape Ecology 33:1481-1488.

Scheller RM, Gustafson EJ, Sturtevant BR, Ward BC, Mladenoff DJ. 2010. Increasing the research and management value of ecological models using modern software engineering techniques. Frontiers in Ecology and the Environment 8:253-260.

Scheller RM, Spencer WD, Rustigian-Romsos H, Syphard AD, Ward BC, Strittholt JR. 2011. Using stochastic simulation to evaluate competing risks of wildfires and fuels management on an isolated forest carnivore. Landscape Ecology 26:1491-1504.

Schlaepfer MA, Sax DF, Olden JD. 2011. The potential conservation value of non-native species. Conservation Biology 25:428-437.

Schlueter M, Mcallister RR, Arlinghaus R, Bunnefeld N, Eisenack K, Hoelker F, Milner-Gulland EJ, Müller B, Nicholson E, Quaas M, Stöven M. 2012. New horizons for managing the environment: A review of coupled social‐ecological systems modeling. Natural Resource Modeling 25:219-272.

Seddon PJ, Moehrenschlager A, Ewen J. 2014. Reintroducing resurrected species: selecting DeExtinction candidates. Trends in Ecology & Evolution 29:140-147.

Seidl R, Thom D, Kautz M, Martin-Benito D, Peltoniemi M, Vacchiano G, Wild J, Ascoli D, Petr M, Honkaniemi J, Lexer MJ. 2017. Forest disturbances under climate change. Nature climate change 7:395-402.

Shackelford N, Hobbs RJ, Burgar JM, Erickson TE, Fontaine JB, Laliberte ́ E, Ramalho CE, Perring MP, Standish RJ. 2013. Primed for change: developing ecological restoration for the 21st Century. Restoration Ecology 21:297-304.

Shapiro B. 2017. Pathways to de-extinction: how close can we get to resurrection of an extinct species? Functional Ecology 31:996-1002.

Silver WL, Vergara SE, Mayer A. 2018. Carbon sequestration and greenhouse gas mitigation potential of composting and soil amendments on California’s rangelands. California Natural Resources Agency.

Soares-Filho BS, Nepstad DC, Curran LM, Cerqueira GC, Garcia RA, Ramos CA, Voll E, McDonald A, Lefebvre P, Schlesinger P. 2006. Modelling conservation in the Amazon basin. Nature 440:520-523.

Sotnik G. 2018. The SOSIEL platform: Knowledge-based, cognitive, and multi-agent. Biologically Inspired Cognitive Architectures 26:103-117.

Spencer W, Rustigian-Romsos H, Strittholt J, Scheller RM, Zielinski W, Truex R. 2011. Using occupancy and population models to assess habitat conservation opportunities for an isolated carnivore population. Biological conservation 144:788-803.

Spies TA, White E, Ager A, Kline JD, Bolte JP, Platt EK, Olsen KA, Pabst RJ, Barros AM, Bailey JD, Charnley S. 2017. Using an agent-based model to examine forest management outcomes in a fire-prone landscape in Oregon, USA. Ecology and Society 22:1-25.

Stanturf JA, Palik BJ, Dumroese RK. 2014. Contemporary forest restoration: A review emphasizing function. Forest Ecology and Management 331:292-323.

Suding KN, Gross KL, Houseman G. 2004. Alternative states and positive feedbacks in restoration ecology. Trends in Ecology and Evolution 193:46-53.

Svenning JC, Pedersen PBM, Donlan CJ, Ejrnæsd R, Faurby S, Galetti M, Hansen DM, Sandel B, Sandom CJ, Terborgh JW, Verai FWM. 2016. Science for a wilder Anthropocene: Synthesis and future directions for trophic rewilding research. PNAS 113:898-906.

Syphard AD, Scheller RM, Ward BC, Spencer WD, Strittholt JR. 2011. Simulating landscape- scale effects of fuels treatments in the Sierra Nevada, California, USA. International Journal of Wildland Fire 20:364-383.

Tabrizian P, Petrasova A, Harmon B, Vaclav P, Mitasova H, Meentemeyer RK. 2016. Immersive tangible geospatial modeling. Paper in proceedings of 24th ACM SIGSPATIAL international conference on advances in geographic information systems, Burlingame, CA.

Taleb NN. 2007. The black swan: The impact of the highly improbable. 2nd Edition. Random house, New York, 1-400.

Thompson JR, Foster DR, Scheller R, Kittredge D. 2011. The influence of land use and climate change on forest biomass and composition in Massachusetts, USA. Ecological Applications 21:2425-2444.

Thompson JR, Wiek A, Swanson FJ, Carpenter SR, Fresco N, Hollingsworth T, Spies TA, Foster DR. 2012. Scenario studies as a synthetic and integrative research activity for long-term ecological research. BioScience 62:367-376.

Torrubia S, McRae BH, Lawler JJ, Hall SA, Halabisky M, Langdon J, Case M. 2014. Getting the most connectivity per conservation dollar. Frontiers in Ecology and the Environment 12:491-497.

Tulloch AIT, Hagger V, Greenville AC. 2020. Ecological forecasts to inform near-term management of threats to biodiversity. Global Change Biology 26:5816-5828.

Twilley RR, Rivera-Monroy VH, Chen R, Botero L. 1999. Adapting an ecological mangrove model to simulate trajectories in restoration ecology. Marine Pollution Bulletin 37:404- 419.

Vose JM, Peterson DL, Domke GM, Fettig CJ, Joyce LA, Keane RE, Luce CH, Prestemon JP, Band LE, Clark JS, Cooley NE, D’Amato A, Halofsky JE. 2018. Forests. In: (Impacts, Risks, and Adaptation in the United States: Fourth National Climate Assessment, Volume II), {[eds.] [Reidmiller DR, Avery CW, Easterling DR, Kunkel KE, Lewis KLM, Maycock TK, Stewart BC.]}. U.S. Global Change Research Program, Washington, DC, 232-267. doi: 10.7930/NCA4.2018.CH6

Vukomanovic J, Skrip MM, Meentemeyer RK. 2019. Making it spatial makes it personal: engaging stakeholders with geospatial participatory modeling. Land 8 doi:10.3390/land8020038.

Wesbrook JW, Holliday JA, Newhouse AE, Powell WA. 2020. A plan to diversify a transgenic blight tolerant American chestnut population using citizen science. Plants People Planet 2:84-95.

Westbrook JW, Zhang Q, Mandal MK, Jenkins EV, Barth LE, Jenkins JW, Grimwood J, Schmutz J, Holliday JA. 2020. Optimizing genomic selection for blight resistance in American chestnut backcross populations: A trade‐off with American chestnut ancestry implies resistance is polygenic. Evolutionary applications 13:31-47.

Westgate MJ, Likens GE, Lindenmayer DB. 2013. Adaptive management of biological systems: a review. Biological Conservation 158:128-139

Wright AS. 2020. Preventing extinctions: Architecting an accretocene. Canada's National Observer. Available at: [https://www.nationalobserver.com/2020/04/30/preventing- extinctions]. Last accessed 30 April 2020.

Zavaleta ES, Hobbs RJ, Mooney HA. 2001. Viewing invasive species removal in a whole- ecosystem context. Trends in Ecology & Evolution 16:454-459.

Zielinski WJ, Kucera TE, Barrett RH. 1995. Current distribution of the fisher, Martes pennanti, in California. California Fish and Game 8:1104-112.

Chapter 2: Conservation Velocity

2.1 Abstract

Global conservation efforts are facing mounting challenges to halt biodiversity loss under accelerating anthropogenic change. Conservation practitioners must reconcile limited resources with escalating demands, and do so under deep uncertainty as landscapes are transformed and existing management paradigms are challenged. Efficient conservation action that supports improved biodiversity outcomes must identify effective strategies. Here we systematically forecasted and evaluated conservation strategy performance at the landscape level prior to implementation. Using a landscape change model parameterized for a landscape of conservation interest in North Carolina, we simulated the acquisition and restoration of land based on four different conservation strategies (acquiring the lowest cost land, acquiring land clustered around

established conservation cores, acquiring land with high geodiversity characteristics, and

acquiring land opportunistically) from 2020 to 2100. We developed a novel metric to evaluate

how quickly a conservation action moves the landscape towards the intended target, the

‘conservation velocity’. Quantifying habitat quality pre- and post-restoration, we assessed conservation velocity for each of the four conservation strategies and three restoration targets

(longleaf pine restoration, pine mix restoration, and hardwood restoration). Using this approach, we found that targeting land clustered around established conservation cores resulted in the highest conservation velocity across restoration targets regardless of initial habitat quality. Our results demonstrate that conservation strategy substantially influences conservation velocity, even given identical on-the-ground actions. This approach for the proactive assessment of conservation strategy performance under uncertainty represents a promising new advance in

conservation planning, especially as managers face unforeseen and growing challenges

associated with global change.

2.2 Introduction

There are more demands being placed on conservation science, and on conservation

practitioners, than ever before. Global climate change is inducing changes in species phenology

and decoupling species interactions, triggering geographical displacement, and disaggregating

ecological communities (Walther et al. 2002; Parmesan & Yohe 2003; Parmesan 2006). Land use and land cover changes are causing habitat loss, conversion to agriculture and infrastructure, and fragmentation (Fahrig 2003; Fischer & Lindenmayer 2007; de Chazal & Rounsevell 2009)

and producing widespread changes to ecosystem service provision (Metzger et al. 2006). Human

population growth is increasing demands on natural resources around the globe (Tilman et al.

2017). Collectively, these anthropogenic forcings are culminating in widespread extinctions and

biodiversity loss, posing unprecedented challenges to conservation (Heller & Zavaleta 2009;

Pimm et al. 2014). To meet these challenges, proactive, innovative, and future-focused conservation planning is critical (Prober et al. 2019).

Increasing conservation demands coupled with limited resources require conservationists and natural resource managers to make difficult decisions about what outcomes to prioritize and the most effective and efficient conservation actions to achieve them (Bower et al. 2018). Even under the umbrella of a specific conservation intervention, practitioners must decide where and when to allocate resources, and how much to allocate. These decisions are made more difficult as anthropogenic change is challenging established approaches to conservation and adding uncertainty in landscape responses to management actions (Lindenmayer et al. 2008; Kujala et

al. 2013). This places a premium on the capacity to select effective strategies as landscapes change (Myers et al. 2000; Williams & Jackson 2007).

In this regard, simulation models are powerful tools to forecast and evaluate the potential outcomes of conservation decisions (Larson et al. 2004; Ravenscroft et al. 2010; Brum et al.

2019; Tulloch et al. 2020). They can help bound the conservation decision-making process, provide critical information about costs and benefits, and illuminate trends in performance that are essential to make robust conservation decisions under uncertainty (Mozelewski & Scheller

2021). Conservation forecasting projects ecosystem changes in response to environmental drivers like climate change and disturbance and in response to land use change and management, and can be used to evaluate how conservation strategies perform over a range of landscape futures and severities of landscape change. It is especially useful for comparing conservation strategy outcomes across scenarios of climate change, encapsulating conservation strategy performance across a range of conditions, and at the landscape scale where full cost and benefit tradeoffs are revealed (Torrubia et al. 2014). Yet such forecasting has typically tested the static implementation of one-time conservation actions or lacked consideration of landscape dynamics

(Carlson et al. 2019). Conservation science needs forecasts that include the systematic, continuous, and dynamic implementation multiple of strategies to better support conservation planning under uncertainty.

Here we introduce an innovative approach for evaluating conservation strategies on a dynamic landscape undergoing anthropogenic change and provide an example of this application. We define ‘conservation velocity’, a metric that estimates the rate at which a conservation strategy moves the landscape towards its intended conservation target, and we demonstrate its utility for informing conservation planning under global change. Velocity has

been used to describe the amount and direction of change in climate (Loarie et al. 2009) and land use change projections (Ordonez et al. 2014), and to describe how quickly species must move to track their climatic niche, or biotic velocity (Carroll et al. 2015). These concepts have been used to evaluate species vulnerability to climate change (Ackerly et al. 2010) and to inform conservation (Garcia et al. 2014; Carroll et al. 2017) and connectivity planning (Costanza &

Terando 2019). However, velocity has not been used to assess conservation actions to evaluate how quickly biodiversity targets are achieved. We used conservation velocity to estimate the rate at which on-the-ground management approaches achieve conservation gains.

Using the landscape change model LANDIS-II (Scheller et al. 2007), we simulated the gradual acquisition and restoration of protected areas under multiple scenarios of climate change in central North Carolina, USA, a biodiverse region of significant conservation interest (Noss et al. 2015). We simulated land acquisition based on four different conservation strategies, derived from conservation and protected area development theory, and three different restoration targets

(e.g. restoration to promote longleaf pine, Pinus palustris Mill.). We then calculated the conservation velocity for each restoration target and conservation strategy. This spatially and temporally explicit framework enables the evaluation of complex, landscape-level responses to specific conservation strategies, allowing conservation planners and natural resource managers to optimize in situ conservation planning, via in silico forecasting and assessment (Cantarello et al.

2011; Scheller et al. 2011; Mina et al. 2021).

2.3 Methods

2.3.1 Study area

We focused on a 1.5 million hectare landscape in central North Carolina, USA (Figure

2.1). The climate of this region is humid subtropical with mild winters and hot, humid summers.

Mean monthly temperatures are 3.3°C and 25°C in January and July, respectively (NC State

Climate Office). Mean annual precipitation averages between 1016-1398 mm with no distinct wet and dry seasons (NC State Climate Office). Forests in the study area consist of longleaf pine

(Pinus palustris) forest, loblolly pine (Pinus taeda L.) plantation, and hardwood, hardwood-pine mixed, and pine mix forest types (Dewitz 2019). Urban, exurban, and agricultural land uses are present throughout. Frequent, low intensity fires were once common on this landscape but were excluded for much of the 20th century (Frost 1993). After years of fire suppression, an effort is

being made to return fire to the landscape using prescribed burning as a management tool

(Mitchell et al. 2006). Hurricanes are also a significant source of tree mortality and forest

community change (Duryea et al. 2007). Collectively, this area is of substantial conservation

interest because of the rare and endemic plant communities and high degree of species richness

found there (Van Berkel et al. 2019). Significant loss of longleaf pine forests from forest

clearing, conversion, and fire suppression; growing urbanization pressures; and the endangered

status of several plant and animal species have resulted in multi-agency partnerships dedicated to large-scale conservation (Wear & Greis 2002; Terando et al. 2014; Rua Mordecai pers. comm).

Figure 2.1. Study area in central North Carolina for assessing conservation strategies using a spatially explicit modeling framework, land types derived from 2016 NLCD data.

2.3.2 Model description and parameterization

We modeled landscape-scale forest dynamics and conservation action implementation using LANDIS-II (Scheller et al. 2007), a spatially explicit landscape simulation model parameterized for the study area in central North Carolina from 2020 to 2100 (Figure 2.2).

LANDIS-II incorporates ecological processes including vegetation dynamics and disturbance regimes to simulate the dynamics of forested landscapes over time (Nitschke et al. 2020). In

LANDIS-II, vegetation dynamics are simulated as the regeneration, growth, mortality, and

dispersal of tree and shrub species and age-specific cohorts that compete for resources (soil

moisture, nitrogen, light) and undergo disturbances and management actions within a landscape

of gridded cells (Scheller et al. 2007). Species are defined by a combination of traits (e.g.

probability of establishment, longevity, age of sexual maturity, shade tolerance) that determine

landscape and site-level demographics (e.g. Krofcheck et al. 2018). Each cell on the landscape, representing 1-ha, was assigned soil characteristics (SSURGO), climate conditions, and disturbance and management regimes.

We initialized LANDIS-II with current forest conditions including tree species-age cohorts assigned to every forested cell using US Forest Service Forest Inventory and Analysis

(FIA) data (USDA Forest Service 2015). We included tree species that represented greater than two percent of forest biomass based on the FIA plots in North Carolina, South Carolina, and

Virginia. We also chose to include two species that fell below the two percent cutoff but are considered species of great conservation interest and important community type indicators: longleaf pine (Pinus palustris) and turkey oak (Quercus laevis). In total, we simulated 11 species.

We simulated three harvest prescriptions and three restoration prescriptions to capture forest management and conservation practices in central North Carolina (Table 2.1). Harvest prescriptions consisted of loblolly pine clear cutting and replanting, longleaf pine thinning and burning to simulate the prescribed burning frequently done on Ft. Bragg and throughout the

restoration was enacted. Restoration prescriptions were only applied to stands identified for

restoration. In addition to harvest and management, we simulated tropical storm disturbance

using data from Schrum et al. (2020).

Table 2.1. Harvest and restoration prescriptions simulated on the study landscape in central North Carolina.

Targets per 3yr Prescription Actions taken harvest rotation Complete clearcut of loblolly 4% of private family Loblolly clearcut pine plantation stands followed land, 10% of private by replanting of loblolly pine. corporate land Removal of >90% of young, fire 18% of state land, intolerant hardwood trees with 2% of private family Longleaf thin/burn some mortality of oak species land, 20% of federal and non-longleaf pines to land, and 12% of simulate prescribed burning. conservation lands Remove 50% of all cohorts > 15 9% of private family Mixed forest thinning years old in these stands. land 100% removal of all major hardwood spp. competitors, 50% of protected Longleaf restoration moderate removal of non- area hectares longleaf pine species, replant with longleaf pine. 80% of more removal of hardwood species, 50% removal 25% of protected Pine mix restoration of loblolly pine, replant with area hectares Virginia, shortleaf, and longleaf pine.

We used global climate model (GCM) projections derived from the Coupled Model

Intercomparison Project Phase 5 downscaled using the Multivariate Adaptive Constructed

Analogs (MACA) method (Abatzoglou 2013). From this ensemble, we chose outputs from five

GCMs for the Representative Concentration Pathway (RCP) 4.5 that bracketed the temperature and precipitation projections for the study area and acted as our model replicates to capture a range of possible future conditions in the study area: bcc-csm1, CNRM-CM5, Had-GEM2,

IPSL-CM5A-LR, and Nor-ESM1-M (Grey 2018). For detailed information about model

extensions, parameterization, and calibration, see the Appendix A.

Figure 2.2. Workflow of conservation actions integrated into a spatially explicit landscape change model parameterized for the study area

2.3.3 Conservation scenarios

We integrated continuous, dynamic conservation action into a spatially explicit vegetation change model that simulated ecological processes affecting forest habitat composition

and vegetation structure. This approach allowed us to test conservation strategy performance and

evaluate landscape level responses to differing management objectives over time using a

scenario framework designed to capture uncertainty from both climate change and system

stochasticity. Using this framework, we simulated four protected area network development

strategies derived from conservation theories and enacted by conservation and land management

agencies globally to proactively evaluate conservation strategy outcomes across scenarios of global change (Burkey 1989; Murdoch et al. 2007; Hjort et al. 2015).

2.3.3.1 Economic

Our ‘economic strategy’ maximized conservation return on investment by weighing

economic costs with intended biodiversity benefits when evaluating what land to conserve

(Polasky 2008; Robillard & Kerr 2017). For this strategy, we focused on the acquisition of lower

cost parcels of land that allowed us to conserve 25% more land at each time step. We simplified

the conservation return on investment framework, which usually incorporates metrics of

biodiversity and/or habitat quality in addition to economic data when assessing where to invest

for the greatest return. Rather, we considered the ability to acquire additional land for

conservation to represent increased conservation return on investment. For the economic

strategy, we calculated cost per acre across the study extent using statewide parcel valuation data

(NC OneMap 2016) and assigned an average cost per acre to individual forest stands (Figure 2.3)

(ESRI 2018). We excluded parcels that lacked parcel value data assigned, and any stands with a

zero total dollar value were removed from consideration for acquisition. We randomly selected

stands via a weighted random sort (Appendix C) with stands of the lowest cost per acre having

the highest weights (average cost per hectare of economic stands= $9,293.26; average cost of cluster stands= $60,557.44; average cost of geodiversity stands= $11,358.60; average cost of random stands= $22,306.49).

Figure 2.3. Cost per hectare of land across the study area, used in the selection of forest stands for conservation under the economic strategy.

2.3.3.2 Cluster

Our ‘cluster strategy’ clustered protected area development around already established conservation cores, creating larger core areas of protected habitat. This strategy has been suggested as a more effective way to combat anthropogenic change than small, scattered protected areas (Maiorano et al. 2008). We created buffers at 100, 500, 1000, 2500 meters around already protected areas on our landscape identified in the US Geological Survey’s

Protected Areas database (USGS 2018; Figure 2.4). Forest stands whose centers fell within the

100m buffer had the highest probability of being selected for conservation under this strategy,

followed by stands that had centers within the 500m buffer, stands with centers in the 1000m

buffer, and finally, stands with centers in the 2500m buffer via a weighted random sort

(Appendix C). Stands that fell outside of these four buffers had the lowest probability of selection.

Figure 2.4. Forest stands that fell within 100m, 500m, 1000m, and 2500m buffers of an already established protected area, used in the selection of forest stands for conservation under the cluster strategy.

2.3.3.3 Geodiversity

Our ‘geodiversity strategy’ preferentially selected stands with the highest geodiversity characteristics, a proxy for the ‘conserving nature’s stage’ theory (Lawler et al. 2015). The term geodiversity encompasses the natural range of variation in geological, geomorphological, and 45

soil characteristics on a landscape (Gray 2008). High geodiversity in a landscape often means more environmental niches, resulting in higher levels of overall biodiversity (Comer et al. 2015).

Prioritizing geodiversity in conservation may prove more resilient to climate change, as geodiversity indicates a greater capacity to maintain species diversity and ecological function as the climate changes (Anderson et al. 2015). For the geodiversity strategy, we used geodiversity data from the Nature Conservancy’s Resilient and Connected Landscapes map assign an average geodiversity score to each stand (Figure 2.5) (Anderson et al. 2016). We then performed a weighted random sort with forest stands scoring the highest in geodiversity having the highest probability of being selected (Appendix C).

Figure 2.5. Geodiversity scores across the study area as described in TNC’s Resilient and Connected Landscapes map, used in the selection of forest stands for conservation under the geodiversity strategy.

2.3.3.4 Opportunistic

Last, for our ‘opportunistic strategy’ we randomly selected forest stands, representing the ad hoc nature of land acquisition that can sometimes occur when a parcel becomes available for conservation acquisition (e.g. when land is donated or when a conservation group or agency is proactively approached by a parcel owner who is willing to sell). This strategy also served as a null model for comparison to the other conservation strategies.

We chose to conserve roughly one percent of the total landscape at each 5-year time step,

resulting in approximately 17 percent of total land conserved over 80 years in accordance with

terrestrial conservation targets established by the Convention on Biological Diversity’s Strategic

Plan for Biodiversity 2011-2020 (SCBD 2010). The economic strategy supported 25 percent

additional land acquisition and restoration. The stands that were designated for protection then

underwent simulated restoration actions. Already conserved lands, identified using the Protected

Areas database from the US Geological Survey, were not considered for land acquisition. Areas

identified as open water or as high, medium, or low density residential in the 2016 NLCD

(Dewitz 2019) were also removed from consideration. Habitat was allowed to grow and mature

over time reflecting the change in habitat quality as trees developed and matured.

2.3.4 Habitat targets

We simulated the restoration of three key habitat types in central North Carolina. The

first habitat type that we experimentally restored was longleaf pine (Pinus palustris) forest.

Longleaf pine forests were once prolific throughout the southeastern United States but are now relegated to less than 5% of their former range (Frost 2007, Holland et al. 2019). Longleaf pine

forests have been the focus of restoration efforts throughout the study area, in part due to the

listing of the red cockaded woodpecker and because of the rare and endemic understory plant

species associated with this habitat type. The second habitat type we restored was mixed pine

forest. Containing a mix of longleaf pine, short leaf pine (Pinus echinata Mill.), and Virginia

pine (Pinus virginiana Mill.), this habitat type has been suggested as a suitable alternative to pure longleaf pine stands, and declining shortleaf pine has been the focus of regional conservation efforts (Stanturf et al. 2004, Guldin 2019). The third habitat type we restored was mixed hardwood forest, emphasizing white oak (Quercus alba L.), a crucial food and habitat

source for commercial wildlife species (Hicks et al. 2004).Oaks have experienced widespread

regeneration failure (Dey 2014) warranting conservation concern in the study area.

2.3.5 Conservation velocity

Here we present a new metric, ‘conservation velocity’, which estimates the rate at which a conservation strategy moves the landscape towards (or away from) its intended conservation target. Conservation velocity is represented as the change in some standardized metric of conservation response (e.g. habitat quality, habitat amount, species conserved, etc.) over the difference between the time when a conservation action is initiated (ti) and some user-defined

time step or end time (tn).

To assess the conservation velocity (Vcons) of each conservation strategy and restoration

target simulated, we compared the habitat quality of forest stands immediately prior to

restoration to their habitat quality at the end of the model run for each restoration target

(longleaf, pine mix, and hardwood mix).

Vcons = [ ] ∆𝐻𝐻𝐻𝐻

∆𝑡𝑡 𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦 We quantified conservation velocity by calculating the rate of change in habitat quality

for each conservation strategy to assess how quickly stands moved towards their respective

restoration targets. Conservation velocity for each conservation strategy and restoration target

was averaged across model replicates. Habitat quality (Hq) was defined as the sum of the

standardized ratio of target species biomass to total site biomass for each stand, the standardized

mean target species age of each stand, and the standardized number of habitat cells (habitat size

in hectares) in the stand. Habitat cells were defined as the number of hectares in the stand that

contained greater than 50% of stand biomass from the restoration target species.

Hq = Hqfinal - Hqinitial

∆ where

Hq= ( . [ ] ÷ [ ]) + ( [ −2 ]) + z( [ −2]) 𝑧𝑧 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑠𝑠𝑠𝑠𝑠𝑠 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑔𝑔 𝑚𝑚 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏𝑏 𝑔𝑔 𝑚𝑚 𝑧𝑧 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 𝑎𝑎𝑎𝑎𝑎𝑎 𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦𝑦 ℎ𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎𝑎 𝑠𝑠𝑠𝑠𝑠𝑠𝑠𝑠 ℎ𝑎𝑎 We assessed how the vegetation characteristics of each conservation strategy and restoration approach changed over time without biasing any particular strategy. For example, geodiversity may be inversely related to land use intensity (Tukiainen et al. 2017), so incorporating proximity to development as a detractor of habitat quality could bias the geodiversity strategy’s habitat quality score.

We individually standardized each habitat quality variable (z-score) across conservation strategy and time step for each restoration target and gave equal weight to each variable’s contribution to habitat quality. Future applications of conservation velocity could apply varying weights to each variable to reflect relative contributions to the conservation response metric.

Standardization of the conservation response variables allows conservation velocity to be calculated for any suite of conservation indicators, enabling broader application of the metric, and can highlight how each variable influences overall conservation velocity.

2.4 Results

2.4.1 Habitat quality scores

Understanding pre- and post-restoration habitat quality is necessary context for understanding conservation velocity. Longleaf pine restoration sites had the most variable habitat quality at the beginning and end of our simulations (22-59% initial; 13-79% final). Initial habitat quality scores varied between 3% and 56% for pine mix restoration sites and post- restoration habitat quality scores varied between <1% and 47%. Mixed hardwood restoration

sites showed the smallest variance in pre-and post-restoration scores, with initial and post- restoration habitat quality scores varying between 2% and 19% scores. Mean habitat quality improvement for the longleaf restoration target varied by 3% to 22% between conservation strategies, between 2% and 6% for mixed pine restoration, and between <1% and 14% for mixed hardwood restoration. The mixed hardwood restoration strategy showed the greatest improvement in habitat quality on average while the longleaf pine strategy showed the smallest improvement in habitat quality after restoration (Figure 2.6).

Figure 2.6. Comparison of changes in average pre- and post-restoration habitat quality scores across conservation strategies and restoration targets.

2.4.2 Conservation velocity

Conservation velocity varied both between conservation strategies and among restoration

targets (Figure 2.7). The clustering strategy achieved the highest conservation velocity for all

restoration targets, especially for longleaf restoration targets. Conservation velocities for the

other three strategies were variable in comparison, though the economic strategy typically had

the lowest Vcons. Longleaf restoration had the greatest variability in conservation velocity across conservation strategies, while the pine mix restoration target had comparable conservation velocities across strategies (Table 2.2). The hardwood restoration target had the highest conservation velocity on average, followed closely by pine mix restoration. Our results suggest that it would take longleaf pine restoration implemented under the geodiversity strategy more than twice as long as longleaf restoration implemented under the clustering strategy to achieve

the same habitat quality improvements (Figure 2.8). Though the amount of habitat quality

improvement was similar across conservation strategies for each restoration target, conservation

velocity showed disproportionately greater variation. The conservation strategy that achieved the

highest resulting habitat quality for each restoration target did not necessarily have the highest

conservation velocity, and that the highest initial habitat quality score did not mandate the

highest final habitat quality score.

Figure 2.7. The conservation velocity of each conservation strategy and restoration target, averaged across model replicates.

Table 2.2. Relative percent of each conservation velocity to the fastest conservation velocity (% max Vcons) for each restoration target. %max Vcons Cluster Economic Geodiversity Opportunistic Longleaf restoration 100.0 74.4 47.7 71.4 Pine mix restoration 100.0 91.0 93.9 97.8 Hwood mix restoration 100.0 81.2 97.1 87.3

a) b) c)

Figure 2.8. Comparison of the conservation velocities of each conservation strategy for all three restoration targets, plotted with a fixed intercept. Targets were: a) longleaf restoration b) pine mix restoration c) hardwood mix restoration shown with standard error shading.

2.5 Discussion

Conservation decision-makers are facing an urgent problem: If landscapes are changing,

conservation strategy outcomes are shifting, and system-wide uncertainty is mounting, how can one possibly decide when, where, and how to optimally invest their limited resources for targeted biodiversity outcomes (Meir et al. 2004; Hobbs et al. 2009; Polasky et al. 2011)? By enabling the assessment of dynamic conservation strategies across scenarios of global change before they are implemented, our framework elucidates landscape-level trends in strategy performance and cost/benefit tradeoffs and facilitates robust conservation decision-making. This approach represents a novel evolution in conservation planning and risk analysis as suites of possible

conservation actions can be evaluated in silico (Reside et al. 2018), before any substantial investment of capital (Mozelewski and Scheller 2021). We see this approach as a parsimonious tool for weighing the risks and benefits of potential conservation actions and assessing conservation actions as part of a larger systematic conservation planning (Margules & Pressey

2000) or modern portfolio theory framework (Ando & Mallory 2012).

Applying our simulation framework to North Carolina, we found that the conservation strategies tested arrived at similar endpoints although over much different time horizons. In comparing the conservation velocity of land acquisition and restoration for four different conservation strategies, we found that clustering protected area development around already established conservation cores achieved improvements in habitat quality the fastest across all restoration targets, though it did not necessarily result in a greater final habitat condition. Forest composition over time showed higher concentrations of longleaf pine and white oak biomass in already protected areas such as the Ft. Bragg Army base and Uwharrie National Forest, respectively, than elsewhere on the landscape, supporting the idea that established protected areas provided propagules that sped up the conservation velocity of land acquired and restored nearby. Additionally, the geodiversity strategy had a far lower rate of conservation velocity for longleaf pine restoration than the other three strategies, compared to how it performed for the other two restoration targets. This may be because geodiverse portions of the study area were often correlated with thinner soil substrates and more clay dominant soils, while longleaf pine forests in much of the region are characterized by deep, sandy soils (Gilliam et al. 1993; Peet

2007). As such, longleaf restoration efforts may need to be guided by soil suitability rather than the aim of enhancing geodiversity in the study area. Additionally, this could indicate that more management is required for longleaf restoration beyond planting alone. Ultimately, even small

differences in conservation velocity could have disproportionately large impacts on how quickly

a conservation action approaches its intended target, which could offer new insight for

conservation decision-making.

The introduction of conservation velocity could be a key consideration for long-term conservation planning, allowing managers to evaluate not just the forecasted outcomes of conservation intervention but how quickly those outcomes are received, enabling a more tailored and time-sensitive conservation portfolio. In some instances, a greater conservation velocity might be preferable or even essential when immediacy of conservation results is critical (e.g. protecting and restoring habitat for species on the brink of extinction (Martin et al. 2012) or when the cost of an action may drastically increase in the future (Balmford et al. 2002)).

Choosing a strategy with the greatest velocity could also allow for earlier biodiversity gains, offering some insurance against resources lost as management objectives change and giving managers the greatest number of options if/when it becomes necessary to change course. In other instances, a more gradual conservation velocity could be useful (e.g. when social resistance to change or conservation regulations (Holmes 2007) necessitates gradual acclimation to novel landscape configurations and conservation intervention, or when planning conservation to facilitate large spatial and temporal scale climate migration that will be carried out over years

(Meir et al. 2004; Hannah et al. 2014)).

Identifying the conservation velocity of management actions could also improve climate adaptation planning. For instance, managers focusing on enhancing landscape connectivity to facilitate climate migration could implement high velocity restoration or conservation actions immediately, both in space and time, while staggering lower velocity actions out at higher latitudes and elevations to create climate stepping stones (Nuñez et al. 2013; Saura et al. 2014).

Identifying high velocity conservation actions could also be critical information for protecting and restoring climate refugia, which will be essential for the persistence and climate range shifts of some species as the planet warms (Keppel et al. 2012). Conservation velocity could be coupled with estimates of climate change (Burrows et al. 2014) and land use change velocities

(Ordonez et al. 2014) to assess whether a conservation action can achieve its target in time for climate migrants to arrive (Lawler et al. 2006) or whether a conservation approach can withstand climate and land use changes. This will be essential as anthropogenic change accelerates and conservation actions struggle to keep pace with the magnitude and direction of change.

Collectively, the framework we have introduced could facilitate more effective and efficient conservation planning for biodiversity and climate adaptation.

2.5.1 Limitations and future research directions

While the utility simulation models in conservation decision-making is clear (i.e. Bini et al. 2006; Briot et al. 2007; Carlson et al. 2019; Mina et al. 2020), they cannot accurately predict the future given any conservation action. Rather, they can help to bound the decision-making process and provide useful information as managers and decision-makers navigate novel ecosystems and an uncertain future (Hobbs et al. 2014). Additionally, though LANDIS-II has been rigorously tested and widely used throughout North America and globally (www.landis- ii.org), like with any model, uncertainty in parameter estimates and model representation of ecological processes exists. We focused on a limited set of landscape-scale dynamics. For example, we did not model insect disturbance (Sturtevant et al. 2015), deer browsing (Stromayer and Warren 1997), or species expanding their range into the study area as the climate warms

(Van Houtven et al. 2019), which have been shown to shift community composition in the eastern United States. Though we did not include estimates of land use change, a major agent of

anthropogenic change in the southeastern United States (Terando et al. 2014), research currently underway using this same framework in central North Carolina will incorporate land use change projections as a next modeling step (Mozelewski et al., in prep). Finally, our estimates of habitat quality as a metric for our conservation targets represents a simplified understanding of habitat quality that is not species specific. Even with these limitations, the framework we have developed can help to improve the likelihood of favorable and long-lasting conservation outcomes that can withstand global change.

2.6 References

Abatzoglou J. T. 2013. Development of gridded surface meteorological data for ecological applications and modelling. International Journal of Climatology 33:121-131.

Ackerly DD, Loarie SR, Cornwell WK, Weiss SB, Hamilton H, Branciforte R, Kraft NJB. 2010. The geography of climate change: Implications for conservation biogeography. Diversity & Distributions 16:476-487.

Anderson MG, Barnett A, Clark M., Prince J, Olivero Sheldon A, Vickery B. 2016. Resilient and Connected Landscapes for Terrestrial Conservation. The Nature Conservancy, Eastern Conservation Science, Eastern Regional Office. Boston, MA.

Anderson MG, Comer PJ, Beier P, Lawler JJ, Schloss CA, Buttrick S, Albano CM, Faith DP. 2015. Case studies of conservation plans that incorporate geodiversity. Conservation Biology 29:680–691.

Ando AW, Mallory ML. 2012. Optimal portfolio design to reduce climate-related conservation uncertainty in the Prairie Pothole Region. Proceedings of the National Academy of Sciences of the United States of America 109:6484–6489.

Balmford A et al. 2002. Economic reasons for conserving wild nature. Science 297:950–3.

Bini LM, Diniz-Filho JAF, Rangel TFLVB, Bastos RP, Pinto MP. 2006. Challenging Wallacean and Linnean shortfalls: knowledge gradients and conservation planning in a biodiversity hotspot. Diversity & Distributions 12:475–482.

Bower SD, Brownscombe JW, Birnie-Gauvin K, Ford MI, Moraga AD, Pusiak RJP, Turenne ED, Zolderdo AJ, Cooke SJ, Bennett JR. 2018. Making Tough Choices: Picking the Appropriate Conservation Decision-Making Tool. Conservation Letters 11:e12418.

Briot J-P, Semiocast PG, Guyot P, Irving M. 2007. Participatory simulation for collective management of protected areas for biodiversity conservation and social inclusion. AIS- CMS 7:183-188

Brum FT, Pressey RL, Bini LM, Loyola R. 2019. Forecasting conservation impact to pinpoint spatial priorities in the Brazilian Cerrado. Biological Conservation 240:108283.

Burkey TV. 1989. Extinction in Nature Reserves: The Effect of Fragmentation and the Importance of Migration between Reserve Fragments. Oikos 55:75.

Burrows MT et al. 2014. Geographical limits to species-range shifts are suggested by climate velocity. Nature 507:492–495.

Cantarello E, Newton AC, Hill RA, Tejedor-Garavito N, Williams-Linera G, López-Barrera F, Manson RH, Golicher DJ. 2011. Simulating the potential for ecological restoration of

dryland forests in Mexico under different disturbance regimes. Ecological Modelling 222:1112–1128.

Carlson M, Browne D, Callaghan C. 2019. Application of land-use simulation to protected area selection for efficient avoidance of biodiversity loss in Canada’s western boreal region. Land Use Policy 82:821–831.

Carroll C, Lawler JJ, Roberts DR, Hamann A. 2015. Biotic and Climatic Velocity Identify Contrasting Areas of Vulnerability to Climate Change. PLOS ONE 10:e0140486.

Carroll C, Roberts DR, Michalak JL, Lawler JJ, Nielsen SE, Stralberg D, Hamann A, Mcrae BH, Wang T. 2017. Scale-dependent complementarity of climatic velocity and environmental diversity for identifying priority areas for conservation under climate change. Global Change Biology 23:4508–4520.

Comer PJ, Pressey RL, Hunter ML, Schloss CA, Buttrick SC, Heller NE, Tirpak JM, Faith DP, Cross MS, Shaffer ML. 2015. Incorporating geodiversity into conservation decisions. Conservation Biology 29:692–701.

Costanza JK, Terando AJ. 2019. Landscape Connectivity Planning for Adaptation to Future Climate and Land-Use Change. Current Landscape Ecology Reports 4:1–13.

de Chazal J, Rounsevell MDA. 2009. Land-use and climate change within assessments of biodiversity change: A review. Global Environmental Change 19:306–315.

Dewitz J. 2019. National Land Cover Database (NLCD) 2016 Products: U.S. Geological Survey data release, https://doi.org/10.5066/P96HHBIE.

Dey DC. 2014. Sustaining Oak Forests in Eastern North America: Regeneration and Recruitment, the Pillars of Sustainability. Forest Science 60:926–942.

Duryea ML, Kampf E, Littell RC. 2007. Hurricanes and the Urban Forest: I. Effects on Southeastern United States Coastal Plain Tree Species. Agriculture & Urban Forestry 33:83-97.

ESRI 2018. ArcGIS Desktop: Release 10.6. Redlands, CA: Environmental Systems Research Institute.

Fahrig L. 2003. Effects of Habitat Fragmentation on Biodiversity. Annual Review of Ecology, Evolution, and Systematics 34:487–515.

Fischer J, Lindenmayer DB. 2007. Landscape modification and habitat fragmentation: a synthesis. Global Ecology and Biogeography 16:265–280.

Frost C. 2007. History and Future of the Longleaf Pine Ecosystem. Pages 9–48 The Longleaf Pine Ecosystem. Springer New York.

Frost CC. 1993. Four centuries of changing landscape patterns in the longleaf pine ecosystem. Proceedings of the Tall Timbers Fire Ecology Conference 18:17–43.

Garcia RA, Cabeza M, Rahbek C, Araújo MB. 2014. Multiple Dimensions of Climate Change and Their Implications for Biodiversity. Science 344:6183.

Gilliam FS, Yurish BM, Goodwin LM. 1993. Community Composition of an Old Growth Longleaf Pine Forest: Relationship of Soil Texture. Bulletin of the Torrey Botanical Club 120:287.

Gray M. 2008. Geodiversity: developing the paradigm. Proceedings of the Geologists’ Association 119:287–298.

Grey GME. 2018. Ensemble creation of downscaled climate projections in the Southeast United States. North Carolina State University. [Unpublished master’s thesis]

Guldin JM. 2019. Restoration of Native Fire-Adapted Southern Pine-Dominated Forest Ecosystems: Diversifying the Tools in the Silvicultural Toolbox. Forest Science 65:508– 518.

Hannah L, Flint L, Syphard AD, Moritz MA, Buckley LB, Mccullough IM. 2014. Fine-grain modeling of species’ response to climate change: holdouts, stepping-stones, and microrefugia. Trends in Ecology & Evolution 29:390–397.

Heller NE, Zavaleta ES. 2009. Biodiversity management in the face of climate change: A review of 22 years of recommendations. Biological Conservation 142:14–32.

Hicks RR, Conner WH, Kellison RC, Lear D Van. 2004. Silviculture and Management Strategies Applicable to. Page In: Gen. Tech. Rep. SRS–75. Asheville, NC: U.S. Department of Agriculture, Forest Service, Southern Research Station. Chapter 7. p. 51-62.

Hijmans RJ, van Etten J. 2012. raster: Geographic analysis and modeling with raster data . R package version 3.4-5. http://CRAN.R-project.org/package=raster.

Hjort J, Gordon JE, Gray M, Hunter ML. 2015. Why geodiversity matters in valuing nature’s stage. Conservation Biology 29:630–639.

Hobbs RJ et al. 2014. Managing the whole landscape: historical, hybrid, and novel ecosystems. Frontiers in Ecology and the Environment 12:557–564.

Hobbs RJ, Higgs E, Harris JA. 2009. Novel ecosystems: implications for conservation and restoration. Trends in Ecology and Evolution 24:599–605.

Holland AM, Rutledge BT, Jack SB, Stober JM. 2019. The longleaf pine forest: Long-term monitoring and restoration of a management dependent ecosystem. Journal for Nature Conservation 47:38–50.

Holmes G. 2007. Protection, politics and protest: Understanding resistance to conservation. Conservation & Society 5:184-201.

Keppel G, Van Niel KP, Wardell-Johnson GW, Yates CJ, Byrne M, Mucina L, Schut AGT, Hopper SD, Franklin SE. 2012. Refugia: identifying and understanding safe havens for biodiversity under climate change. Global Ecology and Biogeography 21:393–404.

Krofcheck DJ, Hurteau MD, Scheller RM, Loudermilk EL. 2018. Prioritizing forest fuels treatments based on the probability of high-severity fire restores adaptive capacity in Sierran forests. Global Change Biology 24:729–737.

Kujala H, Moilanen A, Araújo MB, Cabeza M. 2013. Conservation Planning with Uncertain Climate Change Projections. PLoS ONE 8:e53315.

Larson MA, Thompson FR, Millspaugh JJ, Dijak WD, Shifley SR. 2004. Linking population viability, habitat suitability, and landscape simulation models for conservation planning. Ecological Modelling 180:103–118.

Lawler JJ, Ackerly DD, Albano CM, Anderson MG, Dobrowski SZ, Gill JL, Heller NE, Pressey RL, Sanderson EW, Weiss SB. 2015. The theory behind, and the challenges of, conserving nature’s stage in a time of rapid change. Conservation Biology 29:618–629.

Lawler JJ, White D, Neilson, RP, Blaustein, AR. 2006. Predicting climate-induced range shifts: model differences and model reliability. Global Change Biology 12:1568-1584.

Lindenmayer DB, Fischer J, Felton A, Crane M, Michael D, Macgregor C, Montague-Drake R, Manning A, Hobbs RJ. 2008. Novel ecosystems resulting from landscape transformation create dilemmas for modern conservation practice. Conservation Letters 1:129–135.

Loarie SR, Duffy PB, Hamilton H, Asner GP, Field CB, Ackerly DD. 2009. The velocity of climate change. Nature 462:1052–1055.

Maiorano L, Falcucci A, Boitani L. 2008. Size-dependent resistance of protected areas to land- use change. Proceedings of the Royal Society B 275:1297–1304.

Margules CR, Pressey RL. 2000. Systematic conservation planning. Nature 405:243–253.

Martin TG, Nally S, Burbidge AA, Arnall S, Garnett ST, Hayward MW, Lumsden LF, Menkhorst P, Mcdonald-Madden E, Possingham HP. 2012. Acting fast helps avoid extinction. Conservation Letters 5:274–280.

Meir E, Andelman S, Possingham HP. 2004. Does conservation planning matter in a dynamic and uncertain world? Ecology Letters 7:615–622.

Metzger MJ, Rounsevell MDA, Acosta-Michlik L, Leemans R, Schröter D. 2006. The vulnerability of ecosystem services to land use change. Agriculture, Ecosystems &

Environment 114:69-85.

Mina M, Messier C, Duveneck M, Fortin M, Aquilué N. 2021. Network analysis can guide resilience‐based management in forest landscapes under global change. Ecological Applications 31:e2221. Ecological Society of America.

Mitchell RJ, Hiers JK, O’brien JJ, Jack SB, Engstrom RT. 2006. Silviculture that sustains: the nexus between silviculture, frequent prescribed fire, and conservation of biodiversity in longleaf pine forests of the southeastern United States. Canadian Journal of Forest Research 36:2724-2736.

Mozelewski TG, Scheller RM. 2021. Forecasting for intended consequences. Conservation Science and Practice:e370.

Murdoch W, Polasky S, Wilson KA, Possingham HP, Kareiva P, Shaw R. 2007. Maximizing return on investment in conservation. Biological Conservation 139:375-388.

Myers N, Mittermeierf RA, Mittermeierf CG, Da Fonsecat GAB, Kent J. 2000. Biodiversity hotspots for conservation priorities. Nature 403:853-858.

NC OneMap Geospatial Portal. 2021. North Carolina Department of Information Technology, Government Data Analytics Center, Center for Geographic Information and Analysis. Available at https://www.nconemap.gov.

Nitschke CR, Trouvé R, Lumsden LF, Bennett LT, Fedrigo M, Robinson AP, Baker PJ. 2020. Spatial and temporal dynamics of habitat availability and stability for a critically endangered arboreal marsupial: implications for conservation planning in a fire‐prone landscape. Landscape Ecology 35:1553-1570.

North Carolina State Climate Office. NC Climate Synopsis. https://climate.ncsu.edu/climate/synopsis

Noss RF, Platt WJ, Sorrie BA, Weakley AS, Means DB, Costanza J, Peet RK. 2015. How global biodiversity hotspots may go unrecognized: Lessons from the North American Coastal Plain. Diversity and Distributions.

Nuñez TA, Lawler JJ, McRae BH, Pierce DJ, Krosby MB, Kavanagh DM, Singleton PH, Tewksbury JJ. 2013. Connectivity Planning to Address Climate Change. Conservation Biology 27:407–416.

Ordonez A, Martinuzzi S, Radeloff VC, Williams JW. 2014. Combined speeds of climate and land-use change of the conterminous US until 2050. Nature Climate Change 4:811–816.

Parmesan C. 2006. Ecological and Evolutionary Responses to Recent Climate Change. Annual Review of Ecology, Evolution, and Systematics 37:637–669.

Parmesan C & Yohe G. 2003. A globally coherent fingerprint of climate change impacts across natural systems. Nature 421:37-42.

Pimm SL, Jenkins CN, Abell R, Brooks TM, Gittleman JL, Joppa LN, Raven PH, Roberts CM, Sexton JO. 2014. The biodiversity of species and their rates of extinction, distribution, and protection. Science 344:1246752–1246752.

Polasky S. 2008, May 6. Why conservation planning needs socioeconomic data. PNAS 105:6505-6506.

Polasky S, Carpenter SR, Folke C, Keeler B. 2011. Decision-making under great uncertainty: Environmental management in an era of global change. Trends Ecology & Evolution 26:398-404.

Prober SM, Doerr VAJ, Broadhurst LM, Williams KJ, Dickson F. 2019. Shifting the conservation paradigm: a synthesis of options for renovating nature under climate change. Ecological Monographs 89:e01333.

R Core Team. R: A Language and Environment for Statistical Computing. (R Foundation for Statistical Computing, 2017).

Ravenscroft C, Scheller RM, Mladenoff DJ, White MA. 2010. Forest restoration in a mixed- ownership landscape under climate change. Ecological Applications 20:327–346.

Reside AE, Butt N, Adams VM. 2018. Adapting systematic conservation planning for climate change. Biodiversity & Conservation 27:1-29.

Robillard CM, Kerr JT. 2017. Assessing the shelf life of cost-efficient conservation plans for species at risk across gradients of agricultural land use. Conservation Biology 31:837–847.

Saura S, Bodin Ö, Fortin M-J. 2014. Stepping stones are crucial for species’ long-distance dispersal and range expansion through habitat networks. Journal of Applied Ecology 51:171–182.

Scheller RM, Domingo JB, Sturtevant BR, Williams JS, Rudy A, Gustafson EJ, Mladenoff DJ. 2007. Design, development, and application of LANDIS-II, a spatial landscape simulation model with flexible temporal and spatial resolution. Ecological Modelling 201:409–419.

Schrum P, Scheller RM, Duveneck MJ, Lucash MS. 2020. Base-Hurricane: A new extension for the Landis-II forest landscape model. Environmental Modelling and Software 133:104833.

Soil Survey Staff, Natural Resources Conservation Service, United States Department of

Agriculture. Soil Survey Geographic (SSURGO) Database. Available online at https://sdmdataaccess.sc.egov.usda.gov.

Sturtevant BR, Cooke BJ, Kneeshaw DD, Mac Lean DA. 2015. Modeling insect disturbance across forested landscapes: Insights from the spruce budworm. Simulation Modeling of Forest Landscape Disturbances 93-134.

Terando AJ, Costanza J, Belyea C, Dunn RR, McKerrow A, Collazo JA. 2014. The Southern Megalopolis: Using the Past to Predict the Future of Urban Sprawl in the Southeast U.S. PLoS ONE 9:e102261.

Tilman D, Clark M, Williams DR, Kimmel K, Polasky S, Packer C. 2017. Future threats to biodiversity and pathways to their prevention. Nature 546:73-81.

Torrubia S, McRae BH, Lawler JJ, Hall SA, Halabisky M, Langdon J, Case M. 2014. Getting the most connectivity per conservation dollar. Frontiers in Ecology and the Environment 12:491–497.

Tukiainen H, Bailey JJ, Field R, Kangas K, Hjort J. 2017. Combining geodiversity with climate and topography to account for threatened species richness. Conservation Biology 31:364– 375.

Tulloch AIT, Hagger V, Greenville AC. 2020. Ecological forecasts to inform near-term management of threats to biodiversity. Global Change Biology 26:5816–5828.

USDA Forest Service. 2015. Forest inventory and analysis database: Database description and user guide for phase 2. U.S. Department of Agriculture, Forest Service, Northeastern Research Station, Forest Inventory and Analysis Unit, Newtown Square, PA. 758 p.

U.S. Geological Survey (USGS) Gap Analysis Project (GAP). 2018. Protected Areas Database of the United States (PAD-US): U.S. Geological Survey data release, https://doi.org/10.5066/P955KPLE.

Van Berkel D, Shashidharan A, Mordecai R, Vatsavai R, Petrasova A, Petras V, Mitasova H, Vogler J, Meentemeyer R. 2019. Projecting Urbanization and Landscape Change at Large Scale Using the FUTURES Model. Land 8:144.

Van Houtven G, Phelan J, Clark C, Sabo RD, Buckley J, Thomas RQ, Horn K, LeDuc SD. 2019. Nitrogen deposition and climate change effects on tree species composition and ecosystem services for a forest cohort. Ecological Monographs 89:e01345.

Walther G-R, Post E, Convey P, Menzel A, Parmesank C, Beebee TJC, Fromentin J-M, Hoegh- Guldberg O, Bairlein F. 2002. Ecological responses to recent climate change. Nature 416:389-395.

Wear DN, Greis JG. 2002. Southern Forest Resource Assessment: Summary of findings. Journal

of Forestry 100:6–15.

Williams JW, Jackson ST. 2007. Novel climates, no-analog communities, and ecological surprises. Ecological Society of America. Diversity & Distributions 16:476-487.

Chapter 3: Forecasting Conservation Strategy Selection Influences on Landscape

Connectivity

3.1 Abstract

Maintaining and enhancing landscape connectivity reduces the effects of habitat fragmentation on biodiversity decline due to rapid anthropogenic change. However, there is large uncertainty about how effective different conservation strategies will be at enhancing

connectivity for multiple species, especially over long temporal scales on dynamic landscapes.

Using a landscape change model we simulated dynamic landscape connectivity across a 1.5 million hectare landscape in central North Carolina over an 80-year time period (2020-2100). To assess differences in connectivity among varying conservation strategies, we simulated protected area acquisition and restoration at 5-year time steps using four distinct conservation strategies: acquiring the lowest cost land, acquiring land clustered around already established conservation areas, acquiring land with high geodiversity characteristics, and acquiring land opportunistically.

We used graph theoretic metrics to quantify landscape-level connectivity across these strategies, evaluating connectivity for high and low dispersing species as well as for habitat specialists and generalists. Our study showed that any conservation strategy used to guide protected area network development significantly improved landscape connectivity, but that these improvements were highly variable depending on the conservation strategy selected. Clustering the development of new protected areas around land already designated for conservation yielded the strongest improvements in connectivity for all species guilds, increases several orders of magnitude beyond baseline connectivity. Meanwhile, conserving the lowest cost land showed the smallest contributions to connectivity. By providing an approach to evaluate the influence of conservation strategy selection on landscape-level connectivity prior to deployment, this study

can help managers to strategically incorporate connectivity enhancement in conservation decision-making and protected area network development to maximize conservation benefit.

3.2 Introduction

Habitat loss and fragmentation represent substantial threats to global biodiversity and are widely regarded as proximal drivers of extinction (Kuussaari et al. 2009; Crooks et al. 2017).

Maintaining and enhancing landscape connectivity, the degree to which the landscape facilitates biotic movement (Taylor et al. 1993), can reduce the impacts of habitat fragmentation and decrease vulnerability to extinction risk (Keeley et al. 2019). Connectivity is integral to migration, dispersal, and gene flow, which are required for wildlife population persistence and essential for adaptation to environmental change (Saura et al. 2011). Because of this, landscape- scale connectivity has become a major emphasis for conservation organizations and land managers aiming to conserve biodiversity, reduce the adverse effects of fragmentation, and facilitate species ranges shifts in response to climate change (Keeley et al. 2018).

The creation of new protected areas can significantly contribute to improving landscape connectivity, but identifying what conservation targets to prioritize in protected area development is a key challenge for future-focused conservation, and can have significant effects on landscape permeability (Bergsten et al. 2013; Reside et al. 2018). Decisions about which conservation strategies to employ to meet these targets will dictate where protected areas are established and result in protected area networks with variable landscape connectivity, an important consideration for long-term conservation planning. Such decisions may be aided by technological advances in simulation modeling, which can now forecast ecosystem and landscape change in response to both anthropogenic and environmental drivers allowing for the

evaluation of conservation actions prior to implementation (Perring et al. 2015; Mozelewski &

Scheller 2021).

To date, many studies of landscape connectivity assume either a static landscape matrix

or take place over short temporal scales (Martensen et al. 2017). The inclusion of future

conservation strategies in connectivity forecasting has been similarly narrow in scope, focusing

on a single subset of candidate actions or evaluating conservation at coarse scales (Carlson et al.

2019). Ideally, conservation planning would anticipate landscape change and would evaluate the

outcomes of a suite of feasible conservation alternatives in order to provide the best defense

against biodiversity loss (Pressey et al. 2007). Connectivity assessments should integrate

dynamic landscapes and longer time horizons and should assess the relative contributions of

different conservation strategies to optimize conservation decision-making (Bishop-Taylor et al.

2018). As such, the evaluation of landscape connectivity under multiple conservation strategies

on a spatiotemporally dynamic landscape over long temporal scales represents an important but

understudied research need in terrestrial systems (Zeller et al. 2020). Our study aims to address

this gap, leveraging a well-established simulation model to integrate conservation actions into a

dynamic landscape over time, enabling the testing of conservation strategies for long-term

conservation and connectivity planning.

Our goal was to assess the implications of protected area development conducted under

the umbrella of specific conservation strategies for landscape-level connectivity. Specifically, we assessed the connectivity generated by four different strategies of protected area development:

1) acquiring land as cheaply as possible (economic strategy), 2) acquiring land near already

established conservation cores (cluster strategy), 3) maximizing geodiverse lands (geodiversity

strategy), and 4) acquiring land opportunistically (opportunistic strategy). Using a landscape

change model we forecasted the trajectory of forests in central North Carolina from 2020 to

2100, an area with significant conservation interest in enhancing connectivity for threatened and endangered species. We assessed connectivity for four trait-based species guilds: habitat specialist and generalist species, each with high and low dispersal abilities. Combining landscape change modeling and network analysis, we assessed connectivity associated with each conservation strategy to track changes over time as additional land was acquired and to compare the variance in connectivity among strategies, all on a spatiotemporally dynamic landscape.

3.3 Methods

3.3.1 Study area

Our analysis focused on a 1.5 million hectare landscape located in central North Carolina,

USA (Figure 3.1). The climate is humid subtropical with mild winters (January mean

temperature: 3.3°C), hot, humid summers (July mean temperature: 25°C), and mean annual

rainfall between 1016-1398 mm with no distinct wet and dry seasons (NC State Climate Office).

The vegetation is typical of the three ecoregions found in the study area, the Atlantic Coastal

Plain, Sandhills, and Piedmont, with a combination of longleaf pine (Pinus palustris Mill.)

forest, loblolly pine (Pinus taeda L.) plantation, and pine and hardwood mixed forest types.

Approximately 58% of the study area is forested with urban, suburban, and agricultural land uses

also present. Past and current forest management has strongly shaped forest structure and

composition; the frequent, low intensity fires historically common on this landscape were

suppressed for much of the 20th century shifting pyrophytic pine forests to more shade tolerant

hardwoods (Frost 1993). Returning fire to this landscape through prescribed burning has become

a major focus for natural resource agencies, though population growth in the adjacent urban

cores make prescribed burning more contentious (Costanza et al. 2013). The study area has been

the focus of multi-agency partnerships dedicated to enhancing connectivity for the benefit of threatened and endangered species, such as the red cockaded woodpecker (Leuconotopicus

borealis), particularly between established natural resource areas (Wear & Greis 2002; Terando

et al. 2014).

Figure 3.1. Study area in central North Carolina for assessing the influence of conservation strategies on landscape connectivity, land cover types derived from 2016 NLCD data.

3.3.2 Conservation strategies

Here we define a conservation strategy as the portfolio of land acquisition and restoration actions intended to increase habitat area and enhance connectivity. We chose to conserve roughly one percent of the total landscape area at each 5-year time step, resulting in approximately 17 percent of total land conserved over 80 years in line with global terrestrial conservation targets established by the Convention on Biological Diversity’s Strategic Plan for Biodiversity 2011-

2020 (SCBD 2010). Already conserved lands, identified using the Protected Areas database from the US Geological Survey (USGS 2018), were not considered for land acquisition. Areas identified as open water or as high, medium, or low density residential in the 2016 NLCD

(Dewitz 2019) were also removed from consideration.

We evaluated the connectivity facilitated by four conservation strategies guiding protected area development (Mozelewski and Scheller in review). First, we assessed a conservation strategy that maximizes conservation return on investment (Polasky 2008;

Robillard & Kerr 2017) by focusing on the lowest cost parcels of land, enabling more land to be acquired and restored. Using assessor’s parcel valuation data for the state of North Carolina, we assigned an average cost per acre to individual forest stands and did a weighted random sort with stands of the lowest cost per acre having the highest probability of being selected (NC OneMap

2016, ESRI 2018). Approximately two percent of stands were removed from consideration for acquisition because they either did not have a parcel value or had a parcel value of 0. Second, we evaluated the clustering of protected area development around already established conservation cores by creating buffers at 100, 500, 1000, 2500 meters around already protected areas on our landscape identified in the US Geological Survey’s Protected Areas database (USGS 2018).

Forest stands whose centers fell within the 100m buffer had the highest probability of being

selected for conservation under this strategy, followed by stands that had centers within the

500m buffer, and so on, with stands that fell outside of these three buffers having the lowest probability of selection through a weighted random sort. Third, we evaluated the conservation of land with the highest geodiversity characteristics, a proxy for the ‘conserving nature’s stage’ theory (Lawler et al. 2015) by using geodiversity data from the Nature Conservancy’s resilient and protected landscapes map and zonal statistics to assign an average geodiversity score to each stand (Anderson et al. 2016). We then performed a weighted random sort with forest stands scoring the highest in geodiversity having the highest probability of being selected. Last, we assessed an opportunistic acquisition and restoration of land strategy. To do this, forest stands were randomly selected, representing the ad hoc nature of land acquisition that can sometimes occur when a parcel becomes available for conservation acquisition (e.g. when land is donated or voluntarily protected by easement). For additional details on the probabilities assigned to each conservation strategy for weighted random sorts, please see Appendix C.

At each time step, stands were selected for acquisition and restoration until approximately 1% of the total landscape amount was reached at each time step. The economic strategy, emphasizing the acquisition of lower cost land that enabled greater rates of acquisition for conservation, acquired approximately 1.25% of the total landscape for conservation at each time step. Following land acquisition, stands underwent simulated restoration and/or habitat maintenance actions to promote longleaf pine and mixed pine forest habitat types and improve habitat quality. These actions included prescribed burns in longleaf pine and longleaf mixed stands, hardwood removal, and planting of longleaf, shortleaf (Pinus echinata Mill.), and

Virginia pine (Pinus virginiana Mill.). For additional information on conservation strategy simulation, please see Appendix A.

3.3.3 Species and replicates

We evaluated connectivity over time on a dynamic landscape for four theoretical, trait- based species guilds, species that exploit the same resources, which allowed us to represent landscape permeability at multiple dispersal scales for broader conservation insight and greater generalization of findings (Blaum et al. 2011; Lechner et al. 2017). In this way, we focused on facilitating connectedness and managing ecosystems for functional capacity rather than focusing on particular species as recommended by Barnosky et al. (2017). Our species guilds were based on two traits known to influence the accessibility of landscape elements: degree of habitat specialization and dispersal ability. These two traits were combined to produce four simulated species: habitat specialist/high dispersal ability, habitat generalist/high dispersal ability, habitat specialist/low dispersal ability, and habitat generalist/low dispersal ability. We designated habitat specialists as those that preferred mature (> 21 year median stand age) longleaf pine forests while habitat generalists were designated as having an affinity for any mature non-plantation pine forest habitat type. High and low dispersal distances of 5,000m and

1,000m were selected to represent the maximum dispersal abilities of a range of common reptiles and small mammals from the southeastern US (Sutherland et al. 2000; Smith & Green 2005) similar to methods described in Bishop-Taylor et al. (2017) and Saura et al. (2011).

3.3.4 Landscape modeling

We simulated the gradual implementation of land acquisition and restoration for protected area network development on the study area from 2020-2100 using a spatially explicit landscape simulation model, LANDIS-II (Scheller et al. 2007) (Figure 3.2). LANDIS-II models landscape-scale forest succession, disturbance, and management, allowing us to evaluate landscape-level responses to conservation and track changes in habitat networks over time as

restored habitat matured for a dynamic assessment of potential connectivity. In LANDIS-II, vegetation dynamics are simulated via competitive processes whereby tree and shrub species- age-specific cohorts compete for resources (soil moisture, nitrogen, light) and experience disturbances and management actions on a landscape of gridded cells (Scheller et al. 2007).

LANDIS-II simulates vegetation regeneration, growth, mortality, and dispersal based on species life history traits (e.g. longevity, age of sexual maturity, shade tolerance) that determine landscape and site-level demographics (Scheller et al. 2011). Each 1-hectare cell on the landscape was assigned climate conditions (Livneh et al. 2015), soil characteristics (SSURGO

2017), and disturbance (tropical storm events, Shrum et al. 2020) and management regimes

(forest stand harvest and thinning) obtained from local managers and forestry extension specialists. For a detailed description of LANDIS-II parameterization for the study area, see

Appendix A.

We simulated three harvest prescriptions and three restoration prescriptions to capture forest management and conservation practices in central North Carolina (Table 3.1). Harvest prescriptions consisted of loblolly pine clear cutting and replanting, longleaf pine thinning and burning to simulate the prescribed burning frequently done on Ft. Bragg and throughout the

Sandhills ecoregion, and mixed forest thinning. Restoration prescriptions, each of which were simulated for every conservation strategy, included longleaf pine restoration, mixed pine restoration, and mixed hardwood restoration. Harvest prescriptions were applied across the entire landscape, including stands identified for restoration only leading up to the time step in which restoration was enacted. Restoration prescriptions were only applied to stands identified for restoration. In addition to harvest and management, we simulated tropical storm disturbance using data from Schrum et al. (2020).

Table 3.1. Harvest and restoration prescriptions simulated on the study landscape in central North Carolina.

We simulated four conservation strategies and four species guilds assuming static

climate, downloaded from the USGS Geodata portal daily Observed Gridded Meteorological

Data from 1949 to 2010 at 1/8th degree spatial resolution (Livneh et al. 2015) as daily

precipitation, maximum and minimum temperatures, and wind speed; and static land use, downloaded from the 2016 National Land Cover Database (Dewitz 2019), for a total of 16 scenarios. We simulated five replicates for each conservation scenario for a total of 80 simulations (4 conservation scenarios × 4 species guilds × 5 replicates). Future research will focus on climate change and land use change impacts to connectivity and the ability of conservation to keep pace with anthropogenic change (Mozelewski et al. in prep).

3.3.5 Connectivity

To process and visualize LANDIS-II outputs we used the raster library in the statistical software package R (v3.4.3) (Hijmans and van Etten 2012, R Development Core Team 2017). To analyze connectivity, we used the gDistance (Van Etten 2017), spatstat (Baddeley and Turner

2005), igraph (Csardi and Nepusz 2006), and Conefor command line (v2.6- Saura and Torne

2009) R packages. We conducted our connectivity analysis on North Carolina State University’s high performance computing cluster.

We used structural and functional graph theoretic metrics to quantify potential connectivity across the landscape over time. Graph theory network analysis efficiently assesses landscape connectivity across large study areas, incorporating both landscape structure and organism movement information to provide indices of connectivity well-suited for assessing landscape-level trends over time (Minor & Urban 2008; Galpern et al. 2011; Tulbure et al. 2014).

In this approach, the landscape is represented as a set of discrete habitat patches, or nodes, connected by links that represent the ability of an organism to disperse between nodes (Calabrese

& Fagan 2004). We defined links using least-cost paths as a measure of effective distance between nodes (Bishop-Taylor et al. 2015). Least-cost paths use cost-distance surfaces to account for landscape resistance to movement, representing the influence of a heterogeneous landscape matrix on species dispersal ability (Bunn et al. 2000; Dilts et al. 2016). We defined resistance values based on the assumption that forest species, especially specialist species, will face greater dispersal difficulty as they move through land cover types with characteristics increasingly disparate from those of the forested areas where they reside, similar to Saura et al.

(2011). Our definition accounted for potential barriers to dispersal including major roads, urban infrastructure, and water bodies. A rating scale of 1-100 was used for resistance values

(Appendix B), consistent with the magnitude of possible resistances found in previous studies

(e.g. Greenwald et al. 2009; Stevenson-Holt et al. 2014; Shirk et al. 2015; Blazquez-Cabrera et al. 2016) with 1 representing the easiest land cover type to move through (e.g mature preferred forest type) and 100 being the most difficult (e.g. an urban core). As gDistance calculates least- cost paths using conductance instead of resistance surfaces, every cell on the landscape was assigned a conductance value (1/resistance) based on whether the species was a habitat specialist or generalist.

Least-cost distances were calculated between the centroids of habitat patches similar to

Theobald et al. (2012) and reflect the minimum cost accumulated along the shortest path between two habitat nodes ( Bishop-Taylor et al. 2015; van Etten 2017). Using habitat patch centroids reduced computation time and allowed us to simulate species movement between and within habitat patches as a continuous process rather than assuming within-patch homogenization or an abrupt end to species movement at a patch edge (Dickson et al. 2017). For habitat specialists, a cell on the landscape was considered habitat if at least 25% of total biomass was longleaf pine and the median stand age was at least 21. Habitat cells for generalist species were defined as any cell with at least 25% of biomass from longleaf pine or at least 65% of biomass from any mix of pine species with the same median stand age requirements. Cells with

90% or more of their total biomass from loblolly pine were considered to be loblolly plantations and not considered habitat.

To identify habitat patches and capture habitat asynchrony, we separately grouped contiguous habitat pixels that co-occurred in both time steps t and t-1 and those that only occurred in time step t for each time step and replicate using an eight neighbor approach for each species guild to accommodate changes to connectivity across time and space (Appendix D).

Doing so allowed us to treat habitat added over time as discrete habitat patches, which more realistically reflects how an animal would perceive spatiotemporal habitat fluctuations and better represented intra- and inter-patch movement dynamics (Hanski 1999; Wimberly 2006). Nodes were weighted by area. Links were restricted to pairs of nodes with a pair-wise Euclidean distance that fell within each species group’s maximum dispersal distance. The probability of direct movement between patches was obtained using a negative exponential function of inter- patch least cost path value.

We used the Equivalent Connected Area (ECA) metric to quantify changes in functional connectivity facilitated by each conservation strategy over time (Saura et al. 2011) (Figure 3.2).

ECA measures the size of a single, maximally connected patch of habitat that if added would provide the same probability of connectivity as the observed habitat network (Saura et al. 2011).

In other words, ECA is the size of a forest habitat patch that would provide the same amount of landscape connectivity as all of the existing forest habitat patches and corridors (Rojas et al.

2020). ECA is a network-based index that measures connectivity by accounting for both the habitat area within each node and the habitat made available as organisms disperse, incorporating the contribution of patches and links as stepping stones that support connectivity between other habitat areas (Saura & Rubio 2010). The area units associated with ECA (e.g. hectares) enable a direct comparison between changes in total habitat area and network spatial configuration to assess how the addition or loss of habitat, whether from land acquisition or landscape dynamics, affects connectivity (Bishop-Taylor et al. 2018; McIntyre et al. 2018).

In addition to ECA, we quantified connectivity for three ECA subcomponents: (i) intra- patch connectivity, the connectivity contributed solely by the area within habitat nodes

(ECAintra), (ii) the connectivity contributed by direct connections between neighboring nodes

(ECAdirect), and (iii) connectivity contributions from intermediate stepping stones that enabled

longer distance movements (ECAstep) (Saura and Rubio 2010, Saura et al. 2011, Bishop-Taylor et

al. 2018). Evaluating the subcomponents of ECA (ECAintra, ECAdirect, and ECAstep) provided additional insights into the influence of different aspects of connectivity facilitated by each

conservation strategy and among species guilds. Each subcomponent was measured by the

percent it contributed to overall landscape connectivity. All indicators were calculated using

the probabilistic formulation of ECA at 10 year time steps.

Figure 3.2. Workflow of conservation actions integrated into a spatially explicit landscape change model parameterized for the study area, with model outputs of future forest habitat projections used to conduct a graph theory network analysis using Conefor.

3.4 Results

3.4.1 Nodes and links

Over our 80-year modeled timescale, the habitat networks facilitated by each

conservation strategy varied significantly. At time step zero the landscape contained 4,608

habitat nodes for specialist species and 11,228 nodes for generalist species (the total number of

nodes was not influenced by dispersal distance) (Figure 3.3). For specialist species, the economic

strategy resulted in the greatest number of habitat nodes on average after 80 years (n= 25,371,

Figure 3.3b) and for generalist species, the cluster strategy averaged the greatest number of nodes (n= 31,704, Figure 3.3a). Habitat nodes for generalist species were relatively similar across conservation strategies in both abundance and average size (avg. node size = ~6 ha), while habitat nodes for specialist species varied considerably by strategy. The cluster strategy had 19-

22% fewer nodes for specialist species than the other strategies (Figures 3.3a-d) and the highest average node size while average node size of the economic strategy was almost 50% less (~7.5 ha and ~4 ha, respectively). Generalist species had a greater number of habitat nodes than specialist species across all conservation strategies.

Interestingly, while link counts were closely correlated with nodes for each strategy

(Figures 3.3e-h), links did not considerably differ between specialists and generalists of the same dispersal distance for any strategy other than cluster (Figure 3.3e). The number of links for longer distance dispersers (5000m, n= 2,281,189) were two orders of magnitude higher than short distance dispersers (1000m, n= 120,057). Out of the four conservation strategies simulated, the economic strategy was the most variable in numbers of nodes and links for specialist species

(Figures 3.3b,f) and the cluster strategy was the most variable for generalist species between model replicates (Figures 3.3a,e).

Figure 3.3. Habitat networks showing the number of habitat nodes (panels a through d) and links (panels e through h) between habitats for four conservation strategies: cluster (a, d), economic (b, f), geodiversity (c, g), and opportunistic (d, h). Metrics are shown in blue for specialist species and orange for generalist species. Darker colors in the bottom row of graphs refer to short-distance (1000 m) dispersal organisms and lighter colors refer to long-distance (5000 m) dispersers.

3.4.2 Equivalent Connected Area

Every conservation strategy employed increased the Equivalent Connected Area (ECA)

indicator of connectivity for all species guilds, improving functional connectivity although by

different magnitudes (Table 3.2, Figure 3.4). For both short and long dispersal specialist species, the cluster strategy had the highest Equivalent Connected Area (Figure 3.4a) and the economic strategy had the lowest ECA (Figure 3.4e).

ECA patterns for both specialist species guilds were strongly correlated while ECA indices for generalist species varied by dispersal distance for each conservation strategy (Table

3.2). Longer dispersers had higher ECAs for all conservation strategies than their shorter dispersing counterparts (Figures 3.4a,e,i,m). For both specialist species, ECA increased rapidly over the first ~25 years. After that, the cluster and opportunistic strategies continued with a more gradual increase in connectivity over time (Figures 3.4a and 3.4m). The economic strategy experienced a slight decline in connectivity after year 25 before leveling out for 1000m specialist species and beginning a very gradual increase for 5000m specialist species (Figure 3.4e).

Connectivity for specialists under the geodiversity strategy peaked around year 40 after which connectivity experienced a very slight decline for short dispersal specialists and leveled off for long distance dispersal specialists (Figure 3.4i). Overall, connectivity for 1000m specialist species increased by 666% in the cluster strategy (increasing ECA by 29,535 hectares), 390% in the economic strategy (increasing ECA by 15,116 hectares), 524% in the geodiversity strategy

(increasing ECA by 22,140 hectares), and 618% in the opportunistic strategy (increasing ECA by

27,059 hectares). Connectivity for the 5000m specialist species increased by 308% in the cluster strategy (increasing ECA by 26,738 hectares), 180% in the economic strategy (increasing ECA by 10,262 hectares), 243% in the geodiversity strategy (increasing ECA by 18,377 hectares), and

284% in the opportunistic strategy (increasing ECA by 23,775 hectares). For short dispersal generalist species, ECA increased gradually (~133%) over the first 30 years before leveling off

(Figures 3.4a,i,m) or, in the case of the economic strategy, experiencing a slight decline (Figure

3.4e). ECA for high dispersal generalist species stayed constant over time for every conservation strategy tested.

Table 3.2. Changes to Equivalent Connected Area (ECA) over time for each conservation strategy (cluster, economic, geodiversity, and opportunistic) and species guild (habitat specialists and generalists with 1000m and 5000m dispersal abilities). Specialist 1000m Specialist 5000m % Difference % Difference Time ECA (ha) ECA (ha) Change (ha) Change (ha) Cluster 0 5219.5 100.0 0.0 12859.7 100.0 0.0 20 30266.6 579.9 25047.2 35370.8 275.1 22511.1 40 30869.5 591.4 25650.0 36027.4 280.2 23167.7 60 34807.5 666.9 29588.1 40508.9 315.0 27649.2 80 34754.4 665.9 29534.9 39597.8 307.9 26738.1 Economic 0 5219.5 100.0 0.0 12859.7 100.0 0.0 20 27946.5 535.4 22727.0 32222.1 250.6 19362.4 40 22449.1 430.1 17229.7 25649.4 199.5 12789.7 60 24163.5 462.9 18944.0 27060.0 210.4 14200.3 80 20335.6 389.6 15116.1 23121.6 179.8 10261.9 Geodiversity 0 5219.5 100.0 0.0 12859.7 100.0 0.0 20 27661.1 530.0 22441.7 31233.7 242.9 18374.0 40 31977.2 612.7 26757.8 35916.4 279.3 23056.7 60 27279.6 522.7 22060.2 31193.8 242.6 18334.1 80 27359.4 524.2 22140.0 31236.5 242.9 18376.8 Opportunistic 0 5219.5 100.0 0.0 12859.7 100.0 0.0 20 26560.4 508.9 21341.0 30164.2 234.6 17304.5 40 26008.4 498.3 20789.0 28985.4 225.4 16125.7 60 29386.8 563.0 24167.3 33181.5 258.0 20321.8 80 32278.8 618.4 27059.3 36634.6 284.9 23774.9

Generalist 1000m Generalist 5000m % Difference % Difference Time ECA (ha) ECA (ha) Change (ha) Change (ha) Cluster 0 29342.8 100.0 0.0 41440.2 100.0 0.0 20 37690.1 128.4 8347.3 44075.6 106.4 2635.4 40 39208.0 133.6 9865.2 41623.9 100.4 183.6 60 39590.5 134.9 10247.7 45765.2 110.4 4324.9 80 38879.5 132.5 9536.7 41221.4 99.5 -218.8

Table 3.2 (continued). Economic 0 29342.8 100.0 0.0 40012.3 100.0 0.0 20 35992.6 122.7 6649.8 40889.1 102.2 876.8 40 35577.9 121.2 6235.1 39842.5 99.6 -169.8 60 37417.8 127.5 8075.0 43920.4 109.8 3908.1 80 30686.2 104.6 1343.4 37586.2 93.9 -2426.1 Geodiversity 0 29342.8 100.0 0.0 41764.5 100.0 0.0 20 37707.7 128.5 8364.9 44506.8 106.6 2742.3 40 39130.7 133.4 9787.9 41852.5 100.2 88.0 60 38105.0 129.9 8762.2 45457.4 108.8 3692.9 80 37282.5 127.1 7939.7 40617.3 97.3 -1147.2 Opportunistic 0 29342.8 100.0 0.0 41973.3 100.0 0.0 20 36996.8 126.1 7654.0 42595.1 101.5 621.8 40 36992.7 126.1 7649.9 41251.7 98.3 -721.6 60 38395.0 130.8 9052.2 44312.8 105.6 2339.5 80 36593.3 124.7 7250.5 40630.9 96.8 -1342.4

3.4.3 ECAintra, ECA direct, and ECAstep

For specialist species, intra-patch connectivity (ECAintra, Figures 4b,f,j,n) was much higher for the cluster strategy than other strategies, and up to three times higher than the economic strategy. For short dispersal specialists, ECAintra connectivity declined by 50% or more

for all conservation strategies except the cluster strategy over time as additional land was

acquired and restored. For high dispersal specialists the contribution of intra-patch connectivity increased for 30 years (2050) after which its importance declined. ECAintra was more important

to connectivity for short dispersal species than long dispersal species.

The decline in ECAintra was particularly steep for the economic strategy and co-occurred with a steep increase in stepping stone (ECAstep, Figures 3.4c,g,k,o) importance in facilitating

connectivity. Stepping stone importance was consistently higher for the economic strategy than

the other strategies, particularly for short dispersal specialists, and up to 50% higher for the

economic strategy than the cluster strategy.

Connectivity between neighboring habitat nodes (ECA direct, Figures 3.4d,h,l,p) for

specialist species was consistently at least twice as important for long distance dispersers as for

short distance dispersers. For short distance dispersers, ECA direct only contributed about 1% to

overall connectivity.

For both generalist species, ECAintra, ECAdirect, and ECAstep were consistent across model

runs. For short dispersal generalists, intra-patch connectivity accounted for nearly 100% of overall landscape connectivity (Figures 3.4b,f,j,n). For high dispersal generalists, intra-patch connectivity accounted for about 70% of overall connectivity, direct connections accounted for about 8% (Figures 3.4c,g,k,o), and connectivity facilitated by stepping stones accounted for about 22% of ECA (Figures 3.4d,h,l,p) for all conservation strategies

Figure 3.4. Equivalent connected area metrics and ECA components for each species guild and conservation strategy from 2020- 2100. Metrics are shown in blue for specialist species and orange for generalist species. Darker colors refer to short-distance (1000 m) dispersal organisms and lighter colors refer to long-distance (5000 m) dispersers.

3.5 Discussion

The creation of protected area networks under any of the conservation strategies tested resulted in significant connectivity gains for both specialist species guilds (1000m and 5000m dispersal). On the other hand, land acquisition and restoration resulted in only small improvements in connectivity for low dispersal generalists and no improvement for high dispersal generalists, likely because habitat elements are already readily accessible. The cluster strategy, targeting the acquisition and restoration of land in close proximity to established conservation cores, resulted in the greatest connectivity improvements regardless of specialist species dispersal ability. This is largely because current longleaf pine habitat is primarily relegated to managed and protected areas in the southeast region of the study area. The cluster strategy, targeting land around conservation areas, created immediate connections to existing habitat and enabled a more feasible and continuous spatial progression for both short- and long- distance dispersers. In addition to reducing fragmentation, clustering new protected areas around areas already managed for conservation can expand habitat cores, reducing edge-to-area ratios and buffering species from edge effects like interspecific competition and predation (Herse et al.

2018). Larger habitat areas have also been suggested as an anthropogenic change-adaptive conservation strategy (Oliver & Morecroft 2014), so prioritizing protected area development around existing conservation cores would allow the integration of multiple conservation objectives, especially for specialist species, and may increase overall biodiversity outcomes

(Albert et al. 2017).

Our analysis also revealed that different conservation strategies facilitated connectivity in different ways for specialist species while being largely the same for generalist species (Figures

3.5 and 3.6). While the cluster strategy created larger core habitat patches and facilitated high

intra-patch connectivity, the habitat created by the economic strategy was more likely to serve as stepping stones resulting in longer distance dispersal opportunities (Figure 3.5) (Hannah et al.

2014). Though the economic strategy did not perform as well as the other strategies in overall connectivity because of its small average node size, the higher contribution of stepping stones to connectivity documented in this strategy could support long-term population persistence, facilitating metapopulation dynamics and enabling species range expansion in response to environmental change (Saura et al. 2014). Conservation practitioners seeking to prioritize stepping stones for movement between larger habitat patches or as stopovers for seasonal or climate change-induced species migration (Mawdsley et al. 2009) may be better served investing in low cost parcels of land, enabling a greater rate of land acquisition and restoration to create habitat islands throughout the landscape.

Figure 3.5. Differences in habitat networks between the cluster and economic strategies for long dispersal habitat specialists.

Figure 3.6. Differences in habitat networks between the cluster and economic strategies for long dispersal habitat generalists.

In general, the shift in connectivity importance from ECAintra to ECAstep for specialist species across the conservation strategies showed that dispersal ability across the landscape increased for specialist species as land was acquired and restored. By comparing the subcomponents of ECA for each conservation strategy, we were also able to evaluate how changes in habitat networks affected connectivity for species with different habitat associations and dispersal abilities. Intra-patch connectivity was most important for short distance dispersers while direct connections between habitat patches were more important for longer distance dispersers regardless of habitat preference, findings that echo those of Martensen et al. (2017) and Bishop-Taylor et al. (2018). The importance of stepping stones was relatively equal across dispersal distances for specialist species while being more important for high dispersal generalists than short dispersers. Collectively, these findings could help identify appropriate conservation responses for both single and multi-species concerns, as managers could prioritize the strategy that best met the connectivity needs of their focal species (Keeley et al. 2018).

By providing insight into the varied connectivity contributions of different conservation

strategies on a dynamic landscape, our approach can help to ensure that the allocation of limited

conservation resources support multi-criteria spatial prioritization (Moilanen et al. 2005). More

broadly, our approach to evaluating changes to complex habitat networks following land

acquisition and restoration represents an evolution in conservation planning as landscape-level

responses to dynamic and continuous conservation can be evaluated in silico prior to on the

ground deployment. This framework, enabling the evaluation of changes to complex habitat

networks on stochastic landscapes, can help planners and managers to design protected area

networks that account for the needs of multiple species and facilitate movement at multiple

spatial scales for targeted biodiversity outcomes.

3.5.1 Limitations and Future Work

The purpose of this study was to establish a baseline understanding of how conservation

strategy selection can influence connectivity on a dynamic landscape over time, even under a no-

change scenario. Yet we know that a business-as-usual future is highly unlikely. Incorporating projections of global change into conservation planning is critical to improve conservation

outcomes and to meet the challenge of increasing rates of biodiversity loss (Thuiller et al. 2008).

This is especially true in the southeastern United States, where land use change has been

accelerating for decades and climate change is shifting available management windows in

longleaf pine ecosystems (Costanza et al. 2015). Forthcoming work will assess the influence of

both climate and land use change on landscape-level connectivity, and variance in connectivity will be compared across conservation strategies and global change scenarios. The ultimate goal of this research is to assess whether aggressive protected area network development can counter

connectivity losses initiated by anthropogenic change.

While the use of simulation modeling is beneficial to conservation decision-making, this approach is not intended to predict the future of any given conservation action. Instead, landscape change models can be used to develop scenarios that examine possible intended and unintended consequences of conservation, helping to bound the decision-making process and providing support for managers seeking to proactively manage their landscapes for change

(Hobbs et al. 2014). While LANDIS is a widely-used and rigorously tested model for evaluating landscape-level responses to management, as with any model, uncertainty in parameter estimates and model representation of ecological processes exists. Additionally, we focused on a limited set of landscape-scale dynamics to include in the model. For example, insect disturbance

(Sturtevant et al. 2015) or species expanding their range into the study area as the climate warms

(Van Houtven et al. 2019), which have been shown to shift community composition in the eastern United States, were not included in our modeling efforts. It is also important to note that estimates of the cost of restoration were not included in the economic strategy valuation process, the inclusion of which could substantially alter the conservation costs of those parcels.

Our connectivity analysis relied on least cost path resistance distances based on the assumption that that forest species will face greater dispersal difficulty as they move through land cover types increasingly disparate from their preferred habitat as described in Saura et al.

(2011). Yet the use of least cost paths rely on two simplifying assumptions: that a single optimal path between habitats conveys the movement potential of that segment of the landscape, and that an organism has complete knowledge of the landscape and can select a movement path accordingly (Fahrig 2007; McRae et al. 2008; Pinto & Keitt 2009; Bishop-Taylor et al. 2015).

Additionally, the landscape matrix simulated through our modeling efforts is not a perfect representation of the land cover and vegetation types of study area. In our analysis, habitat nodes

were weighted by area but habitat quality metrics could be incorporated (e.g. weight nodes by

habitat size x some metric of habitat quality) to further account for variation in habitat suitability.

Finally, we use species functional types to convey differences in landscape connectivity that are

broadly applicable to conservation decision-makers, but these species guilds are not representative of any one species.

3.5.2 Conclusion

The use of simulation models to evaluate the landscape-level effects of conservation strategy implementation prior to on-the-ground deployment represents a promising advance in conservation decision-making (Soares-Filho et al. 2006; Cantarello et al. 2011; Rojas et al. 2020;

Mina et al. 2021; Mozelewski & Scheller 2021). In this study, we used a landscape change model to simulate the systematic acquisition and restoration of land under four different conservation strategies and evaluated landscape connectivity responses for four species guilds.

Through our combined use of landscape change modeling and graph theory network analysis we

forecasted dynamic functional connectivity under a business-as-usual scenario, quantifying changes to complex habitat networks differentially facilitated by conservation strategy selection.

Importantly, this approach allowed us to directly relate spatiotemporal changes in habitat

availability to the dynamic connectivity contributions of each conservation strategy. Our study

highlights significant conservation strategy-induced differences in connectivity both

spatiotemporally and by species guild type, demonstrating the importance of strategically

considering connectivity enhancement in conservation decision-making and protected area

network development to maximize conservation benefit. To our knowledge this study represents

one of the largest ecological network analyses to date, containing more than 15 million nodes,

nearly one billion links, and requiring more than 20,000 high performance computing hours to process.

3.6 References

Albert CH, Rayfield B, Dumitru M, Gonzalez A. 2017. Applying network theory to prioritize multispecies habitat networks that are robust to climate and land-use change. Conservation Biology 31:1383–1396.

Anderson MG, Barnett A, Clark M, Prince J, Olivero Sheldon A, Vickery B. 2016. Resilient and Connected Landscapes for Terrestrial Conservation. The Nature Conservancy, Eastern Conservation Science, Eastern Regional Office. Boston, MA.

Baddeley A, Turner R. 2005. spatstat: An R package for analyzing spation point patterns. Jornal of Statistical Software 12:1-42.

Barnosky AD et al. 2017. Merging paleobiology with conservation biology to guide the future of terrestrial ecosystems. Science 355:eaah4787.

Bergsten A, Bodin örjan, Ecke F. 2013. Protected areas in a landscape dominated by logging - A connectivity analysis that integrates varying protection levels with competition-colonization tradeoffs. Biological Conservation 160:279–288.

Bishop-Taylor R, Tulbure MG, Broich M. 2015. Surface water network structure, landscape resistance to movement and flooding vital for maintaining ecological connectivity across Australia’s largest river basin. Landscape Ecology 30:2045–2065.

Bishop-Taylor R, Tulbure MG, Broich M. 2017. Impact of hydroclimatic variability on regional- scale landscape connectivity across a dynamic dryland region. Ecological Applications 27:1124-1137.

Bishop-Taylor R, Tulbure MG, Broich M. 2018. Impact of hydroclimatic variability on regional- scale landscape connectivity across a dynamic dryland region. Ecological Indicators 94:142–150.

Blaum N, Mosner E, Schwager M, Jeltsch F. 2011. How functional is functional? Ecological groupings in terrestrial animal ecology: Towards an animal functional type approach. Biodiversity and Conservation 20:2333-2345.

Blazquez-Cabrera S, Gastón A, Beier P, Garrote G, Simón MÁ, Saura S. 2016. Influence of separating home range and dispersal movements on characterizing corridors and effective distances. Landscape Ecology 31:2355–2366.

Bunn AG, Urban DL, Keitt TH. 2000. Landscape connectivity: A conservation application of graph theory. Journal of Environmental Management 59:265–278.

Calabrese JM, Fagan WF. 2004. A comparison-shopper’s guide to connectivity metrics. Frontiers in Ecology and the Environment 2:529-536.

Cantarello E, Newton AC, Hill RA, Tejedor-Garavito N, Williams-Linera G, López-Barrera F, Manson RH, Golicher DJ. 2011. Simulating the potential for ecological restoration of dryland forests in Mexico under different disturbance regimes. Ecological Modelling 222:1112–1128.

Costanza JK, Terando AJ, McKerrow AJ, Collazo JA. 2015. Modeling climate change, urbanization, and fire effects on Pinus palustris ecosystems of the southeastern U.S. Journal of Environmental Management 151:186–199.

Costanza JK, Weiss J, Moody A. 2013. Examining the knowing-doing gap in the conservation of a fire-dependent ecosystem. Biological Conservation 158:107–115.

Crooks KR, Burdett CL, Theobald DM, King SRB, Marco M Di, Rondinini C, Boitani L. 2017. | Quantification of habitat fragmentation reveals extinction risk in terrestrial mammals. PNAS 114:7635-7640.

Csardi G, Nepusz T. 2006. The igraph software package for complex network research. InterJournal Complex Systems 1695.

Dewitz J. 2019. National Land Cover Database (NLCD) 2016 Products: U.S. Geological Survey data release, https://doi.org/10.5066/P96HHBIE.

Dickson BG, Albano CM, McRae BH, Anderson JJ, Theobald DM, Zachmann LJ, Sisk TD, Dombeck MP. 2017. Informing Strategic Efforts to Expand and Connect Protected Areas Using a Model of Ecological Flow, with Application to the Western United States. Conservation Letters 10:564–571.

Dilts TE, Weisberg PJ, Leitner P, Matocq MD, Inman RD, Nussear KE, Esque TC. 2016. Multiscale connectivity and graph theory highlight critical areas for conservation under climate change. Ecological Applications 26:1223–1237.

ESRI 2018. ArcGIS Desktop: Release 10.6. Redlands, CA: Environmental Systems Research Institute.

Fahrig L. 2007. Non-optimal animal movement in human-altered landscapes. Functional Ecology 21:1003-1015.

Frost C. 2007. History and Future of the Longleaf Pine Ecosystem. Pages 9–48 The Longleaf Pine Ecosystem. Springer New York. Available from https://link.springer.com/chapter/10.1007/978-0-387-30687-2_2 (accessed March 11, 2021).

Frost CC. 1993. Four centuries of changing landscape patterns in the longleaf pine ecosystem.

Proceedings of the Tall Timbers Fire Ecology Conference 18:17–43.

Galpern P, Manseau M, Fall A. 2011. Patch-based graphs of landscape connectivity: A guide to construction, analysis and application for conservation. Biological Conservation 144:44-55.

Greenwald KR, Purrenhage JL, Savage WK. 2009. Landcover predicts isolation in Ambystoma salamanders across region and species. Biological Conservation 142:2493–2500.

Hannah L, Flint L, Syphard AD, Moritz MA, Buckley LB, McCullough IM. 2014. Fine-grain modeling of species’ response to climate change: Holdouts, stepping-stones, and microrefugia. Trends in Ecology & Evolution 29:390-397.

Hanski I. 1999. Habitat Connectivity, Habitat Continuity, and Metapopulations in Dynamic Landscapes. Oikos 87:209.

Herse MR, With KA, Boyle WA. 2018. The importance of core habitat for a threatened species in changing landscapes. Journal of Applied Ecology 55:2241–2252.

Hijmans RJ, van Etten J. 2012. raster: Geographic analysis and modeling with raster data . R package version 3.4-5. http://CRAN.R-project.org/package=raster

Hobbs RJ et al. 2014. Managing the whole landscape: historical, hybrid, and novel ecosystems. Frontiers in Ecology and the Environment 12:557–564.

Keeley ATH, Basson G, Cameron DR, Heller NE, Huber PR, Schloss CA, Thorne JH, Merenlender AM. 2018. Making habitat connectivity a reality. Conservation Biology 32:1221–1232.

Keeley ATH, Beier P, Creech T, Jones K, Jongman RH, Stonecipher G, Tabor GM. 2019. Thirty years of connectivity conservation planning: An assessment of factors influencing plan implementation. Environmental Research Letters 14:103001.

Kuussaari M et al. 2009. Extinction debt: a challenge for biodiversity conservation. Trends in Ecology & Evolution 24:564-571.

Lechner AM, Sprod D, Carter O, Lefroy EC. 2017. Characterising landscape connectivity for conservation planning using a dispersal guild approach. Landscape Ecology 32:99–113.

Livneh B, Bohn TJ, Pierce DS, Munoz-Ariola F, Nijssen B, Vose R, Cayan D, Brekke LD. 2015. A spatially comprehensive, hydrometeorological data set for Mexico, the U.S., and southern Canada 1950-2013, Nature Scientific Data 5:150042, doi:10.1038/sdata.2015.42.

Martensen AC, Saura S, Fortin M-J. 2017. Spatio-temporal connectivity: assessing the amount of reachable habitat in dynamic landscapes. Methods in Ecology and Evolution 8:1253-1264.

Mawdsley JR, O’Malley R, Ojima DS. 2009. A review of climate-change adaptation strategies for wildlife management and biodiversity conservation. Conservation Biology 23:1080- 1089. McIntyre NE, Collins SD, Heintzman LJ, Starr SM, van Gestel N. 2018. The challenge of assaying landscape connectivity in a changing world: A 27-year case study in the southern Great Plains (USA) playa network. Ecological Indicators 91:607–616.

McRae BH, Dickson BG, Keitt TH, Shah VB. 2008. Using circuit theory to model connectivity in ecology, evolution, and conservation. Ecology 89:2712–2724.

Mina M, Messier C, Duveneck M, Fortin M, Aquilué N. 2021. Network analysis can guide resilience‐based management in forest landscapes under global change. Ecological Applications 31:e2221.

Minor ES, Urban DL. 2008. A graph-theory framework for evaluating landscape connectivity and conservation planning. Conservation Biology 22:297–307.

Moilanen A, Franco AMA, Early RI, Fox R, Wintle B, Thomas CD. 2005. Prioritizing multiple- use landscapes for conservation: methods for large multi-species planning problems. Proceedings. Biological sciences 272:1885–91.

Mozelewski TG, Scheller RM. 2021. Forecasting for intended consequences. Conservation Science and Practice 3:e370.

North Carolina State Climate Office. NC Climate Synopsis. https://climate.ncsu.edu/climate/synopsis.

Oliver TH, Morecroft MD. 2014. Interactions between climate change and land use change on biodiversity: attribution problems, risks, and opportunities. Wiley Interdisciplinary Reviews: Climate Change 5:317–335.

Perring MP, Standish RJ, Price JN, Craig MD, Erickson TE, Ruthrof KX, Whiteley AS, Valentine LE, Hobbs RJ. 2015. Advances in restoration ecology: Rising to the challenges of the coming decades. Ecosphere 6:1-25.

Pinto N, Keitt TH. 2009. Beyond the least-cost path: Evaluating corridor redundancy using a graph-theoretic approach. Landscape Ecology 24:253–266.

Polasky S. 2008. Why conservation planning needs socioeconomic data. PNAS 105:6505-6506.

Pressey RL, Cabeza M, Watts ME, Cowling RM, Wilson KA. 2007. Conservation planning in a changing world. Trends in Ecology & Evolution 22:583–592.

R Core Team. R: A Language and Environment for Statistical Computing. (R Foundation for Statistical Computing, 2017).

Reside AE, Butt N, Adams VM. 2018. Adapting systematic conservation planning for climate change. Biodiversity & Conservation 27:1-29.

Robillard CM, Kerr JT. 2017. Assessing the shelf life of cost-efficient conservation plans for species at risk across gradients of agricultural land use. Conservation Biology 31:837–847.

Rojas IM, Pidgeon AM, Radeloff VC. 2020. Restoring riparian forests according to existing regulations could greatly improve connectivity for forest fauna in Chile. Landscape and Urban Planning 203:103895.

Saura S, Bodin Ö, Fortin M-J. 2014. EDITOR’S CHOICE: Stepping stones are crucial for species’ long-distance dispersal and range expansion through habitat networks. Journal of Applied Ecology 51:171–182.

Saura S, Estreguil C, Mouton C, Rodríguez-Freire M. 2011. Network analysis to assess landscape connectivity trends: Application to European forests (1990-2000). Ecological Indicators 11:407–416.

Saura S, Rubio L. 2010. A common currency for the different ways in which patches and links can contribute to habitat availability and connectivity in the landscape. Ecography 33:523- 537.

Saura S, Torne J. 2009. Conefor Sensinode 2.2: a software package for quantifying the importance of habitat patches for landscape connectivity. Environment Modeling & Software 24:135-139.

SCBD. In COP 10 Decision X/2: Strategic Plan for Biodiversity 2011–2020 (Secretariat of the Convention on Biological Diversity, 2010).

Scheller RM, van Tuyl S, Clark KL, Hom J, La Puma I. 2011. Carbon Sequestration in the New Jersey Pine Barrens Under Different Scenarios of Fire Management. Ecosystems 14:987– 1004.

Shirk AJ, Schroeder MA, Robb LA, Cushman SA. 2015. Empirical validation of landscape resistance models: insights from the Greater Sage-Grouse (Centrocercus urophasianus). Landscape Ecology 30:1837–1850.

100

Schrum P, Scheller RM, Duveneck MJ, Lucash MS. 2020. Base-Hurricane: A new extension for the Landis-II forest landscape model. Environmental Modelling and Software 133:104833.

Smith AM, M. Green D. 2005. Dispersal and the metapopulation paradigm in amphibian ecology and conservation: are all amphibian populations metapopulations? Ecography 28:110–128.

Soares-Filho BS, Nepstad DC, Curran LM, Cerqueira GC, Garcia RA, Ramos CA, Voll E, McDonald A, Lefebvre P, Schlesinger P. 2006. Modelling conservation in the Amazon basin. Nature 440:520–523.

Stevenson-Holt CD, Watts K, Bellamy CC, Nevin OT, Ramsey AD. 2014. Defining Landscape Resistance Values in Least-Cost Connectivity Models for the Invasive Grey Squirrel: A Comparison of Approaches Using Expert-Opinion and Habitat Suitability Modelling. PLoS ONE 9:e112119.

Sturtevant BR, Cooke BJ, Kneeshaw DD, Mac Lean DA. 2015. Modeling insect disturbance across forested landscapes: Insights from the spruce budworm. In: Perera A, Sturtevant B, Buse L. (eds) Simulation Modeling of Forest Landscape Disturbances. Springer, Cham. Sutherland, G. D., A. S. Harestad, K. Price, and K. P. Lertzman. 2000. Scaling of natal dispersal distances in terrestrial birds and mammals. Conservation Ecology 4:1-31. Taylor PD, Fahrig L, Henein K. 1993. Connectivity Is a Vital Element of Landscape Structure. Oikos 68:57-573.

Terando AJ, Costanza J, Belyea C, Dunn RR, McKerrow A, Collazo JA. 2014. The Southern Megalopolis: Using the Past to Predict the Future of Urban Sprawl in the Southeast U.S. PLoS ONE 9:e102261.

Theobald DM, Reed SE, Fields K, Soulé M. 2012. Connecting natural landscapes using a landscape permeability model to prioritize conservation activities in the United States. Conservation Letters 5:123–133.

Thuiller W et al. 2008. Predicting global change impacts on plant species’ distributions: Future challenges. Perspectives in Plant Ecology, Evolution and Systematics 9:137–152.

Tulbure MG, Kininmonth S, Broich M. 2014. Spatiotemporal dynamics of surface water networks across a global biodiversity hotspot - Implications for conservation. Environmental Research Letters 9:114012.

U.S. Geological Survey (USGS) Gap Analysis Project (GAP). 2018. Protected Areas Database of the United States (PAD-US): U.S. Geological Survey data release, https://doi.org/10.5066/P955KPLE.

101

Wear DN, Greis JG. 2002. Southern Forest Resource Assessment: Summary of findings. Journal of Forestry 100:6–15.

Wimberly MC. 2006. Species dynamics in disturbed landscapes: When does a shifting habitat mosaic enhance connectivity? Landscape Ecology 21:35–46.

Zeller KA, Lewsion R, Fletcher RJ, Tulbure MG, Jennings MK. 2020. Understanding the importance of dynamic landscape connectivity. Land 9:303.

102

Chapter 4: Conservation Contributions to Landscape Connectivity under Global Change

4.1 Abstract

Both environmental variability and anthropogenic change cause spatiotemporal

fluctuations in the availability of habitat resources on a landscape. Active land acquisition and

restoration for conservation also facilitates dynamic habitat accessibility and suitability over

time. This can make forecasting landscape permeability a challenge, the improvement of which

is a core conservation strategy for reducing the fragmenting effects of anthropogenic change.

Accounting for these spatiotemporal dynamics is critical to the success of conservation planning for connected landscapes. We assessed the combined effects of anthropogenic change, environmental variability, and conservation on dynamic landscape connectivity across a 1.5 million ha landscape in central North Carolina over an 80-year time period (2020-2100). Using graph theoretic metrics, we quantified landscape-level connectivity across multiple scenarios of anthropogenic change and tropical cyclone events. We simulated the creation and expansion of

protected areas through land acquisition and restoration and forecasted the variance in connectivity created by spatiotemporal changes to complex habitat networks. Finally, we

evaluated the capacity of aggressive conservation action to keep pace with global change. Our

study showed that under the most extreme climate and land use change scenarios tested, no

conservation actions were able to achieve substantial connectivity improvement objectives.

4.2 Introduction

More than a century of accelerating anthropogenic change has resulted in a biodiversity

crisis (Goodwin & Fahrig 2002; Thomas et al. 2005; Zeller et al. 2012; De Vos et al. 2015).

Global climate change is triggering changes in species phenology and shifting species

distributions, disaggregating ecological communities and decoupling coevolved relationships,

103

and altering habitat availability (Walther et al. 2002; Parmesan & Yohe 2003; Parmesan 2006;

Mawdsley et al. 2009). Land use and land cover changes are driving habitat conversion, fragmenting intact habitat (Fahrig 2003; Fischer & Lindenmayer 2007; de Chazal & Rounsevell

2009), and disrupting ecological service provision (Metzger et al. 2006). Climate and land use

change represent two of the most significant drivers of habitat loss worldwide (Costanza &

Terando 2019) and are expected to act synergistically, exacerbating habitat decline and reducing

landscape connectivity, increasing the vulnerability of fragmented populations to extinction

(Mantyka-Pringle et al. 2015; Dilts et al. 2016).

Connected landscapes facilitate species movement between resource patches (Taylor et

al. 1993) and enable movement in response to changing environments, allowing species to move

following their climate envelopes as the climate changes and reducing the fragmenting effects of

land use change (Perring et al. 2015). By promoting migration, dispersal, and gene flow,

connected landscapes support the persistence of wildlife populations and their adaptation to

environmental change (Santini et al. 2016). Because of this, maintaining and enhancing

landscape connectivity has been increasingly implemented as a core conservation strategy to

combat biodiversity decline (Heller & Zavaleta 2009; Nuñez et al. 2013).

One approach to both enhance landscape connectivity and counteract habitat loss and

conversion is the expansion of protected area networks (Andrello et al. 2015). Protected areas are

essential for meeting global conservation targets and supporting ecosystem functioning,

providing and protecting additional habitat and augmenting connectivity (Le Saout et al. 2013;

Elsen et al. 2020). This emphasis on protected area establishment for both biodiversity

conservation and connectivity enhancement was highlighted in The Convention on Biological

Diversity’s Strategic Plan for Biodiversity 2011-2020, which targeted the expansion of protected

104

area networks to cover 17% of the terrestrial environment while maintaining and improving network connectivity (SCBD 2010).

Integrating landscape-level interactions between climate change and land use/land cover change into conservation planning for connectivity is critical as their synergistic effects will likely change where and how to best allocate resources and alter conservation strategy effectiveness (de Chazal & Rounsevell 2009). Studies on conservation strategy prioritization often incorporate future climate change or future land use change but seldom consider both drivers jointly, especially when forecasting future habitat connectivity (Costanza & Terando

2019). Yet these effects will strongly influence protected area network success and species dispersal ability, crucial considerations when evaluating whether conservation can keep pace with accelerating global change (Mantyka-Pringle et al. 2015). Because of this, the intersection of the landscape-level effects of global change and landscape-level responses to future conservation represents an understudied research need.

In this study, we applied graph theory to scenario-based projections of future climate and land use change to evaluate changes to complex habitat networks on a 1.5 million hectare study landscape in North Carolina of immense conservation interest from 2020-2100. We then systematically integrated two adaptive conservation strategies that seek to address anthropogenic change (conserving geodiverse lands and conserving lands near already established conservation cores) to assess whether aggressive conservation action will be able to counteract climate and land use change effects on connectivity, all on a spatiotemporally dynamic landscape. This framework represents a promising new approach for evaluating the ability of conservation actions to adapt to anthropogenic change.

105

4.3. Methods

4.3.1 Study area

We focused our study on a 1.5 million hectare landscape located in central North

Carolina, USA (Figure 4.1). The climate of this region is humid subtropical with hot, humid summers (July mean temperature: 25°C) and mild winters (January mean temperature: 3.3°C).

There is no distinct wet and dry season and mean annual rainfall is between 1016-1398 mm (NC

State Climate Office). The study area contains three ecoregions: the Atlantic Coastal Plain, the

Sandhills, and the Piedmont, each with a combination of longleaf pine (Pinus palustris Mill.)

forests, loblolly pine (Pinus taeda L.) plantations, and pine and hardwood mixed forest types and

distinct understory plant communities. Forests make up about 58% of the study area with urban,

suburban, and agricultural land uses also present throughout. This region is experiencing

increasing urbanization pressures and accelerating land use change as urban cores such as

Charlotte and Raleigh/Durham expand (Terando et al. 2014). Climate change is also playing a

substantial role in driving landscape change, altering the feasibility and timing of management

approaches like prescribed burning and potentially shifting forest type composition (Mitchell et

al. 2014). The study area has been the center of multi-agency partnerships dedicated to

enhancing connectivity between established natural resources areas (e.g. Uwharrie National

Forest and Ft. Bragg Army base) for the benefit of threatened and endangered species and to

promote longleaf pine restoration (Costanza et al. 2020). These partnerships have been incredibly

active in the acquisition, restoration, and management of land for conservation in the study area.

106

Figure 4.1. The study area in central North Carolina, shown with abbreviated NLCD 2016 land use types.

4.3.2 Landscape change modeling

Scenario-based projections of landscape responses to future climate and land use change and conservation can help to ensure that habitat networks enhanced through land acquisition and restoration are robust to as many feasible scenarios of change as possible (Albert et al. 2017). We used LANDIS-II (Scheller et al. 2007), a spatially explicit landscape change model, to forecast landscape-scale forest dynamics under climate change, land use change, and conservation action implementation. LANDIS-II simulates changes to forest landscape conditions over time, incorporating ecological processes including vegetation dynamics, such as regeneration, growth, mortality, and dispersal, and disturbance regimes (Nitschke et al. 2020). In LANDIS-II, tree and shrub species compete for resources (soil moisture, nitrogen, light) and are exposed to disturbances and management actions within a landscape composed of grid cells (Scheller et al.

2007). Landscape and site-level demographics are determined by species and functional group

107

traits (e.g. longevity, age of sexual maturity, and shade tolerance) that influence the ability of species-age cohorts to survive and reproduce (Scheller & Mladenoff 2008). We parameterized

LANDIS-II for the study area in central North Carolina from 2020 to 2100 on a one hectare grid, evaluating changes to habitat networks and the surrounding matrix. For detailed information about model extensions, parameterization, and calibration, see Appendix A.

We simulated three harvest prescriptions and three restoration prescriptions to capture forest management and conservation practices in central North Carolina (Table 4.1). Harvest prescriptions consisted of loblolly pine clear cutting and replanting, longleaf pine thinning and burning to simulate the prescribed burning frequently done on Ft. Bragg and throughout the

108

Table 4.1. Harvest and restoration prescriptions simulated on the study landscape in central North Carolina.

We used global climate model (GCM) projections derived from the Coupled Model

Intercomparison Project Phase 5 downscaled using the Multivariate Adaptive Constructed

Analogs (MACA) method (Abatzoglou 2013). From this ensemble, we chose outputs from five

GCMs for the Representative Concentration Pathways (RCP) 4.5 and 8.5 that bracketed the

temperature and precipitation projections for the study area and served as our model replicates to

capture a range of possible future conditions in the study area: bcc-csm1, CNRM-CM5, Had-

GEM2, IPSL-CM5A-LR, and Nor-ESM1-M (Grey 2018). These GCMs were selected to

encompass a range of temperature and precipitation changes, from warm to hottest and wettest to

driest. To understand the influence of land use change on landscape connectivity, we used two shared socioeconomic pathway (SSP) scenarios from the EPA’s Integrated Climate and Land-

109

Use Scenarios (ICLUS) v2.1.1 (EPA 2016). SSP2 represents a middle of the road land use change scenario in which population growth is moderate and levels off in the second half of the century. SSP5 represents a scenario in which strong market forces and technological progress support both economic and social development leading to greater population growth and urban expansion.

In total, we simulated four climate and land use change scenarios (RCP 4.5 and static land use, RCP 8.5 and static land use, RCP 4.5 and SSP2 land use change, RCP 8.5 and SSP5 land use change) with and without conservation for a total of 60 simulations (4 climate and land use change scenarios × 3 conservation scenarios × 5 replicates).

4.3.3 Conservation scenarios

We simulated two protected area network development strategies designed to support global change-adaptive conservation: i) focusing on the acquisition and restoration of lands in close proximity to existing protected areas (hereafter referred to as the cluster strategy) and ii) emphasizing the acquisition and restoration of areas exhibiting strong environmental niche gradients (hereafter, the geodiversity strategy; see below for details of both strategies). Each strategy simulated an aggressive rate of conservation, acquiring and restoring 1% of the study area every five years for a total of 17% of the landscape in line with the Convention on

Biological Diversity’s terrestrial land restoration targets (SCBD 2010). Land acquired and restored for conservation in model simulations remained in conservation status in perpetuity and was not subject to land use changes. Connectivity was evaluated for a functional species guild of high-dispersal (5000m maximum dispersal distance) habitat specialists with longleaf pine forest habitat preferences.

110

Our ‘cluster strategy’ simulated the creation new protected areas around land already

managed for conservation, creating larger core areas of habitat and reducing edge effects, and

increasing the effective size of populations; increasing overall resilience to anthropogenic change

(Oliver & Morecroft 2014; Herse et al. 2018). We created buffers at 100, 500, 1000, 2500 meters

around established protected areas identified using the US Geological Survey’s Protected Areas

database (USGS). Forest stands whose centers fell within the 100m buffer had the highest

probability of being selected for conservation under this strategy, followed by stands that had

centers within the 500m buffer, then 1000m, then 2500m via a weighted random sort (Appendix

C). Stands with centers that fell outside of these buffers had the lowest probability of selection.

Our ‘geodiversity strategy’ preferentially selected stands with the highest geodiversity

attributes, those that possess a diversity of abiotic conditions, for acquisition and restoration

(Grey 2008; Lawler et al. 2015). Geodiverse lands often result in greater environmental niches

(Comer et al. 2015) and as such possess a greater capacity to maintain species diversity and ecological function as the climate changes which may confer greater resilience (Anderson et al.

2015). We assigned an average geodiversity score to each forest stand using geodiversity data from the Nature Conservancy’s resilient and connected landscapes map (Anderson et al. 2016).

We then performed a weighted random sort with forest stands scoring the highest in geodiversity having the highest probability of being selected (Appendix C).

4.3.4 Connectivity analysis

We quantified connectivity for the study area by constructing habitat networks based on least cost path resistance distances between all pairs of habitat patches within 5000 meters pair- wise Euclidean distance from one another. We chose 5000m as our maximum dispersal distance to encompass a range of possible species and consistent with other dispersal distance thresholds

111

seen in the literature (e.g Saura et al. 2011; Bishop-Taylor et al. 2018). Resistance values

represent the influence of a heterogeneous landscape matrix on dispersal and were based on the

assumption that forest species will face greater dispersal difficulty as they move through land

cover types increasingly disparate from the forested areas where they reside (Saura et al. 2011).

We used a rating scale of 1-100 for resistance values, consistent with the scale of possible

resistances found in previous studies (e.g. Greenwald et al. 2009; Stevenson-Holt et al. 2014;

Shirk et al. 2015; Blazquez-Cabrera et al. 2016), with 1 representing the lowest cost to

movement (e.g. mature longleaf pine forest) and 100 representing the highest (e.g. a highway or

urban core). Least-cost paths were calculated between the centroids of each pair of habitat

patches and reflect the minimum cost accumulated along the shortest path between them

(Bishop-Taylor et al. 2015; van Etten 2017). By using habitat patch centroids, we were able to

simulate species movement between and within habitat patches as a continuous process rather

than assuming patch homogenization or an immediate end to species movement at a patch edge

(Dickson et al. 2017). We considered a cell on the landscape to be habitat if at least 25% of

total biomass was longleaf pine and the median stand age was at least 21. We targeted

longleaf pine because this habitat type has been the focus of regional conservation and

restoration efforts for decades (Costanza et al. 2015).

To identify habitat nodes and network changes in response to climate change, land use

change, and conservation, we grouped contiguous habitat cells that co-occurred in both time steps t and t-1 discretely from those that only occurred in time step t for each time step and replicate using an eight neighbor rule (Appendix D). Doing so allowed us to capture habitat asynchrony and more realistically represent how an organism would perceive spatiotemporal

112

habitat fluctuations, better capturing intra- and inter-patch movement dynamics (Hanski 1999;

Wimberly 2006). Habitat nodes were weighted by area.

We used the Equivalent Connected Area (ECA) index to quantify landscape-level trends

in functional connectivity over time across each scenario of global change with and without

conservation intervention (Saura et al. 2011; Tulbure et al. 2014). ECA measures the size of a

single habitat patch, maximally connected, that if added to the landscape would provide the same

probability of connectivity as the existing habitat network (Saura et al. 2011). ECA measures

connectivity by accounting for both the habitat area within each node and the habitat made

available through dispersal, incorporating the connectivity contribution of nodes and links that serve as stepping stones between other habitat areas (Saura & Rubio 2010). In this approach, the landscape is represented as a set of discrete habitat patches, or nodes, connected by least cost path links that characterize species dispersal ability (Calabrese & Fagan 2004). ECA indices use area units (e.g. hectares) enabling a direct comparison between changes in total habitat area and network spatial configuration to assess how the addition or loss of habitat, whether from land acquisition or landscape dynamics, affects connectivity (Bishop-Taylor et al. 2018; McIntyre et al. 2018).

As part of our analysis, we also quantified connectivity for three ECA subcomponents:

(i) intra-patch connectivity (ECAintra), which describes the connectivity contributed by the area

within each habitat node; (ii) the connectivity contributed by direct connections between

neighboring habitat nodes (ECAdirect), and (iii) the connectivity contributed by intermediate

stepping stones that enabled longer distance dispersal (ECAstep) (Saura and Rubio 2010, Saura et

al. 2011, Bishop-Taylor et al. 2018). All indicators were calculated at 10-year time steps using

113

the probabilistic PC formulation of ECA obtained via a negative exponential function of inter- patch least cost path.

LANDIS-II outputs were processed using the raster library in the statistical software package R (v3.4.3) (Hijmans and van Etten 2012, R Development Core Team 2017). To conduct our connectivity analysis we used the gDistance (Van Etten 2017), spatstat (Baddeley and Turner

2005), igraph (Csardi and Nepusz 2006), and Conefor command line (v2.6- Saura and Torne

2009) R packages. We used the high performance computing cluster at North Carolina State to process connectivity indices.

4.4 Results

4.4.1 Nodes and links

Our analysis of dynamic landscape connectivity across scenarios of climate and land use change, with and without conservation intervention, revealed that habitat networks varied substantially between scenarios. Under both RCPs, the geodiversity conservation strategy without the inclusion of land use change averaged the greatest number of nodes (Figures 4.2a and 4.2b). Targeting conservation on geodiverse lands also yielded the greatest number of nodes for both RCPs when land use change was included. We observed an approximate 25% reduction in habitat nodes between no land use change and land use change scenarios under the RCP 4.5 climate change scenarios (Figure 4.2a) and an approximate 34% reduction in nodes between no

LUC and LUC scenarios under RCP 8.5 (Figure 4.2b). The addition of land acquisition and restoration for conservation increased the number of available habitat nodes by approximately

11-25% for RCP 4.5 with no LUC (11%-cluster, 25% geodiversity), 11-22% for RCP 4.5 with

LUC (cluster-11%, geodiversity 22%), by approximately 25% for RCP 8.5 with no LUC, and between 10-20% for RCP 8.5 with LUC (cluster-10%, geodiversity-20%). The geodiversity

114

strategy without land use change incorporated also had the greatest number of links between habitat nodes across both RCPs (Figure 4.2c and 4.2d), though the cluster strategy without LUC facilitated a similar number of links under RCP 8.5. The addition of land use change significantly reduced the number of links between habitats under both RCPs. LUC reduced links by 23% without conservation, by 25% for the cluster strategy, and by 32% for the geodiversity strategy under RCP 4.5 (Figure 4.2c). LUC reduced links by 39% without conservation, by 47% for the clusters strategy, and by 51% for the geodiversity strategy under RCP 8.5.

115

Figure 4.2. Habitat networks from 2020-2100 showing the number of habitat nodes (panels a and b) and links between nodes (panels c and d) for four scenarios of global change (RCP 4.5 and 8.5 with and without land use change) with and without conservation (conservation strategies include clustering conservation around established conservation cores and prioritizing geodiverse lands for conservation).

116

4.4.2 Equivalent Connected Area

Equivalent connected area indices were strongly influenced by both the climate change

and inclusion of land use change and, to a lesser degree, by the conservation strategy employed

under both climate change scenarios (Tables 4.2 and 4.3; Figure 4.3a and 4.3b). Under RCP 4.5,

all scenarios that included land use change, as well as the no LUC and no conservation scenario,

showed strong increases in ECA up until model year 40, after which connectivity began to

decline. This early increase resulted in a nearly 330% increase in connectivity on average in the

first half of the model run for all LUC scenarios, an improvement in ECA of about 28,000 hectares, and a 350% increase for the no LUC no conservation scenario, an improvement in ECA of about 32,000 hectares. After declining, average final ECA indices showed a 225% increase for the LUC no conservation scenario (an ECA improvement of 15,111 hectares), a 260% increase in the no LUC no conservation scenario (an ECA improvement of 20,540 hectares), a 292% increase in connectivity for the LUC and cluster scenario (an ECA improvement of 23,192 hectares), and a 276% increase in connectivity for the LUC and geodiversity scenario (an ECA improvement of 21,261 hectares) by year 2100. Without land use change, the two conservation strategies under RCP 4.5 resulted in an average 349% increase in connectivity for the cluster scenario (an ECA improvement of 31,575 hectares) and an average 374% increase in connectivity for the geodiversity scenario (an ECA improvement of 28,811 hectares; Figure

4.3a).

117

Table 4.2. Changes to Equivalent Connected Area (ECA) over time under RCP 4.5 with and without land use change (LUC) and conservation. No Conservation LUC No LUC Difference Difference Time ECA (ha) % Change (ha) ECA (ha) % Change (ha) 0 12062 100 0 12859.7 100 0 20 34453.8 285.6392 22391.8 37037.84 288.0148 24178.14 40 39371.76 326.4115 27309.76 45177.58 351.3113 32317.88 60 35783.48 296.6629 23721.48 40923.32 318.2292 28063.62 80 27173.24 225.2797 15111.24 33400.16 259.7274 20540.46

Cluster LUC No LUC Difference Difference Time ECA (ha) % Change (ha) ECA (ha) % Change (ha) 0 12054.38 100 0 12692.54 100 0 20 34346.78 284.932 22292.4 37811.6 297.9041 25119.06 40 39987.86 331.7289 27933.48 43980.48 346.5065 31287.94 60 38866.84 322.4292 26812.46 44407.04 349.8672 31714.5 80 35246.14 292.3928 23191.76 44267.58 348.7685 31575.04

Geodiversity LUC No LUC Difference Difference Time ECA (ha) % Change (ha) ECA (ha) % Change (ha) 0 12061.9 100 0 12692.54 100 0 20 37408.4 310.1369 25346.5 36967.28 291.252 24274.74 40 40207.66 333.3443 28145.76 45270.12 325.4932 32577.58 60 34778.92 288.337 22717.02 42603.78 335.66 29911.24 80 33323.36 276.2696 21261.46 41503.22 374.3416 28810.68

Connectivity increases were greater under RCP 8.5 than RCP 4.5 for every scenario,

especially when land use change was considered (Tables 4.2 and 4.3). Without the inclusion of

land use change, the no conservation scenario increased in ECA by 335% (an ECA improvement of 30,269 hectares). Average final ECA indices showed a 316% increase for the LUC no conservation scenario (an ECA improvement of 25,974 hectares), a 350% increase in connectivity for the LUC and cluster scenario (an ECA improvement of 30,088 hectares), and a

118

335% increase in connectivity for the LUC and geodiversity scenario (an ECA improvement of

28,627 hectares) by year 2100. Without land use change, the two conservation strategies under

RCP 8.5 resulted in an average 376% increase in connectivity for the cluster scenario (an ECA improvement of 35,027 hectares) and an average 334% increase in connectivity for the geodiversity scenario (an ECA improvement of 29,696 hectares). Without land use change, the geodiversity scenario had a much sharper initial increase while the cluster scenario had a more gradual increase in ECA under RCP 8.5 (Figure 4.3b).

119

Table 4.3. Changes to Equivalent Connected Area (ECA) over time under RCP 8.5 with and without land use change (LUC) and conservation. No Conservation LUC No LUC Difference Difference Time ECA (ha) % Change (ha) ECA (ha) % Change (ha) 0 12012.9 100 0 12859.7 100 0 20 31095.06 258.8472 19082.16 37147.84 288.8702 24288.14 40 41282.76 343.6536 29269.86 41905.28 325.8651 29045.58 60 42822.78 356.4733 30809.88 42595.82 331.2349 29736.12 80 37987.26 316.2206 25974.36 43128.26 335.3753 30268.56

Cluster LUC No LUC Difference Difference Time ECA (ha) % Change (ha) ECA (ha) % Change (ha) 0 12015.1 100 0 12692.54 100 0 20 34201.4 284.6535 22186.3 32775.2 258.2241 20082.66 40 41448.76 344.9722 29433.66 37874.6 298.4005 25182.06 60 38318.24 318.9174 26303.14 41711.64 328.6311 29019.1 80 42103.04 350.4177 30087.94 47719.46 375.9646 35026.92

Geodiversity LUC No LUC Difference Difference Time ECA (ha) % Change (ha) ECA (ha) % Change (ha) 0 12023.9 100 0 12692.54 100 0 20 37475.62 311.6761 25451.72 35045.14 276.1082 22352.6 40 42984.7 357.4938 30960.8 44143.42 347.7903 31450.88 60 38305.1 318.5747 26281.2 45521.14 358.6448 32828.6 80 40290.9 335.0901 28267 42388.1 333.9607 29695.56

4.4.3 ECAintra, ECA direct, and ECAstep

Intra-patch connectivity importance (ECAintra, Figures 4.3c and 4.3d) was higher for all scenarios that included land use change relative to their no land use change counterparts under both representative concentration pathways. Under RCP 4.5 the importance of ECAintra peaked for all scenarios approximately halfway through model runs (around year 2060), after which it declined (Figure 4.3c). Both with and without land use change, the importance of intra-patch

120

connectivity was higher for conservation scenarios than no conservations scenarios under RCP

4.5. Under RCP 8.5, ECAintra increased over time for all scenarios that included land use change, especially for the no conservation scenario. Scenarios without land use change were highest at year 2050 before starting a gradual decline and slightly increasing again (Figure 4.3d). Under this RCP, intra-patch connectivity was higher in the no conservation scenarios than either conservation scenario with and without land use change incorporated.

Connectivity between neighboring habitat nodes (ECA direct, Figures 4.3e and 4.3f) declined in its contribution to overall landscape connectivity until year 2050. ECA direct then retained a constant rate of contribution to connectivity for all scenarios under RCP 4.5 (Figure

4.3e) and before beginning a gradual increase in contribution for all scenarios under RCP 8.5

(Figure 4.3f).

The importance of stepping stones to landscape connectivity (ECAstep) was higher for all scenarios under RCP 4.5 (Figure 4.3e) than their respective counterparts under RCP 8.5 (Figure

4.3f). Stepping stone importance declined under both RCPs and across all scenarios until 2050, after which ECAstep began to increase for all RCP 4.5 scenarios. After 2050, the importance of stepping stones stayed relatively constant for all scenarios under RCP 8.5 except for the no conservation with land use change scenario, which declined for the entirety of the 80-year modeled timescale.

121

Figure 4.3. Equivalent connected area indices (panels a and b) and ECA components (panels c through h) for global change and conservation strategies from 2020-2100.

122

4.5 Discussion

We observed that connectivity consistently improved more under RCP 8.5 and shared socioeconomic pathway 5 than RCP 4.5 and SSP2, though land use change under SSP5 is expected to be more severe as population growth and urbanization increase. With land use change excluded, connectivity improved even without conservation under both RCPs. This supports past studies that found that a changing climate is likely to have positive impacts on longleaf pine, increasing its dominance in the southeast, in part because longleaf pine is tolerant of drought and heat stress (Prasad et al. 2007; Samuelson et al. 2012; Costanza et al. 2015).

While a warmer climate could favor the expansion of longleaf pine forests, increasing habitat availability and improving connectivity, it could also make the implementation of prescribed burning, which is currently the principal strategy for maintaining and restoring the longleaf ecosystem, more difficult (Mitchell et al. 2014; Kupfer et al. 2020). Longleaf pine is a fire- adapted ecosystem and prescribed burning increases soil nutrient availability, reduces hardwood competition, and improves seedling establishment (Heuberger & Putz 2003). Urbanization will likely exacerbate limitations on prescribed burning as the wildland-urban interface expands, potentially decreasing longleaf area (Costanza and Moody 2011). We see the integration of changing burn windows and locations as an important next step in forecasting future connectivity in the study area.

Our analysis also found that stepping stones will be more important to future connectivity in scenarios without land use change while intra-patch connectivity will contribute more heavily to connectivity when land use change is incorporated (Figure 4.4). This was especially true for the no conservation with land use change scenario under RCP 8.5. As longleaf pine was added to the landscape through conservation or through increasing landscape dominance as the climate

123

warmed, the shift in connectivity importance from ECAintra to ECAstep reflects increasing dispersal potential as new habitat patches were formed (Bishop-Taylor et al. 2018) without land use change pressures. When land use change was considered, intra-patch connectivity was most important for overall landscape connectivity as land use changes made stepping stones more scarce or difficult to reach (also observed in Piquer-Rodríguez et al. 2015) and connectivity relied on large, intact habitat holdouts. This finding is especially important because protected areas added to the landscape through simulated land acquisition and restoration remained protected for the remainder of the model run and not subject to land use change. The inclusion of land use change strongly influenced the ability of species to move across the landscape.

Figure 4.4. Habitat network changes under RCP 8.5 with and without conservation for both no land use change and land use change scenarios.

124

We observed connectivity declines in the second half of model time steps under RCP 4.5 across nearly all scenarios. These declines were largely due to decreases in the number of possible links between habitat nodes between time steps 40 (2060) and 80 (2100). This may be because of an increase in red maple, which increased both in the number of hectares occupied

(by approximately 10%) and in mean biomass (g m-2) per hectare (by approximately 20%) during this time. This expansion of red maple, which may have occurred after warmer temperatures increased long leaf pine establishment but a lack of additional management to suppress hardwood species on the landscape lead to long leaf eventually being outcompeted, reduced the ability of specialist species to move from habitat patch to habitat patch, reducing connectivity.

Finally, we found that the capacity of both conservation strategies to facilitate connectivity was reduced by land use change (Riordan & Rundel 2014). Under RCP 4.5 the inclusion of land use change reduced the equivalent connected area by ~9,000 hectares, or about

25%, compared to scenarios without land use change. The inclusion of land use change under

RCP 8.5 resulted in a less substantial, but still apparent, loss of connectivity. Interestingly, while conservation enacted in land use change scenarios under RCP 8.5 resulted still resulted in small connectivity improvements, the effect of conservation without land use change incorporated was negligible. Under the most extreme climate and land use change scenarios simulated, conservation was unable to substantially increase connectivity beyond baseline, no conservation improvements. This information could be incredibly useful for decision makers who want to develop conservation strategies that will be robust to the most extreme scenarios of change. Our analysis highlights the need for the expansion of protected area networks beyond a 17% target to see improvements in dynamic landscape connectivity under more aggressive climate and land

125

use change scenarios. Furthermore, our analysis underscores the importance of incorporating both climate and land use change into conservation planning to make habitat networks robust to global change (Albert et al. 2017).

4.5.1 Limitations and Future Work

While we believe the utility of simulation modeling to inform conservation decision- making is clear, this approach is not intended to predict the exact outcomes of landscape connectivity in response to global change and conservation. Instead, our use of a landscape change model operated under a scenario framework is intended to evaluate possible intended and unintended consequences of conservation (Mozelewski and Scheller 2021), helping to bound the decision-making process and assessing whether conservation can keep pace with global change

(Hobbs et al. 2014). While LANDIS has been widely used and rigorously tested for evaluating landscape-level responses to management and disturbance drivers (Scheller et al. 2011; Lucash et al. 2018; Boulanger et al. 2019; Maxwell et al. 2020), as with any model, uncertainty in parameter estimates and representation of ecological processes exists. Additionally, our focus on a limited set of landscape-scale dynamics excluded insect disturbance (Sturtevant et al. 2015) and species range expansions into the study area as the climate warms (Van Houtven et al. 2019), which have been shown to shift community composition in the eastern United States (Clark et al.

2021). During this analysis, we did not simulate a reduction in prescribed burn windows due to climate change or in prescribed burn acreage in response to urbanization, but we feel that this presents an exciting avenue for future research.

The use of least cost path resistance distances in our analysis is based on the assumption that that forest species will face greater dispersal difficulty as they move through land cover types increasingly disparate from their preferred habitat as described in Saura et al. (2011). Yet

126

the use of least cost paths rely on two simplifying assumptions: that a single optimal path between habitats conveys the movement potential of that segment of the landscape, and that an organism has complete knowledge of the landscape and can select a movement path accordingly

(Fahrig 2007; McRae et al. 2008; Pinto & Keitt 2009; Bishop-Taylor et al. 2015). Additionally, the landscape matrix simulated through our modeling efforts is not a perfect representation of the land cover and vegetation types of study area. In our analysis, habitat nodes were weighted by area, but habitat quality metrics could be incorporated to further account for variation in habitat suitability.

4.5.2 Conclusion

To be effective in the face of climate and land use change, networks of protected areas must be able to simultaneously accomplish multiple conservation objectives and continue to do so as global change accelerates (Zetterberg et al. 2010; Albert et al. 2017). Climate and land use change are expected to act in concert, resulting in optimal locations for protected area networks that differ from those when only a single driver is considered (de Chazal & Rounsevell 2009).

Yet conservation prioritization seldom considers both drivers jointly, especially in regards to habitat connectivity (Costanza & Terando 2019), which could impact the ability of conservation to keep pace with global change (Mantyka-Pringle et al. 2015). Our study addressed this research need (Correa Ayram et al. 2016), combining landscape change modeling and graph theory network analysis to forecast dynamic connectivity across scenarios of climate and land use change with and without conservation intervention. In doing so, we evaluated whether aggressive conservation action- the acquisition and restoration of land for protected areas- was able to counteract climate and land use change effects on connectivity. Our study highlights significant differences in connectivity when both global change drivers are considered compared to climate

127

change alone and underscores the need for conservation action that goes beyond SCBD terrestrial land protection targets under more extreme climate and land use change scenarios.

128

4.6 References

Abatzoglou JT. 2013. Development of gridded surface meteorological data for ecological applications and modelling. International Journal of Climatology 33:121–131. Albert CH, Rayfield B, Dumitru M, Gonzalez A. 2017. Applying network theory to prioritize multispecies habitat networks that are robust to climate and land-use change. Conservation Biology 31:1383–1396. Anderson MG, Barnett A, Clark M, Prince J, Olivero Sheldon A, Vickery B. 2016. Resilient and Connected Landscapes for Terrestrial Conservation. The Nature Conservancy, Eastern Conservation Science, Eastern Regional Office. Boston, MA.

Anderson MG, Comer PJ, Beier P, Lawler JJ, Schloss CA, Buttrick S, Albano CM, Faith DP. 2015. Case studies of conservation plans that incorporate geodiversity. Conservation Biology 29:680–691. Andrello M, Jacobi MN, Manel S, Thuiller W, Mouillot D. 2015. Extending networks of protected areas to optimize connectivity and population growth rate. Ecography 38:273– 282.

Baddeley A, Turner R. 2005. spatstat: An R package for analyzing spation point patterns. Jornal of Statistical Software 12:1-42. Bishop-Taylor R, Tulbure MG, Broich M. 2015. Surface water network structure, landscape resistance to movement and flooding vital for maintaining ecological connectivity across Australia’s largest river basin. Landscape Ecology 30:2045–2065. Bishop-Taylor R, Tulbure MG, Broich M. 2018. Impact of hydroclimatic variability on regional- scale landscape connectivity across a dynamic dryland region. Ecological Indicators 94:142–150. Blazquez-Cabrera S, Gastón A, Beier P, Garrote G, Simón MÁ, Saura S. 2016. Influence of separating home range and dispersal movements on characterizing corridors and effective distances. Landscape Ecology 31:2355–2366. Boulanger Y, Arseneault D, Boucher Y, Gauthier S, Cyr D, Taylor AR, Price DT, Dupuis S. 2019. Climate change will affect the ability of forest management to reduce gaps between current and presettlement forest composition in southeastern Canada. Landscape Ecology 34:159–174. Calabrese JM, Fagan WF. 2004. A comparison-shopper’s guide to connectivity metrics. Frontiers in Ecology and the Environment 2:529-536. Correa Ayram CA, Mendoza ME, Etter A, Salicrup DRP. 2016. Habitat connectivity in biodiversity conservation. Progress in Physical Geography: Earth and Environment 40:7– 37.

Costanza JK, Moody A. 2011. Deciding where to burn: stakeholder priorities for prescribed burning of a fire-dependent ecosystem. Ecology and Society 16:14.

129

Costanza JK, Terando AJ. 2019. Landscape Connectivity Planning for Adaptation to Future Climate and Land-Use Change. Current Landscape Ecology Reports 4:1–13. Costanza JK, Terando AJ, McKerrow AJ, Collazo JA. 2015. Modeling climate change, urbanization, and fire effects on Pinus palustris ecosystems of the southeastern U.S. Journal of Environmental Management 151:186–199. Costanza JK, Watling J, Sutherland R, Belyea C, Dilkina B, Cayton H, Bucklin D, Romañach SS, Haddad NM. 2020. Preserving connectivity under climate and land-use change: No one- size-fits-all approach for focal species in similar habitats. Biological Conservation 248:108678.

Csardi G, Nepusz T. 2006. The igraph software package for complex network research. InterJournal Complex Systems 1695. Clark JS, Andrus R, Aubry-Kientz M et al. 2021. Continent-wide tree fecundity driven by indirect climate effects. Nature Communications 12:1242. de Chazal J, Rounsevell MDA. 2009. Land-use and climate change within assessments of biodiversity change: A review. Global Environmental Change 19:306–315. De Vos JM, Joppa LN, Gittleman JL, Stephens PR, Pimm SL. 2015. Estimating the normal background rate of species extinction. Conservation Biology 29:452–462. Dickson BG, Albano CM, McRae BH, Anderson JJ, Theobald DM, Zachmann LJ, Sisk TD, Dombeck MP. 2017. Informing Strategic Efforts to Expand and Connect Protected Areas Using a Model of Ecological Flow, with Application to the Western United States. Conservation Letters 10:564–571. Dilts TE, Weisberg PJ, Leitner P, Matocq MD, Inman RD, Nussear KE, Esque TC. 2016. Multiscale connectivity and graph theory highlight critical areas for conservation under climate change. Ecological Applications 26:1223–1237. Elsen PR, Monahan WB, Dougherty ER, Merenlender AM. 2020. Keeping pace with climate change in global terrestrial protected areas. Available from http://advances.sciencemag.org/.

EPA Updates to the Demographic and Spatial Allocation Models to Produce Integrated Climate and Land Use Scenarios (ICLUS) Version 2, EPA/600/R-16/366F (National Center for Environmental Assessment, 2016); http://www.epa.gov/ncea Fahrig L. 2003. Effects of Habitat Fragmentation on Biodiversity. Annual Review of Ecology, Evolution, and Systematics 34:487–515. Fahrig L. 2007. Non-optimal animal movement in human-altered landscapes. Functional Ecology 21:1003-1015. Fischer J, Lindenmayer DB. 2007. Landscape modification and habitat fragmentation: a synthesis. Global Ecology and Biogeography 16:265–280. Goodwin BJ, Fahrig L. 2002. How does landscape structure influence landscape connectivity? Oikos 99:552–570.

130

Gray M. 2008. Geodiversity: developing the paradigm. Proceedings of the Geologists’ Association 119:287–298. Greenwald KR, Purrenhage JL, Savage WK. 2009. Landcover predicts isolation in Ambystoma salamanders across region and species. Biological Conservation 142:2493–2500.

Grey GME. 2018. Ensemble creation of downscaled climate projections in the Southeast United States. North Carolina State University. [Unpublished master’s thesis]. Hanski I. 1999. Habitat Connectivity, Habitat Continuity, and Metapopulations in Dynamic Landscapes. Oikos 87:209. Heller NE, Zavaleta ES. 2009. Biodiversity management in the face of climate change: A review of 22 years of recommendations. Biological Conservation 142:14–32. Herse MR, With KA, Boyle WA. 2018. The importance of core habitat for a threatened species in changing landscapes. Journal of Applied Ecology 55:2241–2252. Heuberger KA, Putz FE. 2003. Fire in the Suburbs: Ecological Impacts of Prescribed Fire in Small Remnants of Longleaf Pine (Pinus palustris) Sandhill. Restoration Ecology 11:72–81.

Hijmans RJ, van Etten J. 2012. raster: Geographic analysis and modeling with raster data . R package version 3.4-5. http://CRAN.R-project.org/package=raster Hobbs RJ et al. 2014. Managing the whole landscape: historical, hybrid, and novel ecosystems. Frontiers in Ecology and the Environment 12:557–564. Kupfer JA, Terando AJ, Gao P, Teske C, Hiers JK. 2020. Climate change projected to reduce prescribed burning opportunities in the south-eastern United States. International Journal of Wildland Fire 29:764. Lawler JJ, Ackerly DD, Albano CM, Anderson MG, Dobrowski SZ, Gill JL, Heller NE, Pressey RL, Sanderson EW, Weiss SB. 2015. The theory behind, and the challenges of, conserving nature’s stage in a time of rapid change. Conservation Biology 29:618–629. Le Saout S et al. 2013. Protected areas and effective biodiversity conservation. Science 342:803- 805. Lucash MS, Scheller RM, Sturtevant BR, Gustafson EJ, Kretchun AM, Foster JR. 2018. More than the sum of its parts: how disturbance interactions shape forest dynamics under climate change. Ecosphere 9:e02293. Mantyka-Pringle CS, Visconti P, Di Marco M, Martin TG, Rondinini C, Rhodes JR. 2015. Climate change modifies risk of global biodiversity loss due to land-cover change. Biological Conservation 187:103–111. Mawdsley JR, O’Malley R, Ojima DS. 2009. A review of climate-change adaptation strategies for wildlife management and biodiversity conservation. Conservation Biology 23:1080- 1089. Maxwell CJ, Serra‐Diaz JM, Scheller RM, Thompson JR. 2020. Co‐designed management scenarios shape the responses of seasonally dry forests to changing climate and fire regimes.

131

Journal of Applied Ecology 57:1328–1340. McIntyre NE, Collins SD, Heintzman LJ, Starr SM, van Gestel N. 2018. The challenge of assaying landscape connectivity in a changing world: A 27-year case study in the southern Great Plains (USA) playa network. Ecological Indicators 91:607–616. McRae BH, Dickson BG, Keitt TH, Shah VB. 2008. Using circuit theory to model connectivity in ecology, evolution, and conservation. Ecology 89:2712–2724. Metzger MJ, Rounsevell MDA, Acosta-Michlik L, Leemans R, Schröter D. 2006. The vulnerability of ecosystem services to land use change. Agriculture, Ecosystems and Environment 114:69-85. Mitchell RJ, Liu Y, O’Brien JJ, Elliott KJ, Starr G, Miniat CF, Hiers JK. 2014. Future climate and fire interactions in the southeastern region of the United States. Forest Ecology and Management 327:316–326.

Mozelewski TG, Scheller RM. 2021. Forecasting for intended consequences. Conservation Science and Practice 3:e370. Nitschke CR, Trouvé R, Lumsden LF, Bennett LT, Fedrigo M, Robinson AP, Baker PJ. 2020. Spatial and temporal dynamics of habitat availability and stability for a critically endangered arboreal marsupial: implications for conservation planning in a fire-prone landscape. Landscape Ecology 35:1553–1570.

North Carolina State Climate Office. NC Climate Synopsis. https://climate.ncsu.edu/climate/synopsis. Nuñez TA, Lawler JJ, McRae BH, Pierce DJ, Krosby MB, Kavanagh DM, Singleton PH, Tewksbury JJ. 2013. Connectivity Planning to Address Climate Change. Conservation Biology 27:407–416. Oliver TH, Morecroft MD. 2014. Interactions between climate change and land use change on biodiversity: attribution problems, risks, and opportunities. Wiley Interdisciplinary Reviews: Climate Change 5:317–335. Parmesan C. 2006. Ecological and Evolutionary Responses to Recent Climate Change. Annual Review of Ecology, Evolution, and Systematics 37:637–669. Parmesan C, Yohe G. 2003. A globally coherent fingerprint of climate change impacts across natural systems. Nature 421:37-43. Perring MP, Standish RJ, Price JN, Craig MD, Erickson TE, Ruthrof KX, Whiteley AS, Valentine LE, Hobbs RJ. 2015. Advances in restoration ecology: Rising to the challenges of the coming decades. Ecosphere 6:1-25. Pinto N, Keitt TH. 2009. Beyond the least-cost path: Evaluating corridor redundancy using a graph-theoretic approach. Landscape Ecology 24:253–266. Piquer-Rodríguez M, Torella S, Gavier-Pizarro G, Volante J, Somma D, Ginzburg R, Kuemmerle T. 2015. Effects of past and future land conversions on forest connectivity in

132

the Argentine Chaco. Landscape Ecology 30:817–833.

Prasad AM, Iverson LR, Matthews S, Peters M. 2007. A Climate Change Atlas for 134 Forest Tree Species of the Eastern United States [database]. Northern Research Station, USDA Forest Service, Delaware, OH, USA. Available at: http://www.nrs.fs.fed.us/atlas/tree.

R Core Team. R: A Language and Environment for Statistical Computing. (R Foundation for Statistical Computing, 2017). Riordan EC, Rundel PW. 2014. Land Use Compounds Habitat Losses under Projected Climate Change in a Threatened California Ecosystem. PLoS ONE 9:e86487. Samuelson LJ, Stokes TA, Johnsen KH. 2012. Ecophysiological comparison of 50-year-old longleaf pine, slash pine and loblolly pine. Forest Ecology and Management 274:108–115. Santini L, Saura S, Rondinini C. 2016. Connectivity of the global network of protected areas. Diversity and Distributions 22:199–211. Saura S, Estreguil C, Mouton C, Rodríguez-Freire M. 2011. Network analysis to assess landscape connectivity trends: Application to European forests (1990-2000). Ecological Indicators 11:407–416. Saura S, Rubio L. 2010. A common currency for the different ways in which patches and links can contribute to habitat availability and connectivity in the landscape. Ecography 33:523- 537. Saura S, Torne J. 2009. Conefor Sensinode 2.2: a software package for quantifying the importance of habitat patches for landscape connectivity. Environment Modeling & Software 24:135-139. SCBD. In COP 10 Decision X/2: Strategic Plan for Biodiversity 2011–2020 (Secretariat of the Convention on Biological Diversity, 2010). Scheller R, Mladenoff D. 2008. Simulated effects of climate change, fragmentation, and interspecific competition on tree species migration in northern Wisconsin, USA. Climate Research 36:191–202. Scheller RM, Domingo JB, Sturtevant BR, Williams JS, Rudy A, Gustafson EJ, Mladenoff DJ. 2007. Design, development, and application of LANDIS-II, a spatial landscape simulation model with flexible temporal and spatial resolution. Ecological Modelling 201:409–419. Scheller RM, Hua D, Bolstad P V., Birdsey RA, Mladenoff DJ. 2011. The effects of forest harvest intensity in combination with wind disturbance on carbon dynamics in Lake States Mesic Forests. Ecological Modelling 222:144–153.

Schrum P, Scheller RM, Duveneck MJ, Lucash MS. 2020. Base-Hurricane: A new extension for the Landis-II forest landscape model. Environmental Modelling and Software 133:104833.

133

Soil Survey Staff, Natural Resources Conservation Service, United States Department of Agriculture. Soil Survey Geographic (SSURGO) Database. Available online at https://sdmdataaccess.sc.egov.usda.gov. Stevenson-Holt CD, Watts K, Bellamy CC, Nevin OT, Ramsey AD. 2014. Defining Landscape Resistance Values in Least-Cost Connectivity Models for the Invasive Grey Squirrel: A Comparison of Approaches Using Expert-Opinion and Habitat Suitability Modelling. PLoS ONE 9:e112119. Sturtevant BR, Cooke BJ, Kneeshaw DD, Mac Lean DA. 2015. Modeling insect disturbance across forested landscapes: Insights from the spruce budworm. In: Perera A, Sturtevant B, Buse L. (eds) Simulation Modeling of Forest Landscape Disturbances. Springer, Cham. Taylor PD, Fahrig L, Henein K. 1993. Connectivity Is a Vital Element of Landscape Structure. Oikos 68:57-573. Terando AJ, Costanza J, Belyea C, Dunn RR, McKerrow A, Collazo JA. 2014. The Southern Megalopolis: Using the Past to Predict the Future of Urban Sprawl in the Southeast U.S. PLoS ONE 9:e102261. Tulbure MG, Kininmonth S, Broich M. 2014. Spatiotemporal dynamics of surface water networks across a global biodiversity hotspot - Implications for conservation. Environmental Research Letters 9:114012.

U.S. Geological Survey (USGS) Gap Analysis Project (GAP). 2018. Protected Areas Database of the United States (PAD-US): U.S. Geological Survey data release, https://doi.org/10.5066/P955KPLE.

U.S. Geological Survey (USGS) Gridded Observed Meteorological Data: 1949-2010. https://catalog.data.gov/dataset/gridded-observed-meteorological-data-1949-2010 van Etten J. 2017. R package gdistance: Distances and routes on geographical grids. Journal of Statistical Software 76:1–21. Van Houtven G, Phelan J, Clark C, Sabo RD, Buckley J, Thomas RQ, Horn K, LeDuc SD. 2019. Nitrogen deposition and climate change effects on tree species composition and ecosystem services for a forest cohort. Ecological Monographs 89:e01345. Walther GR, Post E, Convey P, Menzel A, Parmesank C, Beebee TJC, Fromentin JM, Hoegh- Guldberg O, Bairlein F. 2002. Ecological responses to recent climate change. Nature 416:389-395. Wimberly MC. 2006. Species dynamics in disturbed landscapes: When does a shifting habitat mosaic enhance connectivity? Landscape Ecology 21:35–46. Zeller KA, McGarigal K, Whiteley AR. 2012. Estimating landscape resistance to movement: A review. Landsacpe Ecology 27:777-797. Zetterberg A, Mörtberg UM, Balfors B. 2010. Making graph theory operational for landscape ecological assessments, planning, and design. Landscape and Urban Planning 95:181–191.

134

APPENDICES

135

Appendix A

Initial communities We initialized LANDIS-II with current forest conditions including tree species-age cohorts assigned to every forested cell using both US Forest Service’s Forest Inventory and Analysis (FIA) data (USFS) and the Wilson et al. (2013) maps of live tree species basal area for the eastern United States (Figure A.1). This combined approach was necessary because FIA data, while highly detailed, is not contiguous and publicly accessible FIA data is fuzzed which can provide false spatial relations between trees. Meanwhile, imputation data can estimate the amount of each tree species expected at each location but lacks the detail required in the initial communities map, including tree species ages and biomass. We generated our Initial Communities map by imputing FIA data across ecologically suitable areas through environmental niche modeling. To create a contiguous spatial distribution of our focal species and their relative densities, we cropped the Wilson et al. (2013) raster maps of our individual species down to our study extent and extracted the tree species total basal area for each cell on the landscape. Because LANDIS-II requires age and biomass data to simulate initial forest conditions, we matched FIA plots to tree species compositions in each cell using Sorenson’s correlation coefficient and imputed them across the landscape. FIA establishes one plot per 2428 ha of forested land sampled every 5-10 years, so this matching process is necessary to generate a seamless map of initial forest conditions across the landscape (Bechtold and Patterson 2005). We used the allometric site index curves from Carmean, Hahn, and Jacobs (1989) to estimate the age of each tree. We then binned trees into species-age cohorts at each cell. Each cell on the landscape corresponds to an FIA plot with associated ages and biomass for each tree species-age cohort, making up initial forest conditions. For additional information on the creation of the initial communities map including data sources, similarity indices tested, and biomass and species calibration, please visit: https://github.com/LANDIS-II-Foundation/Project-NC- Conservation-Global- Change/tree/master/Parameterization%20and%20calibration/Initial%20communities

136

Figure A.1. Difference in biomass between LANDIS-II initial communities map and Wilson et al. (2013) maps of live tree species basal area for the study landscape.

Species included + parameters We identified the top 20 occurring tree species in the FIA plots in North Carolina, South Carolina, and Virginia and chose to include in our model species that represented greater than two percent of forest biomass in these plots (Figure A.2). After consulting with local experts on our included species list, we added longleaf pine and turkey oak, which fell well below the 2% threshold but are crucial components of the Sandhills and Coastal Atlantic Plain ecosystems of North Carolina. This is a small but biologically distinct portion of the study extent and an area of significant conservation focus. Species represented in the model are: Pinus taeda (FIA code PITA), Liquidamber styraciflua (FIA code LIST2), Acer rubrum (FIA code ACRU), Liriodendron tulipifera (FIA code LITU), Pinus echinata (FIA code PIEC2), Quercus alba (FIA code QUAL), Cornus florida (FIA code COFL2), Oxydendrum arboretum (FIA code OXAR), Pinus virginiana (FIA code PIVI2), Pinus palustris (FIA code PIPA2), and Quercus laevis (FIA code QULA2). Parameters for each species were derived from the literature, silvics manuals, and the TRY database (Adler et al. 2014, Burns and Honkala 1990, De Jager et al. 2017, Green 2009, He et al. 2011, He et al. 2012, Iversen et al. 2017, Kattge et al. 2009, Kattge et al. 2020, Martin et al. 2015, Scheller et al. 2011a, White et al. 2000, Wirth et al. 2009, Wright et al. 2004, Wright et al. 2006). For additional information on the species included, their respective biomasses in the states considered, species considered by not included, and data sources for species parameters please visit: https://github.com/LANDIS-II-Foundation/Project-NC-Conservation-Global-

137

Change/tree/master/Parameterization%20and%20calibration/Species. To view the species input parameters used, including longevity, age of sexual maturity, fire and shade tolerances, and effective seed dispersal distances, please view species_TM.txt at https://github.com/LANDIS-II- Foundation/Project-NC-Conservation-Global-Change/tree/master/Model%20inputs

Figure A.2. Top 20 species by biomass in FIA plots in North Carolina, South Carolina, and Virginia.

Net Ecosystem Carbon and Nitrogen (NECN) Succession Extension The Net Ecosystem Carbon and Nitrogen (NECN) Succession extension simulates the life cycle of each species-age cohort as they grow, reproduce, age, and die (Scheller et al. 2011b). NECN simulates all components of the carbon cycle and tracks carbon pools from live and dead trees (including leaf, wood, fine and coarse root, coarse woody debris, litter, and surface residue) (e.g. Creutzberg et al. 2017). It also tracks pools of active, passive, and slow pools of soil organic matter as well as above ground biomass and leaf area index (Scheller et al. 2011a). NECN operates on a monthly time step and incorporates temperature, precipitation, and wind data from historic climate data or modeled future climate scenarios. Each species‐age cohort is limited by temperature, water, nitrogen, leaf area index, and growing space available to simulate tree cohort growth and competition and requires soil and climatic inputs. For any additional information related to NECN parameterization, calibration, and validation, please see the subsections below and visit: https://github.com/LANDIS-II-Foundation/Project-NC-Conservation-Global- Change/tree/master/Parameterization%20and%20calibration/NECN%20Succession

Species parameters and functional groups Species values including C:N ratios and lignin parameters were derived from existing LANDIS- II papers and supplemented with values from the TRY database (Kattge et al. 2020). Try data was queried by species and by trait and in the event of multiple returns, values were averaged. GGDmin, GDDmax, frost, leaf longevity, and max drought were taken from the original linkages manual (Post and Pastor, 2013) and existing LANDIS-II papers (De Jager et al. 2017, Martin et

138

al. 2015, Scheller et al. 2011b). Species data that could not be found using the above approaches were adapted from a qualitative assessment of range in comparison to known values for other species.

We chose four functional groups: A pines group (loblolly, shortleaf, and longleaf), a group solely for Virginia pine (because Virginia pine has a range that extends further north than the other pines and can tolerate much colder temperatures), a hardwoods group (white oak, sweet gum, red maple, yellow poplar, flowering dogwood, and sourwood), and a group solely for turkey oak (while white oak is a generalist, turkey oak only exists in the southern portion of white oak’s range and has temperature and precipitation limit differences).

To view NECN species and functional group parameters please view NECN_input_TM.txt at https://github.com/LANDIS-II-Foundation/Project-NC-Conservation-Global- Change/tree/master/Model%20inputs

Calibration and Validation To calibrate the NECN extension, we began with single-cell, single species simulations. We did this first for three dominant species on the landscape: loblolly pine, red maple, and white oak (Figure A.3 and A.4). Once those individual species were calibrated, we moved on to other single cell, single species calibrations for the rest of our 11 species. After these calibrations, we moved on to single cell, single species, multiple cohort calibrations for loblolly, red maple, and white oak. We then calibrated a single cell with all 11 species before moving on calibrating the entire landscape. To calibrate above ground biomass and validate species parametrization, we isolated the top 25% of FIA plots by above ground carbon per age, assuming this to represent plots with ideal growing conditions for each species. These were then plotted as box plots and compared against simulated above ground biomass of each species (Figure A.3). LAI was validated against established values for forests from He et al. (2012; Figure A.4). This was done for every modeled species. Initial mineral N was an estimate similar to the values found in other LANDIS-II projects (https://github.com/LANDIS-II-Foundation). Atmospheric N slope intercept was attained by looking up National Atmospheric Deposition Program nitrogen deposition data (http://nadp.slh.wisc.edu/) for the past 20 years and plotting it as a function of annual precipitation (from U.S. Climate data, https://www.usclimatedata.com/climate/north- carolina/united-states/3203) in excel and finding the slope and intercept of the trend line (Figure A.5). Denitrification rates and all of the soil organic matter decay rates were originally taken from the Lake Tahoe LANDIS-II project (https://github.com/LANDIS-II-Foundation/Project- Lake-Tahoe-2017) single cell calibration text file and then calibrated for North Carolina.

139

Figure A.3. Single cell, single species calibration of simulated above ground biomass for red maple (AcerRubr) and white oak (QuerAlba), validated against the top 25% of FIA plots by above ground carbon per age to represent ideal growing conditions.

Figure A.4. Calibration of leaf area index (LAI) for red maple and white oak single cells, validated against oak/gum/cypress forest values values found in He et al. 2012.

140

Figure A.5. National Atmospheric Deposition Program nitrogen deposition data for the past 20 years plotted as a function of annual precipitation from U.S. Climate Data to parameterize atmospheric nitrogen slope intercept in the model.

Climate Climate regions used in LANDIS-II are typically delineated by substantial changes in topography, mean temperature, or annual or seasonal precipitation across the study area. We decided on only one climate region after determining that the study area had only small differences in temperature and precipitation along with low to moderate topographical variation. We used climate projections from the Coupled Model Intercomparison Project Phase 5 downscaled using the Multivariate Adaptive Constructed Analogs (MACA) method (Abatzoglou 2013). We chose five global climate models (GCMs) for the Representative Concentration Pathways (RCP) 4.5 and 8.5 that bracketed the temperature and precipitation projections for the study area and acted as our model replicates: bcc-csm1, CNRM-CM5, Had-GEM2, IPSL- CM5A-LR, and Nor-ESM1-M (Grey 2018). The five GCMs acted as our model replicates for each conservation strategy. Climate data were downloaded as monthly precipitation, maximum and minimum temperature, and wind speed. For additional information on climate inputs, please visit: https://github.com/LANDIS-II-Foundation/Project-NC-Conservation-Global- Change/tree/master/Parameterization%20and%20calibration/Climate

Soils Soil carbon was estimated from West’s (2014) soil carbon estimates for the conterminous US and attributed to the different pools of soil organic matter (active, passive, and slow) using parameter guidelines from the CENTURY model (Parton 1996). Soil nitrogen was estimated from soil carbon using the same parameter guidelines from Century (Parton 2013). Other soils maps, including drainage, flood frequency, field capacity, wilting point, percent sand, percent clay, and soil depth were derived from the USGS SSURGO soils database (Soil Survey Staff). To calculate dead wood biomass, we isolated each FIA plot in the study area within the last survey cycle and interpolated the total carbon down dead and across the whole study area

141

(USFS). For addition soils information, please visit: https://github.com/LANDIS-II- Foundation/Project-NC-Conservation-Global- Change/tree/master/Parameterization%20and%20calibration/Soils

Biomass Harvest Management units and forest stands The Biomass Harvest extension selects and removes tree species biomass based on specific management prescriptions that specify the timing of harvest, the species to be harvested, and the amount of biomass removed (Gustafson et al. 2000). These management prescriptions allow the user to specify how a stand is harvested (e.g., thinning or clearcut), how much of each species- age cohort is harvested, and whether the stand is replanted after harvesting, among other prescription criteria. Maps of management areas and forest stands determine where harvesting occurs on the landscape. Stands are ranked based on user-defined characteristics (e.g., based on the economic values of the species-age cohorts in that stand) and selected for harvest according to their rank. The amount of the landscape harvested in each management area/ prescription combination for each time step is determined by the rotation period (Gustafson et al. 2000).

Every non-water cell on the landscape was grouped into management units and forest stands. These assignments were used to specify which forest stands received which management prescriptions. Management units were delineated using the land ownership types described in Forest Service Forest Ownership Types in the Conterminous US map (Hewes et al. 2014) (Figure A.6). Forest stands were delineated using an extensive network of major and minor roads and streams from the North Carolina Department of Transportation's NCRouteCharacteristics data layer (NCDoT) and the North Carolina DEQ Division of Water Resources Major Hydrography dataset (NCDEQ). We integrated the acquisition and restoration of land over time into the model by assigning forest stands to be conserved management unit values unique to each restoration time step. These management areas received the same harvest treatments as the rest of the landscape up until the model time step in which they were to be restored, upon which they switched over to the restoration prescriptions for the rest of the model run. Stand size was validated using North Carolina private land and forest holdings data from the 2018 National Woodland Owner Survey (Butler et al. 2020) (Table A.1). Already conserved lands, identified using the Protected Areas database from the US Geological Survey, were not considered for land acquisition. Areas identified as open water or as high, medium, or low density residential in the NLCD (2016) were also removed from consideration and these land uses were assigned to their respective cells on the landscape.

142

Figure A.6. Management units without conservation strategies added to landscape.

143

Table A.1. Comparison of our forest stand sizes to private land owner stand data collected by North Carolina State University’s Forestry Extension specialists. NC National Woodland Owner Our stands (acres) Survey (acres) <40 0 1-20 11% 40-55 27% 20-49 18% 56-100 38% 50-99 16% 100-199 17% 100-199 17% 200-499 12% 200-499 17% 500-999 2% 500-999 11% >=1000 0.5% >=1000 9%

Management prescriptions We decided on three harvest prescriptions: (1) loblolly pine clearcutting for family and private industrial forests, (2) hardwood removal to simulate prescribed burning in the longleaf pine forests of Ft. Bragg and other private (conservation and natural resource agency) lands, and (3) a prescription to simulate mixed forest thinning that occasionally happens on private land (Table 3). We also simulated three restoration prescriptions on land acquired for conservation: longleaf restoration, pine mix restoration, and hardwood mix restoration (Table A.2). For the loblolly clearcut, longleaf thin/burn, longleaf restoration, and hardwood restoration prescriptions, stands were ranked on an index of economic value. With economic ranking, stands that most closely meet the economic criteria specified by the user are the first to receive that management prescription. In the loblolly clearcut prescription, stands with loblolly pine of at least 25 years old were assigned the highest economic value and targeted for harvest. In the longleaf thin/burn and longleaf restoration prescriptions, stands with longleaf pine of any age were prioritized for management. In the hardwood restoration prescription, stands with red maple and white oak 20 years or older were ranked highest for restoration. The mixed forest thinning and pine mix restoration prescriptions selected stands at random for management that had any combination of species besides longleaf pine and any combination of pine species besides loblolly pine, respectively. Harvest prescriptions were applied across the entire landscape, including stands identified for restoration only leading up to the time step in which restoration was enacted. Restoration prescriptions were only applied to stands identified for restoration. The number of hectares that each management prescription was applied to at each time step was validated against eVALIDator data (USFS), with expert opinion, and through speaking to land managers actively working in the study area. The Biomass Harvest extension operated at a 3-year time step. For additional information on the development of harvest prescriptions please visit: https://github.com/LANDIS-II-Foundation/Project-NC-Conservation-Global- Change/tree/master/Parameterization%20and%20calibration/Harvest

144

Table A.2 Detailed harvest and restoration prescriptions.

Prescription Actions taken Rank Targets per 3 yrs Economic- Pinus 4% of private non- Complete clearcut of Pinus taeda pine taeda ≥ least 25 corporate (family) Loblolly plantation stands followed by replanting yo (rank 100) land, 10% of private clearcut of Pinus taeda. corporate land Removal of 100% Acer rubrum, Economic- Pinus 18% of state land, Liquidamber styraciflua, and palustris any age 2% of private non- Liriodendron tulipifera ages 1-40 and (rank 100) corporate land, 20% 90% ages ≥ 41. Removal of 20% Quercus of federal land, and laevis ages 1-10 and Pinus palustris ages 12% of conservation 1-3 and 30% of Quercus laevis ages ≥ 11. lands Removal of 100% of Pinus taeda ages 1- 80. Removal of 50% Quercus alba ages 1-10 and 90% ages ≥ 11. Removal of 25% Cornus florida ages 1-10 and 70% ages ≥ 11. Removal of 90% Oxydenrum arboretum all ages. Removal of 80% of Longleaf Pinus virginiana ages 1-10 and 40% ages thin/burn ≥ 11. Mixed forest Remove 50% of all cohorts ≥ 15 years Random 9% of private non- thinning old in these stands. corporate land 100% removal of Acer rubrum, Economic- Pinus 50% of conservation Liquidamber styraciflua, Liriodendron palustris any age area hectares tulipifera, and Pinus taeda. 95% removal (rank 100) of Quercus alba, Oxydenrum arboretum, and Cornus florida. 40% removal of Pinus echinata, 30% removal of Quercus Longleaf laevis, and 25% removal of Pinus restoration virginiana. Replant with longleaf pine. 100% removal ofAcer rubrum, Random 25% of conservation Liquidamber styraciflua, Liriodendron area hectares tulipifera, Cornus florida, and Oxydendron arboretum. 80% removal of Quercus laevis and Quercus alba. 50% Pine mix removal of Pinus taeda. Replant with restoration Pinus virginiana, echinata, and palustris. Economic- Acer 25% of conservation 100% removal of Pinus taeda, 50% rubrum (rank 80) area hectares removal of all other cohorts (except Pinus and Quercus alba Hardwood virginiana- 40% removal), replant with (rank 75) ≥ 20 restoration Quercus alba. yo

145

Base Hurricane The Base Hurricane extension simulates tropical cyclones of varying severities making landfall in North Carolina (Schrum et al. 2020). The extension generates tropical cyclones that may or may not hit the study area using storm occurrence probabilities. For each hurricane, initiation parameters of landfall latitude, maximum wind speed at landfall, and storm track are created. These parameters are used to generate a maximum wind speed field on a continental grid that then determines maximum wind speed in each cell on the study area. The extension estimates the maximum wind speed parameter field from a wind speed equation according to the distance between each site and the hurricane path. The damage done by a storm is a product of cohort mortality probabilities estimated from empirical data about species and ages and these maximum wind speeds (Schrum et al. 2020).

We simulated hurricanes using the Base Hurricane extension, parameterized for landfall in North Carolina by from Schrum et al. (2020), who parameterized the Base Hurricane extension for the Southeastern United States. Schrum et al. (2020) obtained hurricane data for storms occurring on the coast of Georgia, South Carolina, North Carolina, and Virginia from 1969 to 2018 to derive estimates of storm frequency, wind field size, maximum sustained wind speed, and storm size. Simulated cohort mortalities vary based on those maximum sustained wind speeds. For additional information on hurricane simulation, please visit: https://github.com/LANDIS-II- Foundation/Project-NC-Conservation-Global- Change/tree/master/Parameterization%20and%20calibration/Hurricanes. Hurricane input parameters can be viewed in BaseHurricane.txt at https://github.com/LANDIS-II- Foundation/Project-NC-Conservation-Global-Change/tree/master/Model%20inputs

Output Extensions

To summarize model outputs, we used the Biomass Output extension to track individual species biomass and total site biomass for each cell on the landscape over time. We used the Cohort Statistics Output extension to calculate median stand and species ages for every cell on the landscape. These metrics were used to create estimates of habitat quality for every forested cell on the landscape.

146

References:

Abatzoglou JT. 2013. Development of gridded surface meteorological data for ecological applications and modelling. International Journal of Climatology 33:121-131. Adler PB, Salguero-Gómez R, Compagnoni A, Hsu JS, Ray-Mukherjee J, Mbeau-Ache C, Franco M. 2014. Functional traits explain variation in plant life history strategies. PNAS 111:740-745. doi: 10.1073/pnas.1315179111 Bechtold WA, Patterson PL. 2005. The enhanced forest inventory and analysis program: national sampling design and estimation procedures. USDA Forest Service Southern Research Station General Technical Report SRS‐80. Asheville, NC. Burns RM, Honkala BH. 1990. Silvics of North America: 1. Conifers. Agriculture Handbook 654, U.S. Department of Agriculture, Forest Service, Washington, DC. Butler BJ, Butler SM, Caputo J, Dias J, Robillard A, Sass EM. 2020. Family Forest Ownerships of the United States, 2018: Results from the USDA Forest Service, National Woodland Owner Survey. Gen. Tech. Rep. NRS-199. Madison, WI: U.S. Department of Agriculture, Forest Service, Northern Research Station. 56 p. https://doi.org/10.2737/NRS-GTR-199 Carmean WH, Hahn JT, Jacobs RD. 1989. Site index curves for forest tree species in the eastern United States. General Technical Report NC-128. St. Paul, MN: US Dept. of Agriculture, Forest Service, North Central Forest Experiment Station, 128. Creutzburg MK, Scheller RM, Lucash MS, LeDuc SD, Johnson MG. 2017. Forest management scenarios in a changing climate: trade‐offs between carbon, timber, and old forest. Ecological Applications 27:503–518. De Jager NR, Drohan PJ, Miranda BM, Sturtevant BR, Stout SL, Royo AA, Gustafson EJ, Romanski MC. Simulating ungulate herbivory across forest landscapes: A browsing extension for LANDIS-II. Ecological Modelling 350:11-29. Dewitz J. 2019. National Land Cover Database (NLCD) 2016 Products: U.S. Geological Survey data release, https://doi.org/10.5066/P96HHBIE. Green W. 2009. USDA PLANTS Compilation, version 1, 09-02-02. (http://bricol.net/downloads/data/PLANTSdatabase/) NRCS: The PLANTS Database (http://plants.usda.gov, 1 Feb 2009). National Plant Data Center: Baton Rouge, LA 70874- 74490 USA. He L, Chen JM, Pan Y, Birdsey R, Kattge J. 2012. Relationships between net primary productivity and forest stand age in US forests. Global Biogeochemical Cycles, 26(3). He T, Lamont BB, Downs KS. 2011. Banksias born to burn. New Phytologist 191:184-196. doi:10.1111/j.1469-8137.2011.03663.x. He T, Pausas JP, Belcher CM, Schwilk DW, Lamont BB. 2012. Fire-adapted traits of Pinus arose in the fiery Cretaceous. New Phytologist 194: 751-759. doi: 10.1111/j.1469- 8137.2012.04079.x. Hewes JH, Butler BJ, Liknes GC, Nelson MD, Snyder SA. 2014. Map of distribution of six forest ownership types in the conterminous United States Res. Map nrs-6, U.S. Department of Agriculture, Forest Service, Northern Research Station, Newtown Square, PA. Iversen CM, McCormack ML, Powell AS, Blackwood CB, Freschet GT, Kattge J, Roumet C, Stover DB, Soudzilovskaia NA, Valverde Barrantes OJ, Bodegom PM, Violle C. 2017. A global fine root ecology database to address below ground challenges in plant ecology. New Phytologist 215:15-26. doi:10.1111/nph.14486 Grey GME. 2018. Ensemble creation of downscaled climate projections in the Southeast United States. North Carolina State University. [Unpublished master’s thesis] 147

Gustafson EJ, Shifley SR, Mladenoff DJ, Nimerfro KK, He HS. 2000. Spatial simulation of forest succession and timber harvesting using LANDIS. Canadian Journal of Forest Research 30:32-43. Kattge J, Knorr W, Raddatz T, and Wirth C. 2009. Quantifying photosynthetic capacity and its relationship to leaf nitrogen content for global-scale terrestrial biosphere models. Global Change Biology 15:976-991. Kattge J, Bönisch G, Díaz S, Lavorel S, Prentice IC, Leadley P, et al. 2020. TRY plant trait database – enhanced coverage and open access. Global Change Biology 26:119-188. https://doi.org/10.1111/gcb.14904 Martin KL, Hurteau MD, Hungate BA, Koch GW, North MP. 2014. Carbon Tradeoffs of Restoration and Provision of Endangered Species Habitat in a Fire-Maintained Forest. Ecosystems 18:76–88. North Carolina Department of Environmental Quality. MajorHydro dataset. https://data- ncdenr.opendata.arcgis.com/datasets/2a5324f732bb44d9a1ca29f1c17370ee_0 North Carolina Department of Transportation. NC Route Characteristics dataset. https://connect.ncdot.gov/resources/gis/Pages/GIS-Data-Layers.aspx Parton WJ. 1996. The CENTURY model. Pages 283–291 Evaluation of Soil Organic Matter Models. Springer Berlin Heidelberg. Parton, W. 2013. CENTURY Soil Organic Matter Model Environment. Technical Documentation. Agroecosystem Version 3.0. USDA-ARS, Forest Collins, CO. West, T.O. 2014. Soil Carbon Estimates in 20-cm Layers to 1-m Depth for the Conterminous US, 1970- 1993. Data set. Available on-line [http://daac.ornl.gov] from Oak Ridge National Laboratory Distributed Active Archive Center, Oak Ridge, Tennessee, USA. http://dx.doi.org/10.3334/ORNLDAAC/1238 Soil Survey Staff, Natural Resources Conservation Service, United States Department of Agriculture. Soil Survey Geographic (SSURGO) Database for NC. Post WM, Pastor J. 2013. LINKAGES: An Individual-based Forest Ecosystem Biogeochemistry Model. ORNL DAAC, Oak Ridge, Tennessee, USA. Scheller RM, Domingo JB, Sturtevant BR, Williams JS, Rudy A, Gustafson EJ, Mladenoff DJ. 2007. Design, development, and application of LANDIS-II, a spatial landscape simulation model with flexible temporal and spatial resolution. Ecological Modelling 201:409–419. Scheller RM, Hua D, Bolstad P V., Birdsey RA, Mladenoff DJ. 2011a. The effects of forest harvest intensity in combination with wind disturbance on carbon dynamics in Lake States Mesic Forests. Ecological Modelling 222:144–153. Scheller RM, van Tuyl S, Clark KL, Hom J, La Puma I. 2011b. Carbon Sequestration in the New Jersey Pine Barrens Under Different Scenarios of Fire Management. Ecosystems 14:987– 1004. Schrum P, Scheller RM, Duveneck MJ, Lucash MS. 2020. Base-Hurricane: A new extension for the Landis-II forest landscape model. Environmental Modelling and Software 133:104833. Soil Survey Staff, Natural Resources Conservation Service, United States Department of Agriculture. Soil Survey Geographic (SSURGO) Database. Available online at https://sdmdataaccess.sc.egov.usda.gov. USDA Forest Service. 2015. Forest inventory and analysis database: Database description and user guide for phase 2. U.S. Department of Agriculture, Forest Service, Northeastern Research Station, Forest Inventory and Analysis Unit, Newtown Square, PA. 758 p. USDA Forest Service. EVALIDator User Guide (Version 4.01). U.S. Department of Agriculture,

148

Forest Service. https://www.fia.fs.fed.us/tools- data/tutorials_training/docs/EVALIDator_user_guide.pdf U.S. Geological Survey (USGS) Gap Analysis Project (GAP). 2018. Protected Areas Database of the United States (PAD-US): U.S. Geological Survey data release, https://doi.org/10.5066/P955KPLE. White MA, Thornton PE, S. Running W, Nemani RR. 2000. Parameterization and sensitivity analysis of the BIOME-BGC terrestrial ecosystem model: Net primary production controls. Earth Interactions 4:1-85. Wilson BT, Lister AJ, Riemann RI, Griffith DM. 2013. Live tree species basal area of the contiguous United States (2000-2009). Newtown Square, PA: USDA Forest Service, Rocky Mountain Research Station. https://doi.org/10.2737/RDS-2013-0013 West TO. 2014. Soil Carbon Estimates in 20-cm Layers to 1-m Depth for the Conterminous US, 1970-1993. Data set. Available on-line from Oak Ridge National Laboratory Distributed Active Archive Center, Oak Ridge, Tennessee, USA. http://dx.doi.org/10.3334/ORNLDAAC/1238 Wirth C, Lichstein JW. 2009. The Imprint of Species Turnover on Old-Growth Forest Carbon Balances - Insights from a Trait-Based Model of Forest Dynamics. Pages 81-113 in C. Wirth, G. Gleixner, and M. Heimann, editors. Old-Growth Forests: Function, Fate and Value. Springer, New York, Berlin, Heidelberg. Wright IJ, Reich PB, Westoby M, Ackerly DD, Baruch Z, Bongers F, Cavender-Bares J, et al. 2004. The worldwide leaf economics spectrum. Nature 428:821-827. Wright IJ, Reich PB, Atkin OK, Lusk CH, Tjoelker MG, Westoby M. 2006. Irradiance, temperature and rainfall influence leaf dark respiration in woody plants: evidence from comparisons across 20 sites. New Phytologist 169:309-319.

149

Appendix B

Table B.1. Resistance values used to characterize landscape resistance to movement for specialist and generalist species guilds.

Specialist Generalist Resistance Resistance Land cover type score Land cover type score Longleaf age 0:1 90 Longleaf age 0:5 90 Longleaf age 2:5 80 Longleaf age 6:10 50 Longleaf age 6:7 60 Longleaf age 11:20 20 Longleaf age 8:9 40 Longleaf age 21:34 1 Longleaf age 10:21 10 Longleaf age >= 35 1 Longleaf age 21:34 1 Longleaf age >= 35 1

Pine plantation age 0:5 90 Pine plantation age 0:5 90 Pine plantation age 6:10 70 Pine plantation age 6:10 70 Pine plantation age 11:20 60 Pine plantation age 11:20 50 Pine plantation age 21:30 50 Pine plantation age 21:30 40 Pine plantation age >= 31 40 Pine plantation age >= 31 30

Pine mix age 0:5 90 Pine mix age 0:5 90 Pine mix age 6:10 70 Pine mix age 6:10 50 Pine mix age 11:20 40 Pine mix age 11:20 20 Pine mix age 21:34 30 Pine mix age 21:34 1 Pine mix age >=35 20 Pine mix age >=35 1

Hardwood age 0:10 90 Hardwood age 0:10 90 Hardwood age 11:20 80 Hardwood age 11:20 80 Hardwood age 21:30 70 Hardwood age 21:30 70 Hardwood age >=31 60 Hardwood age >=31 60

Mixed forest age 0:10 90 Mixed forest age 0:10 90 Mixed forest age 11:20 70 Mixed forest age 11:20 70 Mixed forest age 21:30 60 Mixed forest age 21:30 50 Mixed forest age >= 31 50 Mixed forest age >= 31 40

Cropland 90 Cropland 90 Hay/pasture 90 Hay/pasture 90 Water 100 Water 100 Developed, high intensity 100 Developed, high intensity 100 Developed, med intensity 90 Developed, med intensity 90

150

Table B.1 (continued). Developed, low intensity 80 Developed, low intensity 80 Barren land 90 Barren land 90 Mining 100 Mining 100

LUC land cover types Cropland 90 Hay/pasture 90 Grazing 90 Water 100 Reservoirs 90 Wetlands 70 Urban, high intensity 100 Urban, low intensity 100 Suburban 90 Exurban, high intensity 80 Exurban, low intensity 70 Mining 100 Commercial 100 Industrial 100 Transportation 100

151

Appendix C

Weights for weighted random sort of forest stands for conservation

Economic

• Probability of selection from 0.0001 (high cost per hectare) to 0.96 (lowest cost per hectare) by 0.00015 • Remaining highest cost per acre stands assigned 0.00001

Geodiversity

Table C.1. Probability of selection weights used for forest stand selection under the geodiversity strategy. Geodiversity score Probability of selection >3000 0.95 <3000 & >=2500 0.925 <2500 & >=2000 0.9 <2000 & >= 1500 0.85 <1500 & >=1000 0.825 <1000 & >= 500 0.775 <500 & >= 0 0.65 <0 & >=-500 0.6 <-500 & >=-1000 0.3 <-1000 & >=-1500 0.1 <-1500 & >=-2000 0.05 <-2000 0.001

Cluster

Table C.2. Probability of selection weights used for forest stand selection under the cluster strategy. Buffer distance Probability of selection 100m 0.9 500m 0.7 1000m 0.5 2500m 0.3 >2500m 0.1

152

Appendix D

Figure D.1. Conceptual depiction of habitat node identification by grouping contiguous habitat cells that co-occurred in both time steps t and t-1 discretely from those that only occurred in time step t.

153