Download The

ACCOUNTING FOR LONG-TERM PERSISTENCE OF MULTIPLE SPECIES IN

SYSTEMATIC CONSERVATION PLANNING

Matthew Strimas-Mackey

BSc, University of Guelph, 2012 MSc, University of Toronto, 2006 BSc, University of Toronto, 2005

A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF

MASTER OF SCIENCE

THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES

(Zoology)

THE UNIVERSITY OF BRITISH COLUMBIA

(Vancouver)

July 2016

Protected areas form the cornerstone of global efforts to conserve biodiversity. The goal of systematic conservation planning is to design protected area networks that secure the long term persistence of biodiversity. However, most current methods focus on maximizing the representation of species and don’t explicitly plan for the persistence of those species in the protected landscapes into the future. In this thesis, I present a new tool for systematic reserve design that optimizes the configuration of reserve networks to maximize persistence across multiple species. This method is based on metapopulation capacity, a relative, asymptotic metric of persistence derived from a spatially explicit metapopulation model. This metric requires few parameters to calculate, and incorporates the size and spatial configuration of reserves as well as species-specific dispersal dynamics among them. I demonstrate this method using a case study in Indonesian New

Guinea with 114 terrestrial mammal species. Compared to Marxan, the most popular representation-based reserve design tool, my persistence-based method led to a 2.3- times increase in mean metapopulation capacity across all species. At the level of individual species, I identified two distinct groups: those that experienced significant benefits from the persistence-based approach and those for which the Marxan solution was nearly as good or slightly better. This thesis demonstrates that systematic reserve design can account for species persistence in an ecologically meaningful way, and that this approach can yield significant gains compared to traditional methods.

Preface

This initial idea for this research came from discussions with Dr. Jedediah Brodie. I was responsible for all background research, collection and processing of data, analysis, and tool development.

A version of Chapter 2 will be submitted for publication and co-authored with Dr. Jedediah

Brodie.

iii

Table of Contents

Abstract ...... ii

Preface ...... iii

Table of Contents ...... iv

List of Tables ...... vi

List of Figures ...... vii

Acknowledgements ...... ix

Chapter 1 Introduction ...... 1

1.1 Systematic conservation planning ...... 1

1.2 Metapopulation theory ...... 5

1.3 Planning for persistence ...... 6

1.4 Study area: Indonesian New Guinea ...... 9

Chapter 2 Accounting for long-term persistence of multiple species in systematic conservation planning ...... 11

2.1 Introduction ...... 11

2.2 Methods ...... 14

2.2.1 Metapopulation capacity ...... 14

2.2.2 Simulated annealing ...... 15

2.2.3 Mathematical formulation ...... 16

2.2.4 Case study ...... 17

2.3 Results ...... 22

2.4 Discussion ...... 28

Chapter 3 Conclusion ...... 33

References ...... 35

Appendices ...... 43

Appendix A Simulated annealing ...... 43

Appendix B Species list ...... 45

List of Tables

Table 2.1 Summary of reserve performance using three reserve selection methods: representation-based selection using Marxan, our method maximizing metapopulation capacity, and our method excluding patches less than 4 planning units in size. λ is the mean scaled metapopulation capacity, which ranges from 0 (no protection) to 1 (entire study area protected). Targets missed gives the number of species (out of 114 total) for which the 20% representation target was not met and, for these species, % gap gives the percent difference between the actual and targeted representation levels...... 23

List of Figures

Figure 2.1 Map of the study area in Indonesian New Guinea, showing 4,399 hexagonal

100 km2 planning units. (a) Cost of planning units, composed of current and 2050 potential oil palm yield (IIASA/FAO 2012), percent crop cover (Ramankutty et al. 2002), and population density (CIESIN/IFPRI/WB/CIAT 2011). Each of the four components are standardized to 0-1, then averaged. (b) Richness of 114 terrestrial mammal species used in this study based on IUCN Red List range polygons (IUCN 2015)...... 21

Figure 2.2 Candidate reserves resulting from the three reserve selection methods: (a) representation-based selection using Marxan, (b) our method maximizing metapopulation capacity, and (c) our method as in (b) but excluding patches less than 4 planning units in size. Green indicates selected planning units, light grey indicates unselected planning units, and Papua New Guinea (not part of the case study) is shown in dark grey. In (c), small patches were removed to reduce the noise that arose because computational limitations restricted the number of simulated annealing iterations...... 24

Figure 2.3 Distribution of the differences in species-specific scaled metapopulation capacity between candidate reserves from a Marxan analysis versus our persistence- based selection method. Positive values indicate an improvement in metapopulation capacity. Bars are coloured according to species’ IUCN Red List status...... 26

Figure 2.4 Relationship between change in scaled metapopulation capacity (between -1 and 1) and range size (from IUCN distribution polygons) within Indonesian New Guinea for

vii

each of the 114 mammal species in the case study. The y-axis shows the difference in scaled metapopulation capacity between candidate reserves from Marxan and our persistence-based selection method. Range restricted species see a greater improvement in metapopulation capacity compared to widespread species...... 27

viii

Acknowledgements

Thank you to my advisor Jedediah Brodie for helping me develop interesting questions and for his unwavering positive attitude. Thanks also to my committee, Mary O’Connor and

Sarah Gergel, for offering useful feedback and asking thoughtful questions. Thanks to my lab mates for their support and friendship during my time at UBC.

Throughout my many years in school, I’ve always received love and support from my amazing family. Thanks to my parents, Gina and Eric, and all my siblings: Josh, Sascha,

Seth, Rowen, Jonah, Ezra, and Rosie-Deer. Finally, thanks to my best friend and partner

Katie. Love you, boo.

Chapter 1

Introduction

1.1 Systematic conservation planning

Protected areas form the cornerstone of global efforts to conserve biodiversity in the current age of extinction (Pimm and Raven 2000, Pimm et al. 2014). Their function is to preserve intact ecosystems and ensure long term persistence of species by demarcating geographical regions that exclude anthropogenic threats to biodiversity features (Margules and Pressey 2000). The central role of protected areas in global conservation efforts is reflected in the protected area targets in the 2010 Convention on Biological Diversity

(CBD). The 193 signatories of the CBD agreed to increase the coverage of the global protected area network from the current 13% to 20% of the terrestrial realm (Convention on Biological Diversity 2011).

As protected area networks expand they must do so under social, political, and economic constraints. Chief among these constraints is a limit on the total amount of land that can be protected due to limited conservation resources and, in many regions, a declining amount of intact habitat left to protect (Hansen et al. 2013). As a result of these constraints, decision must be made about where to allocate conservation funds and which sites to prioritize for protection (Wilson et al. 2009). These decisions will have important consequences for the fate of biodiversity globally.

Historically, the locations of most protected areas were not chosen with the explicit goal of biodiversity conservation in mind. Rather sites were often chosen for cultural, scenic, or recreation value (Possingham et al. 2000). The locations of some of North

America’s oldest and most famous National Parks, such as Yosemite and Banff, may have been picked for this reason. In other cases, sites are chosen to protect a small number of charismatic focal species, with no assurance of the persistence of the ecosystem as a whole (Possingham et al. 2000). Furthermore, new protected areas are often located in areas of low economic value, resulting in a significant bias towards high elevation, infertile, and inaccessible land (Joppa and Pfaff 2009). As a consequence, many biodiversity features remain unprotected or poorly represented within the global protected area network (Rodrigues et al. 2004). For example, Venter et al. (2014) found that 17% of threatened vertebrate species do not occur within any protected area, and 85% are not afforded sufficient protection by existing protected areas to ensure their long-term persistence.

Systematic conservation planning is the process of using systematic tools and techniques to identify locations that efficiently meet conservation objectives while minimizing socioeconomic cost (Margules and Pressey 2000). This method can be applied to a variety of conservation actions, however, it is most commonly used to select locations for new protected areas, a process known as systematic reserve design. In this context, the fundamental objective is to identify the set of sites that if protected will ensure the long- term persistence of biodiversity in the most cost effective manner (Cabeza and Moilanen

2001). The resulting reserve network should comprehensively represent the full range of 2

biodiversity in the study system (Wilson et al. 2009). In practice, due to computational limitations and gaps in our knowledge of the spatial distribution of most biodiversity features, a subset of well-known biodiversity features is typically used as a surrogate for biodiversity as a whole (Wilson et al. 2009). These surrogates, often referred to as the conservation features, generally consist of species or habitat types for which spatial distribution maps exist or can be generated.

One of the principles of effective systematic reserve design is complementarity. To ensure that all biodiversity features are represented in the reserve network for the lowest possible cost, individual reserves in the network must complement each other (Wilson et al. 2009). A failure to consider complementarity is evident in the existing global protected area network where some geographic regions, biomes, and species are over represented, while many others are insufficiently represented (Jenkins and Joppa 2009, Venter et al.

2014). Complementarity precludes simple approaches to reserve selection based on scoring or ranking of individual sites, such as by species richness (Wilson et al. 2009). For example, a low diversity site that includes a species not already represented in the reserve network may be of higher priority than a high diversity site with species already captured within the existing reserve network. Complementarity requires that the full distribution of all conservation features across all sites must be considered explicitly. Furthermore, all sites that are candidates for protection must be considered simultaneously since the selection of any given site will influence the importance of all other sites.

The reserve design problem is inherently complex and requires specialized tools to solve. Many such tools exist, however, most address the problem in a similar fashion. The 3

study area is broken up into a discrete set of planning units (often a regular grid of squares or hexagons) and the cost of protection and conservation value are estimated for each planning unit. The cost of a planning unit is typically represented by the acquisition cost of the land, the costs of management, the opportunity cost of foregone commercial activities, or simply the area. Conservation value is a measure of the level of representation of a given conservation feature within a given planning unit. It is typically estimated by occupancy (i.e. presence or absence), some measure of abundance, or area of habitat within the planning unit (Beyer et al. 2016). The goal is then to identify the set of planning units that maximizes representation for a given fixed cost (known as the Maximum Coverage Problem), or to minimize cost while meeting a set of fixed representation targets (known as the Minimum

Set Cover Problem). This latter formulation is used by Marxan (Ball et al. 2009), the most widely used systematic reserve design software. Within this formulation the representation targets are area-based targets specifying the amount of habitat that should be protected for each conservation feature.

While the fundamental objective of systematic conservation planning is to ensure the long-term persistence of conservation features (Cabeza and Moilanen 2001), in the dominant reserve design paradigm just described persistence is rarely treated explicitly.

Instead, reserve design exercises typically focus on maximizing the representation of conservation features within reserves, with no guarantee that the spatial configuration of these reserve will ensure persistence (Cabeza and Moilanen 2003, Nicholson et al. 2006).

Some existing tools include simple approaches for reducing fragmentation or increasing connectivity with the goal of producing spatial configurations that are more conducive to 4

persistence. For example, Marxan can produce more compact reserve networks by minimizing the overall perimeter of the network. However, such metrics do not account for the spatial population dynamics of species or the spatial structure of the landscape, and should therefore be viewed more as rules of thumb than valid estimates of species persistence. Conversely, spatial decision support tools, such as CONEFOR (Saura and

Torné 2009), do explicitly address landscape connectivity, however, they do not account cost nor are they easily incorporated into a reserve design framework.

1.2 Metapopulation theory

A metapopulation is a set of spatially discrete populations of the same species that interact through dispersal (Akçakaya et al. 2007). Individual populations inhabit patches and patches are embedded in a matrix of non-habitat that permits the movement of species but will not support populations (Akçakaya et al. 2007). These are inherently dynamic systems characterized by repeated local extinction and re-colonization events (Hanski and

Gilpin 1991). Since it’s inception, metapopulation theory has proven to be a valuable framework for investigating population dynamics of spatially structured populations.

Habitat destruction is the main proximate driver of species extinctions (Pimm et al.

1995, Brook et al. 2008). In some cases, all existing habitat for a species may be destroyed resulting in immediate and direct extinction, however, the more common scenario is for habitat to become greatly reduced in extent and highly fragmented (Brook et al. 2008). The result is a landscape of isolated habitat patches embedded in an inhospitable matrix. This scenario meets the requirements of a metapopulation, therefore

metapopulation theory has direct applications for biodiversity conservation and management (Akçakaya et al. 2007). In particular, metapopulation models are widely used to predict how changes in the amount and configuration of habitat will affect the spatial population dynamics, and hence extinction risk, of species (Cabeza and Moilanen 2001).

In the context of reserve design, metrics derived from metapopulation models can be used to compare alternate configurations of reserve networks in terms of their ability to support viable populations of species (Margules and Pressey 2000, Cabeza and Moilanen

2001). However, the use of these metrics is often constrained to planning for single, well- studied species due to computational limitations and availability of species-specific data to parameterize the models (Nicholson and Possingham 2007).

1.3 Planning for persistence

Despite substantial interest in both metapopulation theory and systematic conservation planning, exceedingly few attempts have been made to combine them. Systematic reserve design is an inherently computationally intensive optimization problem because the space of possible solutions is extremely large; there are 2# possible solutions and the number of planning units, $, is usually on the order of tens of thousands. In order to find a near optimal reserve configuration, a large portion of the solution space must be sampled and, for each potential solution, the conservation benefit must be calculated. Therefore, any metric that is incorporated into the conservation benefit function must be efficient to calculate. This precludes the use of most of the more complex metapopulation models, such as those requiring Monte Carlo simulations (Nicholson et al. 2006). The use of

metapopulation models in systematic reserve design is also hindered by the need for species-specific parameters, such as home range size and dispersal distance, which are not available for the majority of species.

As a result of these challenges, systematic reserve design tools rarely explicitly plan for the persistence of conservation features, focusing instead on maximizing the area of representation of features and ignoring their spatial dynamics. The small number of studies that do successfully account for persistence using metapopulation models or metrics have done so in an overly simplified fashion and are of limited utility for real reserve design exercises. Moilanen and Cabeza (2002) used a spatially explicit metapopulation model to select from 125 sites the subset that minimized the extinction probability of a single species for a fixed cost. Using a stochastic patch occupancy model, Westphal et al. (2003) compared the effect of different conservation strategies (enlarging existing reserves, connecting reserves via corridors, or creating new reserves) on the extinction probability of a single species. In both studies, only a single species was considered and the set of potential solutions was small. This contrasts with most real systematic reserve design exercises, which consider many species and typically divide the study area into thousands of planning units.

An alternative approach has been to use metapopulation models to assign a conservation value to individual planning units and select the optimal set of planning units based on these static values. These methods only evaluate the metapopulation model once at the start of the reserve design exercise rather than building the model directly into the optimization process. This results in greatly improved efficiency, but such approaches 7

do not explicitly optimize the spatial configuration of reserves to maximize species persistence. For example, Nalle et al. (2004) used individual-based metapopulation simulations to score 28,000 planning units based on their importance for the persistence of two species. They used these scores as the conservation value metric, which was optimized under economic constraints (Nalle et al. 2004). However, since the assigned scores were static, they do not take into account the effect of spatial context on planning unit importance. In particular, a planning unit’s value for species persistence depends on the protection status of neighbouring units. In addition, Nalle et al. (2004) note that they are aware of only 20 species for which enough data exists to parameterize their models.

There is currently only one example of a systematic reserve design exercise that accounts for the persistence of multiple species by incorporating a process-based metapopulation model directly into the selection algorithm. Nicholson et al. (2006) considered a system of 10 species in a landscape of 39 forest fragments, selecting the optimal set of sites to protect to maximize persistence across all species for a fixed conservation budget. Their method showed a 20-fold improvement in estimated species persistence compared to a traditional area-based reserve design method (Nicholson et al.

2006). However, their approach was constrained to a very small set of planning units (39 forest fragments) and was based on a metapopulation model (Frank and Wissel 2002) that requires parameters such as home range size that are unknown for most species.

In this thesis, I develop a new reserve design method that explicitly maximizes regional persistence for a large assemblage of species. My approach is based on maximizing metapopulation capacity, a relative, asymptotic metric of persistence derived from a 8

spatially explicit metapopulation model (Hanski 1999, Hanski and Ovaskainen 2000). This metric requires few parameters to calculate, and incorporates the size and spatial configuration of reserves as well as species-specific dispersal dynamics among them. I demonstrate this method with a reserve planning exercise for 114 terrestrial mammal species in Indonesian New Guinea. Compared to a traditional representation-based reserve planning exercise performed in Marxan, this new approaches yields a significant increase in metapopulation persistence for only a modest reduction in representation.

1.4 Study area: Indonesian New Guinea

New Guinea is a tropical island located in the Southwestern Pacific, between Australia and the equator. At 785,753 km2, it is the world’s second largest island. Politically the island is divided into the eastern half, which comprises the country of Papua New Guinea, and the western half, which is part of Indonesia. In this thesis I will use Indonesian New Guinea as a case study to demonstrate my persistence-based reserve design method.

New Guinea is extremely diverse geophysically. The New Guinea Highlands form a central spine stretching across the length of the island from east to west. Many of the peaks in this range achieve elevations above 4,000m, and the highest peaks are permanently glaciated. Straddling the highlands to the north and south are vast, mostly flat lowlands composed of tropical rainforest, savannah, wetlands, and mangroves.

This geophysical diversity is paralleled by high biological diversity and endemicity

(Myers et al. 2000), with some estimates suggesting 10% of global diversity is found on the island. The fauna of New Guinea is closely allied with Australia, reflecting the fact that

these islands are both located on the Australian Plate and were connected by land during the last glaciation. For example, apart from bats and rats, all native New Guinean mammals are monotremes or marsupials. In contrast, the Southeast Asian islands to the west of New Guinea, such as Borneo, have a distinctly Asian fauna, with few species of

Australian origin. This pattern exists across many animal taxa, and the dividing line between the two faunas is famously known as Wallace’s Line.

The island of New Guinea houses the world’s third largest remaining area of intact tropical rainforest. However, the region is experiencing high rates of deforestation

(Shearman et al. 2009) and increasing pressure for conversion of forest to oil palm plantations (Obidzinski et al. 2014). Thus, New Guinea presents an interesting case study since it contains large areas of currently pristine habitat with high land conversion pressures in the near term.

Chapter 2

Accounting for long-term persistence of multiple species in systematic conservation planning

2.1 Introduction

Protected areas are a key component of global efforts to conserve biodiversity. Over the last 30 years, the terrestrial protected area network has expanded substantially, from 3.5% of global land area in 1985 (Zimmerer et al. 2004) to nearly 13% today (IUCN and UNEP-

WCMC 2016). However, despite this expansion, many biodiversity features remain unprotected or poorly represented within reserves (Rodrigues et al. 2004). Protected areas tend to be significantly biased towards land with low economic value with little pressure for conversion to agriculture (Joppa and Pfaff 2009). The result of this bias is a mismatch between the location of reserves and the spatial distribution of biodiversity. For example,

Venter et al. (2014) found that 17% of threatened vertebrate species do not occur within any protected area, and 85% are not afforded sufficient protection by existing protected areas to ensure their long-term persistence.

Systematic conservation planning offers a structured and scientifically defensible approach to designing new reserve networks that efficiently meet conservation objectives while minimizing socioeconomic cost (Margules and Pressey 2000). The fundamental objective of conservation planning is to ensure the long term persistence of biodiversity

(Cabeza and Moilanen 2001); however, under the current paradigm, conservation planning

rarely accounts for persistence explicitly (Cabeza and Moilanen 2001, Nicholson et al.

2006). Rather, reserve design exercises typically focus on maximizing the representation of species within reserves, with limited consideration of how the spatial configuration of these reserve will affect persistence (Cabeza and Moilanen 2003, Nicholson et al. 2006).

Some reserve design tools do include indirect means of accounting for species persistence. For example, Marxan (Ball et al. 2009), the most widely used conservation planning tool, can produce more compact and less fragmented spatial configurations of protected areas by minimizing the cumulative perimeter of the reserve network. Species- specific habitat requirements can also be incorporated in the form of minimum patch sizes, and the risk of simultaneous stochastic extinctions can be reduced by specifying the minimum number of patches and minimum separation distance between them (Ball et al.

2009). However, the relationship between patch size or configuration and local or regional persistence is never explicitly incorporated into these analyses; therefore none of these approaches assess the population ecology of the focal species in a process-based fashion.

The location and spatial configuration of reserve networks play a critical role in shaping the population dynamics, and hence persistence, of species in a landscape

(Cabeza and Moilanen 2001). Metapopulation theory provides a framework within which to study these spatial population dynamics, and to estimate how alternative reserve configurations affect species persistence (Hanski 1999, Hanski and Ovaskainen 2000).

However, despite substantial interest in both metapopulation theory and systematic conservation planning, exceedingly few attempts have been made to combine them. One 12

of the only examples of such a synthesis was provided by Nicholson et al. (2006), who developed a method for reserve selection based on minimizing the expected number of extinctions across multiple species. While their approach did explicitly account for species persistence, it was based on spatially explicit, stochastic metapopulation models (Frank and Wissel 2002) that require parameters such as home range size that are unknown for most species. In addition, Nicholson et al. (2006) constrained the application of their method to a fixed landscape of only 39 forest fragments, whereas most real-world reserve design problems divide landscapes into many thousands of planning units.

Here I present a method for systematic reserve design that explicitly maximizes regional persistence for a large assemblage of species. My approach is based on maximizing metapopulation capacity, a relative, asymptotic metric of persistence derived from a spatially explicit metapopulation model (Hanski 1999, Hanski and Ovaskainen

2000). This metric requires few parameters to calculate, and incorporates the size and spatial configuration of reserves as well as species-specific dispersal dynamics among them. I use simulated annealing, an optimization heuristic widely used in conservation planning, to maximize metapopulation capacity and, by extension, species persistence. I demonstrate our approach with a reserve planning exercise for 114 terrestrial mammal species in Indonesian New Guinea. For comparison, I also performed a traditional representation-based reserve planning exercise using Marxan.

As with other systematic reserve design tools, such as Marxan (Ball et al. 2009), mine requires dividing the study area into planning units that are candidates for protection.

The objective was to find the set of planning units that maximizes persistence across a 13

suite of species for a fixed cost. I used metapopulation capacity to estimate species persistence for candidate protected area networks and simulated annealing to perform the optimization. This reserve design tool was implemented in R (R Core Team 2015).

2.2 Methods

2.2.1 Metapopulation capacity

Metapopulation capacity is a relative, asymptotic approximation of the ability of a network of habitat patches to support a viable metapopulation of a species (Hanski and

Ovaskainen 2000). I used this metric because it is a proxy for species persistence that accounts for metapopulation processes such as dispersal and local persistence as well as landscape structure (Hanski and Ovaskainen 2000). This contrasts with graph theoretic metrics, which typically only address proximate correlates of persistence such as total landscape connectivity, and purely spatial metrics of landscape fragmentation such as those available in FRAGSTATS (Calabrese and Fagan 2004). Visconti and Elkin (2009) found that metapopulation capacity performs well compared to other measure of landscape connectivity in ranking habitat patches according to their contribution to population viability.

Metapopulation capacity is defined as the dominant eigenvalue, %, of the landscape

/.1 matrix M with elements &'( = *(,'().(.' , where .' is the area of patch 2 and ,'( is the edge-to-edge distance between patches 2 and 3. *(,'() is the species-specific dispersal- survival function, which indicates the proportion of individuals leaving patch 2 that make it

to patch 3 at distance ,'( away. I assume * decays linearly from 1 (when ,'( = 0) to 0 at the maximum dispersal distance of the species. Thus the only species-specific parameter required to calculate % is maximum dispersal distance, which can be estimated even for data-poor species using body-size allometries (Santini et al. 2013).

The original formulation, &'( = 0 for 2 = 3, leads to the undesirable behaviour that single-patch landscapes have % = 0 regardless of patch size. To address this, I followed

Schnell et al. (2013) in allowing for self-colonization (i.e., within patches) by setting &'( =

/.1 .(.' . Furthermore, since metapopulation capacity is a relative metric, which can vary by several orders of magnitude between species, to facilitate aggregation across different species I scaled % to be between 0 and 1 by dividing by its maximum possible value, which is attained when the entire study area is protected. This normalization results in a metric that assesses alternative reserve configurations based on relative, rather than absolute, differences in metapopulation persistence. Without the normalization, this reserve selection method prioritizes common, widespread species with overall high metapopulation capacity at the expense or rare, range-restricted species.

2.2.2 Simulated annealing

Simulated annealing is a stochastic heuristic for approximating global optima of complex functions (Kirkpatrick et al. 1983). This method is widely used for reserve design problems, notably by Marxan (Ball et al. 2009), the most widely used systematic conservation planning software. In this context, simulated annealing can be used to find a near-optimal set of planning units that minimizes a given objective function encapsulating the

conservation objective. I used the implementation of simulated annealing provided by R (R

Core Team 2015).

Given some initial starting set of planning units to include in the candidate reserve network, simulated annealing works by switching the status of a random planning unit from selected to not selected, or vice versa. Changes that result in a lower objective function are always kept, while changes that increase the objective function are sometimes kept, with a probability that decreases as the heuristic progresses. This probabilistic acceptance ensures that the heuristic avoids getting trapped in local minima. For further details on simulated annealing see Appendix A .

Simulated annealing typically uses a random starting point; however, I used Marxan

(Ball et al. 2009) to generate the starting reserve network. This approach reduces the number of simulated annealing iterations required by assuming that a Marxan-generated solution does a reasonable job of ensuring metapopulation persistence. This tool can therefore be seen as a method for refining Marxan solution to improve species persistence.

2.2.3 Mathematical formulation

I formulated the conservation objective (to maximize persistence across species for a fixed cost) mathematically as an optimization problem. In particular, given a set of & planning units, with costs 5(, and $ species the objective was

# J

Maximize > %' ? subject to 5(?( ≤ K '@A (@A where ?( is a binary indicator variable specifying whether planning unit 3 is included or excluded from the candidate protected area, %' ? is the metapopulation capacity (scaled 16

to 0-1) for species 2 and reserve configuration ?, > is the benefit function, and K is the total conservation budget. The benefit function, >, translates changes in metapopulation capacity to a measure of the conservation benefit associated with that change. I used a logarithmic benefit function (log N? + 1 /log (N + 1)) in this study, that preserves the 0-1 range of the scaled metapopulation capacity, while prioritizing changes that benefit species with low metapopulation capacity. This benefit function ensures that simulated annealing does not favour changes to the reserve configuration that benefit species with already high persistence at the expense of species with low persistence.

The above formulation consists of maximizing an objective function subject to a budgetary constraint; however, simulating annealing is significantly more efficient without explicit constraints (Nicholson et al. 2006). Therefore, I modified the objective function to incorporate the constraint directly

# J

*(?) = > %' ? − S ∙ max 0, 5(?( − K '@A (@A where the second term is a penalty for going over budget and S measures the relative importance of staying within budget versus increasing metapopulation persistence. It is this objective function that I sought to maximize with simulated annealing.

2.2.4 Case study

To demonstrate the use of this persistence-based reserve design method, I applied it to the western half of the island of New Guinea, which is composed on the Indonesian states

Papua and West Papua. New Guinea has extremely high biodiversity and endemicity

(Myers et al. 2000), and the world’s third largest remaining area of contiguous tropical rainforest. The region is also experiencing high rates of deforestation (Shearman et al.

2009) and increasing pressure for conversion of forest to oil palm plantations (Obidzinski et al. 2014). Thus, New Guinea presents an interesting case study since it contains large areas of currently pristine habitat with high land conversion pressures in the near term.

I divided the study region into 100km2 hexagonal planning units (& = 4,399). This planning unit size is arbitrary, but represented a balance between computational feasibility and planning flexibility, and is on a spatial scale relevant for this study, given the low population density of many tropical rainforest mammals (for example, Brodie and Giordano

2012). Within each planning unit, I estimated the cost for protection by combining four factors: current oil palm yield, predicted 2050 oil palm yield, population density, and percent crop cover. Oil palm yield came from the Global Agro-Ecological Zones dataset

(GAEZ; IIASA/FAO 2012) and represents the opportunity cost for protection, assuming that the most lucrative alternative land use is oil palm cultivation (Runting et al. 2015).

Population density and percent crop cover were used as metrics of current land use intensity and were derived from the Global Rural-Urban Mapping Project (GRUMP;

CIESIN/IFPRI/WB/CIAT 2011) and EarthStat Cropland and Pasture Area dataset

(Ramankutty et al. 2002), respectively. These four cost metrics were normalized, then averaged to produce a final measure of cost of protection within each cell (Figure 2.1a).

I used 114 native terrestrial mammal species as the conservation features (Figure

2.1b). Range polygons for these species came from the IUCN Red List (IUCN 2015). Most of these polygons are based on expert opinion because data are lacking for so many 18

tropical species; while range estimates almost certainly have inaccuracies for any particular species, I had no reason to expect any systematic bias when combining estimates across so many taxa. Body mass and home range size for these species were taken from the

PanTHERIA mammal trait database (Jones et al. 2009) and used to estimate maximum dispersal distance via the allometric relationships developed by Santini et al. (2013).

Maximum dispersal distances were used to parameterize dispersal survival functions, which in turn were used to calculate metapopulation capacity. For the full list of species, including maximum dispersal distances, see Appendix B .

I used Marxan to generate a candidate reserve network both to serve as a starting point for the persistence-based reserve design, and as a baseline for comparison of the performance of the method. In contrast to the persistence-based approach, Marxan requires representation targets for each species being considered. I set these targets to

20% of each species’ current range. Ideally, species-specific targets would be set in a systematic fashion through viability analysis or elicitation of expert opinion (Ardron et al.

2010); however, since this case study is meant as a proof of concept, I felt that a 20% target for all species was a reasonable starting point. To set the Species Penalty Factors

(SPFs) and Boundary Length Modifier (BLM) I followed the best practices for calibration outlined in the Marxan Good Practices Handbook (Ardron et al. 2010). I ran Marxan for

100 simulated annealing runs and used the best solution (i.e. the one with the lowest objective function) as the starting point for the persistence-based selection. To aid comparison between approaches, the cost of this best solution was used to set the

budget for our method. I used the R package marxan (Hanson and Watts 2015) as an interface to the Marxan command-line tool (Ball et al. 2009).

Finally, I ran the metapopulation capacity-based reserve design method for 5,000 simulated annealing iterations using an Amazon Web Services EC2 cluster. Simulated annealing is typically run for a larger number of iterations for a problem of this type; however, I was restricted to 5,000 iterations because of the high computational requirements of this task. Given that I started with a sensible Marxan solution, rather than a random starting point, and that this study is intended to provide a proof of concept rather than precise conservation recommendations, I feel this to be sufficient to demonstrate the utility of the method. One side effect of the relatively small number of iterations was noise in the solution in the form of small, isolated collections of planning units. I removed these from the raw simulated annealing output to produce the final solution.

Figure 2.1 Map of the study area in Indonesian New Guinea, showing 4,399 hexagonal 100 km2 planning units. (a) Cost of planning units, composed of current and 2050 potential oil palm yield (IIASA/FAO 2012), percent crop cover (Ramankutty et al. 2002), and population density (CIESIN/IFPRI/WB/CIAT 2011). Each of the four components are standardized to 0-1, then averaged. (b) Richness of 114 terrestrial mammal species used in this study based on IUCN Red List range polygons (IUCN 2015). The low cost, high richness area at the center of the island corresponds to the New Guinean Highlands, which is largely unsuitable for palm due to the high altitude.

2.3 Results

Comparison of the best reserve output from Marxan with the results from the persistence- based optimization, showed that substantial gains in metapopulation persistence can be achieved for only modest reductions in representation (Table 2.1). In particular, the reserve output from my method had a mean scaled metapopulation capacity that is 2.4 times higher than the Marxan reserve. This result applied to both the raw simulated annealing output (Figure 2.2b) and the final reserve after removal of small patches (i.e. noise) (Figure

2.2c). Henceforth I will focus on the final reserve with noise removed. The Marxan reserve

(Figure 2.2a) met the 20% representation targets for all species (Table 2.1), which was expected given that this is the task that Marxan was designed for. In contrast, for 20 of the

114 species (17.5%), my persistence-based selection method failed to meet the representation target. For these 20 species, there was on average a 27% gap between the actual representation and the target level.

The spatial configuration of the different candidate reserves (Figure 2.2) revealed that the persistence-based method refined the Marxan solution by connecting the discrete reserves in the center of the study area into one large contiguous reserve spanning much of the high-diversity highlands along the spine of New Guinea. Compared to the Marxan results, the refined reserve was 12.3% larger in area and 20.2% lower in cost (Table 2.1), showing that low cost, marginal planning units were added to form connections between formerly unconnected reserve patches.

Table 2.1 Summary of reserve performance using three reserve selection methods: representation-based selection using Marxan, our method maximizing metapopulation capacity, and our method excluding patches less than 4 planning units in size. Y is the mean scaled metapopulation capacity, which ranges from 0 (no protection) to 1 (entire study area protected). Targets missed gives the number of species (out of 114 total) for which the 20% representation target was not met and, for these species, % gap gives the percent difference between the actual and targeted representation levels. Method Cost Area (km2) Y Targets Missed % Gap Marxan 289.1 81,853 0.1703 0/114

Maximize % 288.4 103,970 0.4171 17/114 14% Maximize % (simplified) 230.8 91,958 0.4171 20/114 27%

Explicitly accounting for persistence in the reserve selection process led to an increase in metapopulation capacity for a large majority (87%) of species when compared to the Marxan solution (Figure 2.3). While this method had a substantial positive effect on persistence for most species, two Critically Endangered species saw a decrease in metapopulation capacity: the western long-beaked echidna, for which the decrease was effectively zero, and the golden-mantled tree kangaroo, which saw a more significant decrease.

The distribution of changes in metapopulation capacity was distinctly bimodal

(Figure 2.3), suggesting two categories of species: those for which persistence-based reserve design is important and those for which traditional representation-based approaches may be sufficient. One potential factor underlying this difference in the response of species to my reserve selection method is the variation in species ranges, which span several orders of magnitude (16km2-40,000km2). Range-restricted species experienced a greater increase in metapopulation capacity than did widespread species

(Figure 2.4) and are therefore better candidates for persistence-based planning.

Figure 2.3 Distribution of the differences in species-specific scaled metapopulation capacity between candidate reserves from a Marxan analysis versus our persistence-based selection method. Positive values indicate an improvement in metapopulation capacity. Bars are coloured according to species’ IUCN Red List status.

Figure 2.4 Relationship between change in scaled metapopulation capacity (between -1 and 1) and range size (from IUCN distribution polygons) within Indonesian New Guinea for each of the 114 mammal species in the case study. The y-axis shows the difference in scaled metapopulation capacity between candidate reserves from Marxan and our persistence-based selection method. Range restricted species see a greater improvement in metapopulation capacity compared to widespread species.

2.4 Discussion

In this study, I developed a novel method for systematic reserve design that explicitly maximizes long-term species persistence while achieving only marginally lower species representation than traditional conservation planning strategies. This method is grounded in ecological theory and accounts for the interaction between the spatial configuration of reserves and the metapopulation dynamics of species. This contrasts with the majority of systematic conservation planning tools which focus on meeting species representation targets and either ignore the spatial configuration of habitat or account for it very indirectly via simple landscape metrics that are unrelated to ecological processes (Saura and

Pascual-Hortal 2007). Addressing spatial configuration in reserve design is challenging because suitable metapopulation models are typically difficult to parameterize and computational limitations preclude them from inclusion in an optimization framework. I overcame these challenges by using metapopulation capacity, which requires few species- specific parameters and offers a balance between ecological realism and computational efficiency.

While this method led to an overall increase in mean metapopulation capacity across the entire mammal assemblage, at the level of individual species I identified two distinct groups: those that experienced significant benefits from my persistence-based approach and those for which the Marxan solution was nearly as good or slightly better.

This pattern is partly explained by differences in range sizes between species, with range- restricted species apparently exhibiting a greater need for persistence-based planning. 28

However, further work is needed to identify the ecological or geographic traits that are associated with this distinction. Once identified, species that require explicit planning for persistence can be treated separately from those for which more efficient, representation- based approaches are sufficient.

The net effect of my method on the spatial configuration of the reserve network was to connect groups of close, but unconnected, reserve into larger reserves. These findings are consistent with a study by Brodie et al. (2016) on the importance of different links between patches in a network. They found that short links connecting large patches into

“super-patches” were of greatest importance for metapopulation persistence and metacommunity stability. Starting from the Marxan solution (Figure 2.2a), my method connected the chain of five large reserves along the center of the island into a single reserve (Figure 2.2c), and the two reserves in the north into another large reserve. Thus these findings reinforce the importance of integrating reserve design with wildlife corridors design to improve connectivity between reserves (Beier et al. 2011).

Underlying this reserve design method is a set of assumptions resulting from the use of metapopulation capacity as a persistence metric. The metapopulation capacity metric is based on a simple spatially explicit metapopulation model consisting of colonization and extinction within a network of habitat patches embedded in a matrix of non-habitat (Hanski 1994, Hanski and Ovaskainen 2000). Apart from difference in size and isolation, all patches are assumed to be identical in terms of habitat quality, and colonization and extinction rates. Furthermore, this model assumes that the unprotected matrix is homogenous and permits dispersal, but cannot support populations of species. 29

In reality, there is considerable heterogeneity in the landscape that is not captured by this binary patch-matrix dichotomy, yet may have an impact on metapopulation dynamics

(Moilanen and Hanski 1998, Prevedello and Vieira 2009). More complex metapopulation models that address these issues have been developed; however, as noted above, computational and data limitations make them unsuitable for the present application.

Finally, metapopulation capacity is based on a model that assumes the landscape has reached colonization-extinction equilibrium. Therefore my method should be used to predict long-term persistence, but could be biased in the immediate aftermath of fragmentation.

In addition to limitations and assumptions inherent in our metapopulation model, uncertainty in our knowledge of the species under consideration affects the robustness of the results. To calculate metapopulation capacity, I used estimates of species-specific maximum dispersal distance (Santini et al. 2013). The error associated with this estimation propagates into uncertainty in the outcome of the reserve selection exercise. However,

Nicholson and Possingham (2007) found that rankings of different reserve configurations based on metapopulation metrics were generally robust to uncertainty in parameter estimates, thus conservation decision can be robust to uncertainty.

My results were also affected by issues of spatial correlation in the pattern of diversity. The 114 species in this study were not independently distributed across the study area; rather most species occurred in the central highlands (Figure 2.1b). Thus this region was overrepresented in the final reserve configuration, which consisted primarily of one contiguous “super-reserve” spanning the high diversity highlands (Figure 2.2c). 30

Species occurring outside of this high diversity region may experience shortfalls in protection. In particular, none of the 16 species that experienced a decline in metapopulation capacity with our method were found within the “super-reserve”. This suggest that my method may be a good candidate for a focal species approach in which a subset of representative species is chosen based on geographical distribution and life history characteristics (Nicholson et al. 2013). This approach would have the dual benefit of working with a spatially independent set of species and reduced computation time due to fewer species under consideration.

In the case study, there were some threatened species that saw a decrease in metapopulation capacity relative to the Marxan solution. This scenario is likely to occur when certain planning units are critical for the protection of a single species, but of limited value to remaining species. To address this, I recommend identifying range restricted species that are at high risk of extinction, and locking their ranges into the final solution.

Alternatively, species-specific weighting factors (equivalent to species penalty factors in

Marxan) could be included in the objective function, and higher weights could be given to threatened species that are experiencing declines in metapopulation capacity.

Despite these caveats, as a proof of concept my method demonstrates that significant gains in persistence can be achieved for only modest losses in representation. I argue that this trade-off is justified since the importance of representation is largely due to its contribution to ensuring persistence. More broadly, this method highlights the importance of considering the spatial configuration of reserves in a species-specific, ecologically meaningful way. For many species, the dominant paradigm of representation- 31

based reserve design may yield reserve networks that do not adequately ensure long-term persistence. Computational limitations preclude the method I have described in this study from being used for complex reserve design problems. However, in cases where persistence cannot be explicitly incorporated into the optimization framework, metapopulation capacity can still be a valuable tool in reserve design, for example as a means of ranking different reserve configurations output from tools such as Marxan. I also highlight the importance of effective wildlife corridor design to ensure landscape connectivity.

Chapter 3

Conclusion

Systematic reserve design is a powerful framework for designing new reserve networks that efficiently meeting conservation objectives in a cost effective manner. While the overarching goal of most reserve design exercises is to ensure the long term persistence of biodiversity, due to computational constraints and data limitations, persistence is rarely accounted for explicitly. Where persistence has been incorporated into the reserve design processes, it has been done using metrics that do not account for the spatial population ecology of species.

This thesis represents the second example of a systematic reserve design exercise that explicitly accounts for persistence by incorporating a process-based metapopulation model directly into the selection algorithm. Furthermore, it is the only example of a method explicitly designed to be applicable to real-world reserve design problems involving multiple data-limited species and large numbers of planning units. Using a case study from

Indonesian New Guinea, I demonstrated that this method can yield significant gains in metapopulation persistence across a suite of 114 mammal species compared to traditional representation-based approaches.

The results of this case study suggest some best practices for including persistence in the reserve design process. First, rare or range-restricted species are more likely to benefit from persistence-based reserve design than widespread species. In cases of extremely range restricted species, especially those isolated from areas of high diversity, it 33

is recommended to lock the ranges of these species in to the final reserve network.

Second, persistence-based reserve design is a particularly good candidate for focal species approaches. Analyses that focus on a functionally, taxonomically, or geographically representative subset of species, will benefit from a spatially independent set of species and reduced computation time due to fewer species under consideration.

Finally, my analysis highlights the importance of linking habitat patches with wildlife corridors to increase connectivity.

More broadly, this thesis also highlights the importance of explicitly considering the spatial configuration of reserve networks in an ecologically meaningful way. The dominant reserve design paradigm, typified by Marxan, focuses on maximizing the amount of habitat protected. However, this approach takes a simplified view of the effect of reserve configuration on species persistence, and therefore may not secure biodiversity over the long term. In this thesis I have demonstrated the utility of process-based metapopulation models for optimizing reserve configuration to support species persistence.

References

Akçakaya HR, Mills G, Doncaster CP. 2007. The role of metapopulations in conservation.

In: DW MacDonald and K Service, editor. Key topics in conservation biology

Blackwell Publishing. p. 64–84.

Ardron JA, Possingham HP, Klein CJ. 2010. Marxan Good Practices Handbook, Version

2. Pacific Marine Analysis and Research Association, Victoria, BC, Canada.

Ball IR, Possingham HP, Watts M. 2009. Marxan and relatives: software for spatial

conservation prioritisation. In: A Moilanen, KA Wilson, and HP Possingham, editor.

Spatial Conservation Prioritisation: Quantitative Methods and Computational Tools

Oxford University Press, Oxford, UK. p. 185–195.

Beier P, Spencer W, Baldwin RF, McRae B. 2011. Toward best practices for developing

regional connectivity maps. Conservation Biology 25: 879–892.

Beyer HL, Dujardin Y, Watts ME, Possingham HP. 2016. Solving conservation planning

problems with integer linear programming. Ecological Modelling 328: 14–22.

Brodie JF, Mohd-Azlan J, Schnell JK. 2016. How individual links affect network stability in

a large-scale, heterogeneous metacommunity. Ecology Accepted.

Brodie J, Giordano AJ. 2012. Density of the vulnerable Sunda clouded leopard Neofelis

diardi in a protected area in Sabah, Malaysian Borneo. Oryx 46: 427–430.

Brook BW, Sodhi NS, Bradshaw CJ. 2008. Synergies among extinction drivers under

global change. Trends in Ecology & Evolution 23: 453–460.

Cabeza M, Moilanen A. 2001. Design of reserve networks and the persistence of

biodiversity. Trends in Ecology & Evolution 16: 242–248.

Cabeza M, Moilanen A. 2003. Site-selection algorithms and habitat loss. Conservation

Biology 17: 1402–1413.

Calabrese JM, Fagan WF. 2004. A comparison-shopper’s guide to connectivity metrics.

Frontiers in Ecology and the Environment 2: 529–536.

CIESIN/IFPRI/WB/CIAT. 2011. Global Rural-Urban Mapping Project, Version 1

(GRUMPv1): Population Density Grid.

Convention on Biological Diversity. 2011. Conference of the Parties Decision X/2: Strategic

plan for biodiversity 2011–2020.

Frank K, Wissel C. 2002. A formula for the mean lifetime of metapopulations in

heterogeneous landscapes. The American Naturalist 159: 530–552.

Hansen MC, Potapov PV, Moore R, Hancher M, Turubanova SA, Tyukavina A, Thau D,

Stehman SV, Goetz SJ, Loveland TR, Kommareddy A, Egorov A, Chini L, Justice

CO, Townshend JRG. 2013. High-resolution global maps of 21st-century forest

cover change. Science 342: 850–853.

Hanski I. 1994. A practical model of metapopulation dynamics. Journal of animal ecology

151–162.

Hanski I. 1999. Habitat connectivity, habitat continuity, and metapopulations in dynamic

landscapes. Oikos 209–219.

Hanski I, Gilpin M. 1991. Metapopulation dynamics: Brief history and conceptual domain.

Biological journal of the Linnean Society 42: 3–16.

Hanski I, Ovaskainen O. 2000. The metapopulation capacity of a fragmented landscape.

Nature 404: 755–758.

Hanson JO, Watts ME. 2015. marxan: Decision support tools for reserve selection in R

using Marxan.

IIASA/FAO. 2012. Global Agro-ecological Zones (GAEZ v3.0).

IUCN. 2015. The IUCN Red List of Threatened Species.

IUCN and UNEP-WCMC. 2016. The World Database on Protected Areas (WDPA).

Jenkins CN, Joppa L. 2009. Expansion of the global terrestrial protected area system.

Biological Conservation 142: 2166–2174.

Jones KE, Bielby J, Cardillo M, Fritz SA, O’Dell J, Orme CDL, Safi K, Sechrest W, Boakes

EH, Carbone C, Connolly C, Cutts MJ, Foster JK, Grenyer R, Habib M, Plaster CA,

Price SA, Rigby EA, Rist J, Teacher A, Bininda-Emonds ORP, Gittleman JL, Mace

GM, Purvis A. 2009. PanTHERIA: a species-level database of life history, ecology,

and geography of extant and recently extinct mammals. Ecology 90: 2648–2648.

Joppa LN, Pfaff A. 2009. High and far: biases in the location of protected areas. PLoS

ONE 4: e8273.

Kirkpatrick S, Gelatt CD, Vecchi MP. 1983. Optimization by simulated annealing. Science

220: 671–680.

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

Moilanen A, Cabeza M. 2002. Single-species dynamic site selection. Ecological

Applications 12: 913–926.

Moilanen A, Hanski I. 1998. Metapopulation dynamics: Effects of habitat quality and

landscape structure. Ecology 79: 2503–2515.

Myers N, Mittermeier RA, Mittermeier CG, Da Fonseca GA, Kent J. 2000. Biodiversity

hotspots for conservation priorities. Nature 403: 853–858.

Nalle DJ, Montgomery CA, Arthur JL, Polasky S, Schumaker NH. 2004. Modeling joint

production of wildlife and timber. Journal of Environmental Economics and

Management 48: 997–1017.

Nicholson E, Lindenmayer DB, Frank K, Possingham HP. 2013. Testing the focal species

approach to making conservation decisions for species persistence. Diversity and

Distributions 19: 530–540.

Nicholson E, Possingham HP. 2007. Making conservation decisions under uncertainty for

the persistence of multiple species. Ecological applications 17: 251–265.

Nicholson E, Westphal MI, Frank K, Rochester WA, Pressey RL, Lindenmayer DB,

Possingham HP. 2006. A new method for conservation planning for the persistence

of multiple species: Multiple species conservation planning. Ecology Letters 9:

1049–1060.

Obidzinski K, Dermawan A, Hadianto A. 2014. Oil palm plantation investments in

Indonesia’s forest frontiers: Limited economic multipliers and uncertain benefits for

local communities. Environment, Development and Sustainability 16: 1177–1196.

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.

Pimm SL, Raven P. 2000. Biodiversity: Extinction by numbers. Nature 403: 843–845.

Pimm SL, Russell GJ, Gittleman JL, Brooks TM. 1995. The future of biodiversity. Science

269: 347.

Possingham H, Ball I, Andelman S. 2000. Mathematical methods for identifying

representative reserve networks. Quantitative methods for conservation biology

Springer. p. 291–306.

Prevedello JA, Vieira MV. 2009. Does the type of matrix matter? A quantitative review of

the evidence. Biodiversity and Conservation 19: 1205–1223.

Ramankutty N, Foley JA, Norman J, McSweeney K. 2002. The global distribution of

cultivable lands: current patterns and sensitivity to possible climate change. Global

Ecology and Biogeography 11: 377–392.

R Core Team. 2015. R: A Language and Environment for Statistical Computing. Vienna,

Austria: R Foundation for Statistical Computing.

Rodrigues ASL, Andelman SJ, Bakarr MI, Boitani L, Brooks TM, Cowling RM, Fishpool

LDC, da Fonseca GAB, Gaston KJ, Hoffmann M, Long JS, Marquet PA, Pilgrim JD,

Pressey RL, Schipper J, Sechrest W, Stuart SN, Underhill LG, Waller RW, Watts

MEJ, Yan X. 2004. Effectiveness of the global protected area network in

representing species diversity. Nature 428: 640–643.

Runting RK, Meijaard E, Abram NK, Wells JA, Gaveau DLA, Ancrenaz M, Posssingham

HP, Wich SA, Ardiansyah F, Gumal MT, Ambu LN, Wilson KA. 2015. Alternative

futures for Borneo show the value of integrating economic and conservation targets

across borders. Nature Communications 6: 6819.

Santini L, Di Marco M, Visconti P, Baisero D, Boitani L, Rondinini C. 2013. Ecological

correlates of dispersal distance in terrestrial mammals. Hystrix, the Italian Journal of

Mammalogy 24: 181–186.

Saura S, Pascual-Hortal L. 2007. A new habitat availability index to integrate connectivity

in landscape conservation planning: Comparison with existing indices and

application to a case study. Landscape and Urban Planning 83: 91–103.

Saura S, Torné J. 2009. Conefor Sensinode 2.2: a software package for quantifying the

importance of habitat patches for landscape connectivity. Environmental Modelling

& Software 24: 135–139.

Schnell JK, Harris GM, Pimm SL, Russell GJ. 2013. Estimating extinction risk with

metapopulation models of large-scale fragmentation. Conservation Biology 27:

520–530.

Shearman PL, Ash J, Mackey B, Bryan JE, Lokes B. 2009. Forest conversion and

degradation in Papua New Guinea 1972–2002. Biotropica 41: 379–390.

Venter O, Fuller RA, Segan DB, Carwardine J, Brooks T, Butchart SHM, Marco MD,

Iwamura T, Joseph L, O’Grady D, Possingham HP, Rondinini C, Smith RJ, Venter

M, Watson JEM. 2014. Targeting global protected area expansion for imperiled

biodiversity. PLoS Biology 12: e1001891.

Visconti P, Elkin C. 2009. Using connectivity metrics in conservation planning–when does

habitat quality matter? Diversity and Distributions 15: 602–612.

Westphal MI, Pickett M, Getz WM, Possingham HP. 2003. The use of stochastic dynamic

programming in optimal landscape reconstruction for metapopulations. Ecological

Applications 13: 543–555.

Wilson KA, Cabeza M, Klein CJ. 2009. Fundamental concepts of spatial conservation

prioritization. In: A Moilanen, KA Wilson, and HP Possingham, editor. Spatial

Conservation Prioritisation: Quantitative Methods and Computational Tools Oxford

University Press, Oxford, UK. p. 185–195.

Zimmerer KS, Galt RE, Buck MV. 2004. Globalization and multi-spatial trends in the

coverage of protected-area conservation (1980-2000). Ambio: A Journal of the

Human Environment 33: 520–529.

Appendices

Appendix A Simulated annealing

Simulated annealing is a stochastic heuristic for approximating global optima of complex functions with many local optima. It widely used in conservation planning because it can efficiently be applied to highly non-linear objective functions with many decision variables.

For these problems, exact optimization methods are either unavailable or their inefficiency precludes them from being applied to all but the simplest problems. However, Simulated annealing does not find the true minimum of the objective function. Rather it finds a near minimal solution and, since it is stochastic, each time simulated annealing is run, a different near solution is produced. In the context of conservation planning, simulated annealing is performed many times to generate a suite of possible reserve networks. Thus, multiple options can be presented to decision makers, and the importance of particular planning units can be measured by the selection frequency (i.e. the number of solutions containing that planning unit).

To determine which combination of planning units minimizes the objective function, simulated annealing stochastically and iteratively explores the state space of the decision variables by:

1. Randomly choosing a planning unit and changing its status from selected to not-

selected, or vice versa.

2. If the change decreases the objective function it is accepted. If the change

increases the objective function, it is accepted according to an acceptance

probability, which depends on the magnitude of the increase in objective function

and a global temperature parameter.

3. As the process progresses, the temperature parameter gradually decreases

according to an annealing schedule. As a result, changes that increase the

objective function become less likely to be accepted.

Initially having some non-zero acceptance probability for changes that increase the objective function, reduces the likelihood of being caught in a local minimum. This acceptance probability is given by:

Δ* min 1, exp − ]' where Δ* is the difference in objective function and ]' is the temperature parameter in iteration 2.

The annealing schedule determines the rate at which ]' decreases, and alternate implementations of simulated annealing use different functional forms for the schedule. In particular, Marxan uses

' ]' = S ∙ ]/

and the R implementation of simulated annealing uses

] ] = / ' log (2 + ^) where ]/ is the starting temperature and S is a cooling factor, both of which are user- defined parameters.

Appendix B Species list

Table B.1 List of 114 terrestrial mammal species included in this study. IUCN status and ranges are taken from the IUCN Red List (IUCN 2015), mass comes from the PanTHERIA mammal trait database (Jones et al. 2009), and maximum dispersal distance is based on allometric relationships from Santini et al. (2013). For species with no mass, dispersal distance was based on the mean dispersal distance for the genus. IUCN Range in Study Mass Maximum Dispersal Order Species Status Area (km2) (kg) Distance (km) Dasyuromorphia Dasyurus albopunctatus NT 267,926 0.60 9.73 Dasyuromorphia Dasyurus spartacus NT 8,199 0.89 12.39 Dasyuromorphia Murexia habbema LC 35,978 0.04 1.79 Dasyuromorphia Murexia longicaudata LC 204,371 0.13 3.64 Dasyuromorphia Murexia melanurus LC 95,085 0.04 1.71 Dasyuromorphia Murexia so LC 86,076 0.13 3.64 Dasyuromorphia Myoictis melas LC 277,519 0.22 5.18

Dasyuromorphia Myoictis wallacei LC 12,278 - 5.18 Dasyuromorphia Neophascogale lorentzii LC 78,223 0.21 5.03 Dasyuromorphia Phascolosorex doriae LC 81,667 0.17 4.36 Dasyuromorphia Phascolosorex dorsalis LC 67,980 0.12 3.57 Dasyuromorphia Planigale novaeguineae LC 8,181 0.01 0.95 Dasyuromorphia Sminthopsis virginiae LC 37,725 0.04 1.63 Diprotodontia Cercartetus caudatus LC 108,832 0.02 0.28

IUCN Range in Study Mass Maximum Dispersal Order Species Status Area (km2) (kg) Distance (km) Diprotodontia Dactylopsila megalura LC 54,477 0.31 6.46 Diprotodontia Dactylopsila trivirgata LC 349,750 0.41 7.67 Diprotodontia Dactylox palpator LC 80,004 0.37 7.15 Diprotodontia Dendrolagus inustus VU 115,218 12.63 12.24 Diprotodontia Dendrolagus mbaiso EN 5,547 9.40 10.25

Diprotodontia Dendrolagus pulcherrimus CR 152 - 9.83

Diprotodontia Dendrolagus stellarum VU 38,992 - 9.83 Diprotodontia Dendrolagus ursinus VU 27,925 13.25 12.60 Diprotodontia Distoechurus pentus LC 200,210 0.05 0.46 Diprotodontia Dorcopsis hageni LC 64,498 5.50 7.43 Diprotodontia Dorcopsis luctuosa VU 41,816 4.94 6.97 Diprotodontia Dorcopsis muelleri LC 118,757 5.37 7.33 Diprotodontia Dorcopsulus vanheurni NT 39,346 1.89 3.92 Diprotodontia Macropus agilis LC 48,999 11.82 11.76 Diprotodontia Petaurus breviceps LC 390,153 0.12 0.94 Diprotodontia Phalanger carmelitae LC 36,842 1.82 19.51 Diprotodontia Phalanger gymnotis LC 288,823 2.60 24.41

Diprotodontia Phalanger mimicus LC 113,092 - 19.77

IUCN Range in Study Mass Maximum Dispersal Order Species Status Area (km2) (kg) Distance (km) Diprotodontia Phalanger orientalis LC 177,329 2.49 23.75 Diprotodontia Phalanger sericeus LC 45,015 2.00 20.72 Diprotodontia Phalanger vestitus LC 15,334 1.85 19.71 Diprotodontia Pseudochirops albertisii NT 20,120 0.78 11.48 Diprotodontia Pseudochirops corine NT 49,503 1.12 14.39 Diprotodontia Pseudochirops corotus VU 7,451 1.40 16.50 Diprotodontia Pseudochirops cupreus LC 68,729 1.76 19.13 Diprotodontia Pseudochirulus canescens LC 187,135 0.30 6.27 Diprotodontia Pseudochirulus caroli LC 37,932 0.46 8.16

Diprotodontia Pseudochirulus larvatus LC 1,160 - 8.80 Diprotodontia Pseudochirulus mayeri LC 45,656 0.15 4.08 Diprotodontia Pseudochirulus schlegeli VU 10,312 0.26 5.66 Diprotodontia Spilocuscus maculatus LC 390,153 4.06 32.34 Diprotodontia Spilocuscus rufoniger CR 74,058 6.00 41.36 Diprotodontia Thylogale browni VU 17,125 5.48 7.41 Diprotodontia Thylogale brunii VU 42,045 4.03 6.17 Diprotodontia Thylogale stigmatica LC 873 4.51 3.86 Monotremata Tachyglossus aculeatus LC 9,306 4.50 34.50

IUCN Range in Study Mass Maximum Dispersal Order Species Status Area (km2) (kg) Distance (km)

Monotremata Zaglossus attenboroughi CR 16 - 53.22

Monotremata Zaglossus bartoni CR 49,574 - 53.22 Monotremata Zaglossus bruijnii CR 87,688 8.95 53.22 Peramelemorphia Echymipera clara LC 85,074 1.20 15.04 Peramelemorphia Echymipera kalubu LC 390,741 0.83 11.85 Peramelemorphia Echymipera rufescens LC 389,138 1.05 13.81 Peramelemorphia Isoodon macrourus LC 5,589 1.51 17.31 Peramelemorphia Microperoryctes aplini DD 76 0.23 5.29 Peramelemorphia Microperoryctes longicauda LC 79,887 0.54 9.10 Peramelemorphia Microperoryctes muri DD 609 0.23 5.29 Peramelemorphia Peroryctes raffraya LC 100,085 0.91 12.57 Rodentia Anisomys imitator LC 67,113 0.51 1.79 Rodentia Baiyankamys habbema DD 774 0.09 0.63 Rodentia Coccymys albidens DD 87 0.05 0.42 Rodentia Coccymys ruemmleri LC 16,998 0.03 0.32 Rodentia Crossomys moncktoni LC 17,723 0.17 0.92 Rodentia Hydromys chrysogaster LC 329,164 0.63 2.02 Rodentia Hydromys hussoni DD 838 0.10 0.69

IUCN Range in Study Mass Maximum Dispersal Order Species Status Area (km2) (kg) Distance (km) Rodentia Hyomys dammermani DD 67,504 0.93 2.56 Rodentia Lorentzimys nouhuysi LC 34,312 0.01 0.21 Rodentia Macruromys elegans DD 340 0.12 0.74 Rodentia Macruromys major LC 25,134 0.36 1.44 Rodentia Mallomys gunung EN 926 2.04 4.10 Rodentia Mallomys istapantap LC 15,172 1.99 4.04 Rodentia Mallomys rothschildi LC 75,040 1.16 2.92 Rodentia Mammelomys lanosus LC 31,288 0.12 0.74 Rodentia Mammelomys rattoides LC 71,387 0.21 1.05 Rodentia Melomys burtoni LC 11,067 0.07 0.55 Rodentia Melomys frigicola LC 2,691 0.06 0.51 Rodentia Melomys leucogaster LC 102,968 0.10 0.69 Rodentia Melomys lutillus LC 123,239 0.04 0.36 Rodentia Melomys rufescens LC 355,301 0.06 0.49 Rodentia Microhydromys richardsoni DD 34,979 0.01 0.17 Rodentia Parahydromys asper LC 72,614 0.53 1.82 Rodentia Paraleptomys rufilatus EN 262 0.05 0.47 Rodentia Paraleptomys wilhelmi DD 4,368 0.03 0.35

IUCN Range in Study Mass Maximum Dispersal Order Species Status Area (km2) (kg) Distance (km) Rodentia Paramelomys lorentzii LC 94,561 0.15 0.86 Rodentia Paramelomys mollis LC 85,408 0.09 0.63 Rodentia Paramelomys platyops LC 197,853 0.09 0.61 Rodentia Paramelomys rubex LC 83,680 0.05 0.63 Rodentia Paramelomys so LC 112,050 0.14 0.84 Rodentia Paramelomys steini DD 405 0.06 0.49 Rodentia Pogonomelomys brassi LC 34,126 0.16 0.89 Rodentia Pogonomelomys bruijnii NT 26,430 0.16 0.89 Rodentia Pogonomelomys mayeri LC 38,619 0.11 0.72 Rodentia Pogonomys loriae LC 7,751 0.10 0.65 Rodentia Pogonomys macrourus LC 64,804 0.05 0.43 Rodentia Pogonomys sylvestris LC 16,856 0.04 0.40 Rodentia Pseudohydromys ellermani LC 26,484 0.02 0.27 Rodentia Pseudohydromys occidentalis DD 916 0.02 0.27 Rodentia Rattus arfakienis DD 80 0.05 0.36 Rodentia Rattus arrogans LC 12,218 0.07 0.53 Rodentia Rattus leucopus LC 134,427 0.20 1.02 Rodentia Rattus norvegicus LC 391,215 0.28 0.90

IUCN Range in Study Mass Maximum Dispersal Order Species Status Area (km2) (kg) Distance (km) Rodentia Rattus omichlodes DD 293 0.07 0.54 Rodentia Rattus pococki LC 58,492 0.06 0.52 Rodentia Rattus praetor LC 196,520 0.19 1.00 Rodentia Rattus richardsoni VU 6,977 0.06 0.51 Rodentia Rattus sordidus LC 7,380 0.16 0.88 Rodentia Rattus steini LC 88,428 0.15 0.86 Rodentia Rattus verecundus LC 11,401 0.10 0.47 Rodentia Uromys ak LC 73,427 0.75 2.24 Rodentia Uromys caudimaculatus LC 388,550 0.64 2.05 Rodentia Xenuromys barbatus LC 272,998 0.98 2.63