UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Conservation planning at multiple scales : a density model and spatial planning tool to facilitate the… Morrell, Nina 2018

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2019_february_morrell_nina.pdf [ 4.22MB ]
Metadata
JSON: 24-1.0373204.json
JSON-LD: 24-1.0373204-ld.json
RDF/XML (Pretty): 24-1.0373204-rdf.xml
RDF/JSON: 24-1.0373204-rdf.json
Turtle: 24-1.0373204-turtle.txt
N-Triples: 24-1.0373204-rdf-ntriples.txt
Original Record: 24-1.0373204-source.json
Full Text
24-1.0373204-fulltext.txt
Citation
24-1.0373204.ris

Full Text

i   Conservation planning at multiple scales: A density model and spatial planning tool to facilitate the conservation of Andean bears (Tremarctos ornatus) and the Northern Andes by Nina Morrell B.Sc., The University of British Columbia, 2014  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE in THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES (Forestry) THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  October 2018 © Nina Morrell, 2018   ii The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, a thesis/dissertation entitled: Conservation planning at multiple scales: A density model and spatial planning tool to facilitate the conservation of Andean bears (Tremarctos ornatus) and the Northern Andes  submitted by Nina Morrell  in partial fulfillment of the requirements for the degree of Master of Science in Forestry  Examining Committee: Dr. Peter Arcese, Forestry Supervisor    Dr. Cole Burton, Forestry Supervisory Committee Member  Dr. Richard Schuster, Liber Ero Fellow, Carleton University Supervisory Committee Member Dr. Angela Fuller, Department of Natural Resources, Cornell University Supervisory Committee Member     iii Abstract Global declines in large-bodied terrestrial vertebrates have been widely linked to human disturbance and habitat loss. Consequently, identifying remaining opportunities to conserve habitats likely to maximize the persistence of such species remains a key challenge to conserving biological diversity globally. Andean bears (Tremarctos ornatus) are among the least-known wide-ranging large mammal species, but occupy montane ecoregions throughout the Northern Andes, a global biodiversity hotspot. Recent evidence suggests the Andean bear range includes forests at the fringe of human development, including the equatorial dry forests of Peru. If Andean bears are to be conserved in this region, a better understanding of their potential densities, use of habitats, and response to human influence is required.  This thesis used spatial density models, existing field and remote-sensed data, and spatial planning tools to relate species detection to environmental variables and optimize conservation plans for Andean bears and the Northern Andes. I assessed habitat use by Andean bears in equatorial dry forest at a local scale in Northwestern Peru, using camera-trap data to construct spatially-explicit capture recapture (SCR) models. I compared models for resources thought to affect bear density by their association with threats and food availability; including elevation, slope, forest cover, and proximity to roads. I found that proximity to roads reduced the density of Andean bears, and the influence of factors other than roads varied seasonally. I identified potential areas of equatorial dry forest outside the IUCN range that could support Andean bears, but noted that unmapped roads and smallholder agriculture affected the reliability of results.  I also employed systematic prioritization methods to identify configurations of land parcels that maximized biodiversity features at the least cost. I found that the Andean bear range performed better at capturing species richness than a random control feature. I also found that planning for multiple goals in a systematic planning framework greatly increased the area efficiency of the solutions, compared with planning for biodiversity features separately. Overall, this thesis highlights the importance of working at multiple scales for efficient conservation planning, and provides a common framework for conducting such analyses in the future.  Resumen Los declives poblacionales globales de vertebrados terrestres mayores han sido ampliamente asociados a perturbaciones humanas, así como a la pérdida de hábitats. Como consecuencia, la identificación de oportunidades para conservar hábitats cuya protección maximice la probabilidad de sostener a sus poblaciones en el tiempo, es probable que siga siendo uno de los principales retos globales para la conservación de la biodiversidad. El oso andino (Tremarctos ornatus) es una de las especies de mamíferos mayores de amplio rango menos iv conocidas a nivel global. Sin embargo, esta especie ocupa un rango diverso de bosques a lo largo de los Andes norte, un hotspot de biodiversidad global. Evidencia reciente sugiere que el rango de distribución del oso andino incluye áreas de bosque ubicados al margen de asentamientos humanos, como en el bosque seco ecuatorial de Perú. Por lo tanto, la conservación del oso andino requiere un mejor entendimiento de la densidad poblacional, el uso de hábitat y la respuesta de estas poblaciones a la influencia humana en esta región. Esta tesis uso modelos de distribución espacial (MDEs), datos de campo y detección remota, así como herramientas de planificación espacial, para optimizar los planes de conservación del oso andino y de los Andes norte. Evalué el uso de hábitat del oso andino en el bosque seco ecuatorial a escala local en Perú, utilizando cámaras trampa para construir modelos espacialmente explícitos de captura y recaptura (MEECRs). Comparé modelos alternativos para recursos que podrían afectar la densidad del oso andino por su asociación con amenazas conocidas y disponibilidad de alimento; incluyendo también parámetros como: elevación, pendiente, cobertura forestal, y cercanía a carreteras. Encontré que la cercanía a carreteras es un factor que reduce la densidad de osos, y que la influencia de los otros factores varia de manera estacional. Luego expandí los resultados locales para evaluar el potencial del bosque seco ecuatorial para sostener poblaciones de oso andino. Mis resultados identificaron paisajes con alto potencial para sostener altas densidades de osos andinos que habitan fuera del rango establecido por la IUCN.  Adicionalmente, empleé métodos de priorización sistemática para explorar como usos alternativos de la tierra podrían incrementar la integridad y persistencia de los objetos de conservación y de las cuencas hidrográficas en los Andes norte. Encontré que el uso de la distribución planteada por el IUCN para esta especie era capaz de capturar mejor la riqueza de especies de la zona, que distribuciones uniformes aleatorias sobre el mismo espacio. Encontré también que la planificación simultánea de múltiples objetivos, en el marco de la planificación sistemática para la conservación, incrementó altamente la eficiencia de cobertura (área) de las soluciones, en comparación con ejercicios planificación cuya aproximación es la evaluación de múltiples objetivos de conservación por separado. En síntesis, esta tesis resalta la importancia de trabajar en múltiples escalas para mejorar la eficiencia de la planificación para la conservación, y presenta un marco metodológico para desarrollar dicho proceso en el futuro.   v Lay summary Large-bodied vertebrate species are in global decline, largely due to habitat loss. With many countries committed to meeting goals for conservation under a limited budget, conservation has largely become a task of selecting the best available habitat at the least cost. In the Northern Andes, a global biodiversity hotspot, a rapid shift to human land use has put many species and ecosystems at risk. The Andean bear (Tremarctos ornatus) is a focal species for conservation in the Northern Andes and one of the least understood large carnivore species. This work i) evaluates the associations that Andean bears have with habitat in the equatorial dry forests of Peru; and ii) evaluates conservation plans that use Andean bears as a focal species. This thesis contributes to a growing body of knowledge supporting the conservation of Andean bears, and aims to provide a common framework for land managers to conduct future analyses.  Resumen para legos Las poblaciones de vertebrados mayores a nivel global se encuentran en declive, principalmente debido a la pérdida del hábitat que ocupan. Con muchos países comprometidos con objetivos de conservación, pero con presupuestos limitados para dicha tarea, el problema de la conservación se convierte en un triaje: seleccionar el mejor hábitat disponible cuya protección genere el menor costo. En los Andes norte, un hotspot de biodiversidad mundial, la rápida transición hacia un uso del territorio dominado por los humanos a puesto en riesgo a muchas especies y ecosistemas. El oso andino (Tremarctos ornatus) es una especie objetivo para la conservación en los Andes norte, y una de las especies de carnívoros menos comprendidas de la tierra. Este trabajo: (i) evalúa la asociación del oso andino con variables de hábitat de fácil medición en el bosque seco ecuatorial de Perú; y (ii) evalúa la utilidad del oso andino como especie objetivo para acciones de conservación. Esta investigación contribuye a incrementar el conocimiento a favor de la conservación de los osos andinos, y busca generar un marco común para que los administradores de territorio puedan realizar evaluaciones futuras.  vi Preface This thesis uses 4 years out of a much larger dataset of camera trap data collected by field teams in the mountains around Rio La Leche, Lambayeque region, Peru. All field work was conducted by the dedicated staff and volunteers at the Spectacled Bear Conservation Society (SBC) under the supervision of Robyn Appleton. Drs. Peter Arcese, Richard Schuster, and Joanna Burgar advised my statistical analysis. My supervisor Dr. Peter Arcese provided great ideas, advice, and helpful comments on my thesis. My committee members, Drs. Cole Burton, Angela Fuller, and Richard Schuster, and my friend, Carolyn King, also provided useful feedback and comments on my thesis. I adapted methods for spatial capture-recapture originally designed by Andy Royle, Richard Chandler, and others. I used methods for systematic planning designed and written into the prioritizr R package by Jeffrey Hansen, Richard Schuster, myself, and others.  Chapter 2 will be submitted as a paper co-authored by Peter Arcese and Robyn Appleton. Peter Arcese guided the structure and direction of the chapter and provided substantial edits and statistical support. Robyn Appleton conducted field surveys to collect the camera trap data. Chapter 3 will be submitted as a paper co-authored by Peter Arcese and Richard Schuster. Peter Arcese again provided substantial guidance and edits, and Richard Schuster provided useful programming code (R) and processed GIS data layers that enabled my analyses. I conducted all analyses on my own with the guidance of co–authors, and wrote the original drafts of all sections, which were subsequently improved upon by comments and edits by Dr. Peter Arcese. I translated my abstract and lay summary into Spanish with the help of Santiago de la Puente, because my research project took place in Latin America.   vii Table of Contents Abstract ....................................................................................................................................iii Resumen ................................................................................................................................ iii Lay summary ............................................................................................................................ v Resumen para legos ............................................................................................................... v Preface ......................................................................................................................................vi Table of Contents ....................................................................................................................vii List of tables .............................................................................................................................ix List of figures ..........................................................................................................................xii Acknowledgements ............................................................................................................... xiv Dedication .............................................................................................................................. xvi Chapter 1. General introduction .............................................................................................. 1 1.1 Context in conservation .................................................................................................... 1 1.2 Andean bears as a focal species ...................................................................................... 3 1.3 Rationale and research approach ..................................................................................... 6 Chapter 2. Human influence and the density of Andean bears (Tremarctos ornatus) in dry forests of the northern Andes ................................................................................................. 9 2.1 Introduction ....................................................................................................................... 9 2.2 Methods ..........................................................................................................................12 2.2.1 Study area.................................................................................................................12 2.2.2 Field protocol ............................................................................................................13 2.2.3 Habitat data ...............................................................................................................15 2.2.4 Data analysis ............................................................................................................17 2.3 Results ............................................................................................................................20 2.3.1 Field surveys .............................................................................................................20 2.3.2 Habitat effects on density ..........................................................................................20 2.3.3 Regional predictions ..................................................................................................24 viii 2.4 Discussion .......................................................................................................................27 2.4.1 Regional Predictions .................................................................................................29 2.4.2 Conclusions ..............................................................................................................30 Chapter 3. Systematic conservation prioritisation in the Northern Andes: integrating focal species and multiple goals .....................................................................................................32 3.1 Introduction ......................................................................................................................32 3.2 Methods ..........................................................................................................................34 3.2.1 Study area.................................................................................................................34 3.2.2 Systematic conservation planning .............................................................................37 3.2.3 Scenario development and post processing ..............................................................38 3.3 Results ............................................................................................................................40 3.3.1 Biodiversity representation ........................................................................................40 3.3.2 Landscape linkages ..................................................................................................41 3.3.3 Efficacy of Andean bears as an umbrella species .....................................................41 3.4 Discussion .......................................................................................................................51 Chapter 4. General conclusion ..............................................................................................54 4.1 Implications .....................................................................................................................54 4.2 Key findings, limitations, and future steps ........................................................................55 References ..............................................................................................................................57 Appendix  A. Correlation Matrix for Chapter 2 .....................................................................69 Appendix  B. Model selection and MLE tables for chapter 2 ...............................................70 Appendix  C. Model selection tables for observation models in chapter 2 ........................75 Appendix  D. Distribution of Andean bears predicted for equatorial dry forest ................80 Appendix  E. Feature representation tables for chapter 3 ...................................................81 Appendix  F. Bears will be bears ..........................................................................................90    ix List of tables Table 2.1. Variables selected to determine Andean bear habitat availability and their ranges in continuous rasters of the Cerro Venado and Laquipampa study areas, Lambayeque, Peru, 2012-2016. Range values are shown in the original units before standardization for modelling. ........17 Table 2.2. Camera trap survey carried out in Cerro Venado in 2012-2013 and Laquipampa in 2015-2016. ................................................................................................................................20 Table 2.3. Model selection for spatial-capture recapture models to explain density of Andean bears (Tremarctos ornatus) at Cerro Venado. The best-fitting observation model (po ~ season*sex; σ ~season) is used in each model. K is the number of parameters in the model. ΔAIC is the difference in Akaike’s information criterion (AIC) between the best model and each successive model. wi is the Akaike weight, representing the probability that a model is the best in the model set. See Apprendix B  for full results. ........................................................................22 Table 2.4. Model selection for spatial-capture recapture models to explain density of Andean bears (Tremarctos ornatus) at Laquipampa Wildlife Refuge. Predictor variables are built into the activity center for density. The best-fitting observation model (po ~ 1; σ ~season*sex) is used in each model. See table 2.3 for definitions. ..................................................................................23 Table 2.5. Model selection for spatial-capture recapture models to explain density of Andean bears (Tremarctos ornatus) across the combined datasets of Cerro Venado and Laquipampa Wildlife Refuge. The best-fitting observation model (po ~ 1; σ ~season + sex) is used in each model. See table 2.3 for definitions. ..........................................................................................23 Table 3.1. Scenarios used to identify priority conservation areas in the Northern Andes. .........39 Table 3.2. Feature representation table for vertebrate species and water risk in three separate target setting scenarios. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. Scenario 1 supplied equal targets for all feature layers, scenario 2 supplied equal targets for all feature layers but dropped water risk, and scenario 3 only supplied targets for threatened vertebrate species. For full feature representations, including ecoregions, see Appendix E. ...........................................................44 Table 3.3. Feature representation table for vertebrate species and water risk in two single-feature target setting scenarios. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. a) uses Andean bear range; and b) uses a random uniformly generated feature covering the same total area as Andean bear range. For full feature representations, including ecoregions, see Appendix E. ...............................................49 Table A.1. Correlation matrix for spatial habitat variables used in initial and final modelling .....69 x Table B.1. Estimated average effect-size and standard error for all predictors of the density of Andean bears (Tremarctos ornatus) at Cerro Venado. Effect sizes indicate the standardized number of units of change in the predictor variable for a one-unit increase in the response. Coefficients are averaged by model weight across models within the model set containing 95% cumulative model weight. Adjusted estimate and adjusted standard errors include models where the parameter did not occur. Relative Value Index indicates the percentage of models that the parameter occurred in. ..............................................................................................................70 Table B.2. Estimated average effect-size and standard error for all predictors of the density of Andean bears (Tremarctos ornatus) at Laquipampa Wildlife Refuge. See table B.1 for definitions. .................................................................................................................................................71 Table B.3. Estimated average effect-size and standard error for all predictors of the density of Andean bears (Tremarctos ornatus) at Laquipampa Wildlife Refuge. See table B.1 for definitions. .................................................................................................................................................71 Table B.4. Full model selection table of spatial-capture recapture models to predict the density of Andean bears (Tremarctos ornatus) in Cerro Venado. The best-fitting observation model (po ~ season*sex; σ ~season) is used in each model. K is the number of parameters in the model. ΔAIC is the difference in Akaike’s information criterion (AIC) between the best model and each successive model. wi is the Akaike weight, representing the probability that a model is the best in the model set. ...........................................................................................................................72 Table B.5. Full model selection table of spatial-capture recapture models to predict the density of Andean bears (Tremarctos ornatus) in Laquipampa Wildlife Refuge. The best-fitting observation model (po ~ 1; σ ~season*sex) is used in each model. See Table B.4 for definitions. ...............73 Table B.6. Full model selection table of spatial-capture recapture models to predict the density of Andean bears (Tremarctos ornatus) in pooled datasets from Cerro Venado and Laquipampa Wildlife Refuge. The best-fitting observation model (po ~ 1; σ ~season + sex) is used in each model. See Table B.4 for definitions. .........................................................................................74 Table C.1. Model selection for base observation model used for Cerro Venado dataset, Lambayeque, Peru. D is the density model, p0 is the model for baseline capture probability, and σ is the model for the home range shape parameter. K is the number of parameters in the model. ΔAIC is the difference in Akaike’s information criterion (AIC) between the best model and each successive model. wi is the Akaike weight, representing the probability that a model is the best in the model set. ...........................................................................................................................75 xi Table C.2. Model selection for base observation model used for Laquipampa dataset, Lambayeque, Peru. See table C.1 for definitions. .....................................................................76 Table C.3. Model selection for base observation model used for combined Laquipampa and Cerro Venado datasets, Lambayeque, Peru. See table C.1 for definitions. .........................................78 Table E.1. Feature representation table for vertebrate species, water risk, and ecoregions in Scenario 1, where targets were set for all features. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. ....................81 Table E.2. Feature representation table for vertebrate species, water risk, and ecoregions in Scenario 2, where targets were set for all features except water risk. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. .................................................................................................................................................82 Table E.3. Feature representation table for vertebrate species, water risk, and ecoregions in Scenario 3, where targets were only set for threatened species. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature.  .................................................................................................................................................84 Table E.4. Feature representation table for vertebrate species, water risk, and ecoregions in Scenario 4, where targets were only set for the Andean bear range. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. 86 Table E.5. Feature representation table for vertebrate species, water risk, and ecoregions in Scenario 5, where targets were only set for a random species range. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. .................................................................................................................................................87   xii List of figures Figure 1.1. Example camera trap photo of an Andean bear (Tremarctos ornatus) from Cerro Venado field site, Lambayeque, Peru. ........................................................................................ 4 Figure 2.1. Location of the equatorial dry forest area used in this study and the Andean bear (Tremarctos ornatus) IUCN range (Goldstein et al. 2008) in northern Peru, South America. .....13 Figure 2.2. Camera trap locations for surveys conducted 2012-2016 in the Lambayeque region, Peru ..........................................................................................................................................14 Figure 2.3. Example camera trap photos from Cerro Venado and Laquipampa showing unique facial markings of Andean bears (SBC-Peru 2017). ..................................................................15 Figure 2.4. Averaged, natural log-scale effect-size estimates (β coefficients) and 95% confidence intervals for top habitat variables explaining spatial variation in estimated density of Andean bears (Tremarctos ornatus). Effect sizes indicate the standardized number of units of change in the response variable for a one-unit increase in the predictor. Coefficients are averaged by model weight across models within the model set containing 95% cumulative model weight. See Appendix B for detailed results. .................................................................................................24 Figure 2.5. Predicted density of Andean bear (Tremarctos ornatus) habitat in the equatorial dry forest study region of Peru. Values for each 1km pixel are predicted using model-averaged effect sizes from both Cerro Venado and Laquipampa Wildlife Refuge surveys and remotely-sensed habitat data available for the region using the equation {ln(β0 + β1(nearest road) + β2(slope) + β3(forest cover) + β4(elevation)} (Manly et al., 2002). ................................................................25 Figure 2.6. Regional predictions for Andean bears and land cover comparison for the equatorial dry forest ecoregion in Peru, where details of numbered areas show a) density predictions; b) land cover; and c) satellite imagery. Areas 1,2 and 3 show areas of high predicted density that may have high co-occurrence with small-holder agricultural areas far from roads. Area 4 shows a comparative high density area with low predicted agriculture cover. The bar chart shows the predicted density breakdown for each land cover type region-wide, in increasing order of predicted suitability. ..................................................................................................................................26 Figure 3.1. North Andes study area overlaying digital elevation model (DEM) of South America. Andean bear range is indicated in white. ...................................................................................34 Figure 3.2. Biodiversity feature layers used in spatial prioritization; a) overall amphibian species richness; b) threatened amphibian species richness; c) overall bird species richness; d) small-ranged threatened bird species richness; e) threatened bird species richness; f) overall mammal xiii species richness; g) threatened mammal species richness; h) overall reptile species richness; i) overall water risk by sub-basin (Gassert et al., 2014); and  j) ecoregions (Olson et al., 2001). ..36 Figure 3.3. Cost layers used in prioritization ..............................................................................38 Figure 3.4. Selected priority areas in prioritisation scenario 1, where all features were given equal targets. Targets were set at 17%, 30%, and 50% levels; shown here overlaid because planning units selected for 17% targets are always selected again for higher target level. ......................43 Figure 3.5. Composite of least-cost corridors connecting existing core protected areas that intersect the known range of Andean bears. Corridors only considered for nearest neighbouring protected area to each protected area, and are shown here truncated at a total cost-weighted distance of 200 000 km. ............................................................................................................46 Figure 3.6. Representation of auxiliary features in single-feature prioritization with targets at 17%, 30%, 50%, and 100% of the total feature area; where a) uses Andean bear range; and b) uses a random uniformly generated feature covering the same total area as A ....................................47 Figure 3.7. Representation of auxiliary features in single-feature prioritization with targets at 17%, 30%, 50%, and 100% of the total feature area; where a) uses Andean bear range; and b) uses a random uniformly generated feature covering the same total area as Andean bear range. .......48 Figure 3.8. Combined results showing 1) existing core protected areas; 2) priority areas selected when 17% targets were applied to all features; and 3) least-cost corridors truncated at a total cost-weighted distance of 200 000 km. .............................................................................................50 Figure D.1. Predicted distribution of Andean bears (Tremarctos ornatus; ≥ 4/100 km2) in equatorial dry forest of northwestern Peru. Values for each 1km pixel are predicted using model-averaged effect sizes from both Cerro Venado and Laquipampa Wildlife Refuge surveys and remotely-sensed habitat data available for the region using the equation {log(β0 + β1(nearest road) + β2(slope) + β3(forest cover) + β4(elevation)} (Manly et al., 2002). ...........................................80     xiv Acknowledgements First, I would like to extend my most sincere gratitude to Dr. Peter Arcese, whose optimism, philosophy, occasional jam sessions, and brilliant big picture thinking have kept me buoyant in the murky waters of applied conservation science these last few years.  I was once a bright-eyed first year undergraduate looking for some career advice, and it was actually Peter who inspired me to pursue the field of conservation biology in his “Conservation 101” class at UBC. From one pivotal conversation (topic: why bother?) to the completion of this thesis, it has truly been a privilege working with him. My start in this field would also not have happened without Merle Crombie and Kate Johnson, who first hired me into the Arcese lab to work a memorable summer fending off seagulls on Mandarte Island, and who have provided invaluable support, advice, and laughs ever since. I’ve been very lucky to have a brilliant cohort of students and friends around me in Forestry, including Hannah Visty, Cora Skaien, Jessica Krippel, Devin DeZwann, Roseanna Gamlen-Greene, and many others. My friends have been my family in the last three years, and I’m grateful to have so many to thank, including the most wonderful crew I conned into befriending me over at the Institute of Fisheries and the Biodiversity Research Center. Aaron Purdy, Madeline Cashion, Sarah Fortune, and Natalie Mahara in particular, thanks for helping me through the stickiest spots. Manspie Brewing, thanks for nothing.  There were several points that I was genuinely concerned that this project might be beyond my abilities, and I have many people to thank for helping me overcome those hurdles. Most notably, Richard Schuster, who is a data wizard, provided me with code, support, and helpful suggestions on many components of this work. Joanna Burgar helped me troubleshoot my models when they weren’t working, and Cole Burton, Angela Fuller, Alex Pulwicki, Carolyn King, and Kurt Trzcinski also offered invaluable statistical or structural advice. Thank you also to Santiago de la Puente for helping with Spanish translations.  This thesis was made possible largely because of the hard work and dedication of the team at the Spectacled Bear Conservation Society, who have been in Batangrande, Peru for more than 10 years collecting data, building trust in the community, and campaigning tirelessly for the conservation of bears - sometimes through literal fire and flood. ¡Gracias a todos! I am especially grateful to Robyn Appleton for trusting me to work with her data.  xv I also want to acknowledge my funders. My work was supported by the Natural Sciences and Engineering Research Council of Canada (NSERC) and by grants from UBC’s Faculty of Graduate Studies and Department of Forest and Conservation Sciences. Dr. Peter Arcese provided funding from his NSERC, and helped in my successful application for an NSERC CGS–M (2016). Lastly, I want to thank my wonderful, loving parents, Bob and Margot, for having my back through all kinds of craziness and for showing me I don’t have to walk the straight and narrow to be happy. And, for being hilarious and making me happy - my niece, Kasia, my nephew, Kallen, and of course Jeff the dog, who let me hug him even though he was skeptical.   xvi Dedication For the bears               1   Chapter 1. General introduction 1.1 Context in conservation Global declines in large terrestrial vertebrates are widely-reported and often linked to over-harvest and habitat fragmentation and loss (Dirzo et al., 2014). To curb extinction and protect ecosystems, environmental planners employ a variety of tools, including habitat protection (Dinerstein et al., 2017), community engagement  (Ban et al., 2013), and enforcement and monitoring (Keane et al., 2008), as well as analytical tools designed to optimize conservation outcomes given constraints. Establishing protected areas is one of several approaches to combat extinctions that can sustain threatened species and ecosystems, especially when they are large, well-managed, connected to other protected habitats at large spatial scales, and distributed across diverse and representative range of habitats (Gray et al., 2016). As a consequence, the 1992 Convention on Biological Diversity (CBD) commits 177 signatory countries to increase the extent of effectively managed, protected area to 17% by 2020, by emphasizing “areas of particular importance for biodiversity” (Target 11; Secretariat CBD 2010: 9).  However, this target remains largely unmet due to lack of funding, political instability, and limits on ecological knowledge and enforcement (Wilson, Carwardine and Possingham, 2009; Dinerstein et al., 2017). This is particularly true where conflict between wildlife and humans exposes potential trade-offs between conservation and human well-being (McBride et al., 2007; Wilson et al., 2007). It is therefore imperative to find opportunities to conserve habitats likely to maximize the persistence of at-risk species while minimizing impacts on human well-being.  In areas such as the Northern Andes, the degradation of watersheds poses a direct threat to wildlife, but also to food production, drinking water and hydroelectricity for humans (Peyton, 1999).  To the degree that protecting or restoring habitat can maintain or increase such services to local communities and also address biodiversity conservation, the resulting synergies may offer incentives to find common solutions (Ban et al., 2013).  High endemic species richness and rates of land conversion by humans combine to make the Northern Andes a global biodiversity hotspot (Myers et al., 2000). The region spans from Venezuela to the Bolivia-Argentina border, covering 38 degrees of latitude (11° N to 27° S) in a strip 200-700km wide. Although the entire region is classified as the tropical Andes, it comprises 44 ecoregions, including montane cloud forests, equatorial dry forests, and alpine grassland, or páramos (Olson et al., 2001). Compared to other tropical hotspots, the Northern Andes has not 2 been a focus of conservation planning (Rodríguez and Young, 2000; Sierra, Campos and Chamberlin, 2002), and no planning efforts have considered the entire range. Given ongoing climate change and growing human populations, there is an urgent need to identify key areas for the conservation of species and ecosystem values identified by local and international experts (Kintz, Young and Crews-Meyer, 2006).  With the challenge of balancing limited resources and immense, sometimes competing conservation goals, numerous approaches have arisen to identify and prioritize conservation areas (Flather et al., 1997; Margules, Pressey and Williams, 2002). The rise of systematic planning approaches that use advanced computer algorithms to find optimal solutions has drastically increased the feasibility of incorporating multiple goals into land use planning (Margules and Pressey, 2000; Ball, Possingham and Watts, 2009). In particular, integer linear programming techniques have emerged as a way to quantify trade-offs and solve complex conservation planning problems in a fraction of the time of common optimization algorithms (Beyer et al., 2016). However, while these algorithms make it possible to efficiently identify the configuration of planning units that fulfill targets for representation of diverse features, identifying features and a scale in which to focus efforts remains a point of debate (Schwartz, 1999).  One common approach is to use a charismatic focal species as a vehicle to capture biodiversity values and also enjoin local or international communities in conservation funding or action (Bowen-Jones and Entwistle, 2002). In theory, effective conservation for focal species with large area requirements should have a positive spill-over effect on species that co-occur in those habitats (Caro, 2003; Thorne, Cameron and Quinn, 2006; Breckheimer et al., 2014). Although widely applied in conservation planning, focal species approaches are also criticized if narrowly focused because many species are limited by factors not considered when planning for a single species (Caro, 2003; Roberge and Angelstam, 2004; Wang et al., 2018). Bennett et al. (2015) suggested that focal species approaches were highly effective at securing new private funding, but returned biodiversity gains only when focal species shared habitats and threats with species that would otherwise not receive conservation actions, henceforth referred to as “beneficiary” species. In sum, careful selection of focal species is an important factor in their success in a broader conservation planning framework.  Bears are charismatic species and often a focus of conservation planning, including: the Kermode subspecies of the American black bear, or “spirit bears” (Ursus americanus kermodei; Howlett et al., 2009), polar bears (Ursus maritimus; Peacock et al., 2011), giant pandas (Ailuropoda melanoleuca; Wang et al., 2018), and grizzly bears (Ursus arctos; Nielsen, 2011). 3 The large home range requirements of bears make them ideal candidates as umbrella species (Servheen, Herrero and Peyton, 1999; Carroll, Noss and Paquet, 2001; Nielsen, 2011; Wang et al., 2018), but the approach has yet to be applied in South America. It has been suggested Andean bears (Tremarctos ornatus (F.G. Cuvier, 1825)) occupy a wide range of habitats throughout the Northern Andes, and may therefore qualify as a potential umbrella species for the region (Peyton et al., 1995; Peralvo, Cuesta and van Manen, 2005; Velez–Liendo, Adriaensen and Matthysen, 2014), but to date this assumption has not been formally tested.  A central theme of landscape ecology is the effects of spatial scale on ecological processes and their observation (Orians and Wittenberger, 1991; Macarthur et al., 1992; Wheatley and Johnson, 2009). Animals typically choose habitats within their range based on the availability of patches that provide resources, and make subsequent decisions about patch use based on local scale characteristics, such as ability of the habitat to support foraging or breeding (Orians and Wittenberger, 1991). Relationships between species abundance and habitat variables often change depending on the scale being considered (Apps et al., 2001), and as such, analyses of habitat use should always be run at a scale applicable to management (Elith and Leathwick, 2009; Toews, Juanes and Burton, 2017). However, global and regional goals for conservation require that planning take place at the scale of landscapes and regions. The mismatch between studies conducted at a scale useful to management and those at a scale useful to broad conservation goals represents a key knowledge gap in understanding species-habitat relationships, particularly in systems such as the Northern Andes, where data are limited. Building knowledge at multiple-scales can uncover novel information on species biology, and represents an important first step in moving conservation planning exercises out of the esoteric and into real-world implementation.  1.2 Andean bears as a focal species Andean bears, also known as spectacled bears, are a short-faced, mostly herbivorous species that occupies a wide range of habitats in the Northern Andes, including dry thorn forest, high-altitude cloud forest, steppe, and páramos (Figure 1.1. Example camera trap photo of an Andean bear (Tremarctos ornatus) from Cerro Venado field site, Lambayeque, Peru (SBC-Peru 2017).; Peyton et al., 1995). The mapped Andean bear range covers only 3% of the land area of South America, but ostensibly shares habitat with as many as 76% of all South American vertebrates (Peyton, 1999). High rates of habitat conversion to commercial or subsistence agriculture, and hunting and illegal trade in bear parts have resulted in the Andean bear being listed as vulnerable to extinction (Goldstein et al., 2008; García-Rangel, 2012). The Andean bear’s 4 requirement for large home ranges therefore make it a candidate umbrella species for the conservation of hundreds of endemic plants and animals (Peyton, 1999) and the winter ranges of many North American Neotropical migrant birds (Schuster et al., 2018). This in combination with their iconic status in the region make Andean bears an ideal focal species for landscape-scale conservation plans (Kattan et al., 2004). In addition, because the mapped range of Andean bears overlaps thousands of water catchments that provide critical water provisioning services to humans, potential synergies may also exist to facilitate conservation gains that simultaneously benefit human well-being, particularly given regional trends in climate (Vera et al., 2006) and deforestation (Harden, 2006).  Figure 1.1. Example camera trap photo of an Andean bear (Tremarctos ornatus) from Cerro Venado field site, Lambayeque, Peru (SBC-Peru 2017). Despite their potential utility, little action has been taken to incorporate knowledge on the distribution models for Andean bears in systematic conservation planning. Challenging terrain and political environments, and their naturally elusive behavior has also rendered Andean bears as one of the least-known bear species globally (Peyton et al., 1995; García-Rangel, 2012).  A current inability to estimate their abundance or predict their distribution therefore represents a key knowledge gap (Garshelis, 2011).  5 Methods for estimating Andean bear population density are varied, ranging from studies that extrapolated North American black bear densities over the range area (Peyton 1999) to those that use camera trap and telemetry data in a small-scale capture-recapture framework (Ríos-Uzeda, Gómez and Wallace, 2007). For planning purposes, a major shortcoming of these early estimates was that they assumed a uniform density and occupancy probability across the species range. Early attempts to bridge such gaps involved modifying density estimates developed for North American black bears. For example, Yerena & Torres (1994) used topographic data and models developed for North American black bears to identify potential corridor habitat in an expanded reserve system for Andean bears in Venezuela. More recently, Kattan et al. (2004) analyzed satellite imagery to determine the degree of fragmentation in Andean bear habitat to predict the long term viability of Andean bear sub-populations.  It is only with recent methodological advancements that conservation planners have begun developing spatial models of the occurrence and density of focal species based on their multivariate response to various mapped landscape features potentially affecting habitat distribution and quality. Such ‘species distribution models’ (SDMs) are typically regression-based, numerical models that combine observations of species occurrence or abundance with environmental data to estimate linear combinations of factors which identify sites more or less likely to support the species of interest (Elith and Leathwick, 2009). However, there are a number of distinct modelling methods used to create SDMs, the use of which depends on the nature of the species data available (Elith et al., 2006). For example, given ‘presence-only data’ (e.g., sign, sightings, or historic records of occurrence at a site), Phillips et al. (2004) recommend a maximum-entropy (maxent) approach to distribution mapping, but often at the expense of significant sampling biases due to the lack of known ‘absences.’ Surveys and data that include records of species absence can help minimize such biases, especially when also taking into account factors affecting the probability of detecting individuals, given that they are present at the sample site (Royle et al., 2005; Schuster and Arcese, 2013). However, such data are not always available for cryptic, rare, or little-known ‘species at-risk’ (Peterman, Crawford and Kuhns, 2013). Most recently, camera trap surveys have enabled the identification of individual animals in addition to simple presence-absence, facilitating the development of spatially explicit capture-recapture (SCR) models capable of producing detailed estimates of density and abundance across environmental or other feature gradients mapped over the landscape of interest (Otis et al., 1978; Royle et al., 2014). In all such cases, careful selection and sourcing of predictor environmental variables is critical for high quality outputs (Austin, 2002). 6  To date, only a handful of studies have attempted to develop species distribution models for Andean bears. In Colombia, Cuesta et al. (2003) recorded presence-only bear sign on transects and used those data to predict bear occurrence based on habitat variables derived from satellite imagery. They estimated an index of habitat suitability and suggested that bears avoided roads and displayed seasonal differences in habitat use. Two studies have also applied occupancy modelling: one using historic presence-only data (Venezuela; Sánchez-Mercado et al., 2014), and a similar study in Peru (Figueroa, Stucchi and Rojas-Verapinto, 2016). Most recently, Molina et al. (2017) reported a species distribution model for Andean bears in Ecuador based on presence-absence data for individuals, and developed in an SCR framework. Molina et al. highlight the potential of SCR modelling by demonstrating an ability to infer habitat use. It is widely believed that Andean bears preferentially occupy high-altitude cloud forest and páramo ecosystems (Peyton et al., 1995; Cuesta, Peralvo and Van Manen, 2003; Peralvo, Cuesta and van Manen, 2005; García-Rangel, 2012). However, detailed studies of individually-identified bears indicates that Andean bears can exist at relatively high densities in other montane ecoregions, such as equatorial dry forest (Appleton et al., 2018), which is recognized as a globally significant ecoregion given high endemism and diversity (Linares-Palomino et al., 2010).  Although deforestation and agricultural expansion has reduced or fragmented much of the original extent of dry forest, no studies other than Appleton et al. have endeavoured to estimate the distribution or abundance of Andean bears in this region.  1.3 Rationale and research approach I set out to evaluate the use of Andean bears as a focal species in conservation planning in northern Peru, and by extension to the Northern Andes ecoregion. If Andean bears are to be effectively incorporated into the conservation planning process as a focal species, better spatial characterization of their distribution and habitat use in all regions of the Andean landscape is necessary. Specifically, I built density models that relate locations of individual bears from camera trap data to habitat variables from remote sensed data, in order to: a) improve habitat and range-wide distribution maps for Andean bears, and b) identify priority conservation areas for bears that may also protect Andean endemic species across ecoregions. By combining these goals with a further goal of protection of critically important water catchments, I also identify potential synergies between conservation planning and human well-being, of interest to international organizations whose goal is maximizing return on conservation investments by also securing water for human agriculture and population centers.  7 Successful regional planning for focal species requires that data inputs on ranges and distribution are refined to include limiting resources and their potential effect on habitat selection (Elith and Leathwick, 2009). Spatially-explicit habitat models derived from empirical data at small spatial extents provide a preliminary means to account for future supply of habitat better than delineating by habitat-type alone, and can help decision makers ‘scale up’ to regions and landscapes, especially if combined with multi-scale studies from the same area (Toews, Juanes and Burton, 2017).  The aim of my thesis is to use species distribution models (SDMs), existing field and remote-sensed data, and spatial planning tools to optimize conservation plans for Andean bears. In Chapter 2, I use camera-trap data at two well-sampled cloud forest and equatorial dry forest sites (Laquipampa, Cerro Venado, Peru) and construct spatial density models that relate observations to habitat characteristics. I test whether forest cover, topography, road density, distance to roads, or climate influenced the density or distribution of Andean bear home range centres in dry forest habitats of northwestern Peru. My objectives for this chapter are 1) to develop a common framework for estimating Andean bear habitat use in the Northern Andes with the best available data; 2) speculate on the potential historical, present-day and future distribution of Andean bears in northwestern Peru; 3) ask how my results compared to an IUCN range map for Andean bears (Goldstein et al., 2008) and correspond to the distribution of existing protected areas identified by the World Database on Protected Areas (WDPA; UNEP-WCMC & IUCN 2016).   In Chapter 3, I employ systematic prioritization methods to explore alternate land use plans likely to overlap with the most species and the most intact and persistent ecosystems and watersheds in the Northern Andes. My overall objective is to deliver a multi-species GIS decision support tool that can be used to prioritize, and explore the potential to conserve, landscapes likely to enhance conservation. Specifically, to 1) prioritize areas within the Andean bear range to maximize biodiversity and minimize the potential impact of roads; 2) identify least cost paths between existing protected areas that may represent potential corridors for the dispersal of Andean bears and by extension, other large carnivores; and 3) ask what fraction of these potential priority areas are currently contained within protected areas and, more generally, determine the coverage of other species ranges and ecoregions within the Andean bear range and proposed priority areas. Given that species distribution models for Andean bears have so far been limited in spatial extent, my study represents a first attempt to scale-up such models to a regional extent to explore the influence of roads, seasonality, and several factors thought to affect primary production on the predicted distribution and abundance of Andean bear habitat. Because the 8 models I developed relied largely on prior relationships reported for Andean bears across much of the IUCN Andean bear range, my goal was to represent and develop further testable hypotheses that could be applied at multiple scales in the future. Overall, by systematically characterizing Andean bear habitat use, I aim to identify conservation areas in the Northern Andes likely to maximize the persistence of Andean bears, while simultaneously protecting the largest number of migrant and endemic species, and water catchments.  9 Chapter 2. Human influence and the density of Andean bears (Tremarctos ornatus) in dry forests of the northern Andes 2.1 Introduction Habitat destruction and degradation by humans are pervasive factors that affect the movement and distribution of wide-ranging species (Tucker et al., 2018). An often unavoidable consequence of human expansion into wildlife habitat is an increase in human-wildlife conflict, especially with large mammals, such as bears, that range widely to exploit seasonal foods (Woodroffe, 2000; Treves and Karanth, 2003). The expansion of human-dominated landscapes, agricultural frontiers, and natural resource development has prompted efforts to reduce such conflict by engaging in strategic conservation planning at landscape scales (Cushman, McKelvey and Schwartz, 2009; Chester, 2015). However, existing efforts have yet to identify or protect sufficient corridor or high quality habitat to insure the persistence of large carnivores globally (Woodroffe et al., 2007; Newmark, 2008; Laurance et al., 2012; Xun, Yu and Wang, 2017) which depend in part on their ability to coexist with humans at the boundaries of anthropogenic development (Takahata et al., 2014).  As a consequence, there is an urgent need to develop spatial tools to reliably predict the density and abundance of free-living carnivore species and their response to human development and land use. Identifying what factors limit species density and abundance can be facilitated by testing their response to factors hypothesized to attract or repel individuals. In theory, habitat use is thought to be influenced by a variety of factors linked to individual phenotype and fitness, such that higher quality habitats are assumed to support higher densities of individuals and more persistent populations (Morris, 1987; Dahle and Swenson, 2003). For wide-ranging carnivores, habitat quality is often characterized as the product of several interacting factors linked to seasonality, prey abundance, and interactions with humans (e.g., Carroll et al., 2001; Fisher & Burton 2018). In the absence of detailed spatial data on such factors, edaphic and climatic variables are often adopted as proxies for habitat quality (Clark, Palmer and Clark, 1999; Cuesta, Peralvo and Van Manen, 2003; Sánchez-Mercado et al., 2014). For example, Carroll et al. (2001) and Homeier et al. (2010) characterized ecosystem productivity using a combination of variables linked to seasonality, elevation, and slope in montane habitat. Forest cover is also used widely to predict the distribution of refuge habitats and food availability (Fernández et al., 2012; Hansen et al., 2013; Velez–Liendo, Strubbe and Matthysen, 2013). As a consequence, where such data are available and relatively precise, it is also possible to develop hypothetical models of habitat use 10 to further understanding about the habitat preferences and demography of rare and little-known species for which high-quality habitat maps are urgently needed, but unavailable. We provide such an example here by developing testable hypotheses about the influence of human development, edaphic and topographic factors on the abundance and distribution of Andean bears (Tremarctos ornatus) in a global biodiversity hotspot.  Human development is often characterized by expanding road networks (Venter et al., 2016) which are often used to estimate human impacts on wide-ranging species (Forman and Alexander, 1998; Basille et al., 2013; Červinka et al., 2015). For example, survival in bears tends to decline in areas of overlap with humans, such as near roads (Doak, 1995), making bears a potential indicator of the likely influence of roads on other wide-ranging carnivores (Schoen, 1990; Rajpurohit and Krausman, 2000; Falcucci et al., 2009). The building of new roads, or access management on existing roads, are tools available to land managers that can have broad consequences, and thus a good candidate for action (Forman and Alexander, 1998). However, whereas  associations between roads and bears have been quantified in parts of North America (Nielsen, 2011; Simek et al., 2015; Lamb et al., 2018) and Eurasia (Rajpurohit and Krausman, 2000; Takahata et al., 2014; Scotson et al., 2017), they remain largely unexplored in South America and have never been quantified precisely for Andean bears. A better understanding of how roads influence habitat use and species distribution could therefore advance the identification of key conservation areas and actions substantially (Margules and Pressey, 2000; Bennett et al., 2015).  Andean bears are thought to have declined dramatically over their range in South America due to habitat loss and human hunting, but the challenging political and geographic landscapes they occupy and their elusive behavior have made them among the least-known large mammal species (Peyton et al., 1995; Goldstein et al., 2006). It is widely believed that Andean bears preferentially occupy high-altitude cloud forest and páramo ecosystems (Peyton et al., 1995; Cuesta, Peralvo and Van Manen, 2003; Peralvo, Cuesta and van Manen, 2005; García-Rangel, 2012). However, detailed studies of individually-identified bears indicates that Andean bears can exist at relatively high densities in equatorial dry forest (Appleton et al., 2018), an area recognized globally as a critically-endangered biodiversity hotspot but largely outside of the described range of Andean bears (Janzen, 1988; Hoekstra et al., 2005; Linares-Palomino et al., 2010).  Given their potential role as an umbrella species (Bennett et al., 2015), their wide-spread promotion in conservation and education (Yerena 1998; Espinosa & Jacobson 2012), and a near-absence of data on their use of dry forests (Appleton et al., 2018), Andean bears offer an outstanding 11 opportunity to improve understanding of such forests and test the hypothesis that increasing road density has contributed to their reduced distribution in the face of human development in lowland forests of northwestern Peru.  To date, studies of habitat use by Andean bears have been limited to sign surveys (Cuesta, Peralvo and Van Manen, 2003), presence-only data (Velez–Liendo, Strubbe and Matthysen, 2013; Figueroa, Stucchi and Rojas-Verapinto, 2016), limited camera trap data (Rios-Uzeda, Gomez and Wallace, 2006; Molina et al., 2017), or based on home range estimates for a small number of radio-collared bears (Castellanos, 2011). However, recent methodological advances suggest more precise models of density for Andean bears can be developed using a sufficient number of observations of the presence and absence of individual animals in systematic surveys (Garshelis, 2011; Molina et al., 2017).  In this chapter, we used multi-year camera trap data and spatial capture-recapture models to test how individually-identified Andean bears were influenced by roads and a suite of edaphic and topographic variables potentially indicative of food resources in tropical dry forest habitats. To do so, we first evaluated local and global digital data sources to identify feature layers which have provided sufficient to predict the densities of bears in Asia (Scotson et al., 2017), North America (Nielsen et al., 2010; Simek et al., 2015), and South America (Cuesta, Peralvo and Van Manen, 2003; Sánchez-Mercado et al., 2014) including forest cover, elevation, slope, distance to roads, and climate. We then employed spatially-explicit capture-recapture (SCR) models to estimate the density of Andean bears and the response to these variables in our dry forest study area. Last, we developed a hypothetical spatial model to explore the potential application of our local results to regional scales by: a) projecting the influence of roads and environment on predicted bear density throughout the tropical dry forest ecoregion; b) testing whether our predictions varied as expected given published accounts of habitat use, and c) identifying short-comings of our local model applied to large spatial scales by examining qualitatively how areas predicted to support low and high densities of bears differ in digital and visual representations of the landscape. Following Yerena & Torres (1994), we used the suggested lowest density estimate of 4 bears/100 km2 for Andean bears to delineate areas capable of supporting realistic densities of Andean bears. We next compare our hypothetical distribution map with an IUCN range map based on expert elicitation (Goldstein et al., 2008), and to the distribution of existing protected areas (World Database on Protected Areas, WDPA; UNEP-WCMC & IUCN 2016) to identify potential gaps in existing knowledge of Andean bear distribution.  Based on literature and experience in the region (Appleton et al., 2018), we also made several a priori predictions about 12 factors potentially affecting the abundance and seasonal distribution of Andean bears in tropical dry forest habitats, including that density will be: i) positively related to seasonal food availability as predicted by topography, climate, and forest cover, but ii) decline with proximity to human influence.  2.2 Methods 2.2.1 Study area All observations of bears were collected in mountainous terrain of the Lambayeque region adjacent to Rio La Leche, c. 60 km northeast of Chiclayo, Peru (Figure 2.1; Appleton et al., 2018).  The study area comprised ~607 km2 of equatorial dry forest (140 -1800 m elevation a.s.l) where the average road density is 0.10 km/ km2, and ~61 km2 of transitional dry and montane cloud forest (1800 m and 2600 m a.s.l.) where the average road density is 0.17 km/km2.  Equatorial dry forest occurs in a narrow band between the coast and western slopes of the Andes in northern Peru and Ecuador and is a globally significant biodiversity hotspot given high endemism and diversity (Linares-Palomino et al., 2010). However, deforestation and agricultural expansion has reduced and fragmented much of its original extent.  Regional climate is dominated by the El Niño-Southern oscillation (Rodríguez et al., 2005) and mean annual rainfall (100-2000 mm) increases with elevation and distance from the coast (Linares-Palomino et al., 2010). Precipitation is highly seasonal, falling almost exclusively in November to May. Vegetation cover is dominated by trees in the Fabaceae family (e.g., algarrobo; Prosopis spp.), but can include high relative abundances of hualtaco (Loxopterigium huasango), palo santo (Bursera graveolens), pasallo (Eriotheca ruizii), overo (Cordia lutea), sapote (Capparis scabrida), and cactus (Cactacaeae). Sapote is a particularly significant component of tree cover in this region because its fruits make up a large fraction of the Andean bear diet seasonally (Appleton et al., 2018) but is critically endangered in Peru (Rodríguez et al., 2007). 13 2.2.2 Field protocol We conducted camera-trap surveys at Cerro Venado and Laquipampa Wildlife Refuge (Figure 2.2). At Cerro Venado, we surveyed in the dry and wet season from June to October 2012 (2535 trap-days) and November 2012 to April 2013 (2552 trap-days), respectively.  We used infrared motion-activated cameras (Reconyx RapidFire® RM 45 and MC 65; Bushnell Trophy Cam™) deployed at 40 stations between 140 and 1300 m a.s.l. in an approximate 1km2 grid  that varied due to access (Figure 2.2). At Laquipampa, we surveyed over 4688 trap-days from August 2015 to March 2016, using 61 camera stations from 200 to 2600 m a.s.l. in an approximate 1km2 Figure 2.1. Location of the equatorial dry forest area used in this study and the Andean bear (Tremarctos ornatus) IUCN range (Goldstein et al. 2008) in northern Peru, South America.  14 grid that varied somewhat due to access (Figure 2.2).  All stations had two cameras positions c. 50 cm above ground to photograph the bears’ face and chest, running 24 hours/day to facilitate identification of individual bears from facial markings (Figure 2.3). Cameras were checked every 10-30 days to change batteries and ensure proper operation.  Figure 2.2. Camera trap locations for surveys conducted 2012-2016 in the Lambayeque region, Peru 15  Figure 2.3. Example camera trap photos from Cerro Venado and Laquipampa showing unique facial markings of Andean bears (SBC-Peru 2017).  Andean bears undertake altitudinal migrations in response to seasonal fruit availability in our study area (Appleton et al., 2018; see also Cuesta et al., 2003; Castellanos, 2011; Molina et al., 2017), causing us to subset data into two, four-month periods representing the wet (June to October) and dry seasons (late November to April), respectively; and to satisfy the modelling assumption of demographic closure (i.e. no mortality, recruitment, or dispersal). Individual bears were identified based on unique facial markings which remain stable over a bear’s life (Van Horn et al., 2014, 2015), and photographs of bears where markings could not be observed were not included. Photographs of an individual bear were considered independent detections if separated by more than 12 hours.  2.2.3 Habitat data We used 4 spatial variables from remotely-sensed satellite data to assess habitat condition, by aggregating the data within 1km2 pixels and extracting to each camera trap location with a geographic information system (GIS; Clark et al., 1993; Table 2.1). The variables - elevation, slope, forest cover, and distance to roads - were used for a resource selection function in spatial capture-recapture models to estimate habitat use by Andean bears at Cerro Venado and Laquipampa and selected based on the literature and local knowledge (Appleton et al., 2018). We selected elevation as a variable to capture productivity and therefore food availability instead 16 of the normalized difference vegetation index (NDVI) or precipitation because these variables were very closely correlated to elevation but lower in resolution (Pearson’s correlation coefficient|>0.7|; see appendix A). Two principal components extracted from climate data for the region (Hamann et al., 2013) were also closely related to elevation (Pearson’s correlation coefficient |0.99|; see appendix A), but performed worse than elevation on its own when substituted into initial models. Thus, because elevation is easily interpreted and extrapolated to larger scales, we used elevation instead of climate and NDVI in final analyses. We selected slope as an additional topographic variable because positive associations with steep terrain are well documented for this species (Goldstein, 2002; García-Rangel, 2012). We selected forest cover as a proxy for food and shelter, given the importance of trees as nest, den, and feeding sites (Velez–Liendo, Adriaensen and Matthysen, 2014; Appleton et al., 2018). This layer, expressed as percent canopy cover, was derived from a global dataset based on 30m resolution Landsat imagery from Hansen et al. (2013). We did not use land cover data in our models because of the poor resolution and reliability of land cover data available for the region.  Given the well-documented and mainly deleterious effects of roads on the behavior, movement and survival of large mammals (Basille et al., 2013; Ordiz et al., 2014; Tucker et al., 2018), including bears (Doak, 1995; Kattan et al., 2004), we used distance to any type of road as a proxy for anthropogenic land use, following numerous studies (Rice et al., 2009; Sánchez-Mercado et al., 2014; Simek et al., 2015; Stillfried et al., 2015). While we acknowledge there is likely some variability in effect depending on the type of road, here we assume all road types equal, as all roads in the study area were secondary roadways.  For modelling, we standardized all variables using mean and standard deviation from the broader dry forest region. All spatial variables were developed in ArcGIS (ESRI, 2011) and R version 3.3.3 (R Core Development Team, 2014).    17 Table 2.1. Variables selected to determine Andean bear habitat availability and their ranges in continuous rasters of the Cerro Venado and Laquipampa study areas, Lambayeque, Peru, 2012-2016. Range values are shown in the original units before standardization for modelling.  Variable Range in study areas Range at camera locations Data source Elevation (m a.s.l.) 112 m – 2942 m 404 m – 2576 m NASA Shuttle Radar Topographic Mission (SRTM) 90m Digital Elevation Model (DEM) Slope (degrees) 0˚ - 65˚ 2˚ - 49˚ Derived from NASA Shuttle Radar Topographic Mission (SRTM) 90m DEM with ArcGIS 10 Slope tool Forest cover (percent) 0% - 100% 0% – 95% Hansen et al., 2013 Distance to nearest road 0.0 km – 12.9 km 1.0 km – 10.4 km Derived from Open Street Map Database with ArcGIS 10 Near tool  2.2.4 Data analysis We used a spatial capture-recapture model (Royle et al., 2014) to estimate Andean bear density and individual resource selection. Spatial capture-recapture models assume that individuals occupy bivariate normal home ranges centered on activity centers (si) that are distributed over a two-dimensional plane referred to as the state-space (Si). Data are modeled in hierarchical nested models, where an observation model using the observed data is conditional on an activity center model. Observations (y) of individuals (i =1, 2, …, I) at sites (j = 1, 2, …, J) are assumed to be a binomially distributed function of the number of sampling occasions (Kj) and the encounter probability for each individual (pij). Encounter probability is a bivariate normal model that depends on the Euclidean distance between the trap location (xj) and the modelled activity center (si).  Baseline capture probability (p0) and the movement parameter (σ) control the shape of the detection function, and we varied these by both the sex of the individual and the season of capture.   𝑦𝑖𝑗  ~ 𝐵𝑖𝑛𝑜𝑚𝑖𝑎𝑙(𝐾𝑗, 𝑝𝑖𝑗)  where: 𝑝𝑖𝑗 = 𝑝0  × 𝑒−12𝜎2‖𝑥𝑗−𝑠𝑖‖2  18 We tested all combinations and interactions of sex and season for each dataset, and selected the best fitting model for each dataset based on Akaike’s Information Criterion (AIC) values, under the assumption that these two variables account for much of the variation in the density and distribution of home ranges (Castellanos, 2011). Home ranges for Andean bears are typically larger for males and during the wet season (Castellanos, 2011). With sex and season thus controlled for, this top observation model was used as the base model over which combinations of spatial variables and activity centers were modelled. In discrete space, the distribution of activity centers may vary in relation to gridded habitat variables. Individuals are hypothesized to center home ranges in areas of high quality habitat, therefore distribution of activity centers could be modelled as function of habitat variables. The activity center model is defined as a categorically-distributed function of the log-linear probability (π) that an activity center (si) is in a given pixel (g): 𝑆𝑖~𝐶𝑎𝑡𝑒𝑔𝑜𝑟𝑖𝑐𝑎𝑙(𝜋𝑔) Where: 𝜋𝑔 =𝜇(𝑠𝑖)∑ 𝜇(𝑠𝑖) And: ln(𝜇(𝑠𝑖, 𝐵)) =  𝛽0 + 𝛽1 ∗ 𝑓𝑜𝑟𝑒𝑠𝑡(𝑠𝑖) +  𝛽2 ∗ 𝑒𝑙𝑒𝑣𝑎𝑡𝑖𝑜𝑛(𝑠𝑖) + 𝛽3 ∗ 𝑠𝑙𝑜𝑝𝑒(𝑠𝑖) +  𝛽4 ∗ 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑡𝑜 𝑟𝑜𝑎𝑑𝑠(𝑠𝑖) Density is derived from these models as the expected number of activity centers in each 1km2 pixel, given the modelled distribution of activity centers in the state space. We defined the state-space as the outermost extent of the trap array, plus a 5 km buffer, based on the distance after which we found density estimates were no longer sensitive to state space size (Karanth and Nichols, 1998; Sollmann, Gardner and Belant, 2012).   We fit all combinations of habitat variables using maximum likelihood implemented through functions from the package oSCR (Sutherland, Fuller and Royle, 2015). We ran model sets for each survey separately as multi-session models with two seasons. We then also ran one combined model set with both surveys where data were pooled together, but still separated by season. Because there was no overlap in the individuals present in each survey area, we assume that the assumption of demographic closure was not violated by pooling the data in this way. We also tested for interactions between season and elevation, and season and distance to road, 19 because seasonality in habitat use at Cerro Venado brings bears to low elevations near human habitation when sapote and other fruits become available.  However, no model with interactions had an AIC value within the cumulative weight cutoff of 0.95; hence, we excluded interactions from further analysis. We assessed model fit using AIC, where smaller AIC indicated better model fit (Burnham & Anderson 2004). We identified top models as having a low AIC and within a delta AIC ≤ 2 of the top-ranked model (Burnham & Anderson 2004; Arnold 2010; Tables 2.3-2.5). Models with a ΔAIC ≤ 2 from the best-fitting model were considered well-supported, and those with a ΔAIC from 2 to 4 to have limited support. Effect sizes and standard errors were averaged by model weight and parameter estimates reported for all models with a cumulative weight of 0.95 (Burnham and Anderson, 2004). We estimated the mean size of an individual bear home range using a bivariate normal approximation calculated as {√(5.99) × 𝜎} following Royle et al. (2014). Model-averaged effect sizes were considered significant if 95% confidence intervals did not overlap zero.  Following Schuster and Arcese (2013), model averaged effect sizes were then interpolated using a resource selection function (RSF) over the study area in a continuous 1km x 1km grid to visualize the predicted habitat of Andean bears in each study area. We used the resource selection function equation {ln(β0 + β1(nearest road) + β2(slope) + β3(forest cover) + β4(elevation)} (Manly et al., 2002) to estimate the number of activity centres predicted to be in each 1km pixel, given the habitat conditions at each location. We then explored the result of extrapolating this resource selection function to other parts of equatorial dry forest where the habitat variables had similar ranges, assuming that if bears were present in these areas they would respond similarly to the habitat conditions as in the study area. We applied a density threshold of 4 individuals/100 km2 to delineate areas that were most likely to be able to support bears, based on the lowest probably density levels proposed for Andean bears by Yerena & Torres (1994). We used this layer to draw comparisons with the IUCN bear range (Goldstein et al., 2008), based mostly on expert elicitation and data from North American black bears, and with protected areas identified by the World Database on Protected Areas (UNEP-WCMC & IUCN 2016).   The lack of georeferenced location data on Andean bear density in this region made it necessary to validate the results of the regional predictions using a qualitative comparison with available land cover data (Defourney, 2015), and look for evidence that relatively high densities were predicted in areas of cover typically associated with Andean bear habitat, such as cloud forest, and that relatively low densities were predicted in agricultural and urban areas.  20 2.3 Results 2.3.1 Field surveys We identified 20 individual adult bears over 5087 trap-nights at Cerro Venado (Table 2.2) and 18 individual adult bears over 4688 trap-nights at Laquipampa Wildlife Refuge (Table 2.2). Out of 160 independent detections at Cerro Venado, 140 were recaptures of the same individuals. Out of 76 independent detections at Laquipampa, 58 were recaptures. Individuals were detected from 1 to 27 times overall, and identified at 1 to 10 trap locations, but no individual contributed more than 20% of detections.  Table 2.2. Camera trap survey carried out in Cerro Venado in 2012-2013 and Laquipampa in 2015-2016.  Survey Season Start Date End Date Stations Camera Days # of independent bear photos Individuals  N (M/F/Sex unk.) Cerro Venado Dry June 15th, 2012 October 25th,  2012 40 2535 66 15 (7/4/4) Wet November 25th, 2012 April 1st, 2013 40 2552 94 17 (7/6/4) Laquipampa Dry August 25th, 2015 October 25th, 2015 61 1697 29 13 (5/3/5) Wet November 25th, 2015 March 23rd, 2016 61 2991 47 16 (6/3/7) 2.3.2 Habitat effects on density For each study area and season, and each dataset analyzed alone or pooled, distance to road had the most strongly supported effect on bear density, or in other words the number of activity centers predicted to be in each 1km pixel. Effect coefficients always indicated an increase in bear density as distance to a road increased. Elevation, slope and tree cover also gained moderate support in most models, but their predicted effect on bear density varied across surveys (Tables 2.3-2.5; Figure 2.4). Density estimates from the top model for each dataset were 0.001 individuals per 1km2 pixel for Cerro Venado, 0.011 individuals per 1km2 pixel for Laquipampa, and 0.039 individuals per 1km2 pixel in the pooled dataset, although caution is warranted in these estimates because the inclusion of habitat variables may result in the underestimation of density (Sutherland, Fuller and Royle, 2015). This translates to an average of 3.9 bears /100km2 using the pooled dataset.  21 For Cerro Venado, the best model to explain spatial variation in density included distance to the nearest road and elevation (wi = 0.44; Table 2.3; see correlation matrix in Appendix A). Two other models received similar support (ΔAIC ≤ 2), and included elevation, distance to road, and tree cover or slope (wi = 0.20 and 0.19; respectively).  Distance to nearest road was included in all top models within a cumulative weight of 0.95, indicating that distance to road was a more consistent predictor of density than elevation, tree cover, or slope. Elevation was included in three of four top models, with a cumulative weight of 0.83. Slope and tree cover received less consistent support, but were each included in at least one top model (cumulative weight: 0.28 and 0.20, respectively). Estimates for models including tree cover were numerically unstable due to low variation in the tree cover covariate for Cerro Venado.  For Laquipampa, the best model included distance to nearest road, elevation, tree cover, and slope (wi = 0.53; Table 2.4). No other model received similar levels of support (ΔAIC ≤ 2). Distance to nearest road was included in all 7 top models within a cumulative weight of 0.95. Elevation was included in four of 7 top models, with a cumulative weight of 0.80. Slope and tree cover were each included in three and two top models with a cumulative weight of 0.71 and 0.68, respectively.  With surveys combined, the best model included distance to road, slope, and tree cover (wi = 0.43; Table 2.5).  Two other models received similar support (ΔAIC ≤ 2), one with elevation, distance to road, and slope; the second adding tree cover (wi = 0.20 and 0.17; respectively).  Distance to nearest road was again included in all six top models within a cumulative weight of 0.95. Slope and tree cover received more consistent support by combining surveys; each were included in four and three top models (cumulative weights 0.84 and 0.64, respectively). In contrast, elevation received less support with surveys combined, but and was included in two of six top models, with a cumulative weight of 0.37. In all three model sets, a base observation model with only parameters for sex and season had negligible support compared to the top model (evidence ratio of Akaike weights, wi).  Averaged effect sizes for all top models (cumulative wi = 0.95) indicate that adult bears occurred at higher densities as distance to roads increased (Cerro Venado (2.60 ± 0.68); Laquipampa (3.87 ± 1.22); Both (2.34 ± 0.41); Figure 2.4). Slope also had a negative effect on bear density with surveys combined (-1.26 ± 0.53; Figure 2.4). A different pattern was observed with respect to elevation. In Laquipampa, bear density increased with elevation (0.97 ± 0.41; Figure 2.4). Effects for elevation for Cerro Venado and across pooled datasets had 95% confidence intervals that included 0, and were therefore not significant. Nevertheless, density 22 increased slightly with elevation in the pooled datasets (0.70 ± 0.45; Figure 2.4) and decreased slightly at Cerro Venado, (-5.85 ± 2.99; Figure 2.4).  Contrasting patterns also prevailed with respect to tree cover, although estimated effect sizes for tree cover in each survey had 95% confidence intervals including zero. Nevertheless, bear density tended to increase with tree cover in Cerro Venado (1.61±1.29; Figure 2.4) and decrease with tree cover in Laquipampa (-2.32 ± 3.06; Figure 2.4). Combining surveys suggested a modest positive effect of tree cover on density (0.66 ± 0.34; Figure 2.4) with the 95% confidence interval just including zero.  Although seasonal differences were not detected in habitat effects through interactions, our model results suggested that home range sizes and baseline capture probability were substantially higher in wet season in both study areas (Cerro Venado, σ = 1.13/0.68 (wet/dry); Laquipampa, σ = 0.96/0.55 (wet/dry); see Appendix B). Across both data sets and sexes, averaged estimates for 95% home range area (based on sigma parameter in density models) were 53% larger in the wet season (31.50 km2) than dry season (20.58 km2). This difference was especially pronounced at Laquipampa, with a 173% increase in home range size from the dry (17.88 km2) to wet (48.96 km2) season. Male home ranges were substantially larger than female home ranges (pooled datasets, σ = 1.06/0.73 (M/F); see appendix B). Table 2.3. Model selection for spatial-capture recapture models to explain density of Andean bears (Tremarctos ornatus) at Cerro Venado. The best-fitting observation model (po ~ season*sex; σ ~season) is used in each model. K is the number of parameters in the model. ΔAIC is the difference in Akaike’s information criterion (AIC) between the best model and each successive model. w i is the Akaike weight, representing the probability that a model is the best in the model set. See Apprendix B for full results.  Density model K AIC ΔAIC wi cumulative wi elevation +  nearest road 10 1997.00 0.00 0.44 0.44 elevation + tree cover + nearest road 11 1999.00 1.60 0.20 0.64 elevation +  slope + nearest road 11 1999.00 1.70 0.19 0.83 slope + nearest road 10 2001.00 3.10 0.09 0.92 elevation + tree cover + slope + nearest road 12 2001.00 3.40 0.08 1.00 Null (~1) 8 2029.00 31.40 0.00 1.00    23 Table 2.4. Model selection for spatial-capture recapture models to explain density of Andean bears (Tremarctos ornatus) at Laquipampa Wildlife Refuge. Predictor variables are built into the activity center for density. The best-fitting observation model (po ~ 1; σ ~season*sex) is used in each model. See table 2.3 for definitions. Density model K AIC ΔAIC wi cumulative wi elevation + tree cover + slope + nearest road 11 998.00 0.00 0.53 0.53 elevation +  nearest road 9 1001.00 2.80 0.13 0.66 elevation +  slope + nearest road 10 1002.00 3.70 0.09 0.75 slope + nearest road 9 1002.00 4.40 0.06 0.81 elevation + tree cover + nearest road 10 1003.00 4.40 0.06 0.87 tree cover + nearest road 9 1003.00 4.50 0.06 0.92 nearest road 8 1003.00 4.90 0.05 0.97 tree cover + slope + nearest road 10 1004.00 5.60 0.03 1.00 Null (~1) 7 1024.00 25.90 0.00 1.00 Table 2.5. Model selection for spatial-capture recapture models to explain density of Andean bears (Tremarctos ornatus) across the combined datasets of Cerro Venado and Laquipampa Wildlife Refuge. The best-fitting observation model (po ~ 1; σ ~season + sex) is used in each model. See table 2.3 for definitions. Density model K AIC ΔAIC wi cumulative wi tree cover + slope + nearest road 9 3055.00 0.00 0.43 0.43 elevation +  slope + nearest road 9 3057.00 1.50 0.20 0.63 elevation +  slope + nearest road + tree cover 10 3057.00 1.80 0.17 0.80 nearest road 7 3059.00 3.60 0.07 0.87 tree cover + nearest road 8 3060.00 4.80 0.04 0.91 slope + nearest road 8 3060.00 5.00 0.04 0.94 elevation +  nearest road 8 3060.00 5.10 0.03 0.97 elevation + tree cover + nearest road 9 3061.00 5.60 0.03 1.00 Null (~1) 6 3091.00 36.20 0.00 1.00   24  Figure 2.4. Averaged, natural log-scale effect-size estimates (β coefficients) and 95% confidence intervals for top habitat variables explaining spatial variation in estimated density of Andean bears (Tremarctos ornatus). Effect sizes indicate the standardized number of units of change in the response variable for a one-unit increase in the predictor. Coefficients are averaged by model weight across models within the model set containing 95% cumulative model weight. See Appendix B for detailed results. 2.3.3 Regional predictions Model-averaged effect sizes projected across the 96 000 km2 equatorial dry forest study area suggested a strong east-west gradient in bear density, increasing with elevation and areas of lower human density, and with increasing elevation (Figure 2.5). Using the low density estimate of 4 bears/100 km2 as a threshold value to delineate likely and unlikely habitat, we identified 7349 km2 of potential bear habitat in the equatorial dry forest study area; corresponding to roughly 35% of the total study area (See Appendix D). Of this area, 16% (1162 km2) overlapped with the existing Andean bear IUCN range, and 3% (226 km2) overlapped with existing protected areas.  25  Figure 2.5. Predicted density of Andean bear (Tremarctos ornatus) habitat in the equatorial dry forest study region of Peru. Values for each 1km pixel are predicted using model-averaged effect sizes from both Cerro Venado and Laquipampa Wildlife Refuge surveys and remotely-sensed habitat data available for the region using the equation {ln(β0 + β1(nearest road) + β2(slope) + β3(forest cover) + β4(elevation)} (Manly et al., 2002).  A comparison with available land cover data indicates that high density areas are most strongly associated with intact paramo and montane forest, likely due to the greater elevation or forest cover observed in those sites, but also show a high association with both dry forest and agricultural areas (Figure 2.6). Additional qualitative comparisons show areas of high predicted density that potentially co-occurred with small-holder agricultural cover with few or no roads. 26 However, most high density areas were associated with relatively intact cover in montane forest, dry forest, or paramo.   Figure 2.6. Regional predictions for Andean bears and land cover comparison for the equatorial dry forest ecoregion in Peru, where details of numbered areas show a) density predictions; b) land cover; and c) satellite imagery. Areas 1,2 and 3 show areas of high predicted density that may have high co-occurrence with small-holder agricultural areas far from roads. Area 4 shows a comparative high density area with low predicted agriculture cover. The bar chart shows the predicted density breakdown for each land cover type region-wide, in increasing order of predicted suitability. 27 2.4 Discussion We set out to test whether forest cover, topographic features, or distance to roads influenced spatial variation in of Andean bears within a 600 km2 area of the equatorial dry forests of northwestern Peru. We then used that information to estimate the potential density of bears throughout the equatorial dry forest region to help identify potential knowledge and data gaps for the species and region. Roads had a large and most consistent influence on Andean bear density, measured as the number of predicted activity centers per unit area, mirroring responses of bear species worldwide and declining in density as proximity to roads increased (Doak, 1995; Fisher and Burton, 2018; Lamb et al., 2018). Less consistent effects of topography and tree cover were observed across seasons and study areas, potentially reflecting variation in habitat use by bears occupying different elevation ranges (Treves and Karanth, 2003).  It is widely believed that Andean bears preferentially occupy high-altitude cloud forest and paramó ecosystems (Peyton et al., 1995; Cuesta, Peralvo and Van Manen, 2003; Peralvo, Cuesta and van Manen, 2005; García-Rangel, 2012). However, our results indicate that Andean bears also reside year-round in equatorial dry forest at relatively high densities of 4 bears per 100km2, thus approaching estimates of ~7 bears per 100km2 in montane cloud forest (Molina et al., 2017). In some areas bears appeared to prefer lower elevation areas with sparse tree cover. Appleton et al. (2018) noted that bear food availability varied seasonally? at low elevation, which probably also contributed to seasonal variation in predictions between studies areas. Proximity to roads was a consistent negative predictor of Andean bear density in all models. Roads are also well-known to reduce the density of large mammals, and especially carnivores (Forman and Alexander, 1998; Basille et al., 2013), because they are avoided, act as demographic sinks, or both (Doak 1995). For example, vehicles are a significant source of direct mortality in many carnivores (Červinka et al., 2015), but adverse effects of roads may also arise from avoidance behavior (Doak, 1995) linked to traffic noise or visual effects that detectable hundreds of meters from roadsides (Forman and Alexander, 1998). Consequently, reductions in habitat quality can also reduce mobility and gene flow between populations (Cushman and Lewis, 2010).  The avoidance of areas close to roads has also been documented for bear species worldwide, including American black bears (Gantchoff and Belant, 2017), Asiatic black bears (Liu et al., 2009), grizzly bears (McLellan and Shackleton, 1988; Lamb et al., 2018), sun bears (Wong, Leader-Williams and Linkie, 2013), and sloth bears (Ratnayeke et al., 2007). Roads have also 28 been used to predict human-wildlife conflict in areas where humans and high quality habitat intersect, especially where roads and agriculture co-occur (Treves et al., 2004). Similar patterns might be expected in Andean bears, given that their perceived impacts on livestock and crops also contributes to their persecution by humans (Goldstein et al., 2006).  In that context, our results imply that historic road development may have reduced the overall abundance of Andean bears in tropical dry forest substantially over the last century. As a consequence, future work on the response of Andean bears to roads may wish to develop spatial data layers that more precisely record the occurrence of roads by type of road and traffic volume (e.g. McLellan & Shackleton 1988), or investigate the mechanistic links between roads and Andean bear population dynamics.  Forest cover and topography had opposite effects in Laquipampa and Cerro Venado, in contrast to the consistent negative effects of roads on Andean bear density. Our ability to interpret these results is limited due to high uncertainty in effect size. Nevertheless, bear density was negatively associated with elevation in Cerro Venado but positively associated with elevation in Laquipampa. Food availability often drives habitat use by bears, so while these results seem contradictory, they likely reflect local variation in resource availability and use at each site. Cerro Venado is lower in elevation and bears rely annually on the fruit of sapote (Capparis scabrida; Appleton et al., 2018); in such places bears may seek out low elevation habitat in search of food. We observed the opposite pattern in Laquipampa, perhaps because sapote was scarce and altitudinal migrations to access fruit less common (García-Rangel, 2012).  It is well-known that Andean bears use topographic features such as cliffs, steep slopes, and ridges as travel corridors and for refuge (Peyton, 1980; Goldstein, 2002), but we found no consistent effect of slope on bear density. Andean bears in Colombia were found to occupy a wide range of slope and terrain conditions provided there were available food resources (Cuesta, Peralvo and Van Manen, 2003). Therefore, a positive effect of slope on density at Laquipampa may reflect a number of underlying processes such as shelter and accessibility by humans, whereas at Cerro Venado, these processes may be masked by a strong seasonal use of fruit, or seasonal reliance on waterholes, which may be found in areas with certain topographic and vegetative characteristics. Future research may wish to explore the use of topographic complexity variables rather than simply slope.  We also expected to find positive relationships between forest cover and bear density, given Velez–Liendo et al.'s (2014) suggestion that forest cover predicts the availability of shelter and bromeliad food for Andean bears in Bolivia; a trend also observed in sun bears (Scotson et 29 al., 2017). We found some support for this in Cerro Venado and across pooled datasets, but not in Laquipampa; perhaps because food availability changed differently with respect to elevation in each study area, because of limitations due to model complexity, or due to factors not captured in our analysis, such as the distribution of cattle. Cattle were also more common in Laquipampa than Cerro Venado and known to influence carnivore abundance elsewhere (e.g., Sharma et al., 2015).  Seasonal variation in habitat use has been documented for many bear species (Weaver et al., 1996; Nielsen et al., 2010; Takahata et al., 2014) including Andean bears (García-Rangel, 2012; Velez–Liendo, Adriaensen and Matthysen, 2014), but not previously in tropical dry forest (cf Appleton et al., 2018). We detected a 50% increase in home range size from the dry to wet season, indicating a seasonal shift in the intensity of habitat use. This finding supports earlier studies of Andean bears based on many fewer individuals (Peyton, 1980; Castellanos, 2011). Patterns of resource selection and intensity are often mediated by seasonal variation in resource availability and linked to the intrinsic traits of individual animals such as sex and age (e.g., Ibex, Capra pyrenaica., Viana et al., 2018; Grey seal, Halichoerus grypus, Harvey et al., 2008; Lynx, Lynx lynx, Basille et al., 2013).  Overall, therefore, our results agree with those reported for a wide range of large mammals and suggest that, even within ecoregions, variation in habitat selection and quality can occur at fine temporal and spatial scales. 2.4.1 Regional Predictions We pooled data across study sites to estimate the abundance of Andean bears regionally in equatorial dry forests of northwestern Peru, a globally threatened biodiversity hotspot (Janzen, 1988; Hoekstra et al., 2005; Linares-Palomino et al., 2010). We took this approach to evaluate hypotheses on how models might fail, and to identify areas that could be used in the future to validate the model by conducting new surveys in areas predicted to be of higher quality for bears and, by extension, to rare or endemic species that may co-occur with Andean bears in relatively intact dry tropical forests. Specifically, low overlap between some areas predicted by our model to be capable of supporting Andean bears, existing protected areas, and the IUCN range map for Andean bears (Figure 2.5), suggests that the potential to improve conservation outcomes in equatorial dry forest still exists (Appleton et al., 2018). In such areas, Andean bears may also serve as ‘umbrella species’ (Bennett et al., 2015), including for functional plant species such as sapote, which provide annual resource pulses critical to Andean bears at Cero Venado but are globally endangered (Rodríguez et al., 2007; Linares-Palomino et al., 2010; Appleton et al., 2018). 30 However, given the relatively weak effects of forest cover and topography in our local model, and very limited understanding of the food distribution and availability of food resources, our regional predictions necessarily relied largely on the avoidance of roads, as an index of human impact at the landscape scale (e.g., Trombulak & Frissell 2000). However, by examining unclassified images of areas predicted to be capable of supporting bear, we also noted marked variation in the occurrence of small-holder agriculture which remains un-mapped or imprecisely mapped. Given that agriculture may reduce bear density via persecution by humans (Goldstein et al., 2006), our results also suggest that there is value in estimating the independent effects of small-holder agriculture and roads on Andean bear density in tropical dry forest, particularly to the degree that small-holder farms continue to move upslope due to climate change, as predicted for coffee, other shade crops, and ranchers (Buytaert et al., 2006; Mulligan et al., 2010). But taking that step will require high-resolution land use mapping and additional surveys for bears in areas with and without small-holder agriculture and high predicted bear density (Figure 2.5 & Figure 2.6).  Despite existing limits on the precision and availability of spatial data for the tropical dry forest ecoregion we studied, our results revealed potential errors in model predictions that should help prioritize management and research effort in future. Given the prevalence of small-holder agriculture in the Andes (Mulligan et al., 2010), and the influence of roads on bear density in our study and globally, our results suggest that improved mapping of agricultural impacts, roads, and other human land uses are urgently needed to help identify remaining options for the conservation of tropical dry forests of Northern Peru. Establishing camera trap surveys in equatorial dry forest predicted to support high Andean bear densities, but including or excluding small-holder agriculture and a variety of road densities also has the potential to dramatically refine our existing model, reveal remaining opportunities for conservation, and refine existing range maps. 2.4.2 Conclusions We used multi-year camera trap data and spatial capture-recapture models to estimate the spatial variation in local density of Andean bears in an area representative of equatorial dry forests of northwestern Peru. While there is likely the potential for scale-dependent habitat selection, fine-scale analyses like this have the power to detect local scale patterns and variation in density resulting from habitat type which can be validated at larger scales. Working within our 668 km2 study area, we found that roads reduced the density of Andean bears, as expected given many reports for large carnivores globally. Moreover, we found that the same dominant variables used to explain bear densities globally captured much of the variation in density that we observed 31 in Andean bears within our study areas.  However, the influence of factors other than roads varied seasonally, probably due to unmapped seasonal factors related to pulsed food resources and mating behavior (Appleton et al., 2018). Although such factors may contribute bias and uncertainty in model predictions, our results nevertheless demonstrate clearly that some high-quality habitats for Andean bears remain unmapped (Appleton et al., 2018), and suggest that outstanding opportunities to conserve tropical dry forest still exist outside the currently recognized Andean bear range. Improvements to our current model will require the creation of high resolution land classification and further surveys to test its predictions when applied at regional scales, as well as testing the assumption that the covariate coverages did not change over the 4 year study period. However, given the influence of roads on Andean bears estimated here, reported previously (Cuesta, Peralvo and Van Manen, 2003; Kattan et al., 2004; Sánchez-Mercado et al., 2014), and recognized globally among bears (Simek et al., 2015; Scotson et al., 2017), the most cost-efficient next step is to apply our improved understanding of the influence of roads on Andean bears to modify the existing range map by combining coefficients on the influence of roads from other studies in a meta-analytical framework. Such maps can be viewed as hypotheses of ‘best available knowledge’ and must be validated (Bino, Ramp and Kingsford, 2014; Maréchaux, Rodrigues and Charpentier, 2017).  Overall, our results should help facilitate decisions on the allocation of conservation effort for Andean bears by facilitating systematic conservation planning Andean bears and associated rare and endemic species of equatorial dry forest.   32 Chapter 3. Systematic conservation prioritisation in the Northern Andes: integrating focal species and multiple goals 3.1 Introduction Declines in large-bodied terrestrial vertebrates have occurred globally and are often linked to human disturbance and habitat loss and fragmentation (Dirzo et al., 2014). Identifying areas to conserve large-bodied carnivores is particularly challenging due to poor understanding of the ecology and distribution of these species (Wilson, Carwardine and Possingham, 2009), and to perceived and real trade-offs between conservation outcomes and human well-being (McBride et al., 2007; Wilson et al., 2007). It therefore remains imperative to find opportunities to conserve habitats likely to maximize the persistence of at-risk species while also maximizing positive outcomes for human well-being (Wilson, Carwardine and Possingham, 2009; Ban et al., 2013).  One approach to finding such opportunities is to use charismatic focal species to guide conservation planning at large spatial scales (Roberge and Angelstam, 2004), and to test for potential synergies between conservation outcomes and human well-being. ‘Umbrella species’ represent a type of focal species with large area requirements, for which their conservation is presumed to have ‘spill-over effects’ on ‘beneficiary’ species that co-occur in those habitats (Caro, 2003; Thorne, Cameron and Quinn, 2006; Breckheimer et al., 2014). When combined with goals for species with shared habitats and threats, umbrella species approaches can result in overall biodiversity gains, and charismatic species have the distinct advantage of being highly effective at securing otherwise inaccessible private funding (Bowen-Jones and Entwistle, 2002; Bennett et al., 2015). Carroll et al. (2004) also found that umbrella species approaches to conservation planning were most effective when they focused on restoring connectivity for wide ranging mammals with fragmented habitat. Multi-species and multiple goal planning constitute an extension of the original umbrella concept that can include targets for focal species as well as for more general biodiversity goals and complementary ecosystem services  (Margules, Pressey and Williams, 2002). The rise of systematic planning approaches that use advanced computer algorithms to find optimal solutions has drastically increased the feasibility of incorporating multiple goals into land use planning (Flather et al., 1997; Margules, Pressey and Williams, 2002). Historically, conservation land acquisitions occurred mostly in an ad hoc manner due to the limited resources and funding for conservation (Gonzales et al., 2003); disproportionately on lands of low 33 productivity and economic value, and only rarely with the goal of maximizing biodiversity (Pressey, Possingham and Margules, 1996; Margules and Pressey, 2000). Quantifying trade-offs in terms of the ‘cost’ of conservation to humans is an essential step in improving the accessibility of systematic conservation plans to land managers, but has only become practical with the advent of better computing methods. In particular, integer linear programming techniques have emerged as a way to quantify trade-offs and solve complex conservation planning problems in a fraction of the time of common optimization algorithms (Beyer et al., 2016). These algorithms make it possible to efficiently identify the configuration of planning units that fulfill targets for representation of diverse features at the least cost, where cost is often quantified as the area required to form a solution. However, the efficacy of systematic conservation planning approaches that use focal species remain untested in many systems, particularly where information on the distribution and abundance of at-risk species is limited or unknown. The Northern Andes is a global biodiversity hotspot that has undergone drastic land use change in the last century (Myers et al., 2000), and has been identified as one of the most vulnerable areas to climate change (Bush, Silman and Urrego, 2004).  Compared to other tropical areas, relatively little focus has been put on conservation planning in the Northern Andes (Rodríguez and Young, 2000; Sierra, Campos and Chamberlin, 2002), and none across the entire range. Watersheds are an important resource for both humans and wildlife populations; in some areas, including the Northern Andes, the degradation of watersheds poses a direct threat to food production, drinking water and hydroelectricity for humans (Bush, Silman and Urrego, 2004). Given predicted climate change scenarios and a growing human population density in the area, there is an urgent need to identify key conservation areas for important ecosystem values identified by local and international experts (Kintz, Young and Crews-Meyer, 2006). In this analysis, we included targets for representation of species ranges (International Union on the Conservation of Nature; IUCN 2018), high water-risk watersheds (Harden, 2006; Vera et al., 2006; Gassert et al., 2014), ecoregions (Dinerstein et al., 2017), and iconic species.  Andean bears (Tremarctos ornatus) are the only extant species of bear in South America and are thought to have declined dramatically over their range in the Northern Andes due to habitat loss and human hunting (Peyton et al., 1995). Given their potential role as an umbrella species (Caro, 2003; Roberge and Angelstam, 2004), their sensitivity to human development (see Chapter 1; Kattan et al. 2004), and their coexistence with countless threatened species in a key global biodiversity hotspot (Myers et al., 2000), Andean bears offer an outstanding opportunity to test the hypothesis that combining conservation for large carnivores with multiple conservation goals can improve the quality and accessibility of strategic conservation plans.   34 We used the principles of systematic conservation planning to explore alternate land use plans likely to maximize the integrity and persistence of focal communities in the Northern Andes. Our overall objective was to deliver a multi-species GIS decision support tool that could be used to prioritize land acquisition and conservation investment. Our specific goals were threefold; i) develop and compare scenarios to prioritize areas in the Northern Andes that maximize the representation of 4 vertebrate classes, 44 ecoregions, and one ecosystem service, and minimize the area cost; ii) explore how systematic planning methods might be used to identify and prioritize least cost paths between existing protected areas as corridors for large carnivore movement; and iii) ask what fraction of proposed priority areas are protected and, more generally, ask about the coverage of other species ranges and ecoregions within the Andean bear range and proposed priority areas.  3.2 Methods 3.2.1 Study area We delineated our study area using topographic and land cover features across the Northern Andean mountain range from Venezuela to Bolivia and defined a 1km grid of planning units within this area (Figure 3.1). This area included the Andean bear range as mapped by the IUCN using a combination of known occurrences, expert opinion, and published information on American black bears (Ursus americanus; Goldstein et al. 2008). We represented biodiversity in the study area by aggregating 4494 ranges for vertebrate species mapped by the IUCN to create: a) a species richness layer for each vertebrate class based on all Figure 3.1. North Andes study area overlaying digital elevation model (DEM) of South America. Andean bear range is indicated in white. 35 species, and b) on a sub-set of vulnerable, threatened, or critically endangered species in each class of vertebrates (Figure 3.2; IUCN, 2018; Jenkins & Van Houtan, 2016).  We compiled data on Andean ecoregions from Olson et al. (2001; terrestrial ecoregions dataset; Figure 3.2) and the areas of each ecoregion currently protected from the World Database on Protected Areas (WDPA; UNEP-WCMC & IUCN 2016). We also delineated protected areas into two broad categories: core areas and buffer zones or transitional areas. Core areas include national parks and areas where alternative human land uses are limited or prohibited. Buffer or transitional areas include integrated management zones where some alternative land uses are permitted. We used data on overall risk of water provisioning or quality scarcities (hereafter water risk) by sub-basin from the Aqueduct Water Risk Atlas (Gassert et al., 2014; Figure 3.2). All spatial variables were developed in ArcGIS (ESRI, 2011) and R version 3.3.3 (R Core Development Team, 2014).   36   Figure 3.2. Biodiversity feature layers used in spatial prioritization; a) overall amphibian species richness; b) threatened amphibian species richness; c) overall bird species richness; d) small-ranged threatened bird species richness; e) threatened bird species richness; f) overall mammal species richness; g) threatened mammal species richness; h) overall reptile species richness; i) overall water risk by sub-basin (Gassert et al., 2014); and  j) ecoregions (Olson et al., 2001). 37 3.2.2 Systematic conservation planning We applied well-accepted concepts in systematic conservation planning (Margules and Pressey, 2000) to identify priority areas for conservation within our study area using the prioritizr R package (Hanson et al., 2018).  Prioritizr uses integer linear programming (ILP) techniques to build and solve conservation planning problems, following theory and principles widely-applied in Marxan (Ball, Possingham and Watts, 2009; Beyer et al., 2016). Our objective here was to find configurations of planning units that met biodiversity targets at the least cost, otherwise known as ‘the minimum set problem’. The general form of a minimum set problem using ILP can be expressed in matrix notation as: 𝑀𝑖𝑛𝑖𝑚𝑖𝑧𝑒 𝑐𝑇𝑥 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝐴𝑥 ≥ 𝑏 Where x is a vector of decision variables, c and b are vectors of known coefficients, and A is the constraint matrix (Beyer et al., 2016). The final term specifies a series of structural constraints where relational operators for the constraint can be either ≥ the coefficients. In the context of a conservation planning problem, c is used to represent the planning unit costs, A is used to store the data showing the presence/absence (or amount) of each feature in each planning unit, and b represents the minimum target representation required for each feature in the solution.  Because of the lack of detailed information on acquisition cost or some other monetary measure, we selected ‘human impact’ cost metrics for our analysis, under the assumption that conservation of areas with a high degree of human impact would have high monetary or opportunity costs for humans and be less viable for the long-term conservation of biodiversity.  We explored two cost metrics: a) human footprint, based on 2009 data (Venter et al., 2016) and b) road impact, based on empirical estimates for Andean bears estimated at Cerro Venado and Laquipampa Peru (Chapter 2; Figure 3.3). The latter was created using empirically-estimated coefficients derived from robust spatial capture-recapture modelling with a suite of other habitat variables, which have been kept constant for the purposes of this analysis (See chapter 2), but the relationship is supported by findings in multiple studies on Andean bears (Cuesta, Peralvo and Van Manen, 2003; Kattan et al., 2004; Sánchez-Mercado et al., 2008). A final combined cost metric was created that used both the human footprint and the empirically-estimated road impact layer together, assuming equal weights for each layer.  Although interrelated, solutions with a smaller overall area were considered to have high area efficiency, and solutions with a smaller combined ‘human impact’ cost were considered to have more monetary or opportunity cost efficiency.  38  Figure 3.3. Cost layers used in prioritization 3.2.3 Scenario development and post processing  We set relative area targets for all biodiversity features, including species richness layers, ecoregions, and water risk, at 17, 30 and 50% to identify priority areas required to meet Aichi target 11 (Secretariat for the Convention on Biological Diversity; SCBD 2010), address the ‘Nature Needs Half’ (hereafter NNH) objective outlined by Dinerstein et al. (2017), and included an intermediate value for comparison (Table 3.1). All runs we implemented using the combined cost metric that incorporated both the human footprint and the road impact. We did not incorporate any constraints or penalties to increase reserve design compactness, and we did not ‘lock in’ planning units in existing protected areas. We created multiple scenarios to explore the sensitivity of solutions to the features included. These scenarios were divided into two categories: multiple features approaches, which included some combination of vertebrate classes, ecoregions, and water risk; and single-species approaches, which only included one species feature layer, in order to address the umbrella species potential of those species (Table 3.1). For the single species scenarios, we used the IUCN range of Andean bears (Goldstein et al., 2008), and a ‘random’ uniformly distributed feature covering the same area.   39 Table 3.1. Scenarios used to identify priority conservation areas in the Northern Andes.  Multiple features approach    Area targets tested Scenario # Scenario Name Scenario description 17% 30% 50% 100% 1 Full scenario All features layers included: 8 vertebrate classes, 44 ecoregions, and water risk ecosystem service layer ✓ ✓ ✓ X 2 Biodiversity focus 8 vertebrate classes, 44 ecoregions, but no water risk layer ✓ ✓ ✓ X 3 Threatened species focus Only feature layers for threatened vertebrate species included ✓ ✓ ✓ X Single species approach 4 Andean bear focus Only IUCN Andean bear range included ✓ ✓ ✓ ✓ 5 Random feature focus Only randomly generated feature included: uniformly distributed feature covering the same area as Andean bear range ✓ ✓ ✓ ✓  We further analyzed the single-species results by exploring the potential degree of human impact within the existing Andean bear range. To do this we combined the road avoidance cost layer (Chapter 2) with the Andean bear range using a simplified resource selection function (Manly et al., 2002),  and delineated areas likely to support more than 4 individuals/100km2, the proposed lowest realistic density of Andean bears identified by Yerena and Torres (1994) based on limited historical field data and density estimates on North American black bears. This analysis assumes that Andean bears avoid roads to the same extent across their range as in equatorial dry forest in Peru, and does not differentiate roads by traffic volume or road type, but offers potential insight as it implies that some already-protected areas may be impacted by roads in and around them, and thus potentially more difficult to manage over the long term. For the corridor analysis, we ran scenarios that iteratively increased biodiversity target from the 17% level previously endorsed by the SCBD, to the 50% level advocated by NNH (in 2% increments) to create a ramped resistance surface to help identify potential wildlife corridors between existing and potential protected areas. The values in this resistance surface are the inverse of the number of times that each planning unit was selected in a solution, where ‘resistance’ is lower in cells more frequently selected in solutions, and therefore more likely to 40 support a suite of species that require intact habitat for dispersal.  We then identified potential corridors using Linkage Mapper (McRae, Shah and Mohapatra, 2013), a software plug-in that finds the ‘least cost path’ using circuit theory (McRae and Beier, 2007). Our objective was to identify potential pathways that large vertebrates might traverse between habitat patches, or ‘nodes’, where nodes represent a location within a landscape wherein Andean bears are predicted to be highly likely to occur. In our case, we used all existing core protected areas which intersected the IUCN range for Andean bears, assuming that Andean bears are present in those protected areas. Following Mcrae et al. (2008) we created a continuous surface where the value in each pixel represents the cost accumulated moving along the most efficient possible route that passes through the cell from one node to the other. We visualized key corridors by only showing areas within a cost-weighted distance of less than 200 000 from the least cost path, assuming that beyond that value the landscape has high resistance values and therefore a low probability of being used as a corridor (Mcrae et al., 2008). We considered linkages to be unrealistic if the least cost path distance was more than 56 km, one value for maximum dispersal distance observed for Andean bears (Jorgenson and Sandoval-A, 2005), or if they traversed areas above 4700m a.s.l., the upper altitudinal limit of Andean bear range (García-Rangel, 2012).  3.3 Results 3.3.1 Biodiversity representation Including area target requirements for multiple features improved the number of features represented and the area of those features being represented by an average 257% (± 96%) versus scenarios that only incorporated targets for Andean bears (Table 3.2 and Table 3.3). We selected Scenario 1, where equal targets were set for all features, as the best representation of biodiversity across the North Andes at the least cost, given the input data (Figure 3.4). In this solution, most area targets were exceeded for features, indicating high complementarity, particularly for overall amphibian (by 15±6%), bird (by 9±3%), and reptile species richness (by 6±2%). Targets for threatened amphibian species and small-ranged bird species were only just met, which may indicate that these taxa are limiting.  However, despite high overall representation of vertebrate classes and ecoregions, Scenario 1 required more than double the land area to form a solution (Table 3.2) compared with the single-species approach used in Scenario 4 (Table 3.3).  Our results indicated that dropping biodiversity features tended to improve the area efficiency of solutions. However, we did not observe a consistent proportional increase in the area of other features being represented when features were dropped, which would be expected if there was no overlap or complementarity between features. For example, when the target for 41 water risk was dropped from analysis in Scenario 2, we found that there was a slight 2% (±1.6%), improvement in average representation of vertebrate species richness compared with Scenario 1, which may indicate a mismatch between areas of highest risk for water provisioning and areas of highest species richness (Table 3.2). Whereas scenario solutions with or without water risk layers included the same physical area (17%, 30%, and 50% of the land base), the area cost of the solution for Scenario 2, which did not include water risk, was on average 1% higher than the solution for Scenario 1; indicating that the tool was forced to select areas of higher human impact in order to meet the target for water risk. In contrast, setting targets for threatened species only in Scenario 3 cut the total area in the solution by an average of 42% across target levels compared to Scenario 1, and the cost of the solution by 41%, but water risk representation was at least 43% below each target level and many ecoregions were not captured at all (See Appendix E).  3.3.2 Landscape linkages The circuit model identified 100 potential corridors that connected nodes for Andean bears; where nodes are defined here as core protected areas which intersected the known range of Andean bears (Figure 3.5). Of these corridors, 29 had a least cost path length longer than 56km, and therefore lower realism for Andean bear dispersal. All corridors traversed areas of high resistance habitat near roads within major river valleys, and a qualitative comparison shows that some of the low resistance corridors cross areas of high elevation above 4700 meters, higher than Andean bears are known to occur. As a result, it is likely that many of these corridors are effectively unpassable for Andean bears without restorative actions or by re-routing through other areas also predicted to impede dispersal, indicating that populations on either side of these predicted bottlenecks may become more isolated over time. Despite this, the 41 corridors in the top quantile of cost-weighted distances may represent potential corridors for Andean bears which co-occur with high biodiversity areas. 3.3.3 Efficacy of Andean bears as an umbrella species We found that Scenario 4, representing a single-species approach to planning over the Andean bear range, captured an average of 20% (±12%) more vertebrate species as compared to Scenario 5, which used a randomly-generated, uniformly-distributed feature covering the same cumulative area as the Andean bear range within the region (Figure 3.6 and Figure 3.7). Ecoregions well-represented in the Scenario 4 solution included several páramo regions, Bolivian yungas, and montane forests. There were a number of ecoregions that were not included in the solution, and areas of high water risk were under-represented, meaning the area included was lower than the target set for Andean bear range. Overall, the representation of vertebrate species 42 richness in Scenario 4 reached a maximum of 28% of species for birds, 28% for mammals, 28% for amphibians, and 26% for reptiles (Table 3.3). These percentages were higher than when using a random feature in each case, but were also substantially lower than the maximums obtained by multi-feature approach (53%, 52%, 56%, and 52%, respectively; Table 3.2), indicating that planning for Andean bears alone is unlikely to be sufficient to meet representation targets for other taxa. However, the area required for the Scenario 4 solution is a fraction of that required for the multiple features scenarios, and the area is also lower for Scenario 4 than for a randomly generated feature (by ~ 234 km2), indicating that some complementarity between features was achieved.  When comparing the Andean bear range to the human impact cost layer using a coefficient for road avoidance behaviour, we identified 208 684 km2 of suitable habitat likely to support more than 4 individuals/100km2. This represents an approximately 31% reduction of the original Andean bear range.  We found that 35% of this reduced range had some kind of protected status. Our solution overlapped with 95 622 km2, or 46% of the total area identified as prime Andean bear habitat. If protected, this area would nearly double the protected areas in highly suitable habitat for Andean bears, bringing the value to 63%. Lastly, by combining results from the 17% target solution for Scenario 1 with landscape linkages we identified an ‘overall solution’, represented by a candidate set of areas to satisfy area-based goals outlined in Aichi target 11 and incorporate potential movement corridors for large-carnivores (Figure 3.8). The configuration covered 294 586 km2, of which 36% was identified as potential corridor area. At present, 11% of our Northern Andes study area exists in a core protected area. Of the 222 268 km2 identified in our solution, 46 448 km2 were in existing core protected areas, and 42 439 km2 were in a buffer area or integrated management area. This indicated that about 60% of the areas identified here as containing the best representation of biodiversity features remain unprotected.    43  Figure 3.4. Selected priority areas in prioritisation scenario 1, where all features were given equal targets. Targets were set at 17%, 30%, and 50% levels; shown here overlaid because planning units selected for 17% targets are always selected again for higher target level.  44 Table 3.2. Feature representation table for vertebrate species and water risk in three separate target setting scenarios. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. Scenario 1 supplied equal targets for all feature layers, scenario 2 supplied equal targets for all feature layers but dropped water risk, and scenario 3 only supplied targets for threatened vertebrate species. For full feature representations, including ecoregions, see Appendix E. Scenario 1. All features given equal targets Feature 17% 30% 50% All amphibian species 20.60 34.81 53.80 Threatened amphibian species 17.00 30.00 50.00 Andean bear distribution 25.64 42.47 62.99 All bird species 18.94 32.90 52.67 Small-ranged bird species 17.00 30.00 50.00 Threatened bird species 19.25 32.97 52.81 All mammal species 17.82 31.38 51.24 Threatened mammal species 17.84 31.37 51.22 All reptile species 18.44 32.06 51.58 Water risk 17.00 30.00 50.00 Total area (km2) 222269 392220 653673 Total cost 11525550 20588810 34815550 Scenario 2. All features given equal targets, no water risk Feature 17% 30% 50% All amphibian species 21.97 36.03 55.58 Threatened amphibian species 17.00 30.00 50.00 Andean bear distribution 26.89 43.77 64.34 All bird species 19.51 33.52 53.46 Small-ranged bird species 17.00 30.00 50.17 Threatened bird species 19.76 33.69 53.64 All mammal species 18.33 32.02 51.92 Threatened mammal species 18.35 32.01 51.89 All reptile species 19.24 32.82 52.46 Water risk 15.63 28.18 48.22 Total area (km2) 222268 392219 653673 Total cost 11510270 20569060 34802890     45 Scenario 3. Targets for threatened species only Feature 17% 30% 50% All amphibian species 31.85 45.35 63.95 Threatened amphibian species 17.00 30.00 50.00 Andean bear distribution 21.02 35.86 55.95 All bird species 18.31 31.51 51.24 Small-ranged bird species 10.37 21.67 41.95 Threatened bird species 17.41 30.98 50.95 All mammal species 17.06 30.14 50.32 Threatened mammal species 17.00 30.00 50.00 All reptile species 17.53 30.00 49.69 Water risk 8.43 15.63 28.50 Total area (km2) 152206 275879 475961 Total cost 7913914 14617800 25789360 46  Figure 3.5. Composite of least-cost corridors connecting existing core protected areas that intersect the known range of Andean bears. Corridors only considered for nearest neighbouring protected area to each protected area, and are shown here truncated at a total cost-weighted distance of 200 000 km.   47   Figure 3.6. Representation of auxiliary features in single-feature prioritization with targets at 17%, 30%, 50%, and 100% of the total feature area; where a) uses Andean bear range; and b) uses a random uniformly generated feature covering the same total area as A  48   Figure 3.7. Representation of auxiliary features in single-feature prioritization with targets at 17%, 30%, 50%, and 100% of the total feature area; where a) uses Andean bear range; and b) uses a random uniformly generated feature covering the same total area as Andean bear range.   49 Table 3.3. Feature representation table for vertebrate species and water risk in two single-feature target setting scenarios. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. a) uses Andean bear range; and b) uses a random uniformly generated feature covering the same total area as Andean bear range. For full feature representations, including ecoregions, see Appendix E. Scenario 4: Single-feature targets for Andean bears Feature 17% 30% 50% 100% All amphibian species 6.21 10.00 16.30 27.59 Threatened amphibian species 3.74 7.20 15.41 35.51 Andean bear distribution 17.00 30.00 50.00 100.00 All bird species 5.59 9.39 15.05 28.32 Small-ranged bird species 4.61 8.19 14.61 31.92 Threatened bird species 6.05 9.93 15.93 30.32 All mammal species 5.12 8.65 14.04 27.86 Threatened mammal species 5.08 8.62 14.04 27.78 All reptile species 4.61 8.16 13.56 26.20 Water risk 3.06 5.41 9.17 18.87      Total area (km^2) 51569 91004 151673 303346 Total cost 2583634 4638898 7873614 16523200           Scenario 5: Single-feature target for random feature Feature 17% 30% 50% 100% All amphibian species 6.50 10.00 14.14 23.12 Threatened amphibian species 2.83 5.80 10.36 22.88 Andean bear distribution 6.04 10.73 16.01 23.28 All bird species 5.22 8.33 12.58 23.02 Small-ranged bird species 3.56 6.55 11.26 22.92 Threatened bird species 5.11 8.13 12.51 22.99 All mammal species 4.66 7.44 11.62 22.95 Threatened mammal species 4.65 7.46 11.67 22.96 All reptile species 4.68 7.62 11.76 22.95 Water risk 3.38 6.40 11.29 22.98      Total area (km^2) 51693 91223 152037 301034 Total cost 2628930 4721795 8030882 16886390  50  Figure 3.8. Combined results showing 1) existing core protected areas; 2) priority areas selected when 17% targets were applied to all features; and 3) least-cost corridors truncated at a total cost-weighted distance of 200 000 km.  51 3.4 Discussion We applied systematic planning tools and publicly available data to identify potential options to conserve habitat in support of Andean bear populations and 4494 other vertebrates species inhabiting the Northern Andes biodiversity hotspot (Myers et al., 2000). Key synergies and potential trade-offs identified when planning for multiple goals included that a) planning for multiple goals together used less area than planning to conserve those features separately, but b) increased the overall area needed to reach conservation targets and protect high water risk watersheds versus planning for fewer features. The area needed to meet targets for 8 vertebrate classes, 44 ecoregions, and one ecosystem service was the same as the target supplied for each layer.  Incorporating watersheds with high water risk had little impact on the areas prioritized, the area cost of the solution, or the representation of beneficiary species, suggesting potential synergies with biodiversity conservation. Watersheds are an important resource for human and wildlife populations, but in some areas including the Northern Andes, watershed degradation  poses a direct threat to food production, drinking water and hydroelectricity for humans (Bush, Silman and Urrego, 2004). Although not investigated here, other ecosystem services such as carbon storage and sequestration are also likely to accrue in recovering montane forests (Spracklen & Righelato 2016). Given predicted increases in the amplitude and rate of climate change in the tropical Andes (Urrutia and Vuille, 2009), incorporating the direct and indirect economic values of biodiversity conservation on ecosystem services and human well-being may be essential to achieving success in both arenas.  We also identified potential linkages between existing core protected areas to facilitate dispersal by Andean bears and by extension other species that avoid humans in the region. Because many mountainous regions of the world display high topographic and climatic diversity, species inhabiting such regions may be at risk of demographic and genetic isolation as humans convert lowland habitat in ways that reduce dispersal, especially by species that respond to climate change by shifting their ranges to more favorable habitats (Flagstad et al., 2001; Tovar et al., 2013). Our current results suggest that some populations of Andean bears may already be functionally isolated within the Northern Andes, but also point to areas where connectivity might be restored. In particular, we identified large areas of as yet unprotected habitat in Northern Peru which, if developed, may prevent gene flow between the Northern and Southern Andean bear range.  52 Given the potential demographic and genetic isolation of Andean bear populations in future (Figure 3.5), and limited knowledge about the permeability of landscapes to Andean bears, the open-source planning tools and feature layers we describe and provide herein represent a base model for examining the potential role of unprotected and private lands in maintaining landscape linkages (Pimm et al., 1995; Rabinowitz and Zeller, 2010). Schuster et al. (2018) reported that including working landscapes in conservation planning scenarios for 109 species of neotropical migrant birds that winter in Central and South America dramatically increased the efficiency of conservation plans. In the case of Andean bears, to adequately assess the permeability of landscapes to movement would require surveys of habitat use within and outside of protected area boundaries, as well as studies of dispersal and movement ecology, which have not yet been conducted.  Given that we identify many potential corridors that traversed established buffer areas or integrated management zones (Figure 3.8), such mixed-use landscapes may serve a key role in maintaining connectivity among Andean bear and other native vertebrate populations.   We also assessed the potential for Andean bears to act as a focal ‘umbrella species’ in regional planning. As expected, the Andean bear range performed better than random features at capturing species richness, as defined by the presence or absence of vertebrate species, rather than the portion of their range included in a solution. Single-species approaches have been criticized because many species are limited by factors not considered when planning for a single species (Caro, 2003; Roberge and Angelstam, 2004; Wang et al., 2018).  Our analysis was also unable to account for specific habitats which may be critical to the persistence of particular beneficiary species (Roberge and Angelstam, 2004). However, even conserving 100% of Andean bear range left gaps in conservation for some of the 4494 vertebrate species we considered whose ranges do not fall well-within that range, confirming a need for fine-filter approaches such as local-scale habitat assessments and species distribution modelling in conjunction with regional planning (Schwartz, 1999). Nevertheless, our results indicate that many areas likely to be critical to the conservation of Andean bears will also be critical to the conservation of many species which also occur in highly biodiverse areas within the Northern Andes, suggesting tremendous potential for strategic planning in favor of global species conservation. Our analysis represents a first step in demonstrating the potential utility of an open planning platform for strategic reserve design in the Northern Andes. However, the scenarios presented here are strictly illustrative, not prescriptive. Moreover, it is very clear that all scenarios could be improved by enhanced data inputs, especially of species distributions, as yet 53 unrepresented taxa, and on the distribution of human impacts and potential ecosystem services. Our results are subject to errors due to data inputs and therefore require validation. In the case of Andean bears, we employed spatial algorithms under the assumption that all bears react similarly to human-influence as they do in dry forest habitat in Peru, an uncommon habitat for Andean bears. In reality, this behavioral relationship may be scale-dependent and vary by population or by factors not captured in our analysis, including unmapped agriculture far from roads (Chapter 2; Treves and Karanth, 2003; Goldstein et al., 2006).  Also, existing maps fail to identify all roads and do not differentiate roads by size or volume, factors that affect the magnitude of road effects in other species, including black bears (Simek et al., 2015). Sánchez-Mercado et al. (2014) also showed that Andean bears may inadvertently choose habitats subject to high poaching risk. The spatial arrangement of potential ecological traps for bears will also affect the effectiveness of conservation actions (Doak, 1995; Sánchez-Mercado et al., 2014; Scotson et al., 2017), but the data necessary to do so are not yet available across the Andean bear range. Identifying and acquiring the most valuable new data products, and increasing the precision of existing input layers, should therefore be of real value organizations engaging in land use planning for species conservation and to support human well-being (Wilson et al., 2007; Wilson, Carwardine and Possingham, 2009).  Our results demonstrate the potential application of systematic planning tools for identifying priority habitats within a global biodiversity hotspot, and they provide a framework for incorporating cultural and socioeconomic issues in future, whereby spatial layers of socioeconomic features can be included as they become available (e.g. Chan et al., 2012; Ban et al., 2013; Durán et al., 2014). The rapid development of ecosystems in the Northern Andes highlight the urgent need for area-based plans to curb the decline in species and ecosystem integrity (Myers et al., 2000; Kintz, Young and Crews-Meyer, 2006; Wassenaar et al., 2007). The CBD commits 177 signatory countries to increase the area of protected areas to 17% by 2020, emphasizing ‘areas of particular importance for biodiversity’ (Target 11; SCBD 2010). Ongoing population declines and range contractions of many vertebrates resident in the Northern Andes (Bush, Silman and Urrego, 2004; Jenkins, Pimm and Joppa, 2013), and over-wintering there annually (Schuster et al., 2018) are now evident, but a response has been limited by financial and other constraints (McCarthy et al., 2012). This points to a need for landscape-scale plans that maximize the efficiency of conservation investment while minimizing the cost (Margules, Pressey and Williams, 2002). As a complement to expert input, systematic planning tools help identify efficient, alternative land use plans than improve on ad hoc approaches (Gonzales et al., 2003).  54 Chapter 4. General conclusion 4.1 Implications The planet is in an era of unprecedented change, and nowhere is this more apparent than in the tropics (Laurance et al., 2012). According to the 2005 Millennium Ecosystem Assessment, the current rate of biodiversity loss is exceeding historic rates by several orders of magnitude (MEA, 2005). Tropical forests are being lost at a rate of 0.58%/year, and that number continues to climb as a result of expanding pastoral and agricultural lands (Wassenaar et al., 2007; Wright, 2010). Some regions have been identified as key biodiversity hotspots because of the high richness of endemic species and the imminent risk of loss (Myers et al., 2000). One of these areas is the tropical Northern Andes, although relatively little attention has been given to strategic conservation planning in this region compared to other biodiversity hotspots (Rodríguez and Young, 2000; Sierra, Campos and Chamberlin, 2002), partially due to limitations on resources for conservation (McCarthy et al., 2012), and conflict with human needs for resource extraction (Kintz, Young and Crews-Meyer, 2006; Wassenaar et al., 2007). For these reasons, it is more important than ever to identify priority areas for conservation actions in the Northern Andes that represent functional ecosystems at the least cost to humans.  Large carnivores are among the most at-risk groups in the tropics, and also have the potential function as umbrella species for co-occurring species in the habitats they occupy (Roberge and Angelstam, 2004). These types of charismatic focal species have the added importance of helping to increase the efficiency and accessibility of complex systematic conservation plans, and may help secure funding for conservation that would otherwise be unavailable (Bennett et al., 2015). However, the successful conservation of any species with a large home range requires detailed knowledge on the ways in which they use their habitat, as well as their distribution and abundance within their range. Andean bears are vulnerable to extinction and are thought to have declined dramatically due to fragmentation and human hunting (Peyton et al., 1995; Goldstein et al., 2008). There is an urgent need to address the causes of decline and prioritize conservation areas in the Northern Andes likely to maximize the persistence of Andean bears, while simultaneously protecting the largest number of migrant and endemic species. In this thesis, I used a robust systematic camera trap dataset to conduct a detailed study on the habitat use of Andean bears in equatorial dry forest. I used habitat parameters that are easily assessed and available across the Northern Andes, including topography, forest cover, and 55 road density. As a result, this study may be used as a credible template to identify and assess areas likely to sustain populations of Andean bears in the future, particularly for areas close to human habitation where the potential for conflict exists. I extended this analysis to speculate on the impact of roads on Andean bears throughout their range, which I combined with human footprint data to identify areas with the highest human impact on the landscape and, by proxy, the highest opportunity cost for humans. By minimizing this cost and maximizing the representation of other features on the landscape, I was able to highlight key areas for conservation across the Andes, which offer several key insights into the performance and efficiency of existing protected areas and the ability of Andean bears to act as an umbrella species for the region. Generally, my work offers a common framework for using both species distribution modelling and systematic conservation planning methods to inform and defend managerial decisions for at-risk species and ecosystems at the local and landscape level. Especially where available data are limited, these methods make it possible to extract the maximum information from the data as possible, by using multiple data sources and state-of-the-art computing techniques. Robust studies such as detailed in this thesis are increasingly important as we move into a future of environmental uncertainty.  4.2 Key findings, limitations, and future steps In Chapter 2, I assessed local habitat use by Andean bears in equatorial dry forest, where little to no documentation of Andean bears exists and there is an urgent need for conservation actions. One of the key findings of the chapter was that roads, widely-known to influence on carnivore density and demography globally (Doak, 1995; Fisher and Burton, 2018; Lamb et al., 2018), also reduced the occurrence of Andean bears in the dry forest habitats I studied. I also found that the same dominant variables used to explain bear densities globally captured much of the variation in occurrence that we observed in Andean bears; however, the influence of factors other than roads varied seasonally. This highlights the importance of spatial and temporal scale in identifying key habitat for species involved in conservation planning and management actions. I identified several areas of habitat that may support high densities of Andean bears, or otherwise likely to support rare or endemic species that co-occur with Andean bears or that prefer relatively intact dry tropical forest habitats. Although I am confident that the results of my species distribution models represent the best available data on Andean bears in this region, it is possible that limitations in the fine-scale data for human impacts, such as smallholder agriculture and unmapped roads, impact the reliability of the results. This finding alone highlights the need to improve data layers for future iterations, if Andean bear habitats are to be successfully incorporated into conservation efforts. 56 In Chapter 3, I assessed the potential of Andean bears to act as umbrella species across their range, and I identified priority areas for conservation in the Northern Andes likely to support the maximum number of species, ecoregions, and at-risk watersheds at the least cost. I found that the Andean bear range performed better than a randomly generated feature at capturing species richness. I also found that planning for multiple goals in a systematic planning framework greatly increased the area efficiency of the solutions, compared with planning for each feature separately. However, in my comparison of richness, I only explored the presence or absence of beneficiary species, not portion of their range covered by solution, which may be a better measure of the potential for species to act as an umbrella (Caro, 2003; Breckheimer et al., 2014), but requires more detailed data on species distribution and density than is available for the Andes region. As with Chapter 2, results might be improved with high resolution input data. However, quantitative validations of such models are not yet possible due to limited reliability of land use data. Until those data are assembled, my results offer compelling predictions of high quality areas for biodiversity conservation in the Northern Andes.  While much of my thesis is focused on identifying habitats or areas deserving greater protection, it is important to note that protected areas are only part of the solution, particularly in countries with low average income, and where enforcement and monitoring is weak or absent (McCarthy et al., 2012). Conventional wisdom holds that governments should manage protected areas under a “fences-and-fines” regime, but protected areas often fail where local communities are not engaged, or are forced to abandon resource extraction in favour of newly designated parks (Barrett et al., 2001). Community-based natural resource management represents a critical next step in on-the-ground conservation implementation, but problems of weak institutions, transaction costs, and externalities continue to hamper the efforts of land managers (Ostrom, 1992; Agrawal and Gibson, 1999).  Although it is outside the scope of my thesis, it is worth mentioning that in the realm of practical implementation there is much more work to be done. That said, when the aim of conservation is also the protection of key ecosystem services, communities can be the foundation of successful management (Barrett et al., 2001). My thesis showed that integrated management areas and buffer zones were critical to meeting overall conservation goals, but additional community work may be required to ensure that these habitats do not become population sinks.  I included the water risk features as one example of an ecosystem service that could be included to increase incentives for local communities to participate in the conservation process. Overall, testing the biological assumptions behind conservation area prioritization is the first important step towards truly effective management of protected areas.   57 References Agrawal, A. and Gibson, C. C. (1999) Enchantment and disenchantment: The role of community in natural resource conservation, World Development. doi: 10.1016/S0305-750X(98)00161-2. Appleton, R. D. et al. (2018) ‘Phenotypic plasticity in the timing of reproduction in Andean bears’, Journal of Zoology, pp. 1–7. doi: 10.1111/jzo.12553. Apps, C. D. et al. (2001) Scale-dependent habitat selection by mountain caribou, Columbia Mountains, British Columbia, Journal of Wildlife Management. doi: 10.2307/3803278. Arnold, T. W. (2010) ‘Uninformative Parameters and Model Selection Using Akaike’s Information Criterion’, The Journal of Wildlife Management. Blackwell Publishing Ltd, 74(6), pp. 1175–1178. doi: 10.1111/j.1937-2817.2010.tb01236.x. Assessment, M. E. (2005) Ecosystems and human well-being: Synthesis, Island Press. Washington, DC. doi: 10.1196/annals.1439.003. Austin, M. P. (2002) ‘Spatial prediction of species distribution: An interface between ecological theory and statistical modelling’, Ecological Modelling, 157(2–3), pp. 101–118. doi: 10.1016/S0304-3800(02)00205-3. Ball, I. R., Possingham, H. P. and Watts, M. E. (2009) ‘Marxan and relatives: Software for spatial conservation prioritisation’, in Moilanen, A., Wilson, K. A., and Possingham, H. P. (eds) Spatial conservation prioritisation: Quantitiative methods and computational tools. Oxford, UK: Oxford University Press, pp. 185–195. Ban, N. C. et al. (2013) ‘A social-ecological approach to conservation planning: Embedding social considerations’, Frontiers in Ecology and the Environment, pp. 194–202. doi: 10.1890/110205. Barrett, C. B. et al. (2001) ‘Conserving Tropical Biodiversity amid Weak Institutions’, BioScience. Oxford University Press, 51(6), pp. 497–502. doi: 10.1641/0006-3568(2001)051[0497:ctbawi]2.0.co;2. Basille, M. et al. (2013) ‘Selecting Habitat to Survive: The Impact of Road Density on Survival in a Large Carnivore’, PLoS ONE. Edited by M. S. Boyce. Public Library of Science, 8(7), p. e65493. doi: 10.1371/journal.pone.0065493. Bennett, J. R. et al. (2015) ‘Biodiversity gains from efficient use of private sponsorship for flagship species conservation’, Proceedings of the Royal Society B: Biological Sciences, 282(1805), pp. 1–7. doi: 10.1098/rspb.2014.2693. Beyer, H. L. et al. (2016) ‘Solving conservation planning problems with integer linear programming’, Ecological Modelling. Elsevier B.V., 328, pp. 14–22. doi: 10.1016/j.ecolmodel.2016.02.005. Bino, G., Ramp, D. and Kingsford, R. T. (2014) ‘Identifying minimal sets of survey techniques for multi-species monitoring across landscapes: an approach utilising species distribution models’, International Journal of Geographical Information Science, 28(8), pp. 1674–1708. doi: 10.1080/13658816.2013.871016. Bowen-Jones, E. and Entwistle, A. (2002) ‘Identifying appropriate flagship species: the 58 importance of culture and local contexts’, Oryx. Cambridge University Press, 36(02), pp. 189–195. doi: 10.1017/S0030605302000261. Breckheimer, I. et al. (2014) ‘Defining and Evaluating the Umbrella Species Concept for Conserving and Restoring Landscape Connectivity’, Conservation Biology, 28(6), pp. 1584–1593. doi: 10.1111/cobi.12362. Burnham, K. P. and Anderson, D. R. (2004) ‘Multimodel inference: understanding AIC and BIC in model selection.’, Sociological Methods and Research, 33, pp. 261–304. Bush, M. B., Silman, M. R. and Urrego, D. H. (2004) ‘48 , 000 Years of Climate and Forest Change in a Biodiversity Hot Spot’, Science, 303(February), pp. 827–829. doi: 10.1126/science.1090795. Buytaert, W. et al. (2006) ‘Human impact on the hydrology of the Andean páramos’, Earth-Science Reviews, 79(1–2), pp. 53–72. doi: 10.1016/j.earscirev.2006.06.002. Caro, T. M. (2003) ‘Umbrella species: Critique and lessons from East Africa’, Animal Conservation. Wiley/Blackwell (10.1111), 6(2), pp. 171–181. doi: 10.1017/S1367943003003214. Carroll, C. et al. (2004) Extinction Debt of Protected Areas in Developing Landscapes, Conservation Biology. Available at: https://onlinelibrary.wiley.com/doi/pdf/10.1111/j.1523-1739.2004.00083.x (Accessed: 7 August 2018). Carroll, C., Noss, R. F. and Paquet, P. C. (2001) ‘Carnivores as focal species for conservation planning in the rocky mountain region’, Ecological Applications, 11(4), pp. 961–980. doi: 10.2307/3061005. Castellanos, A. (2011) ‘Andean bear home ranges in the Intag region, Ecuador’, Ursus, 22(1), pp. 65–73. doi: 10.2192/URSUS-D-10-00006.1. Červinka, J. et al. (2015) ‘Large-scale evaluation of carnivore road mortality: the effect of landscape and local scale characteristics’, Mammal Research. Springer Berlin Heidelberg, 60(3), pp. 233–243. doi: 10.1007/s13364-015-0226-0. Chan, K. M. A. et al. (2012) ‘Where are Cultural and Social in Ecosystem Services? A Framework for Constructive Engagement’, BioScience, 62(8), pp. 744–756. doi: 10.1525/bio.2012.62.8.7. Chester, C. C. (2015) ‘Yellowstone to Yukon: Transborder conservation across a vast international landscape’, Environmental Science and Policy, 49, pp. 75–84. doi: 10.1016/j.envsci.2014.08.009. Clark, D. B., Palmer, M. W. and Clark, D. A. (1999) ‘Edaphic factors and the landscape-scale distributions of tropical rain forest trees’, Ecology, 80, pp. 2662–2675. Clark, J. D. et al. (1993) ‘A multivariate model of Female Black Bear Habitat use for a Geographic Information System’, The Journal of Wildlife Management, 57(3), pp. 519–526. Available at: http://www.jstor.org/stable/3809276 (Accessed: 31 January 2018). Cuesta, F., Peralvo, M. F. and Van Manen, F. T. (2003) ‘Andean Bear Habitat Use in the Oyacachi River Basin’, Ursus, 14(2), pp. 198–209. doi: 10.2307/3873019. Cushman, S. A. and Lewis, J. S. (2010) ‘Movement behavior explains genetic differentiation in American black bears’, Landscape Ecology, 25(10), pp. 1613–1625. doi: 10.1007/s10980-010-9534-6. 59 Cushman, S. A., McKelvey, K. S. and Schwartz, M. K. (2009) ‘Use of Empirically Derived Source-Destination Models to Map Regional Conservation Corridors’, Conservation Biology. Wiley/Blackwell (10.1111), 23(2), pp. 368–376. doi: 10.1111/j.1523-1739.2008.01111.x. Dahle, B. and Swenson, J. E. (2003) ‘Home ranges in adult Scandinavian brown bears (Ursus arctos): Effect of mass, sex, reproductive category, population density and habitat type’, Journal of Zoology. Cambridge University Press, 260(4), pp. 329–335. doi: 10.1017/S0952836903003753. Defourney, P. (2015) ‘ESA Land Cover Climate Change Initiative (ESA LC_cci) data: Global land cover mapping at 300m’, Centre for Environmental Data Analysis. Dinerstein, E. et al. (2017) ‘An Ecoregion-Based Approach to Protecting Half the Terrestrial Realm’, BioScience. Oxford University Press, 67(6), pp. 534–545. doi: 10.1093/biosci/bix014. Dirzo, R. et al. (2014) ‘Defaunation in the Anthropocene’, Science, 345(6195), pp. 401–406. doi: 10.1126/science.1251817. Doak, D. F. (1995) ‘Source-Sink Models and the Problem of Habitat Degradation: General Models and Applications to the Yellowstone Grizzly’, Conservation Biology, 9(6), pp. 1370–1379. doi: 10.1046/j.1523-1739.1995.09061370.x. Durán, A. P. et al. (2014) ‘Exclusion of agricultural lands in spatial conservation prioritization strategies: consequences for biodiversity and ecosystem service representation.’, Proceedings. Biological sciences / The Royal Society, 281(1792), pp. 323–324. doi: 10.1098/rspb.2014.1529. Elith, J. et al. (2006) ‘Novel methods improve prediction of species’ distributions from occurrence data’, Ecography, 29(2), pp. 129–151. doi: 10.1111/j.2006.0906-7590.04596.x. Elith, J. and Leathwick, J. (2009) ‘Species distribution models: ecological explanation and prediction across space and time’, Annual Review of Ecology, Evolution, …, 40, pp. 677–697. doi: 10.1146/annurev.ecolsys.110308.120159. Espinosa, S. and Jacobson, S. K. (2012) ‘Human-wildlife conflict and environmental education: Evaluating a community program to protect the andean bear in ecuador’, Journal of Environmental Education, 43(1), pp. 55–65. doi: 10.1080/00958964.2011.579642. ESRI (2011) ‘ArcGIS Desktop: Release 10.’ Redlands, CA: Environmental Systems Research Institute. Falcucci, A. et al. (2009) ‘Assessing habitat quality for conservation using an integrated occurrence-mortality model’, Journal of Applied Ecology, 46(3), pp. 600–609. doi: 10.1111/j.1365-2664.2009.01634.x. Fernández, N. et al. (2012) ‘Brown bears at the edge: Modeling habitat constrains at the periphery of the Carpathian population’, Biological Conservation, 153, pp. 134–142. doi: 10.1016/j.biocon.2012.04.013. Figueroa, J., Stucchi, M. and Rojas-Verapinto, R. (2016) ‘Modelación de la distribución del oso andino Tremarctos ornatus en el bosque seco del Marañón (Perú)’, Revista Mexicana de Biodiversidad. Universidad Nacional Autónoma de México, Instituto de Biología, 87(1), pp. 230–238. doi: 10.1016/j.rmb.2016.01.008. Fisher, J. T. and Burton, A. C. (2018) ‘Wildlife winners and losers in an oil sands landscape’, Frontiers in Ecology and the Environment. Wiley-Blackwell. doi: 10.1002/fee.1807. 60 Flagstad, O. et al. (2001) ‘Environmental change and rates of evolution: the phylogeographic pattern within the hartebeest complex as related to climatic variation’, Proceedings of the Royal Society B: Biological Sciences, 268(1468), pp. 667–677. doi: 10.1098/rspb.2000.1416. Flather, C. H. et al. (1997) ‘Identifying gaps in conservation networks: Of indicators and uncertainty in geographic-based analyses’, Ecological Applications. Ecological Society of America, 7(2), pp. 531–542. doi: 10.1890/1051-0761(1997)007[0531:IGICNO]2.0.CO;2. Forman, R. T. T. and Alexander, L. E. (1998) ‘Roads and Their Major Ecological Effects’, Annual Review of Ecology and Systematics, 29(1), pp. 207–231. doi: 10.1146/annurev.ecolsys.29.1.207. Gantchoff, M. G. and Belant, J. L. (2017) ‘Regional connectivity for recolonizing American black bears (Ursus americanus) in southcentral USA’, Biological Conservation, 214, pp. 66–75. doi: 10.1016/j.biocon.2017.07.023. García-Rangel, S. (2012) ‘Andean bear Tremarctos ornatus natural history and conservation’, Mammal Review, pp. 85–119. doi: 10.1111/j.1365-2907.2011.00207.x. Garshelis, D. L. (2011) ‘Andean bear density and abundance estimates — How reliable and useful are they?’, Ursus, 22(1), pp. 47–64. doi: 10.2192/URSUS-D-10-00030.1. Gassert, F. et al. (2014) Aqueduct global maps 2.1: Constructing decision-relevant global water risk indicators. Washington, DC. Available at: http://www.wri.org/publication/aqueduct-global- (Accessed: 5 July 2018). Goldstein, I. (2002) ‘Andean bear-cattle interactions and tree nest use in Bolivia and Venezuela’, Ursus, 13(2002), pp. 369–372. Available at: http://www.jstor.org.ezproxy.library.ubc.ca/stable/pdf/3873218.pdf (Accessed: 26 May 2018). Goldstein, I. et al. (2006) ‘Andean bear–livestock conflicts: a review’, Ursus, 17, pp. 8–15. Available at: https://www.bearbiology.com/wp-content/uploads/2017/10/Goldstein_Paisley_et_al_Vol_17_1_.pdf (Accessed: 12 January 2018). Goldstein, I. et al. (2008) ‘Tremarctos ornatus’, The IUCN Red List of Threatened Species. Gonzales, E. K. et al. (2003) ‘Strategic reserve design in the central coast of British Columbia: integrating ecological and industrial goals’, Canadian Journal of Forest Research, 33(11), pp. 2129–2140. doi: 10.1139/x03-133. Gray, C. L. et al. (2016) ‘Local biodiversity is higher inside than outside terrestrial protected areas worldwide’, Nature Communications. Nature Publishing Group, 7, p. 12306. doi: 10.1038/ncomms12306. Hamann, A. et al. (2013) ‘A Comprehensive, High-Resolution Database of Historical and Projected Climate Surfaces for Western North America’, Bulletin of the American Meteorological Society.  American Meteorological Society , 94(9), pp. 1307–1309. doi: 10.1175/BAMS-D-12-00145.1. Hansen, M. C. et al. (2013) ‘High-Resolution Global Maps of 21st-Century Forest Cover Change’, Science, 342, pp. 850–853. doi: 10.1126/science.1239552. Hanson, J. et al. (2018) ‘prioritizr: Systematic Conservation Prioritization in R’. Available at: https://github.com/prioritizr/prioritizr. 61 Harden, C. P. (2006) ‘Human impacts on headwater fluvial systems in the northern and central Andes’, Geomorphology, 79(3–4), pp. 249–263. doi: 10.1016/j.geomorph.2006.06.021. Harvey, V., Côté, S. D. and Hammill, M. O. (2008) ‘The ecology of 3-D space use in a sexually dimorphic mammal’, Ecography, 31(3), pp. 371–380. doi: 10.1111/j.0906-7590.2008.05218.x. Hoekstra, J. M. et al. (2005) ‘Confronting a biome crisis: Global disparities of habitat loss and protection’, Ecology Letters, 8(1), pp. 23–29. doi: 10.1111/j.1461-0248.2004.00686.x. Homeier, J. et al. (2010) ‘Tree diversity, forest structure and productivity along altitudinal and topographical gradients in a species-rich Ecuadorian montane rain forest’, Biotropica. Wiley/Blackwell (10.1111), 42(2), pp. 140–148. doi: 10.1111/j.1744-7429.2009.00547.x. Van Horn, R. C. et al. (2014) ‘Human visual identification of individual Andean bears Tremarctos ornatus’, Wildlife Biology. Nordic Board for Wildlife Research, 20(5), pp. 291–299. doi: 10.2981/wlb.00023. Van Horn, R. C. et al. (2015) ‘Photos provide information on age, but not kinship, of Andean bear’, PeerJ. PeerJ Inc., 3, p. e1042. doi: 10.7717/peerj.1042. Howlett, M., Rayner, J. and Tollefson, C. (2009) ‘From government to governance in forest planning? Lessons from the case of the British Columbia Great Bear Rainforest initiative’, Forest Policy and Economics. Elsevier B.V., 11(5–6), pp. 383–391. doi: 10.1016/j.forpol.2009.01.003. IUCN (2018) The IUCN Red List of Threatened Species. Version 2018-1. Available at: http://www.iucnredlist.org (Accessed: 5 July 2018). Janzen, D. H. (1988) ‘Tropical Dry Forests: The Most Endangered Major Tropical Ecosystem’, in Wilson, E. O. and Peter, F. M. (eds) Biodiversity. Washington, DC: National Academies Press (US). Available at: https://www.ncbi.nlm.nih.gov/books/NBK219281/ (Accessed: 11 June 2018). Jenkins, C. N. and Van Houtan, K. S. (2016) ‘Global and regional priorities for marine biodiversity protection’, Biological Conservation, 204(November), pp. 333–339. doi: 10.1016/j.biocon.2016.10.005. Jenkins, C. N., Pimm, S. L. and Joppa, L. N. (2013) ‘Global patterns of terrestrial vertebrate diversity and conservation’, Proceedings of the National Academy of Sciences. National Academy of Sciences, 110(28), pp. E2602–E2610. doi: 10.1073/pnas.1302251110. Jorgenson, J. P. and Sandoval-A, S. (2005) ‘Andean bear management needs and interactions with humans in Colombia’, Ursus, 16(1), pp. 108–116. doi: 10.2192/1537-6176(2005)016[0108:ABMNAI]2.0.CO;2. Karanth, K. U. and Nichols, J. D. (1998) ‘Estimation of tiger densities in India using photographic captures and recaptures’, Ecology. Ecological Society of America, 79(8), pp. 2852–2862. doi: 10.1890/0012-9658(1998)079[2852:EOTDII]2.0.CO;2. Kattan, G. et al. (2004) ‘Range fragmentation of the spectacled bear tremarctos ornatus in the northern Andes’, Oryx, 38(2), pp. 1–7. doi: 10.1017/S0030605303000000. Keane, A. et al. (2008) ‘The sleeping policeman: Understanding issues of enforcement and compliance in conservation’, Animal Conservation, 11(2), pp. 75–82. doi: 10.1111/j.1469-1795.2008.00170.x. 62 Kintz, D. B., Young, K. R. and Crews-Meyer, K. A. (2006) ‘Implications of Land Use/Land Cover Change in the Buffer Zone of a National Park in the Tropical Andes’, Environmental Management. Springer-Verlag, 38(2), pp. 238–252. doi: 10.1007/s00267-005-0147-9. Lamb, C. T. et al. (2018) ‘Effects of habitat quality and access management on the density of a recovering grizzly bear population’, Journal of Applied Ecology, (November 2017), pp. 1–12. doi: 10.1111/1365-2664.13056. Laurance, W. F. et al. (2012) ‘Averting biodiversity collapse in tropical forest protected areas’, Nature. Nature Publishing Group, 489(7415), pp. 290–293. doi: 10.1038/nature11318. Linares-Palomino, R. et al. (2010) ‘Diversity and endemism of woody plant species in the Equatorial Pacific seasonally dry forests’, Biodiversity and Conservation. Springer Netherlands, 19(1), pp. 169–185. doi: 10.1007/s10531-009-9713-4. Liu, F. et al. (2009) ‘Spatial distribution as a measure of conservation needs: An example with Asiatic black bears in south-western China’, Diversity and Distributions. Blackwell Publishing Ltd, 15(4), pp. 649–659. doi: 10.1111/j.1472-4642.2009.00571.x. Macarthur, R. H. et al. (1992) ‘the Problem of Pattern and Scale in Ecology’, Ecology, 73(6), pp. 1943–1967. doi: https://doi.org/10.2307/1941447. Manly, B. F. J. et al. (2002) Resource selection by animals: Statistical design and analysis for field studies. Second. Dordrecht, The Netherlands: Kluwer Academic Publishers. Maréchaux, I., Rodrigues, A. S. L. and Charpentier, A. (2017) ‘The value of coarse species range maps to inform local biodiversity conservation in a global context’, Ecography. Wiley/Blackwell (10.1111), 40(10), pp. 1166–1176. doi: 10.1111/ecog.02598. Margules, C. R. and Pressey, R. L. (2000) ‘Systematic conservation planning.’, Nature, 405(6783), pp. 243–53. doi: 10.1038/35012251. Margules, C. R., Pressey, R. L. and Williams, P. H. (2002) ‘Representing biodiversity: Data and procedures for identifying priority areas for conservation’, Journal of Biosciences. Springer India, 27(4), pp. 309–326. doi: 10.1007/BF02704962. McBride, M. F. et al. (2007) ‘Incorporating the effects of socioeconomic uncertainty into priority setting for conservation investment’, Conservation Biology, 21(6), pp. 1463–1474. doi: 10.1111/j.1523-1739.2007.00832.x. McCarthy, D. P. et al. (2012) ‘Financial costs of meeting global biodiversity conservation targets: Current spending and unmet needs’, Science, 338(6109), pp. 946–949. doi: 10.1126/science.1229803. McLellan, B. N. and Shackleton, D. M. (1988) ‘Grizzly Bears and Resource-Extraction Industries - Effects of Roads on Behavior, Habitat Use and Demography’, Journal of Applied Ecology, 25(2), pp. 451–460. doi: 2. Mcrae, B. H. et al. (2008) Using Circuit Theory to Model Connectivity in Ecology, Evolution, and Conservation, Source: Ecology. Available at: http://www.ripuc.org/efsb/efsb/SB2015_06_Invenergy_comings_J.pdf (Accessed: 26 September 2018). McRae, B. H. and Beier, P. (2007) Circuit theory predicts gene flow in plant and animal populations., Proceedings of the National Academy of Sciences of the United States of America. doi: 10.1073/pnas.0706568104. 63 McRae, B. H., Shah, V. and Mohapatra, T. (2013) ‘Circuitscape 4 User Guide’, The Nature Conservancy. Available at: www.circuitscape.org (Accessed: 5 July 2018). Molina, S. et al. (2017) ‘Use of spatial capture–recapture to estimate density of Andean bears in northern Ecuador’, Ursus, 28(1), pp. 117–126. doi: 10.2192/URSU-D-16-00030.1. Morris, D. W. (1987) Ecological scale and habitat use, Ecology. doi: 10.2307/1939267. Mulligan, M. et al. (2010) ‘The andes basins: Biophysical and developmental diversity in a climate of change’, Water International, 35(5), pp. 472–492. doi: 10.1080/02508060.2010.516330. Myers, N. et al. (2000) ‘Biodiversity hotspots for conservation priorities’, Nature. Nature Publishing Group, 403(6772), pp. 853–858. doi: 10.1038/35002501. Newmark, W. D. (2008) ‘Isolation of African protected areas’, Frontiers in Ecology and the Environment, 6(6), pp. 321–328. doi: 10.1890/070003. Nielsen, S. E. et al. (2010) ‘Dynamic wildlife habitat models: Seasonal foods and mortality risk predict occupancy-abundance and habitat selection in grizzly bears’, Biological Conservation, 143(7), pp. 1623–1634. doi: 10.1016/j.biocon.2010.04.007. Nielsen, S. E. (2011) ‘Relationships between grizzly bear source-sink habitats and prioritized biodiversity sites in Central British Columbia’, BC Journal of Ecosystems and Management, 12(1), pp. 136–147. Available at: P & D. Olson, D. M. et al. (2001) ‘Terrestrial Ecoregions of the World: A New Map of Life on Earth’, BioScience. Oxford University Press, 51(11), p. 933. doi: 10.1641/0006-3568(2001)051[0933:TEOTWA]2.0.CO;2. Ordiz, A. et al. (2014) ‘Brown bear circadian behavior reveals human environmental encroachment’, Biological Conservation. Elsevier, 173, pp. 1–9. doi: 10.1016/j.biocon.2014.03.006. Orians, G. H. and Wittenberger, J. F. (1991) Spatial and Temporal Scales in Habitat Selection, The American Naturalist. doi: 10.1086/285138. Ostrom, E. (1992) Governing the commons: The evolution of institutions for collective action, Natural Resources Jounal. doi: 10.1017/CBO9781316423936. Otis, D. L. et al. (1978) ‘Statistical inference from capture data on closed animal populations’, Wildlife Monographs, (62), pp. 3–135. doi: 10.2307/2287873. Peacock, E. et al. (2011) ‘Conservation and management of Canada’s polar bears (Ursus maritimus) in a changing Arctic’, Canadian Journal of Zoology, 89(5), pp. 371–385. doi: 10.1139/z11-021. Peralvo, M. F., Cuesta, F. and van Manen, F. (2005) ‘Delineating priority habitat areas for the conservation of Andean bears in northern Ecuador’, Ursus, 16, pp. 222–233. Available at: https://www.bearbiology.com/publications/ursus-archive/delineating-priority-habitat-areas-for-the-conservation-of-andean-bears-in-northern-ecuador/ (Accessed: 12 January 2018). Peterman, W. E., Crawford, J. A. and Kuhns, A. R. (2013) ‘Using species distribution and occupancy modeling to guide survey efforts and assess species status’, Journal for Nature Conservation. Elsevier GmbH., 21(2), pp. 114–121. doi: 10.1016/j.jnc.2012.11.005. Peyton, B. (1980) ‘Ecology, Distribution, and Food Habits of Spectacled Bears, Tremarctos ornatus, in Peru’, Journal of Mammalogy, 61(4), pp. 639–652. doi: 10.2307/1380309. 64 Peyton, B. et al. (1995) ‘Status Of Wild Andean Bears And Policies For Their Management’, Ursus, 10, pp. 87–100. Peyton, B. (1999) ‘Spectacled bear conservation action plan’, Bears: status survey and conservation action plan, pp. 157–198. Available at: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.196.5692&rep=rep1&type=pdf. Phillips, S., Dudík, M. and Schapire, R. (2004) ‘A maximum entropy approach to species distribution modeling’, Proceedings of the Twenty-First International Conference on Machine Learning, pp. 655–662. doi: 10.1145/1015330.1015412. Pimm, S. L. et al. (1995) The Future of Biodiversity, Science. Available at: https://www-jstor-org.ezproxy.library.ubc.ca/stable/pdf/2888268.pdf?refreqid=excelsior%3Af10678ffc4f68519bfeab46bb13efb08 (Accessed: 7 September 2018). Pressey, R. L., Possingham, H. P. and Margules, C. R. (1996) ‘Optimality in reserve selection algorithms: When does it matter and how much?’, Biological Conservation, 76(3), pp. 259–267. doi: 10.1016/0006-3207(95)00120-4. R Core Development Team (2014) ‘A language and environment for statistical computing’. Vienna, Austria: R Foundation for Statistical Computing. Available at: http://www.r-project.org/. Rabinowitz, A. and Zeller, K. A. (2010) ‘A range-wide model of landscape connectivity and conservation for the jaguar, Panthera onca’, Biological Conservation, 143(4), pp. 939–945. doi: 10.1016/j.biocon.2010.01.002. Rajpurohit, K. S. and Krausman, P. R. (2000) ‘Human-sloth-bear conflicts in Madhya Pradesh, India’, Wildlife Society Bulletin, 28(2), pp. 393–399. Available at: http://www.jstor.org/stable/3783697 (Accessed: 13 March 2018). Ratnayeke, S. et al. (2007) ‘Landscape characteristics of sloth bear range in Sri Lanka’, Ursus, 18(1), pp. 189–202. doi: 10.2192/1537-6176(2007)18. Rice, M. B. et al. (2009) ‘Habitat-distribution modeling of a recolonizing Black Bear, Ursus americanus, Population in the Trans-Pecos region of Texas’, Canadian Field-Naturalist, 123(3), pp. 246–254. doi: 10.22621/cfn.v123i3.972. Rios-Uzeda, B., Gomez, H. and Wallace, R. B. (2006) ‘Habitat preferences of the Andean bear (Tremarctos ornatus) in the Bolivian Andes’, Journal of Zoology. Blackwell Science Ltd, 268(3), pp. 271–278. doi: 10.1111/j.1469-7998.2005.00013.x. Ríos-Uzeda, B., Gómez, H. and Wallace, R. B. (2007) ‘A preliminary density estimate for Andean bear using camera-trapping methods’, Ursus, 18(1), pp. 124–128. doi: 10.2192/1537-6176(2007)18[124:APDEFA]2.0.CO;2. Roberge, J. M. and Angelstam, P. (2004) ‘Usefulness of the Umbrella Species Concept as a Conservation Tool’, Conservation Biology. Wiley/Blackwell (10.1111), 18(1), pp. 76–85. doi: 10.1111/j.1523-1739.2004.00450.x. Rodríguez, E. F. et al. (2007) ‘Capparis scabrida (Capparaceae) a species from Peru and Ecuador in urgent need of conservation plans’, Arnaldoa, 14, pp. 269–282. Available at: https://www.researchgate.net/profile/Rainer_Bussmann/publication/228108008_Capparis_scabrida_Capparaceae_una_especie_del_Peru_y_Ecuador_que_necesita_planes_de_conservacion_urgente_Capparis_scabrida_Capparaceae_a_species_from_Peru_and_Ecuador_in_urgent_nee (Accessed: 17 January 2018). 65 Rodríguez, L. O. and Young, K. R. (2000) ‘Biological Diversity of Peru: Determining Priority Areas for Conservation’, AMBIO: A Journal of the Human Environment, 29(6), pp. 329–337. doi: 10.1579/0044-7447-29.6.329. Rodríguez, R. et al. (2005) ‘“El Niño” events recorded in dry-forest species of the lowlands of northwest Peru’, Dendrochronologia. Urban & Fischer, 22(3), pp. 181–186. doi: 10.1016/J.DENDRO.2005.05.002. Royle, J. A. et al. (2005) ‘Modelling Occurrence and Abundance of Species When Detection Is Imperfect’, Oikos, 110(2), pp. 353–359. Royle, J. A. et al. (2014) Spatial Capture-Recapture, Spatial Capture-recapture. Oxford, UK: Academic Press. doi: 10.1016/B978-0-12-405939-9.00020-7. Sánchez-Mercado, A. et al. (2008) ‘Factors affecting poaching risk to Vulnerable Andean bears Tremarctos ornatus in the Cordillera de Merida, Venezuela: space, parks and people’, Oryx, 42(3), pp. 437–447. doi: 10.1017/S0030605308006996. Sánchez-Mercado, A. et al. (2014) ‘Combining threat and occurrence models to predict potential ecological traps for Andean bears in the Cordillera de Mérida, Venezuela’, Animal Conservation, 17(4), pp. 388–398. doi: 10.1111/acv.12106. Schoen, J. W. (1990) ‘Bear habitat management: a review and future perspective’, International Conference on Bear Research and Management, 8(Table 2), pp. 143–154. Available at: P & D (Accessed: 11 April 2018). Schuster, R. et al. (2018) ‘Optimizing conservation of migratory species over their full annual cycle in the Western Hemisphere’, bioRxiv. Cold Spring Harbor Laboratory, 268805, p. 268805. doi: https://doi.org/10.1101/268805. Schuster, R. and Arcese, P. (2013) ‘Using bird species community occurrence to prioritize forests for old growth restoration’, Ecography, 36(4), pp. 499–507. doi: 10.1111/j.1600-0587.2012.07681.x. Schwartz, M. W. (1999) ‘Choosing the Appropriate Scale of Reserves for Conservation’, Annual Review of Ecology and Systematics.  Annual Reviews  4139 El Camino Way, P.O. Box 10139, Palo Alto, CA 94303-0139, USA  , 30(1), pp. 83–108. doi: 10.1146/annurev.ecolsys.30.1.83. Scotson, L. et al. (2017) ‘Projecting range-wide sun bear population trends using tree cover and camera-trap bycatch data’, PLoS ONE, 12(9). doi: 10.1371/journal.pone.0185336. Secretariat CBD (2010) The Strategic Plan for Biodiversity 2011-2020 and the Aichi Biodiversity Targets, Document UNEP/CBD/COP/DEC/X/2. Nagoya, Japan. doi: 10.1111/cobi.12383. Servheen, C., Herrero, S. and Peyton, B. (1999) ‘Status, survey and conservation action plan: Bears’, IUCN/SSC Bear Specialist Group. 321pp, pp. 1–310. Available at: http://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:Status+survey+andconservation+action+plan+Bears#5%5Cnhttp://scholar.google.com/scholar?hl=en&btnG=Search&q=intitle:Status+survey+and+conservation+action+plan+Bears%235. Sharma, R. K., Bhatnagar, Y. V. and Mishra, C. (2015) ‘Does livestock benefit or harm snow leopards?’, Biological Conservation. Elsevier, 190, pp. 8–13. doi: 10.1016/j.biocon.2015.04.026. Sierra, R., Campos, F. and Chamberlin, J. (2002) ‘Assessing biodiversity conservation priorities: Ecosystem risk and representativeness in continental Ecuador’, Landscape and Urban 66 Planning. Elsevier, 59(2), pp. 95–110. doi: 10.1016/S0169-2046(02)00006-3. Simek, S. L. et al. (2015) ‘Source populations and roads affect American black bear recolonization’, European Journal of Wildlife Research. Springer Berlin Heidelberg, 61(4), pp. 583–590. doi: 10.1007/s10344-015-0933-5. Sollmann, R., Gardner, B. and Belant, J. L. (2012) ‘How does spatial study design influence density estimates from spatial capture-recapture models?’, PLoS ONE, 7(4), p. 34575. doi: 10.1371/journal.pone.0034575. Spracklen, D. V. and Righelato, R. (2016) ‘Carbon storage and sequestration of re-growing montane forests in southern Ecuador’, Forest Ecology and Management. Elsevier, 364, pp. 139–144. doi: 10.1016/j.foreco.2016.01.001. Stillfried, M. et al. (2015) ‘When top predators become prey: Black bears alter movement behaviour in response to hunting pressure’, Behavioural Processes. Elsevier, 120, pp. 30–39. doi: 10.1016/j.beproc.2015.08.003. Sutherland, C., Fuller, A. K. and Royle, J. A. (2015) ‘Modelling non-Euclidean movement and landscape connectivity in highly structured ecological networks’, Methods in Ecology and Evolution. Edited by O. Gimenez, 6(2), pp. 169–177. doi: 10.1111/2041-210X.12316. Takahata, C. et al. (2014) ‘Habitat selection of a large carnivore along human-wildlife boundaries in a highly modified landscape’, PLoS ONE, 9(1). doi: 10.1371/journal.pone.0086181. Thorne, J. H., Cameron, D. and Quinn, J. F. (2006) ‘A Conservation Design for the Central Coast of California and the Evaluation of Mountain Lion as an Umbrella Species’, Natural Areas Journal.  Natural Areas Association , 26(2), pp. 137–148. doi: 10.3375/0885-8608(2006)26[137:ACDFTC]2.0.CO;2. Toews, M., Juanes, F. and Burton, A. C. (2017) ‘Mammal responses to human footprint vary with spatial extent but not with spatial grain’, Ecosphere, 8(3). doi: 10.1002/ecs2.1735. Tovar, C. et al. (2013) ‘Diverging Responses of Tropical Andean Biomes under Future Climate Conditions’, PLoS ONE. Edited by K. A. Crandall. Public Library of Science, 8(5), p. e63634. doi: 10.1371/journal.pone.0063634. Treves, A. et al. (2004) ‘Predicting Human-Carnivore Conflict: A Spatial Model Derived from 25 Years of Data on Wolf Predation on Livestock’, Conservation Biology, 18(1), pp. 114–125. doi: 10.1111/j.1523-1739.2004.00189.x. Treves, A. and Karanth, K. U. (2003) ‘Human-Carnivore Conflict and Perspectives on Carnivore Management Worldwide’, Conservation Biology, 17(6), pp. 1491–1499. doi: 10.1111/j.1523-1739.2003.00059.x. Trombulak, S. C. and Frissell, C. A. (2000) ‘Review of ecological effects of roads on terrestrial and aquatic communities’, Conservation Biology. Wiley/Blackwell (10.1111), 14(1), pp. 18–30. doi: 10.1046/j.1523-1739.2000.99084.x. Tucker, M. A. et al. (2018) ‘Moving in the Anthropocene: Global reductions in terrestrial mammalian movements’, Science, 359, pp. 466–469. Available at: http://science.sciencemag.org.ezproxy.library.ubc.ca/content/sci/359/6374/466.full.pdf (Accessed: 5 March 2018). UNEP-WCMC and IUCN (2016) Protected Planet Report 2016. Cambridge, UK. Available at: https://wdpa.s3.amazonaws.com/Protected_Planet_Reports/2445 Global Protected Planet 67 2016_WEB.pdf (Accessed: 15 May 2018). Urrutia, R. and Vuille, M. (2009) ‘Climate change projections for the tropical Andes using a regional climate model: Temperature and precipitation simulations for the end of the 21st century’, Journal of Geophysical Research Atmospheres. Wiley-Blackwell, 114(2), p. D02108. doi: 10.1029/2008JD011021. Velez–Liendo, X., Adriaensen, F. and Matthysen, E. (2014) ‘Landscape assessment of habitat suitability and connectivity for Andean bears in the Bolivian Tropical Andes’, Ursus, 25(2), pp. 172–187. doi: 10.2192/URSUS-D-14-00012.1. Velez–Liendo, X., Strubbe, D. and Matthysen, E. (2013) ‘Effects of variable selection on modelling habitat and potential distribution of the Andean bear in Bolivia’, Ursus, 24(2), pp. 127–138. doi: 10.2192/URSUS-D-12-00027R4.1. Venter, O. et al. (2016) ‘Sixteen years of change in the global terrestrial human footprint and implications for biodiversity conservation’, Nature Communications, 7. doi: 10.1038/ncomms12558. Vera, C. et al. (2006) ‘Climate change scenarios for seasonal precipitation in South America from IPCC-AR4 models’, Geophysical Research Letters, 33(13), pp. 1–5. doi: 10.1029/2006GL025759. Viana, D. S. et al. (2018) ‘Linking seasonal home range size with habitat selection and movement in a mountain ungulate’, Movement Ecology. BioMed Central, 6(1), p. 1. doi: 10.1186/s40462-017-0119-8. Wang, F. et al. (2018) ‘Does one size fit all? A multispecies approach to regional landscape corridor planning’, Diversity and Distributions. Edited by J. Lahoz-Monfort. Wiley/Blackwell (10.1111), 24(3), pp. 415–425. doi: 10.1111/ddi.12692. Wassenaar, T. et al. (2007) ‘Projecting land use changes in the Neotropics: The geography of pasture expansion into forest’, Global Environmental Change. Pergamon, 17(1), pp. 86–104. doi: 10.1016/J.GLOENVCHA.2006.03.007. Weaver, J. L. et al. (1996) ‘Resilience and Conservation of Large Carnivores in the Rocky Mountains’, Conservation Biology, 10(4), pp. 964–976. Available at: http://www.jstor.org/stable/2387133 (Accessed: 13 March 2018). Wheatley, M. and Johnson, C. (2009) ‘Factors limiting our understanding of ecological scale’, Ecological Complexity, 6(2), pp. 150–159. doi: 10.1016/j.ecocom.2008.10.011. Wilson, K. A. et al. (2007) ‘Conserving biodiversity efficiently: What to do, where, and when’, PLoS Biology, 5(9), pp. 1850–1861. doi: 10.1371/journal.pbio.0050223. Wilson, K. A., Carwardine, J. and Possingham, H. P. (2009) ‘Setting conservation priorities’, Annals of the New York Academy of Sciences, 1162, pp. 237–264. doi: 10.1111/j.1749-6632.2009.04149.x. Wong, W. M., Leader-Williams, N. and Linkie, M. (2013) ‘Quantifying changes in sun bear distribution and their forest habitat in Sumatra’, Animal Conservation, 16(2), pp. 216–223. doi: 10.1111/j.1469-1795.2012.00587.x. Woodroffe, R. (2000) ‘Predators and people: Using human densities to interpret declines of large carnivores’, Animal Conservation, 3(2), pp. 165–173. doi: 10.1017/S136794300000086X. 68 Woodroffe, R. et al. (2007) ‘Edge effects and the extinction of populations inside protected areas edge effects and the extinction of populations inside protected areas’, Science, 280(5372), pp. 2126–2128. doi: 10.1126/science.280.5372.2126. Wright, S. J. (2010) ‘The future of tropical forests’, Annals of the New York Academy of Sciences, 1195, pp. 1–27. doi: 10.1111/j.1749-6632.2010.05455.x. Xun, B., Yu, D. and Wang, X. (2017) ‘Prioritizing habitat conservation outside protected areas in rapidly urbanizing landscapes: A patch network approach’, Landscape and Urban Planning, 157, pp. 532–541. doi: 10.1016/j.landurbplan.2016.09.013. Yerena, E. (1998) ‘Protected Areas for the Andean Bear in South America’, Ursus, 10(1), pp. 101–106. Available at: https://www.bearbiology.com/publications/ursus-archive/protected-areas-for-the-andean-bear-in-south-america/ (Accessed: 12 January 2018). Yerena, E. and Torres, D. (1994a) ‘Spectacled Bear Conservation and Dispersal Corridors in Venezuela’, Bears: Their Biology and Management, 9(1), p. 169. doi: 10.2307/3872698. Yerena, E. and Torres, D. (1994b) ‘Spectacled Bear Conservation and Dispersal Corridors in Venezuela’, Bears: Their Biology and Management. International Association for Bear Research and Management, 9, p. 169. doi: 10.2307/3872698.   69 Appendix  A. Correlation Matrix for Chapter 2 Table A.1. Correlation matrix for spatial habitat variables used in initial and final modelling   pixel ID distance to roads elevation slope tree cover climate PC1 climate PC2 pixel ID 1.00       distance to road -0.04 1.00      elevation -0.30 -0.03 1.00     slope -0.17 0.14 0.53 1.00    tree cover -0.22 -0.02 0.47 0.41 1.00   climate PC1 -0.29 -0.02 0.99 0.48 0.41 1.00  climate PC2 -0.19 -0.06 0.12 0.33 0.43 0.00 1.00   70 Appendix  B. Model selection and MLE tables for chapter 2 Table B.1. Estimated average effect-size and standard error for all predictors of the density of Andean bears (Tremarctos ornatus) at Cerro Venado. Effect sizes indicate the standardized number of units of change in the predictor variable for a one-unit increase in the response. Coefficients are averaged by model weight across models within the model set containing 95% cumulative model weight. Adjusted estimate and adjusted standard errors include models where the parameter did not occur. Relative Value Index indicates the percentage of models that the parameter occurred in. Parameter Estimate Standard Error Adjusted Estimate Adjusted Standard Error Relative Value Index β (Nearest road) 2.6 0.68 2.6 0.68 1.00 β (Slope) -1.29 1.38 -0.46 0.85 0.36 β (Tree cover) 1.61 1.29 0.45 0.81 0.28 β (Elevation) -5.85 2.99 -5.31 3.28 0.91 D0 (Intercept) -7.51 2.47 -7.51 2.47 1.00 p0 (Intercept) -4.04 0.32 -4.04 0.32 1.00 p0 (sex (F) * season (dry)) -0.47 0.52 -0.47 0.52 1.00 p0 (sex (M) * season (wet)) -0.15 0.38 -0.15 0.38 1.00 p0 (sex (M) * season (dry)) 1.23 0.47 1.23 0.47 1.00 ψ constant 0.31 0.43 0.31 0.43 1.00 σ (Intercept) 1.13 0.07 1.13 0.07 1.00 σ (season (dry)) -0.45 0.1 -0.45 0.1 1.00    71 Table B.2. Estimated average effect-size and standard error for all predictors of the density of Andean bears (Tremarctos ornatus) at Laquipampa Wildlife Refuge. See table B.1 for definitions. Parameter Estimate Standard Error Estimate* Standard Error* Relative Value Index β (Nearest road) 3.87 1.22 3.87 1.22 1.00 β (Slope) 1.04 0.44 0.76 0.58 0.73 β (Tree cover) -2.32 3.06 -1.62 2.65 0.70 β (Elevation) 0.97 0.41 0.85 0.48 0.88 D0 (Intercept) -4.50 1.56 -4.50 1.56 1.00 p0 (Intercept) -4.41 0.25 -4.41 0.25 1.00 ψ constant 0.15 0.52 0.15 0.52 1.00 σ (Intercept) 0.91 0.11 0.91 0.11 1.00 σ (season (dry)) -0.36 0.22 -0.36 0.22 1.00 σ (season (dry))*sex (M)) 1.00 0.28 1.00 0.28 1.00 σ (sex (M)) -0.45 0.16 -0.45 0.16 1.00  Table B.3. Estimated average effect-size and standard error for all predictors of the density of Andean bears (Tremarctos ornatus) at Laquipampa Wildlife Refuge. See table B.1 for definitions. Parameter Estimate Standard Error Estimate* Standard Error* Relative Value Index β (Nearest road) 2.34 0.41 2.34 0.41 1.00 β (Slope) -1.26 0.53 -1.11 0.62 0.89 β (Tree cover) 0.66 0.34 0.44 0.42 0.68 β (Elevation) 0.70 0.45 0.27 0.42 0.39 D0 (Intercept) -3.08 0.30 -3.08 0.30 1.00 p0 (Intercept) -4.08 0.14 -4.08 0.14 1.00 ψ constant 0.07 0.30 0.07 0.30 1.00 σ (Intercept) 0.73 0.06 0.73 0.06 1.00 σ (season (dry)) -0.14 0.06 -0.14 0.06 1.00 σ (sex (M)) 0.33 0.07 0.33 0.07 1.00  72 Table B.4. Full model selection table of spatial-capture recapture models to predict the density of Andean bears (Tremarctos ornatus) in Cerro Venado. The best-fitting observation model (po ~ season*sex; σ ~season) is used in each model. K is the number of parameters in the model. ΔAIC is the difference in Akaike’s information criterion (AIC) between the best model and each successive model. wi is the Akaike weight, representing the probability that a model is the best in the model set. model log-likelihood K AIC ΔAIC weight cumulative weight elevation +  nearest road 989 10 1997.00 0.00 0.44 0.44 elevation + tree cover + nearest road 989 11 1999.00 1.60 0.20 0.64 elevation +  slope + nearest road 989 11 1999.00 1.70 0.19 0.83 slope + nearest road 990 10 2001.00 3.10 0.09 0.92 elevation + tree cover + slope + nearest road 988 12 2001.00 3.40 0.08 1.00 nearest road 1001 9 2020.00 22.80 0.00 1.00 slope 1001 9 2021.00 23.30 0.00 1.00 elevation 1001 9 2021.00 23.50 0.00 1.00 elevation +  slope 1001 10 2021.00 23.70 0.00 1.00 Null (~1) 1006 8 2029.00 31.40 0.00 1.00    73 Table B.5. Full model selection table of spatial-capture recapture models to predict the density of Andean bears (Tremarctos ornatus) in Laquipampa Wildlife Refuge. The best-fitting observation model (po ~ 1; σ ~season*sex) is used in each model. See Table B.4 for definitions. model log-likelihood K AIC ΔAIC weight cumulative weight elevation + tree cover + slope + nearest road 488 11 998.00 0.00 0.53 0.53 elevation +  nearest road 491 9 1001.00 2.80 0.13 0.66 elevation +  slope + nearest road 491 10 1002.00 3.70 0.09 0.75 slope + nearest road 492 9 1002.00 4.40 0.06 0.81 elevation + tree cover + nearest road 491 10 1003.00 4.40 0.06 0.87 tree cover + nearest road 492 9 1003.00 4.50 0.06 0.92 nearest road 493 8 1003.00 4.90 0.05 0.97 tree cover + slope + nearest road 492 10 1004.00 5.60 0.03 1.00 elevation +  slope 499 9 1017.00 18.60 0.00 1.00 slope 501 8 1017.00 19.10 0.00 1.00 elevation + slope +  tree cover 499 10 1017.00 19.20 0.00 1.00 slope +  tree cover 500 9 1018.00 20.40 0.00 1.00 elevation 503 8 1022.00 23.70 0.00 1.00 tree cover 503 8 1023.00 24.60 0.00 1.00 elevation +  tree cover 503 9 1024.00 25.70 0.00 1.00 1 505 7 1024.00 25.90 0.00 1.00    74 Table B.6. Full model selection table of spatial-capture recapture models to predict the density of Andean bears (Tremarctos ornatus) in pooled datasets from Cerro Venado and Laquipampa Wildlife Refuge. The best-fitting observation model (po ~ 1; σ ~season + sex) is used in each model. See Table B.4 for definitions. model log-likelihood K AIC ΔAIC weight cumulative weight tree cover + slope + nearest road 1519 9 3055.00 0.00 0.43 0.43 elevation +  slope + nearest road 1519 9 3057.00 1.50 0.20 0.63 elevation + tree cover + slope + nearest road 1519 10 3057.00 1.80 0.17 0.80 nearest road 1522 7 3059.00 3.60 0.07 0.87 tree cover + nearest road 1522 8 3060.00 4.80 0.04 0.91 slope + nearest road 1522 8 3060.00 5.00 0.04 0.94 elevation +  nearest road 1522 8 3060.00 5.10 0.03 0.97 elevation + tree cover + nearest road 1521 9 3061.00 5.60 0.03 1.00 1 1540 6 3091.00 36.20 0.00 1.00    75 Appendix  C. Model selection tables for observation models in chapter 2 Table C.1. Model selection for base observation model used for Cerro Venado dataset, Lambayeque, Peru. D is the density model, p0 is the model for baseline capture probability, and σ is the model for the home range shape parameter. K is the number of parameters in the model. ΔAIC is the difference in Akaike’s information criterion (AIC) between the best model and each successive model. wi is the Akaike weight, representing the probability that a model is the best in the model set. model logL K AIC ΔAIC weight D(~1) p0(~session*sex) σ(~session) 1006 8 2029 0.00 0.35 D(~1) p0(~session*sex) σ(~session + sex) 1006 9 2030 1.50 0.17 D(~session) p0(~session*sex) σ(~session) 1006 9 2031 1.90 0.13 D(~1) p0(~session*sex) σ(~session*sex) 1006 10 2032 3.20 0.07 D(~session) p0(~session*sex) σ(~session + sex) 1006 10 2032 3.40 0.07 D(~session) p0(~session*sex) σ(~session*sex) 1006 11 2034 5.10 0.03 D(~1) p0(~session) σ(~session) 1011 6 2034 5.30 0.03 D(~1) p0(~session) σ(~session*sex) 1009 8 2034 5.30 0.03 D(~1) p0(~sex) σ(~session) 1012 6 2035 6.60 0.01 D(~1) p0(~1) σ(~session) 1013 5 2035 6.60 0.01 D(~1) p0(~1) σ(~session*sex) 1011 7 2036 6.70 0.01 D(~1) p0(~session + sex) σ(~session*sex) 1009 9 2036 7.20 0.01 D(~session) p0(~session) σ(~session*sex) 1009 9 2036 7.30 0.01 D(~1) p0(~1) σ(~session + sex) 1012 6 2036 7.30 0.01 D(~session) p0(~session) σ(~session) 1011 7 2036 7.30 0.01 D(~session) p0(~session) σ(~session + sex) 1010 8 2037 7.80 0.01 D(~1) p0(~sex) σ(~session*sex) 1010 8 2037 8.10 0.01 D(~session) p0(~session + sex) σ(~session) 1010 8 2037 8.10 0.01 D(~1) p0(~sex) σ(~session + sex) 1012 7 2037 8.50 0.01 D(~session) p0(~sex) σ(~session) 1012 7 2037 8.60 0.00 D(~session) p0(~1) σ(~session*sex) 1011 8 2037 8.60 0.00 D(~session) p0(~1) σ(~session) 1013 6 2037 8.60 0.00 D(~session) p0(~sex + session) σ(~session*sex) 1009 10 2038 9.10 0.00 D(~session) p0(~1) σ(~session + sex) 1012 7 2038 9.20 0.00 D(~session) p0(~session + sex) σ(~session + sex) 1010 9 2038 9.70 0.00 D(~session) p0(~sex) σ(~session*sex) 1010 9 2039 9.90 0.00 D(~session) p0(~sex) σ(~session + sex) 1012 8 2039 10.40 0.00 D(~1) p0(~session*sex) σ(~1) 1015 7 2043 14.20 0.00 76 model logL K AIC ΔAIC weight D(~session) p0(~session*sex) σ(~1) 1015 8 2045 16.20 0.00 D(~1) p0(~session*sex) σ(~sex) 1015 8 2045 16.20 0.00 D(~session) p0(~session*sex) σ(~sex) 1015 9 2047 18.20 0.00 D(~1) p0(~1) σ(~1) 1020 4 2047 18.50 0.00 D(~1) p0(~session) σ(~1) 1019 5 2047 18.60 0.00 D(~1) p0(~session + sex) σ(~1) 1018 6 2048 19.60 0.00 D(~1) p0(~sex) σ(~1) 1020 5 2049 20.20 0.00 D(~session) p0(~1) σ(~1) 1020 5 2049 20.40 0.00 D(~session) p0(~session) σ(~1) 1019 6 2049 20.50 0.00 D(~1) p0(~session) σ(~sex) 1019 6 2049 20.50 0.00 D(~1) p0(~1) σ(~sex) 1020 5 2049 20.50 0.00 D(~1) p0(~session + sex) σ(~sex) 1018 7 2050 21.40 0.00 D(~session) p0(~session + sex) σ(~1) 1018 7 2050 21.60 0.00 D(~1) p0(~sex) σ(~sex) 1019 6 2051 21.90 0.00 D(~session) p0(~sex) σ(~1) 1019 6 2051 22.10 0.00 D(~session) p0(~1) σ(~sex) 1020 6 2051 22.40 0.00 D(~session) p0(~session) σ(~sex) 1019 7 2051 22.50 0.00 D(~session) p0(~session + sex) σ(~sex) 1018 8 2052 23.30 0.00 D(~session) p0(~sex) σ(~sex) 1019 7 2053 23.80 0.00  Table C.2. Model selection for base observation model used for Laquipampa dataset, Lambayeque, Peru. See table C.1 for definitions. model logL K AIC ΔAIC weight D(~1) p0(~1) σ(~session*sex) 505 7 1024 0.00 0.26 D(~1) p0(~session) σ(~session*sex) 505 8 1025 1.10 0.15 D(~1) p0(~session + sex) σ(~session*sex) 504 9 1025 1.50 0.12 D(~session) p0(~1) σ(~session*sex) 505 8 1026 1.90 0.10 D(~1) p0(~session*sex) σ(~session*sex) 503 10 1027 3.00 0.06 D(~session) p0(~session) σ(~session*sex) 505 9 1027 3.00 0.06 D(~session) p0(~sex) σ(~session*sex) 505 9 1027 3.40 0.05 D(~session) p0(~sex + session) σ(~session*sex) 504 10 1027 3.50 0.04 D(~session) p0(~session*sex) σ(~session*sex) 503 11 1029 5.00 0.02 D(~1) p0(~sex) σ(~1) 510 5 1029 5.10 0.02 D(~1) p0(~sex) σ(~session) 509 6 1029 5.30 0.02 D(~1) p0(~sex) σ(~session*sex) 507 8 1029 5.40 0.02 D(~1) p0(~session*sex) σ(~session) 507 8 1031 6.80 0.01 77 model logL K AIC ΔAIC weight D(~1) p0(~sex) σ(~sex) 509 6 1031 7.00 0.01 D(~1) p0(~session + sex) σ(~1) 510 6 1031 7.10 0.01 D(~session) p0(~sex) σ(~1) 510 6 1031 7.10 0.01 D(~session) p0(~sex) σ(~session) 509 7 1031 7.10 0.01 D(~1) p0(~sex) σ(~session + sex) 509 7 1031 7.20 0.01 D(~1) p0(~session*sex) σ(~1) 509 7 1031 7.50 0.01 D(~session) p0(~session + sex) σ(~session) 508 8 1032 8.50 0.00 D(~1) p0(~session*sex) σ(~session + sex) 507 9 1033 8.60 0.00 D(~session) p0(~session*sex) σ(~session) 507 9 1033 8.80 0.00 D(~1) p0(~1) σ(~1) 512 4 1033 8.80 0.00 D(~session) p0(~sex) σ(~session + sex) 508 8 1033 8.90 0.00 D(~1) p0(~session + sex) σ(~sex) 509 7 1033 9.00 0.00 D(~session) p0(~sex) σ(~sex) 509 7 1033 9.00 0.00 D(~session) p0(~session + sex) σ(~1) 510 7 1033 9.10 0.00 D(~session) p0(~session*sex) σ(~1) 509 8 1033 9.30 0.00 D(~1) p0(~session*sex) σ(~sex) 509 8 1033 9.50 0.00 D(~1) p0(~1) σ(~session) 512 5 1034 9.60 0.00 D(~1) p0(~1) σ(~sex) 512 5 1034 9.80 0.00 D(~session) p0(~session + sex) σ(~session + sex) 508 9 1034 10.10 0.00 D(~session) p0(~session*sex) σ(~session + sex) 507 10 1035 10.60 0.00 D(~1) p0(~session) σ(~1) 512 5 1035 10.80 0.00 D(~session) p0(~1) σ(~1) 512 5 1035 10.80 0.00 D(~1) p0(~1) σ(~session + sex) 511 6 1035 10.80 0.00 D(~1) p0(~session) σ(~session) 511 6 1035 10.90 0.00 D(~session) p0(~session + sex) σ(~sex) 509 8 1035 11.00 0.00 D(~session) p0(~session*sex) σ(~sex) 509 9 1035 11.30 0.00 D(~session) p0(~1) σ(~session) 512 6 1035 11.40 0.00 D(~1) p0(~session) σ(~sex) 512 6 1036 11.70 0.00 D(~session) p0(~1) σ(~sex) 512 6 1036 11.80 0.00 D(~session) p0(~1) σ(~session + sex) 511 7 1037 12.70 0.00 D(~session) p0(~session) σ(~session) 511 7 1037 12.80 0.00 D(~session) p0(~session) σ(~1) 512 6 1037 12.80 0.00 D(~session) p0(~session) σ(~sex) 512 7 1038 13.70 0.00 D(~session) p0(~session) σ(~session + sex) 511 8 1038 14.40 0.00    78 Table C.3. Model selection for base observation model used for combined Laquipampa and Cerro Venado datasets, Lambayeque, Peru. See table C.1 for definitions. model logL K AIC ΔAIC weight D(~1) p0(~sex) σ(~session + sex) 1539 7 3091 0.00 0.14 D(~1) p0(~1) σ(~session + sex) 1540 6 3091 0.13 0.14 D(~1) p0(~1) σ(~session*sex) 1539 7 3092 1.09 0.08 D(~1) p0(~sex) σ(~session*sex) 1538 8 3092 1.16 0.08 D(~session) p0(~sex) σ(~session + sex) 1539 8 3093 2.00 0.05 D(~1) p0(~session + sex) σ(~session*sex) 1538 9 3093 2.08 0.05 D(~1) p0(~session) σ(~session*sex) 1539 8 3093 2.13 0.05 D(~session) p0(~1) σ(~session + sex) 1540 7 3093 2.13 0.05 D(~1) p0(~session*sex) σ(~session + sex) 1538 9 3094 2.59 0.04 D(~session) p0(~session + sex) σ(~session + sex) 1538 9 3094 2.86 0.04 D(~session) p0(~session) σ(~session + sex) 1539 8 3094 3.01 0.03 D(~session) p0(~1) σ(~session*sex) 1539 8 3094 3.05 0.03 D(~session) p0(~sex) σ(~session*sex) 1538 9 3094 3.11 0.03 D(~1) p0(~sex) σ(~session) 1542 6 3095 3.72 0.02 D(~session) p0(~sex + session) σ(~session*sex) 1538 10 3095 4.06 0.02 D(~1) p0(~session*sex) σ(~session*sex) 1538 10 3095 4.08 0.02 D(~session) p0(~session) σ(~session*sex) 1539 9 3095 4.11 0.02 D(~session) p0(~session*sex) σ(~session + sex) 1538 10 3096 4.58 0.02 D(~1) p0(~sex) σ(~sex) 1542 6 3096 4.90 0.01 D(~1) p0(~1) σ(~sex) 1543 5 3097 5.27 0.01 D(~session) p0(~sex) σ(~session) 1542 7 3097 5.72 0.01 D(~1) p0(~session + sex) σ(~sex) 1542 7 3097 5.98 0.01 D(~session) p0(~session*sex) σ(~session*sex) 1538 11 3097 6.05 0.01 D(~1) p0(~session) σ(~sex) 1543 6 3098 6.32 0.01 D(~session) p0(~sex) σ(~sex) 1542 7 3098 6.72 0.01 D(~1) p0(~session*sex) σ(~session) 1541 8 3098 6.78 0.00 D(~session) p0(~session + sex) σ(~session) 1541 8 3098 6.96 0.00 D(~session) p0(~1) σ(~sex) 1543 6 3098 7.09 0.00 D(~1) p0(~session*sex) σ(~sex) 1542 8 3099 7.74 0.00 D(~session) p0(~session + sex) σ(~sex) 1542 8 3099 7.90 0.00 D(~1) p0(~sex) σ(~1) 1545 5 3099 8.00 0.00 D(~session) p0(~session) σ(~sex) 1543 7 3100 8.23 0.00 D(~session) p0(~session*sex) σ(~session) 1541 9 3100 8.78 0.00 D(~1) p0(~session + sex) σ(~1) 1544 6 3100 8.91 0.00 D(~session) p0(~session*sex) σ(~sex) 1542 9 3101 9.70 0.00 79 model logL K AIC ΔAIC weight D(~session) p0(~sex) σ(~1) 1545 6 3101 9.83 0.00 D(~1) p0(~session*sex) σ(~1) 1544 7 3102 10.73 0.00 D(~session) p0(~session + sex) σ(~1) 1544 7 3102 10.85 0.00 D(~session) p0(~session*sex) σ(~1) 1544 8 3104 12.70 0.00 D(~1) p0(~1) σ(~session) 1547 5 3105 13.64 0.00 D(~1) p0(~session) σ(~session) 1547 6 3106 14.74 0.00 D(~session) p0(~1) σ(~session) 1547 6 3107 15.63 0.00 D(~1) p0(~1) σ(~1) 1550 4 3107 16.17 0.00 D(~session) p0(~session) σ(~session) 1547 7 3108 16.72 0.00 D(~1) p0(~session) σ(~1) 1549 5 3109 17.59 0.00 D(~session) p0(~1) σ(~1) 1550 5 3109 17.98 0.00 D(~session) p0(~session) σ(~1) 1549 6 3111 19.47 0.00   80 Appendix  D. Distribution of Andean bears predicted for equatorial dry forest  Figure D.1. Predicted distribution of Andean bears (Tremarctos ornatus; ≥ 4/100 km2) in equatorial dry forest of northwestern Peru. Values for each 1km pixel are predicted using model-averaged effect sizes from both Cerro Venado and Laquipampa Wildlife Refuge surveys and remotely-sensed habitat data available for the region using the equation {log(β0 + β1(nearest road) + β2(slope) + β3(forest cover) + β4(elevation)} (Manly et al., 2002). 81 Appendix  E. Feature representation tables for chapter 3 Table E.1. Feature representation table for vertebrate species, water risk, and ecoregions in Scenario 1, where targets were set for all features. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. Scenario 1: All features Feature Category 17% 30% 50% All amphibian species vertebrate species ranges 20.60 34.81 53.80 Threatened amphibian species vertebrate species ranges 17.00 30.00 50.00 Andean bear distribution vertebrate species ranges 25.64 42.47 62.99 All bird species vertebrate species ranges 18.94 32.90 52.67 Small-ranged bird species vertebrate species ranges 17.00 30.00 50.00 Threatened bird species vertebrate species ranges 19.25 32.97 52.81 All mammal species vertebrate species ranges 17.82 31.38 51.24 Threatened mammal species vertebrate species ranges 17.84 31.37 51.22 All reptile species vertebrate species ranges 18.44 32.06 51.58 Water risk water risk 17.00 30.00 50.00 Guajira-Barranquilla xeric scrub ecoregions 17.08 30.04 50.04 Cordillera La Costa montane forests ecoregions 17.06 30.08 50.06 Catatumbo moist forests ecoregions 17.01 30.01 50.00 Venezuelan Andes montane forests ecoregions 17.00 30.00 50.00 Southern Andean Yungas ecoregions 17.00 30.00 50.00 Bolivian montane dry forests ecoregions 17.00 30.00 50.00 Cordillera Central paramo ecoregions 17.00 30.00 50.00 Iquitos varzea ecoregions 17.13 30.06 50.00 Magdalena Valley montane forests ecoregions 17.00 30.00 50.00 Eastern Cordillera real montane forests ecoregions 17.00 30.00 50.00 Southwest Amazon moist forests ecoregions 17.00 30.00 50.00 Sinu Valley dry forests ecoregions 17.01 30.01 50.00 Central Andean wet puna ecoregions 17.00 30.00 50.00 Llanos ecoregions 17.12 30.09 50.09 Paraguana xeric scrub ecoregions 17.02 30.02 50.02 Lara-Falcon dry forests ecoregions 17.17 30.05 50.00 Cordillera Oriental montane forests ecoregions 17.00 30.00 50.00 Maracaibo dry forests ecoregions 17.01 30.04 50.00 Santa Marta paramo ecoregions 17.07 30.05 50.00 La Costa xeric shrublands ecoregions 17.17 30.09 50.09 Magdalena-Uraba moist forests ecoregions 17.01 30.01 50.01 82 Feature Category 17% 30% 50% Choco-Darien moist forests ecoregions 17.01 30.01 50.01 Santa Marta montane forests ecoregions 17.01 30.00 50.00 Northwestern Andean montane forests ecoregions 17.00 30.00 50.00 Cauca Valley dry forests ecoregions 17.01 30.01 50.01 Magdalena Valley dry forests ecoregions 17.01 30.00 50.00 Napo moist forests ecoregions 17.00 30.00 50.00 Western Ecuador moist forests ecoregions 17.39 30.43 50.00 Guayaquil flooded grasslands ecoregions 22.22 33.33 55.56 Tumbes-Piura dry forests ecoregions 17.00 30.00 50.00 Dry Chaco ecoregions 17.00 30.00 50.00 Peruvian Yungas ecoregions 17.00 30.00 50.00 Sechura desert ecoregions 17.00 30.00 50.00 Northern Andean paramo ecoregions 17.00 30.00 50.00 Maranon dry forests ecoregions 17.01 30.01 50.00 Central Andean puna ecoregions 17.00 30.00 50.00 Bolivian Yungas ecoregions 17.00 30.00 50.00 Central Andean dry puna ecoregions 17.00 30.00 50.00 Patia Valley dry forests ecoregions 17.01 30.04 50.02 Apure-Villavicencio dry forests ecoregions 17.01 30.00 50.01 Cordillera de Merida paramo ecoregions 17.01 30.00 50.00 Ucayali moist forests ecoregions 17.00 30.00 50.00 Cauca Valley montane forests ecoregions 17.00 30.00 50.00 Caqueta moist forests ecoregions 17.02 30.02 50.00  Table E.2. Feature representation table for vertebrate species, water risk, and ecoregions in Scenario 2, where targets were set for all features except water risk. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. Scenario 2: All except water risk Feature Category 17% 30% 50% All amphibian species vertebrate species ranges 21.97 36.03 55.58 Threatened amphibian species vertebrate species ranges 17.00 30.00 50.00 Andean bear distribution vertebrate species ranges 26.89 43.77 64.34 All bird species vertebrate species ranges 19.51 33.52 53.46 Small-ranged bird species vertebrate species ranges 17.00 30.00 50.17 Threatened bird species vertebrate species ranges 19.76 33.69 53.64 All mammal species vertebrate species ranges 18.33 32.02 51.92 83 Feature Category 17% 30% 50% Threatened mammal species vertebrate species ranges 18.35 32.01 51.89 All reptile species vertebrate species ranges 19.24 32.82 52.46 Water risk water risk 15.63 28.18 48.22 Guajira-Barranquilla xeric scrub ecoregions 17.08 30.04 50.04 Cordillera La Costa montane forests ecoregions 17.06 30.08 50.06 Catatumbo moist forests ecoregions 17.01 30.01 50.00 Venezuelan Andes montane forests ecoregions 17.00 30.00 50.00 Southern Andean Yungas ecoregions 17.00 30.00 50.00 Bolivian montane dry forests ecoregions 17.00 30.00 50.00 Cordillera Central paramo ecoregions 17.00 30.00 50.00 Iquitos varzea ecoregions 17.13 30.06 50.00 Magdalena Valley montane forests ecoregions 17.00 30.00 50.00 Eastern Cordillera real montane forests ecoregions 17.00 30.00 50.00 Southwest Amazon moist forests ecoregions 17.00 30.00 50.00 Sinu Valley dry forests ecoregions 17.01 30.01 50.00 Central Andean wet puna ecoregions 17.00 30.00 50.00 Llanos ecoregions 17.12 30.09 50.09 Paraguana xeric scrub ecoregions 17.02 30.02 50.02 Lara-Falcon dry forests ecoregions 17.17 30.05 50.00 Cordillera Oriental montane forests ecoregions 17.00 30.00 50.00 Maracaibo dry forests ecoregions 17.01 30.04 50.00 Santa Marta paramo ecoregions 17.07 30.05 50.00 La Costa xeric shrublands ecoregions 17.17 30.09 50.09 Magdalena-Uraba moist forests ecoregions 17.01 30.01 50.01 Choco-Darien moist forests ecoregions 17.01 30.01 50.01 Santa Marta montane forests ecoregions 17.01 30.00 50.00 Northwestern Andean montane forests ecoregions 17.00 30.00 50.00 Cauca Valley dry forests ecoregions 17.01 30.01 50.01 Magdalena Valley dry forests ecoregions 17.01 30.00 50.00 Napo moist forests ecoregions 17.00 30.00 50.00 Western Ecuador moist forests ecoregions 17.39 30.43 50.00 Guayaquil flooded grasslands ecoregions 22.22 33.33 55.56 Tumbes-Piura dry forests ecoregions 17.00 30.00 50.00 Dry Chaco ecoregions 17.00 30.00 50.00 Peruvian Yungas ecoregions 17.00 30.00 50.00 Sechura desert ecoregions 17.00 30.00 50.00 Northern Andean paramo ecoregions 17.00 30.00 50.00 Maranon dry forests ecoregions 17.01 30.01 50.00 84 Feature Category 17% 30% 50% Central Andean puna ecoregions 17.00 30.00 50.00 Bolivian Yungas ecoregions 17.00 30.00 50.00 Central Andean dry puna ecoregions 17.00 30.00 50.00 Patia Valley dry forests ecoregions 17.01 30.04 50.02 Apure-Villavicencio dry forests ecoregions 17.01 30.00 50.01 Cordillera de Merida paramo ecoregions 17.01 30.00 50.00 Ucayali moist forests ecoregions 17.00 30.00 50.00 Cauca Valley montane forests ecoregions 17.00 30.00 50.00 Caqueta moist forests ecoregions 17.02 30.02 50.00  Table E.3. Feature representation table for vertebrate species, water risk, and ecoregions in Scenario 3, where targets were only set for threatened species. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. Scenario 3: Threatened species only Feature Category 17% 30% 50% All amphibian species vertebrate species ranges 31.85 45.35 63.95 Threatened amphibian species vertebrate species ranges 17.00 30.00 50.00 Andean bear distribution vertebrate species ranges 21.02 35.86 55.95 All bird species vertebrate species ranges 18.31 31.51 51.24 Small-ranged bird species vertebrate species ranges 10.37 21.67 41.95 Threatened bird species vertebrate species ranges 17.41 30.98 50.95 All mammal species vertebrate species ranges 17.06 30.14 50.32 Threatened mammal species vertebrate species ranges 17.00 30.00 50.00 All reptile species vertebrate species ranges 17.53 30.00 49.69 Water risk water risk 8.43 15.63 28.50 Guajira-Barranquilla xeric scrub ecoregions 0.00 1.58 50.83 Cordillera La Costa montane forests ecoregions 2.24 73.51 93.04 Catatumbo moist forests ecoregions 3.80 33.17 80.40 Venezuelan Andes montane forests ecoregions 35.83 67.78 89.08 Southern Andean Yungas ecoregions 0.00 0.00 0.00 Bolivian montane dry forests ecoregions 0.11 0.56 0.93 Cordillera Central paramo ecoregions 0.00 0.00 1.57 Iquitos varzea ecoregions 98.46 99.86 100.00 Magdalena Valley montane forests ecoregions 1.18 7.14 55.34 Eastern Cordillera real montane forests ecoregions 15.37 30.65 40.65 Southwest Amazon moist forests ecoregions 29.97 49.24 67.17 85 Feature Category 17% 30% 50% Sinu Valley dry forests ecoregions 0.00 7.59 82.64 Central Andean wet puna ecoregions 0.00 0.09 0.94 Llanos ecoregions 3.60 53.51 88.29 Paraguana xeric scrub ecoregions 0.00 2.47 60.84 Lara-Falcon dry forests ecoregions 0.00 8.08 25.76 Cordillera Oriental montane forests ecoregions 29.92 61.82 90.75 Maracaibo dry forests ecoregions 1.57 11.65 42.05 Santa Marta paramo ecoregions 0.00 0.24 93.91 La Costa xeric shrublands ecoregions 0.00 2.48 37.52 Magdalena-Uraba moist forests ecoregions 0.00 9.18 73.29 Choco-Darien moist forests ecoregions 11.80 44.61 87.63 Santa Marta montane forests ecoregions 0.00 11.16 89.79 Northwestern Andean montane forests ecoregions 11.82 24.93 50.48 Cauca Valley dry forests ecoregions 0.00 0.03 7.41 Magdalena Valley dry forests ecoregions 0.00 0.00 16.61 Napo moist forests ecoregions 62.38 85.57 95.81 Western Ecuador moist forests ecoregions 0.00 0.00 14.49 Guayaquil flooded grasslands ecoregions 0.00 0.00 0.00 Tumbes-Piura dry forests ecoregions 0.00 0.00 0.00 Dry Chaco ecoregions 0.00 0.00 0.00 Peruvian Yungas ecoregions 12.77 21.16 31.51 Sechura desert ecoregions 0.00 0.00 0.00 Northern Andean paramo ecoregions 3.21 14.49 44.59 Maranon dry forests ecoregions 0.00 0.34 5.42 Central Andean puna ecoregions 0.00 0.00 0.00 Bolivian Yungas ecoregions 2.91 10.14 25.70 Central Andean dry puna ecoregions 0.00 0.00 0.00 Patia Valley dry forests ecoregions 0.00 0.00 10.14 Apure-Villavicencio dry forests ecoregions 15.04 58.21 86.81 Cordillera de Merida paramo ecoregions 72.95 93.58 98.99 Ucayali moist forests ecoregions 56.18 76.27 93.57 Cauca Valley montane forests ecoregions 0.23 3.74 32.13 Caqueta moist forests ecoregions 80.54 92.82 99.62   86 Table E.4. Feature representation table for vertebrate species, water risk, and ecoregions in Scenario 4, where targets were only set for the Andean bear range. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. Scenario 4: Single-species focus - Andean bear  Feature Category 17% 30% 50% 100% All amphibian species vertebrate species ranges 0.06 0.10 0.16 0.28 Threatened amphibian species vertebrate species ranges 0.04 0.07 0.15 0.36 Andean bear distribution vertebrate species ranges 0.17 0.30 0.50 1.00 All bird species vertebrate species ranges 0.06 0.09 0.15 0.28 Small-ranged bird species vertebrate species ranges 0.05 0.08 0.15 0.32 Threatened bird species vertebrate species ranges 0.06 0.10 0.16 0.30 All mammal species vertebrate species ranges 0.05 0.09 0.14 0.28 Threatened mammal species vertebrate species ranges 0.05 0.09 0.14 0.28 All reptile species vertebrate species ranges 0.05 0.08 0.14 0.26 Water risk water risk 0.03 0.05 0.09 0.19 Guajira-Barranquilla xeric scrub ecoregions 0.00 0.00 0.00 0.00 Cordillera La Costa montane forests ecoregions 0.00 0.00 0.00 0.00 Catatumbo moist forests ecoregions 0.00 0.00 0.00 0.01 Venezuelan Andes montane forests ecoregions 0.00 0.00 0.02 0.33 Southern Andean Yungas ecoregions 0.05 0.18 0.29 0.44 Bolivian montane dry forests ecoregions 0.01 0.04 0.07 0.17 Cordillera Central paramo ecoregions 0.00 0.02 0.04 0.09 Iquitos varzea ecoregions 0.00 0.00 0.05 0.11 Magdalena Valley montane forests ecoregions 0.00 0.01 0.04 0.18 Eastern Cordillera real montane forests ecoregions 0.07 0.12 0.18 0.31 Southwest Amazon moist forests ecoregions 0.05 0.06 0.07 0.08 Sinu Valley dry forests ecoregions 0.00 0.00 0.00 0.00 Central Andean wet puna ecoregions 0.00 0.01 0.02 0.07 Llanos ecoregions 0.00 0.00 0.00 0.00 Paraguana xeric scrub ecoregions 0.00 0.00 0.00 0.00 Lara-Falcon dry forests ecoregions 0.00 0.00 0.00 0.00 Cordillera Oriental montane forests ecoregions 0.01 0.09 0.19 0.44 Maracaibo dry forests ecoregions 0.00 0.00 0.00 0.00 Santa Marta paramo ecoregions 0.00 0.00 0.00 0.00 La Costa xeric shrublands ecoregions 0.00 0.00 0.00 0.00 Magdalena-Uraba moist forests ecoregions 0.00 0.00 0.01 0.01 Choco-Darien moist forests ecoregions 0.00 0.03 0.15 0.40 Santa Marta montane forests ecoregions 0.00 0.00 0.00 0.00 87 Feature Category 17% 30% 50% 100% Northwestern Andean montane forests ecoregions 0.01 0.04 0.14 0.34 Cauca Valley dry forests ecoregions 0.00 0.00 0.00 0.00 Magdalena Valley dry forests ecoregions 0.00 0.00 0.00 0.00 Napo moist forests ecoregions 0.00 0.00 0.00 0.00 Western Ecuador moist forests ecoregions 0.00 0.00 0.00 0.01 Guayaquil flooded grasslands ecoregions 0.00 0.00 0.00 0.00 Tumbes-Piura dry forests ecoregions 0.00 0.00 0.01 0.11 Dry Chaco ecoregions 0.08 0.14 0.15 0.22 Peruvian Yungas ecoregions 0.12 0.15 0.19 0.26 Sechura desert ecoregions 0.00 0.00 0.00 0.00 Northern Andean paramo ecoregions 0.01 0.04 0.12 0.57 Maranon dry forests ecoregions 0.00 0.00 0.00 0.01 Central Andean puna ecoregions 0.00 0.00 0.01 0.03 Bolivian Yungas ecoregions 0.03 0.13 0.31 0.61 Central Andean dry puna ecoregions 0.00 0.00 0.00 0.01 Patia Valley dry forests ecoregions 0.00 0.00 0.02 0.04 Apure-Villavicencio dry forests ecoregions 0.00 0.00 0.00 0.00 Cordillera de Merida paramo ecoregions 0.00 0.00 0.00 0.78 Ucayali moist forests ecoregions 0.19 0.25 0.32 0.39 Cauca Valley montane forests ecoregions 0.00 0.00 0.00 0.10 Caqueta moist forests ecoregions 0.00 0.00 0.00 0.00  Table E.5. Feature representation table for vertebrate species, water risk, and ecoregions in Scenario 5, where targets were only set for a random species range. Values are the percentage of each feature captured in the solution for each run, based on the total coverage of that feature. Scenario 5: Single-species focus - Random feature  Feature Category 17% 30% 50% 100% All amphibian species vertebrate species ranges 0.07 0.10 0.14 0.23 Threatened amphibian species vertebrate species ranges 0.03 0.06 0.10 0.23 Andean bear distribution vertebrate species ranges 0.06 0.11 0.16 0.23 All bird species vertebrate species ranges 0.05 0.08 0.13 0.23 Small-ranged bird species vertebrate species ranges 0.04 0.07 0.11 0.23 Threatened bird species vertebrate species ranges 0.05 0.08 0.13 0.23 All mammal species vertebrate species ranges 0.05 0.07 0.12 0.23 Threatened mammal species vertebrate species ranges 0.05 0.07 0.12 0.23 All reptile species vertebrate species ranges 0.05 0.08 0.12 0.23 88 Feature Category 17% 30% 50% 100% Water risk water risk 0.03 0.06 0.11 0.23 Guajira-Barranquilla xeric scrub ecoregions 0.00 0.00 0.00 0.22 Cordillera La Costa montane forests ecoregions 0.00 0.00 0.02 0.25 Catatumbo moist forests ecoregions 0.00 0.01 0.04 0.22 Venezuelan Andes montane forests ecoregions 0.00 0.00 0.03 0.23 Southern Andean Yungas ecoregions 0.05 0.12 0.16 0.24 Bolivian montane dry forests ecoregions 0.02 0.06 0.13 0.23 Cordillera Central paramo ecoregions 0.05 0.08 0.13 0.23 Iquitos varzea ecoregions 0.15 0.18 0.21 0.22 Magdalena Valley montane forests ecoregions 0.00 0.01 0.04 0.23 Eastern Cordillera real montane forests ecoregions 0.06 0.09 0.14 0.23 Southwest Amazon moist forests ecoregions 0.14 0.17 0.19 0.23 Sinu Valley dry forests ecoregions 0.00 0.00 0.01 0.25 Central Andean wet puna ecoregions 0.01 0.03 0.09 0.23 Llanos ecoregions 0.00 0.01 0.05 0.21 Paraguana xeric scrub ecoregions 0.00 0.00 0.03 0.20 Lara-Falcon dry forests ecoregions 0.00 0.00 0.00 0.19 Cordillera Oriental montane forests ecoregions 0.03 0.05 0.11 0.23 Maracaibo dry forests ecoregions 0.00 0.00 0.00 0.20 Santa Marta paramo ecoregions 0.00 0.00 0.03 0.24 La Costa xeric shrublands ecoregions 0.00 0.00 0.00 0.18 Magdalena-Uraba moist forests ecoregions 0.00 0.03 0.12 0.24 Choco-Darien moist forests ecoregions 0.03 0.08 0.18 0.23 Santa Marta montane forests ecoregions 0.00 0.00 0.03 0.23 Northwestern Andean montane forests ecoregions 0.01 0.04 0.10 0.23 Cauca Valley dry forests ecoregions 0.00 0.00 0.00 0.20 Magdalena Valley dry forests ecoregions 0.00 0.00 0.00 0.22 Napo moist forests ecoregions 0.11 0.13 0.16 0.24 Western Ecuador moist forests ecoregions 0.03 0.07 0.07 0.22 Guayaquil flooded grasslands ecoregions 0.00 0.00 0.00 0.33 Tumbes-Piura dry forests ecoregions 0.00 0.04 0.13 0.24 Dry Chaco ecoregions 0.12 0.14 0.16 0.23 Peruvian Yungas ecoregions 0.06 0.09 0.13 0.23 Sechura desert ecoregions 0.02 0.06 0.13 0.23 Northern Andean paramo ecoregions 0.01 0.03 0.08 0.23 Maranon dry forests ecoregions 0.01 0.04 0.12 0.23 Central Andean puna ecoregions 0.03 0.07 0.13 0.23 Bolivian Yungas ecoregions 0.05 0.12 0.17 0.23 89 Feature Category 17% 30% 50% 100% Central Andean dry puna ecoregions 0.05 0.09 0.14 0.23 Patia Valley dry forests ecoregions 0.00 0.00 0.01 0.24 Apure-Villavicencio dry forests ecoregions 0.02 0.03 0.08 0.23 Cordillera de Merida paramo ecoregions 0.00 0.00 0.02 0.23 Ucayali moist forests ecoregions 0.14 0.18 0.20 0.23 Cauca Valley montane forests ecoregions 0.00 0.00 0.01 0.22 Caqueta moist forests ecoregions 0.14 0.18 0.19 0.23    90 Appendix  F. Bears will be bears  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0373204/manifest

Comment

Related Items