University of Florida Thesis Or Dissertation

AVIAN BEHAVIORAL AND COMMUNITY-LEVEL RESPONSES TO FOREST FRAGMENTATION IN THE WESTERN ANDES OF COLOMBIA

HARRISON HENRY JONES

A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2020

To the many people who have given so much to help me get here

ACKNOWLEDGMENTS

I would like to acknowledge the tremendous help and support of so many to make this document a reality. My parents have been a well of positivity, support, and understanding, and have consistently pushed me to pursue my passions and develop my intellectual curiosity. I am indebted to them for taking the time to introduce me to the natural world and the wonders it contains. I had the priviledge of being part of a vibrant intellectual community at the University of Florida that helped to shape me as a scientist and a conservationist. Dr. Scott Robinson, in addition to being the finest birder I have ever met, has been a consistently supportive and helpful presence. I am truly lucky to have such a great scientist, and good person, for a co-adviser. I’d also like to acknowledge the other members of my committee for their input and advice. Dr. Rob

Fletcher has been a fantastic resource on statistical analyses and landscape ecology in general.

Dr. Ben Baiser has been helpful in framing analyses of multivariate statistics, while Dr. Collette

St. Mary provided input on behavioral mechanisms. Dr. Gabriel Colorado was invaluable in discussing Andean birds and mixed species flocks and providing another perspective on my study system. Finally, Dr. Todd Palmer, while often on another continent, has been a font of academic wisdom about publishing, jobs, and life in general. While not (technically) on my committee, many of the statistical analyses in this dissertation would not have been possible without the mathematical genius that is Dr. Zach Siders, who has patiently explained to me how multi-species occupancy models work at least a dosen times…

I am also indebeted to the graduate student community and University of Florida and my rockstar lab mates who push me to be a better ornithologist. I would not be where I am today without the support and academic guidance of Drs. Kristen Malone and Larissa Boesing, who took time out of their busy schedules to take a befuddled master’s student under their wing (pun very much intended). I received incredible near-daily feedback and help from Ian Ausprey,

Felicity Newell, Stephanie Wheeler, Diego Garcia, Mitch Walters, Daniel Montalvo, Liz

Hurtado and Wenyi Zhou. It is hard to imagine a finer collection of intelligent and dedicated field ornithologists. My field work in Colombia would not have become a reality were it not for the tremendous help of (soon-to-be Dr.) Juliana Bedoya and the entire staff of the Serraniagua organization, particularly Cristhian Cardona. Dr. Oscar Murillo also helped with getting research permits in Colombia. I had the true priviledge of working with an amazing group of field assistants, who uncomplainingly walked up steep mountains before dawn and otherwise endured the worst of what the Andes could throw at them. I owe a great debt to Julio Bermúdez, Edwin

Muñera, Duvan García, Felipe Castro, María Fernanda Restrepo, and Carolina Revelo.

TABLE OF CONTENTS

page

ACKNOWLEDGMENTS ...... 4

LIST OF TABLES ...... 9

LIST OF FIGURES ...... 11

ABSTRACT ...... 12

CHAPTER

1 INTRODUCTION ...... 14

Conserving Fragmented Andean Landscapes ...... 14 Mixed-Species Flocks: Importance and Response to Fragmentation ...... 15 Changes to Species Roles and Species Association Patterns with Fragmentation ...... 16 Community-level Changes: Turnover or Nestedness? ...... 18

2 PATCH SIZE AND VEGETATION STRUCTURE DRIVE CHANGES TO MIXED- SPECIES FLOCK DIVERSITY AND COMPOSITION ACROSS A GRADIENT OF FRAGMENT SIZES IN THE WESTERN ANDES OF COLOMBIA ...... 20

Methods ...... 23 Study System and Sites ...... 23 Transect Surveys for Mixed-Species Flocks ...... 25 Calculation of Lanscape-level Variables ...... 25 Vegetation Measurements and Principal Components Analysis ...... 27 Calculation and Standardization of Functional and Phylogenetic Diversity Metrics ...... 28 Generalized Linear Mixed Model Analyses of Flock Size and Diversity ...... 30 Canonical Correspondance Analysis ...... 32 Results...... 33 Transect Surveys of Flocks ...... 33 Changes to Flock Diversity and Encounter Rate ...... 33 Changes to Flock Species Richness of Taxonomic/Functional Groups ...... 35 Canonical Correspondance Analysis ...... 35 Discussion ...... 36 Persistance of Flocking Behavior: Open Membership and Nuclear Species Redundancy...... 37 Importance of Patch Size and Vegetation Structure ...... 39 Changes to Flock Composition: Loss of Forest Interior Suboscines ...... 40 Other Factors: Seasonal Decreases in Flock Diversity ...... 42 Management Implications: Maintaining Old-growth, Primary Forests ...... 42

3 VEGETATION STRUCTURE DRIVES MIXED-SPECIES FLOCK INTERACTION STRENGTH AND NUCLEAR SPECIES ROLES ...... 51

Methods ...... 54 Study System and Sites ...... 54 Transect Surveys for Mixed-species Flocks ...... 56 Calculation of Landscape Level Variables ...... 56 Vegetation Measurements and Principal Component Analysis ...... 57 Construction, Measurement, and Analysis of Social Networks ...... 59 Calculation and Analysis of Network Dissimilarity ...... 62 Results...... 63 Flock Observations and Network Metrics ...... 63 LMM Analysis of Network Metrics and Species Nuclearity...... 63 Partitioning Network Dissimilarity Using dbRDA ...... 65 Discussion ...... 66 Maintenance of Flock Cohesion Amidst Species Turnover: Open-membership Flocks? ...... 66 Vegetation Structure Affects Strength of Flocking Interactions ...... 68 Nuclear Species Redundancy and Turnover ...... 70 Seasonal Changes to Interaction Strength and Nuclear Species Importance ...... 71 Conservation Implications ...... 72

4 SIMULTANEOUS LOSS OF FOREST DEPENDENT SPECIES AND SPATIAL TURNOVER DRIVE CHANGES TO AVIAN RICHNESS AND TURNOVER IN ANDEAN FOREST FRAGMENTS ...... 81

Methods ...... 83 Study Area and Sites ...... 83 Andean Bird Surveys ...... 84 Quantifying Local Vegetation ...... 86 Quantifying Landscape Composition and Configuration ...... 87 Calculating Alpha Functional Diversity Metrics ...... 88 Multi-Species, Muti-survey-type Occupancy Model ...... 89 Analysis of Alpha Diversity ...... 90 Beta Diversity Partitioning of Taxonomic and Functional Diversity ...... 91 Results...... 92 Avifaunal Surveys ...... 92 MS-MSOM Results ...... 93 Alpha Taxonomic and Functional Diversity ...... 93 Beta Taxonomic and Functional Diversity ...... 94 Discussion ...... 95 Negative Patch, Edge, and Logging Effects on Forest Specialist Species ...... 96 Functional and Taxonomic Turnover Across a Fragment Size Gradient ...... 97 Conservation Implications: Large Reserves Matter ...... 98

5 CONCLUSIONS AND CONSERVATION IMPLICATIONS ...... 108

APPENDIX

A ADDITIONAL RESULTS, TABLES, AND FIGURES FOR CHAPTER 2 ...... 111

B ADDITIONAL METHODS, TABLES, AND FIGURES FOR CHAPTER 4 ...... 122

LIST OF REFERENCES ...... 131

BIOGRAPHICAL SKETCH ...... 149

LIST OF TABLES

Table page

2-1 Field sites in El Cairo municipality, Valle del Cauca, Colombia...... 44

2-2 Model-averaged estimates of predictor variables on flock species richness, size, and diversity...... 45

2-3 Model-averaged estimates of predictor variables on flock species richness (FSR) of functional and taxonomic groups...... 46

3-1 Field sites in El Cairo municipality, Valle del Cauca, Colombia...... 74

3-2 Linear mixed model estimates of fragmentation predictor variables on network metrics...... 75

3-3 Linear mixed model estimates of the effect of fragmentation on the centrality of nuclear species in mixed-species flock social networks...... 76

3-4 Marginal significance of dbRDA predictor variables on partitioned network dissimilarity...... 77

4-1 Study sites and survey results in from forest fragments in El Cairo municipality...... 100

4-2 Mean beta estimates of fragmentation covariates from the MG-MSOM...... 101

4-3 Linear mixed model estimates of fragmentation effects on alpha functional diversity. ..102

4-4 Beta diversity partitioning of taxonomic and functional diversity...... 103

A-1 Eigenvalues and proportion of variance explained of the principal component analysis of understory vegetation density...... 112

A-2 Loadings of the understory vegetation principal component axes...... 112

A-3 Eigenvalues and proportion of variance explained of the principal component analysis of tree dbh category density...... 112

A-4 Loadings of the tree dbh class principal component axes...... 113

A-5 Eigenvalues and proportions of variance explained for the six CCA constrained axes ..113

A-6 Bi-plot scores for constraining variables used in the CCA...... 113

A-7 Species scores for the first four CCA constrained axes...... 114

A-8 Predictor variables used for generalized linear mixed model analysis of flock size, species richness, and diversity measures...... 116

A-9 Model selection table for eight predictor variables on flock encounter rate...... 117

A-10 Bird species observed in mixed-species flocks during boreal summer (June-August) and boreal winter (January-March) sampling...... 118

B-1 Prior distributions for all parameters used in the MSOM...... 127

B-2 Eigenvalues and proportion of variance explained in beta estimate PCA...... 127

B-3 Predictor variable loadings on the PCA axes...... 127

LIST OF FIGURES

Figure page

2-1 Map of study fragments in the Western Andes of Colombia...... 47

2-2 Landscape effects on mixed-species flock diversity...... 48

2-3 Importance of vegetation variables in driving mixed-species flock diversity and encounter rate...... 49

2-4 Biplot of the first two CCA axes...... 50

3-1 Social networks of mixed-species flocking interactions across a gradient of canopy height...... 78

3-2 Landscape and local vegetation effects on nuclear species network centrality...... 79

3-3 Ternary plot of pairwise network dissimilarity...... 80

4-1 Phylogeny of Andean bird species included in the MG-MSOM...... 1044

4-2 Effect of forest fragmentation on taxonomic and functional diversity of Andean birds...... 105

4-3 Changes to Andean bird community composition in response to forest fragmentation. .106

A-1 Biplot of first two principal component axes of the ordinations of the understory vegetation densities and the tree diameters...... 121

B-1 Phylogeny of Andean bird species included in the MS-MSOM...... 1288

B-2 Phylogeny of Andean bird species with median covariate betas...... 129

B-3 Correlation among occupancy and detection covariance effect parameters {푢, 푣푀푁푆, 푣푀푁푊, 푣푂푆, 푣푇}...... 130

Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy

AVIAN BEHAVIORAL AND COMMUNITY-LEVEL RESPONSES TO FOREST FRAGMENTATION IN THE WESTERN ANDES OF COLOMBIA

Harrison Henry Jones

May 2020

Chair: Todd Palmer Cochair: Scott Robinson Major: Zoology

The montane forests of the tropical Andes, a global hotspot of biodiversity, have undergone extensive deforestation and fragmentation, yet little is known about how Andean avifauna respond to this disturbance. We looked at responses of a facilitative behavior, mixed species flocking, and the full bird community along a gradient of forest fragment sizes (10-170 ha), as well as a continuous forest reference site, in the same landscape of the Colombian

Western Andes. Response variables were measured along 500-meter transects; we used buffer analysis to quantify local landscape composition within 1 km and multivariate measures of the vegetation structure to measure disturbance effects. Species richness of mixed species flocks were negatively affected by declining patch size and vegetation structure, with extensive turnover in the flocking bird community. We found that a portion of the functional diversity in flocks was restricted to primary, unlogged forest. When using network analysis to look at the strength of interactions in Andean flocks, we found no effect of patch size or edge effects; flocks across the gradient were ‘open membership’ and characterized by numerous weak interactions and a lack of flock sub-groupings like those in lowland systems. Multiple leader (‘nuclear’) species existed which shifted in importance along the gradient. Overall community composition

underwent important changes, with strong negative area and edge effects exerted on forest dependent species, which were replaced in smaller fragments with disturbance-adapted and non- forest species. The variance in functional traits in the community was negatively impacted by edge effects, possibly influencing ecosystem functioning.

CHAPTER 1 INTRODUCTION

Conserving Fragmented Andean Landscapes

The cutting of tropical forests, which produce patchy, fragmented landscapes, is a leading driver of biodiversity loss worldwide (Fahrig 2003, Geist and Lambin 2002) and a driver of extinction risk in vertebrate species (Tracewski et al. 2016). Though over 7 million hectares of tropical forest are cleared per year (Achard et al. 2014), one ecosystem particularly at-risk to ongoing forest fragmentation is the montane cloud forest of the northern Andes (Aldrich et al.

1997, Tejedor Garavito et al. 2012, Hermes et al. 2018). These forests represent a global hotspot of endemic, restricted-range, and at-risk bird species richness (Orme et al. 2005, Kier et al. 2009) and are undergoing high levels of deforestation (Tejedor Garavito et al. 2012, Tracewski et al.

2016). In Colombia for example, only ~30% of originally forested lands above 1,500 meters remain (Etter et al. 2006), with conversion to agricultural lands and urban expansion driving deforestation (Armenteras et al. 2007, Armenteras et al. 2011). Andean birds are vulnerable to habitat loss because they occur in narrow altitudinal bands (Graves 1988), leading to small range sizes and restricting them to specific climatic conditions. Andean birds therefore respond negatively to fragmentation, with a loss of species occurring in cloud forest fragments (Kattan et al. 1994, Renjifo 1999, Aubad et al. 2010). Other studies have detected negative effects of patch isolation (Aubad et al. 2010) and edge effects (Restrepo and Gomez 1998), suggesting a sensitivity to landscape variables.

Although fragmentation can be treated as a single concept, it affects ecosystems and species interactions in multiple ways which reflect different scales (Fahrig 2017). On the one hand, communities and behaviors are affected indirectly by effects steming from habitat loss at the landscape scale and associated loss of species (Bagchi et al. 2018, Zou et al. 2018).

Biodiversity loss is associated with reduction in patch size, changes to landscape configuration and composition, and increased patch isolation (Wilson et al. 2016), which can lead to a loss of species in mutualistic interaction networks (Cramer et al. 2007, Cordeiro et al. 2015, Emer et al.

2018), and fewer interactions even where species persist (Mokross et al. 2014). However, species interactions can also suffer direct effects at a local scale (Bagchi et al. 2018, Zou et al., 2018).

Novel species interactions and climate conditions at the edge of habitat patches (Fagan et al.

1999) can lead to ‘edge effects’ on positive interactions (Chen et al. 2017, Galetti et al. 2003).

Fragmentation of forests can also lead to within-patch disturbance (Magrach et al. 2012), which has destabilizing effects on positive species interactions (Knowlton and Graham 2011). Because of these multiple spatial scales, it is important to understand the relative impact of landscape and local factors in driving breakdowns of species interactions.

Mixed-Species Flocks: Importance and Response to Fragmentation

Positive species interactions are particularly susceptible to fragmentation (Magrach et al.

2014) and are therefore increasingly of interest as targets of conservation (Tylianakis et al.

2010). Loss of positive species interactions may occur prior to, and independently of, loss of species (Valiente-Banuet et al. 2015), such that species become ‘functionally extinct’ even while persisting within a community (e.g. Klein et al. 2003, McConkey and O'Farrill 2016). One important positive species interaction for forest birds is mixed-species flocking behavior, defined as associations of two or more species moving and foraging together (Morse 1970). Flocks consist of one or more leader (‘nuclear’) species and many associate or follower species

(Moynihan 1962), and flocking leads to greater foraging efficiency and antipredator vigilance

(Sridhar et al. 2009). Furthermore, flocks allow for species to expand their foraging niches to more exposed microhabitats (Darrah and Smith 2013, Martínez et al. 2018), higher strata in the canopy (Dinesen 1995, Zou et al. 2011, Farine and Milburn 2013), and adjacent matrix habitat

(Dolby and Grubb 2000, Tubelis et al. 2006). Flocking positively affects the survival and condition of flock-following species (Dolby and Grubb 1998) and allows birds to persist in disturbed habitat (Mammides et al. 2015). Because 20% of all bird species, and 50 to 60 percent of tropical bird communities, participate in flocks (Zou et al. 2018), they can be considered a form of large-scale facilitation. Finally, flocking species themselves are conservation targets, as flocking propensity is inversely correlated with persistence in human-modified landscapes (Lees and Peres 2008, Mammides et al., 2015).

Flocking, however, is negatively affected by fragmentation, with smaller, less-speciose flocks in fragments (Tellería et al. 2001, Maldonado-Coelho and Marini 2004, Mokross et al.

2014, Cordeiro et al. 2015). Even when they persist in fragments, birds have a lower propensity to join flocks (Mokross et al. 2014). Landscape-level factors partially determine flock size and composition, with fragment size (Tellería et al. 2001, Fernandez-Juricic 2002, Cordeiro et al.

2015) and percentage forest cover (Brandt et al. 2009, Colorado Zuluaga and Rodewald 2015) exerting significant effects. Within-patch foraging microhabitat also plays a role, with more complex vegetation structure associated with larger and more diverse flocks (McDermott and

Rodewald 2014, Mokross et al. 2014, Colorado Zuluaga and Rodewald 2015). Flocks forage preferentially in microhabitats with greater canopy density and vertical vegetation structure

(Potts et al. 2014, McDermott et al. 2015, Mokross et al. 2018), but fragmentation changes vegetation structure and available foraging microhabitats (Stratford and Stouffer 2015). Finally, flocking associations are affected by the loss of nuclear species, which can lead to ‘non-trophic cascades’ of secondary extinctions (Mammides et al. 2015). There is therefore an urgent need to evaluate the relative contributions of these diverse factors to changes in flock behavior.

Changes to Species Roles and Species Association Patterns with Fragmentation

While the loss of species in disturbed landscapes can lead to fewer species interactions, interaction networks can also ‘rewire’ in response to disturbance, creating novel species interactions and/or changing the ecological context of existing interactions (Valiente-Banuet et al. 2015). Pairwise species co-occurrences within a flock, are not random- species with more similar body sizes (Bell 1983, King and Rappole 2008, Mammides et al. 2018), height strata within the forest (Munn 1985, Zou et al. 2011, Srinivasan et al. 2012), and foraging ecology

(Sridhar et al. 2012) are all more likely to co-occur, and positive species associations characterize undisturbed flocking systems worldwide (Sridhar et al. 2012). Mutualistic species interactions are context-specific (Miguel et al. 2018), however, and species co-occurrences across disturbance gradients show little correlation (Mammides et al. 2018). Changes to resource distribution or flock composition in fragments might increase the costs associated with joining flocks such as interspecific competition (Graves and Gotelli 1993), or non-optimal foraging and movement speeds (Hutto 1988), causing flocking species to abandon flocking behavior.

Alternatively, a ‘re-wiring’ of flocking interactions in the presence of disturbance could instead increase the number of positive species associations (e.g. Borah et al. 2018).

Some species also play leadership roles within flocks (nuclear species, see above). While such species have traditionally been classified categorically, a quantitative measure of

‘nuclearity’ is likely more appropriate (Srinivasan et al. 2010), especially because species roles can shift across habitats (Gram 1998, Morse 1970). To our knowledge, the ecological drivers of changes to nuclearity have never been investigated, yet the fact that species roles can shift across nearby sites and altitudinal gradients suggests that ecological factors may play a role (Marín-

Gomez and Arbeláez-Cortés 2015, Arbeláez-Cortés and Marín-Gomez 2012). Different nuclear

species show different susceptibilities to anthropogenic change, so ‘nuclear species resilience’ in the face of disturbance can help buffer the loss of flocking (Mammides et al. 2015). However,

‘nuclear species redundancy’, the ability of secondary ‘redundant’ nuclear species to assume leadership roles where ‘primary’ nuclear species have been extirpated, could also contribute to flock persistence. The fact that multiple species of tanagers can play nuclear roles in Andean flocks suggests this may be an important mechanism (Bohórquez 2003, Guevara et al. 2011); changes to nuclearity of different leader species along a fragmentation gradient therefore merit additional study.

Community-level Changes: Turnover or Nestedness?

To acquire a more wholistic view of the effects of fragmentation on Andean bird communities, it is important to quantify community-level changes. While both forest fragmentation and habitat disturbance have negative effects on avian biodiversity (e.g. Carrara et al. 2015, Alroy 2017), it is important to understand how changes to biodiversity are occurring across a fragment size gradient. On the one hand, biological communities often suffer non- random species loss with declining patch size, a pattern known as nested species loss (Patterson

1987). Nested species loss is commonly observed among avifaunas in fragmented landscapes

(Wethered and Lawes 2005, Hill et al. 2011, Smith et al. 2018), and, where this occurs, large and medium size fragments will be able to support a subset of forest dependent species diversity.

However, changes to community composition with forest fragmentation can also be caused by species turnover (Baselga 2010), where forest dependent species are replaced with disturbance- adapted or non-forest species. Where extensive species turnover occurs, fragments may be supporting an entirely different community from continuous forest (Carrara et al. 2015, Cordeiro et al. 2015, Keinath et al. 2017). Andean avifaunas are known to lose species in forest fragments

(Kattan et al. 1994, Aubad et al. 2010), yet extensive species turnover is reported among Andean

avifauna in isolated fragments and secondary forest (Renjifo 1999, O'Dea and Whittaker 2007).

There is therefore a pressing need to identify mechanisms of change to community composition; this requires studies of gradients of fragmentation rather than isolated case studies of specific fragments (e.g. Kattan et al. 1994, Renjifo 1999).

CHAPTER 2 PATCH SIZE AND VEGETATION STRUCTURE DRIVE CHANGES TO MIXED-SPECIES FLOCK DIVERSITY AND COMPOSITION ACROSS A GRADIENT OF FRAGMENT SIZES IN THE WESTERN ANDES OF COLOMBIA

The cutting of tropical forests, which produces patchy, fragmented landscapes, is a leading driver of biodiversity loss worldwide (Geist and Lambin 2002, Fahrig 2003). Indeed, over seven million hectares of tropical forest are cleared per year (Achard et al. 2014). Smaller, more isolated fragments of habitat have lower biodiversity generally (Wilson et al. 2016), and tropical forest communities in particular are negatively affected by this fragmentation (Bregman et al. 2014). Positive species interactions such as pollination and seed dispersal are particularly susceptible to forest fragmentation (Magrach et al. 2014) and therefore of interest as targets of conservation (Tylianakis et al. 2010). Loss of positive species interactions may occur prior to, and independently of, loss of species (Valiente-Banuet et al. 2015), such that species may lose their functional role even while persisting within a community (e.g. McConkey and O'Farrill

2016). Such functional extinctions have important carry-over effects on ecosystem functioning

(Anderson et al. 2011), and may lead to secondary species extinctions (Säterberg et al. 2013).

Conserving ecological communities therefore depends not only on which species are present, but also on the structure and consequences of their interactions (Tylianakis et al. 2010, Jordano

2016), and how such interactions respond to habitat loss and fragmentation.

One important positive species interaction for forest birds is mixed species flocking

(hereafter flocking) behavior, defined as associations of two or more species moving and foraging together (Morse 1970). Flocks consist of one or more leader (‘nuclear’) species and many associate or follower species (Moynihan 1962), and flocking leads to greater foraging efficiency and antipredator vigilance (Sridhar et al. 2009). Furthermore, flocks allow species to expand their foraging niches to more exposed microhabitats (Martínez et al. 2018), higher strata

in the canopy (Zou et al. 2011, Farine and Milburn 2013), and adjacent matrix habitat (Tubelis et al. 2006). Flocking positively affects the survival and condition of flock-following species

(Dolby and Grubb 1998, Srinivasan 2019) and allows birds to persist in disturbed habitat

(Mammides et al. 2015). Because 20% of all bird species, and 50 to 60% of bird communities, participate in flocks (Zou et al. 2018), they can be considered a form of large-scale interspecific facilitation.

Flocking is negatively affected by fragmentation, with smaller, less-speciose flocks reported in lowland fragments compared to continuous forest (Tellería et al. 2001, Maldonado-

Coelho and Marini 2004, Mokross et al. 2014, Cordeiro et al. 2015), although the extent to which these trends hold in highland systems is unknown. Landscape-level factors partially determine flock size and composition, with patch size (Tellería et al. 2001, Cordeiro et al. 2015) and percentage forest cover (Brandt et al. 2009, Colorado Zuluaga and Rodewald 2015) exerting significant effects. Within-patch vegetation structure also plays a role, with more complex structure associated with larger and more diverse flocks (McDermott and Rodewald 2014,

Mokross et al. 2014, Colorado Zuluaga and Rodewald 2015). Flocks forage non-randomly in microhabitats with greater canopy density and vertical vegetation structure (Potts et al. 2014,

McDermott et al. 2015, Mokross et al. 2018), both of which change with fragmentation

(Stratford and Stouffer 2015). Finally, flocking associations change with the loss of nuclear species, which can lead to ‘non-trophic cascades’ of secondary extinctions (Mammides et al.

2015). Edge effects, which have been extensively explored in the fragmentation literature

(Pfeifer et al. 2017), have, to our knowledge, not yet been studied in flocks. The effects of these diverse factors on flock diversity metrics have never been simultaneously assessed, however, which limits our ability to identify the relative importance of each process on flocking behavior.

The effects of anthropogenic disturbance on flocking behavior have traditionally been measured by looking at the species richness and size of the flock (Zou et al. 2018). Biodiversity, however, can be measured in additional dimensions not captured by these traditional metrics.

Functional diversity, for example, describes the diversity of functional roles, or, put another way, of ‘what organisms do’ (Petchey and Gaston 2006) within an assemblage. In flocks, this includes how and where a species forages and is particularly important because it drives ecosystem functioning (Laureto et al. 2015). Similarly, phylogenetic diversity is an important measure of how the evolutionary history of an assemblage, or the length of branches of the phylogenetic tree

(Tucker et al. 2017), is affected by disturbance. Lastly, compositional changes to mixed-species flocks are of great interest but have been little studied. Such changes can result not only from species loss, but also from species turnover, producing novel compositions without corresponding changes to species richness or other metrics of biodiversity (Baselga 2010). In many flocking systems, forest-interior species are replaced by generalist or edge-associated species in flocks subject to disturbance (Sidhu et al. 2010, Goodale et al. 2014, Cordeiro et al.

2015). It is therefore important to look at what drives higher species richness of both forest- interior specialist bird families (Furnariidae, Tyrannidae; Powell et al. 2015a) and edge-tolerant or edge-associated groups (Thraupidae, Boreal migrant species) in flocks.

In this study, we investigated changes to mixed-species flocks across a patch size gradient in the Western Andes of Colombia. We used transects to sample flocks in fragments of primary cloud forest to ask the following questions: (1) Is there a loss of highland flocking behavior in small fragments as documented in lowland flocks? (2) What is the relative importance of patch size, vegetation structure, and edge effects in driving changes to flock diversity? (3) Do the different aspects of flock diversity (species richness, phylogenetic, and

functional) respond differently to fragmentation? And (4) how does the composition of mixed- species flocks change with fragmentation metrics? If only patch size is affecting flocks, then we would expect smaller, less diverse flocks in smaller fragments, regardless of other factors.

Alternatively, if distance to edge or vegetation structure are also important, then we would expect that the distance of the flock to the forest edge and the structural complexity of the local vegetation will be positively correlated with flock diversity regardless of patch size.

Methods

Study System and Sites

We conducted all fieldwork in subtropical humid forests located within the municipality of El Cairo, Valle del Cauca department in Colombia. The study region is part of the Western

Andes mountain range, a center of avian threatened species diversity and endemism within

Colombia (Ocampo-Peñuela and Pimm 2014). Andean forests in Colombia are highly fragmented, with only 30% of original forest cover remaining (Etter et al. 2006), and the landscape in this municipality consists of the typical patchwork of forest fragments embedded in a matrix of cattle pasture, regenerating scrub, and coffee farms (Armenteras et al. 2011). Within this landscape, we selected eight fragments representing a gradient in patch sizes (range 10 to

170 ha; Table 2-1, Figure 2-1). Sites are in the same altitudinal belt (1900-2200 m.a.s.l.) and matrix type (cattle pasture) to control for effects of altitude (Marín-Gomez and Arbeláez-Cortés

2015) and matrix type (Renjifo 2001) on flock size and composition. Within-patch disturbance is common in fragmented Andean forests in Colombia, particularly illegal selective logging, which can reduce tree diversity and shift vegetation structure of logged patches towards early successional forests (Aubad et al. 2008). In our landscape this typically occurred as removal of select old-growth trees for lumber by landowners; logging histories varied considerably from historical to ongoing, and extensive to limited, both within and between patches.

We established 500-meter transects through forest interior (n = 14 total transects; Table

2-1) as our sample unit. Transects were opportunistically placed on existing trails, which were at variable distances from the edge of the fragments. Where possible, we placed one transect in a more disturbed (logged) section of the forest and a second in a relatively undisturbed site within each fragment. We further divided each transect into 100-meter segments to account for heterogeneity in vegetation structure within transects. We accounted for edge effects by measuring the distance to forest edge of each transect segment.

We stratified forest fragments into large (≥ 100 ha), medium (~30-50 ha), and small (≤ 20 ha) size categories and selected a minimum of two replicates of each (Table 2-1); these represent the full range of fragment sizes available in our study landscape (Figure 2-1). We also included a non-fragmented reference site (Reserva Natural Comunitária Cerro El Inglés, ~750 ha) connected to over 10,000 ha of continuous forest to the north and west along the spine of the

Serranía de los Paraguas. Fragments were privately owned, and we collaborated with a local

NGO (Serraniagua; http://www.serraniagua.org) to ensure access. We only selected fragments with primary or late-successional secondary forest; vegetation structure and canopy height varied substantially between patches based on intensities of selective logging and land-use histories (see above). Fragments were all separated by ≥ 100 meters to minimize among-patch movement of birds, and all transects in different fragments were at least 250 meters apart. Mixed-species flocks in Colombian sub-montane forests contain both understory and canopy species, with ~100 participating species and up to 50 individuals per flock (Arbeláez-Cortés et al. 2011, Colorado

Zuluaga and Rodewald 2015). These flocks are frequently joined by migratory bird species during the boreal winter, which are among the most common flocking species during this period

(McDermott and Rodewald 2014, Marín-Gomez and Arbeláez-Cortés 2015).

Transect Surveys for Mixed-Species Flocks

We performed transect surveys for mixed-species flocks, adapted from Goodale et al.

(2014), within forest fragments from June-August 2017 (boreal migrants absent) and January-

March 2018 (boreal migrants present). Both sampling periods corresponded to a dry season in the Western Andes, which has a bimodal two-dry, two-wet seasonality pattern. For each transect, we spent two and a half sequential field days performing continuous transect surveys; we conducted surveys in small fragments, large fragments, and continuous forest sites in random order to avoid a temporal bias in sampling. Surveys were distributed across the morning (7:30-

11:30) and evening (15:00-17:30) hours when flocking behavior is most common in Andean forests (Arbeláez-Cortés et al. 2011). Transects were walked slowly and continuously by 2-3 observers, including local birdwatchers familiar with all species (HHJ present for all surveys); flocking birds were identified by both sight and sound. When a flock was encountered, we noted the time of day and transect segment in which it was observed and spent up to a maximum of 45 minutes characterizing it with 10x binoculars. At least 5 minutes were spent with each flock, following it if possible. Because detection of species in flocks was imperfect, we only included a flock observation in the analysis if we felt that at least 80% of the individuals were observed

(e.g. after spending several minutes of continuous observation at the end of the survey period without observing a new species or individual); incomplete flock observations were not included in analyses. We are confident that our survey methodology accurately described flock composition because birds moved and called frequently in flocks, leading to high detectability.

We noted the start and end time of each survey, and the presence of incomplete flocks to calculate flock encounter rate. We also supplemented the transect surveys with data from flocks

opportunistically observed on a transect while performing other fieldwork. Avian taxonomy of flocking species follows Quiñones (2018).

Calculation of Lanscape-level Variables

We obtained landscape-level variables for analyses using geographic information software (GIS) analysis in ArcGIS (ArcMap 10.3.1, Esri; Redlands, CA). To quantify landscape composition and configuration, we buffered each transect (n = 14) by 1 km; buffers extended from the entire length of the transect. We selected the 1-km scale because tropical forest birds have been shown to respond up to this scale to landscape variables (Cerezo et al. 2010, Carrara et al. 2015). We then calculated measures of landscape composition and configuration using a recent land-cover/use categorization made by the Corporación Autónoma Regional del Valle del

Cauca, converted to a 25-m cell-size raster. To quantify landscape composition, we calculated percentages of the forest-cover type within each buffer using the ‘isectpolyrst’ tool in Geospatial

Modelling Environment (version 0.7.4.0; Beyer 2015). Following Carrara et al. (2015), we selected percentage of forest cover as a proxy for patch size because some of our transects were located in continuous forest with no patch size measurement. Because the matrix in our landscape consisted of unforested cattle pasture, we feel that this is a good measure of patch area, since there was no other forested habitat in the buffer areas. We measured landscape configuration for each transect within the buffers using edge density, or length of all forest edges

(in meters) divided by total buffer area (in hectares), as described by Carrara et al. (2015). We did not analyze metrics from buffers of different sizes (e.g. 500-m) because composition and configuration metrics correlated heavily with the 1-km scale. The distance to edge was calculated in meters for each 100-meter transect segment (n = 70) as the shortest straight-line distance between the center point of the segment and the nearest edge of the fragment. For the flock

encounter rate analyses, we calculated a single value for each transect by averaging the five distances-to-edge associated with each segment.

Vegetation Measurements and Principal Components Analysis

We measured vegetation structure in each 100-m transect segment used for flock sampling. Vegetation measurements were made from June-August 2017; based on our observations of vegetation, we assumed variation between the two sampling periods was minimal. We used the sampling methodology of James and Shugart (1970), following the modifications made by Stratford and Stouffer (2013), and further modified to be used with belt transects. Broadly, the methodology comprises two components for every 100-meter transect segment: (1) the quantification of canopy cover, ground cover, canopy height, and foliage height diversity of vegetation using point sampling every 10 meters and (2) the quantification of shrub, vine, fern, palm, and tree fern and tree density using 3-meter-wide belt sampling.

For the point sampling, we measured eight variables at ten-meter intervals, for 10 points per 100-meter segment. As a measure of foliage height diversity along the transect, we noted the presence or absence of live vegetation at five heights: <0.5 m, >0.5–3 m, >3–10 m, >10–20 m, and >20 m. Above 3 meters, we used a rangefinder to determine heights while sighting through a tube with crosshairs. Canopy and ground cover were calculated to the nearest 1/8th of the field of view by sighting through a vertical canopy densiometer (GRS Densiometer, Geographic

Resource Solutions, Arcata, CA). For each segment, we averaged values for canopy cover, and ground cover, and calculated the proportion of points at which vegetation was present for each height category. For the belt transect sampling, we surveyed vegetation along the same transects and calculated densities for each 100-m transect interval. We counted all shrubs, vines, ferns, tree ferns, and palms encountered on 1.5 meters to either side. Secondly, we counted all trees

(woody vegetation > 2 m in height) within 1.5 meters of the transect and measured their diameter

at breast height (DBH). Trees were later categorized into six DBH size classes for analysis: 1-7 cm, 8-15 cm, 16-23 cm, 24-30 cm, 31-50 cm, and > 50 cm. We additionally recorded the largest tree’s DBH.

To quantify foliage height diversity, we calculated the Shannon Diversity Index of the proportion of points with vegetation present in each of the five height bands for each segment (n

= 70 segments). To reduce redundancy and minimize correlation between variables, we

(separately) ordinated our tree DBH and understory plant density data using principal component analysis (PCA: McGarigal et al. 2000) for each 100-meter transect segment. We column (Z score) standardized data prior to ordination to account for differences in the units of measurement and used the covariance matrix to run the PCA. The principal components were interpreted using the significance of the principal component loadings. All analyses were performed in R (version 3.5.1; R Core Team 2020). The PCA was run using the princomp function in the stats package. The Shannon Index was calculated using the diversity function of the vegan package (Oksanen et al. 2019). The results of the vegetation data ordination are reported in the Appendix A.

Calculation and Standardization of Functional and Phylogenetic Diversity Metrics

We calculated metrics of functional and phylogenetic diversity of birds from flocks with complete (80+ %) composition data. We define functional diversity as the diversity of foraging niches and behaviors present within a given flock. We calculated two measures of functional diversity: functional richness (Villéger et al. 2008) and functional dispersion (Laliberté and

Legendre 2010). These are multivariate measures calculated using a distance framework from a matrix of quantitative and categorical traits of all species observed (Laliberté and Legendre

2010). Briefly, functional richness represents the volume of available functional trait space occupied by a given flock (Mason et al. 2005), while functional dispersion represents the mean

distance in multidimensional trait space of all species in a flock to the centroid for all flocking species and is equivalent to the functional diversity variance in a flock (Laliberté and Legendre

2010).

We built a trait matrix for all species observed in flocks based on information from the

Handbook of the Birds of the World Alive website (del Hoyo et al. 2020), supplemented with family-specific references where appropriate. We included body mass, diet (degree of insectivory, frugivory, nectivory, and granivory), foraging maneuvers, foraging substrates, foraging strata (ground, understory, midstory, sub-canopy, and canopy), and habitat preferences

(forest interior, forest edge, secondary forest, open habitat) as functional traits. Categories were not mutually exclusive; species’ diet preferences were classified on a 0 to 3 scale based on reported frequency of consumption, while use of foraging maneuvers, substrates, strata and habitat types were keyed as present (1) or absent (0). Foraging maneuvers and substrates were classified according to the Remsen and Robinson (1990) typology. Functional diversity metrics were calculated from a matrix of species-by-flock abundances using the dbFD function of the

FD package (Laliberté et al. 2014). Functional dispersion was weighted by the abundance of each species in each flock. We standardized functional richness by the ‘global’ functional richness to constrain it between 0 and 1.

We also calculated two phylogeny-based measures selected for their widespread use and ability to capture the richness and divergence dimensions of phylogenetic diversity (Tucker et al.

2017). We selected phylogenetic diversity as a measure of the quantity of phylogenetic differences present in each flock. This is measured as the summed branch lengths of all species present on a phylogeny of all species observed (Faith 1992). The divergence of phylogenetic lineages present in each flock was measured using the mean pairwise distance measure, which

calculates the mean branch lengths between all species pairs in a given flock (Webb et al. 2002).

We created a phylogeny of all species observed in flocks by sub-setting the global Jetz (2012) phylogeny. We downloaded 1000 trees from the Bird Tree website (www.birdtree.org) using the

Hackett (2008) backbone phylogeny. We calculated a 50% majority-rule consensus tree using mean edge lengths with the consensus.edge function of the phytools package (Revell 2019); the observed metrics were then calculated using the pd and mpd functions of the picante package

(Kembel et al. 2019), using the same species-by-flock abundances matrix as for the functional diversity measures.

Because measures of functional and phylogenetic diversity are often correlated with species richness (Vellend et al. 2014, Weiher 2014), we used standardized effect sizes (SES) to control for species richness when calculating these metrics. To create a null model of functional and phylogenetic diversity we used 999 iterations of the tip swapping method (Webb et al. 2008) to randomly shuffle taxa labels across the tips of the functional trait matrix and flocking species phylogeny, respectively. We then calculated the relevant diversity metrics on each of the null communities and compared it to the observed value. We standardized SES values for each metric by subtracting the mean of the null values from the observed value and then dividing this by the standard deviation of the null values. We used the ses.pd and ses.mpd functions of the picante package to calculate phylogenetic diversity SES values; functional diversity values were calculated based on novel code developed by Dr. Ben Baiser.

Generalized Linear Mixed Model Analyses of Flock Size and Diversity

We ran generalized linear mixed models (lmer function, lme4 package; Bates et al. 2015) for 11 flock response variables to determine the relative importance of different fragmentation- related variables on flocks. Our replicate was the individual flock, and we limited our sample size to flocks observed on a transect (n = 433 flock compositions). We used an identity link

function to model total observed individuals in the flock (flock size), observed flock species richness, functional richness, functional dispersion, phylogenetic diversity, and mean pairwise distance. To understand differences in flock composition, we also modeled species richness of boreal migrants, ovenbirds (Furnariidae), tyrant flycatchers (Tyrannidae), and tanagers

(Thraupidae) using the glmer function with a log link function. Models of migrant richness used only the subset of flock data collected during the January-March sampling period. We log- transformed the flock richness, functional richness, and flock size variables for normalcy.

Models included 9 fixed-effect predictor variables (Table A-8), encompassing patch size, edge effects, and vegetation characteristics as well as nuclear species presence and season. We also included transect ID (n = 14) and site (n = 9; Table 2-1) as two random effects to account for the non-independence of flocks observed on the same transect and transects surveyed within the same forest fragment. We checked for multicollinearity of the predictor variables using variance inflation factors; no predictor variable had a VIF of ≥ 3.

We used an information theoretic framework to determine best models explaining flock diversity measures, ranking and comparing models using AIC values adjusted for small sample size (AICc: Burnham and Anderson 2002). We ranked all subsets of the global model (9 fixed effects, 2 random effects; n = 512 models), and considered models equivalent if their ΔAICc value was two or less. We performed full model averaging for all response variables in the candidate model set (all models within 2 ΔAICc of the best model) because a best model did not emerge from the AICc ranking of the full model set (Richards et al. 2011). We did not model average the 95% confidence model sets because they contained thousands of models. We first calculated AICc and Akaike weights for all models using the dredge function of the MuMIn package (Bartoń 2019). We then performed full model averaging using the model.avg function,

as this method is recommended when the relative importance of different predictors is of interest

(Nakagawa and Freckleton 2011). Goodness-of-fit was evaluated using the marginal and conditional r2 values (Nakagawa and Schielzeth 2013) implemented using the r.squaredGLMM function of the MuMIn package.

Lastly, we ran a separate linear mixed model analysis for flock encounter rate, calculated for each transect during both survey periods (n = 28 transect by season combinations). Because the predictor variables for this analysis showed evidence of multicollinearity (variance inflation factors > 3), we compared eight single predictor models (the same predictors as for the above analyses except for presence of C. canigularis) using AICc with random effects of transect and site.

Canonical Correspondance Analysis

We evaluated the effects of fragmentation on flock composition using canonical correspondence analysis (CCA: Ter Braak 1986). Rare species can bias ordination, so we removed 47 species occurring in fewer than 3% of flocks (pooling both sampling periods) for 54 species included in the final analysis (listed in Table A-5); no species occurred in more than 95% of flocks. We took the log10(x+1) of the flock composition data because these data were zero- inflated and therefore right-skewed. We selected the same a priori predictor variables used in other analyses based on the fragment size, edge effects, and vegetation characteristics. These included six environmental variables (seasonal effects, distance to edge, foliage height diversity, understory vegetation density, tree DBH, and proportion of forest in a 1-km buffer, see Table A-

8), which were column standardized prior to the analysis. We used a detrended correspondence analysis (decorana function, vegan package; Oksanen et al. 2019) to determine that a unimodal model better fit the data. We performed the CCA using the cca function and evaluated

significance and goodness-of-fit using the anova.cca and goodness.cca functions (vegan package).

Results

Transect Surveys of Flocks

Over the two three-month sampling periods we observed 502 complete foraging flock compositions of which 436 were observed on a transect. Flocks contained two mammal and 98 bird species, including 11 boreal migrants and 87 resident species (see Table A-10 for full species list). We observed 232 (46%) complete flocks between June and August (boreal summer), of which 188 were observed on a transect. Between January and March (boreal winter), we observed 270 complete flocks (54%) of which 248 were recorded on a transect

(Table 2-1). During the boreal summer, we walked 177 transects (range = 3-19 walks per site) for a total of 116 hours of observation across all sites, while during the boreal winter we walked

315 transects (range = 9-30 walks per site) for over 178 hours of surveying. On transects, a total of 83 (resident) species were observed in flocks between June and August, whereas 79 species

(68 residents, 11 migrants) were observed in flocks between January and March. We also observed squirrels (Sciuridae; 2 species) in flocks on 16 occasions; in each case only a single individual was observed. In this system the Ashy-throated Chlorospingus, Chlorospingus canigularis, which frequently occurred in conspicuous and vocal family groups, appeared to play a nuclear role in the flocks.

Changes to Flock Diversity and Encounter Rate

Our linear mixed modeling suggested that flock richness and flock size responded in similar ways to our predictor variables; these variables were significantly affected by patch size and vegetation structure as well as season and the presence of the nuclear species (Table 2-2).

Both richness (beta = 0.086, P = 0.01) and size (beta = 0.120, P = < 0.001) increased with

percentage of forest in a 1-kilometer radius of the transect, suggesting a patch size effect (Figure

2-2a). We found contrasting effects of vegetation structure, with significant increases in the richness and size of flocks in forest with greater foliage height diversity (Figure 2-3a), but a decrease in richness and size associated with an increase in the density of large-diameter, old growth trees. We found no support for edge effects on richness and size, from either distance to edge or landscape edge density. In addition to the fragmentation variables, flocks were larger and more speciose when C. canigularis was present, and we detected a small but significant decrease in both richness (beta = -0.103, P = 0.03) and size (beta = -0.107, P = 0.03) during the boreal winter compared to the boreal summer sampling period.

Indices of functional and phylogenetic diversity responded differently to fragmentation variables than flock richness and size (Table 2-2). Functional diversity in flocks was affected by density of large trees, presence of the nuclear species, and season, but the sign of the effect was often reversed compared to flock richness and size. In contrast to flock richness, functional richness increased with greater densities of large-diameter trees (beta = 0.093, P = < 0.001;

Figure 2-3c) and was not significantly correlated with patch size. Neither foliage height diversity nor distance to edge significantly affected our two measures of functional diversity in flocks.

However, both functional richness and dispersion were significantly smaller when the nuclear C. canigularis was present in the flock, and during the boreal winter (Table 2-2). Phylogenetic diversity of flocks was not affected by patch size, edge effects, or vegetation structure. However, like functional diversity metrics, phylogenetic diversity in flocks was significantly lower when

C. canigularis was present in the flock, and both phylogenetic diversity and mean pairwise distance were also significantly smaller during the boreal winter than the summer (Table 2-2).

We found a significant positive correlation between the encounter rate of flocks and the foliage height diversity, with no support for either an edge effect or a patch size effect. The best

nd 2 2 model (ΔAICc over 2 model = 4.20, Akaike weight = 0.76, conditional r = 0.30, marginal r =

0.30; Table A-9) contained vertical vegetation complexity, which was positively correlated with flock encounter rate (beta = 0.363, P = 0.006; Figure 2-3b).

Changes to Flock Species Richness of Taxonomic/Functional Groups

We found significant differences in the species richness of taxonomic and functional groups within flocks across the patch size gradient, suggesting important spatial turnover in flock composition (Table 2-3). All four taxonomic and functional groups modeled showed a significant positive correlation between their richness and flock size. Three of the four groups also showed significant responses to fragmentation variables. The flock richness of furnariids showed negative responses to anthropogenic disturbance, as richness was positively affected by increasing distance to edge (beta = 0.174, P = 0.008) and increasing vertical vegetation structure

(beta = 0.166, P = 0.010). Many small flock-following tyrannids were also only detected in flocks in large patches of forest (Table 7, Appendix A). The richness of Thraupidae showed an opposite trend in response to disturbance, increasing with decreasing density of large-diameter trees (beta = -0.055, P = 0.04). The number of thraupid tanagers in flocks also showed significant seasonal effects, with reduced numbers during the boreal winter. Lastly, the species richness of boreal migrant species was negatively correlated with distance to forest edge (beta = -

0.168, P = 0.055).

Canonical Correspondance Analysis

Our results suggested a significant effect of both forest fragmentation and season on the composition of mixed-species flocks (Figure 2-4). The CCA species scores mirror the results of our analyses of taxonomic group richness within flocks, with forest-interior Furnariids and

Tyranniids being replaced by generalist Thraupids and migratory species with increasing fragmentation (Table 7, Appendix A). The first two CCA axes represent 7% and 2% of the total variance and 58% and 18% of the constrained variance, respectively (Table 5, Appendix A). The first CCA axis showed high biplot scores for large-diameter tree density, distance from edge, and proportion of forest cover within 1 km of the transect. More positive scores were associated with greater tree densities, percentages of forest cover, and distances from edge. We interpreted this axis as a measure of degree of fragmentation and selective logging on the composition of flocks

(Table 6, Appendix A). The second CCA axis showed a high biplot score for season and we interpreted this axis as a measure of seasonal differences in flock composition. Neither of the other vegetation variables showed high biplot scores in association with either axis. In spite of within-site variability in flock composition, an ANOVA test showed that both the

‘fragmentation’ (CCA 1; df = 1, F = 35.45, P = < 0.001) and ‘season’ (CCA 2; df = 1, F = 10.70,

P = < 0.001) axes significantly affected flock composition. The goodness-of-fit, averaged across all 54 species for each axis, was 0.06 for CCA 1 and 0.09 for CCA 2.

Discussion

We found that the diversity of Andean mixed-species flocks was negatively affected by declining patch size and changes to vegetation structure. Flocking behavior did persist in small

10-ha fragments, although declining patch size and vegetation structure negatively affected several aspects of the size, richness, and composition of flocks. Flock richness and size responded differently to fragmentation effects compared to functional and phylogenetic diversity metrics. Both richness and size increased with the area of the forest patch (Figure 2-2a), but this variable had little effect on functional and phylogenetic diversity. Phylogenetic diversity was relatively unaffected by fragmentation variables, and flock species richness and size may therefore be poor proxies for other dimensions of diversity in flocks. Vegetation disturbance

effects were significantly correlated with both richness and size as well as measures of functional diversity. Two distinct mechanisms appear to be at work: increasing foliage height diversity

(presence of vegetation in all five height bands) was positively correlated with flock richness, flock size, and flock encounter rate (Figure 2-3a,b), while greater densities of large-diameter, old growth trees were correlated with smaller, less speciose flocks, but greater functional richness

(Figure 2-3c). We additionally found species turnover in flock composition in response to fragmentation and seasonal effects (Figure 2-4). The flock richness of furnariids increased in response to increasing distance to edge and vertical vegetation complexity, whereas that of thraupids and boreal migrants increased in early successional and forest edge flocks respectively.

Persistance of Flocking Behavior: Open Membership and Nuclear Species Redundancy

Mixed-species flocking behavior persisted even in 10-ha fragments, although this behavior can disappear in smaller fragments (HHJ, unpublished data). While the limited geographic scope of our study limits our ability to generalize to other Andean flocking systems, this stands in contrast to the collapse of lowland Amazonian flocking systems in 10-ha fragments

(Ferraz et al. 2007). The persistence of Andean flocks may result from their open membership, as reflected in the high degree of species turnover across the patch size gradient. Amazonian understory flocks are highly coevolved networks with stable species composition and joint territorial defense (Munn and Terborgh 1979; Jullien and Thiollay 1998). Andean flocks, by contrast, have a variable composition, as species join and leave the flock as it moves through their territory (Poulsen 1996; HHJ pers. obs.). This lack of compositional stability may allow

Andean flocking networks to ‘rewire’ as forest-interior species are replaced by edge species, increasing resilience (Valiente-Banuet et al. 2015). Andean flocks also do not show the clear differentiation of understory and canopy flocks (Poulsen 1996; Bohorquez 2003; Guevara et al.

2011; Colorado Zuluaga and Rodewald 2015) reported from Amazonia (Munn 1985).

Understory flocks appear to be more vulnerable to fragmentation (e.g. Maldonado-Coelho and

Marini, 2004); the few flocks composed solely of understory species were observed in continuous forest (HHJ, pers. obs.). We caution, however, that persistence of flocking behavior does not mean persistence of forest-interior flock-following species (e.g. Colorado Zuluaga and

Rodewald 2015, see below).

Andean montane flocks are often led by Chlorospingus spp. (Bohorquez 2003; Arbeláez-

Cortés et al. 2011; Colorado Zuluaga and Rodewald 2015; Marín-Gómez and Arbeláez-Cortés

2015), and presence of C. canigularis was associated with larger, and more speciose flocks in our system, although flocks also had significantly less functional and phylogenetic diversity when this species was present (Table 2-2). This species is resilient to fragmentation, being found in forest edge and secondary forest (Isler and Isler 1999, authors’ pers. obs.), possibly because

Chlorospingus spp. are not obligate flockers and consume fruit as well as insects (Valburg 1992).

The resilience of nuclear species can maintain flocking behavior in disturbed landscapes

(Mammides et al., 2015), but the presence of ‘redundant’ nuclear species could achieve the same effect. Several other species, particularly Anisognathus somptuosus and Tangara tanagers, can act as nuclear species in the Andes (Bohórquez 2003; Guevara et al. 2011; HHJ, pers. obs.).

Chlorospingus spp., however, may be followed more by forest-interior insectivores than the several Thraupid tanagers that act as nuclear species, which might explain the reduced diversity in such flocks. Understanding effects of fragmentation and vegetation structure on flocks therefore requires additional studies of how network structure varies among flocks of different nuclear species and how the relative importance of different nuclear species shifts along gradients of fragmentation and vegetation structure (Srinivasan et al. 2010; Arbeláez-Cortés and

Marín-Gómez 2012; Marín-Gómez and Arbeláez-Cortés 2015).

Importance of Patch Size and Vegetation Structure

Our results suggest that fragmentation degrades Andean flock diversity through two distinct mechanisms: reduction in habitat area and simplification of vegetation structure through illegal selective logging. Changes to vegetation structure were just as important as patch area and affected flock diversity in two ways. On one hand, foliage height diversity (closely correlated with canopy height) was associated with increases in flock richness, flock size, and flock encounter rate (Figure 2-3a,b). On the other hand, functional richness increased with the density of large-diameter (DBH > 24 cm) trees (Figure 2-3c), although this variable was negatively correlated with flock size and richness. This suggests that greater flock species richness and size largely depend upon the presence of foraging niches in multiple vertical height bands, but that a portion of flocking bird functional richness only occurs in unlogged, old-growth forests with large-diameter trees. This agrees with previous findings that flock diversity changes along successional and disturbance gradients in vegetation (Knowlton and Graham 2011; Zhang et al.

2013), and that flocks are more diverse in habitats with greater foliage height diversity (Lee et al.

2005; Sidhu et al. 2010; Colorado Zuluaga and Rodewald 2015). Increasing evidence shows that flocks prefer forest with taller canopy height (Mokross et al. 2014; Potts et al. 2014; McDermott et al. 2015), and spend more time foraging in such forest (Mokross et al. 2018). Flock diversity is likely affected by vegetation structure through changes to the availability of foraging niches.

Tropical forest birds forage in narrow height bands and on specific substrates (Marra and

Remsen 1997; Walther 2002; Kotagama and Goodale 2004), and simplification of the vegetation structure could therefore filter out flock participants as foraging niches are lost (Zou et al. 2018).

Reduction in habitat area was also important, as flock species richness and size increased in response to increasing percentage of forest cover (Table 2-2; Figure 2-2a), mirroring similar findings in lowland systems (Maldonado-Coelho and Marini 2004; Brandt et al. 2009; Colorado

Zuluaga and Rodewald 2015). Andean bird communities respond negatively to forest fragmentation (Renjifo 1999), less permeable matrices (Renjifo 2001), and disturbed land use types (O'Dea and Whittaker 2007), and at our study sites the total number of observed flocking species significantly increased with patch size (linear model, beta = 0.267, P = 0.019; Figure 2-

2b) and canopy height (linear model, beta = -3.877, P = 0.027). Any loss of forest interior species, however, was also partially compensated for by the participation of edge-associated species in smaller patches (Figure 2-4, Table 7, Appendix A; see below). The results of the CCA showed that flock composition was affected by patch size, distance to edge, and density of large- diameter trees, which matches previous results showing that both edge flocks (Péron and Crochet

2009), as well as those in disturbed habitat (Sridhar and Sankar 2008; Cordeiro et al. 2015), have different compositions from forest interior flocks. Species richness may therefore be a poor measure of changes to flock functional diversity, phylogenetic diversity and composition at a given site, especially where species loss accompanies species turnover.

Changes to Flock Composition: Loss of Forest Interior Suboscines

Our analyses detected significant changes to flock composition in response to forest fragmentation as well as changes to the flock species richness of taxonomic and functional groups, which we believe explain changes to functional diversity and flock composition along our gradient. Flocks on less fragmented transects with more forest cover located farther from an edge contained a greater species richness of ovenbirds (Furnariidae) and tyrant flycatchers

(Tyrannidae). Furnariid richness was negatively affected by proximity to forest edge and simplified vegetation structure, whereas the participation in the flock of many species of small flock-following tyrannids was positively associated with greater forest cover (e.g. Phyllomyias cinereiceps, Leptopogon rufipectus, Nephelomyias pulcher; Table 7, Appendix A). By contrast, flocks on transects in disturbed patches had greater numbers of tanagers and migrant species.

The richness of thraupid species increased in secondary forest with smaller-diameter trees, while that of migrants significantly increased at forest edges. These trends, as well as the results of the

CCA (Figure 2-4), suggest important taxonomic and functional turnover along the successional and patch size gradients, with suboscine (suborder Tyranni) passerine richness decreasing in more disturbed and fragmented patches. Such species turnover has previously been reported for functional groups, notably forest specialists and open habitat species (Sidhu et al. 2010; Goodale et al. 2014; Cordeiro et al. 2015). Importantly, most of the flock-joining furnariids and tyrannids in this system are forest understory insectivores, a group globally vulnerable to disturbance

(Bregman et al. 2014; Powell et al. 2015a) and often lost from Andean fragments (Renjifo 1999).

In the Neotropics, suboscine forest passerines may be particularly vulnerable because of their specialized foraging niches and behavioral stereotypy (Tobias et al. 2012). Suboscines tend to be specialized on specific substrates in their foraging (Marra and Remsen 1997; Stratford and

Stouffer 2013); the greater vertical vegetation complexity of mature forest may allow more of these specialized insectivores to persist (MacArthur et al. 1966). Loss of vegetation complexity to disturbances such as selective logging could therefore be as important as patch size in determining the persistence of specialized insectivores (Gray et al. 2007; Arriaga-Weiss et al.

2008). Suboscines are also often stereotyped in their diet (Macedo Mestre et al. 2010) and space use (many species are sedentary and avoid edges: Hansbauer et al. 2008; Powell et al. 2013,

2015b), hindering their ability to persist in fragments and recolonize patches following extirpation. Powell et al. (2015a) hypothesized that suboscine stereotypy may explain why

Neotropical insectivores are more strongly associated with certain vegetation structures then their Afrotropical counterparts. Clearly, we need more behavioral studies of how suboscines respond to fragmentation (Tobias et al. 2012).

Other Factors: Seasonal Decreases in Flock Diversity

Although ‘winter’ flocks were joined by many migrants, we observed a small, but significant, seasonal decrease in flock diversity, species richness, and size (Table 2-2), and significant seasonal changes to flock composition (Figure 2-4). We found that flock size, richness, and all measures of functional and phylogenetic diversity in flocks were significantly smaller from January to March relative to June to August. The presence of boreal migrant species may therefore drive a decrease in flocking by resident species from January to March.

Fewer resident species flocked during this period (83 ‘summer’ versus 68 ‘winter’), and most attended proportionately fewer flocks (Table 10, Appendix A). This decrease in percentage of flocks attended was greater in species that infrequently joined flocks from June to August (linear model, beta = 0.918, P = 0.012), suggesting the effect was driven by the loss of irregularly flocking species. It is tempting to conclude that the presence of boreal migrant species, which were recorded in 235 of 270 winter flocks (87%) and on all transects, could be influencing the flocking behavior of resident birds. Non-breeding migrants have traditionally been thought to avoid competition with resident species (Rappole 1995), but foraging shifts by resident birds when migrants are present suggest that exploitation competition may be occurring (Jedlicka et al.

2006). Tropical birds, however, also change their flocking propensities in response to seasonal changes in temperature and rainfall (Mangini and Areta 2018), so it is possible our sampling was capturing this variation. Our data suggest seasonal changes to tropical flocking systems are more important than previously assumed, and sampling at multiple times of year may be necessary to fully describe them.

Management Implications: Maintaining Old-growth, Primary Forests

Our results suggest that while Andean mixed-species flocks persist in small 10-ha fragments, conserving the majority of flocking species biodiversity requires extensive tracts of

tall-canopy forest. Managers should prioritize tracts of mature primary forest with >18-meter canopies that contain canopy emergent trees as well as foliage in the sub-canopy, midstory, and understory. Maximizing the diversity of foraging strata will maximize the encounter rate, species richness, and size of sub-tropical Andean mixed-species flocks (Figure 2-2). Furthermore, old- growth forests with large-diameter (e.g. 30-50 DBH) trees contain a unique subset of flocking bird diversity; selective logging within forest fragments may therefore be especially detrimental to flock diversity. These old-growth trees are characterized by numerous epiphytic plants, lianas, and hanging dead plant matter, which may provide foraging habitat for specialized forest-interior insectivores. Managers should also aim to conserve large areas of forest which minimize the edge-to-interior ratio of the patch. Furnariid species in flocks appear to be edge-sensitive, while some flock-following tyrannids appear to be area-sensitive. Even large (170 ha) fragments had reduced flocking species richness relative to continuous forest, and some flocking species (e.g.

Chlorochrysa nitidissima) appear to be lost from fragments smaller than ~130 ha. We find it unlikely that loss of the nuclear species could contribute to collapse of flocking behavior, because C. canigularis is found in disturbed habitat and other species (Anisognathus, Tangara) can also play nuclear roles.

Table 2-1. Field sites in El Cairo municipality, Valle del Cauca, Colombia. Numbered sites are fragments while the named site is a private reserve connected to continuous forest; latitude and longitude are given for the approximate center point of each transect location. Area of the fragment or reserve is given in hectares. Summer sampling corresponds to the boreal summer (June-August 2017) while winter sampling corresponds to the boreal winter (migrant species present; January-March 2018). CF = number of complete (over 80% composition) flocks observed. TF = number of complete flocks observed on the transect. #T = total number of transects walked. Hrs = total number of hours of transect surveys. FS = total number of flocking species observed on the transect across both seasons. F% = percentage of forest in a 1 km buffer of the transect. ED = edge density calculated as length of forest edge in meters divided by the total area within a 1 km buffer. Summer Winter Site Latitude Longitude Area Transect FS F% ED CF TF #T Hrs CF TF #T Hrs 1 4° 46.984' 76° 12.511' 10 La Cancana 10 9 5 7.7 16 16 9 7.9 33 19.26 18.73 2 4° 47.866' 76° 11.744' 14 La Gitana 12 11 11 5.6 16 14 21 10.7 43 28.93 25.22 3 4° 47.793' 76° 11.286' 28 La Tulia 14 14 15 9.8 13 13 18 9.8 38 13.09 20.15 4 4° 45.904' 76° 08.260' 37 Las Brisas 17 15 19 9.1 17 15 30 12.7 31 32.39 37.91 4° 45.711' 76° 08.101' El Tigre 7 7 13 7.2 3 3 25 10.6 18 26.71 30.88 5 4° 46.857' 76° 13.017' 43 Altamira 18 9 3 2.6 26 18 13 11.5 47 23.77 24.09 6 4° 43.700' 76° 14.706' 107 Altomira 16 9 11 6.5 17 16 28 13.8 48 39.11 25.30 4° 43.474' 76° 15.087' El Eden 20 20 12 9.6 25 25 25 16.2 48 40.01 22.96 7 4° 42.277' 76° 14.630' 147 El Lagito 11 9 17 9.3 19 18 24 15.3 49 41.88 23.47 4° 42.400' 76° 14.986' La Sonora 18 15 12 8.7 15 14 29 14.4 39 51.47 29.33 8 4° 42.384' 76° 12.930' 173 La Guardia 21 15 10 6.5 28 23 24 13.3 48 49.42 26.51 4° 41.964' 76° 13.254' El Rocio 17 13 11 6.7 24 24 16 10.5 42 43.19 25.16 RN El 4° 44.752' 76° 17.448' 750 El Brillante 33 24 19 14.1 28 26 27 16.4 60 87.50 12.38 Ingles 4° 44.526' 76° 17.655' El Ingles 18 18 19 12.3 23 23 26 15.1 51 97.11 7.08 232 188 177 115.7 270 248 315 178.2

Table 2-2. Model-averaged estimates of predictor variables on flock species richness, size, and diversity. Models are full model averages of a candidate model set consisting of all models within 2 ΔAICc of the best model, calculated from all possible model subsets. Sample size of flocks corresponds to the number of complete (80+ % composition) flocks used for each analysis. Average conditional and marginal r2 were calculated across all models in the candidate set. RVI = relative variable importance, corresponding to the proportion of the total Akaike weight represented by all models in which the variable was included. Full descriptions of predictor variables can be found in Table A-8. Richness and size correspond to the observed flock species richness and size respectively. Functional richness and dispersion represent measures of functional diversity, while phylogenetic diversity and mean pairwise distance are indices of phylogenetic diversity; full descriptions of these metrics can be found in Methods. We used standardized effect sizes (SES) of indices of functional and phylogenetic diversity to control for the effect of species richness on these factors.

SES Functional SES Functional SES Phylogenetic SES Mean Pairwise Species Richness Flock Size Richness Dispersion Diversity Distance N Flocks 433 433 426 433 433 433 Candidate Models 6 7 7 18 5 22 Avg. Conditional r2 0.22 0.21 0.14 0.32 0.04 0.05 Avg. Marginal r2 0.22 0.21 0.18 0.37 0.10 0.18 Beta p RVI Beta p RVI Beta p RVI Beta p RVI Beta p RVI Beta p RVI Intercept 1.953 <0.001 2.393 <0.001 -0.411 <0.001 -0.680 <0.001 -0.108 0.23 -0.566 <0.001 Season -0.103 0.03 1.00 -0.107 0.03 1.00 -0.358 <0.001 1.00 -0.508 <0.001 1.00 -0.205 0.01 1.00 -0.291 0.006 1.00 Chlorospingus 0.425 <0.001 1.00 0.436 <0.001 1.00 -0.202 0.002 1.00 -0.640 <0.001 1.00 -0.192 0.03 1.00 -0.029 0.70 0.24 presence Pct. forest 1 km 0.086 0.01 1.00 0.120 <0.001 1.00 -0.090 0.21 0.77 -0.056 0.44 0.49 -0.095 0.32 0.67 -0.001 0.95 0.03 Edge density 1 km 0.003 0.84 0.13 0.002 0.87 0.10 -0.003 0.89 0.10 -0.012 0.76 0.14 -0.090 0.55 0.38 Distance to edge 0.051 0.23 0.74 0.025 0.44 0.52 -0.004 0.82 0.12 -0.062 0.43 0.52 0.034 0.62 0.25 0.060 0.55 0.37 Foliage height 0.118 <0.001 1.00 0.098 <0.001 1.00 0.008 0.74 0.18 0.083 0.16 0.83 0.07 0.33 0.65 0.098 0.36 0.60 diversity Understory density 0.007 0.63 0.31 0.004 0.78 0.23 0.041 0.19 0.77 0.001 0.88 0.06 0.003 0.87 0.07 Large-diameter -0.065 0.002 1.00 -0.052 0.005 1.00 0.093 <0.001 1.00 0.048 0.20 0.78 0.015 0.70 0.24 trees Canopy cover -0.002 0.84 0.13 -0.003 0.83 0.22 -0.036 0.41 0.57 0.002 0.92 0.04

Table 2-3. Model-averaged estimates of predictor variables on flock species richness (FSR) of functional and taxonomic groups. Best models are full model averages of a candidate model set consisting of all models within 2 ΔAICc of the best model, calculated from all possible model subsets. Number of flocks corresponds to the number of complete (80+ % composition) flocks used for each analysis. Average conditional and marginal r2 were calculated across all models in the candidate set. RVI = relative variable importance, corresponding to the proportion of the total Akaike weight represented by all models in which the variable was included. Full descriptions of predictor variables can be found in Table A-8. Response variables correspond to the species richness of ovenbirds (Furnariidae), tyrant flycatchers (Tyrannidae), tanagers (Thraupidae) and boreal migrant species observed in each flock. Furnariid FSR Tyrannid FSR Thraupid FSR Migrant FSR N Flocks 433 433 433 248 Candidate Models 8 13 15 12 Avg. Conditional r2 0.24 0.18 0.36 0.15 Avg. Marginal r2 0.28 0.23 0.40 0.19 Beta p RVI Beta p RVI Beta p RVI Beta p RVI Intercept 0.268 0.002 -0.335 0.10 0.998 <0.001 0.279 <0.001 Season -0.028 0.65 0.32 -0.004 0.90 0.06 -0.273 <0.001 1.00 Flock size 0.338 <0.001 1.00 0.335 <0.001 1.00 0.466 <0.001 1.00 0.294 <0.001 1.00 Chlorospingus presence 0.067 0.47 0.50 -0.100 0.21 0.79 0.005 0.89 0.07 Pct. forest 1 km 0.120 0.53 0.42 -0.017 0.69 0.24 -0.060 0.46 0.51 Edge density 1 km -0.002 0.92 0.08 -0.263 0.17 0.81 0.068 0.30 0.69 -0.006 0.85 0.08 Distance to edge 0.174 0.008 1.00 0.001 0.95 0.05 -0.168 0.055 0.94 Foliage height diversity 0.166 0.010 1.00 0.011 0.78 0.24 -0.002 0.86 0.07 0.007 0.82 0.14 Understory density 0.012 0.67 0.31 0.011 0.74 0.19 -0.044 0.14 0.87 0.002 0.88 0.12 Large-diameter trees -0.004 0.85 0.12 -0.055 0.04 1.00 -0.004 0.84 0.14 Canopy cover 0.008 0.78 0.15 -0.019 0.57 0.38 -0.001 0.92 0.06

Figure 2-1. Map of study fragments in the Western Andes of Colombia. The study landscape is located in El Cairo municipality in the Valle del Cauca department. Black outlines indicate fragmented patches of Andean forest in the landscape, while red lines represent the 500-meter transects used for mixed-species flock transects (n = 14 transects, 8 fragments). The blue line demarcates the boundaries of the Cerro El Inglés private reserve used as a continuous forest reference site for the study. Numbers refer to the fragment study sites described in Table 2-1. Background colors represent land use categories assigned to the landscape from satellite imagery using remote sensing.

Figure 2-2. Landscape effects on mixed-species flock diversity. A) Both the number of participating individuals (flock size; blue circles) and the species richness of foraging flocks (red triangles) increased significantly with percentage forest cover within 1 km of the transect, though the effect size was small (Table 2-2). B) The total number of species observed in mixed-species flocks (pooling across both sampling periods) increased significantly with percentage forest cover (linear model; beta = 0.267, P = 0.02). Percentage forest cover was used as a proxy for patch size and is highly correlated with this variable in our study landscape.

Figure 2-3. Importance of vegetation variables in driving mixed-species flock diversity and encounter rate. Increasing diversity of foliage in different height bands within subtropical Andean forest was associated with increasing A) species richness and B) encounter rate of mixed-species flocks in the Western Andes. C) The functional richness of species in the flock was instead correlated with the density of large- diameter trees on a transect. Vertical vegetation complexity was measured as the Shannon Diversity Index of the proportion of points with vegetation present in each of the five height bands for each 100-meter transect segment (n = 14 transects, 70 segments). Trend lines represent linear model estimates.

Figure 2-4. Biplot of the first two CCA axes. Axis labels show percentage of constrained variance explained by each CCA axis. Blue circles denote site (flock) scores, while red (filled) circles denote species scores (n = 436 flocks, 54 species observed in greater than 3% of flocks). Black arrows and text represent bi-plot scores for the constraining variables. Ellipses represent dispersion ellipses of the standard deviation of site scores, grouped into four categories based on patch size. Species scores for each CCA axis are available in Table A-7 and constraining variable bi-plot scores are reported in Table A-6. Predictor variables are explained in detail in Table A-8.

CHAPTER 3 VEGETATION STRUCTURE DRIVES MIXED-SPECIES FLOCK INTERACTION STRENGTH AND NUCLEAR SPECIES ROLES

The world is currently undergoing a biodiversity crisis, characterized by elevated extinction rates (Ceballos et al. 2015) and large-scale abundance declines of even common species of wildlife (Ceballos et al. 2017, Rosenberg et al. 2019). These declines have resulted largely from the loss of forest habitat and its conversion to other land uses (Young et al. 2016), which has been particularly severe in the humid tropics (Lewis et al. 2015, Barlow et al. 2018).

Species experiencing high rates of forest loss are more likely to be listed as threatened

(Tracewski et al. 2016), particularly because forest loss is accompanied by forest fragmentation, or division of remaining habitat into small patches that contain higher proportions of edge habitat and are often isolated from one another by inhospitable ‘matrix’ habitat (Haddad et al. 2015). In addition to the loss of species, however, interspecific interactions can be vulnerable to fragmentation (Tylianakis et al. 2010, Magrach et al. 2014) and may disappear prior to species loss (Valiente-Banuet et al. 2015). This can result in the ‘functional extinction’ of species that persist in fragments but no longer fulfill their functional role (McConkey and O'Farrill 2016), which can have negative effects on the long-term persistence of species in fragments (e.g. through seed dispersal and recruitment; Cordeiro and Howe 2003).

Although fragmentation is often treated as a single concept, it affects species interactions in multiple ways and at multiple scales (Fahrig 2017). First, species interactions may be affected indirectly by habitat loss at the patch scale (patch size effect) and associated loss of species

(Bagchi et al. 2018, Zou et al. 2018). Biodiversity loss is associated with reduction in patch size, changes to landscape configuration and composition, and increased patch isolation (Wilson et al.

2016, Fletcher et al. 2018), which can lead to a loss of species in mutualistic interaction networks

(Cramer et al. 2007, Cordeiro et al. 2015, Emer et al. 2018). Second, species interactions are also

directly affected by fragmentation effects at a local scale (Bagchi et al. 2018, Zou et al. 2018).

Novel species interactions and climate conditions at the edge of habitat patches (Fagan et al.

1999) can lead to ‘edge effects’ on positive interactions (Galetti et al. 2003, Chen et al. 2017).

Fragmentation of forests can also facilitate habitat degradation within patches (within-patch disturbance effects; e.g. Magrach et al. 2012), which can destabilize positive species interactions

(Knowlton and Graham 2011, Borah et al. 2018). In spite of these multiple mechanisms, however, no study to date has simultaneously assessed the effects of patch size, proximity to edge, and habitat disturbance on positive species interactions.

Arguably the most important positive species interactions for tropical forest birds occur in mixed-species flocks- interspecific groups of mostly insectivorous species that move and forage together (Morse 1970). These interspecific associations are led by ‘nuclear speices’ that contribute to the formation and coherence of the flock (Moynihan 1962), which are, in turn, followed by numerous ‘attendant’ species. In some systems, species may form several flocking sub-types based on shared foraging stratum (Munn 1985; Srinivasan et al. 2012) or body size

(King and Rappole 2001), likely by associating with more similar nuclear species (Mammides et al. 2015). Participation in flocks is ubiquitous among forest bird communities (Zou et al. 2018), presumably because participants obtain benefits in terms of reduced predation risk and increased foraging efficiency (Sridhar et al. 2009). Flocking species also benefit from social information about predators, often gained by eavesdropping on gregarious nuclear species (Goodale and

Beauchamp 2010, Sridhar and Shanker 2014, Pagani-Nuñez et al. 2018), which allows them to exploit riskier microhabitats (Darrah and Smith 2013, Martínez et al. 2018). Flocking increases the survival of forest birds (Dolby and Grubb 1998, Srinivasan 2019), and allows them to persist in disturbed matrix habitat (Goodale et al. 2014, Zhou et al. 2019). Though the size and species

richness of mixed-species flocks are known to be negatively affected by fragmentation (Zou et al. 2018), changes to species interactions themselves have rarely been studied.

Recently, network analysis (Farine and Whitehead 2015) has been used to quantify changes to interspecific flocking interactions (Farine et al. 2012). Collapse or resilience of flocking behavior can be quantified as the strength of association (co-occurrence) between each pair of species in a flocking network. If flocking behavior loses coherence in disturbed habitats, we would predict that species participation would be less frequent and less temporally stable, resulting in fewer, weaker interspecific co-occurences (e.g. Mokross et al. 2014, Borah et al.

2018, Zhou et al. 2019). Alternatively, changes to flocking interactions could be driven by turnover among participating species from forest-dependent to generalist species along gradients of increasing fragmentation (Sridhar and Sankar 2008, Cordeiro et al. 2015, Chapter 2) in addition to changes in co-occurrence patterns (e.g. Mammides et al. 2018). Partitioning of network dissimilarity allows us to quantify the proportional change in flocking interactions explained by each mechanism (Poisot et al. 2012). If nuclear species in flocks drop out in response to fragmentation or disturbance gradients (Mammides et al. 2015, Zhou et al. 2019), then species interactions could be adversely affected. Alternatively, nuclear species might either exhibit resilience in the face of fragmentation, or forest-dependent nuclear species might be replaced by novel or redundant nuclear species in more fragmented habitats. Network analyses allow the ‘nuclearity’ of a species to be quantified as their centrality in the network, and to be statistically modeled in response to environmental covariates.

To quantify changes to flocking interactions in response to forest fragmentation, we surveyed mixed-species flock composition along a patch size gradient (10-173 ha) and in a continuous forest reference site in subtropical montane forest of the Western Andes of Colombia.

We constructed 27 social networks of flocking interactions observed on 14 transects sampled during both the boreal summer (June-August 2017) and boreal winter (January-March 2018). We then used social network analysis to answer the following questions: (1) Do Andean flocking interactions show a loss of coherence in response to forest fragmentation? (2) What types of fragmentation effects (patch area, edge effects, vegetation disturbance) drive changes in flocking interactions? (3) Are changes in flocking networks driven by turnover in species composition or changes in species co-occurance patterns within flocks? (4) Does the importance of different nuclear species change in response to fragmentation? And (5), do flock sub-types dissappear in forest fragments? If flock coherence is negatively affected by patch size, then we would predict that overall connectedness, strength of co-occurrence interactions, and degree of clustering in the network will decline with patch size. Alternatively, if disturbance to vegetation structure or edge effects are more important, than we would predict that network structure would be better explained by vegetation structure or density of forest edge near the sampling transect.

Methods

Study System and Sites

We conducted field work in subtropical humid montane forests located within the municipality of El Cairo, Valle del Cauca department in Colombia. The study region is located in the Serranía de los Paraguas mountain range, part of the Western Andes, and a center of avian threatened species diversity and endemism in Colombia (Ocampo-Peñuela and Pimm 2014).

Andean forests in Colombia are highly fragmented, with only 30% of original forest cover remaining (Etter et al. 2006). Our focal landscape consists of the typical patchwork of forest fragments embedded in a matrix of cattle pasture, regenerating scrub, and coffee farms

(Armenteras et al. 2011). Within this landscape, we selected eight fragments representing a gradient in patch sizes (range 10 to 170 ha; Table 3-1, Figure 2-1). We chose sites in the same

altitudinal belt (1900-2200 masl) and matrix type (cattle pasture), because Andean flock composition changes with altitude (Marín-Gomez and Arbeláez-Cortés 2015), and matrix type affects species composition in Andean fragments (Renjifo 2001). Within-patch disturbance is common in Andean fragments in Colombia, especially illegal selective logging which can reduce tree diversity and shift vegetation structure towards early successional forests (Aubad et al.

2008). We therefore established 500-meter transects through forest interior (n = 14 total transects; Table 3-1) as our sample replicate. Where possible, we placed a transect in a more disturbed (logged) area and another in a relatively undisturbed site within each fragment.

We stratified forest fragments into large (≥ 100 ha), medium (30-50 ha), and small (≤ 20 ha) size categories and selected a minimum of two replicates of each (Table 3-1). We also performed field work in a non-fragmented reference site (Reserva Natural Comunitária Cerro El

Inglés) connected to thousands of hectares of continuous forest. Fragments were privately-owned forests, and we collaborated with a local NGO (Serraniagua; http://www.serraniagua.org) to ensure access. We only selected fragments with primary or late-successional secondary forest; vegetation structure and canopy height varied substantially between patches based on intensities of selective logging and land use histories. Fragments were separated by ≥ 100 meters to minimize cross-patch movement of birds, and all transects in different patches were separated by at least 250 m. Mixed-species flocks in Colombian sub-montane forests are large and diverse, with ~100 participating species and a maximum of 50 individuals per flock (Colorado Zuluaga and Rodewald 2015, Chapter 2). Boreal migrant species commonly join these flocks while overwintering and are among the most common flocking species during this period (McDermott and Rodewald 2014, Marín-Gomez and Arbeláez-Cortés 2015).

Transect Surveys for Mixed-species Flocks

We adapted transect surveys for mixed-species flocks from Goodale et al. (2014), which were conducted within forest fragments from June-August 2017 (boreal migrants absent) and

January-March 2018 (boreal migrants present). Both sampling periods correspond to a dry season in the Western Andes, which has a bimodal two dry, two wet seasonality. For each sampling period, we sampled each transect for 2.5 consecutive field days by continuously walking back and forth; we staggered visits to continuous forest, large fragments, and small fragments to avoid a temporal bias in our sampling. Transects were walked slowly by 2-3 observers familiar with all bird species by sight and sound (HHJ present for all surveys), with visits distributed across the morning (7:30-11:30) and evening (15:00-17:30) hours when flocking behavior is most common in Andean forests (Arbeláez-Cortés et al. 2011). When a flock was encountered, we spent up to a maximum of 45 minutes characterizing it with 10x binoculars. We identified flocking species by sight and sound, as flock members often called and sang conspicuously while foraging. At least 5 minutes were spent with each flock, following it if possible. Because detection of bird species in mixed-species flocks is imperfect, we considered flock composition to be ‘complete’ when we estimated that at least 80% of individuals in the flock had been observed; we only used ‘complete’ flock compositions in our analyses. We considered the flock composition ‘complete’ when we did not detect new species or individuals over the past 5 minutes of observation. Avian taxonomy of flocking species follows Quiñones

(2018).

Calculation of Landscape Level Variables

We obtained landscape-level variables for analyses using geographic information software (GIS) analysis in ArcGIS (ArcMap 10.3.1; Esri, Redlands, CA). We quantified landscape composition and configuration by buffering each transect (n = 14) by 1 km, a scale

that has been shown to affect the richness of tropical forest birds in fragmented landscapes

(Cerezo et al. 2010, Carrara et al. 2015). We then calculated measures of landscape composition and configuration using a land cover use categorization made by the Corporación Autónoma

Regional del Valle del Cauca, converted to a 25m cell-size raster. To quantify landscape composition, we calculated percentages of forest land use type within each 1 km buffer using the

‘isectpolyrst’ tool in Geospatial Modelling Environment (version 0.7.4.0; Beyer 2015).

Following Carrara et al. (2015), we selected percentage of forest cover as a proxy for patch size because some of our transects were located in continuous forest with no patch size measurement.

Because the matrix in our landscape consisted of unforested cattle pasture, we feel that this is a good measure of patch area, since there was no other forested habitat in the buffer area; percentage forest was highly correlated with patch area (correlation coefficient = 0.95). We also measured landscape configuration for each transect at the 1-kilometer buffer scale using edge density, or length of all forest edges (in meters) divided by total buffer area (in hectares), as described by Carrara et al. (2015). We did not include other (e.g. 500 m) buffer scales because composition and configuration metrics correlated heavily with the 1 km scale.

Vegetation Measurements and Principal Component Analysis

To quantify disturbance to local vegetation at each site, we measured vegetation structure and density along each transect used to sample for flocks. Vegetation measurements were made between June and August 2017 and based on our field observations there appeared to be minimal variation between seasons. We used the methodology of James and Shugart (1970) following the modifications made by Stratford and Stouffer (2013), and further modified it to be used with belt transects. The sampling consisted of two components for each transect: (1) the quantification of canopy cover, ground cover, canopy height, and vertical structure of vegetation using point sampling spaced every 10 meters on the transect and (2) the quantification of shrub, vine, fern,

palm, and tree fern and tree density using 3-meter-wide belt transect sampling. Because transects ran along trails, we measured vegetation at least three meters from the trail edge on a randomly selected side for each 100-meter transect segment.

For the point sampling, we measured eight variables at 10-m intervals, for 50 points per transect. As a measure of vertical vegetation structure along the transect, we noted the presence or absence of live vegetation at five heights: <0.5 m, >0.5–3 m, >3–10 m, >10–20 m, and >20 m. Above 3 meters, we used a rangefinder to determine heights and sighted through a tube with crosshairs while straddling the point. The canopy height at each point was measured using a laser rangefinder (Raider 600 Digital Laser Rangefinder; Redfield Inc. Beaverton, OR) pointed at the highest foliage. Canopy and ground cover were calculated to the nearest 1/8th of the field of view by sighting through a vertical canopy densitometer (GRS Densiometer; Geographic Resource

Solutions, Arcata, CA). For each transect, we averaged values for canopy height, canopy cover, and ground cover, and calculated the proportion of points at which vegetation was present for each height category. For the belt transect sampling, we surveyed vegetation along the same transects and calculated densities for each 100-meter transect interval. We counted all shrubs, vines, ferns, tree ferns, and palms encountered on 1.5 meters to either side. Secondly, we counted all trees (woody vegetation > 2 m in height) within 1.5 meters of the transect and measured their diameter at breast height (dbh). Trees were later categorized into six dbh size classes for analysis: 1-7 cm, 8-15 cm, 16-23 cm, 24-30 cm, 31-50 cm, and > 50 cm. We additionally recorded the largest tree as a measure of degree of logging in each fragment.

We retained four measures of local vegetation for our analyses. The average canopy height and canopy cover for each transect were directly used for the analyses. Canopy height was strongly correlated with degree of vertical vegetation complexity (i.e. presence of foliage in

different height categories), as calculated using the Shannon diversity index on the proportion of points with vegetation present in each of the five height bands for each transect (correlation coefficient = 0.89). We also used two principal component analysis (PCA) ordinations of understory vegetation density and ground cover, density of large-diameter trees, calculated for each transect. Ordinated predictor variables were taken from Chapter 2, ordinations are described in Appendix A. We used the first PC axis from each ordination; more negative values of the understory vegetation PC axis indicate higher densities of understory shrubs, vines, palms, ferns, and tree ferns while more positive values of the tree-size PC axis indicate greater densities of large-diameter trees (e.g. 20-50 cm DBH).

Construction, Measurement, and Analysis of Social Networks

We characterized the flocking interactions of the bird community on each 500-meter transect by assembling a social network (Farine and Whitehead 2015). Species move and forage in close association in mixed-species flocks, so we considered two species observed in the same flock to be interacting ecologically (the ‘gambit of the group’ approach; Franks et al. 2010). We therefore defined each node as an individual species and each edge as a co-occurrence of two species in a flock. All statistical analyses were performed in R (version 3.5.1; R Core Team

2020). We used presence-absence, flock-by-species adjacency matrices derived from field observations to create social networks using the get_network function of the asnipe package

(Farine 2013). We constructed one social network for each transect by sampling period combination using all flock compositions observed on the transect during that sampling period

(range = 7-26 flock compositions per network). However, we did not construct a network for one transect during the boreal winter due to insufficient sample size of flock compositions (n = 27 networks). Because detectability of birds in flocks is high, and associations were likely rarely missed, we used the simple ratio index (SRI), an undirected, weighted measure of association, to

calculate an association index for each species pair in the flocking network (Farine and

Whitehead 2015).

For each network, we calculated both global and node-based metrics relevant to our questions of interest. We quantified the strength of flocking associations using mean normalized degree (equivalent to edge density), mean weighted degree (hereafter strength), and skewness of mean weighted degree values for each network. Mean normalized degree represents the average of the number of edges for each node, divided by the total number of nodes minus one. Strength is calculated as the average of all sums of weighted edge lengths for each node in the network.

Skewness of mean weighted degree measures the distribution of weighted degree measurements across all network edges; large values of skewness indicate greater numbers of weak species interactions. Mean normalized degree was calculated using the degree function of the igraph package (Csárdi 2019), strength was calculated using the strength function of the igraph package, and skewness was calculated using the skewness function of the moments package

(Komsta and Novomestky 2015). Following Mokross et al. (2014), we also used the global clustering coefficient, calculated using the transitivity function of the igraph package, as a measure of flock cohesiveness. Lastly, we quantified the extent to which flocks were divided into sub-types versus homogenous in composition by calculating the modularity of each flocking network. First, we assigned each node (species) to a flocking sub-type using the eigenvector modularity method (Newman 2006) applied through the cluster_leading_eigen function of the igraph package. We then calculated the modularity of this optimal solution using the modularity function of the igraph package.

In order to understand how nuclear species importance changed across our patch size gradient, we calculated node-based measures of centrality for six a priori-defined leader species.

Putative nuclear species were identified based on field observations and descriptions of Andean flocking systems in the literature. We chose Chlorospingus canigularis, Tangara labradorides,

T. arthus, and Anisognathus somptuosus because they are gregarious species that often join flocks as social groups, a trait often associated with nuclear species (Goodale and Beauchamp

2010), while Myioborus miniatus and Pachysylvia semibrunnea are commonly flocking species that frequently call and sing in the flock, possibly contributing to cohesion. We quantified

‘nuclearity’ as the normalized betweenness centrality of a species in the network, or the number of shortest paths between nodes passing through the node of interest, standardized by the size of the network. On one transect, C. canigularis was replaced by C. semifuscus, which plays a similar nuclear role in flocks (Bohórquez 2003) and is ecologically similar to C. canigularis

(Isler and Isler 1999). We therefore used the centrality value for C. semifuscus on that transect.

Centrality was calculated using the betweenness function of the igraph package.

To test for the effect of fragmentation variables on our global network metrics and the centrality of each nuclear species, we ran linear mixed models (LMM; Bolker et al. 2009) fit using the lme4 package (Bates et al. 2015). We included two random effects of transect and site to account for the non-independence of ‘summer’ and ‘winter’ networks from the same transect as well as transects from the same forest fragment. We also included five fixed effects: percentage of forest and density of forest edge within 1 km of the transect, average canopy height along the transect, a PCA axis representing density of large-diameter trees, and season.

The mean normalized degree was log transformed for normality prior to analysis. We checked for multicollinearity of the predictor variables using variance inflation factors; no predictor variable had a VIF of ≥ 3 for any model. We evaluated the significance of each predictor variable using a likelihood ratio test to compare the full model with a reduced model without the variable

of interest. Model goodness-of-fit was assessed using the marginal and conditional coefficients of determination (Nakagawa and Schielzeth 2013), implemented using the r.squaredGLMM function in the MuMIn package (Bartoń 2016).

Calculation and Analysis of Network Dissimilarity

To understand if changes to flocking interactions along our patch size gradient were driven by species turnover of flock participants or changes to the interactions themselves, we partitioned pairwise network dissimilarity following Poisot et al.’s decomposition (2012). This divides overall pairwise dissimilarity (βWN) into the dissimilarity attributable to differences in species composition (βST) and the dissimilarity attributable to differences in the interactions of species pairs co-occuring in both netoworks (βOS). Network dissimilarity partitioning was run using the network_betadiversity function of the betalink package (Poisot 2016); we divided the average pairwise turnover and interaction dissimilarity components by the average overall dissimilarity in order to obtain the proportion of changes to flocking networks explained by each.

In order to understand how our predictor variables affected each component of network dissimilarity, we ran a distance-based redundancy analysis (dbRDA; Legendre and Anderson

1999) on both the βST and βOS components of dissimilarity. We used the same five predictor variables used for the generalized linear mixed models, as well as the average canopy cover and density of understory vegetation. We ran the dbRDA using the capscale function of the vegan package (Oksanen et al. 2019); we square root transformed the dissimilarities to make the distances Euclidean, as recommended for such data (Legendre 2014). We tested for the marginal significance of each predictor variable using an ANOVA-like permutation test (Legendre et al.

2011), implemented with 999 permutations using the anova.cca function in the vegan package.

Results

Flock Observations and Network Metrics

During two three-month field seasons (86 days in the field and ~294 hours walking transects), we observed 502 ‘complete’ mixed-species flocks of which 436 were on a transect and retained for analyses (Table 3-1). We observed 188 transect flocks in 177 transect surveys

(range = 3-19 transect walks per site) during the boreal summer and 248 transect flocks in 315 transect surveys (range = 9-30 transect walks per site) during the Boreal winter. In total, we observed 99 bird species in mixed-species flocks, including 79 species during the boreal summer and 78 species during the boreal winter, of which 11 were migrants. The primary families observed included Furnariidae, Tyrannidae, Thraupidae, and Parulidae; we also observed two species of squirrel, Sciurus granatensis and Microsciurus sp., participating in flocks on 16 occasions. The full list of bird species observed in flocks is reported in Table A-10. The species richness (mean = 9.3, range = 2-27) and number of individuals (mean = 14.6, range = 3-51) in flocks varied both within and between transects. The overall number of species observed in flocks declined with the fragment size (linear model; beta = 0.22, P = <0.001), and by consequence so did network size. Observed flocking species richness for each season-by-transect combination therefore varied considerably (mean = 32.6, range = 16-48; Table 3-1).

LMM Analysis of Network Metrics and Species Nuclearity

Our LMM analysis found few effects of fragmentation variables on any of the five network-based metrics measured, including no significant effects of percentage of forest cover within 1 km or edge density (Table 3-2). While there was no effect of patch area or edge density, we found that network strength, a measure of the strength of species co-occurrence patterns in flocks, showed a significant positive relationship with a measure of vegetation structure (canopy height; beta = 0.21, P = 0.03; Figure 3-1). Generally, we found a high proportion of realized

flocking interactions (mean normalized degree = 0.65), a high clustering coefficient (mean =

0.79), and a low degree of modularity (mean = 0.06) regardless of transect, and both skewness of degree distribution and network modularity were not significantly correlated with any predictor variables. In addition to the fragmentation variables, we found significant effects of season on numerous network metrics: flocking networks showed a significantly lower mean normalized degree (beta = -0.11, P = 0.008), strength (beta = -0.97, P = 0.006) and clustering coefficient

(beta = -0.05, P = < 0.001) during the January-March sampling period compared to the June-

August sampling period. Models had a high goodness-of-fit (conditional r2 range = 0.20-0.45;

Table 3-2), though the random effects of transect and site explained comparatively little variance.

In contrast to the network metrics, we found that the centrality of several species in the flocking networks was correlation with percentage of forest with 1 km as well as vegetation variables (Table 3-3). The average betweenness centrality was higher for C. canigularis (0.04),

T. labradorides (0.04), and M. miniatus (0.05), and slightly lower for T. arthus (0.03) and P. semibrunnea (0.02). Four of the six species showed significant, and often contrasting, changes to centrality in response to forest fragmentation. The chlorospingus C. canigularis increased in centrality with the density of large-diameter trees (beta = 0.021, P = < 0.001; Figure 3-2a). By contrast, three other species increased in centrality at more disturbed sites. T. labradorides (beta

= -0.001, P = 0.03; Figure 3-2b) and P. semibrunnea (beta = -0.001, P = 0.002; Figure 3-2b) decreased in centrality with increasing percentage of forest within 1 km of the transect, our proxy for patch size. T. arthus decreased in centrality with increasing canopy height (beta = -0.006, P =

0.01) and large-diameter tree density (beta = -0.016, P = 0.005; Figure 3-2a). Two tanagers

(Thraupidae) also decreased in centrality during the boreal winter: T. labradorides (beta = -

0.051, P = 0.004) and A. somptuosus (beta = -0.031, P = 0.04). C. canigularis, by contrast, showed a non-significant increase in centrality during this sampling period (beta = 0.024, P =

0.053). We did not find any significant effects of season or fragmentation on the centrality of M. miniatus. Model goodness-of-fit was similar to that of the global network metric models

(conditional r2 range = 0.18-0.67; Table 3-3).

Partitioning Network Dissimilarity Using dbRDA

We found a relatively high degree of pairwise dissimilarity across all network pairs (βWN; mean = 0.65; Figure 3-3), of which 66% was partitioned into the species turnover component

(βST; mean = 0.44) and 34% into the differences in interaction component (βOS; mean = 0.21).

When we ran a dbRDA on the partitioned network dissimilarity, we found contrasting effects of fragmentation predictor variables on βST and βOS (Table 3-4). For the βST dbRDA, the constrained ordination explained 49% (2.80) of the total inertia (5.71). The species turnover and loss component of network dissimilarity was significantly explained by the landscape variable percentage of forest within 1 km (F = 3.13, P = 0.002) and the vegetation variables canopy height (F = 3.50, P = 0.001) and density of large-diameter trees (F = 1.84, P = 0.02). There was also a significant effect of season on the turnover component (F = 2.87, P = 0.001). By contrast, the βOS dbRDA only explained 33% (0.97) of the total inertia (2.92), and we found few significant predictors of this component. We found no effect of either edge density or percentage of forest within 1 km on this component of network dissimilarity, and, among the vegetation disturbance variables measured, the only one that significantly explained βOS was the density of understory vegetation (F = 1.52, P = 0.02). We also found a seasonal effect on this network dissimilarity component (F = 2.13, P = 0.001).

Discussion

We found few effects of forest fragmentation on the coherence and modularity of flocking interactions within the range of fragment sizes sampled. Neither percentage of forest nor edge density within 1 km of the sampling transect had a significant effect on any network metric measured, and four of five metrics were not significantly explained by any fragmentation variable. We did find, however, that network strength, a measure of frequency of species co- occurrences in flocks, increased significantly with canopy height (Figure 3-1). Mixed-species flocks therefore appear to have greater cohesion and a more structured composition in primary forest relative to secondary forest. However, 66% of the pairwise network dissimilarity was explained by species turnover (Figure 3-3), which was itself significantly explained by both landscape and vegetation variables. This suggests that coherence of flocking behavior itself is maintained even as extensive species turnover occurs from continuous forest to small fragments.

We also found evidence of redundant nuclear species that replace one another along the fragmentation gradient. C. canigularis persisted in small fragments but was most central in old- growth forest with many large-diameter trees. The omnivorous tanagers T. labradorides and T. arthus, on the other hand, increased in centrality as percentage of forest around the transect and density of large-diameter trees declined, respectively (Figure 3-2). Finally, we detected strong seasonal effects on the number and frequency of species co-occurrences, as well as the centrality of two species of tanager, which declined significantly during the boreal winter (January-March).

Maintenance of Flock Cohesion Amidst Species Turnover: Open-membership Flocks?

Contrary to our expectations, we found little evidence of landscape-level effects on the coherence of flocking behavior itself, at least down to 10 ha fragments. Four of five global network metrics were not significantly affected by any of the fragmentation variables used in our analysis, and only 34% of pairwise network dissimilarity was partitioned into the βOS component.

This result stands in contrast to previous findings of lower strength of association in small 10 ha fragments in Amazonia (Mokross et al. 2014) and reduced normalized degree of flocking networks in disturbed land use types in tropical China (Zhou et al. 2019). Instead, we found that

66% of pairwise dissimilarity in flocks was attributed to βST, and this component was significantly explained by percentage of forest within 1 km as well as the canopy height and density of large trees on a transect. This result mirrors our findings of extensive changes to flock species composition, as well as loss of flocking species richness, along this same fragmentation gradient (Chapter 2). Fragmentation-related changes in our Andean system therefore appear to be primarily driven by ‘indirect’ changes to the flocking bird community (Zou et al. 2018) rather than ‘direct’ changes to flocking behavior itself. Such species turnover in flock composition is commonly observed in the face of anthropogenic change, with differing species compositions reported in fragments (Maldonado-Coelho and Marini 2004; Cordeiro et al. 2015; Chapter 2) and disturbed land-use types (Lee et al. 2005; Zhou et al. 2019; Goodale et al. 2014).

The ‘open membership’ nature of Andean flocks may explain the persistence of flock cohesion in the face of species turnover. Andean flocks have a variable composition, with species frequently joining and leaving the flock (Poulsen 1996b; HHJ pers. obs.). These flocks also consist of a mix of understory and canopy species (Poulsen 1996b, Guevara et al. 2011,

Colorado Zuluaga and Rodewald 2015), unlike the well-defined canopy and understory flocks of many lowland systems (e.g. Munn 1985, Srinivasan et al. 2012). On average, ~65% of possible species co-occurrences were observed in our flocking networks, and strength of co-occurrence was skewed towards numerous weak interactions (mean skewness of normalized degree = 1.2), in contrast to the more numerous strong interactions in Amazonian primary forest flocks

(skewness of normalized degree = ~0.2; Mokross et al. 2014). Our flocks also showed low

modularity (mean = 0.06), unlike the higher values in systems with multiple flock types (e.g.

~0.45; Zhou et al. 2019). Insectivorous understory bird species are frequently lost from fragments (Bregman et al. 2014) and are more likely to be lost from flocks in disturbed habitat

(Lee et al. 2005, Sridhar and Sankar 2008, Cordeiro et al. 2015). Where flocks are ‘open membership’ in nature, consisting of numerous weak interactions, this loss of flocking species can be compensated for by species turnover to generalist species, which maintain the ‘interaction richness’ (sensu Tylianakis et al. 2010) of the behavior. Specialized and highly structured understory flocking assemblages with more strong interactions, on the other hand, appear to lose cohesiveness from the loss of forest-specialist species that are not replaced (Maldonado-Coelho and Marini 2004, Van Houtan et al. 2006, Borah et al. 2018).

Vegetation Structure Affects Strength of Flocking Interactions

Network strength, or the frequency of species co-occurrences in flocks, was positively correlated with canopy height, suggesting that stability of species co-occurrences, but not overall number of co-occurrences, increased with canopy height. This mirrors Mokross et al.’s (2014) finding that Amazonian understory flocks showed greater clustering as canopy height increased and a greater network strength in primary forest. Canopy height at our study sites was highly correlated with the vegetation structural complexity, and our finding agrees with a large body of literature suggesting that flock species richness (Sidhu et al. 2010, Knowlton and Graham 2011,

Zhang et al. 2013, Colorado Zuluaga and Rodewald 2015, Chapter 2), flocking propensity of flock-joining species (Knowlton and Graham 2011, Zhang et al. 2013, Mokross et al. 2014, Zhou et al. 2019), and encounter rate of mixed-species flocks (Zhang et al. 2013, Mokross et al. 2018,

Chapter 2) increase with vegetation structural complexity. Furthermore, flocks preferentially use high-canopy forest (Potts et al. 2014, McDermott et al. 2015), and spend more time foraging in such forest (Mokross et al. 2018). We suggest that this association with greater vegetation

structure may make flocking behavior more sensitive to selective logging practices, which simplify vegetation structure by reducing canopy cover, tree density, and understory vegetation density (Thiollay 1997, Sekercioglu 2002), and removing late-successional tree species (Aubad et al. 2008). Indeed, the coherence of flocks has been shown to be reduced where vegetation structure has been simplified through logging (Borah et al. 2018) and livestock grazing

(Knowlton and Graham 2011).

The specific mechanism that connects flock coherence and vegetation structure is poorly understood but may have to do with changes to the availability of foraging microhabitats and density of mixed-species flocks (Zou et al. 2018). Many tropical insectivorous species are known to be highly specialized on specific foraging substrates (Rosenburg 1990, Marra and Remsen

1997) and foraging height bands (Walther 2002, Mansor et al. 2019), which may help to partition the foraging niches of flock-joining species. Where simplification of vegetation structure occurs, the loss of specialized foraging substrates may lead to the loss of the foraging niche and a switch to foraging outside of the flock, lowering flocking propensity. For example, many cloud forest species specialize on epiphytic plants (Sillett 1994), yet Andean secondary forests with reduced vegetation structure have lower epiphyte diversity relative to primary forest (Koster et al. 2009).

Similarly, insectivorous Andean birds preferentially forage on a subset of tree genera (Tarbox et al. 2018), yet selectively logged Andean forests show reduced tree species richness (Aubad et al.

2008). In some systems, secondary forest also has a lower density of nuclear species (Zhang et al. 2013), and flocks in such forests have larger home ranges (Mokross et al. 2018), leading to lower flock encounter rates (see above). Low flock densities could reduce flock cohesion if widely ranging flocks are harder to encounter, move more rapidly (leading to a loss of ‘slow- foraging’ species), or spend too little time in specialized microhabitats (e.g. treefall gaps).

Nuclear Species Redundancy and Turnover

Our analysis of a priori-defined nuclear species along our patch size gradient suggested that multiple species can play the nuclear role, and that ecological conditions dictate the importance of different species. Many other studies have identified ‘redundant’ nuclear species within the same system that lead flocks along forest edges (Péron and Crochet 2009), at different height strata (Srinivasan et al. 2012, Mammides et al. 2015), and in different land-use types

(Goodale et al. 2014, Zhou et al. 2019). However, this is the first study, to our knowledge, to show that nuclear species can change in importance within the same habitat type in response to fragmentation. In some well-described flocking systems, a single nuclear species plays an outsized role in flock coherence and movement, such as Thamnomanes antshikes in Amazonia

(Williams and Lindell 2019) or Baeolophus titmice in North America (Contreras and Sieving

2011). By contrast, descriptions of middle-elevation Andean flocking systems have classified numerous genera as playing a nuclear role, including Chlorochyrsa, Chlorospingus, Myioborus,

Basileuterus, Tangara, Anisognathus, and Pachysylvia (formerly Hylophilus), with little concordance between studies (Bohórquez 2003, Arbeláez-Cortés et al. 2011, Guevara et al. 2011,

Colorado Zuluaga and Rodewald 2015, Muñoz 2016). Even in Andean field sites in the same region and at similar altitudes, the density and ‘nuclear status’ of species varied considerably from site to site (Marín-Gomez and Arbeláez-Cortés 2015).

We suggest that much of this variability may result from the fact that Andean flocks possess multiple, redundant nuclear species whose importance can vary based on local ecological conditions. Marín-Gomez and Arbeláez-Cortés (2015) suggest that the importance of an Andean nuclear species may be linked to its local density, which agrees with our findings here. While it can persist in secondary forest, C. canigularis had the highest centrality in old-growth forest.

This species is thought to forage preferentially on woody vine tangles and hanging dead

vegetation in the canopy and sub-canopy (Isler and Isler 1999, HHJ pers. obs.), which may be more common in old-growth forest with many large-diameter trees (Dewalt et al. 2000). T. labradorides and T. arthus, on the other hand, are generalist species that are most commonly observed foraging along forest edges and in secondary forest (Isler and Isler 1999, Naoki 2007).

Similarly, P. semibrunnea is a species of edge habitat and early-successional forest (Slager

2011), which could make it more common in smaller forest patches. These species had the highest centrality in small fragments and secondary forest respectively and appear to occur at higher densities in such habitat (authors’ personal observations). Given the turnover of nuclear species in our system, it seems unlikely that flocking behavior is being negatively affected by the loss of nuclear species, as reported in other systems (e.g. Maldonado-Coelho and Marini 2004,

Cordeiro et al. 2015). However, flock-following species are more likely to join ecologically similar nuclear species (Mammides et al. 2015), which could act as a mechanism to shift the composition of Andean flocks across sites.

Seasonal Changes to Interaction Strength and Nuclear Species Importance

Surprisingly, we found that the number and strength of species co-occurrences in flocks, as well as the importance of nuclear species, significantly differed between our boreal summer and boreal winter sampling periods. The winter flocks had a lower number of realized co- occurrences in flocks, a lower clustering coefficient, and a lower network strength, indicating a relative reduction in the coherence of flocking interactions during this period. We also found that two Thraupid tanager species were significantly less central in flocking networks during this period, while the Passerellid Chlorospingus showed a non-significant trend of increasing centrality. These results match our previous finding from this landscape that flock species richness, size, and functional diversity all significantly declined during this period (Chapter 2).

Indeed, the number of resident flocking species declined from 79 during the summer to 67 during

the winter, and we previously (Chapter 2) found that species with low flocking propensities in the boreal summer were more likely to drop out during the winter. Tropical flocking systems are often assumed to be relatively invariant across the yearly cycle (Zou et al. 2018), but cloud forest birds appear to vary their flocking propensities considerably in response to seasonal rainfall and temperature patterns (Mangini and Areta 2018), as well as daily cloud cover (Poulsen 1996a).

However, migratory species were present in 87% of winter flocks, which could change the competitive dynamics in flocks (Jedlicka et al. 2006). Flocking species co-occur more frequently with ecologically similar species (Sridhar et al. 2012, Mammides et al. 2018), so the addition of migratory species could also increase or decrease the frequency with which other species join flocks in relation to their ecological similarity to small Parulid warblers. Future studies should focus on seasonal changes to tropical flocking systems to understand the extent to which biotic or abiotic factors explain variation in flocking propensity.

Conservation Implications

Our results suggest that flocking behavior in the Andes persists into small 10 ha fragments where primary forest is maintained. We found no effect of patch size or patch edges on the coherence of flocking behavior, though the frequency of species co-occurrences was significantly reduced in secondary forest with canopies under 17 m in height, regardless of patch size. Canopy height was highly correlated with vertical structural complexity of the vegetation, so within-patch disturbance in Andean fragments, including selective logging practices, will negatively affect the coherence of flocking behavior. Changes to flocking interactions occur mainly through extensive species turnover in flock participants that was significantly correlated with patch size, canopy height, and stand age. Therefore, persistence of flocking behavior in fragments does not necessarily mean persistence of forest-dependent flocking species, which are frequently lost from fragments (Chapter 2). The resiliency of flocking behavior in the Western

Andes is in part due to the ‘open membership’ nature of flocks, which consist of numerous weak interactions and a lack of flock sub-types. It is also due to the existence of many redundant nuclear species which replace each other along the fragmentation gradient. C. canigularis is the most important nuclear species in primary old-growth forest, while T. arthus, T. labradorides, and P. semibrunnea play a more important nuclear role in smaller fragments and areas of secondary growth. Flock conservation strategies focusing on the conservation of keystone nuclear species are therefore unlikely to be effective in the Andes.

Table 3-1. Field sites in El Cairo municipality, Valle del Cauca, Colombia. Numbered sites are fragments while the named site is a private reserve connected to continuous forest; latitude and longitude are given for the approximate center point of each transect location. Area of the fragment or reserve is given in hectares. Summer sampling corresponds to the boreal summer (June-August 2017) while winter sampling corresponds to the boreal winter (migrant species present; January-March 2018). CF = number of complete (over 80% composition) flocks observed. TF = number of complete flocks observed on the transect. #T = total number of transects walked. Hrs = total number of hours of transect surveys. FS = total number of flocking species observed on the transect during that season. F% = percentage of forest in a 1 km buffer of the transect. ED = edge density calculated as length of forest edge in meters divided by the total area within a 1 km buffer.

Summer Winter Site Latitude Longitude Area Transect F% ED CF TF #T Hrs FS CF TF #T Hrs FS 1 4° 46.984' 76° 12.511' 10 La Cancana 10 9 5 7.7 23 16 16 9 7.9 30 19.26 18.73 2 4° 47.866' 76° 11.744' 14 La Gitana 12 11 11 5.6 28 16 14 21 10.7 36 28.93 25.22 3 4° 47.793' 76° 11.286' 28 La Tulia 14 14 15 9.8 22 13 13 18 9.8 31 13.09 20.15 4 4° 45.904' 76° 08.260' 37 Las Brisas 17 15 19 9.1 23 17 15 30 12.7 21 32.39 37.91 4° 45.711' 76° 08.101' El Tigre 7 7 13 7.2 16 3 3 25 10.6 - 26.71 30.88 5 4° 46.857' 76° 13.017' 43 Altamira 18 9 3 2.6 32 26 18 13 11.5 32 23.77 24.09 6 4° 43.700' 76° 14.706' 107 Altomira 16 9 11 6.5 30 17 16 28 13.8 36 39.11 25.30 4° 43.474' 76° 15.087' El Eden 20 20 12 9.6 36 25 25 25 16.2 37 40.01 22.96 7 4° 42.277' 76° 14.630' 147 El Lagito 11 9 17 9.3 34 19 18 24 15.3 39 41.88 23.47 4° 42.400' 76° 14.986' La Sonora 18 15 12 8.7 31 15 14 29 14.4 33 51.47 29.33 8 4° 42.384' 76° 12.930' 173 La Guardia 21 15 10 6.5 34 28 23 24 13.3 38 49.42 26.51 4° 41.964' 76° 13.254' El Rocio 17 13 11 6.7 29 24 24 16 10.5 37 43.19 25.16 RN El 4° 44.752' 76° 17.448' 750 El Brillante 33 24 19 14.1 48 28 26 27 16.4 43 87.50 12.38 Ingles 4° 44.526' 76° 17.655' El Ingles 18 18 19 12.3 43 23 23 26 15.1 39 97.11 7.08 232 188 177 115.7 79 270 248 315 178.2 78

Table 3-2. Linear mixed model estimates of fragmentation predictor variables on network metrics. Response variables represent network metrics calculated on social networks of species co-occurrence patterns in mixed-species flocks. Each network represents a unique combination of transect by season. Flocks were sampled using transect surveys during both the boreal summer (June-August 2017) and boreal winter (January-March 2018). Model parameter estimates come from a full model incorporating five fixed effects of season and fragmentation variables and two random effects of transect and site (fragment). Marginal goodness-of-fit represents the fit of just the fixed effects, while conditional goodness-of-fit represents the global fit of the model, including random effects.

Skewness of degree Mean normalized degree Mean strength Clustering coefficient Modularity distribution Marginal r2 = 0.34 Marginal r2 = 0.45 Marginal r2 = 0.26 Marginal r2 = 0.38 Marginal r2 = 0.16 Variable Conditional r2 = 0.34 Conditional r2 = 0.45 Conditional r2 = 0.26 Conditional r2 = 0.38 Conditional r2 = 0.20 Std. Std. Std. Std. Std. β p β p β p β p β p error error error error error Intercept -0.67 0.18 1.76 1.56 1.65 0.65 0.73 0.06 0.14 0.06 Pct. Forest Cover 0.00 0.00 0.40 0.01 0.01 0.23 0.00 0.00 0.65 0.00 0.00 0.69 0.00 0.00 0.17 Edge Density 0.01 0.00 0.053 0.00 0.03 0.90 0.01 0.01 0.22 0.00 0.00 0.19 0.00 0.00 0.22 Canopy Height 0.01 0.01 0.51 0.21 0.07 0.03 -0.04 0.03 0.21 0.00 0.00 0.40 0.00 0.00 0.50 Large Tree Density 0.00 0.02 0.79 0.20 0.14 0.16 0.07 0.06 0.23 0.00 0.01 0.84 0.01 0.01 0.28 Season -0.11 0.04 0.008 -0.97 0.33 0.006 -0.24 0.14 0.10 -0.05 0.01 < 0.001 0.01 0.01 0.31

Table 3-3. Linear mixed model estimates of the effect of fragmentation on the centrality of nuclear species in mixed-species flock social networks. Response variables represent the betweenness centrality of four a priori-defined nuclear species in 27 social networks constructed from observations of flock composition along 14 transects during two seasons. Flock composition was sampled both during the boreal summer (June-August 2017) as well as during the boreal winter (January- March 2018). Linear model estimates are from the full model, including five fixed effect predictors of fragmentation and season (see Methods) and two random effects of transect and site (fragment). Marginal goodness-of-fit represents the fit of just the fixed effects, while conditional goodness-of-fit represents the global fit of the model, including random effects. C. canigularis T. arthus T. labradorides A. somptuosus M. miniatus P. semibrunnea Marginal r2 = 0.49 Marginal r2 = 0.59 Marginal r2 = 0.39 Marginal r2 = 0.21 Marginal r2 = 0.18 Marginal r2 = 0.35 Variable Conditional r2 = 0.49 Conditional r2 = 0.67 Conditional r2 = 0.39 Conditional r2 = 0.28 Conditional r2 = 0.18 Conditional r2 = 0.35 Std. Std. Std. Std. Std. Std. β p β p β p β p β p β p error error error error error error Intercept 0.108 0.056 0.172 0.050 0.161 0.078 0.090 0.074 0.106 0.075 0.060 0.041 Pct. Forest 0.000 0.000 0.27 -0.001 0.000 0.07 -0.001 0.000 0.03 0.000 0.000 0.27 -0.001 0.000 0.15 -0.001 0.000 0.002 Cover Edge -0.001 0.001 0.15 0.000 0.001 0.66 -0.001 0.001 0.47 -0.002 0.001 0.17 -0.002 0.001 0.26 -0.001 0.001 0.1 Density Canopy -0.002 0.002 0.45 -0.006 0.002 0.01 -0.002 0.003 0.61 0.001 0.003 0.68 0.001 0.003 0.78 0.001 0.002 0.48 Height Large Tree 0.021 0.005 <0.001 -0.016 0.005 0.005 0.000 0.007 0.97 -0.002 0.007 0.83 -0.004 0.007 0.58 0.001 0.004 0.84 Density Season 0.024 0.012 0.053 -0.012 0.009 0.16 -0.051 0.017 0.004 -0.031 0.015 0.04 -0.026 0.016 0.11 0.007 0.009 0.41

Table 3-4. Marginal significance of dbRDA predictor variables on partitioned network dissimilarity. Pairwise dissimilarity of 27 mixed-species flock social networks was partitioned into a species turnover (βST) and interaction change (βOS) components, and we ran a dbRDA on each component that included seven predictor variables representing the patch area, edge effects, and vegetation disturbance components of fragmentation, as well as season. The tables present marginal tests of significance for each predictor variable using an ANOVA-like permutation test with 999 permutations.

βST (66% of dissimilarity)

Variable Df Sum of squares F p Pct. Forest Cover 1 0.478 3.13 0.002 Edge Density 1 0.230 1.51 0.07

Canopy Height 1 0.535 3.50 0.001 Density Large Trees 1 0.281 1.84 0.02 Understory Density 1 0.175 1.15 0.28 Canopy Cover 1 0.170 1.12 0.33

Season 1 0.439 2.87 0.001 Residual Variance 19 2.904

βOS (34% of dissimilarity)

Variable Df Sum of squares F p Pct. Forest Cover 1 0.123 1.20 0.20 Edge Density 1 0.124 1.21 0.18

Canopy Height 1 0.113 1.11 0.30

Density Large Trees 1 0.111 1.09 0.33 Understory Density 1 0.156 1.52 0.02 Canopy Cover 1 0.123 1.20 0.19

Season 1 0.218 2.13 0.001

Residual Variance 19 1.943

Figure 3-1. Social networks of mixed-species flocking interactions across a gradient of canopy height. Social networks were built from flock compositional data collected along three transects in medium-sized (30-50 ha) fragments within the same altitudinal band and landscape. Networks represent the Las Brisas (37 ha; summer), Tulia (28 ha; winter), and Altamira (43 ha; summer) transects; in each case we selected the season with the higher average network strength. Nodes represent flock- joining species, and node width represents the mean weighted degree (strength) of each species. Edges represent co- occurrences of species pairs within a flock and are weighted by strength of association using the simple ratio index.

Figure 3-2. Landscape and local vegetation effects on nuclear species network centrality. Centrality measurements represent the normalized betweenness centralities of each species from 27 social networks built from flock composition data. Density of large- diameter trees is an ordinated measure of the densities of tree size categories, while percentage forest cover represents the percentage of forest in a 1 km buffer around each sampling transect. This variable correlated closely with patch size. Trend lines represent linear model estimates. A) C. canigularis increased in centrality with the density of large-diameter trees, while T. arthus showed the opposite pattern. B) Both T. labradorides and P. semibrunnea showed a higher centrality in smaller forest patches.

Figure 3-3. Ternary plot of pairwise network dissimilarity. Each point represents a pair of mixed- species flock social networks (n = 27 networks, 702 pairwise comparisons). Axes correspond to the relative proportions of network similarity (1 – βWN), dissimilarity due to species turnover (βST), and dissimilarity due to changes to the interactions between co-occurring species pairs (βOS) for each network pair.

CHAPTER 4 SIMULTANEOUS LOSS OF FOREST DEPENDENT SPECIES AND SPATIAL TURNOVER DRIVE CHANGES TO AVIAN RICHNESS AND TURNOVER IN ANDEAN FOREST FRAGMENTS

Forest loss is among the most important drivers of biodiversity loss worldwide (Achard et al. 2014, Watson et al. 2016), particularly in the humid tropics (Barlow et al. 2018). Even small amounts of habitat loss in intact forested landscapes negatively affect vertebrate biodiversity

(Betts et al. 2017), and ongoing forest loss in biodiversity hotspots is a leading driver of vertebrate endangerment (Tracewski et al. 2016, Betts et al. 2017). Deforestation is accompanied by forest fragmentation, where remaining forest is isolated into small patches (Wilson et al.

2016). Forest fragmentation negatively affects biodiversity, with fewer species encountered in smaller and more isolated patches (Haddad et al. 2015, Fletcher et al. 2018). While some biodiversity can persist in disturbed landscapes, a subset of tropical vertebrate communities is exclusively tied to primary forest (Gibson et al. 2011, Alroy 2017).

Understanding the drivers of biodiversity loss with fragmentation requires an understanding of its effects at different scales (Bhakti et al. 2018). At the patch scale, fragment area determines how many species persist (Ferraz et al. 2007, Rueda-Hernandez et al. 2015,

Ulrich et al. 2016). At the within-patch scale, fragmentation increases ease of human access to tropical forests, leading to additional impacts of disturbance, including selective logging and hunting (Barlow et al. 2016, Benitez-Lopez et al. 2017, Alroy 2017). Selective logging of large trees shifts plant community composition and structure towards secondary forest (Aubad et al.

2008) and changes the richness and composition of bird communities (Thiollay 1997, Burivalova et al. 2014, Arcilla et al. 2015). Finally, fragmentation frequently leads to changes to biotic and abiotic conditions near the edge of forest patches (‘edge effects’; Pfeifer et al. 2017), to which

bird species may respond positively or negatively (Manu et al. 2007, Banks-Leite et al. 2010).

Few studies, however, have addressed these scales at the same time.

In addition to the question of scale, it is important to understand how avian biodiversity is lost. First, measuring functional diversity (Petchey and Gaston 2006) allows us to understand if species loss also engenders a loss of critical functional roles, such as pollination and seed dispersal, in fragments. Second, finding nested patterns of species loss, where small habitat patches contain a subset of the species found in large habitat patches (Patterson 1987), suggests community-level change occurs through gradual, non-random loss of species in fragments.

Nested species loss is commonly observed among avifaunas in fragmented landscapes (Wethered and Lawes 2005, Hill et al. 2011, Smith et al. 2018). Alternatively, changes to species composition between assemblages might also reflect species turnover (Baselga 2010). Generalist and early-successional species can colonize small fragments, causing changes to species composition independent of species loss (Carrara et al. 2015, Keinath et al. 2017). It is therefore important to understand the roles of nestedness and species turnover, and the processes driving each one, when studying fragmented landscapes.

One ecosystem particularly at-risk to ongoing forest fragmentation is the montane cloud forest of the northern Andes (Aldrich et al. 1997, Tejedor Garavito et al. 2012, Hermes et al.

2018). These forests represent a global hotspot of endemic, restricted-range, and at-risk bird species richness (Orme et al. 2005, Kier et al. 2009) and are undergoing high levels of deforestation (Tejedor Garavito et al. 2012, Tracewski et al. 2016). In Colombia for example, only ~30% of originally forested lands above 1,500 meters remain (Etter et al. 2006), with conversion to agricultural lands and urban expansion driving deforestation (Armenteras et al.

2007, Armenteras et al. 2011). Andean birds are vulnerable to habitat loss because they occur in

narrow altitudinal bands (Graves 1988), leading to small range sizes and restricting them to specific climatic conditions. Andean birds therefore respond negatively to fragmentation, with a loss of species occurring in cloud forest fragments (Kattan et al. 1994, Renjifo 1999, Aubad et al.

2010). Other studies have detected negative effects of patch isolation (Aubad et al. 2010) and edge effects (Restrepo and Gomez 1998), suggesting a sensitivity to landscape variables.

Despite these concerns, no study to date has surveyed Andean avifaunal diversity across a fragment size gradient while controlling for altitude. We therefore sampled bird diversity along

500-meter transects in primary forest fragments in the Western Andes of Colombia, using a multi-species occupancy model to infer species richness, functional diversity, and community composition at each site. We ask three questions about Andean bird communities: (1) Are patch area, disturbance to vegetation, or edge effects more important in driving alpha diversity patterns in fragmented Andean landscapes? (2) Do species richness and functional diversity patterns differ in their response to fragmentation? And (3), is beta diversity in the Andes driven more by nested species loss in fragments or species turnover to generalist or early-successional species? If

Andean forest-dependent birds show a nested response to fragment size, we would expect to see a gradual loss of species along our gradient. If edge effects or vegetation structure are more important, then species richness should decrease with greater edge density or logging intensity regardless of patch size. Alternatively, if turnover to non-forest and disturbance adapted species drives changes to small fragment bird communities then we would expect little change to overall species richness and significant changes to community composition across the gradient.

Methods

Study Area and Sites

We surveyed bird communities in the Serranía de los Paraguas range of the Western

Andes, a Birdlife International important bird area (Birdlife International, 2019) and center of endemism within Colombia (Ocampo-Peñuela and Pimm 2014). We sampled primary forest fragments on private lands in the municipality of El Cairo, Valle del Cauca department, in collaboration with a local conservation NGO (Serraniagua; www.serraniagua.org). We selected isolated primary and late-successional secondary fragments of subtropical Andean forest on the east slope of the massif within the same altitudinal band (1950-2300 m.a.s.l.) to control for altitudinal effects on the community. This forest type is characterized by humid conditions, numerous epiphytes, and a diverse assemblage of over 200 tree species (Aubad et al. 2008).

Forests in our study landscape were heavily fragmented, with ongoing forest clearing for conversion to grazing lands and smallholder coffee plantations. To sample a full gradient of fragment sizes, we stratified forest fragments into large (≥ 100 ha), medium (30-50 ha), and small (≤ 20 ha) size categories and selected a minimum of two replicates of each (n = 8 fragments, range = 10-173 ha; Table 4-1); all sites chosen were surrounded by cattle pasture. We also sampled in a continuous forest reference site (RCN Cerro El Inglés), a ~750 ha private reserve connected to thousands of hectares of primary forest. Land use histories, particularly disturbance by selective logging (Aubad et al. 2008), differed substantially within and across sites; we therefore created 500-meter sampling transects (n = 14; Figure 2-1) in forest interior, placed on existing trails. We placed two transects in large-sized fragments, with one placed in a more disturbed site and the other in a less-disturbed forest site.

Andean Bird Surveys

We sampled birds along each transect using three distinct methodologies: transect surveys, mist-netting, and nocturnal playback surveys for owls. We chose complimentary methods to characterize the avifauna because neither mist netting nor audio-visual surveys alone adequately survey full tropical bird communities (Whitman et al. 1997, Derlindati and Caziani

2005). We avoided using point-counts because they cannot accurately survey species that participate in mixed-species foraging flocks (Shiu and Lee 2003), which are particularly large and speciose in the Colombian Andes (Arbeláez-Cortés et al. 2011, Chapter 2). We conducted sampling from June-August 2017 (‘summer’) and from January-March 2018 (‘winter’) to account for migratory species that were only seasonally present. We conducted transect surveys during the winter, owl surveys during the summer, and mist netting sampling during both sampling periods. For each survey technique, we conducted 2.5 sequential days of sampling at each site (weather allowing); survey techniques were not used concurrently on the same transect.

For the transect sampling, 2 or 3 observers continuously walked back and forth along the transect from ~0700 to 1700 hours, recording all species detected by either sight or sound. To reduce inter-observer bias, HHJ was present for all transect surveys; he spent 3 months learning local vocalizations at the field sites prior to the surveys. When a mixed-species flock was encountered during transect surveys, we spent up to 45 minutes characterizing all species present, following the flock along the transect where possible. To calculate proportion of sites where a species joins flocks, we characterized each detection of a species for a given survey date as having been in a mixed-species flock at least once (1), or not observed in a flock (0).

For the mist netting sampling, we deployed twelve 12 m by 3 m understory mist nets (38 mm mesh; Avinet Research Supplies, Portland, ME), placed relatively evenly along the transect.

Mist nets were operated from ~0600 to ~1700 hours by two or more trained technicians and were

closed during inclement weather. All birds captured were taken to a central banding station where they were identified to species and fitted with an aluminum leg band. Lastly, we conducted nocturnal playback surveys for seven species of Strigid owl thought to occur at our sites. For each species, the playback sequence consisted of one minute of playback of territorial vocalizations followed by five minutes of silent listening for response to playback. This methodology is typical for owls that respond slowly to playback (e.g. Pardieck et al. 1996, Trejo et al. 2011). Recordings were of territorial calls, and were broadcast using a Bluetooth speaker from the center point of each transect. Because intra-guild predation is common in raptors

(Sergio and Hiraldo 2008), we ordered playbacks by increasing mass to avoid a negative response bias. Because Glaucidium pygmy owls are partially diurnal (König and Weick 2009), we began our surveys a half hour before sunset (~1730 hours). Playback surveys did not take place in high winds or rain.

Quantifying Local Vegetation

To understand the impact of local vegetation, particularly successional stage and disturbance by illegal logging, we quantified the vegetation along each transect. Vegetation measurements were made from June-August 2017, though annual variation appeared minimal.

We used the sampling methodology of James and Shugart (1970), following the modifications made by Stratford and Stouffer (2013), and further modified to be used with belt transects.

Broadly, the methodology comprised two components: (1) the quantification of canopy height, and foliage height diversity of vegetation using point sampling every 10 meters and (2) the quantification of tree size category density using 3-meter-wide belt sampling. Because transects ran along trails, we measured vegetation at least three meters from the trail edge on a randomly selected side of the trail. For the point sampling, we measured six variables at ten-meter intervals, for 50 points per transect. As a measure of foliage height diversity along the transect,

we noted the presence or absence of live vegetation at five heights: <0.5 m, >0.5–3 m, >3–10 m,

>10–20 m, and >20 m. Above 3 meters, we used a rangefinder to determine heights and sighted through a tube with crosshairs while straddling the point. We averaged values for canopy height for each transect and calculated the proportion of points at which vegetation was present for each height category. For the belt sampling, we surveyed all trees (woody vegetation > 2 m in height) on 1.5 meters to either side of the observer and measured their diameter at breast height (dbh).

Trees were later categorized into six dbh size classes for analysis: 1-7 cm, 8-15 cm, 16-23 cm,

24-30 cm, 31-50 cm, and > 50 cm. We also recorded the largest tree dbh recorded on each transect. Because selective logging targets large, old-growth trees, we consider this to be a proxy measure for current and historical logging pressure at each site.

To quantify foliage height diversity, we calculated the Shannon diversity index of the proportion of points with vegetation present in each of the five height bands for each transect.

However, foliage height diversity was highly correlated with canopy height at our sites

(Pearson’s coefficient = 0.90), so we only retained canopy height for analyses. To reduce redundancy and minimize correlation between variables, we used ordinated measured of our tree

DBH data from principal component analysis (PCA: McGarigal et al. 2000), taken from Chapter

2 for each transect.

Quantifying Landscape Composition and Configuration

We used a buffer analysis to quantify the landscape composition and configuration within

1 km of each sampling transect (n = 14 transects); this scale has been shown to affect the occupancy of tropical bird communities (Cerezo et al. 2010, Carrara et al. 2015), and other buffer scales (e.g. 500 meters) were highly correlated with this one. All landscape analyses were conducted in the ArcGIS software (ArcMap 10.3.1; Esri, Redlands, CA) using the ‘isectpolyrst’ tool in Geospatial Modelling Environment (version 0.7.4.0; Beyer 2015). Buffers were centered

on the full length of the transect, resulting in a non-circular buffer shape. Following Carrara et al.

(2015), we used the proportion of forest cover within the buffer region as a proxy for the patch size of the fragment. Because the matrix in our landscape consisted of cattle pasture there was no forest cover in the matrix, so we consider this variable to be a good proxy for patch size

(correlation coefficient = 0.96). We could not use patch size directly because our continuous forest reference sites have no value for the variable. We quantified edge effects using the ‘edge density’ measure from Carrara et al. (2015), which is defined as the density of forest edge habitat within the 1 km buffer, measured in meters per hectare. We measured proportion of forest cover and edge density using a land-cover use map for our study area from the departmental conservation authority (Corporación Autónoma Regional del Valle del Cauca), which we converted to a 25-m cell-size raster.

Calculating Alpha Functional Diversity Metrics

We calculated alpha functional diversity metrics for the bird community at each transect.

Functional diversity metrics consisted of functional dispersion (FDis; Laliberté and Legendre

2010), which measures the mean distance of each species to the centroid of each assemblage in multidimensional trait space, and functional richness (FRic; Villéger et al. 2008), which measures the convex hull volume of an assemblage in this trait space. These metrics are suitable for occupancy data because they do not require species abundances at a site (Weiher 2014). We assembled a trait matrix for all species using data from the Handbook of the Birds of the World

Alive database (del Hoyo et al. 2020). We selected body size, body mass, use of five diet categories (insects, fruits, seeds, nectar, and vertebrates), use of six forest strata (ground, understory, midstory, subcanopy, canopy, aerial), migratory habit (resident, nomadic, altitudinal migrant, boreal migrant), participation in mixed-species foraging flocks, nest type (cup, platform, dome, and cavity), maximum clutch size, and use of three specific Andean microhabitats: ravines

and small streams, patches of Chusquea bamboo, and treefall gaps. For body size, we used the maximum body length given, while for body mass we took the average of the mass of the male, where sexes differed in mass. Diet use categories and forest strata were not mutually exclusive; we scored diet categories from 0 to 3, while use of a forest stratum was presence/absence only.

Measures of migratory habit and nest type were categorical. We did not include data on breeding biology for boreal migrant species. All analyses were performed in R (version 3.5.1; R Core

Team 2020); functional diversity metrics were calculated using the dbFD function of the FD package (Laliberté et al. 2015).

Measures of functional diversity are often correlated with species richness values

(Weiher 2014), so we used standardized effect sizes (SES) to control for species richness when calculating these metrics. To create a null model of functional diversity we used 100 iterations of the tip swapping method (Webb et al. 2008) to randomly shuffle taxa labels across the row names of the functional trait matrix respectively. We then calculated the relevant diversity metrics on each of the null communities and compared it to the observed value. We calculated

SES values for each metric by subtracting the mean of the null values from the observed value and then dividing this by the standard deviation of the null values.

Multi-Species, Muti-survey-type Occupancy Model

We used a novel multi-species, multi-survey-type occupancy model without data augmentation (Iknayan et al. 2014; hereafter MS-MSOM), implemented in a Bayesian framework, to integrate presence-absence data from each of the three avian survey techniques across 14 transects at nine sites (Table 4-1). The full model specification is given in Appendix B; briefly, the model was formulated as a hierarchical state-space model (Royle and Kéry 2007) with a state process model of site-level occupancy and an observation model of repeated detections from each of four survey methods (summer and winter mist netting were modeled as

separate methods). Each species had a unique detectability for each survey method, and species occupancy and detection parameters were assumed to covary. We fit five fixed effect covariates on site-level occupancy: proportion of sites where present that a species joined a mixed-species foraging flock, percentage forest cover and edge density (m/ha) within 1 km of the transect, mean canopy height along the transect, and a PCA axis representing logging intensity. We also included a random effect of site to account for the continuous forest nature of our reference transects. The MG-MSOM model was implemented in Just Another Gibb Sampler (JAGS) using the jagsUI package (Kellner 2019). The goodness-of-fit of the model was assessed using

Bayesian p-values as per Broms et al. (2016).

Analysis of Alpha Diversity

We analysed alpha diversity using a combination of beta estimates included in the MS-

MSOM and post-hoc analyses performed on the median site-by-species occupancy matrix. For each of the four occupancy covariates in the MS-MSOM we calculated the mean and standard deviation of the median beta estimates. We also determined the number of species whose occupancy was significantly positively and negatively affected by each covariate. Overall effect of each predictor variable was inferred from the sign and magnitude of the median beta estimate, while significance was evaluated based on whether the 90% posterior estimates overlapped zero.

We looked at the same effects on two a priori defined community subsets: foraging guilds and categories of forest dependence. We sorted all species in our community into forest dependent (=

“forest specialist”), forest generalist, and non-forest (or forest visitor) species based on the criteria in Bennun et al. (1996) and using the habitat preferences in del Hoyo et al. (2020). Forest dependent species were defined as only occurring in the interior of primary forest, while forest generalists were commonly encountered in disturbed forest habitat such as treefall gaps, forest edge, secondary forest, or agroforests. Foraging guilds were assigned based on the primary diet

item for a species in the functional trait table; if two or more diet items were commonly exploited, the species was classified as an omnivore.

For functional alpha diversity, we calculated standardized effect sizes of our two functional diversity metrics (see above) using the median site-by-species occupancy matrix derived from the MS-MSOM. We then fitted a general linear mixed model (Bolker et al. 2009) with a random effect of site (to control for the non-independence of transects in the same fragment) and four fixed effects corresponding to the covariates included in the MS-MSOM. We ran models using the lme4 package in R (Bates et al. 2015). Lastly, we visualized changes to species composition across sites and differences in species responses to fragmentation using multivariate techniques. First, we used non-metric multidimensional scaling (NMDS; Kruskal

1964) to visualize similarity of composition between transects. The NMDS was run using the metaMDS function of the vegan package (Oksanen et al. 2019) using the Jaccard similarity index to quantify differences between assemblages. We plotted ellipses around assemblies representing continuous forest and large, medium, and small fragments using the ordihull function in the same package. We also used principal components analysis (PCA) to ordinate the four beta estimates of each predictor variable for each species and visualize differences in species responses to fragmentation.

Beta Diversity Partitioning of Taxonomic and Functional Diversity

We calculated overall beta diversity for taxonomic and functional richness using the median site-by-species occupancy matrix. For taxonomic beta diversity, we calculated three values using the Sørensen dissimilarity index: the overall dissimilarity (βSOR), as well as dissimilarity subsets partitioned into species turnover (βSIM) and nestedness components (βSNE;

(Baselga 2010). We used the Baselga family of dissimilarity indices because we were interested in beta diversity driven by nestedness (Legendre 2014). We calculated beta diversity values for

the full community and our foraging guild and forest dependence community subsets. For the functional beta diversity analysis we used the FD measure of functional diversity (Petchey and

Gaston 2002), the sum of the branch lengths of a functional dendrogram. We built a functional dendrogram from our functional trait matrix by creating a Gower dissimilarity matrix (Gower

1971) of all species pairs and then using the UPGMA clustering method. We then partitioned beta diversity according to Leprieur et al.’s (2012) extension of the Baselga model using the

Jaccard dissimilarity index, yielding overall functional dissimilarity (βJAC), a nestedness component (βJNE), and a turnover component (βJTU). We did not use a multidimensional trait space partitioning method (e.g. Villéger et al. 2013) due to the large number of traits and species in our bird community.

Results

Avifaunal Surveys

Across 89 days of field sampling on our 14 transects we detected 178 bird species when pooling across all sample methodologies, encompassing 35 families and 141 genera. The most common families encountered were Tyrannidae, Thraupidae, Trochilidae, and Furnariidae. We captured 125 species in mist nets (98 and 100 species during ‘summer’ and ‘winter’ surveys respectively) over 696.5 hours of netting on 80 sampling days, detected 135 species on transect surveys over 218 hours on 39 sampling days, and detected 4 owl species during our 32 playback owl surveys (Table 4-1). On transect surveys, we detected 78 species (44% of total species) participating in mixed-species flocks at least once and 68% of 1331 species detections (i.e. detection of a species on a transect on one sampling day) were in mixed-species flocks. Naïve species occupancy was variable, ranging from one transect to all 14, but most species were only detected on a fraction of all transects (mean = 5.6, median = 4; naïve species richness estimates in Table 4-1).

MS-MSOM Results

The multi-survey-type, multi-species occupancy model converged to stable posterior distributions both visually and according to the potential scale reduction statistic (Gelman-Rubin

Diagnostic) value of Ȓ = 1. Bayesian p-values were not significant for 132 of 178 species, indicating the model was well specified for these species. The Bayesian p-value had a significant phylogenetic signal (K = 0.279, P = 0.022) as exemplified by the Turdids (Turdus spp.), some

Cardinalids, and some species of Thraupids (Figure B-1). On average, the estimated species richness at a site added 20.6 species to the observed richness value, while occupancy for each species increased by 1.6 ± 2.0 (mean ± standard deviation) sites. As expected, species detectability differed across the four survey types (Figure 4-1, Figure B-3) and species responses to the fragmentation covariates were highly variable (Figure B-2).

Alpha Taxonomic and Functional Diversity

Forest fragmentation negatively correlated with Andean bird community composition and overall species richness at a site (Table 4-2, Figure 4-2a). Occupancy of Andean forest bird species increased with increasing percentage forest cover within 1 km, our proxy for patch size

(mean β estimate = 0.28). We also found an overall trend of decreasing occupancy with increasing edge density (mean β estimate = -0.17) and increasing occupancy with the density of large-diameter trees (mean β estimate = 0.07). Beta estimates for canopy height were near zero.

However, fragmentation effects differed significantly between community subsets. We found that forest dependent species showed a positive response to percentage forest within 1 km (mean

β estimate = 1.47; 9 significantly positive relationships) and increasing large tree density (mean β estimate = 0.35), and a negative response to edge density (mean β estimate = -0.77). Non-forest species, on the other hand, showed the opposite trend of increasing occupancy with declining percentage of forest and increasing edge density. Forest generalist species did not respond

strongly to fragmentation variables and had near-zero β estimates for all four covariates (Table 4-

2). Diet guilds also differed in their response to fragmentation: carnivore and insectivore occupancy increased with greater percentages of forest cover and lower edge densities, whereas granivores, nectarivores, and frugivores showed the opposite pattern.

Species responses to three of four covariates were correlated (Figure 4-3a). Area sensitive species showed negative responses to increasing edge density and increased in occupancy with higher densities of large-diameter trees, while edge-adapted species had higher occupancy at sites with a lower percentage of forest cover. The first PC axis explained 94% of variance and had high loadings for percentage forest cover (0.88) and edge density (-0.44); Eigenvalues and axis loadings are available in Tables B-2 and B-3. Response to canopy height was uncorrelated with the other covariates, with no loading on PCA 1. Species composition changed considerably across the patch size gradient, with the greater similarity in composition occurring between sites with similar percentages of forest (Figure 4-3b). There was no overlap in the 95% ellipses of large (>100 ha) fragments and continuous forest, while small and medium fragments differed from large fragments but showed partial overlap in composition; the NMDS accurately captured community composition in two dimensions (stress = 0.03). Our analysis of functional diversity

(Table 4-3) found no significant effects of fragmentation on the SES of functional richness. In fact, functional richness was highest in large fragments containing a mix of forest dependent and forest generalist species. However, the SES of functional dispersion was negatively affected by increasing edge density (β = -0.08, P = 0.01; Figure 4-3b).

Beta Taxonomic and Functional Diversity

We found high taxonomic beta diversity across our patch size gradient (0.69), of which

81% was explained by species turnover (Table 4-4). The nestedness beta diversity component was slightly greater for forest dependent species (34%), insectivores (27%), and omnivores

(25%), yet in all cases at least two thirds of beta diversity was explained by species turnover.

Frugivores and nectarivores in particular showed the highest proportion of beta diversity due to turnover (~90%). Functional diversity patterns closely matched taxonomic diversity- we found high overall beta diversity (0.72) of which the majority (80%) was explained by functional turnover (Table 4-4).

Discussion

We found that Andean bird communities are sensitive to declining patch size and increasing edge density and, to a lesser extent, selective logging practices. Fragmentation effects on Andean birds took place through two simultaneous mechanisms. First, forest dependent bird species occupancy declined with fragment size and many species were lost from even large

(>100 ha) fragments. These species were simultaneously edge and area sensitive and associated with unlogged primary forest (Figure 4-3a). Secondly, we found evidence of extensive species turnover occurring in association with declining patch size, mediated by edge density and forest successional stage. Community composition in small fragments was dominated by forest generalist and non-forest species (Figure 4-3b) which may be adapted to natural early successional and disturbed forest habitat (e.g. treefall gaps, landslides). Species richness may therefore be an especially poor indicator of habitat quality for forest dependent species in

Andean fragments. Overall functional diversity did not decline with taxonomic species richness, likely due to the extensive functional turnover occurring simultaneously with taxonomic turnover. However, functional dispersion significantly declined with increasing edge density

(Figure 4-2b), suggesting that these assemblages were increasing dominated by a subset of edge- adapted species.

Negative Patch, Edge, and Logging Effects on Forest Specialist Species

Declining patch size and increasing edge density lead to declines of forest-dependent species, but not forest-generalist or non-forest species. Large fragments lost ~30 species of forest dependent birds relative to continuous forest, and others dropped out in fragments smaller than

100 ha. Our results support the conclusion of Gibson et al. (2011) that a portion of tropical forest biodiversity can only be conserved in undisturbed primary forest. Forest dependent species tend to show greater sensitivity to fragmentation (Morante et al. 2015, Khimoun et al. 2016), and previous work in Ecuador also showed that a subset of cloud forest birds had strong preferences for primary forest (Becker et al. 2008). While our study was conducted in a small number of sites, our results echo other work from the Andes showing that a subset of species are sensitive to patch area (Aubad et al. 2010) and edge effects (Restrepo and Gomez 1998), and that specific functional groups are lost from fragments (Kattan et al. 1994). Like Carrara et al. (2015) and

Cerezo et al. (2010), we find that proportion of forest within 1 km was the primary driver of species presence in fragmented landscapes. However, area sensitivity in forest dependent birds was highly correlated with edge sensitivity, providing support for the idea that edge effects may drive negative species responses to declining patch area (Banks-Leite et al. 2010). Concerningly, we identified many highly area sensitive species currently listed by the IUCN as Least Concern

(e.g. Premnornis guttuliger, Myiophobus flavicans, Pharomachrus auriceps). Accounting for remaining habitat and minimum habitat area requirements in determining conservation statuses

(e.g. Tracewski et al. 2016) should therefore be a priority for Andean forest birds.

Forest dependent species also showed sensitivity to within-fragment disturbance by selective logging, a common practice in the Colombian Andes (Aubad et al. 2008). These species showed higher occupancy on transects with greater densities of large-diameter trees, which decline with ongoing or historical selective logging. Andean bird communities in Bolivia also

showed compositional changes to small scale vegetation disturbance (Montano-Centellas and

Garitano-Zavala 2015), while bird species richness in the Colombian Andes declined with increasing ease of human access (Aubad et al. 2010), a proxy for disturbance. Generally, our results agree with a large body of literature from across the tropics suggesting that selective logging practices negatively impact forest specialist bird species and change forest bird community composition (Politi et al. 2012, Arcilla et al. 2015, Burivalova et al. 2015).

Vegetation structure is simplified in selectively logged forests, with changes to canopy cover, tree density, and understory vegetation density (Thiollay 1997, Sekerçioglü 2002) and a loss of late successional tree species (Aubad et al. 2008). Such changes may lead to the loss of specialized foraging microhabitats (Stratford and Stouffer 2015) as well as the loss of specialized breeding substrates, such large tree cavities (Politi et al. 2010).

Functional and Taxonomic Turnover Across a Fragment Size Gradient

Andean birds have narrow elevational distributions (Graves 1988), leading to high beta diversity and species turnover along elevational gradients (Melo et al. 2009, Jankowski et al.

2013). We find equally high species turnover, ~80% of all taxonomic dissimilarity, along a fragment size gradient; our findings mirror the extensive species turnover documented in Andean fragments and secondary forest (Renjifo 1999, O'Dea and Whittaker 2007). One driver of beta diversity may be changes to vegetation structure and plant community composition along elevational and moisture gradients (Jankowski et al. 2013) combined with the high habitat specialization of tropical montane birds (Jankowski et al. 2009). The same process may occur along successional gradients, with forest fragmentation and associated disturbances mimicking the structure and plant composition of naturally-occurring early successional plant communities

(Tabarelli et al. 2008), such as those caused by landslides (Ohl and Bussmann 2004). Andean bird community composition is certainly tied to vegetation structure (Jankowski et al. 2013) and

forest succession (O'Dea and Whittaker 2007, Rosselli et al. 2017), and we found the highest spatial turnover in nectarivores and frugivores, species most strongly tied to plant composition.

This matches Jankowski et al.’s (2013) finding of high turnover of these species with tree community composition.

Functional diversity was also affected by fragmentation: functional richness did not decline with taxonomic richness, but the variance of functional traits in an assemblage, functional dispersion, decreased with increasing edge density. We found extensive functional turnover, which explained 80% of functional dissimilarity. Functional turnover to early- successional species within the same functional groups present in primary forest therefore appears to delay loss of functional diversity with minor disturbance, though dramatic losses in functional diversity occur in more disturbed land-use types (Flynn et al. 2009, Sekercioglu

2012). However, a loss in the variation of functional traits with increasing edge density suggests a habitat filtering mechanism causing functional trait convergence (Ding et al. 2013). Edge habitat may favor a functional shift towards early successional and pioneer species (Tabarelli et al. 2008), where functionally redundant edge-adapted species are added to the community.

Villéger et al. (2010) found similar increases in functionally redundant species in a disturbed fish community. Functional trait filtering may be particularly troubling if a reduction in functional traits limits the ability of the remaining bird community to effectively perform critical ecosystem functions, such as seed dispersal (Bovo et al. 2018).

Conservation Implications: Large Reserves Matter

Significant effects of patch size and edge effects drive taxonomic richness and community composition of Andean birds. Thirty species of forest dependent birds were lost from continuous forest to 100-150 ha fragments, with further declines in smaller fragments. Managers should therefore aim to conserve intact montane forest landscapes where they still exist. Area

sensitive species were also sensitive to edge effects and associated with primary forest.

Therefore, landscape configuration and core area of existing reserves are important considerations in maintaining forest specialist species, as edge-dominated landscapes are increasingly dominated by edge-adapted early successional and non-forest species. Furthermore, selective logging practices, even within large forest tracts, further degrade habitat for forest dependent species, eventually leading to similar turnover in community composition. Efforts should be made to minimize selective logging in primary forest patches that still contain many primary-forest-adapted species. Compositional change occurred through extensive species turnover rather than nested species loss, suggesting that there may be key thresholds of species loss rather than a gradual species loss. Overall species richness may be an especially poor metric of forest dependent species richness due to the high species turnover. Ultimately, only large areas of primary forest with minimal edge effects can conserve the most area sensitive Andean species.

Table 4-1. Study sites and survey results in from forest fragments in El Cairo municipality. Sites correspond to a gradient of forest fragment sizes sampled in the same landscape in the Western Andes of Colombia, ordered from smallest to largest by patch size. Transects correspond to 500-meter sampling transects within the forest; coordinates were taken at the approximate center point of each transect. Species totals correspond to the total number of species detected using a sample method across all sites. NSR = species richness detected at a site across all survey types. ESR = species richness estimated for the site using the MG-MSOM. Transect Mist Net Owl Survey Survey Site Latitude Longitude Area Transect NSR ESR Δ SR Hrs. Spp. Hrs. Spp. Sur. Spp. 1 4° 46.984' 76° 12.511' 10 La Cancana 45.0 44 16.0 44 2 1 66 91 25 2 4° 47.866' 76° 11.744' 14 La Gitana 44.0 36 13.0 48 2 1 69 88 19 3 4° 47.793' 76° 11.286' 28 La Tulia 45.0 34 11.5 37 2 1 60 82 22 4 4° 45.904' 76° 08.260' 37 Las Brisas 54.5 41 15.0 31 3 1 58 72 14 4° 45.711' 76° 08.101' El Tigre 53.0 37 12.5 26 2 1 49 70 21 5 4° 46.857' 76° 13.017' 43 Altamira 49.0 39 19.0 55 2 2 78 104 26 6 4° 43.700' 76° 14.706' 107 Altomira 52.0 28 15.0 51 2 2 68 77 9 4° 43.474' 76° 15.087' El Eden 49.0 31 18.0 52 3 3 67 81 14 7 4° 42.277' 76° 14.630' 147 El Lagito 51.0 28 10.0 54 2 2 71 97 26 4° 42.400' 76° 14.986' La Sonora 50.0 41 15.5 58 3 4 80 102 22 8 4° 42.384' 76° 12.930' 173 La Guardia 50.0 30 21.0 53 2 2 71 80 9 4° 41.964' 76° 13.254' El Rocio 47.0 16 14.5 59 2 2 66 75 9 RNC El 4° 44.752' 76° 17.448' ~750 El Brillante 53.0 44 19.0 72 3 2 93 133 40 Ingles 4° 44.526' 76° 17.655' El Ingles 54.0 48 18.0 74 2 2 100 133 33 696.5 124 218.0 135 32 4 178

100

Table 4-2. Mean beta estimates of fragmentation covariates from the MG-MSOM. Means and standard deviations are of the median beta estimate for each species in the community subset. Pluses and minuses indicate the number of species with significantly positive or negative effects on occupancy for each covariate. Forest Cover Edge Density Canopy Height Large Tree Density Community subset Mean SD - 0 + Mean SD - 0 + Mean SD - 0 + Mean SD - 0 + Full community 0.28 2.16 16 151 11 -0.17 1.16 2 163 13 0.01 0.31 1 177 0 0.07 0.5 1 176 1 Forest Dependence Forest specialists 1.47 1.59 1 54 9 -0.77 0.80 0 63 1 0.01 0.30 1 63 0 0.35 0.37 0 63 1 Forest generalists -0.08 2.11 10 79 2 0.02 1.20 2 81 8 0.03 0.33 0 91 0 -0.03 0.49 0 91 0 Forest visitors -1.61 1.92 5 18 0 0.73 1.06 0 19 4 -0.08 0.28 0 23 0 -0.29 0.50 1 22 0 Foraging Guild Insectivores 0.69 1.89 2 75 4 -0.43 1.05 1 60 3 0.06 0.30 0 64 0 0.15 0.50 0 63 1 Frugivores -0.22 2.45 3 17 4 0.17 1.14 0 24 0 -0.02 0.28 0 24 0 -0.05 0.47 0 24 0 Nectarivores -0.38 2.49 6 15 2 0.31 1.44 0 16 7 -0.24 0.38 1 22 0 0.05 0.57 0 23 0 Granivore -1.42 2.99 1 2 0 0.45 1.36 0 2 1 -0.11 0.40 0 3 0 -0.05 0.42 0 3 0 Omnivores -0.05 2.12 4 33 1 -0.07 1.14 1 35 2 0.04 0.25 0 38 0 -0.02 0.52 1 37 0 Carnivores 1.55 1.73 0 9 0 -0.64 0.88 0 9 0 0.14 0.24 0 9 0 0.13 0.41 0 9 0

101

Table 4-3. Linear mixed model estimates of fragmentation effects on alpha functional diversity. The response variable was the standardized effect size (SES) of two functional diversity metrics. Marginal r-squared corresponds to the goodness-of-fit of just the fixed effects, while conditional r-squared corresponds to the overall goodness-of-fit of the model. SES Functional Richness (FRic) Marginal r2 = 0.27, Conditional r2 = 0.27 Variable β Estimate Std. Error t value p Intercept -0.50 0.91 -0.55 Pct. forest cover 0.01 0.01 1.03 0.31 Edge density 0.03 0.02 1.82 0.10 Canopy height 0.05 0.04 1.43 0.22 Tree size 0.03 0.09 0.33 0.74 SES Functional Dispersion (FDis) Marginal r2 = 0.60, Conditional r2 = 0.60 Variable β Estimate Std. Error t value p Intercept 1.67 1.15 1.45 Pct. forest cover -0.01 0.01 -1.22 0.23 Edge density -0.08 0.02 -3.46 0.01 Canopy height -0.04 0.05 -0.93 0.36 Tree size 0.19 0.11 1.70 0.13

102

Table 4-4. Beta diversity partitioning of taxonomic and functional diversity. Sample sizes refer to the number of species included in each community subset. Overall beta diversity is calculated across all sites (n = 14 transects). Partitioned diversity represents the proportion of beta diversity attributable to nested species loss (βSNE) and species turnover (βSIM) across the fragment size gradient. βSIM Prop. βSNE Prop. Grouping N βSOR (turnover) βSIM (nestedness) βSNE Full community 178 0.687 0.559 0.81 0.128 0.19 Forest Dependence Forest specialist 64 0.796 0.522 0.66 0.274 0.34 Forest generalist 91 0.596 0.508 0.85 0.087 0.15 Forest visitor 23 0.768 0.619 0.81 0.149 0.19 Diet Guilds Insectivores 81 0.691 0.506 0.73 0.185 0.27 Frugivores 24 0.642 0.576 0.90 0.066 0.10 Nectarivores 23 0.757 0.658 0.87 0.099 0.13 Carnivores 9 0.801 0.355 0.44 0.447 0.56 Omnivores 38 0.642 0.469 0.76 0.146 0.24 βJTU Prop. βJNE Prop. N βJAC (turnover) βJTU (nestedness) βJNE Functional Diversity Full community 178 0.722 0.576 0.80 0.146 0.20

103

Inner Ring Ring Parameter y 1.0 s Inner s u i

E r c E a u a s 0.8 l nd n h p E u u f a c q t h i u MNS r P o g a

u 2 H i o m p e h h m t a 0.6 n n i e n I C h i T c e M c e p n n o a i y i o e rd T m u a t u s a u r d T c o

i e

y n M g c T n u a s q t u n g a C e u r s MNW C i s h x r a s t 0.4 n u o d i y 3 o e u r y u e i i s r a l t s H d a a b u a c n y r d u r r u u a a s p r d l t s i i a a i h u u r p r n e r n g n i e d u g h t l e s e e s r r o u a c s u i r h i a n

b th y s t n c o t n u c u c n d s s n a p e i s s n 0.2 h r a P s p t i o s n c t s s i h q i a e i s v c i e e p u c n OS a u c o a o a e f s e p r i i f r c s e i 4 i r r a i i h g u r t d o s s c t c o o c r s l l g r s a r y a t e w e p g n v h i r n s o t o a i a u r o h n r e h r a s o l g s s o 0.0 m u a y u r y m a a c i v a i n l a o s r l g i u p i a t u m s u a i s a d c c n a a o e h p V n i c t r V s y l b o h a a l y s n a C t t u l r l o h o r q t c o l y s l u o i t e i m p c i o a s e e Outer T h l i O i i r l n t a i y v r s e h r p t r d p s h m p l o u t t s e t i s l r n o i r p a s c c i e i n a d o u i h r i s A i s r t o u a i h A e t a a h o v s C l C l o a c C t o n u t n a e c e n s e e o i t a h r M e t o l l p a a h y r s l r V e i i l c s r n t a C e f g s g a y s i L b i i e z o a h e l r S e y t r e n i a o r l s S s o e h b e y n i m u i c i i c c p o p s c S s v n e s S a d u i r li a o c e u M y r u a r r f s a il n n ib o i a n t a t a v c f t s i B a e i g s c s M c i o o ig r p r u s M a a is i a n i c l u h o h b n g r e rt c e y n u y ra r n s C a a s n i c i n l c ir n r P r n o p e t o a t a x o n y s H i a t o a v s s C p m ri s r a h u le s E o h x p y s e P i lt n p s r u ra a y m n o o d u n tr a P a e o p a r e e r to b m i l c is h r o z e t n h M p p u a c a S p o g i a la a p M y id u s e d h s l r p c i y ho a us yi ia o s f iv e r S h u s a lb c rr im al od M rc n f la e m a C re lo s a ia y ss h y yi h a u v n a p s a n p i p na ar u x m ic tr C a lo s o ia tid ce us P s c s vi ig a is H g s h n ni ro s h te hu tu re a n i lo op o a y uo yll s s b s tu s D ig r h s ph pt P om ch c er ce s D lo op ry or m la h y ry ep cu n h or ch p so a yll ia so h li s C hl o is s ph om s n c al fe C or rn u rii ce yi ig ep ot r hl o ath ffe no Zim as ro ha es C os n rie ya m c ca lu rid og is s c Py er ine pi s I is rn pi rrh ius re llu An ro au ola a om E c ice s lo thr ve ur yia lae hry p Ch ra fla bsc N s c ni so s po ba a o eph inn a fr ps S re piz P elo am an oe os eus laty my om tzii C em ac ollis rinc ias eu As oliv nic Le hus pul s ris cya ptop albo che Tia ara olina ogo gul r ang vitri Mio n ruf aris T gara ei Pog necte ipec Tan hein onotr s str tus gara ervix iccus o iatico Tan rufic Pog phtha llis ngara viridis onotric lmicus Ta a nigro cus poe angar orides Pseudo cilotis T a labrad triccus pe Tangar Hemit lzelni ra gyrola riccus granad Tanga ensis ra aurulenta Lophotriccus pile Tanga atus ara xanthocephala Pachyramphus versicolor Tang Pachyramphus polychopterus Nothocercus bonapartei Ampelioides tschudii C rii hamaepetes gou ipreola rieffe Odo dotii P hus ntophorus cryptolop Piay hyperythru owornis nus a cayana s Sn peruvia Patag upicola pilla Z ioenas R flavica entry albilin opipo terus C gon ea Chlor rysop s hae frenat s ch stictu Eu tura a Masiu holo ps toxe ciner tes tice Pha res eive adec irga ilis P eth aqu ntris hrip tes v nob ha orn ila T dec s ig aris D eth is g ripa cte bal ory orn uy Th ade su llis Co fer is s hrip yla tico s lib a lu yrm T act tria en Sc ri d do ato nd s esc us C his el vic ph Sy thia nn lat r al te ph ae oru er ru tel ge C lip s g ina s ac b s li ii hlo hl eo e ab lex nis ttu au Ch r ox ffr An op or gu e e A a os m oy n ar is nn ra m ly tilb itc i rem rg rn o za us A a bu o he P a o iss a til s A m z ra n lli M n o is ru lu d a ilia b m em b x il r A e zi u el Pr es lla ps s ge g lo li fr ff an t a o pu i is A la m a an on o ap n n m r s C g i s c ii rh ol y e s y la u la o yi au ia y c S X hu cr u n a C o i ce a c e nc do p a g n o e o r m e h u l n m i s H e li ce c e ro us e m s ia u n u O a g r u l t s ha te tr ic n n a p lig e c s a te P r p s p ra a in s U c l e n u k no i lo a y c d u s H r r o n a s in g y l u s t i r t i o e p c c g e p co h te x u c s a H e c a h a o o i n c la e t e n t A e l h t a i m o n p c c is M io u to e e y a id y la m a ip e r r c l r s e li le s C h in z f l n t o i R c io d o d rq g p r o s a ti i z B o a e s e c c e u s s u i d o u ia u t L o o u r o n e i G r p x n n is h o r r a m o i l s G u p p i o b a a r u s i v r i r M t a d a t ip d d r m t i ic p a l e t o h e x o a u a v t o s i C l e e u X n n e i a l v o a l C a r n a i u l a r f l t p r r P e o r r m r r e e u a n T i a n g e h l z o s T a w c a e M i c u s r s n o M g f s H D D g a e e l P h c n p u l c a i r p i u o i C c i e s r t S k u r c c u i i s a i e c t s o a s r r d r u s a c o c r b e e r o o o o r i i a a i t a l a a i e nd l u i i p n t i c s d a h a r r a a g u g a m u m u a i d r l c a c e g r m e m i p n a l b r u i t n l o u u v a t v i o c i t u t i y a b l n s a l c i p t i i o e y o l c o u a u o t i a a a t a 2 - Outer Ring s c o s l i p m m r a r o s p c a t r o g o n r a s b p s i i F t t i r n m g s i u a u s l l d x o g p u o o o t a t p r i G a y s n n s h u n a e s l u s e l d v t i n e c a a c s l i p r c p p l l G u c c n i s u u o i b r i a s o r a d h g a h m a n e r s i u m t l i t c r S p n s a s h u o i i b s g i p i i l p u t l c a l a o s l 5.0 r l o o h m u b b b s f r a h a a a G n a l m o r a s e t h g u l c o r e e u l h m o i i r t p r r p a a c o t c c n s i r s i r s t c m c q i a u a c a o s o a s s m o a s h i s r a n y r u i c a t m h n l u s s u i i a l y u y c n a m e c b a m u s a y i a s s e u d n s n s 0.0 S D y h i a t t i y h o h b t r t p t u a a t r o e a u o r n d c i p a i T u n o o s t D r c D r h n u o V t r e P a c u

E l e e o i m c c e s o a n a p s r n o C n a a n l l s l y i -5.0 i s q y s C s i u m o s M h c r e A

u o S

c e

L -10.0 a l

u A -15.0

Figure 4-1. Phylogeny of Andean bird species included in the MS-MSOM. The innermost ring indicates the occupancy of a given species (휓푖) in probability space. The second through the outermost ring indicate the detection of a given species on each survey gear (휃푖,푔) in logit space. The second ring corresponds to summer mist net captures, the third ring winter mist net captures, the fourth ring owl surveys, and the fifth, and outermost, ring the transect surveys. For both scales, warmer values indicate higher values, of probability in the inner ring and of log-odds in the second through outermost ring.

104

Figure 4-2. Effect of forest fragmentation on taxonomic and functional diversity of Andean birds. A) Taxonomic diversity of Andean birds declined with percentage forest cover within 1 km, a proxy for patch area. Forest specialist species were strongly affected by fragmentation, while forest generalists were less sensitive. B) The variance of functional traits in an assemblage was negatively affected by increasing edge density.

105

Figure 4-3. Changes to Andean bird community composition in response to forest fragmentation. A) Species responses to fragmentation variables were correlated: area sensitive species showed negative edge effects and a higher occupancy in unlogged forest. Ellipses represent the 95% confidence ellipses for each forest dependence category. B) Species composition changed dramatically between continuous forest and fragments, with no overlap in the compositional ellipses. Even large (> 100 ha) fragments lost many forest species in comparison to continuous forest sites.

106

CHAPTER 5 CONCLUSIONS AND CONSERVATION IMPLICATIONS

Our work found negative impacts of forest fragmentation on Andean birds at both the behavioral scale (mixed-species flocks) and the community level. However, effects were largely turnover driven, with some species benefitting from fragmentation and others being lost from forest fragments. Generally, mixed-species flocking behavior was very resilient to fragmentation due to a couple of mechanisms. First, extensive turnover in flock participants from forest interior

Furnariids and Tyrannids to edge-adapted Thraupid tanagers and boreal migrant species allowed for flocking behavior to continue even in the face of an overall loss of flocking species. This was largely due to the ‘open membership’ nature of the flocking system, which contained a mix of canopy and understory species, a lack of flock sub-types, and numerous, weak interactions.

Indeed, the strength of species co-occurrences in flocks was not affected by patch size or edge effects but did decline in secondary forest relative to primary forest. Second, we found both resilience and redundancy of the nuclear species leading flocking behavior. The ‘primary’ leader species, Chlorospingus canigularis, persisted into small fragments and secondary forest, likely facilitating flocking behavior. However, numerous intraspecifically social species were able to play the nuclear role, including numerous species of Tangara tanagers. The importance of nuclear species changed along the fragmentation gradient and in response to vegetation structure, with Tangara spp. replacing C. canigularis in importance as patch size declined. However, persistence of flocking behavior did not mean persistence of many flocking species into small fragments: there was extensive turnover in flock composition, with many species showing sensitivity to forest edge and vegetation structure.

Community-level responses to fragmentation mirrored those of the flocking community.

We documented high taxonomic and functional beta diversity along the fragment size gradient,

107

of which a large majority was explained by species turnover rather than nested species loss. This effect was driven by the loss of forest dependent species in smaller fragments and sites with a greater density of forest edge. Thirty forest dependent species recorded in continuous forest were never observed in any size fragment, and many more species were lost in fragments smaller than

100 ha. By contrast, we found few effects of fragmentation on many ‘forest generalist’ species that may be pre-adapted to forest disturbance and early successional habitat. Forest dependent species were increasingly replaced by such species as well as non-forest species from the matrix in smaller fragments. Fragment bird communities showed a change in the frequency of diet guild members as well, with a reduction in the number of insectivorous species and an increase in the number of granivores and nectarivores. The variation in functional traits in an assemblage also showed a negative response to the increasing densities of forest edge, with an apparent filtering of functional traits towards more edge-adapted species. Such changes in the functional composition of the community could have important carry-over effects on critical ecosystem functions in the Andes, particularly since Andean birds appear to be tightly associated with specific plant community compositions.

In Chapter 2, we found that while Andean mixed-species flocks persist in small 10-ha fragments, conserving the majority of flocking species biodiversity requires extensive tracts of tall-canopy forest. We suggest that managers prioritize tracts of mature primary forest with >18- meter canopies that contain canopy emergent trees as well as foliage in the sub-canopy, midstory, and understory, and vertical vegetation complexity was closely tied to the encounter rate, species richness, and size of sub-tropical Andean mixed-species flocks (Figure 2-2).

Unlogged, old-growth forests with large-diameter (e.g. 30-50 DBH) trees contained a subset of flocking bird diversity lost from disturbed sites; selective logging within forest fragments may

108

therefore be particularly detrimental to some flocking species. These old-growth trees contain numerous epiphytic plants, lianas, and hanging dead plant matter, which may provide foraging habitat for specialized forest-interior insectivores. Managers should also aim to conserve large areas of forest which minimize the edge-to-interior ratio of the patch. Furnariid species in flocks were edge-sensitive, while some flock-following tyrannids showed area-sensitivity. Even large

(170 ha) fragments had reduced flocking species richness relative to continuous forest, and some flocking species (e.g. Chlorochrysa nitidissima) were lost from fragments smaller than ~130 ha.

In Chapter 3, we found no effect of patch size or patch edges on the coherence of flocking behavior, though the frequency of species co-occurrences was significantly reduced in secondary forest with canopies under 17 m in height, regardless of patch size. Canopy height was highly correlated with vertical structural complexity of the vegetation, so within-patch disturbance in Andean fragments, including selective logging practices, will negatively affect the coherence of flocking behavior. Changes to flocking interactions occur mainly through extensive species turnover in flock participants rather than changes to co-occurrence patterns between species pairs that are maintained across the gradient. This result reiterates that persistence of flocking behavior in fragments does not necessarily mean persistence of forest-dependent flocking species (see above). The resiliency of flocking behavior in the Western Andes is in part due to the ‘open membership’ nature of flocks, which consist of numerous weak interactions and a lack of flock sub-types. It is also due to the existence of many redundant nuclear species which replace each other along the fragmentation gradient. C. canigularis is the most important nuclear species in primary old-growth forest, while T. arthus, T. labradorides, and P. semibrunnea play a more important nuclear role in smaller fragments and areas of secondary growth. Flock

109

conservation strategies focusing on the conservation of keystone nuclear species are therefore unlikely to be effective in the Andes.

In Chapter 4, we document significant effects of patch size and edge effects on taxonomic richness and community composition. Thirty species of forest dependent birds were lost from continuous forest to 100-150 ha fragments, with further declines in smaller fragments.

Managers should therefore aim to conserve intact montane forest landscapes where they still exist. Area sensitive species were also sensitive to edge effects and associated with primary forest. Therefore, landscape configuration and core area of existing reserves are important considerations in maintaining forest specialist species, as edge-dominated landscapes are increasingly dominated by edge-adapted early successional and non-forest species. Furthermore, selective logging practices, even within large forest tracts, further degrade habitat for forest dependent species, eventually leading to similar turnover in community composition. Efforts should be made to minimize selective logging in primary forest patches that still contain many primary-forest-adapted species. Compositional change occurred through extensive species turnover rather than nested species loss, suggesting that there may be key thresholds of species loss rather than a gradual species loss. Overall species richness may be an especially poor metric of forest dependent species richness due to the high species turnover. Ultimately, only large areas of primary forest with minimal edge effects can conserve the most area-sensitive Andean species.

110

APPENDIX A ADDITIONAL RESULTS, TABLES, AND FIGURES FOR CHAPTER 2

Principal Component Analysis of Vegetation Data. We retained only the first axis of seven from the PCA of understory vegetation density. This axis has an Eigenvalue of 2.88 and explains 42% of the overall variance (Table A-1). This axis shows high negative loadings for the densities of shrubs, ferns, tree ferns, vines, and small trees (1-7 DBH; Table A-2, Figure A-1a), which we interpret as a measure of understory vegetation density. We also only retained the first

PCA axis from the ordination of tree DBH size classes. It has an Eigenvalue of 2.24 and explains

38% of the overall variance in vegetation measurements (Table A-3). This axis shows high positive loadings for the densities of trees with a DBH of 24-30, 31-50, and larger than 50, as well as the measure of the dbh of the largest tree (Table A-4, Figure A-1b). As such, we interpret this axis as a measure of the presence and number of large trees on the transect segment. Given that selective logging is prevalent at our study sites, this is likely a measure of the degree of selective logging intensity along the transect.

111

Table A-1. Eigenvalues and proportion of variance explained of the principal component analysis of understory vegetation density. Prop. Var. indicates the proportion of total variance explained by each axis, while Cum. Var. indicates the cumulative proportion of variance explained. The first (named) axis was retained for further analyses. PC Axis Eigenvalue Prop. Var. Cum. Var. Understory Vegetation 2.88 0.418 0.418 Density 2 1.29 0.187 0.605 3 0.89 0.128 0.734 4 0.69 0.100 0.834 5 0.52 0.075 0.909 6 0.38 0.056 0.964 7 0.25 0.036 1.000

Table A-2. Loadings of the understory vegetation principal component axes. Rows represent seven vegetation variables measured for each 100-meter transect segment (n = 70 segments, 14 transects) in the field. Columns represent the principal component axes; only the first principal component axis was retained for further analyses. We considered loadings of greater than 0.3, or less than -0.3, to be significant for the purposes of axis interpretation. Variable PCA 1 PCA 2 PCA 3 PCA 4 PCA 5 PCA 6 PCA 7 Percentage Ground Cover 0.061 0.688 0.517 0.161 0.407 0.226 0.117 Shrub Density -0.475 0.060 0.049 0.399 0.216 -0.640 -0.390 Fern Density -0.421 0.150 0.262 0.283 -0.776 0.201 0.099 Vine Density -0.489 0.036 -0.164 -0.298 0.184 0.561 -0.544 Tree Fern Density -0.374 0.382 -0.152 -0.648 -0.053 -0.352 0.381 Palm Density -0.127 -0.495 0.769 -0.366 0.076 -0.087 -0.014 Small Tree (1-7 DBH) Density -0.446 -0.330 -0.142 0.304 0.379 0.229 0.619

Table A-3. Eigenvalues and proportion of variance explained of the principal component analysis of tree dbh category density. Prop. Var. indicates the proportion of total variance explained by each axis, while Cum. Var. indicates the cumulative proportion of variance explained. The first (named) axis was retained for further analyses.

PC Axis Eigenvalue Prop. Var. Cum. Var. Tree Size 2.24 0.325 0.325 2 1.79 0.259 0.584 3 0.97 0.141 0.725 4 0.80 0.115 0.841 5 0.47 0.067 0.908 6 0.37 0.054 0.962 7 0.26 0.038 1.000

112

Table A-4. Loadings of the tree dbh class principal component axes. Rows represent seven vegetation variables measured for each 100-meter transect segment (n = 70 segments, 14 transects) in the field. Columns represent the principal component axes; only the first principal component axis was retained for further analyses. We considered loadings of greater than 0.3, or less than -0.3, to be significant for the purposes of axis interpretation. Variable PCA 1 PCA 2 PCA 3 PCA 4 PCA 5 PCA 6 PCA 7 Tree Density (1-7 DBH) 0.063 0.423 0.764 0.080 0.037 0.462 0.107 Tree Density (8-15 DBH) 0.052 0.608 0.091 -0.446 0.131 -0.597 -0.215 Tree Density (16-23 DBH) 0.171 0.487 -0.562 -0.249 -0.300 0.500 0.124 Tree Density (24-30 DBH) 0.395 0.270 -0.254 0.586 0.601 -0.024 -0.032 Tree Density (31-50 DBH) 0.516 0.047 0.118 0.377 -0.692 -0.307 -0.046 Tree Density (> 50 DBH) 0.528 -0.233 0.082 -0.374 0.189 -0.106 0.688 Largest Tree DBH 0.513 -0.292 0.077 -0.328 0.120 0.272 -0.671

Table A-5. Eigenvalues and proportions of variance explained for the six CCA constrained axes. The proportion of total variance corresponds to the contribution of the axis to the mean squared contingency coefficient, while the proportion of constrained variance describes the proportion of constrained inertia explained. Biplot scores for each component are provided in Table A-6. Variable CCA 1 CCA 2 CCA 3 CCA 4 CCA 5 CCA 6 Season -0.061 0.983 -0.009 0.151 -0.069 0.047 Distance to edge 0.862 0.085 0.485 -0.021 -0.089 0.081 Understory density 0.186 0.182 0.130 -0.572 0.066 -0.763 Foliage height diversity -0.072 0.130 -0.363 -0.572 0.681 0.235 Large-diameter trees 0.529 -0.033 0.124 0.551 0.622 0.116 Percentage forest 1 km 0.862 -0.014 -0.467 -0.080 -0.175 0.033

Table A-6. Bi-plot scores for constraining variables used in the CCA. Variables listed represent the six constraining variables used in the CCA. Variable descriptions and sources are provided in Table A-8. Eigenvalues and proportion of variance explained for each CCA axis is given in Table A-5. Proportion Cumulative Proportion Cumulative Component Eigenvalue Constrained Constrained Total Var. Total Var. Var. Var. CCA 1 0.344 0.072 0.072 0.582 0.582 CCA 2 0.104 0.022 0.093 0.176 0.758 CCA 3 0.059 0.012 0.107 0.099 0.857 CCA 4 0.045 0.009 0.116 0.075 0.932 CCA 5 0.026 0.006 0.121 0.044 0.977 CCA 6 0.014 0.003 0.124 0.023 1.000

113

Table A-7. Species scores for the first four CCA constrained axes. Species are flocking bird species observed in at least 3% of all flocks (n = 54). Species CCA1 CCA2 CCA3 CCA4 Eubucco bourcierii -0.378 -0.307 0.222 -0.090 Veniliornis dignus 0.484 0.077 0.176 0.256 Picoides fumigatus -0.362 -0.532 0.427 0.300 Colaptes rubiginosus -0.273 -0.157 -0.112 -0.258 Piaya cayana -0.441 0.027 0.110 -0.641 Premnornis guttuligera 1.452 -0.399 -0.245 0.021 Margarornis stellatus 2.864 0.263 1.273 -0.223 Xenops rutilans -0.134 0.202 -0.238 -0.089 Syndactyla subalaris -0.079 -0.107 0.277 0.038 Anabacerthia striaticollis -0.125 0.120 0.034 -0.034 Xiphorhynchus triangularis 0.474 0.026 0.098 -0.141 Lepidocolaptes lacrymiger 0.062 -0.131 -0.164 -0.167 Phyllomyias nigrocapillus 0.793 0.124 -0.377 0.493 Phyllomyias cinereiceps 1.407 0.221 0.336 -0.031 Phylloscartes poecilotis 0.139 0.058 -0.168 -0.158 Phylloscartes opthalmicus -0.209 0.062 -0.144 -0.353 Leptopogon rufipectus 1.956 -0.031 -0.146 0.125 Mionectes striaticollis 1.187 -0.518 -0.103 0.158 Myiophobus flavicans 0.284 -0.494 -0.620 0.110 Nephelomyias pulcher 1.700 -0.320 -0.621 0.068 Pachyramphus versicolor 2.163 -0.020 0.104 0.015 Pachyramphus polychopterus -0.484 -0.151 0.169 -0.165 Vireo leucophrys 0.046 0.145 -0.152 -0.141 Pachysylvia semibrunneus -0.552 0.121 0.312 -0.199 Sphenopsis frontalis 1.187 -0.185 -0.441 -0.188 Creurgops verticalis 0.565 0.104 -0.466 0.135 Anisognathus somptuosus 0.025 -0.143 -0.014 0.016 Iridosornis porphyrocephalus 2.891 0.334 1.250 -0.234 Chlorochrysa nitidissima 0.553 -0.045 -0.650 -0.438 Tangara ruficervix -0.148 -0.145 -0.342 -0.012 Tangara heinei -0.336 -0.176 -0.054 -0.102 Tangara nigroviridis 1.213 -0.644 -0.754 0.217 Tangara labradorides -0.254 -0.339 0.160 0.200 Tangara gyrola -0.793 -0.441 0.450 0.387 Tangara xanthocephala 0.173 -0.220 -0.051 0.182 Tangara arthus -0.380 -0.095 0.062 0.053

114

Table A-7. Continued

Species CCA1 CCA2 CCA3 CCA4 Tangara vitriolina -0.812 -0.702 0.330 -0.178 Chlorophanes spiza -0.541 -0.644 0.008 -0.241 Diglossa caerulescens 1.840 0.071 -0.245 -0.108 Saltator atripennis -0.619 0.021 0.223 -0.114 Chlorospingus canigularis -0.243 0.122 0.017 0.245 Chlorospingus semifuscus 2.845 0.292 1.174 -0.241 Piranga flava -0.731 -0.272 0.258 0.426 Setophaga pitiayumi -0.291 0.304 0.014 -0.524 Myiothlypis coronatus 0.375 0.009 0.057 0.113 Basileuterus tristriatus 0.899 -0.063 0.064 0.088 Myioborus miniatus -0.127 -0.185 -0.088 0.085 Euphonia xanthogaster -0.312 -0.249 0.081 0.504 Chlorophonia cyanea -0.291 -0.395 -0.059 -0.529 Mniotilta varia -0.418 0.901 0.076 -0.004 Setophaga fusca -0.032 0.843 -0.138 0.281 Cardellina canadensis -0.374 0.881 -0.016 -0.008 Vireo flavifrons -0.546 0.879 0.180 -0.139 Piranga rubra -0.563 0.932 -0.071 -0.256

115

Table A-8. Predictor variables used for generalized linear mixed model analysis of flock size, species richness, and diversity measures. Definition indicates how variable was defined by the authors, while source indicates where the predictor data comes from and how it was measured, where applicable. Variable type indicates the type of effect or mechanism the variable is intended to quantify. Predictor Definition Source Type Variable Shortest distance from midpoint of Distance to Calculated in Garmin 100-meter transect segment to forest Edge effects Edge Basecamp edge Shannon diversity index value for Calculated from Foliage height Vegetation proportion of vegetation present in vegetation data measured diversity structure five height bands along the transect A PCA axis, from A multivariate measure of density of Understory multivariate density data Vegetation understory shrubs, ferns, vines, and density collected along the structure tree ferns transect A PCA axis, from Large- A multivariate measure of density of multivariate density data Vegetation diameter tree large (>24 cm DBH) trees, sorted by collected along the structure density DBH class transect Averaged measure of canopy cover Calculated using a Vegetation Canopy cover across 10 points on a transect densiometer along the structure segment transect Percentage Percentage of forest cover within a 1 Calculated in ArcGIS Patch size forest 1 km km buffer of the transect using buffer analysis effects The length of all forest edges (in Edge density 1 meters) divided by the area of the Calculated in ArcGIS Edge effects km landscape (in ha) in a 1 km buffer using buffer analysis around the transect Indicates whether Chlorospingus Nuclear Chlorospingus Recorded in the field as canigularis was present (1) or absent species presence part of flock composition (0) in the flock effects Indicates whether flock was sampled Coded as 0 (June- Seasonal Season during boreal summer (June-August) August) or 1 (January- effects or winter (January-March) March) Total number of individuals observed in the flock, only included in models Recorded in the field as Flock Size - for species richness of taxonomic part of flock composition groups and migrants

116

Table A-9. Model selection table for eight predictor variables on flock encounter rate. Models each contained a single fixed effect with two random effects for transect ID and site ID. Marginal and conditional r-squared values quantify the goodness-of-fit of the fixed effects and the model as a whole respectively. Conditional Marginal log Variable df AIC delta w r2 r2 Likelihood c i Foliage height diversity 0.30 0.30 5 -19.11 50.938 0.00 0.76 Season 0.08 0.41 5 -21.21 55.1388 4.20 0.09 Understory density 0.11 0.25 5 -21.69 56.1105 5.17 0.06 Large-diameter trees 0.01 0.30 5 -22.79 58.3123 7.37 0.02 Canopy cover 0.01 0.28 5 -22.83 58.3939 7.46 0.02 Distance to edge 0.01 0.28 5 -22.85 58.4212 7.48 0.02 Edge density 1 km 0.00 0.26 5 -22.92 58.5628 7.62 0.02 Pct. Forest 1 km 0.00 0.27 5 -22.92 58.5741 7.64 0.02

117

Table A-10. Bird species observed in mixed-species flocks during boreal summer (June-August) and boreal winter (January-March) sampling. The number of June-August (‘summer’) and January-March (‘winter’) observations in flocks, the proportion of total flocks in which a species was observed in each sampling period, and the proportion change in attendance between sampling periods are reported for all flocking species (n = 232 ‘summer’ flocks, 270 ‘winter’ flocks, 98 flocking species). Migratory species (n = 11) are denoted with an asterisk. Summer Prop. Winter Prop. Prop. Species Flocks Summer Flocks Winter Change Trogonidae Trogon collaris 5 0.02 0 0.00 -1.000 Trogon personatus 7 0.03 1 0.00 -0.877 Capitonidae Eubucco bourcierii 100 0.44 67 0.25 -0.430 Picidae Picumnus.granadensis 1 0.00 0 0.00 -1.000 Veniliornis dignus 13 0.06 23 0.09 0.521 Picoides fumigatus 18 0.08 8 0.03 -0.620 Colaptes rubiginosus 38 0.16 37 0.14 -0.164 Cuculidae Piaya cayana 10 0.04 12 0.04 0.034 Thamnophilidae Dysithamnus mentalis 1 0.00 0 0.00 -1.000 Myrmotherula schisticolor 0 0.00 2 0.01 Furnariidae Premnornis guttuligera 20 0.09 8 0.03 -0.655 Premnoplex brunnescens 4 0.02 1 0.00 -0.782 Margarornis stellatus 5 0.02 9 0.03 0.515 Pseudocolaptes boissonneautii 1 0.00 1 0.00 -0.074 Xenops rutilans 22 0.10 33 0.12 0.287 Syndactyla subalaris 47 0.20 40 0.15 -0.270 Anabacerthia striaticollis 125 0.54 163 0.60 0.120 Philydor rufum 1 0.00 0 0.00 -1.000 Thripadectes holostictus 1 0.00 1 0.00 -0.074 Thripadectes virgaticeps 2 0.01 0 0.00 -1.000 Dendrocolaptes picumnus 8 0.03 1 0.00 -0.891 Xiphorhynchus triangularis 34 0.15 45 0.17 0.134 Lepidocolaptes lacrymiger 111 0.48 96 0.36 -0.256 Dendrocincla tyrannina 0 0.00 1 0.00 Campylorhamphus pusillus 0 0.00 1 0.00 Tyrannidae Phyllomyias nigrocapillus 5 0.02 10 0.04 0.684

118

Table A-10. Continued

Summer Prop. Winter Prop. Prop. Species Flocks Summer Flocks Winter Change Phyllomyias cinereiceps 6 0.03 8 0.03 0.140 Mecocerculus poecilocercus 1 0.00 0 0.00 -1.000 Phylloscartes poecilotis 108 0.47 138 0.51 0.097 Phylloscartes opthalmicus 52 0.22 66 0.24 0.091 Leptopogon rufipectus 19 0.08 19 0.07 -0.142 Mionectes striaticollis 29 0.13 15 0.06 -0.556 Myiophobus flavicans 9 0.04 4 0.01 -0.620 Nephelomyias pulcher 12 0.05 6 0.02 -0.573 Myiophobus fasciatus 1 0.00 1 0.00 -0.074 Pyrrhomyias cinnamomeus 10 0.04 4 0.01 -0.655 Contopus fumigatus 2 0.01 0 0.00 -1.000 Myiarchus tuberculifer 3 0.01 1 0.00 -0.715 Myiarchus cephalotes 2 0.01 1 0.00 -0.588 Cotingidae Pipreola riefferi 4 0.02 0 0.00 -1.000 Ampelioides tschudii 1 0.00 2 0.01 0.852 Snowornis cryptolophus 2 0.01 0 0.00 -1.000 Tityridae Pachyramphus versicolor 11 0.05 8 0.03 -0.370 Pachyramphus polychopterus 53 0.23 45 0.17 -0.269 Vireonidae Cyclarhis nigrirostris 8 0.03 6 0.02 -0.346 Vireo flavifrons* 0 0.00 32 0.12 Vireo olivaceus* 0 0.00 2 0.01 Vireo leucophrys 57 0.25 83 0.31 0.250 Pachysylvia semibrunneus 76 0.33 87 0.32 -0.018 Corvidae Cyanocorax yncas 1 0.00 0 0.00 -1.000 Troglodytidae Henicorhina leucophrys 1 0.00 0 0.00 -1.000 Cinnycerthia olivacea 0 0.00 3 0.01 Thraupidae Sphenopsis frontalis 12 0.05 7 0.03 -0.501 Creurgops verticalis 19 0.08 30 0.11 0.355 Anisognathus notabilis 1 0.00 0 0.00 -1.000 Anisognathus somptuosus 78 0.34 75 0.28 -0.173 Iridosornis porphyrocephalus 6 0.03 10 0.04 0.425

119

Table A-10. Continued

Summer Prop. Winter Prop. Prop. Species Flocks Summer Flocks Winter Change Chlorochrysa nitidissima 24 0.10 23 0.09 -0.173 Sporothraupis cyanocephala 2 0.01 4 0.01 0.646 Chlorornis riefferii 0 0.00 1 0.00 Tangara ruficervix 18 0.08 19 0.07 -0.098 Tangara nigroviridis 36 0.16 9 0.03 -0.785 Tangara labradorides 117 0.50 87 0.32 -0.361 Tangara gyrola 30 0.13 21 0.08 -0.397 Tangara xanthocephala 45 0.19 43 0.16 -0.179 Tangara arthus 119 0.51 121 0.45 -0.126 Tangara vitriolina 20 0.09 7 0.03 -0.699 Chlorophanes spiza 25 0.11 9 0.03 -0.691 Diglossa caerulescens 7 0.03 8 0.03 -0.012 Diglossa cyanea 2 0.01 0 0.00 -1.000 Diglossa albilatera 1 0.00 0 0.00 -1.000 Tiaris olivaceus 3 0.01 0 0.00 -1.000 Coereba flaveola 2 0.01 0 0.00 -1.000 Saltator atripennis 10 0.04 16 0.06 0.378 Passerellidae Arremonops bruneinucha 1 0.00 0 0.00 -1.000 Atlapetes albinucha 7 0.03 2 0.01 -0.753 Chlorospingus canigularis 93 0.40 149 0.55 0.376 Chlorospingus semifuscus 9 0.04 13 0.05 0.235 Cardinallidae Habia cristata 1 0.00 1 0.00 -0.074 Piranga flava 8 0.03 9 0.03 -0.020 Piranga rubra* 0 0.00 14 0.05 Pheucticus ludovicianus* 0 0.00 3 0.01 Parulidae Vermivora chrysoptera* 0 0.00 2 0.01 Mniotilta varia* 0 0.00 45 0.17 Leiothlypis peregrina* 0 0.00 2 0.01 Setophaga cerulea* 0 0.00 9 0.03 Setophaga fusca* 0 0.00 205 0.76 Setophaga castanea* 0 0.00 1 0.00 Setophaga pitiayumi 40 0.17 68 0.25 0.464 Myiothlypis coronatus 34 0.15 38 0.14 -0.043 Basileuterus tristriatus 46 0.20 45 0.17 -0.158

120

Table A-10. Continued

Summer Prop. Winter Prop. Prop. Species Flocks Summer Flocks Winter Change Myioborus miniatus 152 0.66 120 0.44 -0.321 Cardellina canadensis* 0 0.00 68 0.25 Icteridae Amblycercus holostictus 1 0.00 0 0.00 -1.000 Fringillidae Euphonia laniirostris 4 0.02 0 0.00 -1.000 Euphonia xanthogaster 51 0.22 50 0.19 -0.158 Chlorophonia phyrrophrys 1 0.00 4 0.01 2.704 Chlorophonia cyanea 14 0.06 9 0.03 -0.444

Figure A-1. Biplot of first two principal component axes of the ordinations of the understory vegetation densities and the tree diameters. Percentages in axis labels indicate percentage of total variance explained by each axis. Site scores are represented on plots as asterisks. Diameter at breast height values are in centimeters. A) First and second PCA axes of the vegetation density PCA. We interpret PCA 1 as a measure of vegetation density, high negative values are associated with greater densities of understory plants and saplings. B) First and second PCA axes of the tree diameter data plotted against each other. PCA 1 is interpreted as a measure of logging intensity, with higher values associated with greater densities of large-diameter trees.

121

APPENDIX B ADDITIONAL METHODS, TABLES, AND FIGURES FOR CHAPTER 4

MS-MSOM Specification. A multi-species occupancy model was implemented to estimate the true occupancy of a species, i, at a given site, k. The model was formulated as a state-space model (Royle and Kéry 2007) with a state process model for the site-level occupancy, z, and an observation model for repeated detections, y, during multi-gear surveys.

The occupancy was described by a Bernoulli state process model (Equation B-1) while the detections were described by a binomial observation model (Equation B-2).

푧푖,푘~Bernoulli(휓푖,푘) (Eq. B-1)

푦푖,푘,푔~Binomial(푧푖,푘훿푖,푔, 퐸푘,푔) (Eq. B-2)

where 휓푖푘 was the occupancy probability of species i at site k, 훿푖,푔 was the detection probability of species i on survey gear g, and 퐸푘,푔 was the effort at site k on gear g. For summer and winter mist nets and transects, 퐸푘,푔 was in units of half hours while for owl surveys 퐸푘,푔 was in the units of surveys. The occupancy and detection parameters for all species were assumed to covary, such that every species received correlated draws of a set of occupancy (푢) and detection parameters (푣1…퐺) in logit space (Equation B-3), from multivariate normal hyperprior with mean

{푢̅, 푣1̅ , … , 푣̅퐺} = 0 and covariance 훴 (Equation B-4).

푢푖 푢̅ 푣푖,1 푣̅ { } ~푀푉푁 ({ 1 } , 훴 ) (Eq. B-3) ⋮ ⋮ 푢푣 푣푖,퐺 푣̅퐺

122

2 𝜎푢 … 𝜎푢𝜎푣퐺𝜌푢,푣퐺

훴푢푣 = [ ⋮ ⋱ 𝜎푣퐺−1𝜎푣퐺 𝜌퐺푣퐺−1,푣퐺 ] (Eq. B-4) 2 𝜎푣퐺 𝜎푢𝜌푢,푣퐺 𝜎푣퐺𝜎푣퐺−1𝜌퐺푣퐺−1,푣퐺 𝜎푣퐺

2 where 𝜎푋 was the variance of a given parameter X, 𝜌푋1,푋2 was the correlation between

parameters X1 and X2, and 𝜎푋1𝜎푋2𝜌푋1,푋2 was the covariance.

The occupancy probability, 휓푖,푘, can then be specified as a linear model with intercept 푢푖 and a slope from additive covariate effects (Equation B-5).

logit(휓푖,푘) = 푢푖 + 훾푋푖,푚푥 + {훼푋1,푖 … 훼푋푛,푖} ⊙ {푋푘,1 … 푋푘,푛} + 훼푘 (Eq. B-5)

where 훾 was the slope for the mixed-species flocking propensity (푋푖,푚푥); Eq. B-8),

훼푋1…푛,푖 were the slopes for the environmental covariates for species i: percentage forest cover within 1 km (푋푘,1), edge density within 1 km (푋푘,2), canopy height (푋푘,3), and large-diameter tree density (푋푘,4), and 훼푘 was the random effect for site 푘. The effects for environmental covariates were assumed to come from a multivariate normal hyperdistribution with mean

{훼̅푋1 … 훼̅푋푛} such that a new set was drawn for each species, i.

훼 푋1,푖 훼̅푋1 { ⋮ } ~푀푉푁 ({ ⋮ } , 훴푋) (Eq. B-6) 훼 푋푛,푖 훼̅푋푛

123

2 𝜎훼1 … 𝜎훼1𝜎푎푛𝜌훼1,훼푛 훴푋 = [ ⋮ ⋱ ⋮ ] (Eq. B-7) 2 𝜎훼푛𝜎훼1𝜌훼1,훼푛 … 𝜎훼푛

2 where 𝜎푋 was the variance of a given parameter X, 𝜌푋1,푋2 was the correlation between

parameters X1 and X2, and 𝜎푋1𝜎푋2𝜌푋1,푋2 was the covariance. The mixed-species flocking propensity covariate was calculated as:

퐾 ∑푘=1 훬푖,푘 푋푖,푚푥 = (Eq. B-8) 푛푘

where 퐾 was the number of total sites, 훬푖,푘 was variable describing whether species i at site k was ever detected in a mixed species flock (during transect flock surveys) such that 푋푖,푚푥 was the proportion of sites where a species was detected in a mixed species flock. The detection probability, 휃푖,푔, was assumed to follow a linear model (Equation B-9).

logit(훿푖,푔) = 푣푖,푔 + 훽푔 + 훽푚푥,푔푋푖,푚푥 (Eq. B-9)

In Equation B-9, 푣푖,푔 was the effect of the occupancy and detection covariance, 훽푔 was the intercept for a particular survey gear, 훽푚푥,푔 was the slope effect of the mixed flock proportion (푋푖,푚푥). Priors for each parameter are specified in Table B-1. All priors were made under a hyper-Jaynesian approach, or the use of weakly informative priors, such that each prior provides minor structural information as well as weakly regularizes the inference (Gelman et al.

2017). The end result was to bound the posterior distribution to estimable space, weakly inform the more likely values of the parameter’s posterior estimate and match the data- generating process.

124

The model was implemented in Just Another Gibb Sampler (JAGS) using the jagsUI package (Kellner 2019) with four chains with 25,000 iterations for the adaptation phase, 50,000 for the burn-in phase, and 125,000 for the sampling phase with a thinning rate of 100 resulting in

1,250 samples per chain and 5,000 samples in total. Each chain was initialized randomly with diffuse initial parameter values drawn from constrained priors. Convergence was assessed using the potential scale reduction statistic (Gelman-Rubin Diagnostic; Gelman and Rubin 1992), 푅̂ <

1.1 as suggested by Gelman et al. (2013). Posterior samples were used to generate 90% credible intervals (훼 = 0.1) for all effects. The significance of a given effect was determined by the 90% credible interval not overlapping zero in logit space (Kéry 2010). Additionally, the probability of the parameter (푝(훩)) was calculated by taking the average of parameter values that were less

훼 훼 than zero (Eq. B-8; Kruschke 2013). Parameters were significant if < 푝(Θ) or 푝(훩) > 1 − . 2 2

∑푁 훩<0 푝(훩) = 푖=1 (Eq. B-10) 푁

(푠) (푠) For each species i, site k, and gear g, occurrences, 푧푖̃ ,푘 , and detections, 푦̃푖,푘,푔, were predicted for each sample, s (Equations B-11 and B-12).

(푠) (푠) 푧푖̃ ,푘 ~퐵푒푟푛표푢푙푙푖(휓푖,푘 ) (Eq. B-11)

(푠) (푠) (푠) 푦̃푖,푘,푔~퐵푖푛표푚푖푎푙(푦̃푖,푘 훿푖,푔 , 퐸푘,푔) (Eq. B-12)

The goodness-of-fit of the model was assessed using Bayesian p-value per Broms et al.

(2016). An integrated likelihood, [푦푖푘푔|휓푖푘, 훿푖푘푔], (Equation B-13) was used to determine the

125

deviance of the observations, 퐷(푠), (Equation B-14) as well as the deviance of the detections predicted by the model, 퐷̃(푠) (Equation B-15).

푦 (퐸 −푦 ) ( 푖푘푔 푘푔 푖푘푔 ) [푦푖,푘,푔|휓푖,푘, 훿푖,푔] = 퐼푦푖푘푔>0 휓푖푘훿푖푘푔 (1 − 훿푖푘푔) + (Eq. B-13) 퐸 (( ) 푘푔 ) ((1 − 퐼푦푖푘푔>0) 1 − 휓푖푘 + 휓푖푘훿푖푘푔 )

퐾 퐺 (푠) (푠) (푠) (푠) 퐷푖 = −2 ∑ ∑ log ([푦푖,푘,푔|휓푖,푘 , 훿푖,푔 ]) (Eq. B-14) 푘=1 푔=1

퐾 퐺 ̃(푠) (푠) (푠) (푠) 퐷푖 = −2 ∑ ∑ log ([푦̃푖,푘,푔|휓푖,푘 , 훿푖,푔 ]) (Eq. B-15) 푘=1 푔=1

where 퐼푦푖푘푔>0 was an identity array for detections. The Bayesian p-value was calculated

(푠) as the proportion of samples where deviance of the observed detections, 퐷푖 , was greater than

̃(푠) the deviance of the predicted detections, 퐷푖 . For a model that fits well, the p-value should be

(푠) ̃(푠) close to random (e.g. 퐷푖 is randomly greater than and less than 퐷푖 ) and a poor fitting model

훼 훼 would have p-values less than or greater than 1 − . 2 2

126

Table B-1. Prior distributions for all parameters used in the MSOM. Priors were parameterized with respect to precision rather than variance. Parameter Prior 1 훾 훾~푁 (0, ) 22 훼푘 1 훽 훽 ~푁 (0, ) 1…퐺+1 푔 22 1 훽푚푥,1…퐺+1 훽 ~푁 (0, ) 푚푥,푔 22 1 … 0 −1 훴푢푣 훴푢푣 ~Wishart (퐺 + 1, [⋮ 1 ⋮]) 0 … 1 1 … 0 −1 훴푋 훴푋 ~Wishart (4, [⋮ 1 ⋮]) 0 … 1

Table B-2. Eigenvalues and proportion of variance explained in beta estimate PCA. The ordination included beta estimates from all detected species (n = 178) from four covariates included in the MG-MSOM.

PCA 1 PCA 2 PCA 3 PCA 4

Eigenvalue 5.93 0.29 0.09 0.01 Prop. Variance 0.939 0.046 0.014 0.001 Explained

Cumulative Variance 0.939 0.985 0.999 1.000 Explained

Table B-3. Predictor variable loadings on the PCA axes. Predictor variables represent four covariates included on the occupancy component in the MG-MSOM. Variable PCA 1 PCA 2 PCA 3 PCA 4 Pct. Forest Cover 0.878 0.444 0.163

Edge Density -0.443 0.770 0.453 Canopy Height -0.402 0.707 0.581 Large Tree Density 0.180 -0.221 -0.684 0.672

127

1.0

s u i

E r c E a 0.8 u a s l

n h p E u u f a t c h i u r o g P a H u i o m p e 0.6 h m h a n n t n i e I C h i T c p e M c e p n n o a i i e o T a m u t s u a u r d T c o i y e M c T n n u a s g t u n C e u r s g a C i s h x 0.4 a r s t o d n u i y e o u r y e i i s r a u l t s d a a b u c n y H r d u r r a u u a a s p r d l

t s a a i h i i u u r p r e r n g i n e d u h n e e s l g r e t r s o u u a c s i r h i a n

y s t b n c o t n s u c u c n d s n a p e i s s n P s h p r a i 0.2 n t o s c t s s i e i a s h i e v c i e p u n a c c o a o a e f u e p s r i i c r s e i f r i i r h g u r a d i s s t c o t c o o c r l g r s l r s w y a a t e e p g n v h i s o a r n r t o a i u o h n r e a r s o l g s s o h m u a u r y y a a c m i v a i n 0.0 l a o s r l g i u p

a i s u a i c s t a u m d c a n o a V i c t e h p r V s n y l b o h a a y s a l C n t t u l r l o h o r c o l t i t u y s l c o e i m p i o a e e O l s i h i i r l n t a i y v r s e h t r d p l s h r p m p o u t t s e i s l t i r n o c c i e r i p h a s i s i d o u s r n a r A t u i a i h A e o a a h o v s C t l C l o a c C t o n u t n a e c e n s e e o i p t a h r M e t o l l a a h y r s l r V e i l c s r n t a C h e f g s g a y s i L i b i i e z o a e l r S e y t r e n i u a c o l s S s o e h b e y n is m in r i i c c p o p s c S c v e e s S a d u i r li a a o i a u M y r u a r r f s il n c n b o if n s t a t a s M c v o i p r t i B a e i g a a s s ic n i ic o g ru o u s M h b n i e t a y n u ly r r h s C g r s r ic e i n c a i n r n * P a a n o p n e t c r * * r o lt x n y s * i t a s s C p a r a o ** H a a o h v u e i s r l o h m x p y s e * P i lt n p s r u a a ys E n o o n tr a P e p a r r e r m t b m d u c i * a h ro o z e te n h M p o u ia lc a s S o g i a a a p M y id p s e d h s p l r p c il y o a s yi ia o us f iv e r S h u s lb c rh im lu od M r n la e m a * C e lo sa a ia yr s ha y y ch a fu v n * r p s a n p is p na ia u x m ic tr a * C a lo s o a tid ce s P s rc s v i a is H g s h ni ni ro su h te hu tu ire ga n i lo p o y uo yl s s b s t s * D ig ro h sa ph pt lom ch c e c us * D lo op ry r m a Ph y ry ep rc en * h r h po o al yl ia so h ul s C lo oc is s s ph lom s c al ife h or rn u rii ce y nig ep ot r C hl o th ffe no Zi ias ro h es C os na ie ya mm c ca alu * id og s r c P e ine p s Ir is ni pis yrrh riu re illu * n ror au ola a om E s c ic s A lo thr ve ur yia lae hr ep * Ch ra fla bsc N s c ni yso s po ba a o eph inn a fr ps S re piz P elo am ant oe os eus latyr myi om zii C em vac ollis inch as p eu As oli anic Le us ulc s aris cy a ptop albo her * Ti ara iolin ogo gula ang vitr Mio n ruf ris * T gara ei Pog necte ipec Tan hein onotri s str tus gara ervix ccus o iatico Tan rufic Pog phtha llis * ngara viridis onotric lmicus * Ta a nigro cus poe angar orides Pseudo cilotis T a labrad triccus pe Tangar Hemit lzelni * a gyrola riccus granad Tangar ensis ra aurulenta Lophotriccus pile Tanga atus ara xanthocephala Pachyramphus versicolor Tang us polychopterus Pachyramph Nothocercus es tschudii bonapartei Ampelioid Chamaepe la riefferii * tes goudotii Pipreo Odontoph tolophus P orus hyper ornis cryp s iaya cay ythrus Snow ruvianu Pa ana icola pe illa tagioen Rup avicap Zent as albi pipo fl erus * * rygon linea hloro sopt s Cha frena C chry tictu E etura ta asius olos s utox cine M tes h ticep * * Ph eres reive dec irga ilis aet aqu ntris hripa s v ob Pha horn ila T ecte ign ris D eth is g ipad ctes ala or orn uy Thr de sub llis C yfe is s ripa la ico s olib ra l yrm Th cty riat en * Sc ri ud at nda st sc s * C hi de ovi op Sy hia ne atu a ste lph cae hor ert run ell er * C llip s g ina us ac b st lig ii hl hl eo e ab lex nis ttu au Ch or ox ff An op or gu e e A a os m roy mn ar is nn ra m lyb tilb itc i re rg rn so za us A a u o he P a o is a til s * A m z ra n lli M n o is ru lu d a ilia b m em b x il r * A e zi u el Pr es lla ps s ge g lo lia fra ffo an t a o pu i is * A la m n n o ap n n m r s C g i s c ii rh ol y e s y la u la o yi au ia y c S X hu cr u n a C o i ce a c e nc do p a g n o e o r m e h u l n m i s H e li ce c e ro us e m s ia u n u O a g r u l t s ha te tr ic n n a p lig e c s a te P r p s p ra a in s * U c l e n u k no i lo a y c d u s H r r o n a s in g y l u s t i r t i o e p c c g e p co ch te x u c s a * * H e c a h a o o i n la e t e n t A e l h t a i m o n p c c is M io u to e e y a id y la m a ip le r r c l r s e r li le s C p h o in z f in t o i * R c io d o d q g s r c s u a t s i z * B o i o a u i e e c e s u d a u t L o o u r o o n e i G r p x n n is h o r r a m i l s G u p p i o b a a r u s i v r p i r * M t a d a t ip d d t i ic a l * e t o h e x o a r m u a t o s i * C l e e u a v o * C n n i l a l * r a i u X v a n e a f l t r r * P e r r m r r l r n p * o e e u a * T i a e n g h l z o s T u a w c a e M i c s r s n o s M g f H D D g a e e P h c n p u l l i i c a r p r u o i C c i e s t S k u r c c u i i s a i e c t s o a s r r d r u s a c o c r b e e r o o o o i r i a a i t a l a a e l u i i i p n t i c s d a h a r r a a g u g a m u m u a i d r l c a c e g r m e m i p n a l b u i t r n l o u u v a t v i o c i t u t i y a b l n s a l c i p t i i o e y o l c o u a u o t i a a t a s c a o s l i p m m r a r o s p c a t r o g o n r a s t b m p s i i F t i r n g s i u a u s l l p d x o g u o o o t a t p i r G a y s n n s h u n a e s l u s e l d v t i n e c a a c s l i p r c p p l l G u c c n i s u u o i b r i a s o r a d h g a h m r a n s e i u m t l i t c r S p n s a s h u o i i b s g i p i i l p u t l c a l a o s l r l s o o h m u b b b f r a h a a a G n a e l m o r a s

h g u l t c o r e e u l h m r o i i r t p r p a a c o t c c n s i r s i r a s t c m c q i u a c a o s o a s s m o a s h i s r a n y u r c i t l a m h n u s s a l u i i y u y c n a m e c b a m u s a y i a s s e u d n s n s S D y h i a t t i y h o h b t r t p t u a a t r o e a u o r n d c i p a i T u n o o s t D r c D r h n u o V t r e P a c u

E l e e o i m c c e s o a n a p s r n o C n a a n l s l y l i y i s s C s i u m o s M h c r e A

u o S

c e

a L l

Figure B-1. Phylogeny of Andean bird species included in the MG-MSOM. The Bayesian p- value is shown in the ring with warmer colors indicating higher p-values. Species with significant values (i.e. for which model fit was poor) are denoted by the white asterisk.

128

1st - Outer Ring Ring Parameter 6.0 aFC1 s Inner s u i

E r 4.0 c E a u a s l nd n h p E u u a f a c ED t 2.0 h i u r o g P a 2 H u i o m p e h h m a n t a n i e n I env C h i T c

0.0 c e M e p n n o a i i o e rd T m u a t u s a u r d T c o

i e a y c n n a s CH M T u g t u n C e u r s g a C i s x -2.0 r h a s t d n u o i y 3 e o u r y

u e i i s r a l t s c y H d a a b u a n r d u r r u u a a s p l r d a a h t s i i u i u r r p n g n e r e d h n i u l g e t s e e s r r -4.0 o u a c s u i r h i a n y s t b n c o t n s u c u c n d s n a p e i s s n P s h p r t a i n a o s c t s s i a h TS e i s i e v c u i e p n Outer a c c o a o a e u f e p s r i c -6.0 i e f r i s i r r a i h s g u i r t d c o t s c o o r l r c s l g s a w r y a t e e p g n v h i s o o a u r n r t a i o h n r e a h m u r s o l g s s o u a r y y m a a c i v a i n l a o s r l g i u p a i s u a i s t a u m d c c a n a o c e h p V n i t r V s y l b o h a a l y s n a C t t u l r l o h o r t c o l y s l u o i t e i m p c i o a s e e O l i h i i r l n t a i y v r s e h p t r d p s h r m p l o u t t s e l t i s n o r c c i e i d o u i r i p h r a s i s i n a s r A t o u i a h A e a a h o v s C t l C l o a c C t o n u t n a e c e n s e e o i p t a h r M e t o l l a a h y r s l r V e i i l c s r n t a C h e f g s g a y s i L b i i e z o a e l r S e y t r e n i u a o r l s S s o e h b e y n i m in c i i c c p o p s c S s c v e e s S a d u i r li a i a o i a u M y r u a r tr f s l n c n b o if n s t B a a s v i p r t i a e i g a a s s c M c ic o o g r o u s M h b n i e ti a n y in ly r r u h s C g r s r ic e n u a i n n P a a n o n e t c io lt c x r ry s r t p a s s C p a r a o n H i a a o h v u e o h m ix y s e s P ir t n s r l a s E n o o p n t a l e p p ru r ea ry m t b m d u c ri P a h ro o za e te n h M p o u ia lc a s S o g i a a a p M y i p s e d h s p l r p c il y o a s y i do u iv e r S h u s lb c rh m lu io M ar n s fla e m a e lo a a a r si a dy y c a fu v n C r p ss i py is ph n ia hu x m i t a C a o a n id e s as rc s v ca ri l ss ho ia it oc u Ph te h t ir ig n s H ig o p on n yr os y s us ub e a s D gl ro h a h tu llo c c e sc tu i o p ys rp p P m hr e rc e s D hl ro hr o om la hy yi ys ph u ns C lo c p s ha llo as o a lif h ro nis s ii ep m n ce lo er C lo r thu er oc Z yia igr ph te h so a eff an im s oc a s C do gn ri cy P me cin ap lus Iri so nis is yr riu er ill ni or up la rho s eic us A or ra eo ra my Ela ch ep hl ath lav scu N ias en rys s C or a f ob ep cin ia f op Sp reb iza P hel nam ran s oe osp us laty omy om tzi C m ace llis rinc ias e i Ase oliv nico L hus pu us ris cya epto alb lche Tia ara lina pogo ogu r ng vitrio Mi n ru laris Ta ara i Po onec fipec ang heine gono tes s tus T ara rvix triccus triatic Tang rufice Po ophth ollis gara iridis gonotri almicu Tan nigrov ccus po s angara rides Pseud ecilotis T labrado otriccus p Tangara Hemit elzelni a gyrola riccus granad Tangar ensis ra aurulenta Lophotriccus pile Tanga atus ara xanthocephala Pachyramphus versicolor Tang Pachyramphus polychopterus Nothocercus bonapartei Ampelioides tschudii C rii hamaepetes gou ipreola rieffe Odo dotii P hus ntophorus cryptolop Piaya hyperythru owornis nus cayana s Sn peruvia Patag upicola apilla Z ioenas R flavic s entry albilin ropipo pteru C gon fr ea Chlo hryso tus haet enata ius c stic Eut ura c Mas holo eps oxe inere ctes atic Pha res a iven pade virg bilis P etho quil tris Thri tes no s hae rni a dec s ig lari D th s gu ripa cte ba s ory orn y Th ade su olli Co fer is s hrip yla tic ns S lib a lu yrm T act tria ce c ri d do ato nd s es us C his el vic ph Sy thia nn lat r al te ph ae oru er ru tel ge C lip s g ina s ac x b s li ii hlo hl eo e ab le nis ttu au Ch r ox ffr An op or gu e e A a os m oy mn ar is nn ra m lyb tilb itc i re rg rn so za us A a u o he P Ma o is a til s A m z ra n lli mn o is ru lu d a ilia b m e b x s il er A e zi f u ela Pr es lla p us g g lo lia ra ffo n pt a o p i ris A la m s n n or la yn n s m a s C g io y a ci ii h o S e u ry l u o la c ia u a yn oc X h c u n a C e io e c e c d p la g in H o l c r m er hu u m an m n s e ig e cu e o s se a s ri u n u a O a l e r s la tt P h te t ic n U p ig n c n e r p s p ra a in s c lo e u k o i lo a y c d u s H r r a s in g y l u s t i r t i o e p n c c g e p co h te x u c s a H e c a h a o o i n c la e t e t A e l h t a i m o n p n c is M io u to e e y a id y la c m a ip e r r c l r s e li le s C h in z f l n t o i R c io d o d rq g p r o s a ti i z B o a e s e c c e u s s u i d o u ia u t L o o u r o o n e i G r p x n n is h o r r a m i l s G u p p i o b a a r s i v r i r M t a d a t ip d d ru m t i ic p a l e t o h e x o a u a v t o s i C l e e u X n n e i a l v o a l C a r n a i u l a r f l t p r r P e o r r m r r e e u a n T i a n g w e c h l z o s T c u a a e M i s r s n o M g f s H D D g a e e P h c n p u l l i i c a r p r u o i C c e s t S i k u r c c u i i s a i e c t s o a s r r d u s a c o c r b e e r o r r o i i o o a a i a l t a a e l u i i i p n t i c s d a h a r r a a g u g a m u m u r a i d r l c a c e g m e m p n a i l b r u i t l n o u u v a t v i o c i t u t i y a l n s b a l c i p t i i o e y o l c o u a u o t i a a a t s a o s c l i r p m m r a o s p c a t r o s g o n p r a b m s i i F t t i r n g i u a u s l s l d x o g p u o o o t a t p i r G a y s n n s s h u n a e l l u s i e n t a d v l i e c a c s r p p c l p l G u c c n i s u u a o s o i b r r i a d h g a h m e r a n m l i s i t u r S p n s a s t h c u o i i s g i b i p i l p u t l c a l a l o s o l r s o h m u b b b f r a h a a a a G l m o r a n s e t h g u l c o e l e r u h m r o i i r t p r p a a c o t c c s i n r s i r m a s t c c q i u a c o

a s o a s s m o a s i h s r a n y u r c i t m h n l a u s s u i i a l y u y c n a m e c b a m y u s s a i a s e u d n s n s S D y h i a t t i y h o h b t r t p t u a a t r o e a u o r n d c i p a T u i n o s t D o r c D r h n u V t r o e P a c u E e o l i e m c c e s o a n a s

r n o p C n a a n l l s l y i y i s s C s i u m o s M h r e c A

u o S

c e

a L l

Figure B-2. Phylogeny of Andean bird species with median covariate betas. The first through the outermost rings indicate the median effect size of the environmental covariates: forest cover within 1 km (FC1), edge density within 1 km (ED), canopy height (CH), and density of large-diameter trees (TS). Warmer values indicate a positive correlation between species occupancy and increasing values of the covariate.

129

r1, 2 r1, 3 r1, 4 r1, 5 A B C D

Occupancy

−0.4 0.0 0.4 0.8 −0.4 0.0 0.4 0.0 0.5 1.0 −1.00 −0.85 −0.70

d$x rd$x2, 3 rd$x2, 4 rd$x2, 5 Median E F G 90%CI Mist Net 0.11 Summer

0.85 0.95 −1.0 −0.4 0.2 −0.8 −0.4 0.0

d$x rd$x3, 4 rd$x3, 5 H I Mist Net 0.08 0.95 Winter

−1.0 −0.4 0.0 −0.6 −0.2 0.2

d$x rd$x4, 5 J

0.56 −0.7 −0.72 Owl Survey

−1.0 0.0 0.5

d$x

−0.9 −0.24 −0.21 −0.41 Transect

Figure B-3. Correlation among occupancy and detection covariance effect parameters {푢, 푣푀푁푆, 푣푀푁푊, 푣푂푆, 푣푇}. On the upper diagonal is depicted the posterior distribution of the correlation between parameters (solid line) with the 90% credible interval region (filled region) and the median (vertical line). On the lower diagonal is the Pearson correlation coefficient (i.e. the median correlation from the posterior distributions).

130

LIST OF REFERENCES

Achard, F., R. Beuchle, P. Mayaux, H. J. Stibig, C. Bodart, A. Brink, S. Carboni, B. Desclee, F. Donnay, H. D. Eva, A. Lupi, et al. (2014). Determination of tropical deforestation rates and related carbon losses from 1990 to 2010. Global Change Biology 20:2540-2554.

Aldrich, M., C. Billington, M. Edwards, and R. Laidlaw (1997). Tropical montane cloud forests: an urgent priority for conservation. WCMC Biodiversity Bulletin 2.

Alroy, J. (2017). Effects of habitat disturbance on tropical forest biodiversity. Proceedings of the National Academy of Sciences of the United States of America 114:6056-6061.

Anderson, S. H., D. Kelly, J. J. Ladley, S. Molloy, and J. Terry (2011). Cascading effects of bird functional extinction reduce pollination and plant density. Science 331:1068-1071.

Arbeláez-Cortés, E., and O. H. Marín-Gómez (2012). The composition of mixed-species bird flocks in Alto Quindio, Colombia. Wilson Journal of Ornithology 124:572-580.

Arbeláez-Cortés, E., H. A. Rodriguez-Correa, and M. Restrepo-Chica (2011). Mixed bird flocks: patterns of activity and species composition in a region of the Central Andes of Colombia. Revista Mexicana de Biodiversidad 82:639-651.

Arcilla, N., L. H. Holbech, and S. O'Donnell (2015). Severe declines of understory birds follow illegal logging in Upper Guinea forests of Ghana, West Africa. Biological Conservation 188:41-49.

Armenteras, D., C. Cadena Vargas, and R. P. Moreno (2007). Evaluación del estado de los bosques de niebla y de la meta 2010 en Colombia.

Armenteras, D., N. Rodriguez, J. Retana, and M. Morales (2011). Understanding deforestation in montane and lowland forests of the Colombian Andes. Regional Environmental Change 11:693-705.

Arriaga-Weiss, S. L., S. Calme, and C. Kampichler (2008). Bird communities in rainforest fragments: guild responses to habitat variables in Tabasco, Mexico. Biodiversity and Conservation 17:173-190.

Aubad, J., P. Aragon, M. A. Olalla-Tarraga, and M. A. Rodriguez (2008). Illegal logging, landscape structure and the variation of tree species richness across North Andean forest remnants. Forest Ecology and Management 255:1892-1899.

Aubad, J., P. Aragon, and M. A. Rodríguez (2010). Human access and landscape structure effects on Andean forest bird richness. Acta Oecologica-International Journal of Ecology 36:396-402.

Babweetera, F., and N. Brown (2008). Can remnant frugivore species effectively disperse tree seeds in secondary tropical rainforests? Biodiversity and Conservation 18:1611-1627.

131

Bagchi, R., Brown, L.M., Elphick, C.S., Wagner, D.L., and M. S. Singer (2018). Anthropogenic fragmentation of landscapes: mechanisms for eroding the specificity of plant–herbivore interactions. Oecologia 187:1-13.

Banks-Leite, C., R. M. Ewers, and J. P. Metzger (2010). Edge effects as the principal cause of area effects on birds in fragmented secondary forest. Oikos 119:918-926.

Barlow, J., Franca, F., Gardner, T. A., Hicks, C. C., Lennox, G. D., Berenguer, E., Castello, L., Economo, E. P., Ferreira, J., Guenard, B., Leal, C. G., Isaac, V., Lees, A. C., Parr, C. L., Wilson, S. K., Young, P. J., and N. A. J. Graham (2018). The future of hyperdiverse tropical ecosystems. Nature 559:517-526.

Barlow, J., G. D. Lennox, J. Ferreira, E. Berenguer, A. C. Lees, R. Mac Nally, J. R. Thomson, S. F. D. Ferraz, J. Louzada, V. H. F. Oliveira, L. Parry, et al. (2016). Anthropogenic disturbance in tropical forests can double biodiversity loss from deforestation. Nature 535:144-147.

Bartoń, K. (2019). Package ‘MuMIn’: Multi-Model Inference. Version 1.43.6.

Baselga, A. (2010). Partitioning the turnover and nestedness components of beta diversity. Global Ecology and Biogeography 19:134-143.

Baselga, A., and C. D. L. Orme (2012). betapart: an R package for the study of beta diversity. Methods in Ecology and Evolution 3:808-812.

Bates, D., M. Machler, B. M. Bolker, and S. C. Walker (2015). Fitting linear mixed-effects models using lme4. Journal of Statistical Software 67:1-48.

Becker, C. D., T. M. Loughin, and T. Santander (2008). Identifying forest-obligate birds in tropical moist cloud forest of Andean Ecuador. Journal of Field Ornithology 79:229-244.

Benitez-Lopez, A., R. Alkemade, A. M. Schipper, D. J. Ingram, P. A. Verweij, J. A. J. Eikelboom, and M. A. J. Huijbregts (2017). The impact of hunting on tropical mammal and bird populations. Science 356:180-183.

Bennun, L., C. Dranzoa, and D. Pomeroy (1996). The forest birds of Kenya and Uganda. Journal of East African Natural History 85:23-48.

Betts, M. G., C. Wolf, W. J. Ripple, B. Phalan, K. A. Millers, A. Duarte, S. H. M. Butchart, and T. Levi (2017). Global forest loss disproportionately erodes biodiversity in intact landscapes. Nature 547:441-444.

Bhakti, T., F. Goulart, C. S. de Azevedo, and Y. Antonini (2018). Does scale matter? The influence of three- level spatial scales on forest bird occurrence in a tropical landscape. Plos One 13.

Beyer, H. L. (2015). Geospatial Modelling Environment. Spatial Ecology LLC.

132

Bohórquez, C. I. (2003). Mixed-species bird flocks in a montane cloud forest of Colombia. Ornitología Neotropical 14:67-78.

Bolker, B. M., Brooks, M. E., Clark, C. J., Geange, S. W., Poulsen, J. R., Stevens, M. H. H. and J. S. S. White (2009). Generalized linear mixed models: a practical guide for ecology and evolution. Trends in Ecology & Evolution 24:127-135.

Borah, B., Quader, S. and U. Srinivasan (2018). Responses of interspecific associations in mixed-species bird flocks to selective logging. Journal of Applied Ecology 55:1637- 1646.

Bovo, A. A. A., K. Ferraz, M. Magioli, E. R. Alexandrino, E. Hasui, M. C. Ribeiro, and J. A. Tobias (2018). Habitat fragmentation narrows the distribution of avian functional traits associated with seed dispersal in tropical forest. Perspectives in Ecology and Conservation 16:90-96.

Brandt, C. S., H. Hasenack, R. R. Laps, and S. M. Hartz (2009). Composition of mixed-species bird flocks in forest fragments of southern Brazil. Zoologia 26:488-498.

Bregman, T. P., C. H. Sekercioglu, and J. A. Tobias (2014). Global patterns and predictors of bird species responses to forest fragmentation: Implications for ecosystem function and conservation. Biological Conservation 169:372-383.

Broms, K. M., M. B. Hooten, and R. M. Fitzpatrick (2016). Model selection and assessment for multi-species occupancy models. Ecology 97:1759-1770.

Burivalova, Z., T. M. Lee, X. Giam, C. H. Sekercioglu, D. S. Wilcove, and L. P. Koh (2015). Avian responses to selective logging shaped by species traits and logging practices. Proceedings of the Royal Society B-Biological Sciences 282.

Burnham, K., and D. Anderson (2002). Model selection and multimodel inference: a practical information-theoretic approach. Springer, New York, NY.

Carrara, E., V. Arroyo-Rodriguez, J. H. Vega-Rivera, J. E. Schondube, S. M. de Freitas, and L. Fahrig (2015). Impact of landscape composition and configuration on forest specialist and generalist bird species in the fragmented Lacandona rainforest, Mexico. Biological Conservation 184:117-126.

Ceballos, G., Ehrlich, P., Barnosky, A., García, A., Pringle, R., and T. Palmer (2015). Accelerated modern human-induced species losses: entering the sixth mass extinction? Science Advances 1.

Ceballos, G., Ehrlich, P. R., and R. Dirzo (2017). Biological annihilation via the ongoing sixth mass extinction signaled by vertebrate population losses and declines. Proceedings of the National Academy of Sciences of the United States of America 114:E6089-E6096.

133

Cerezo, A., S. Perelman, and C. S. Robbins (2010). Landscape-level impact of tropical forest loss and fragmentation on bird occurrence in eastern Guatemala. Ecological Modelling 221:512-526.

Chen, Q., Tomlinson, K. W., Cao, L., and B. Wang (2017). Effects of fragmentation on the seed predation and dispersal by rodents differ among species with different seed size. Integrative Zoology 12:468-476.

Colorado Zuluaga, G. J., and A. D. Rodewald (2015). Response of mixed-species flocks to habitat alteration and deforestation in the Andes. Biological Conservation 188:72-81.

Contreras, T. A. and K. E. Sieving (2011). Leadership of winter mixed-species flocks by tufted titmice (Baeolophus bicolor): are titmice passive nuclear species? International Journal of Zoology 2011:1-11.

Cordeiro, N. J., L. Borghesio, M. P. Joho, T. J. Monoski, V. J. Mkongewa, and C. J. Dampf (2015). Forest fragmentation in an African biodiversity hotspot impacts mixed-species bird flocks. Biological Conservation 188:61-71.

Cordeiro, N. J. and H. F. Howe (2003). Forest fragmentation severs mutualism between seed dispersers and an endemic African tree. Proceedings of the National Academy of Sciences of the United States of America 100:14052-14056.

Cramer, J. M., Mesquita, R. C. G., and G. B. Williamson (2007). Forest fragmentation differentially affects seed dispersal of large and small-seeded tropical trees. Biological Conservation 137:415-423.

Csárdi, G. (2019). Package 'igraph': Network Analysis and Visualization.

Darrah, A. J., and K. G. Smith (2013). Comparison of foraging behaviors and movement patterns of the Wedge-billed Woodcreeper (Glyphorynchus spirurus) traveling alone and in mixed-species flocks in Amazonian Ecuador. Auk 130:629-636. del Hoyo, J., A. Elliott, J. Sargatal, D. A. Christie, and E. de Juana (2020). Handbook of the Birds of the World Alive. Lynx Editions, Barcelona.

Derlindati, E. J., and S. M. Caziani (2005). Using canopy and understory mist nets and point counts to study bird assemblages in Chaco forests. Wilson Bulletin 117:92-99.

Dewalt, S. J., Schnitzer, S. A., and J. S. Denslow (2000). Density and diversity of lianas along a chronosequence in a central Panamanian lowland forest. Journal of Tropical Ecology 16:1-19.

Ding, Z. F., K. J. Feeley, Y. P. Wang, R. J. Pakeman, and P. Ding (2013). Patterns of bird functional diversity on land-bridge island fragments. Journal of Animal Ecology 82:781- 790.

134

Dolby, A. S., and T. C. Grubb (1998). Benefits to satellite members in mixed-species foraging groups: an experimental analysis. Animal Behaviour 56:501-509.

Emer, C., Galetti, M., Pizo, M. A., Guimaraes, P. R., Moraes, S., Piratelli, A., and P. Jordano (2018). Seed-dispersal interactions in fragmented landscapes - a metanetwork approach. Ecology Letters 21:484-493.

Etter, A., C. McAlpine, K. Wilson, S. Phinn, and H. P. Possingham (2006). Regional patterns of agricultural land use and deforestation in Colombia. Agriculture Ecosystems & Environment 114:369-386.

Fagan, W. E., Cantrell, R. S., and C. Cosner (1999). How habitat edges change species interactions. American Naturalist 153:165-182.

Fahrig, L. (2003). Effects of habitat fragmentation on biodiversity. Annual Review of Ecology Evolution and Systematics 34:487-515.

Fahrig, L. (2017). Ecological responses to habitat fragmentation per se. Annual Review of Ecology, Evolution, and Systematics 48:1-23.

Faith, D. P. (1992). Conservation evaluation and phylogenetic diversity. Biological Conservation 61:1-10.

Farine, D. R., Garroway, C. J., and B. C. Sheldon (2012). Social network analysis of mixed- species flocks: exploring the structure and evolution of interspecific social behaviour. Animal Behaviour 84:1271-1277.

Farine, D. R., and P. J. Milburn (2013). Social organisation of thornbill-dominated mixed- species flocks using social network analysis. Behavioral Ecology and Sociobiology 67:321-330.

Farine, D. R., and H. Whitehead (2015). Constructing, conducting and interpreting animal social network analysis. Journal of Animal Ecology 84:1144-1163.

Ferraz, G., J. D. Nichols, J. E. Hines, P. C. Stouffer, R. O. Bierregaard Jr., and T. E. Lovejoy (2007). A large-scale deforestation experiment: effects of patch area and isolation on Amazon birds. Science 315:238-241.

Fletcher, R. J., Didham, R. K., Banks-Leite, C., Barlow, J., Ewers, R. M., Rosindell, J., Holt, R. D., Gonzalez, A., Pardini, R., Damschen, E. I., Melo, F. P. L., Ries, L., Prevedello, J. A., Tschartke, T., Laurance, W. F., Lovejoy, T., and N. M. Haddad (2018). Is habitat fragmentation good for biodiversity? Biological Conservation 226:9-15.

Flynn, D. F. B., M. Gogol-Prokurat, T. Nogeire, N. Molinari, B. T. Richers, B. B. Lin, N. Simpson, M. M. Mayfield, and F. DeClerck (2009). Loss of functional diversity under land use intensification across multiple taxa. Ecology Letters 12:22-33.

135

Franks, D. W., Ruxton, G. D., and R. James (2010). Sampling animal association networks with the gambit of the group. Behavioral Ecology and Sociobiology 64:493-503.

Galetti, M., Alves-Costa, C. P., and E. Cazetta (2003). Effects of forest fragmentation, anthropogenic edges and fruit colour on the consumption of ornithocoric fruits. Biological Conservation 111:269-273.

Geist, H. J., and E. F. Lambin (2002). Proximate causes and underlying driving forces of tropical deforestation. Bioscience 52:143-150.

Gelman, A., J. B. Carlin, H. S. Stern, D. B. Dunson, A. Vehtari, and D. B. Rubin (2013). Bayesian Data Analysis, 3rd Edition. Chapman Hall/CRC, Boca Raton, FL.

Gelman, A., and D. B. Rubin (1992). Inference from iterative simulation using multiple sequences. Statistical Science 7:457-472.

Gelman, A., D. Simpson, and M. Betancourt (2017). The prior can often only be understood in the context of the likelihood. Entropy 19.

Gibson, L., T. M. Lee, L. P. Koh, B. W. Brook, T. A. Gardner, J. Barlow, C. A. Peres, C. J. A. Bradshaw, W. F. Laurance, T. E. Lovejoy, and N. S. Sodhi (2011). Primary forests are irreplaceable for sustaining tropical biodiversity. Nature 478:378-381.

Goodale, E. and G. Beauchamp (2010). The relationship between leadership and gregariousness in mixed-species bird flocks. Journal of Avian Biology 41:99-103.

Goodale, E., S. W. Kotagama, T. R. S. Raman, S. Sidhu, U. Goodale, S. Parker, and J. Chen (2014). The response of birds and mixed-species bird flocks to human-modified landscapes in Sri Lanka and southern India. Forest Ecology and Management 329:384- 392.

Gower, J. C. (1971). A general coefficient of similarity and some of its properties. Biometrics 27:857-871.

Graves, G. R. (1988). Linearity of geographic range and its possible effect on the population structure of Andean birds. Auk 105:47-52.

Gray, M. A., S. L. Baldauf, P. J. Mayhew, and J. K. Hill (2007). The response of avian feeding guilds to tropical forest disturbance. Conservation Biology 21:133-141.

Guevara, E. A., J. C. Valarezo, A. Onofa, and F. Cupueran (2011). Mixed-species flock composition in a Northwestern Ecuadorian cloud forest. Ornitología Neotropical 22:379- 386.

Hackett, S. J., R. T. Kimball, S. Reddy, R. C. K. Bowie, E. L. Braun, M. J. Braun, J. L. Chojnowski, W. A. Cox, K. L. Han, J. Harshman, C. J. Huddleston, et al. (2008). A phylogenomic study of birds reveals their evolutionary history. Science 320:1763-1768.

136

Haddad, N. M., Brudvig, L. A., Clobert, J., Davies, K. F., Gonzalez, A., Holt, R. D., Lovejoy, T. E., Sexton, J. O., Austin, M. P., Collins, C. D., Cook, W. M., Damschen, E. I., Ewers, R. M., Foster, B. L., Jenkins, C. N., King, A. J., Laurance, W. F., Levey, D. J., Margules, C. R., Melbourne, B. A., Nichols, A. O., Orrock, J. L., Song, D. X. and J. R. Townsend (2015). Habitat fragmentation and its lasting impact on Earth's ecosystems. Science Advances 1.

Hansbauer, M. M., I. Storch, S. Leu, J.-P. Nieto-Holguin, R. G. Pimentel, F. Knauer, and J. P. W. Metzger (2008). Movements of Neotropical understory passerines affected by anthropogenic forest edges in the Brazilian Atlantic rainforest. Biological Conservation 141:782-791.

Hermes, C., G. Segelbacher, and H. M. Schaefer (2018). A framework for prioritizing areas for conservation in tropical montane cloud forests. Ecoscience 25:97-108.

Hill, J. K., M. A. Gray, C. V. Khen, S. Benedick, N. Tawatao, and K. C. Hamer (2011). Ecological impacts of tropical forest fragmentation: how consistent are patterns in species richness and nestedness? Philosophical Transactions of the Royal Society B-Biological Sciences 366:3265-3276.

Iknayan, K. J., M. W. Tingley, B. J. Furnas, and S. R. Beissinger (2014). Detecting diversity: emerging methods to estimate species diversity. Trends in Ecology & Evolution 29:97- 106.

Isler, M., L., and P. Isler, R. (1999). The Tanagers, Natural History, Distribution, and Identification. Smithsonian Institution Scholarly Press, Washington, D.C.

James, F. C., and H. H. Shugart Jr. (1970). A quantitative method of habitat description. Audubon Field Notes 24:727-736.

Jankowski, J. E., A. L. Ciecka, N. Y. Meyer, and K. N. Rabenold (2009). Beta diversity along environmental gradients: implications of habitat specialization in tropical montane landscapes. Journal of Animal Ecology 78:315-327.

Jankowski, J. E., C. L. Merkord, W. F. Rios, K. G. Cabrera, N. S. Revilla, and M. R. Silman (2013). The relationship of tropical bird communities to tree species composition and vegetation structure along an Andean elevational gradient. Journal of Biogeography 40:950-962.

Jedlicka, J. A., R. Greenberg, I. Perfecto, S. M. Philpottt, and T. V. Dietsch (2006). Seasonal shift in the foraging niche of a tropical avian resident: resource competition at work? Journal of Tropical Ecology 22:385-395.

Jetz, W., G. H. Thomas, J. B. Joy, K. Hartmann, and A. O. Mooers (2012). The global diversity of birds in space and time. Nature 491:444-448.

Jordano, P. (2016). Chasing species interactions. PLoS Biology 14.

137

Jullien, M., and J. M. Thiollay (1998). Multi-species territoriality and dynamic of neotropical forest understorey bird flocks. Journal of Animal Ecology 67:227-252.

Kattan, G. H., H. Alvarezlopez, and M. Giraldo (1994). Forest fragmentation and bird extinctions- San Antionio 80 years later. Conservation Biology 8:138-146.

Keinath, D. A., D. F. Doak, K. E. Hodges, L. R. Prugh, W. Fagan, C. H. Sekerçioglü, S. H. M. Buchart, and M. Kauffman (2017). A global analysis of traits predicting species sensitivity to habitat fragmentation. Global Ecology and Biogeography 26:115-127.

Kellner, K. (2019). Package 'jagsUI': A Wrapper Around 'rjags' to Streamline 'JAGS' Analyses.

Kembel, S. W., D. D. Ackerly, S. P. Blomberg, S. K. Cornwell, P. D. Cowan, M. R. Helmus, H. Morlon, and C. O. Webb (2019). Pacakge ‘picante’: Integrating Phylogenies and Ecology.

Kéry, M. (2010). Introduction to WinBUGS for Ecologists: Bayesian Approach to Regression, ANOVA, Mixed Models and Related Analyses. Academic Press, Burlington, MA.

Khimoun, A., C. Eraud, A. Ollivier, E. Arnoux, V. Rocheteau, M. Bely, E. Lefol, M. Delpuech, M. L. Carpentier, G. Leblond, A. Levesque, et al. (2016). Habitat specialization predicts genetic response to fragmentation in tropical birds. Molecular Ecology 25:3831-3844.

Kier, G., H. Kreft, T. M. Lee, W. Jetz, P. L. Ibisch, C. Nowicki, J. Mutke, and W. Barthlott (2009). A global assessment of endemism and species richness across island and mainland regions. Proceedings of the National Academy of Sciences of the United States of America 106:9322-9327.

King, D. I. and J. H. Rappole (2001). Mixed-species flocks in dipterocarp forest of north-central Burma (Myanmar). Ibis 143:380-390.

Knowlton, J. L., and C. H. Graham (2011). Species interactions are disrupted by habitat degradation in the highly threatened Tumbesian region of Ecuador. Ecological Applications 21:2974-2986.

Komsta, L. and F. Novomestky (2015). Package 'moments': Moments, cumulants, skewness, kurtosis and related tests.

König, C., and F. Weick (2009). Owls of the World, 2nd Ed. Yale University Press, New Haven, CT.

Koster, N., Freidrich, K., Nieder, J. and W. Barthlott (2009). Conservation of epiphyte diversity in an Andean landscape transformed by human land use. Conservation Biology 23:911- 919.

Kotagama, S. W., and E. Goodale (2004). The composition and spatial organisation of mixed- species flocks in a Sri Lankan rainforest. Forktail 20:63-70.

138

Kruschke, J. K. (2013). Bayesian estimation supersedes the t-test. Journal of Experimental Psychology-General 142:573-603.

Kruskal, J. B. (1964). Nonmetric multidimensional scaling- a numerical method. Psychometrika 29:115-129.

Laliberté, E., and P. Legendre (2010). A distance-based framework for measuring functional diversity from multiple traits. Ecology 91:299-305.

Laliberté, E., P. Legendre, and B. Shipley (2015). Package ‘FD’: Measuring functional diversity (FD) from multiple traits, and other tools for functional ecology. version 1.0-12

Laureto, L. M. O., M. V. Cianciaruso, and D. S. M. Samia (2015). Functional diversity: an overview of its history and applicability. Natureza & Conservação 13:112-116.

Lee, T. M., M. C. K. Soh, N. Sodhi, L. P. Koh, and S. L. H. Lim (2005). Effects of habitat disturbance on mixed species bird flocks in a tropical sub-montane rainforest. Biological Conservation 122:193-204.

Legendre, P. (2014). Interpreting the replacement and richness difference components of beta diversity. Global Ecology and Biogeography 23:1324-1334.

Legendre, P. and M. J. Anderson (1999). Distance-based redundancy analysis: Testing multispecies responses in multifactorial ecological experiments. Ecological Monographs 69:1-24.

Legendre, P., Oksanen, J. and C. J. F. Ter Braak (2011). Testing the significance of canonical axes in redundancy analysis. Methods in Ecology and Evolution 2:269-277.

Leprieur, F., C. Albouy, J. De Bortoli, P. F. Cowman, D. R. Bellwood, and D. Mouillot (2012). Quantifying Phylogenetic Beta Diversity: Distinguishing between 'True' Turnover of Lineages and Phylogenetic Diversity Gradients. Plos One 7.

Lewis, S. L., Edwards, D. P. and D. Galbraith (2015). Increasing human dominance of tropical forests. Science 349:827-832.

MacArthur, R., H. Recher, and M. Cody (1966). On the relation between habitat selection and species diversity. American Naturalist 100:319-332.

Macedo Mestre, L. A., M. Cohn-Haft, and M. M. Dias (2010). Diet and prey availability of terrestrial insectivorous birds prone to extinction in Amazonian forest fragments. Brazilian Archives of Biology and Technology 53:1371-1381.

Magrach, A., Larrinaga, A. R. and L. Santamaria (2012). Internal habitat quality determines the effects of fragmentation on austral forest climbing and epiphytic angiosperms. Plos One 7.

139

Magrach, A., W. F. Laurance, A. R. Larrinaga, and L. Santamaria (2014). Meta-analysis of the effects of forest fragmentation on interspecific interactions. Conservation Biology 28:1342-1348.

Maldonado-Coelho, M., and M. A. Marini (2004). Mixed-species bird flocks from the Brazilian Atlantic forest: the effects of forest fragmentation and seasonality on their size, richness and stability. Biological Conservation 116:19-26.

Mammides, C., Chen, J., Goodale, U. M., Kotagama, S. W. and E. Goodale (2018). Measurement of species associations in mixed-species bird flocks across environmental and human disturbance gradients. Ecosphere 9.

Mammides, C., J. Chen, U. M. Goodale, S. W. Kotagama, S. Sidhu, and E. Goodale (2015). Does mixed-species flocking influence how birds respond to a gradient of land-use intensity? Proceedings of the Royal Society B-Biological Sciences 282.

Mangini, G. G., and J. I. Areta (2018). Bird mixed-species flock formation is driven by low temperatures between and within seasons in a Subtropical Andean-foothill forest. Biotropica:1-10.

Mansor, M. S., Rozali, F. Z., Abdullah, N. A., Nor, S. M. and R. Ramli (2019). How important is aerial leaf litter for insectivorous birds foraging in a Malaysian tropical forest? Global Ecology and Conservation 20.

Manu, S., W. Peach, and W. Cresswell (2007). The effects of edge, fragment size and degree of isolation on avian species richness in highly fragmented forest in West Africa. Ibis 149:287-297.

Marín-Gómez, O. H., and E. Arbelaez-Cortes (2015). Variation on species composition and richness in mixed bird flocks along an altitudinal gradient in the Central Andes of Colombia. Studies on Neotropical Fauna and Environment 50:113-129.

Marra, P. P., and J. V. Remsen, Jr. (1997). Insights into the maintenance of high species diversity in the Neotropics: habitat selection and foraging behavior in understory birds of tropical and temperate forests. Ornithological Monographs 48:445-483.

Martínez, A., E. Parra, O. Muellerklein, and V. T. Vredenburg (2018). Fear-based niche shifts in neotropical birds. Ecology 99:1338-1346.

Mason, N. W. H., D. Mouillot, W. G. Lee, and J. B. Wilson (2005). Functional richness, functional evenness and functional divergence: the primary components of functional diversity. Oikos 111:112-118.

McConkey, K. R., and G. O'Farrill (2016). Loss of seed dispersal before the loss of seed dispersers. Biological Conservation 201:38-49.

McDermott, M. E., and A. D. Rodewald (2014). Conservation value of silvopastures to Neotropical migrants in Andean forest flocks. Biological Conservation 175:140-147.

140

McDermott, M. E., A. D. Rodewald, and S. N. Matthews (2015). Managing tropical agroforestry for conservation of flocking migratory birds. Agroforestry Systems 89:383-396.

McGarigal, K., S. Cushman, and S. Stafford (2000). Multivariate statistics for wildlife and ecology research. Springer-Verlag, York, PA.

Melo, A. S., T. Rangel, and J. A. F. Diniz (2009). Environmental drivers of beta-diversity patterns in New-World birds and mammals. Ecography 32:226-236.

Mokross, K., J. R. Potts, C. L. Rutt, and P. C. Stouffer (2018). What can mixed-species flock movement tell us about the value of Amazonian secondary forests? Insights from spatial behavior. Biotropica:1-10.

Mokross, K., T. B. Ryder, M. C. Cortes, J. D. Wolfe, and P. C. Stouffer (2014). Decay of interspecific avian flock networks along a disturbance gradient in Amazonia. Proceedings of the Royal Society B-Biological Sciences 281.

Montano-Centellas, F. A., and A. Garitano-Zavala (2015). Andean bird responses to human disturbances along an elevational gradient. Acta Oecologica-International Journal of Ecology 65-66:51-60.

Morante, J. C., D. Faria, E. Mariano-Neto, and J. Rhodes (2015). Birds in anthropogenic landscapes: The responses of ecological groups to forest loss in the Brazilian atlantic forest. Plos One 10.

Morse, D. H. (1970). Ecological aspects of some mixed-species foraging flocks of birds. Ecological Monographs 40:119-168.

Moynihan, M. (1962). The organization and probable evolution of some mixed species flocks of neotropical birds. Smithsonian Mis Collections 143:1-140.

Munn, C. (1985). Permanent canopy and understory flocks in Amazonia: species composition and population density. Ornithological Monographs 36:683-712.

Munn, C. A., and J. W. Terborgh (1979). Multi-species territoriality in Neotropical foraging flocks. Condor 81:338-347.

Muñoz, J. (2016). The role of facilitation in the structure of tropical bird communities: a case study of mixed-species flocks. Masters Thesis, Department of Zoology. University of British Columbia, Vancouver, B.C.

Nakagawa, S., and R. P. Freckleton (2011). Model averaging, missing data, and multiple imputation: a case study for behavioral ecology. Behavioral Ecology and Sociobiology 65:103-116.

Nakagawa, S., and H. Schielzeth (2013). A general and simple method for obtaining r2 from generalized linear mixed-effects models. Methods in Ecology and Evolution 4:133-142.

141

Naoki, K. (2007). Arthropod resource partitioning among omnivorous Tanagers (Tangara spp.) in Western Ecuador. Auk 124:197-209.

Newman, M. E. J. (2006). Modularity and community structure in networks. Proceedings of the National Academy of Sciences of the United States of America 103:8577-8582.

O'Dea, N., and R. J. Whittaker (2007). How resilient are Andean montane forest bird communities to habitat degradation? Biodiversity and Conservation 16:1131-1159.

Ocampo-Peñuela, N., and S. L. Pimm (2014). Setting practical conservation priorities for birds in the Western Andes of Colombia. Conservation Biology 28:1260-1270.

Ohl, C., and R. Bussmann (2004). Recolonisation of natural landslides in tropical mountain forests of Southern Ecuador. Feddes Repertorium 115:248-264.

Oksanen, J., F. G. Blanchet, M. Friendly, R. Kindt, P. Legendre, D. McGlinn, P. R. Minchin, R. B. O'Hara, G. L. Simpson, P. Solymos, M. Henry, H. Stevens, Eduard Szoecs, and H. Wagner (2019). Package ‘vegan’: Ordination methods, diversity analysis and other functions for community and vegetation ecologists. Version 2.5-5.

Orme, C. D. L., R. G. Davies, M. Burgess, F. Eigenbrod, N. Pickup, V. A. Olson, A. J. Webster, T. S. Ding, P. C. Rasmussen, R. S. Ridgely, A. J. Stattersfield, et al. (2005). Global hotspots of species richness are not congruent with endemism or threat. Nature 436:1016- 1019.

Pagani-Nuñez, E., X. Xia, G. Beauchamp, R. C. He, J. H. D. Husson, D. Liang and E. Goodale (2018). Are vocal characteristics related to leadership patterns in mixed-species bird flocks? Journal of Avian Biology 49.

Pardieck, K. L., J. M. Meyers, and M. Pagan (1996). Surveys of Puerto Rican Screech-Owl populations in large-tract and fragmented forest habitats. Wilson Bulletin 108:776-782.

Patterson, B. D. (1987). The principle of nested subsets and its implications for biological conservation. Conservation Biology 1:323-334.

Péron, G., and P.-A. Crochet (2009). Edge effect and structure of mixed-species bird flocks in an Afrotropical lowland forest. Journal of Ornithology 150:585-599.

Petchey, O. L., and K. J. Gaston (2002). Functional diversity (FD), species richness and community composition. Ecology Letters 5:402-411.

Petchy, O. L., and K. J. Gaston (2006). Functional diversity: back to the basics and looking forward. Ecology Letters 9:741-758.

142

Pfeifer, M., V. Lefebvre, C. A. Peres, C. Banks-Leite, O. R. Wearn, C. J. Marsh, S. H. M. Butchart, V. Arroyo-Rodríguez, J. Barlow, A. Cerezo, L. Cisneros, N. D’Cruze, D. Faria, A. Hadley, S. M. Harris, B. T. Klingbeil, U. Kormann, L. Lens, G. F. Medina-Rangel, J. C. Morante-Filho, P. Olivier, S. L. Peters, A. Pidgeon, D. B. Ribeiro, C. Scherber, L. Scheider-Maunoury, M. Struebig, N. Urbina-Cardona, J. I. Watling, M. R. Willing, E. M. Wood, and R. M. Ewers (2017). Creation of forest edges has a global impact on forest vertebrates. Nature 551:187-191.

Poisot, T. (2016). Package ‘betalink’: Beta-Diversity of Species Interactions.

Poisot, T., E. Canard, D. Mouillot, N. Mouquet and D. Gravel (2012). The dissimilarity of species interaction networks. Ecology Letters 15:1353-1361.

Politi, N., M. Hunter, and L. Rivera (2010). Availability of cavities for avian cavity nesters in selectively logged subtropical montane forests of the Andes. Forest Ecology and Management 260:893-906.

Politi, N., M. Hunter, and L. Rivera (2012). Assessing the effects of selective logging on birds in Neotropical piedmont and cloud montane forests. Biodiversity and Conservation 21:3131-3155.

Potts, J. R., K. Mokross, P. C. Stouffer, and M. A. Lewis (2014). Step selection techniques uncover the environmental predictors of space use patterns in flocks of Amazonian birds. Ecology and Evolution 4:4578-4588.

Poulsen, B. O. (1996a). Relationships between frequency of mixed-species flocks, weather and insect activity in a montane cloud forest in Ecuador. Ibis 138:466-470.

Poulsen, B. O. (1996b). Structure, dynamics, home range and activity pattern of mixed-species bird flocks in a montane alder-dominated secondary forest in Ecuador. Journal of Tropical Ecology 12:333-343.

Powell, L. L., N. J. Cordeiro, and J. A. Stratford (2015a). Ecology and conservation of avian insectivores of the rainforest understory: A pantropical perspective. Biological Conservation 188:1-10.

Powell, L. L., P. C. Stouffer, and E. I. Johnson (2013). Recovery of understory bird movement across the interface of primary and secondary Amazon rainforest. Auk 130:459-468.

Powell, L. L., G. Zurita, J. D. Wolfe, E. I. Johnson, and P. C. Stouffer (2015b). Changes in habitat use at rainforest edges through succession: a case study of understory birds in the Brazilian Amazon. Biotropica 47:723-732.

Quiñones, F. A. (2018). Guía Ilustrada de la Avifauna Colombiana. Panamericana Formas e Impresos S. A., Bogota, Colombia.

Rappole, J. H. (1995). The ecology of migrant birds, a Neotropical perspective. Smithsonian Institution Press, Washington D.C.

143

R Core Team (2020). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

Remsen, J. V. J., and S. K. Robinson (1990). A classification scheme for foraging behavior of birds in terrestrial habitats. In Studies in Avian Biology, No. 13. Avian Foraging: Theory, Methodology, and Applications (Morrison, M. L. et al.) Cooper Ornithological Society: Los Angeles, California, USA. Illus. Paper:144-160.

Renjifo, L. M. (1999). Composition changes in a subandean avifauna after long-term forest fragmentation. Conservation Biology 13:1124-1139.

Renjifo, L. M. (2001). Effect of natural and anthropogenic landscape matrices on the abundance of subandean bird species. Ecological Applications 11:14-31.

Restrepo, C., and N. Gomez (1998). Responses of understory birds to anthropogenic edges in a Neotropical montane forest. Ecological Applications 8:170-183.

Revell, L. J. (2019). Package ‘phytools’: Phylogenetic Tools for Comparative Biology (and Other Things). Version 0.6-99.

Richards, S. A., M. J. Whittingham, and P. A. Stephens (2011). Model selection and model averaging in behavioural ecology: the utility of the IT-AIC framework. Behavioral Ecology and Sociobiology 65:77-89.

Rosenberg, K., Dokter, A., Blancher, P., Sauer, J., Smith, A., Smith, P., Stanton, J., Panjabi, A., Helft, L., Parr, M. and P. Marra (2019). Decline of North American avifauna. Science 366:120-124.

Rosenberg, K. (1990). Dead-leaf foraging specialization in tropical forest birds: measuring resource availability and use. In: Morrison M, Ralph C, Verner J, Jehl J. (eds.) Avian Foraging: Theory, Methodology, and Applications. Allen Press Inc., Lawrence, Kansas.

Rosselli, L., F. G. Stiles, and P. A. Camargo (2017). Changes in the avifauna in a high Andean cloud forest in Colombia over a 24-year period. Journal of Field Ornithology 88:211-228.

Royle, J. A., and M. Kéry (2007). A Bayesian state-space formulation of dynamic occupancy models. Ecology 88:1813-1823.

Rueda-Hernández, R., I. MacGregor-Fors, and K. Renton (2015). Shifts in resident bird communities associated with cloud forest patch size in Central Veracruz, Mexico. Avian Conservation and Ecology 10.

Säterberg, T., S. Sellman, and B. Ebenman (2013). High frequency of functional extinctions in ecological networks. Nature 499:468-470.

Sekerçioglü, C. H. (2002). Effects of forestry practices on vegetation structure and bird community of Kibale National Park, Uganda. Biological Conservation 107:229-240.

144

Sekerçioglü, C. H. (2012). Bird functional diversity and ecosystem services in tropical forests, agroforests and agricultural areas. Journal of Ornithology 153:S153-S161.

Sergio, F., and F. Hiraldo (2008). Intraguild predation in raptor assemblages: a review. Ibis 150:132-145.

Shiu, H. J., and P. F. Lee (2003). Assessing avian point-count duration and sample size using species accumulation functions. Zoological Studies 42:357-367.

Sidhu, S., T. R. S. Raman, and E. Goodale (2010). Effects of plantations and home gardens on tropical forest bird communities and mixed-species bird flocks in the southern Western Ghats. Journal of the Bombay Natural History Society 107:91-108.

Sillett, T. S. (1994). Foraging ecology of epiphyte-searching insectivorous birds in Costa Rica. Condor 96:863-877.

Slager, D. L. (2011). Rufous-naped Greenlet (Pachysylvia semibrunnea), version 1.0. In: Schulenberg, T. S. (ed.) Neotropical Birds Online. Cornell Lab of Ornithology, Ithaca, N.Y.

Smith, D. A. E., X. F. Si, Y. C. E. Smith, R. Kalle, T. Ramesh, and C. T. Downs (2018). Patterns of avian diversity across a decreasing patch-size gradient in a critically endangered subtropical forest system. Journal of Biogeography 45:2118-2132.

Sridhar, H., G. Beauchamp, and K. Shanker (2009). Why do birds participate in mixed-species foraging flocks? A large-scale synthesis. Animal Behaviour 78:337-347.

Sridhar, H., and K. Sankar (2008). Effects of habitat degradation on mixed-species bird flocks in Indian rain forests. Journal of Tropical Ecology 24:135-147.

Sridhar, H. and K. Shanker (2014). Importance of intraspecifically gregarious species in a tropical bird community. Oecologia 176:763-770.

Sridhar, H., Srinivansan, U., Askins, R. A., Canales-Delgadillo, J. C., Chen, C. C., Ewert, D. N., Gale, G. A., Goodale, E., Gram, W. K., Hart, P. J., Hobson, K. A., Hutto, R. L., Kotagama, S. W., Knowlton, J. L., Lee, T. M., Munn, C. A., Nimnuan, S., Nizam, B. Z., Péron, G., Robin, V. V., Rodewald, A. D., Rodewald, P. G., Thomson, R. L., Trivedi, P., Van Wilgenburg, S. L. and K. Shanker (2012). Positive Relationships between association strength and phenotypic similarity characterize the assembly of mixed-species bird flocks worldwide. American Naturalist 180:777-790.

Srinivasan, U. (2019). Morphological and behavioral correlates of long-term bird survival in a selectively logged forest. Frontiers in Ecology and Evolution 7.

Srinivasan, U., R. H. Raza, and S. Quader (2010). The nuclear question: rethinking species importance in multi-species animal groups. Journal of Animal Ecology 79:948-954.

145

Srinivasan, U., Raza, R. H. and S. Quader (2012). Patterns of species participation across multiple mixed-species flock types in a tropical forest in northeastern India. Journal of Natural History 46:2749-2762.

Stratford, J. A., and P. C. Stouffer (2013). Microhabitat associations of terrestrial insectivorous birds in Amazonian rainforest and second-growth forests. Journal of Field Ornithology 84:1-12.

Stratford, J. A., and P. C. Stouffer (2015). Forest fragmentation alters microhabitat availability for Neotropical terrestrial insectivorous birds. Biological Conservation 188:109-115.

Tabarelli, M., A. V. Lopes, and C. A. Peres (2008). Edge-effects Drive Tropical Forest Fragments Towards an Early-Successional System. Biotropica 40:657-661.

Tarbox, B. C., Robinson, S. K., Loiselle, B. and S. L. Flory (2018). Foraging ecology and flocking behavior of insectivorous forest birds inform management of Andean silvopastures for conservation. Condor 120:787-802.

Tejedor Garavito, N., E. Álvarez, S. Arango Caro, A. Araujo Murakami, C. Blundo, T. E. Boza Espinoza, M. A. La Torre Cuadros, J. Gaviria, N. Gutíerrez, P. M. Jørgensen, B. León, et al. (2012). Evaluación del estado de conservación de los bosques montanos en los Andes tropicales. Ecosistemas 21:148-166.

Tellería, J. L., E. Virgos, R. Carbonell, J. Perez-Tris, and T. Santos (2001). Behavioural responses to changing landscapes: flock structure and anti-predator strategies of tits wintering in fragmented forests. Oikos 95:253-264.

Terbraak, C. J. F. (1986). Canonical correspondence analysis- a new Eigenvector technique for multivariate direct gradient analysis. Ecology 67:1167-1179.

Thiollay, J. M. (1997). Disturbance, selective logging and bird diversity: A Neotropical forest study. Biodiversity and Conservation 6:1155-1173.

Tobias, J. A., J. D. Brawn, R. T. Brumfield, E. P. Derryberry, A. N. G. Kirschel, and N. Seddon (2012). The importance of suboscine birds as study systems in ecology and evolution. Ornitología Neotropical 23:261-274.

Tracewski, L., Butchart, S. H. M., Di Marco, M., Ficetola, G. F., Rondini, C., Symes, A., Wheatley, H., Beresford, A. E. and G. M. Buchanan (2016). Toward quantification of the impact of 21st-century deforestation on the extinction risk of terrestrial vertebrates. Conservation Biology 30:1070-1079.

Trejo, A., F. Beaudoin, and V. Ojeda (2011). Response of Rufous-legged Owls to broadcast of conspecific calls in southern temperate forests of Argentina. Journal of Raptor Research 45:267-271.

146

Tubelis, D. P., A. Cowling, and C. Donnelly (2006). Role of mixed-species flocks in the use of adjacent savannas by forest birds in the central Cerrado, Brazil. Austral Ecology 31:38- 45.

Tucker, C. M., M. W. Cadotte, S. B. Carvalho, T. J. Davies, S. Ferrier, S. A. Fritz, R. Grenyer, M. R. Helmus, L. S. Jin, A. O. Mooers, S. Pavoine, O. Purschke, D. W. Redding, D. F. Rosauer, M. Winter, and F. Mazel (2017). A guide to phylogenetic metrics for conservation, community ecology and macroecology. Biological Reviews 92:698-715.

Tylianakis, J. M., E. Laliberte, A. Nielsen, and J. Bascompte (2010). Conservation of species interaction networks. Biological Conservation 143:2270-2279.

Ulrich, W., L. Lens, J. A. Tobias, and J. C. Habel (2016). Contrasting patterns of species richness and functional diversity in bird communities of East African cloud forest fragments. Plos One 11.

Valburg, L. K. (1992). Flocking and frugivory- the effect of social groupings on resource use in the common bush-tanager. Condor 94:358-363.

Valiente-Banuet, A., M. A. Aizen, J. M. Alcantara, J. Arroyo, A. Cocucci, M. Galetti, M. B. Garcia, D. Garcia, J. M. Gomez, P. Jordano, R. Medel, et al. (2015). Beyond species loss: the extinction of ecological interactions in a changing world. Functional Ecology 29:299- 307.

Van Houtan, K. S., Pimm, S. L., Bierregaard, R. O., Lovejoy, T. E. and P. E. Stouffer (2006). Local extinctions in flocking birds in Amazonian forest fragments. Evolutionary Ecology Research 8:129-148.

Vellend, M., W. K. Cornwall, K. Magnuson-Ford, and A. O. Mooers (2014). Measuring phylogenetic biodiversity. In Biological Diversity, Frontiers in Measurement and Assessment (Magurran, A. E., and B. J. McGill). Oxford University Press, Oxford, UK.

Villéger, S., G. Grenouillet, and S. Brosse (2013). Decomposing functional diversity reveals that low functional diversity is driven by low functional turnover in European fish assemblages. Global Ecology and Biogeography 22:671-681.

Villéger, S., N. W. H. Mason, and D. Mouillot (2008). New multidimensional functional diversity indices for a multifaceted framework in functional ecology. Ecology 89:2290- 2301.

Villéger, S., J. R. Miranda, D. F. Hernandez, and D. Mouillot (2010). Contrasting changes in taxonomic vs. functional diversity of tropical fish communities after habitat degradation. Ecological Applications 20:1512-1522.

Walther, B. A. (2002). Grounded ground birds and surfing canopy birds: Variation of foraging stratum breadth observed in neotropical forest birds and tested with simulation models using boundary constraints. Auk 119:658-675.

147

Watson, J. E. M., D. F. Shanahan, M. Di Marco, J. Allan, W. F. Laurance, E. W. Sanderson, B. Mackey, and O. Venter (2016). Catastrophic declines in wilderness areas undermine global environment targets. Current Biology 26:2929-2934.

Webb, C. O., D. D. Ackerly, and S. W. Kembel (2008). Phylocom: software for the analysis of phylogenetic community structure and trait evolution. Bioinformatics 24:2098-2100.

Webb, C. O., D. D. Ackerly, M. A. McPeek, and M. J. Donoghue (2002). Phylogenies and community ecology. Annual Review of Ecology and Systematics 33:475-505.

Weiher, E. (2014). A primer of trait and functional diversity. In Biological Diversity, Frontiers in Measurement and Assessment (Magurran, A. E., and B. J. McGill). Oxford University Press, Oxford, UK.

Wethered, R., and M. J. Lawes (2005). Nestedness of bird assemblages in fragmented Afromontane forest: the effect of plantation forestry in the matrix. Biological Conservation 123:125-137.

Whitman, A. A., J. M. Hagan, and N. V. L. Brokaw (1997). A comparison of two bird survey techniques used in a subtropical forest. Condor 99:955-965.

Williams, S. M. and C. A. Lindell (2019). The influence of a single species on the space use of mixed-species flocks in Amazonian Peru. Movement Ecology 7.

Wilson, M. C., X. Y. Chen, R. T. Corlett, R. K. Didham, P. Ding, R. D. Holt, M. Holyoak, G. Hu, A. C. Hughes, L. Jiang, W. F. Laurance, et al. (2016). Habitat fragmentation and biodiversity conservation: key findings and future challenges. Landscape Ecology 31:219-227.

Young, H. S., D. J. McCauley, M. Galetti and R. Dirzo (2016). Patterns, Causes, and Consequences of Anthropocene Defaunation. In: Futuyuma, D. J. (ed.) Annual Review of Ecology, Evolution, and Systematics, Vol 47, pp. 333-358.

Zhang, Q., R. Han, Z. Huang, and F. Zou (2013). Linking vegetation structure and bird organization: response of mixed-species bird flocks to forest succession in subtropical China. Biodiversity and Conservation 22:1965-1989.

Zou, F., G. Chen, Q. Yang, and J. R. Fellowes (2011). Composition of mixed-species flocks and shifts in foraging location of flocking species on Hainan Island, China. Ibis 153:269-278.

Zou, F., H. H. Jones, G. J. Colorado, D. Jiang, T.-M. Lee, A. Martinez, K. Sieving, M. Zhang, Q. Zhang, and E. Goodale (2018). The conservation implications of mixed-species flocking in terrestrial birds, a globally-distributed species interaction network. Biological Conservation 224:267-276.

148

BIOGRAPHICAL SKETCH

Harrison Jones was born in Indianapolis, Indiana and graduated from the International

School of Indiana in 2008. He received a Bachelor of Science from Haverford College in 2012, double majoring in biology and French literature. After working for a couple of years as a field assistant for a variety of field-based bird research projects, he completed a Master of Science degree from the Wildlife Ecology and Conservation Department at the University of Florida in

2016. He received a PhD in zoology from the Department of Biology in 2020.

149