INFLUENCE OF CLIMATE ON THE ORIGIN AND MAINTENANCE OF TROPICAL BIRD COMMUNITIES
By JUAN PABLO GOMEZ
ADISSERTATIONPRESENTEDTOTHEGRADUATESCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2016 ⃝c 2016 Juan Pablo Gomez to my parents, Elena and Emilio ACKNOWLEDGMENTS First of all I would like to thank both of my advisors, Scott Robinson and Jose Miguel Ponciano. Scott, for giving me the opportunity to be part of his research group in such a great journey. I have learned so much about many things duringthesixyearsIhavespent at University of Florida. One of the things that I admire from him the most is his ability to dream about innumerable research possibilities and for getting excited about almost any idea I came up with, even though some of them were not that great. For that enthusiasm in my research and in advising me through this long journey thank you. Even though Jose Miguel came later into my advisory committee I cannot thank him anymore for all of his advice and help. He allowed me to turn my professional development on a trajectory that Iwouldhaveneverhadthoughtofbefore.Heintroducedmeinto a whole new world that has changed my way of thinking and viewing science in a different way. I also want to thank my committee, David Steadman, Bette Loiselle and Gordon Burleigh for all their help and advice. All of this work would not have been possible without the help and support of the landowners who voluntarily preserved the few remnants of natural forest left in the Magdalena valley and for being so helpful and supportive to me in my research. Specifically, I want to thank Cesar Garcia, Humberto Llara, Constanza Mendoza, Familia Pizano, Gareth Hughes,
Ana Maria Jaramillo, Reserva Natural Rio Claro, Hacienda Pampas, CorAntioquia, Andres Link and Gustavo Campuzano. Also, I would like to thank Juliana Sandoval, Julian Llano, Mario Loaiza, Jacob Drucker, Juan Luis Parra, Andrea Morales, Eliana Yepes and Sergio Arango for their assistance in the field. My funding sources: National Geographic, Sigma Xi, Cooper Ornithological Society, Wilson Ornithological Society, American Ornithologist Union and the
Ordway lab for ecosytsem conservation. Finally I would like to thank the entire Ordway lab, present and past members, who through their support and advice made this work possible.
4 TABLE OF CONTENTS page
ACKNOWLEDGMENTS ...... 4 LIST OF TABLES ...... 7 LIST OF FIGURES ...... 8 ABSTRACT ...... 9
CHAPTER 1INTRODUCTION...... 11 2EFFICIENTEXTENSIONOFN-MIXTUREMODELS...... 18 2.1 Background ...... 18 2.1.1 The Model ...... 20 2.1.2 Maximum Likelihood Estimation ...... 22 2.1.3 Scenarios In Which ps Are Correlated Among Neotropical Bird Species .24 2.2 Methods ...... 25 2.2.1 Sample Size Estimation for Neotropical Birds ...... 25 2.2.2 The Beta N-mixture Model ...... 25 2.2.3 Example Using Real Data ...... 26 2.3 Results ...... 27 2.3.1 Sample Size Estimation for Neotropical Birds ...... 27 2.3.2 The Beta N-mixture Model ...... 27 2.3.3 Example Using Real Data ...... 30 2.4 Discussion ...... 31 3ENVIRONMENTALFILTERINGONTROPICALBIRDCOMMUNITIES...... 38 3.1 Background ...... 38 3.2 Methods ...... 41 3.2.1 Bird Sampling ...... 41 3.2.2 Abundance Estimation ...... 42 3.2.3 Model Fitting ...... 43 3.2.4 Rarity Analysis ...... 45 3.3 Results ...... 46 3.3.1 Abundance Estimation ...... 46 3.3.2 Comparison Among Models ...... 46 3.3.3 Variable Immigration One-metacommunity Model ...... 46 3.3.4 Variable Immigration Two-metacommunity Model ...... 48 3.3.5 Rarity ...... 48 3.4 Discussion ...... 50
5 4DETERMINISTICTURNOVEROFTROPICALBIRDCOMMUNITIES...... 55 4.1 Background ...... 55 4.2 Methods ...... 58 4.2.1 Study Area ...... 58 4.2.2 Bird Sampling ...... 59 4.2.3 Community Turnover ...... 59 4.2.4 Climatic Description of Localities ...... 62 4.2.5 Trait Sampling ...... 63 4.2.5.1 Environmental Filtering ...... 64 4.2.5.2 Biotic interactions: Competition ...... 66 4.3 Results ...... 68 4.3.1 Compositional, Functional and Phylogenetic Turnover ...... 68 4.3.2 Environmental Variables ...... 70 4.3.3 Environmental Filtering ...... 70 4.3.4 Biotic Interactions: Competition ...... 72 4.4 Discussion ...... 74 5CONCLUDINGREMARKS...... 81
APPENDIX ASUPPLEMENTARYFIGURES...... 87 BSUPPLEMENTARYTABLES...... 93
CRCODESFORABUNDANCEESTIMATION...... 114 C.1 Bugs Models ...... 114 C.1.1 Beta N-mixture Model ...... 114 C.1.2 Sampling From the Beta N-Mixture model ...... 116 C.2 R Functions ...... 117 C.3 R Code ...... 120
REFERENCES ...... 127 BIOGRAPHICAL SKETCH ...... 139
6 LIST OF TABLES Table page
2-1 Abundance estimates for understory insectivorous birds...... 33 3-1 Characteristics of the study sites...... 42 3-2 Model Selection for the 13 metacommunities in the Magdalena Valley...... 47 4-1 Location and description of localities sampled...... 63
4-2 Model selection for the relationship between community similarity and rainfall. ... 69 4-3 Comparison of max and variance nest and ambient temperature...... 72 B-1 Population density estimates for species in each of the 13localities...... 94 B-2 Parameter estimation for each of the models evaluated...... 109 B-3 Relationship of Immigration Rate (m, I ), precipitation and distance...... 112
B-4 Relationship of number of rare species and precipitationandtotalnumberofspecies. 113
7 LIST OF FIGURES Figure page
2-1 Mean bias in mean number of individuals per 100 ha λ...... 28 2-2 Bias in the estimated value of λ for both, N-mixture and beta N-mixture model. .. 29 2-3 True and estimated species abundance distributions...... 30 2-4 Distribution of Expected value and variance of detectionprobability...... 31
3-1 Effects of precipitation on immigration rates...... 48 3-2 Differences in Immigration rates between types of forest and models...... 49 3-3 Relationship between precipitation and number of rare species...... 50 4-1 Compositional, functional and phylogenetic turnover along the rainfall gradient ... 71 4-2 Relationship between average community tolerances and rainfall ...... 73
4-3 Eco-morphological and nest structure of communities...... 80 A-1 Mean bias in mean number of individuals per 100 ha (λ)...... 88 A-2 Histogram of estimated detection probabilities of 26 understory insectivorous birds. .89 A-3 Distribution of λˆ using N-mixture and beta N-mixture models...... 90
A-4 Relationship between the mean value of λˆ and true value of λ...... 91 A-5 Relative abundance of Cavity and Cup nests...... 92
8 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
INFLUENCE OF CLIMATE ON THE ORIGIN AND MAINTENANCE OF TROPICAL BIRD COMMUNITIES By Juan Pablo Gomez
May 2016 Chair: Scott K. Robinson Cochair: Jose Miguel Ponciano Major: Zoology One of the main goals of community ecology is to understand theinfluenceofthe abiotic environment on the abundance and distribution of species. It has been hypothesized that dry forests are harsher environments than wet forests, which leads to the prediction that environmental filtering should be a more important determinant of patterns of species abundance and composition than in wet forest, where biotic interactions or random assembly should be more important. My goal is to understand the influence of rainfall on the abundance and distribution of bird species along a precipitation gradient in an inter-Andean valley in
Colombia. Yearly precipitation in the Magdalena Valley ranges from 1000 mm to 5000 mm, which has previously been shown to influence plant distribution. I gathered data on species distributions, abundance and morphological traits to answer four specific questions: (1) What are the relationships between community and climatic similarity? (2) How does the moisture gradient affect the rates of taxonomic, phylogenetic, and morphological turnover of bird communities? (3) Do the phylogenetic relationships of the species within a locality reflect neutral or deterministic mechanisms? And (4) Is there any evidence of reduced immigration of species from the species pool into dry forest localities? I begin by suggesting a new method for estimating abundance from count data, a crucial step for studies in community ecology.
Further, I demonstrate that there is a steep turnover of community composition at the limit of the dry forest. The taxonomic turnover is steeper than the phylogenetic turnover, suggesting
9 that replacement of closely related species accounts for a disproportionate number of changes along the gradient. We found evidence for environmental filtering in dry forest as species tend to be more tolerant of higher temperature ranges, stronger rainfall seasonality and lower minimum rainfall. On the other hand, wet forest species tend to compete actively for nest space but not for the resources associated with the axes we measured. Wet forest species are less able to immigrate into dry forests than the reverse. Our results suggest that rainfall is a strong determinant of community composition when comparing localities above and below the 2400 mm rainfall isocline. The strong non-random assembly of dry forests and wet forests suggest that they are part of two independent metacommunities with little immigration among them.
10 CHAPTER 1 INTRODUCTION In the last decade, there has been a renewed interest in explaining diversity patterns and predicting the relative abundance of species (Simberloff, 2004; Vellend, 2010; Weiher et al., 2011). This revival has also promoted an intense debate about which mechanisms drive community structure (Clark, 2009). On one side of this debate, neutral theory proposes that species randomly partition the niche space available as theycolonizeeachhabitat,resultingin species-abundance distributions that can only be attributed to random (e. g., ecological drift) or purely biogeographical processes (e.g., speciation and/orextinction;Hubbell, 2001). On the other side of the debate, niche theory suggests that populations of species interact with each other and with the environment to give rise to observed patterns of community structure (Cody & Diamond, 1975). Different types of interactions (e.g. competition, facilitation, environmental filtering) should have different influences on the number and identity of the species present in an environment, but each of these mechanisms should yield predictable, repeatable patterns of community structure (Terborgh, 1977; Graves & Gotelli, 1993; Weiher & Keddy, 1995). Afirststepfordeterminingthemechanismsthatdrivecommunity assembly is to explore the shape of the species-abundance distributions (SADs) of the localities being studied. There has been a long interest in ecology in trying to predict the relative abundance of species (Fisher et al., 1943; Preston, 1948; Hubbell, 2001). Generally, SADs follow a log-normal distribution (Preston, 1948)ormoregenerallyanewstatisticaldistributioncalledtheZero-sum
Multinomial distribution (Hubbell, 2001)thatdiffers from the lognormal by having a fat tail in the low abundance class end of the distribution. This distribution was central to the Unified Neutral Theory of Biodiversity and Biogegraphy (UNTB; Hubbell, 2001), which assumes that all individuals in the community are demographically equivalent and the probability of immigration to a local community depends on the relative abundance of the species in the regional pool. The UNTB therefore assumes that interactionsamongspeciesandwiththeir
11 environment are not an important determinant of the relativeabundanceofspecieswithina community. The UNTB has been successful in predicting SADs of highly diverse assemblages such as tree communities in the Barro Colorado Island in Panama andseveralothertropicalplant
assemblages from around the world (Hubbell, 2001). Since the publication of the UNTB, researchers have criticized this simplistic way of predicting relative species abundance and have found alternative models that also predict SADs using more mechanistic approaches (Jabot & Chave, 2011). One of the ways to determine the mechanisms underlying community assembly, for example, is to fit a set of models that start with a completely neutral base and
modify them to include parameters that introduce more deterministic mechanisms (Munoz et al., 2007; Jabot et al., 2008; Jabot & Chave, 2011; Chisholm & Pacala, 2010). These deterministic models are almost always related to niche-based mechanisms (mainly competition and habitat filtering) that would better explain the variation of species abundances. Fitting
different models to predict SADs is not always the only solution particularly because some of the mechanistic models predict observed SADs just as well as neutral models, which makes it impossible to differentiate between neutral and niche-based mechanisms (Jabot & Chave, 2011). Additionally, neutral models can produce false positivesinwhichthemodelspredict accurately the SADs but alternative methods then reveal the actual mechanisms underlying community structure (Harpole & Tilman, 2006; Chisholm & Pacala, 2010). To address these complications, alternative approaches have been proposed to address the question of how species diversity arises. As I have highlighted above, much of the research has traditionally focused on finding common patterns in species-abundance distributions
and also in understanding the turnover of species diversity along environmental gradients (Cody, 1974; Cody & Diamond, 1975; Terborgh, 1977; Connor & Simberloff, 1978; Graves &Gotelli, 1993). Species interact with the environment and with other species through their functional traits (McGill et al., 2006), which are the result of an evolutionary history
that can also be responsible for the diversity patterns observed. Thus, our understanding of
12 how biological diversity arises and is maintained can benefitfrommethodsthatspecifically incorporate the distribution of functional traits within and among assemblages, in addition to the evolutionary history of the species including a phylogenetic measure of biodiversity (Webb et al., 2002; McGill et al., 2006). Many recent studies address questions about
community ecology using phylogenetic and functional trait information, which might allow us to disentangle the mechanisms that drive species diversity (Cavender-Bares et al., 2009). Many other studies have used mathematical models in parallelwithfunctionalmorphologyto explain this diversity patterns. Most of these studies have focused on plants (Vamosi et al., 2009), raising questions about how the findings of such studies areapplicableandcomparable to other organisms. Even though birds have been the model organisms for much of traditional community ecology research (MacArthur, 1958; Cody, 1974), they have scarcely been studied using an approach that incorporates modeling of SADs, functional traits and evolutionary history. Avian ecologists were among the first to use proxies for the degree of evolutionary
and phenotypic relatedness among species to address questions about community assembly (Connor & Simberloff, 1978; Grant & Abbott, 1980). Only a few recent bird studies have taken advantage of trait and/or phylogenetic information (Lovette & Hochachka, 2006; Graham et al., 2009; Vamosi et al., 2009; Gomez et al., 2010; Weinstein et al., 2014).
Conflicts arise among interpretations of the mechanisms underlying SADs mainly because of two issues: (1) different processes might result in similar patterns (i.e. SAD; Rosindell et al., 2012)and(2)thestochasticcomponentsofthemodelsdonotnecessarily mean that neutral forces are acting in a system (Clark, 2009, 2012). Both of these issues can be resolved by collecting additional data that help us differentiate among the processes that generate
the same pattern and at the same time reduce the uncertainty about the system (Rosindell et al., 2012). As more information is available about a particular system, the proportion of the variation explained by a particular model might move from thestochastictothedeterministic sides of the debate (Clark, 2009). The process of gaining more information potentially can
demonstrate that the previously considered neutral mechanisms are in reality the product of the
13 amount of ignorance that we had about a specific system (Clark, 2009, 2012; Rosindell et al., 2012). To get around the problems created by the difficulty of determining which mechanisms generate observed SADs, alternative approaches have been proposed to tackle the problem of how species diversity arises. To date, much of the research has focused on finding common patterns in species-abundance distributions, especially along environmental gradients (Cody, 1974; Terborgh, 1977; Connor & Simberloff, 1978; Graves & Gotelli, 1993). This approach, however, may have its own limitations because it ignores an additional set of mechanisms that may generate community patterns. Species interact withtheenvironmentandwith
other species through their functional traits (McGill et al., 2006), which are the result of an evolutionary history that can also be responsible for the diversity patterns observed. These patterns in present time can be the reflection of a history of evolution and diversification that promote coexistence by the differentiation between similar species and the adaptation of the
species to the environment (MacArthur, 1958). These mechanisms (competition and habitat filtering), might be hidden in current patterns, but become visible when studying the assembly process on an evolutionary timescale (Emerson & Gillespie, 2008). Thus, our understanding of how biological diversity arises and is maintained can benefit from methods that study
present patterns but also incorporate the distribution of functional traits within and among assemblages, in addition to the evolutionary history of the species including a phylogenetic measure of biodiversity (Webb et al., 2002; McGill et al., 2006; Emerson & Gillespie, 2008). Many recent studies address questions about community ecology using phylogenetic and functional trait information, which might allow us to disentangle the mechanisms that
drive species diversity (Cavender-Bares et al., 2009). Most of the evidence suggests that competition acts mainly at local scales and habitat filteringatlargeregionalscalesandthat habitat filtering is more likely to be detected in studies of the mechanisms driving community assembly (Vamosi et al., 2009; HilleRisLambers et al., 2012).Even though an increasing number
of studies focus on organisms other than plants, there is still an imbalance in favor of plants
14 as model organisms for these types of studies (e.g. Ecology 2012, Volume 93, Special Issue in August). The problem with this potential imbalance is thatthemechanismsunderlying community assembly might influence mobile organisms in different ways than they do for sessile organisms such as plants. For example the effect of dispersal limitation in continuous tropical forests might be stronger in plants than it might be in more mobile organisms such as birds (Hubbell, 2005; Van Houtan et al., 2007)althoughindiscontinuousforestpatchesthepattern might be the opposite (Bacles et al., 2006; Moore et al., 2008). In fact, plant dispersal and recruitment patterns often depend on the movement patterns of mobile organisms such as birds and mammals, but are also influenced by their specific interactions with other plants and the environment, increasing the complexity of interpreting patterns (Clark, 2012). Furthermore, mobile organisms have the options of modifying their behavior to overcome specific challenges and temporarily shape patterns of community structure (McGill et al., 2006). Finally, there are potentially fewer axes in which plants can partition the niche space (Vamosi et al., 2009).
These differences in the response to potential mechanisms of community assembly between plants and mobile organisms raise the question about how the findings of such studies are applicable and comparable to other organisms. Recently, bird community ecologists have turned their attention to questions about the distribution and coexistence of birds using phylogenetic and trait-based approaches (Lovette & Hochachka, 2006; Graham et al., 2009; Vamosi et al., 2009; Gomez et al., 2010; Parra et al., 2010, 2011; Gonz´alez-Caro et al., 2012; Graham et al., 2012). Similar to the evidence gathered in plant studies, bird communities have been shown to be strongly structured by habitat filtering, but in both local and regional scales. Sometimes the patterns are equivocal because of the complicated biogeographical history of birdsanddifferences in rates of evolution between clades (Parra et al., 2010, 2011; Graham et al., 2012). Interpretations of the patterns of phylogenetic relationships among species within communities, some studies show that the strength of environmental filtering in lowland hummingbird communities is stronger in dry forests than in wet forests (Graham et al., 2009; Parra et al., 2011). The problem is, that even
15 though interpretations of phylogenetic diversity patternsallowustoidentifythemechanisms responsible for diversity patterns, they do not pinpoint theparticularwaybywhichthe mechanism is operating. Therefore, the mechanisms by which environmental filtering operate more strongly in dry and highly seasonal forest and relaxed inwetforestsarestillunknown
(Parra et al., 2010). Specific information that broadens the taxonomic representation of studies paired with an ample sampling of potential functional traitsthatcanbeunderselectionby environmental variables should help us identify underlyingmechanisms. Studies in community ecology that use phylogenetic trait-based approaches can improve their power to detect mechanisms underlying community structure by focusing
on environmental gradients. Such gradients might serve as natural experiments that allow tests of different factors that might shape community structure if some variables such as temperature are held constant while others (e.g. precipitation) vary (Graham & Fine, 2008). This approach has been previously reported to be useful at thecommunitylevelbecauseit
can also differentiate between mechanisms that operate at local and regional scales (Ackerly & Cornwell, 2007; Cornwell & Ackerly, 2009; Graham et al., 2009; Cornwell & Ackerly, 2010). Environmental gradients might also influence the evolution of species through adaption to different environmental conditions, adding species to the regional source pool and influencing
the ways by which species might assemble in a community. For these reasons it is useful to take advantage of phylogenetic approaches and focus on both the community and the population levels to strengthen our understanding of the mechanisms that drive changes in community composition along gradients (Smith et al., 1997). In Colombia, the inter-Andean valleys, especially the Magdalena River valley, have been
considered regions of high biological relevance and conservation importance (Olson et al., 2001). The Magdalena River Valley is characterized by a marked precipitation gradient with lowest precipitation in the southern region (∼ 1000 mm annually) and highest levels in the northern part (∼ 5000 mm). The Magdalena valley forests also support populations of at least seven Colombian endemic bird species, four of which areendangered(White-mantled
16 Barbet, Tolima Dove, Antioquia Bristle-Tyrant and Blue-billed Curassow) and one that is nearly threatened (Commission et al., 2010), which greatly adds to the urgency of studying birds in this region. Additionally, almost 500 species have been reported for the area, which allows us to ask questions about the role of the habitat isolation and ecotone in shaping and maintaining these high levels of diversity (Hilty & Brown, 1986). The goals of this study are to understand the mechanisms that drive species diversity by testing the role of neutral and niche-based processes organizing bird communities along the moisture gradient in the Magdalena Valley of Colombia. Because different mechanisms are likely to act at different spatial scales, I will focus on just a few mechanisms mainly regarding the effects of abiotic variables in community organization (Swenson et al., 2006). Habitat filtering, a niche-based mechanism, has been shown to influence species distributionsanddiversitypatterns(Weiher & Keddy, 1995; Keddy & Weiher, 1999). Neutral processes, however, also predict species diversity patterns in highly diverse communities such as tropical forests (Hubbell, 2001; Chisholm &
Pacala, 2010; Rosindell et al., 2012). This dichotomy about the relative roles of the niche (environment) versus neutral mechanisms provides a framework to evaluate the importance of abiotic variables in community assembly. Neutral theory hypothesizes that community structure is the result of random colonization, speciation and extinction of individuals within communities and that the differences among species are not relevant to their abundances (Hubbell, 2001; Rosindell et al., 2012). Alternatively, Niche Theory hypothesizes that the species are adapted to a particular set of abiotic variables (Grinellian Niche) and that the relative abundance of a species is the reflection of its fitness in a particular environmental space (Cody & Diamond, 1975; Sokol et al., 2011). I am interested in understanding the portion of the variance among communities that might be explained by precipitation alone.Theoverarchingquestionofmy thesis is: to what extent does precipitation explain the differences in relative abundance and turnover of communities along the precipitation gradient?
17 CHAPTER 2 AN EFFICIENT EXTENSION OF N-MIXTURE MODELS FOR MULTI-SPECIES ABUNDANCE ESTIMATION 2.1 Background
One of the most common complications that ecologists face when estimating abundances
of mobile organisms is that individuals and species differ in their detection probability. Such differences results in the under or overestimation of real abundance when the detection probability is ignored (MacKenzie et al., 2002; Martin et al., 2005; Royle & Dorazio, 2008). As a result, many quantitative ecologists have proposed statistical methods to estimate detection probability and correct the observed individual counts to estimate either density or abundance
(Denes et al., 2015). N-mixture models are a family of hierarchical models in whichthecountsofspeciesy are binomially distributed with N being the total number of individuals available for detection and p the probability of detecting an individual of that species (Royle, 2004). Because N is not
known, it is considered to be a latent variable that is a product of some discrete distribution such as a Poisson. Inferences about the abundance of species therefore relies on estimating the detection probability and the underlying parameter of the distribution giving rise to N (Royle, 2004). Even though their use has been widely advocated, very few examples exist of the use of N-mixture models for estimation of the abundance of neotropical bird species. Most of the density estimation for bird populations in the Neotropics comes from sampling of 100-ha plots using intensive field methods such as spot mapping or repeatedmist-netting(Terborgh et al., 1990; Thiollay, 1994; Robinson et al., 2000; Blake, 2007). A probable reason for the sparse use of N-mixture models to estimate densities of neotropicalbirdsisthewelldocumented
species abundance distributions of tropical organisms, which have a long tail to the right with very few abundant species and many rare ones (Hubbell, 2001). This is reflected in Parker III et al. (1996)database,inwhichtheyconsiderabirdtobecommonintheNeotropics if its population abundance is higher than 15 ind/100 ha. This high proportion of rare species in
the overall community makes it difficult to obtain enough detections during field censuses
18 for appropriate estimation of both abundance and detection probability for many, if not the majority of neotropical bird species. To ensure independence of point counts used to estimate the abundance of neotropical bird species, researchers have suggested that points must beatleast200mapartandthe radius of the point count cannot be larger than 50 m (Ralph et al., 1993, 1995; Matsuoka et al., 2014). If the goal is to estimate the abundance of all species on a 100 ha plot (minimum area suggested to correctly describe a lowland local bird community in the neotropics; Terborgh et al., 1990), considering the restrictions described above, the maximum number of points that fit in a 100 ha plot is 36 points of 0.78 ha each, based on a radius of 50m. Because of the excess of rarity in tropical birds, the majority of species will have abundances below 15 ind/100 ha. Assuming that individuals are homogeneously distributed
individuals across the plot, the expected number of individuals in each point count is λ ≈ 0.12 point count . This value of λ is well below the λ =2used to evaluate the performance of the model (Royle,
2004). Thus, we do not know how the model performs when λ is low, a common scenario in neotropical birds. In addition, several neotropical species are known to be secretive and therefore have low detection probabilities, which imposes stronger challenges for estimating their abundance. Our first objective in this study was to determine the minimum sample size required to estimate the abundance of neotropical bird species using N-mixture models. We believe that this objective will be particularly useful for population ecologists whose goal is to obtain a rough estimate of the density of a species without using one of the field-intensive methods such as spot-mapping. Asecondgoalofthispaperistodevelopamethodtoestimatethe abundance of all of the species present in a community to make inferences aboutmechanismsdrivingspecies abundance distributions, an important issue in the current debates over the Unified theory (Hubbell, 2001; McGill et al., 2007). While performing point counts an observer can easily count all of the individual birds in the area for a particular amount of time irrespective of the identity of the species. Thus, the actual data will have information about all of the species
19 present in the area and an approximation of their abundance. Because of their behavior, foraging strategy and evolutionary relationships, some, ifnotmostofthespeciesinthe community can have correlated detection probabilities. Such correlations in detectabilities potentially allow us to lump counts of species to increase theinformationavailableon abundance corrected by detection probability. Thus, the second objective of this study is to expand the N-mixture models to a scenario in which we can useinformationfrommultiple species to estimate the parameters of a detection probability distribution of a set of species, and use such probability this distribution to estimate the expected abundance per unit area of each of the species in the set. 2.1.1 The Model
In the following section, after summarizing the widely used N-mixture models, we develop a parsimonious, multi-species model extension that allows a more accurate estimation of the abundance of rare species. The essential contribution of ourapproachistheuseofinformation from the counts of ecologically similar species to improve the estimation of both detectability and abundance.
Using an N-mixture model, we usually let yij be the number of individuals for a given species in the i − th sampling unit (a point count) and j − th replicate of the sampling unit (or visit to the point count). Let p be the individual detection probability for that species. Finally, let ni be the fixed number of individuals available for detection in the i − th sampling unit. If we assume that the counts are binomially distributed, the likelihood of the counts for a given species is
r t n i yij ni −yij L(yij ; ni , p)= p (1 − p) . "yij # !i=1 !j=i for i =1,2,3...r and j =1,2,3...t,wherer is the total number of point counts sampled and t is the number of times each point count was visited (Royle, 2004).
The N-mixture model assumes that the number of individuals available for detection is in fact unknown and random. Thus, such a number is considered to be a latent variable, modeled
20 with a Poisson process with mean λ (the mean number of individuals per sampling unit). From
here on, we write Ni ∼ Pois (λ),wherewehaveusedtheconventionthatlowercaseletters such as ni denote a particular realization of the (capitalized) randomvariableNi .Tocompute the likelihood function, one then has to integrate (sum, in this case) the binomial likelihood over all the possible realizations of the Poisson process,
r ∞ t − N e λ Ni i yij Ni −yij λ L(yij ; λ, p)= p (1 − p) , (2–1) "yij # Ni ! !i=1 Ni =max(y$ i) !j=1
where yi = {yi1, yi2,...,yit}.IftheobjectiveistoestimatetheabundanceofS species, the overall likelihood is simply written as the product of all theindividualspecies’likelihoods,i.e.,
S r ∞ t − N e λs Nsi si ysij Nsi −ysij λs L(ysij ; λ, p)= ps (1 − ps ) , (2–2) "ysij # Nsi ! !s=1 !i=1 Nsi =max(y$ si) !j=1
where ysij is a three dimensional array of dimensions r × t × S,andbothλ = {λ1,...,λS }
and p = {p1,...,pS } are vectors of length S.Writingthelikelihoodinthiswaydirectly implies that in order to estimate the abundance of all the species present in a community, one would need to estimate 2 × S parameters (S mean number of individuals λs plus S detection probabilities ps ). To avoid the proliferation of parameters one could assume that all the ps come from a single probability model that describes the community-wide distribution of detection probabilities. These community-wide detection probabilities, for example, can be modeled with a beta distribution in which we let Ps ∼ Beta(α, β).Theprobabilitydensity
Γ(α+β) α−1 β−1 function of the random detection probabilities is then g(ps ; α, β)= Γ(α)Γ(β) ps (1 − ps ) . The overall likelihood function now integrates over all the realizations of the community-wide detection probabilities Ps :
21 S r ∞ t − 1 N e λs Nsi si ysij Nsi −ysij λs L(ysij ; λ, α, β)= ps (1 − ps ) %0 "ysij # Nsi ! !s=1 !i=1 Nsi =max(y$ si) !j=1
Γ(α + β) × pα−1(1 − p )β−1dp . Γ(α)Γ(β) s s s (2–3)
The difference of the former and the latter forms of the N-mixture model is that in the latter you need S +2parameters to estimate the abundance of the full community instead of 2 × S.
In large communities, this might be a significant decrease of parameters. The usefulness of specifying the likelihood is that in the case in which many species are rare, we can use the information on the abundant species to estimate the detection probability, leaving the actual counts to estimate only the abundance of the species. 2.1.2 Maximum Likelihood Estimation
One drawback of the beta-N-mixture model is its computational complexity. To date, many numerical approximations for obtaining the Maximum Likelihood Estimates (MLEs) for hierarchical models have been proposed (de Valpine, 2012). Of these, the so-called ”Data Cloning” methodology has proven to be a reliable approach to not only obtain the MLEs for these types of models, but also for hypothesis testing andmodelselection,aswellas unequivocally measuring the estimability of parameters (Lele et al., 2010; Ponciano et al.,
2009). The method proposed by Lele et al. (2007, 2010)usestheBayesiancomputational approach coupled with Monte Carlo Markov Chain (MCMC) to compute Maximum Likelihood Estimates (MLE) of parameters of hierarchical models and their asymptotic variance estimates (Lele et al., 2007). The advantage of using the data cloning protocol is that there is no need to find the exact or numerical solution to the likelihood function of the hierarchical model in order to find the MLE. Instead, one only needs to compute means and variances of certain posterior distributions.
22 Data Cloning proceeds by performing a typical Bayesian analysis on a dataset that consists of k copies of the originally observed data set. In other words, toimplementthismethod,one has to write the likelihood function of the data as if by pure happenstance, one had observed k identical copies of the data set at hand. Then, Lele et al. (2007, 2010)showthatask
grows large, the mean of the resulting posterior distribution converges to the MLE. In addition, for continuous parameters as λ, α,andβ,thevariancecovariancematrixoftheposterior
1 distribution converges to k times the inverse of the observed Fisher’s information matrix. In this way, the variance estimated by the posterior distribution can be used to calculate Wald-type confidence intervals of the parameters (Lele et al., 2007, 2010). The advantage
of data cloning over traditional Bayesian algorithms is thatwhileinBayesianalgorithmsthe prior distribution might have some influence over the posterior distribution, in data cloning the choice of the prior distribution does not determine the resulting estimates because these are the MLEs. In our case, the hierarchical model is of the form
Y ∼ Binomial (N, P) = f (y|N = n, P = p)(Observation model),
N ∼ Pois (λ) = g(N; λ)(Process model),
P ∼ Beta (α, β) = h(P; α, β)(Process model).
N and P are unobserved quantities or latent variables which are products of a stochastic process given by the Poisson and Beta distributions respectively. Furthermore, the parameters left to be estimated (i.e., λ, α, β)areseenasrandomvariablesthemselvesthathavea posterior distribution π(λ, α, β|Y). A typical Bayesian approach would sample from the following posterior distribution:
π(λ, α, β, N, P|Y) ∝ [f (y|N = n, P = p)g(N; λ)h(P; α, β)] π(λ, α, β), where π(λ, α, β) is the joint prior of the model parameters. This approach would yield many samples of the vector (λ, α, β, N, P) and in order to sample from the marginal posterior
π(λ, α, β|Y) one only needs to look at the samples of the subset of parameters (λ, α, β).
23 The data cloning approach proceeds similarly, except one needs to sample from the following posterior distribution:
π(λ, α, β, N, P|Y)(k) ∝ [f (y|N = n, P = p)g(N; λ)h(P; α, β)]k π(λ, α, β).
The notation (k) on the left side of this equation does not denote an exponent and is only there to denote the number of times the data set was “cloned”. On the right hand side, however, k is an exponent of the likelihood function. The MLEs of λ, α, β are then simply obtained as the empirical average of the posterior distribution π(λ, α, β|Y)(k) and the variance of the estimates
1 are given by k times the variance of this posterior distribution. For a detailed example that illustrates the calculations with posterior and cloned posterior distributions that are analytical and tractable, and where the MLEs can be easily computed, we refer the reader to Ponciano et al. (2012).
2.1.3 Scenarios In Which ps Are Correlated Among Neotropical Bird Species
Several scenarios can arise in which ps are correlated among species. It is widely known that the probability of detecting diurnal species such as birds that defend territories by singing
is highest at or right after dawn and decreases with time of day(Blake, 1992). Also, different types of forests have differences in structural characteristics that allow or hinder the detection of all of the individuals available. Thus, species sharing a microhabitat or even inhabiting a particular ecosystem should have similar detection probabilities. One common phenomenon in the Neotropics (and probably in other regions) is the formation of mixed-species flocks (Munn
&Terborgh, 1979). These flocks are formed by individuals of different species that forage together. Thus the overall detection of the species in the flock is correlated because once you detect one species, you are likely to detect the rest of the species within the flock. Finally, several foraging behaviors and vocal activity patterns makespeciesparticularlyeasyordifficult
to detect. Species that forage using a sit and wait behavior are usually much more difficult to detect than species that forage by gleaning on leaves actively searching for food. Although there are some species that their high vocalization rate makethemeasiertodetectirrespective
24 of their foraging guild. Thus, you can think of species that are sit and wait foragers but are either easy or difficult to detect (e.g. Monasa vs. Malacoptila,puffbirds). 2.2 Methods
2.2.1 Sample Size Estimation for Neotropical Birds
To determine the minimum sample size required for accurate estimation of the abundance
of neotropical species, we used a series of simulations in which we varied the number of points (r), visits to points (t), mean number of individuals in each point (λ)and detection probability (p). We varied r between 5 and 50, t between 2 and 20, λ = 1, 2, 3, 4, 5, 7, 10, 15, 25, 40, 55, 65, 75, 85, 100 and p between 0.1 and 0.9. For each
combination of parameters, we simulated 170 data sets and estimated λ and p using equation 1 for each of the 170 datasets and each of the parameter combinations. In each simulation, we λˆ−λ ˆ computed the relative bias of the abundance estimate by using, bias = λ ,whereλ is the MLE for a particular data set and λ is the true value of the parameter used to simulate the data. Finally, we retained the mean bias for each combinationofparameters.Wecouldnot retain the full distribution of the bias because of the large number of simulations performed (10.935.000). We considered an acceptable bias to be lower than 0.1, which is a 10 percent difference between the estimate and the true population density. All of the simulations were performed using R statistical software v.3.0.2 (RCoreTeam, 2013)andmaximum
likelihood estimation by maximizing the likelihood of eq (1)usingtheoptimfunctionwiththe Nelder-Mead algorithm. The R code used for simulations and maximum likelihood estimation is presented in the Appendix C. 2.2.2 The Beta N-mixture Model
Because 100-ha plots have become the standard for estimatingabundancesofneotropical birds (Terborgh et al., 1990), we developed our example of the use of the beta n-mixture model using a sampling scenario in 100-ha plot. Assuming thataconservativedistancebetween
point counts required for the points to be independent is 200 mweselectedthemaximum number of points that fit in a square 100 ha plot given this requirement. For this example
25 we used 36 point counts and 5 visits, which is reasonable enough to still be part of a rapid inventory but also has enough information to estimate the abundances of rare species. We simulated 1000 data sets that consisted of 15 species with thesameλ values described in the previous simulation and three sets of parameters of the beta distribution. The sets of parameters where α =10,27,30and β =30,27,10,whichaccountforscenariosoflow,mid and high detection probability with the same variance (E[p]=0.25,0.5,0.75;Var[p]=0.004). For each of the simulated data sets we estimated λ and p under the N-mixture model and λ, α and β under the beta N-mixture model. Then, we computed the bias in λ in the same way as presented above. We performed maximum likelihoodestimationoftheparameters
under the N-mixture model by optimizing equation 1, with the optim function in R using the Nelder-Mead algorithm. To estimate the parameters under theBetaN-mixturemodel,wealso used maximum Likelihood estimation but using Data Cloning (Lele et al., 2007). We used the rjags (Plummer, 2014)interfaceforRtobuildthemodelsandruntheanalysiswith2 chains, with 20000 iterations in each chain and retained the parameter values every 100 generations after a burn-in period of 1000 generations. 2.2.3 Example Using Real Data
Finally, we used a data set that consisted of 94 point counts, located in three dry forest patches in Colombia. Each point count was replicated three times from January 2013 to July 2014. From this data set, we selected the understory insectivore species that forage over foliage (Karr et al., 1990; Parker III et al., 1996)tomeettherequirementoftheBetan-mixturemodel
of correlated detection probabilities among species. In total, we estimated the abundance of 26 species using both the n-mixture and Beta n-mixture models and compared their fit using AIC following the protocol presented in Ponciano et al. (2009). We are aware that it is likely that the closed population assumption for this data set does not necessarily hold, but
it is unlikely that populations of species have changed drastically from one year to another during these years. The point counts were performed in three different forest patches in the upper Magdalena valley in Central Colombia. To maximize the sample size for abundance
26 estimation, we lumped the point counts into a single data set,suchthattheinferencesof species abundances are made for the entire region instead of the particular patch. The three forest patches were separated by less than 150 km and were located within the Magdalena valley dry forest ecoregion (Olson et al., 2001). Because they are in the same ecoregion, the structural variables of the forest are similar and thus itisunlikelythatthedetection probabilities vary among patches as well as the abundance of species, allowing us to lump the data together. Maximum likelihood estimation for the n-mixture and beta-mixture models where performed in the same way as described in the previous section. The R code and jags models used for model selection using AIC are presented in the Appendix C 2.3 Results
2.3.1 Sample Size Estimation for Neotropical Birds
We found that the required minimum sample size needed to accurately estimate the abundance of neotropical bird species decreased with increasing both λ and p (Figure 2-1). For the sample sizes evaluated, there is no combination of point counts and replicates that allows the estimation of abundances with less than 7 individuals/100ha using n-mixture models (Figure A-1). In the 7 ind/100 ha threshold, the effort required is very high. For example, for species with a probability of detection of 0.5 the required sample size to obtain a bias lower than 0.1 is around 50 points and more than 6 replicates of each point count or around 40 point counts with more than 10 replicates (Figure 2-1,A-1). As λ increases the sample size required to estimate appropriately the abundance of species decreases. 2.3.2 The Beta N-mixture Model
The Beta n-mixture model performed better than the regular n-mixture model for the data simulated. In both cases, as λ increased, the median of the distribution of λˆ converged to the true value of λ (Figure 2-2). Such results allow us to conclude that using the regular n-mixture model, the minimum abundance that the model is abletoestimateisaround
10 individuals/100 ha, with a sample size of 36 point counts replicated each five times (Figure 2-2). The use of multi-species information to estimate the abundance of single
27 h eanmxuemdl h minimum the model, densitie n-mixture lower beta estimate the to model the of ability the improved niiul/0 aadpsil niiul/0 ha. individuals/100 3 possibly and ha individuals/100 eeto rbblt sa es . (Figure 0.5 least at is probability detection (Figure distribution abundance species the of variance the decreased it 1) results: two mainly had species ha 100 per individuals of number mean in bias Mean 2-1. Figure vrsiaetedniyo aeseis(Figure species rare of density the overestimate vntog the though Even lc iei aho h aes h eut o h niese entire in the for figure results similar The a panels. in isocl the which of presented 0.1 each select of in We abundance line right. of black the estimation in in (ligh presented bias low is acceptable from scale bias color the The represent (black). panel each in grayscale The ubro elcts n reprmtrvle ofrmdlo mid for ( to probabilities values detection parameter and true abundances and replicates, of number
Number of Point Counts λ ˆ 5 15 25 35 45 5 15 25 35 45 5 15 25 35 45 5 15 25 35 45 λ λ λ λ itiuinhshg aine ayotir n,i tends it and, outliers many variance, high has distribution = = = = 100 65 25 7 01 20 15 10 5 p =0.2 Number ofReplicates λ 01 20 15 10 5 A httemdli bet siaei rud4 around is estimate to able is model the that 2-3 p =0.5 2-3 .Ti snttu o h -itr model, n-mixture the for true not is This ). 28 2-2 .Ti sepcal rewe h mean the when true especially is This ). , λ A-3 =7,25,65,100 , A-2 01 20 15 10 5 sw i t hs i m i l a rs a m p l es i z e s .U s i n g ntedsrbto of distribution the in ,i anan h structure the maintains it ), p =0.8 λ o ag fpitcounts, point of range a for and datrsodfor threshold a ed tofsimulationsare
tgray)tohigh wandhigh n speetda a as presented is ine 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 >10 p bias in λ =0.2,0.5,0.8 λ ˆ to n )it 2) and ). lotcnee ntetu au of value true the in centered almost (Figure probability detection community’s of value estimated the in Bias 2-2. Figure bnatseis(Figure overestimat species and abundant species mid-range more many predicts which (Figure eesmltd(Figure simulated were di the across similar h eaNmxuemdlas efrswl netmtn th estimating in well performs also model N-mixture beta The 2-4 .Hwvr h oe vrsiae Var overestimates model the However, ). eeto rbblte n 6pitcut elctdeac replicated counts point di 36 three and under probabilities data detection of simulations 1000 for ff
rn ye fsmltos hc refl which simulations, of types erent ^ Bias in Estimated Abundance (λ) 2-4 0 2000 4000 6000 0 2000 4000 6000 0 2000 4000 6000 2-3 ). 04 75 40 10 4 1 ). mxueMdlBetamixtureModel Nmixture Model p .T h e r ei sas l i g h tu n d e r e s t i m a t i o no f True Abundance ( 2-4 λ o ohteNmxueadbt -itr model N-mixture beta and N-mixture the both for .Tedsrbto fE of distribution The ). 29
0 5 10 15 20 0 10 20 30 40 50 0 20 40 60 80 100 120 140 [ 04 75 40 10 4 1 p ] λ ff ,b u ti te s t i m a t e st h ev a r i a n c et ob e ) rn scenarios erent csteraiywt hc h data the which with reality the ects sb a h bnac of abundance the far by es edistributionofthe hfivetimes. [ p ] flw i n high and mid low, of o h iuain is simulations the for p = 0.75 p = 0.5 p = 0.25 p when p =0.25 ewe h oe a xrml ihgvn eysrn supp strong very giving high extremely was model the between ( n oe ob h eaNmxue following N-mixture, beta the be to 2 model and t in bird E insectivorous an of probability N-mixture detection beta mean the the For model. N-mixture the using estimated iue23 re(lc)adec fteetmtdseisab species estimated the of each and (black) True 2-3. Figure .. xml sn elData Real Using Example 2.3.3 ∆ [ p AIC ]=0.2 epeetteetmtsof estimates the present We =14725.04 h 00smltoso aaudrtredi three under probabilities. data detection of simulations 1000 the with Var ).
[ ^
p Log(Abundance (λ,λ)) ]=0.009 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10 mxueMdlBetamixtureModel Nmixture Model λ ( ˆ α o ohmdl nTable in models both for =3.15 Species Rank , β 30 =12.7
Ponciano 0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10 .Asmn oe ob h N-mixture the be to 1 model Assuming ). ff rn scenarios erent euprMgaeaVle a of was Valley Magdalena upper he tal. et 2-1 nac ga)dsrbtosfrom distributions (gray) undance r nfvro h eamodel beta the of favor in ort ae h oe siae that estimates model the case, ,aswellas ( 2009 ,tedi the ), flw i n high and mid low, of
p = 0.75 p = 0.5 p = 0.25 p ˆ o ahspecies each for ff rnein erence AIC iue24 itiuino xetdvle(E value Expected of Distribution 2-4. Figure niiul e ntae.Bcueti iuainwsper was simulation this Because area. unit per individuals 2 a ayfo n e fseist nte rfo n habitat one from or another to species of set linked one is from but vary probability can detection own its has species each p pce ihls hn7idh n o eeto probabili detection low a and ind/ha 7 are than models less the with that species is estimate result to interesting An point each birds. of tropical replicates of number the and counts htaedtce e ie.I hscs,ee ihhg samp high with high even very case, a this estimates In par times. is few model detected beta are The that species. rare informati of the abundance uses the model, Our estimate replicates. 20 and points 50 of farr pce srsrce ob orltdto correlated be to restricted is species rare a of -itr oescnsilb sdi h ape rai incr is area sampled the if used be still can models N-mixture efudta -itr oesrqiehg apeszsi b in sizes sample high require models N-mixture that found We coste10 iuain efre ne cnro flo of scenarios under performed simulations 1000 the across n h aevrac Var variance same the and
Expected detection probability (Ε[p]) 0.0 0.2 0.4 0.6 0.8 1.0 λ .5050.75 0.5 0.25 n eylow very and True meandetectionprobability(p) p [ .T h i sp r o b l e mi ss o l v e dw i t ht h eb e t am o d e l ,i nw h i c h . Discussion 2.4 p ] . [ p 31 ] )andvariance(Var
p Variance of detection probability (Var[p])
feooial iia pce.I hsway, this In species. similar ecologically of 0.00 0.02 0.04 0.06 0.08 0.10 .5050.75 0.5 0.25 nbet cuaeyetmt rare estimate accurately to unable hog nudryn rcs that process underlying an through omdfrtoia oet,we forests, tropical for formed y( ty ncletdfrasto pce to species of set a for collected on eszs h -itr model N-mixture the sizes, le iual sflfrrr species rare for useful ticularly cuaeyteaudneof abundance the accurately < oanother. to ae oraise to eased 0.5 [ t h ubro point of number the oth p ] ,a es ihsamples with least at ), )i nd e t e c t i o np r o b a b i l i t y ,mdadhg E high and mid w, λ oaround to [ p ] simulated point counts with a 50-m radius. This distance has been proposed to meet the assumption that the detection probability is homogeneous across the whole sampling area and to increase the detection probability of species within it. Other methods have dealt with relaxing this assumption (e.g. distance sampling Buckland et al., 2005), and recent studies have even suggested methods to perform estimation in a multi-species fashion (Dorazio & Royle, 2005; Dorazio et al., 2015; Sollmann et al., 2015; Yamaura et al., 2011). Our objective, however, was to evaluate the N-mixture model for fixed-radiuspointcounts,amethodthat is commonly used for surveying birds. Thus, the only other solution to increase the area is to discourage the use of point counts in favor of fixed width line transects. Nonetheless, keeping in mind that the objective is to sample a bird community in a 100-ha plot, increasing the area would require a decrease in the number of sampling units.Thismightnotbethemost favorable solution because our simulations suggest that theincreaseinsamplingunitsdecreases the bias faster than the increase in replicates of each sampling unit (Figure 2-1;FigureA-1).
If the objective of the study is to estimate the abundance of a single species correcting for its detection probability, then our simulations are a guide to the sampling effort required. Published databases (e.g. Parker III et al., 1996; Karr et al., 1990), include estimates of abundance of many neotropical species, which could provide general guidelines to researchers in the field about the approximate λ they are dealing with and thus the approximate sample sizes needed to correctly estimate the abundance using N-mixture models. For rare species, the solution can be two fold: increase the sample size to a very high number of points and many replicates (>50 point counts and >20 replicates) or to keep the sampling design of a 36 points and 5 replicates and use our proposed model of the multi-species sampling.
We show that by using the information of other more abundant species, the model is able to predict correctly the abundance of rare species with λ =4and better approximate the abundance of species with λ < 4.Byrestrictingthedetectionprobabilityoftargetspecies to be correlated with other species, the beta n-model has moreinformationtoallocateto the estimation of λ (Figure A-2). While in the N-mixture model, allowing p to vary freely for
32 Table 2-1. Abundance estimates for understory insectivorous birds in the dry forest of the Magdalena Valley Colombia. Estimates are in individuals/100 ha N-mixture Beta Species Detections p λλlower upper Atalotriccus pilaris 83 0.315 119.7 119.8 67.9 171.7 Basileuterus rufifrons 104 0.219 215.4 214.9 104.3 325.5 Campylorhynchus griseus 7 0.311 10.2 10.5 0 22.2 Cantorchilus leucotis 3 0.0004 3832.3 32.5 0 193.0 Cnemotriccus fuscatus 31 0.174 80.9 78.7 8.5 149.0 Contopus cinereus 2 0.004 211.4 14.4 0 69.9 Cymbilaimus lineatus 4 0.0005 3663.8 41.1 0 181.2 Dromococcyx phasianellus 1 0.0005 905.9 6.3 0 39.9 Elaenia flavogaster 67 0.126 241.3 231.4 62.1 400.6 Euscarthmus meloryphus 26 0.265 44.6 44.3 14.9 73.8 Formicivora grisea 172 0.280 279.3 279.5 168.7 390.3 Hemitriccus margaritaceiventer 106 0.408 118.1 118.4 81.8 155.0 Henicorhina leucosticta 28 0.124 102.3 95.5 0 201.3 Hylophilus flavipes 144 0.064 1023.2 829.2 0 2086.4 Leptopogon amaurocephalus 23 0.194 53.8 53.0 9.5 96.6 Myrmeciza longipes 64 0.257 113.4 113.0 55.8 170.2 Myrmotherula pacifica 1 0.001 905.9 5.9 0 32.1 Pheugopedius fasciatoventris 83 0.230 164.0 163.1 77.2 249.0 Poecilotriccus sylvia 69 0.239 131.0 130.0 52.6 207.3 Ramphocaenus melanurus 5 0.206 11.0 11.0 0 29.0 Synallaxis albescens 1 0.0005 905.9 6.4 0 48.7 Thamnophilus atrinucha 93 0.255 165.9 165.5 92.4 238.6 Thamnophilus doliatus 192 0.246 354.5 353.8 186.3 521.2 Todirostrum cinereum 51 0.255 91.1 90.3 44.2 136.4 Tolmomyias sulphurescens 80 0.216 168.4 166.5 75.4 257.7 Troglodytes aedon 26 0.322 36.7 37.2 13.5 60.8
every species can produce an identifiability problem between p and λ (i.e. models with high p and λ have similar likelihood to models with low p and high λ), our beta model corrects for this identifiably problem by restricting p to be correlated with other species and actually increasing the information for it estimation (mean detection probability of 0.2 and 2.5% and
97.5% quantiles of 0.04 and 0.41 respectively). Such correction, allows the estimate of λ,even for rare species, to be much closer to the true value (Figure 2-2;FigureA-3 and A-4). Even when mean detection probability is low (p ∼ 0.25)thebetamodeltendstooverestimatethe
33 abundance of the entire set of species, but maintains the structure of the species abundance distribution (Figure 2-3 and Figure A-3). Our model is different from other approaches to multi-speciesabundanceandoccurrence estimation (Dorazio & Royle, 2005; Yamaura et al., 2011; Dorazio et al., 2015; Sollmann et al., 2015)becausewedonotassumethatdetectionprobabilitiesofspecies are independent. The assumption of correlated detection probabilities allowustomakeinferencesaboutthe abundance of rare species that are usually discarded when estimating the composition of communities. Yamaura et al. (2011)madesimilarassumptionsinwhichspeciesrespondasa community to changes in environmental covariates. However,weassumethatthedetections of species are the product of a stochastic process, instead ofdeterministicallypredictingthe detection probability for each species as a function of some other covariates. An important consideration of our approach is that the grouping of speciesusedtoestimatethedistribution of detection probabilities has to be carefully justified and informed by their ecology and vocal behavior. It makes no sense to assume that species that are extremely different in their ecologies will have correlated detection probabilities. Martin et al. (2011)andDorazio et al. (2013)usedasimilarapproachtoours,but for single-species abundance estimation. In their models, they assumed correlated behavior among the individuals of the same species and variation across sites, adding an additional layer of hierarchy to the traditional N-mixture models (Royle, 2004). In this case the binomial distribution is substituted by a beta binomial that assumes that the probability of detecting one individual is slightly different from another, but the result of the same stochastic process. Similarly, we assumed that species have their own detection probability that is correlated with other species with similar ecologies. However, there isnoanalyticalsolutionforour model because it involves solving the integral of the beta distribution on top of the analytical solution for the N-mixture model. Fortunately, the Data Cloning algorithm allows us to make Maximum likelihood inferences without having to solve for the integral instead of using
Bayesian approaches.
34 Because our model is essentially identical to any N-mixture model it has the advantage that it can be adapted to any underlying distribution of abundances. Similarly, the Poisson distribution used to model the mean number of individuals canbereplacedbyanyother distribution that relaxes the homogeneity assumption (e.g. Negative Binomial or Zero Inflated
Poisson). In addition, ecological inferences can be made by incorporating covariates of the abundance process in the model as previously suggested with N-mixture models (Joseph et al., 2009). The detection process can also depend on variables influencing the overall detectability of species by making the parameters of the beta distribution a function of the covariates. For example, one can assume that the detection probability distribution is a function of variables such as the ecological guild a bird belongs to or to the microhabitat used for foraging and nesting. Model selection comparing models with and without abundance and detection covariates can be useful for inferring ecological mechanisms underlying the abundance of species (Joseph et al., 2009). In the beta n-mixture model, the assumption of the correlated
behavior can be tested by comparing it to a regular n-mixture model, and because the main difference is in the assumptions underlying detection probability, it allows us to make inferences about ecological similarity among species in the same guild,habitatorfunctionalgroup. The estimates of the abundance of the understory insectivores of the upper Magdalena
Valley show that the difference between the N-mixture and beta-mixture models relies on the estimation of the abundance of rare species. For example,forspecieswithlessthanfive detections, the N-mixture model estimates the abundance to be extremely high (Table 2-1). Instead, by assuming the detection probability is correlated with the other species in the set, our approach lowers the estimation of the abundance to valuesclosertodensitiesreportedfor
the same species or similar in other regions (e.g., Karr et al., 1990; Parker III et al., 1996). It is worth noting that the abundance of more common species with higher numbers of detections in our dataset might be a little bit higher than in other published data sets. There are three possible reasons for this. First, when the mean detection probability of the species is low,
our simulations showed that the beta-mixture model overestimated the true abundance of
35 species (Figure A-3). The second reason is more ecological: the data presented here comes from the dry forests of the Magdalena valley. Even though thisecosystemisalessspeciesrich than wet forest ecosystems, the biomass of the community doesnotchange(Gomezetal. unpublished data). This means that the populations of most species tend might be higher than in wet forests from which most of the abundance data for neotropical birds have been collected (Terborgh et al., 1990; Thiollay, 1994; Robinson et al., 2000; Blake, 2007). Third, it is also possible that rare species do not have to sing much to defend their territories because they have few neighbors. Common species, on the other hand, face a constant threat of territorial intrusion and may have to sing more.
Overall, our study can be used as a baseline to determine the number of point counts required to estimate the density of neotropical bird speciesusingN-mixturemodels.We showed that for many species in neotropical communities, thesamplesizeneededtocorrectly estimate their density is high and thus we advocate for more field intensive methods such
as spot mapping. Probably over even larger areas than the standard 100-ha plot because of the very large number of species with fewer than 2 territories/100ha. Distance sampling can also be an alternative, but it has been previously shown that it also requires a high number of detections to appropriately estimate the abundance of species (Matsuoka et al., 2014).
Such high number of detections might be impossible to achieve, particularly for rare species. We conclude that our method might be a good alternative when sample sizes are low but information for many species is available. When the communities are large, however, the data cloning algorithm is computationally intensive. Another caveat is that in large neotropical communities, the assumption of correlated detectabilitiesmightnotholdformanyspecies.We
have shown an example in which we estimated the abundance of 26 species of insectivorous birds in the Magdalena Valley, allowing us to demonstrate that for a community of this size, the maximum likelihood estimation with this size of community is feasible. We are certain that the larger the community the easier it will be for the model to estimate the abundance of rare
species (Sollmann et al., 2015), but there is a tradeoff with the computational power needed
36 for quick maximum likelihood estimation. We hope that this manuscript is useful for study planning and in particular for community ecologists interested in estimating abundances of full communities while taking into account differences in detection probabilities among species.
37 CHAPTER 3 TESTING THE ROLE OF ENVIRONMENTAL FILTERING ON THE ASSEMBLY OF TROPICAL BIRD COMMUNITIES ALONG A MOISTURE GRADIENT 3.1 Background
Environmental factors have long been hypothesized to determine species composition of
a community by filtering out species that are not adapted to theparticularconditionsofthe locality (Grinnell, 1917; Keddy & Weiher, 1999; Condit et al., 2002; Tuomisto et al., 2003). If species abundances are used as surrogates of their fitness under local environmental conditions (Sokol et al., 2011), then the abundances of species should change as environmental conditions change through climatic space (Chase & Myers, 2011). Viewed in this context, environmental
gradients can be treated as natural experiments in which the role of environmental conditions in community assembly processes can be tested under varying conditions (McGill et al., 2007; Chase & Myers, 2011). Along environmental gradients, the spatial scale can be small enough that no geographical
barriers limit dispersal between local communities. In these types of environmental gradients, local communities are connected through dispersal and immigration, homogenizing the local communities at broad spatial scales. This group of local communities can be treated as a metacommunity (Leibold et al., 2004), which is defined as a set of local communities that are
connected through dispersal. Within a metacommunity, the assembly process is the result of two independent steps (Etienne, 2007; Munoz et al., 2007; Jabot et al., 2008; Etienne, 2009): 1) formation of species in the metacommunity and 2) immigration/recruitment in the local community (Jabot et al., 2008). Within local communities, the assembly process reflects a zero-sum game, in which an immigrant from the metacommunity or a recruit from a species
already present in the local community immediately replaceseveryindividualthatdies(Hubbell, 2001). In metacommunities distributed within homogeneous climates, the assembly process could be mainly determined by dispersal limitation and biotic factors (e.g. competition, parasitism). Limitations to dispersal should be a function of the distance of a particular
38 community to the rest of the sites (Hubbell, 2001). However, in metacommunities distributed along environmental gradients, dispersal should be additionally influenced by the abilities of the immigrants to recruit in the climatic conditions of each site (Keddy & Weiher, 1999). All individuals might be able to disperse into the local communities, but only individuals from
species that are adapted to the local climatic conditions canrecruitsuccessfully(Grinnell, 1917; Keddy & Weiher, 1999; Jabot et al., 2008). Those species that fail to recruit in a local community may appear as if they were unable to disperse,eventhoughtheirabsence actually reflects their inability to become established. This process has been referred to as the species-sorting hypothesis (SSH; Leibold et al., 2004).
Along environmental gradients, the selection against immigrants is likely to be stronger at the extremes of the gradient where climates are likely to beharsher.Intheextremes,SSH predicts that each time an individual is lost from the community, it is more likely to be replaced by an individual of a species that is already present in the local community than it is by an
immigrant from the metacommunity. Harsh conditions therefore should decrease immigration from the metacommunity (Leibold et al., 2004). In addition, it is likely that species colonize localities with one or few individuals. If such species are marginally adapted to extreme environmental conditions, then strong selection by the environment would likely eliminate entire
populations of these species, favoring the recruitment of already established, well-adapted and abundant species. Communities in harsh environments, therefore, should have reduced species diversity (few rare species and more common species), and should show greater differentiation from the greater metacommunity regardless of the geographicdistancesseparatingthese communities from those in less harsh environments. This latter prediction is identical to the
prediction of dispersal limitation by distance except that in this case, differentiation should be proportional to the differences in climatic variables ratherthangeographicdistanceper se. Rainfall gradients are hypothesized to have stronger selection at the drier end, possibly because of limited water availability and greater temperature variability (Condit et al., 2002;
Engelbrecht et al., 2007; Jabot et al., 2008; Graham et al., 2009). In this case, the SSH
39 would predict that immigration of individuals from the metacommunity to localities in the dry forest should be reduced as a result of the inability of most wet forest species to recruit in dry forests. In the wet end, where climatic conditions are more benign, the immigration of an individual should not depend on its ability to overcome the environmental conditions, which should lead to higher immigration rates from the metacommunity and greater persistence of species with small populations. Thus, the immigration rate should increase with precipitation as individuals have fewer restrictions on immigration into the wetter end of the gradient (Jabot et al., 2008). Alternatively, if there is no effect of climate on the immigration process, then local communities should have similar immigration rates or at least there should not be a relationship between precipitation and immigration rates (Jabot et al., 2008). Instead, immigration rates should be lower in local communities that are at the periphery of the metacommunity (Hubbell, 2001). Empirical values of immigration to local communities are difficult to obtain even at small spatial scales. Several models based on Neutral theory, however, allow the estimation of immigration parameters from observed abundance data collected in the field (Etienne, 2007; Munoz et al., 2007; Jabot et al., 2008; Etienne, 2009). Even though these models are essentially neutral, the reinterpretation of the immigration parameter into a recruitment limitation parameter allows for testing the role of environmental filtering in community assembly (Jabot et al., 2008). Even though bird communities were very important in the development of current community ecology theory (Preston, 1948; Cody, 1974; Cody & Diamond, 1975), plant systems have been mostly used to test the effect of environmental conditions on community assembly and the development of neutral models (Keddy & Weiher, 1999; Hubbell, 2001; Condit et al., 2002; Tuomisto et al., 2003; Engelbrecht et al., 2007). Birds, however, might be a useful system to test the effects of climate on species abundance distributions, because their dispersal abilities might be greater and because they can actively select their habitat and environments.
40 Thus, neutral processes might be less important for birds than they are for plants, allowing us to increase the power to detect the influence of climatic conditions in community assembly. The Magdalena River Valley is an inter-Andean lowland valleylocatedbetweenthecentral and eastern cordilleras in Colombia, and contains a precipitation gradient that ranges from
1000 mm in the south up to 5000 mm in the north, but has little variation in elevation and temperature. There are no geographic barriers that prevent dispersal of species allowing us to make the assumption that the localities within the valley are potentially part of one metacommunity. In this study, we obtained information on thecommunitycompositionfrom13 localities along this gradient; to test if the SSH is a good predictor of the observed distribution patterns of birds in the Magdalena Valley in Colombia. Specifically, we test the prediction that individual immigration rates are lower in localities with harsher climates (i.e. dry forests). Alternatively, the strength of selection is high in both endsofthegradientinwhichcase, the SSH would predict that the localities within the gradientbehaveastwoindependent metacommunities (i.e. dry forest and wet forest) even in the absence of geographical barriers separating them. Finally, we test if dry forests have fewer rare species than wet forests, as predicted by the SSH based on reduced immigration rates under harsher environmental conditions. 3.2 Methods
3.2.1 Bird Sampling
We selected at random 13 localities along the Magdalena Valley distributed to capture the entire rainfall gradient (Table 3-1). We selected only the localities with continuous forest of at least 50 ha, but it varied between 50 and 1000 ha. In each locality we sampled birds using 50-m fixed radius point counts (Hutto et al., 1986)inwhichwecountedallbirds detected both visually and aurally for a period of ten minutesateachpoint.Pointcounts were repeated temporally a maximum of four times, although some of the points where only counted once. Points were separated from the edge of the forest by a minimum distance of 75 mandwereseparatedfromeachotherbyatleast200mtoensureindependence and minimize
41 the sampling of species of the matrix surrounding the patch (Blake & Loiselle, 2001). Each morning we conducted up to ten point counts starting at dawn and until 10:00 AM or until activity dropped considerably. We avoided censusing duringwindyandrainydays.Forthe analyses we did not include Toucans, Parrots, Hummingbirds,Swallows,Swifts,waterbirds or birds that flew over the point while censusing because it wasdifficult to determine the independence of point counts for loud and highly mobile species.
Table 3-1. Characteristics of the study sites. Classification into the Dry forest or Wet forest ecoregions follows Olson et al. (2001) Locality Ecoregion Points Precipitation Species Individuals Rare Species Venadillo DryForest 20 1224 87 1483 17 Bateas Dry Forest 36 1280 102 1868 20 ManaDulce Dry Forest 32 1375 117 2959 14 Potosi DryForest 30 1379 110 2926 15 Jabiru Dry Forest 26 1618 105 2314 12 Boqueron Dry Forest 13 1668 101 2022 15 Rio Manso Wet Forest 25 2240 118 1980 28 Rio Claro Wet Forest 14 2428 141 2350 38 Maceo Wet Forest 13 2528 133 2222 25 La Perla Wet Forest 9 2755 121 1812 29 Remedios Wet Forest 20 2824 126 1736 28 Barbacoas Wet Forest 34 2888 120 1796 29 SanJuan WetForest 10 2997 97 1672 26
3.2.2 Abundance Estimation
We used a zero-inflated Poisson model to estimate the mean number of individuals per species per point count and extrapolated it to 100 ha. Variation in the mean number of individuals among point counts could be modeled using a zero-inflated negative binomial model; however, this model tends to overestimate abundance (Joseph et al., 2009). We selected 100 ha because it allowed comparisons with other studies from tropical lowland forests (Terborgh et al., 1990; Thiollay, 1994; Robinson et al., 2000; Blake & Loiselle, 2009). We excluded hummingbirds, parrots, swifts and swallows, nocturnal birds and birds that flew over the point count during the sampling time. All of the estimateswerederivedusingthepscl package (Zeileis et al., 2008) in R v 3.0.2 (RCoreTeam, 2013). We finally inspected the
42 estimates visually to determine the accuracy of the models to estimate the abundance of each species, and compared them to previous estimates for similarspecies(Terborgh et al., 1990). 3.2.3 Model Fitting
We used the multiple samples estimation approach (Etienne, 2007; Munoz et al., 2007; Jabot et al., 2008; Etienne, 2009)toestimatetheimmigrationparametersthatbestdescribed the data. In this approach, the assembly process is assumed to be a two-step process in which speciation occurs at the metacommunity level, and then localcommunitiesareassembled
through immigration (Munoz et al., 2007). Within the local communities, the dynamics are assumed to follow a zero-sum game, in which any deaths in the local communities have to be immediately replaced by; (1) a recruit from the local community or (2) an immigrant from the metacommunity. The zero-sum game makes the assumption that all niches in the
local community are occupied and that the number of individuals are kept constant. The two-step assembly process is described by two parameters that allow the prediction of the species abundance distribution of each local community. TheFundamentalBiodiversity number (θ; Hubbell, 2001)describesthediversityofthemetacommunitybasedonitssize
(i.e. number of individuals) and the rate of introduction of new species (i.e. speciation rate). The immigration rate (m)parameterisdefinedastheprobabilitythatanimmigrantfrom the metacommunity replaces a death in the local community. Low values of m indicate high restriction to dispersal because it is more likely that recruits from the local community replace dead individuals. Alternatively, m can be reinterpreted as the potential number of immigrants from the metacommunity (I )thatcompetewithlocalindividualsforavailablesites(Etienne &
I Olff, 2004). I and m are related as follows: m = I +j−1 where j is the number individuals in the local community. In the Magdalena Valley, there are no geographic barriers that prevent dispersal of
species allowing us to make the assumption that all of our communities are potentially part of one same metacommunity. In the two-step assembly process two scenarios can arise. The first occurs when immigration is constant from the metacommunities to each local
43 community. In this case, a two-parameter model is sufficient topredictallofthespecies abundance distributions of the 13 localities (i.e., the Constant Immigration model: Etienne, 2007). The second case occurs when the migration rates are different for each locality in the metacommunity (i.e., the Variable Immigration model: Etienne, 2009). If the environment is
not important for determining the species’ abundances along the valley, then the Constant Immigration model with only two parameters will be a better fittothedata.Alternatively,if the environment is important for structuring communities and the dry forests present a stronger environmental filter, then the Variable Immigration model will fit the data better and the immigration rate should increase with precipitation. Thus,wefittedthemodelswithConstant
and Variable immigration using the numerical optimizationspresentedbyEtienne (2007)and Etienne (2009), respectively. Alternatively, it is possible that selection is strong at both ends of the gradient. This would cause immigration between sites in different habitats to be low, such that the gradient
can be split into two or more metacommunities (Leibold et al., 2004). One way to test for this effect is to assume that dry forests and wet forests each represent a separate metacommunity. We used the definition of the WWF ecoregions to group the localities into dry forest and wet forest communities separately (Olson et al., 2001)andfittedtheConstantandVariable
immigration models to the dry forest and wet forest localities separately. Finally, it is also possible that isolation from the metacommunity results in reduced immigration rates (Hubbell, 2001). To account for the effect of distance, we estimated the centroid of the Magdalena Valley metacommunity and calculated the distance from each local community to the centroid. We then used this distance as a measurement of isolation, assuming that local communities that are furthest from the centroid are in the periphery of themetacommuntityandare therefore more likely to receive a lower number of immigrants. To test for the effects of precipitation and distance on immigration rates (m, I ), we fitted a linear model that had precipitation and distance as predictor variables. We compared the AIC
of this model with the models including only precipitation and only distance to determine which
44 variable had the greatest influence on the variation of immigration rates. We then repeated the analysis with the immigration estimates from the variable immigration two-metacommunity model. In the latter case, we calculated the centroid for eachmetacommunityseparately,and found the distance from each local community to the centroid of the metacommunity in which
it belongs. We also compared the immigration estimates between the dry and wet forests and between variable immigration one and two metacommunty models using ANOVAs. We compared the fit of the models using AIC in which the model with the lowest AIC was taken as the best, and models were significantly different if the difference in AIC was greater than 2. To assess the goodness-of-fit of the best model for predicting the species abundance
distribution, we fitted a fully parameterized model in which for each local community we estimated a θ parameter as well as an immigration parameter using the Etienne’s sampling formula (i.e. Full model: Etienne, 2005). The goodness-of-fit is given by a likelihood ratio test between the full model and the best model (Strong et al., 1999). All of the maximum
likelihood estimates were performed following Etienne (2007, 2009)modificationofthe Etienne’s sampling formula (Etienne, 2005) and were performed in PARI/GP v. 2.7.0 (Group et al., 2014)usingtheprogramsavailableinthesupplementaryinformation of Etienne (2007, 2009).
The 95% confidence intervals for the parameters of the models were estimated using parametric bootstrap. Because of computational limitations, we only estimated confidence intervals for the best two models. We simulated 1000 communities using the maximum likelihood estimates of the observed data and the observed size of the metacommunity and each of the local communities. The simulations were performed following the urn procedure of the two-step approach suggested by Etienne (2007). 3.2.4 Rarity Analysis
Because habitat filtering should reduce the level of immigration into local communities, each time an individual is lost from the community, it is more likely to be replaced by an individual of a species that is present in the site than by an immigrant from the metacommunity.
45 For this reason, communities that are structured by environmental filtering should have fewer rare species than communities in which other factors are moreimportantduringcommunity assembly. To test this, we counted the number of species that had 2 or fewer individuals per 100 hectares in each community and related the number of rare species to precipitation.
Because wet forest habitats usually have more species, the increase in the number of rare species can be due to differences in overall number of species per locality. To account for this potential bias, we constructed a model to explain differencesinthenumberofrarespeciesin which we assumed that the sampling error is Poisson distributed, considering precipitation and species richness together and also separately. We then compared the models using AIC. We
repeated the analysis setting rarity to: < 5 individuals/100 ha. 3.3 Results
3.3.1 Abundance Estimation
We estimated the abundance of a total of 222 species of birds along the entire valley. The estimates based on a zero-inflated Poisson model for the abundance of each species resulted in realistic estimates (Table B-1). Detailed description of the localities is presented in Table 3-1. 3.3.2 Comparison Among Models
The model that fitted the data best was Variable immigration with two-metacommunities, with high goodness-of fit (Table 3-2;seeTableB-2 for parameter estimates for all models; Likelihood ratio test: χ2 =9.52,df =11,p =0.6). The second best model was the Variable Immigration with one meta-community (Table 3-2), but the low goodness-of-fit was low
(Likelihood ratio test: χ2 =54.2,df =12,p > 0.01). Immigration estimates were significantly higher in the variable immigration two-metacommunity modelthantheone-metacommunity model (Fig 2; ANOVA: m : F =53.24,df =24,p < 0.001; I : F =21.27,df =24,p < 0.001). 3.3.3 Variable Immigration One-metacommunity Model
Immigration rates showed a positive significant relationship with precipitation (Figure 3-1; m : F =15.05,df =11,p < 0.01, r 2 =0.58,m =0.009+0.007Precipitation; I :
46 Table 3-2. Model Selection for the 13 metacommunities in the Magdalena Valley. We present the number of parameters per model, the log likelihood of the Akaike Information Criterion (AIC) and the difference in AIC between the four models fitted to the data. The Full model is the model with different θ and restriction to dispersal (m < 1), Constant Immigration and Variable Immigration models consider one θ such that the Magdalena Valley would be considered as one metacommunity (One Metacommunity) and two θ such that Dry forests and Wet forests are considered independent metacommunities but with constant and variableimmigrationrates respectively. Models Parameters LogLik AIC deltaAIC Full Model 26 -1287.26 2620.52 6.48
Constant Immigration One Metacommunity 2 -6166.6 12337.2 9723.16 Two Metacommunities 4 -4481.92 8971.84 6357.8
Variable Immigration One Metacommunity 14 -1314.38 2656.75 42.72 TwoMetacommunities 15 -1292.02 2614.04 0
F =11.9,df =11,p < 0.01, r 2 =0.52,I =24.2+13.3Precipitation). Distance of the locality from the metacommunity had little influence onimmigrationestimates(Table B-3). Potential immigrants significantly decreased with distance from the centroid; the best model was the one including both distance and precipitation (Figure 3-1; F =7.16,df = 10, p =0.01,r 2 =0.58,m =0.005+0.008∗ Precipitation +1.4× 10−6 Distance; F =
5.5, df =10,p =0.02,r 2 =0.52,I =20+14Precipitation +0.02Distance). However, the proportion of variance explained by the Precipitation term was considerably higher than the variance explained by the Distance term and the distance coefficient was non-significant (m : Precipitation =67.2,Distance =0.33;I : Precipitation =49.4%;Distance =9.9%).
After correcting for the effects of distance on the immigration estimates, local communities in the dry forests had significantly lower immigration estimates than local communities in the wet forests (Figure 3-2;ANOVA:m : F =30.54,df =11,p < 0.001; I : F =14.5,df = 11, p < 0.01).
47 oa omnte ntedyadwtfrsssoe iia i similar showed forests ANOVA: wet and dry the in communities local m E 3-1. Figure ihes( richness Rarity 3.3.5 2.4, .. aibeImgainTomtcmuiyModel Two-metacommunity Immigration Variable 3.3.4 mirto siae o h oa omnte (Table communities local the for estimates immigration : nte2idh aetebs oe a h n nldn preci including one the was model best the case ind/ha 2 the In h eainhpbtenimgainadpeiiainwa precipitation and immigration between relationship The df F =0.87, =11, m aespecies rare ftemtcmuiyi ipra iiainb sn h fo the using by limitation dispersal e I in the two-met metacommunity for the the account from of presented results values show The panels bottom model. two the and model ( immigrants of itneadPeiiain respectively, Precipitation, and Distance : ff − F cso rcptto nteimgainrate immigration the on precipitation of ects p β =1.51, df 2 =0.15; Distance
=11, Distance corrected Immigration rate (m) (2 0.00 0.05 0.10 0.15 0.00 0.02 0.04 0.06 ind C A df . . 3.5 2.5 1.5 3.5 2.5 1.5 I I / p = =11, =118 )( B ,D ) .T h et w ot o pp a n e l sr e f e rt ot h eo n em e t a c o m m u n i t y ha =0.37; A )= + p β − e 1 =0.2; 2.3+0.42 Precipitation 11.7 m Precipitation (m) =0.048 Precipitation I Precipitation : 48 F
=4.21, Distance corrected Potential Immigrants (I) ,where
− 0 200 400 600 0 40 80 120 A − D B 2.1 n steitreto h model. the of intercept the is and 0.001 . . 3.5 2.5 1.5 . . 3.5 2.5 1.5 B-3 .Dsac a oiflec nthe on influence no had Distance ). ff × df s( β Richness .Atrcretn o distance, for correcting After ). cso distan of ects 1 mgainetmts(Figure estimates mmigration 10 m =11, and sn o ts i g n i fi c a n t( F i g u r e − )( A ,C )a n dp o t e n t i a ln u m b e r 3 ;hwvr h ihesterm richness the however, ); Precipitation β 2 iainadspecies and pitation p r h coe the are =0.06 lwn correction: llowing efo h centroid the from ce acommunity ). ; ffi I insfor cients : F 3-1 = 3-2 ; ; A B Dry Forest Wet Forest Distance corrected Immigration Rate (m) Distance corrected Immigration Distance corrected Potential Immigrants (I) Immigrants Distance corrected Potential 0 50 100 150
0.00One 0.02 0.04 0.06 One Two Metacommunity Metacommunity Metacommunities
Figure 3-2. Differences in (A) Immigration rates (m)and(B)Potentialnumberofimmigrants (I )betweentypesofforestanddifferent models used to estimateimmigration.The error bars show the standard deviation of the mean parameter in each type of forest. was non-significant (Table B-4). In the 5 ind/ha case the best model was the one including
precipitation only (rare species (5 ind/ha)=e3.2+0.2 Precipitation)butwassimilartothefull model (Table B-4). However, in the precipitation plus species richness model, the latter term was not significant (Table B-4). Because species richness had a positive influence on the number of rare species in the
Observed Rare Species analysis with 2 ind/ha, we applied the following correction: rare species = e0.006Richness . After applying the correction, rarity in the local communities increased from the dry forest to the wet forest (Figure 3-3). The analysis suggested that the model with different mean numbers of rare species per metacommunity was more likely than the model with no differences between types of forests in both the 2 ind/ha and 5 ind/ha cases (Magdalena =
28/54, logLik = −56.2/ − 57.3; Dry =19/44, Wet =37/65, logLik = −37.3/ − 44.02; χ2 =37.8/26.5, p < 0.01 in both cases).
49 ihntedyfrs,btteopst ih o etu nwe in true be not might opposite the hab but forest forest, dry the dry colonizes the species within a Once forest. wet the in more move individuals that and indicates dry which independently, from rates immigration the wet estimating to when dry negative from birds (Figure of smaller dispersal cond was that harsher the indications into also birds were of immigration is pattern reduced This predicts reverse. the f than wet forests that dry suggesting into forest, immigrate dry the in lower is birds con forest metacommunity a from rates d immigration from However, limitation versa. immigration strong g is of there absence that the suggests in even metacommunities, two into divided be number and precipitation between Relationship 3-3. Figure idcmuiisaogtepeiiaingain nteMa the in gradient precipitation the along communities Bird aly A hw h eut o h pce ihls hn2 than less with species ind/ha the 5 for than less results with the species shows (A) Valley.
3-2 Number of rare Species (2 ind/Ha)
.Terltosi ewe mirto ae n precipit and rates immigration between relationship The ). 0 15 30 45 60 A . . 3.5 2.5 1.5 . Discussion 3.4 Precipitation (m) 50 Number of rare Species (5 ind/Ha) 15 30 45 60 75 B . . 3.5 2.5 1.5 rs id ih els beto able less be might birds orest reyaogstsi h r than dry the in sites among freely yfrsst e oet n vice and forests wet to forests ry toso h r oet There forest. dry the of itions frr pce nteMagdalena the in species rare of ossetwt h S,which SSH, the with consistent oet a iie,bttee the but limited, was forests tti a naeaylocality any invade can it itat e oetmetacommunities forest wet itn fbt r n wet and dry both of sisting tf o r e s t s .T h u s ,h a b i t a tfi l t e r i n g orpi ares hsresult This barriers. eographic dln alycnessentially can Valley gdalena n/aad()shows (B) and ind/ha to was ation ff ect appears to operate mainly at the regional scale through exclusion of wet forest species from dry forest environments, but alternative mechanisms such as biotic constraints (see below) might operate at a local scale in wet forests. There is little evidence that distance strongly affects similarity among communities, a result suggesting that dispersal limitation has little effect on these communities(Hubbell, 2001). Distance from the metacommunity also had little explanatorypowerfortheobservedvariation in the immigration estimates, which further suggests that dispersal limitations do not affect community organization. Furthermore, the fact that the two-metacommunity model better fits the data suggests that the forest transition zone, which is only about 20 km wide in the
Magdalena Valley (see Chapter 4), drastically affects the abundance of bird species. Thus, there seems to be a threshold effect in which communities shiftfromadryforesttoawet forest state over a short distance, rather than a gradual change as precipitation increases. This statement was also supported by analyses of phylogenetic, compositional, and morphological changes in communities along this gradient (Chapter 4). Rare species seem better able to persist in wet than in dry forests, a result suggesting a relaxation of the selection pressures that would allow immigrants of new species to establish and maintain populations that are rare initially. In contrast, only a small subset of species from the metacommunity can immigrate into dry forests, perhaps because the selection pressure is stronger, which would only allow species that cantoleratetheharshclimateto establish populations. Under more benign conditions, selection may not be strong enough to eliminate the rare species that are only marginally adapted to the environment, or that have highly specialized niches that could not be maintained in drier forests. Such species could be considered as occasional species that come and go in communities during the dynamic process of community assembly (Magurran & Henderson, 2003). However, in dry forests where selection is more intense, marginally adapted, occasional species cannot persist long enough to be captured in the snapshot time frame of typical community studies such as this one.
51 One potentially strong filter that immigrants would have to overcome in the dry forest is high daily temperature variability. Data logger measurements across two years in a dry and a wet forest locality showed that the coefficient of variation ofthedryforestwasalmosttwice that in the wet forest (Coefficient of Variation [CV] in Temperature in dry forest =13.3,CVin
wet forest =8.7; J.P. Gomez, unpublished data, see also Chapter 4). Such extreme variation could impose a challenge to recruitment, especially in earlylifestageswheneggsandnestlings are particularly susceptible to temperature changes (Webb, 1987). Wet forest species, however, might be able to recruit in dry forests along creeks and in riparian habitats where climate is much less variable than in the surrounding dry forest. In such habitats, immigrants can
overcome the selection by the environment and establish viable populations at low abundances. In fact, most of the wet forest species that that inhabit the dry forest were found along creeks or in riparian vegetation (e.g. Leptopogon amaurocephalus, Mionectes oleagineus;TableB-1). An alternative interpretation to the patterns of rarity thatwedocumentedisthatthe
mechanisms underlying abundance are substantially different in the wet end of the gradient. While dry forests allow only the immigration of well-adaptedspeciestotheharshclimate, wet forests allow the immigration of a larger set of species, some of which may depend upon resources and microhabitats that are not available in dry forests. The abundance of such
species that specialize on these resources, however, might be further constrained by other mechanisms such as competition (e.g., Gymnocichla nudiceps, Gymnopythis bicolor;TableB-1; Robinson & Terborgh, 1995; Touchton & Smith, 2011)ormutualismssuchasmixed-species flocks (e.g., Myrmotherula axillaris, Epinechrophylla fulviventris; Greenberg & Gradwohl, 1986; Mart´ınez & Gomez, 2013). Some of the rare species in the community may also be
limited by increased interspecific competition in the more diverse wet forest communities. The abundance of the Spotted Antbird, for example, has been shown to be determined by the presence of Ocellated Antbirds, which are behaviorally dominant at ant swarms (Willis, 1973). When released from this competition by the absence of the Ocellated Antbird, the
Spotted Antbird has been shown to increase greatly in abundance, suggesting that its rarity
52 was previously maintained by competition (Touchton & Smith, 2011). Russo et al. (2003)also showed that smaller species are often less abundant than larger species in some foraging guilds in tropical wet forests. The previous result might be attributed to their subordinate status during aggressive interactions that might limit their access to clumped resources (Robinson &
Terborgh, 1995). It is also possible that the more diverse community of nest predators available in wet forests (Gomez, unpubl Data) may further constrain populations of some species (Ricklefs, 1969; Martin, 1996). Therefore, in the relatively more stable climatic conditions of wet forests, we might predict a greater role of biotic interactions such as mutualisms, nest predation, and competition in structuring communities.
Several other studies have shown effects of environmental filtering on the assembly of tropical communities. The precipitation gradient along thePanamaCanalshowsevidenceof reduced immigration of plant species into the dry forest (Jabot et al., 2008). There is less similarity among plots along this 200-km gradient than thereisoverthousandsofkilometers in Amazonia (Condit et al., 2002). At a global scale, Jabot & Chave (2011)analyzed density-dependent parameters at the community level and concluded that environmental factors might govern the assembly of communities. They showed that inregionswithlowprecipitation, the species with higher abundances have a higher probabilityofrecruitingindividualsthan species with lower abundances. This can be interpreted as a positive density-dependent effect related to the fitness of a species in particular environmental conditions (Sokol et al., 2011). The effect of environmental filtering on structuring bird communities has mainly been demonstrated in temperate regions, where the climate is believed to be harsher than in tropical environments (Meynard & Quinn, 2008; White & Hurlbert, 2010; Ozkan¨ et al., 2013). In tropical regions, Graham et al. (2009, 2012)demonstratedthatdryforesthummingbird communities are composed of species that are closely related, suggesting that dry environments force species with similar ecological traits to coexist in the communities. This has been interpreted as evidence for the effect of environmental filtering (Graham et al., 2009, 2012).
Also, Gomez et al. (2010), demonstrated that antbird assemblages are most likely structured
53 by the environment at regional scale. Species that coexist inecoregionswereshowntohave similar traits related to physiological tolerances (Gomez et al., 2010). The Magdalena Valley has a complex history of climate fluctuations and presumed community composition. During the Pleistocene, the valley was a continuous dry forest that
connected the south end with the Caribbean lowlands in Colombia (Haffer, 1967). Interestingly, dry forest birds that became isolated from the Caribbean lowlands have successfully immigrated or persisted in the edges of the wet forests (Gomez, J.P. pers. obs.). Many of the birds that can be considered as dry forest indicators successfully invade the edges and the surroundings of the wet forest end of the valley (e.g. Conirostrum leucogenys, Euphonia concinna). The reason why they are not reflected in the analysis is because the point counts were located inside the forest. These birds were mainly found in scattered trees in pastures, live fences and small remnants of forest (<1 hectare), where typical wet forest birds were not found (Gomez, J.P. pers. obs.) It appears that these dry forest species can potentially inhabit the entire rainfall gradient, but their establishment inside wet forest is limited by other mechanisms. Colombia has more bird species than any other country in the world, and our data suggest that moisture gradients play a crucial role in maintaining aswellasgeneratingthisdiversity. Much of this regional diversity may result from dispersal limitation through geographic isolation as a product of Colombia’s complex geography (Smith et al., 2014). Yet, our results suggest that moisture gradients, which are ubiquitous in Colombia, may contribute greatly to this diversification. Most elevation gradients in the Andes also show strong precipitation gradients. The harsh environments created by the extremes of these gradients likely act as stronger filters than either variable acting independently. If threshold effects such as those we observed in
the Magdalena Valley are a general phenomenon, then we might observe similar community changes on even smaller spatial scales along these moisture/temperature gradients, a result that should generate rapid community turnover and very high beta diversity.
54 CHAPTER 4 DETERMINISTIC TURNOVER OF TROPICAL BIRD COMMUNITIES ALONG ASTEEP RAINFALL GRADIENT 4.1 Background
Comparisons of species composition of different communitieshavelonghavebeenused to infer the mechanisms underlying community assembly (Whittaker, 1960). Environmental gradients are particularly useful for such purposes becausetheypotentiallyallowustoseparate the influence of stochastic and niche-based processes (Chase & Myers, 2011; Legendre et al., 2005; Tuomisto & Ruokolainen, 2006)andtopredictthedeterminantsofecological communities (Whittaker, 1960; Terborgh, 1977; Jankowski et al., 2009, 2013; Condit
et al., 2002; Swenson et al., 2011). In particular, latitudinal and elevational gradients have been studied intensively to help identify the roles of different biotic and abiotic variables in determining community composition (Terborgh, 1977; Jankowski et al., 2009, 2013; Swenson et al., 2011; Qian & Ricklefs, 2007; Kraft et al., 2011; Rodr´ıguez & T Arita, 2004).
Environmental gradients are almost always associated with a change in the harshness of the abiotic environment. At high elevations, for example,lowtemperaturesandhigh temperature variability are thought to be analogous to the harsher conditions at high latitudes. Such conditions should increase the influence of the abiotic environment on the occurrence and
abundance of species (Graham et al., 2009; Qian & Ricklefs, 2007,butseeKraft et al., 2011). At low elevations and latitudes, where the environment becomes more stable, productivity increases, which also increases the potential for intra- andinter-trophicbioticinteractionsto determine community composition (Jankowski et al., 2012; Martin, 1988; Janzen, 1970). Thus, the turnover of species along elevational and latitudinal gradients is hypothesized to be the
result of a change in the relative importance of abiotic and biotic mechanisms that determine community assembly. Rainfall gradients also potentially vary in climatic stability and harshness. Along the rainfall gradient it is likely that water restricts the distribution of organisms at the dry end and
biotic interactions potentially determining community composition in the wet end (Engelbrecht
55 et al., 2007; Jabot et al., 2008). Plant communities along rainfall gradients, for example, are known to respond dramatically to drought conditions (Condit et al., 2002; Engelbrecht et al., 2007; Jabot et al., 2008). Alternatively, within the same habitat, plants also show niche partitioning, a possible response to competition at different life stages. Pathogens, herbivores,
and seed predators also affect plant community composition. Combining metrics of compositional, functional and phylogenetic beta diversity could increase the power of studies of species turnover on gradients (Stegen & Hurlbert, 2011). Comparisons of functional traits and phylogenetic relationships among species might give additional insights into the mechanisms underlying community composition (McGill et al.,
2006; Petchey & Gaston, 2006; Graham & Fine, 2008; Bryant et al., 2008). The expectations of the functional and phylogenetic turnover differ when the communities are assembled deterministically or stochastically along gradients (Swenson et al., 2011; Graham & Fine, 2008). Stochastic mechanisms such as random colonization and extinction predict that while compositional turnover can be high, functional turnover should be similar to that expected by chance (Swenson et al., 2011).In contrast, deterministic community assembly predicts high functional turnover among habitat types and low turnover when comparing similar types of habitats.
Both stochastic and deterministic turnover have been documented in plant and animal communities at different geographic and environmental scales (Hubbell, 2001; Gomez et al., 2010; Qian & Ricklefs, 2007; Graham et al., 2009). When functional and compositional turnover are paired with phylogenetic turnover, the latter informs us about the lability or conservatism of traits and the potential modes of speciationandbiogeographicalprocess
underlying species distributions (Graham & Fine, 2008). Assuming that the environment plays an important role in determining the rate of species andfunctionalturnoveralongan environmental gradient, a high phylogenetic turnover wouldbeanindicationthatthereishigh niche conservatism that restricts close relatives to particular environments. In contrast, low
phylogenetic turnover would be indicative of ecological speciation caused by local adaptation
56 to different environmental conditions. In this case, the replacement of species along the gradient would happen mainly among close relatives, some of which may have originated in situ (Graham & Fine, 2008) Even though combining the three metrics (i.e. functional, phylogenetic and compositional) of turnover provides a powerful test of niche versus stochastic processes (Graham & Fine, 2008), the structure of traits within local communities should further provide indications about the mechanisms underlying species turnover (McGill et al., 2006; Petchey & Gaston, 2002; Kraft et al., 2008, 2015). Species are a collection of traits that could evolve at different rates and respond differently to selective pressures (Ackerly & Cornwell, 2007). While some traits
may vary stochastically as a product of genetic drift, other traits may vary deterministically according to different mechanisms (Ackerly & Cornwell, 2007). In addition, the scale at which selection operates is likely to vary among species. In plants, for example,traits such as rooting depth, leaf mass per area and the ability to fix nitrogen are traits that should respond locally
to competition; alternatively, the degree to which a plant isdeciduousandhascompoundor simple leaves is likely a response to climatic stressors suchasdroughtandhightemperatures (Kraft et al., 2015; Lebrija-Trejos et al., 2010). In birds, Miles & Ricklefs (1984)andRicklefs (2012)suggestedthatoverallmorphologyshouldrespondtocompetitive interactions. Others have suggested that physiological tolerances of adults and juveniles should reflect adaptations to the environment (Webb, 1987; Kearney & Porter, 2009). Therefore, a combination of the measurement of trait turnover with the change in the structure of species traits along environmental gradients should allow us to not only differentiate between stochastic and deterministic community assembly, but could also reveal themechanismsthatdetermine
community composition (McGill et al., 2006; Kraft et al., 2008, 2015). In this study, we use compositional, functional and phylogenetic metrics of beta diversity to determine if the distribution of bird species along a steepenvironmentalgradientin Colombia is deterministic or stochastic. Furthermore, we use ecological and morphological
traits to test the hypothesis that the turnover in bird communities along the gradient is the
57 product of a change in the mechanisms determining species composition along the gradient. Specifically, we predict that because dry forests are harsherandmorestressfulenvironments, the relative importance of species sorting through environmental filtering should be highest in dry forests. Thus, we expect that the trait space in physiological tolerances occupied by dry forest communities will be smaller than the one occupied by wet forest ones. In contrast, because of higher productivity and relaxed environmental filtering, wet forest communities should respond more to biotic interactions such as competition and predation. In the particular case for competition, we expect wet forest communities to occupy broader eco-morphological trait space than dry forest communities. 4.2 Methods
4.2.1 Study Area
The Magdalena is one of the two lowland inter-Andean valleys occurring in central Colombia. The Magdalena River has been one of the most important rivers for navigation in the history of Colombia and of high importance in the colonization of northern South
America. The river drops quickly from its headwaters to the lowlands in the upper Magdalena Valley, which is characterized by low annual rainfall (1000 mm). The low rainfall in the upper Magdalena is the product of the rain shadow of both the centralandeasternAndeswhichrise above 4000 masl. About 200 km down river, the central Andes drop considerably in elevation allowing rainfall from the Pacific coast to pass over the AndesandfallinthemidMagdalena Valley increasing mean annual precipitation to almost 6000 mm in the western foothills of the eastern Andes. Because of its importance as a colonization route and as the connection for the interior of South America with the Caribbean Sea, the Magdalena Valley has a complex history of deforestation and fragmentation. The geological historyoftheMagdalenaisalsocomplex, because the wet forest has contracted and expanded several times during the last million years during glacial and interglacial periods. During the glacialperiods,theentirevalleywasdry, which provided connections among the dry forest fauna and flora of the Caribbean region of
Venezuela and Colombia and the dry forests in the upper Magdalena Valley (Haffer, 1967).
58 During these periods, the wet forest fauna and flora were most likely restricted to refuges in the lowlands north of the Andes, Choco and southern Central America (Haffer, 1967). 4.2.2 Bird Sampling
We selected at random 15 localities along the Magdalena Valley distributed to capture the entire rainfall gradient (Table 4-1). In each locality we sampled birds using 50-m fixed radius point counts (Hutto et al., 1986)inwhichwecountedallbirdsdetectedbothvisually and aurally for a period of ten minutes at each point. Point counts were repeated temporally
a maximum of four times, although some of the points where onlycountedonce(Table 4-1). Points were separated from the edge of the forest by a minimum distance of 75 m and were separated from each other by at least 200 m to ensure independence and minimize the sampling of species of the matrix surrounding the patch (Blake & Loiselle, 2001). Each
morning we conducted up to ten point counts starting at dawn and until 10:00 AM or until activity dropped considerably. We avoided censusing duringwindyandrainydays.Forthe analyses we did not include Toucans, Parrots, Hummingbirds,Swallows,Swifts,waterbirds or birds that flew over the point while censusing because it wasdifficult to determine the
independence of point counts for loud and highly mobile species. Because we did not have a large enough sample size to estimate the density of birds while correcting for detection probability (see Chapter 2), we estimated the abundance of bird species as the mean number of counts per species per point count. 4.2.3 Community Turnover
We had two particular objectives in this study. The first was totestiftherewasa difference among compositional, functional and phylogenetic turnovers in relation to rainfall. In
this analysis, we used three sources of information: abundance of species in each locality (as described above - mean number of individuals per point count), morphological and behavioral traits of each species and the phylogeny of all of the species we detected in our study. In the sections below, we provide detailed information about which traits we measured. For the
phylogenetic comparisons, we downloaded 1000 trees, to account for phylogenetic uncertainty,
59 from BirdTree.org using Hackett et al. (2008)asthebackboneforthedistributionoftrees (see Jetz et al., 2012 for details on how the trees where constructed). We calculated compositional turnover using the Chao index for assessing similarity of composition among communities while taking into account both abundance and sampling error
(Chao et al., 2005). The Chao index is an extension of the Jaccard index, which incorporates a probabilistic framework to account for species abundancesandthechancethatspecies might be shared but, because of their rarity in either community, they might be considered as absent from one of the communities because of sampling limitations (Chao et al., 2005). The index estimates the probability that any two individualssampledatrandomareshared
by both communities while taking into account that shared species might be present in the communities but not sampled (Chao et al., 2005). Phylogenetic turnover was calculated as the total length of shared and unshared evolutionary history among any two communities denoted by the length of the branches in the phylogenetic tree shared among communities and unique
to each community (Bryant et al., 2008). Because we had a distribution of phylogenetic trees we estimated phylogenetic turnover for each tree and report the mean turnover for the set of 1000 trees. We calculated functional turnover in a similar way to phylogenetic turnover, but instead
of using a molecular tree to determine relationships among species, we used trait data to infer a dendrogram of similarity among species (Petchey & Gaston, 2002). To construct the dendrogram, we used the total morphological matrix using both continuous and categorical traits. A detailed description of the traits and how they weremeasuredisprovidedinthe Trait Sampling section below. To allow categorical traits inthecalculationofthedissimilarity
matrix we used the general coefficient of similarity proposed by Gower (1971). Following the calculation of the dissimilarity matrix we performed a hierarchical clustering using the UPGMA method to construct the dendrogram, which performs better than other traditional methods in estimating species clustering for functional diversity analysis (Podani & Schmera, 2006). We
performed calculations of compositional and functional turnover using vegan (Oksanen et al.,
60 2016)packageandphylogeneticturnoverusingthepicantepackage (Kembel et al., 2010)inR (RCoreTeam, 2013). We determined the relationship between community similarity and rainfall by performing a multidimensional scaling of the beta diversity and relating the first axis of the scaling to rainfall. This methodology allowed us to determine if the turnover happened linearly in relation to rainfall or had a logistic (stepwise) form, in which case wecouldestimatetheamountof rainfall at which the community turnover is maximal. Additionally, using scaling of distance matrices or similar analyses such as canonical correspondence analysis provides stronger statistical power to detect the amount of community turnoverthatcanbeexplainedbythe variation in the environmental gradient (Legendre et al., 2005). We compared the linear model with a logistic function model in which scaled community similarity was the dependent variable and rainfall the independent one. The logistic function was of the form
a a Community Similarity = − (4–1) (1 + e(−b∗(Rainfall−c))) 2 in which a determines the height of the curve and in this case the maximumdifference estimated between types of communities, b determines how fast the transition happens from one type of community to another, and c determines the inflection point in which the community is expected to transition from type x to type y.Toestimatetheparametersofthe logistic function, we used least squares minimization similar to a traditional linear regression. We then compared the models using Akaike Information Criterion (AIC) and r 2.Weperformed the least squares minimization of the logistic function in R using the optim function. Because the rainfall gradient of the Magdalena Valley expands over two ecoregions
(Olson et al., 2001), we determined if the compositional, functional and phylogenetic turnover were higher than expected by chance between localities in different ecoregions and lower within ecoregion. In particular, if environmental filteringoperatesstrongerinthedryforest than in wet forest we expected the turnover to be lower in localities in the dry forest than in localities in the wet forest indicating lower community variability. We constructed 1000 random
61 communities using a swap algorithm that maintains the species abundance distributions as well as the richness of the communities (Hardy, 2008). For each of the 1000 random communities we calculated the compositional, functional and phylogenetic metrics using the observed functional dendrogram and phylogenetic tree. To determine if the turnover was higher or lower than expected by chance we calculated a standardized effect size (SES) for each of the metrics. The SES was computed as X − X SES = obs null (4–2) SD Xnull Overall, SES values higher than 1.96 or lower than -1.96 denote significantly higher or lower turnover than expected by chance, respectively. Finally, we determined if turnover within types of forests was lower than expected by chance using a t-test. 4.2.4 Climatic Description of Localities
To determine the influence of different environmental variables on the turnover of bird communities along the Magdalena Valley, we obtained mean annual rainfall and temperature variability from different sources. We obtained mean annual rainfall data from the closest climatic station to each of the localities sampled (IDEAM; Table 4-1). Climatic stations are run by Instituto de Hidrologia, Meteorologia y Estudios Ambientales (IDEAM) in Colombia. Mean rainfall for the period of 1981 - 2010 and the location of each station are freely available for download from their website. We determined the closest station to the locality by measuring geographic distance. Localities were all within 20 km of the closest station but most of them where much closer (mean = 8.31 km; Table 4-1). To account for possible deviations in rainfall due to distance from the station to the localities, we corroborated rainfall data from mean annual rainfall layer from bioclim (Hijmans et al., 2005). Because one way in which dry forests might be stressful to birds is through its stronger seasonality than wet forests, in addition to mean annual rainfall, we obtained information about rainfall seasonality and rainfall in the driest quarter from bioclim (Hijmans et al., 2005). Mean annual temperature, mean maximum temperature and temperature range were also obtained from bioclim (Hijmans et al., 2005). Finally, to obtain temperature variability, we used five Hobo U23 data loggers that were located
62 in two dry forests and three wet forests. The data loggers weresettomeasuretemperature and relative humidity each hour for an average of 662 days (Mana Dulce = 585 days, Jabiru = 233 days, Rio Manso = 730 days, San Juan = 659 days and Rio Claro =1104 days). Finally, we tested for significant differences in mean annual temperature, temperature range, mean maximum temperature, temperature coefficient of variation, precipitation seasonality and precipitation in driest month using a linear model relating each of this variables to rainfall in each locality.
Table 4-1. Location and description of localities sampled along the rainfall gradient of the Magdalena Valley. We show environmental variables as well asnumberofpoint counts and replicates per point count performed in each forest patch. Elev = Elevation (m), Bio12 = Annual Rainfall (mm), Dist = Distance to climatic station (Km), Bio2 = Mean Diurnal Temperature Range (◦ C), Bio5 = Max Temperature of Warmest Month (◦ C), Bio15 = Precipitation Seasonality (Coefficient of Variation), Bio17 = Precipitation of Driest Quarter (mm), Points = Number of census points for birds, Reps = Number of Replicates each point was censused Locality Elev Bio12 Dist Bio1 Bio2 Bio5 Bio15 Bio17 Points Reps Bateas 429 1193.5 3.29 27.5 11.7 35.1 60 78 36 3 El Triunfo 196 1281.8 8.40 27.3 10.4 34 51 217 7 2 Potosi 400 1330.5 1.47 27.4 11.3 35 58 102 30 2 ManaDulce 490 1456 5.04 27 11.2 33.8 46 176 32 4 Venadillo 335 1599.7 10.39 27.5 11.3 34.5 45 174 20 2 Jabiru 341 1623.5 4.77 27.2 10.4 33.9 48 216 27 4 Boqueron 650 2249.6 6.37 26.2 10.8 32.6 44 219 13 1 Mariquita 475 2263 0.00 26.2 9.8 32.5 43 334 11 2 Maceo 639 2554 17.67 25.9 9.9 31.5 43 265 13 2 Barbacoas 138 2675.2 6.16 27.8 9.9 33.6 45 245 34 3 Rio Manso 160 2697.1 9.22 27.4 10.2 33.6 47 325 26 4 La Perla 300 2714.7 16.79 27.1 9.6 32.5 35 399 9 1 San Juan 168 2888.5 11.07 28.1 9.7 33.8 44 254 20 4 Remedios 718 2906.3 20.22 25.4 9.7 31.1 44 272 20 3 Rio Claro 449 3775.9 3.88 26.1 10.1 32 43 347 15 4
4.2.5 Trait Sampling
In order to determine the influence of different mechanisms that could determine the rate of community turnover along the gradient, we constructed a database with morphological and ecological traits hypothesized to vary according to environmental filtering and competition.
63 Below, we will describe the traits and predictions of how we expect the morphological trait space to vary depending on the mechanisms hypothesized to operate in each locality. 4.2.5.1 Environmental Filtering
To test the hypothesis that the relative importance of environmental filtering is higher in dry forests, we obtained data on species’ climatic preferences. We specifically wanted to test the hypothesis that species that occupy the dry forests in the Magdalena Valley experience more stressful conditions throughout their ranges. By stressful conditions we mean explicitly higher maximum temperatures, wider temperature ranges, higher precipitation seasonality and lower precipitation during the dry seasons.Allfourvariablespotentially affect species distributions directly or indirectly. For example, wider temperature ranges and higher maximum temperatures might be problematic for theeggs(Webb, 1987)and potentially the adults (McKechnie & Wolf, 2009). Additionally, high precipitation seasonality and low precipitation during dry season are problematic for water regulation in both adults and nests but also might affect species through resource availability, which might be much lower during the dry season for most (but not all e.g. nectarivores) foraging guilds. To test the environmental filtering hypothesis, we obtained the meanvaluesofdiurnaltemperature range (bio2), maximum temperature in warmest month (bio5), precipitation seasonality (bio15) and precipitation of driest quarter (bio17) for each speciesthroughouttheirrange.Wethen computed community-wide environmental tolerances by computing the mean of the species present in the community weighted by the abundance of each species. For the former analysis we assumed that climatic variables are a good proxy for environmental stressors for species and thus for their physiological tolerances. Environmentaldatawereobtainedfrombioclim (Hijmans et al., 2005)andspeciesdistributionrangesfrombirdlifeinternational database (Birdlife International & NatureServe, 2014). Another prediction of the environmental filtering hypothesis is that in communities with stressful environments the species should have more similar environmental tolerances among them than species inmorebenignenvironments. Therefore, communities in dry forests should occupy a smaller trait space than wet forest
64 communities. To test this prediction, we estimated functional richness and dispersion using community-wide measurements of temperature range, maximumtemperature,rainfall seasonality and minimum rainfall. Functional richness is defined as the volume of the convex hull polygon delimited by the values of the n traits and s species present in the community
(Cornwell et al., 2006; Vill´eger et al., 2008). Functional Dispersion estimates the morphological centroid of the community in response to species abundances,andthenestimatesthespread of species from the centroid of the community (Lalibert´e& Legendre, 2010). Finally, using the same randomization procedure described previously to test for significance in the turnover of communities, we constructed 1000 random communities, and calculated a SES to determine if functional richness and dispersion of physiological tolerances were smaller than expected by chance particularly in dry forests. Additionally, we sought to test the hypothesis that dry forest species better regulate the temperature of their nests than wet forest birds. Specifically, we wanted to determine if dry forest birds had greater differences between maximum internal and maximum external temperatures throughout their development to determine if selection to avoid temperature extremes may be stronger in dry forests. Also, we explored if dry forest nests had lower internal temperature variability relative to the ambient temperature variability. We followed the development of 57 nests from 23 species in two localities,oneindryforestandone in wet forest (Dome = 6, Cup = 45, Platform = 6). We recorded temperature inside and 10 cm outside of the nest with a Hobo USB data logger for the length of the entire development of the nest or until it was either abandoned, depredated or nestlings died. We then calculated the difference between the maximum ambient and nest temperatures and the ratio between inside and outside temperature variance. As the ratio converges on one, outside and inside temperature variance are similar. If the ratio is less than one, it means that nest temperature variability is lower than ambient temperature variance and thus the nests are able to dampen environmental variability. Thus, nests that are adapted to avoid extreme changes in temperature that would affect the nest should show a larger difference between inside and
65 outside temperature and a lower than one temperature variability ratio. We asked if there were differences in temperature regulation within nest typesamongforesttypesusingtwo sampled t-test. The p-value of all of the t-tests performed were adjusted using bonferroni test correction. 4.2.5.2 Biotic interactions: Competition
To test the competition hypothesis, which predicts that competitively structured communities will be more overdispersed in morphological trait space, we measured eight morphological traits and one ecological trait that have beensuggestedtobecorrelatedwith the ecology of species (Pigot et al., 2016; Ricklefs, 2012; Miles & Ricklefs, 1984). The traits were: body mass, wing length, tail length, bill width, depth and height and tarsus length and foraging strata. The morphological traits were collected inthefieldusingmistnetstocapture birds. Tarsus and bill measurements were collected using a caliper with 0.01 precision and tail and wing length were measured using a wing ruler. Morphological data were available for 123 of the 213 species accounted in the analysis. For most of the species we used the mean of at least two individuals but some were represented by only one specimen (n=1225 individuals, mean number of individuals per species = 10, Number of specieswithtwoormoreindividuals =92,Numberofspeciesrepresentedbyoneindividual=31).Weareawarethatthespecies represented by one individual can potentially bias our analysis, but it is likely that in all of the species within the Magdalena Valley, intraspecific trait variation is smaller than interspecific variation. We estimated body mass for the species that we did not have morphological measurements using the CRC hand book of avian body masses (Dunning Jr, 1992). For foraging strata, we used a recently published database for all the bird species of the world (Wilman et al., 2014). The database separates foraging strata into five separate categories; ground, understory, mid story, canopy and aerial and for eachcategoryassignsaproportionof time that the species spends in each stratum. In that way, foraging stratum can be treated as a quantitative trait instead of a categorical one. To maintain foraging stratum as a single trait we coded the stratum from 1 - 5 sequentially from the ground to aerial foraging. Subsequently,
66 we obtained the weighted average for each species, with weights determined by the percent use of each strata. For example, if a species forages 20% of the time in the ground, 40% in the understory and 40% in the mid story the foraging strata value for that particular species was calculated as Foraging Stratum =0.2× 1+0.4× 2+0.4× 3+0× 4+0× 5=2.2.
Using the nine eco-morphological traits, we performed a principal components analysis (PCA) to reduce collinearity among variables. Because the variables ranged over several orders of magnitude, all of the variables were centered to have a meanofzeroandscaledtohave variance of one prior to the PCA analysis. We used the rotated scores from the first five PCs (the first five components explained 99.5% of the variation) to calculate functional richness and dispersion indices (i.e. eco-morphological richness and evenness). In competitively structured communities, the prediction is that ecomorphological richness and dispersion are higher than in communities structured by environmental filtering. Given the previous definitions of the metrics of functional diversity, the competition hypothesis predicts that there should be an increase in both functional richness and dispersion with rainfall along the gradient (Pigot et al., 2016; Kraft et al., 2008). The significance of the relationship between eco-morphological richness and rainfall was assessed using a least squares linear regression. Additionally, to determine if eco-morphological richness and dispersion were higher in wet forest than expected by chance, we constructed 1000 random communities for each of the 15 localities using the entire source pool of the Magdalena Valley, but maintaining both the frequency of each species and the richness of communities. To assemble the random communities we used the independent swap algorithm over 1000 iterations (Gotelli, 2000; Hardy, 2008). For each of the 15000 communities, we then calculated eco-morphological richness and dispersion. Finally, we calculated a SES richness and dispersion for each community with the expectation that wet forest communities would have SES values of richness and evenness greater than 1.96. In addition to the competition for food resources, species might also compete for nest space (Martin, 1988). This hypothesis predicts that species in the wet forest would have more
67 diversified nesting strategies than in dry forests. To test the latter prediction, we calculated nest diversity and dispersion among communities in a similarwaythanforeco-morphological traits. In this case, because the trait is categorical, functional richness was measured as the number of unique trait combinations in the locality (i.e. number of nest types, Vill´eger et al.,
2008). Functional dispersion was calculated in a similar way as described above. Because in the case of nest richness the data are counts of species with the same type of nest, we tested the significance of the functional richness and rainfall relationship using Poisson regressions. The relationship between nesting dispersion and rainfall was assessed using a beta regression. Functional dispersion and richness metrics were calculatedusingtheFDpackage(Lalibert´e&
Legendre, 2010; Lalibert´e et al., 2014)inR. 4.3 Results
4.3.1 Compositional, Functional and Phylogenetic Turnover
We found support for a stepwise turnover pattern in composition, function and phylogeny. In all of the cases, the logistic model fit the data better than a simple linear model even though it had at least one more parameter (Table 4-2,Figure4-1). Rainfall explained on average 88% of the variance in community similarity along the gradient (Table 4-2). The maximum turnover of the communities occured around the 2300 - 2400 mm isocline consistently for the three types of turnover measurements. Communities above and belowthe2300mmisoclineareon
average 75% different according to the compositional turnover, 64% to the functional turnover and 58% to the phylogenetic.
68 Table 4-2. Results of model selection for the relationship between community similarity and rainfall. Intercept and m are the parameters of the linear model and a, b and c are the parametersestimatedforthelogisticmodel.Thebestmodelisthe one with the lowest AIC and significant differences among models are detemined by differences greater than 2 in their AIC. Model Intercept m a b c R2 AIC Logistic Compositional 0.61(0.23,97) 210.4(8.59,478.4) 2425.2(1840,3772.5) 0.88 -15.32 Functional 0.44(0.08,0.68) 362.7(64,507.8) 2457.5(1899.9,5810.3.8) 0.87 -24.33 Phylogenetic 0.32(-0.03,0.5) 409.24(75.14,625.4) 2317(1987.9,6453.1) 0.91 -38.81 Linear Compositional -0.87(-1.15,-0.59) 4e-04(3e-04,5e-04) 0.79 -8.55 Functional -0.6(-0.84,-0.42) 3e-04(1.9e-04,3.7e-04) 0.79 -17.7 Phylogenetic -0.46(-0.59,-0.32) 2e-04(1.5e-04,2.6e-04) 0.81 -29.8 69 Thirty-seven percent, 71% and 61% of the communities had higher compositional, functional and phylogenetic turnover, respectively, than expected by chance between types of forests (Figure 4-1). Within the dry forest all of the comparisons where smaller than expected by chance according to the compositional turnover. Functional and phylogenetic turnover
showed that 71% and 67% of the comparisons respectively had smaller turnover than expected by chance, respectively. Within wet forest, on the other hand, the rates of compositional, functional and phylogenetic turnover showed that 67%, 52% and 19% of the comparisons where significantly smaller than expected by chance. Finally, the dry forests were significantly less variable than the wet forests, as suggested by a lower mean compositional, functional
and phylogenetic SES (Compositional; mean dry = -5.4, mean wet = -2.92, t = 3.9, df = 38.8, p>0.01; Functional; mean dry = -2.6, mean wet = -1.6, t = -3.5, df = 38.5, p=0.001; Phylogenetic; mean dry = -2.19, mean wet = -1.1, t = 3.29, df = 39.2, p>0.01). 4.3.2 Environmental Variables
We found that while mean annual temperature was constant among localities (Temperature = 27.7 − 3.8 × 10−4 × Rainfall; p =0.16), both temperature range (Temperature Range =
11.9 − 6.8 × 10−4 × Rainfall; p < 0.01, r 2 =0.57)andmeanmaximumtemperature (Max Temperature =35.9− 1.1 × 10−3 × Rainfall; p < 0.01 , r 2 =0.53)decreased with rainfall. Also, the coefficient of variation of hourly temperature decreased significantly with rainfall (Temperature CV =18.19− 3 × 10−3 × Rainfall; p < 0.01 , r 2 =0.93), suggesting that temperature is significantly less variable as rainfall increases. Finally,
rainfall seasonality and rainfall in the driest month significantly increased along the gradient (Seasonality =58.4− 5 × 10−3 × Rainfall, p < 0.01; r 2 =0.45;Min Rainfall = 39.7 + 0.09 × Rainfall, p < 0.01; r 2 =0.62) 4.3.3 Environmental Filtering
We found that community temperature range and rainfall seasonality decreased significantly (Temperature Range =15.05+Rainfall (−0.054); r 2 =0.59;Rainfall Seasonality =
68 − 4.6−3 × Rainfall; p < 0.01; r 2 =0.85)andcommunityminimumrainfalltoincreasewith
70 Compositional A B Functional Phylogenetic 40 4 8 0.2 0.0 0.2 0.4 − Ordination Axis 1 − Standardized Effect Size Effect Standardized 8 0.4 − −
1000 2000 3000 4000 Between Dry Wet Precipitation
Figure 4-1. Compositional, functional and phylogenetic turnover of lowland bird communities along the rainfall gradient of the Magdalena Valley, showing(A)asteepturnover around the 2300 mm rainfall isocline that is consistent among the measurements, but the measurements decrease in strength of turnover from compositional to phylogenetic, and (B) shows the distribution of the Standardized Effect Sizes for three types of comparisons: between and within wet and dry forests, showing higher than expected by chance turnover between types of forest and lower than expected by chance turnover within dry forests.
annual rainfall in each locality (Minimum Rainfall =66.3+0.05× Rainfall, p < 0.01; r 2 = 0.82;Figure4-2). We found no relationship between rainfall and community maximum
temperature (Maximum Temperature =31.9− 1 × 10−4 × Rainfall, p =13;r 2 =0.15; Figure 4-2). Physiological trait structure did not follow our predictions. The trait space in physiological tolerances was not smaller or less dispersed in dry forests as expected. This is shown by the non-significant relationship between community physiological richness or dispersion and rainfall (Richness =20.4+6.8× 10−4 × Rainfall, p =0.8;Dispersion =
1.44 + 6.5 × 10−6 × Rainfall, p =0.9). Finally, neither physiological richness nor dispersion of dry forest communities was lower than expected by chance. Among nest types, we found that cup and platform nests in dry forests had significantly lower differences between internal and external max temperatures. Only cup nests in dry
71 forests had significantly lower internal variance relative to the environmental variance when compared to wet forests (Table 4-3). In fact, the temperature variance in cup nests of dry forests was significantly lower than ambient temperature (Mean =0.5,df =3,p =0.03). This result means that variance in temperature of cup nests indryforestswas50%lowerthan
the variance in ambient temperature . The variance in nest temperature of platform nests in dry forests was also 45% lower than ambient variance, but thisdifference was not significant (Mean =0.55,df =1,p =0.06)
Table 4-3. Results of t-tests comparing max difference and variance ratio between nest and ambient temperature among nests in dry and wet forests. The results are product of multiple t-tests comparing types of nests and each type of nest between localities. For comparison among types of nests the objective was to determine if the difference between nest and ambient temperature was less thanzeroandlessthan one in the case of the ratio of variances. Type of Nest Dry Wet df p Max Difference Cup 4.69 9.6 8.5 > 0.01 Dome 5.93 4.17 2.96 1 Platform 3.65 9.12 2.99 0.02 Variance Ratio Cup 0.5 3.57 36.5 > 0.01 Dome 1.63 1.87 3.71 1 Platform 0.55 1.46 2.07 1
4.3.4 Biotic Interactions: Competition
As rainfall increases, the strength of environmental filtering should decrease and thus competition for resources should be more important in determining community structure. The competition hypothesis predicts that species co-occurringlocallyshouldbeecologicallyand consequently morphologically more diverse to avoid competition (MacArthur & Levins, 1967),
but we found no evidence for change in eco-morphological richness or dispersion with increasing rainfall (Functional Richness =86.7−2×10−4 ×Rainfall, p =0.99;Functional Dispersion = 1.45−9.8×10−6 ×Rainfall, p =0.9). Furthermore, only one site in the wet forest (Barbacoas) had eco-morphological richness higher than expected by chance (Figure 4-3). We found no
relationship between nest richness and rainfall (Nest Richness =5.25+3.9−4 Rainfall, p =
72 0.26), but nest dispersion increased with rainfall as predicted (Nest Dispersion =0.23+2.99× 10−5 Rainfall, p < 0.01; r 2 =0.58).
A B Temperature Range (C) Temperature Maximum Temperature (C) Temperature Maximum 9.5 10.0 10.5 31.25 31.50 31.75 32.00 1500 2500 3500 1500 2500 3500 Rainfall (mm) Rainfall (mm)
C D Rainfall Seasonality Rainfall Minimum Rainfall (mm) Rainfall Minimum 50 55 60 65 125 175 225 275 1500 2500 3500 1500 2500 3500 Rainfall (mm) Rainfall (mm)
Figure 4-2. Relationship between average community physiological tolerances and rainfall, showing, (A) no relationship between community average maximum temperature tolerance; (B) a negative relationship between average community temperature range and rainfall; (C), a decrease in mean community rainfall seasonality with locality rainfall and; (D) an increase in the minimum rainfall that species experience throughout their ranges.
73 4.4 Discussion
Our results suggest that there is deterministic bird community turnover around the
2400 mm annual rainfall isocline in the Magdalena Valley in Colombia. The rainfall gradient promoted a strong compositional, morphological and phylogenetic turnover in which almost the entire community was replaced in a short geographic distance. The models are strongly consistent with a stepwise function replacement of the communities (Table 4-1). Around the 2400 mm isocline there is up to a 75% change in the community, whereas in more than 200 km of dry or wet forest spanning a rainfall gradient of more than 1000mm on either side of this transitional zone, the average turnover among communities within the same type of forest was only 41%. Furthermore, our results partially suggest that environmental filtering might be of higher importance for structuring communities in the dry end of the Valley. Not only did dry forest communities have significantly less turnover thanexpectedbychance,theyalsohad lower turnover rates than wet forest communities (Figure 4-1). Species in the dry forests were also better adapted to higher rainfall seasonality and stronger dry seasons (Figure 4-2). In wet forests, we found evidence that competition for nest sites isstrongerthanindryforestsand the lower phylogenetic turnover compared with compositional turnover might be an indication of replacements among closely related ecologically similarspeciesthatdonotcoexistbecause of competition (Robinson & Terborgh, 1995). Nevertheless, there was little evidence that the communities were more dispersed in traits in wet forests thanindryforestssuggestingthat competitively structured niches are not necessarily more likely in wet than in dry forests.
Differences in temperature regulation within nest types between types of forests also point to the possibility that climate might be a determinant of community composition in dry forests. Cup and platform nests in the dry forests dampen the high environmental variability of the habitat whereas they do not in wet forests (Table 4-3). Our data also show that the difference in maximum inner and outer nest temperature is lower in cup and platform nests of dry forests, suggesting that that species might be more selective of the microclimates of nest sites in the dry than in the wet forests (Table 4-3). Such patterns might also result from
74 higher nesting seasonality in dry forest birds, which may only nest during the wet season when temperature variation is less extreme. Thus, an alternativepredictionthatarisesfromthe environmental filtering hypothesis is that there should be a decrease in nesting seasonality with increasing rainfall. Some studies suggest that in Amazonianwetforestsbirdsnestthroughout
the year, ignoring rainfall seasonality (Stouffer et al., 2013). In dry forests, however, we have no comparable data on the nesting phenology that could potentially support our hypothesis and predictions. One caveat that rises against the environmental filtering hypothesis is the low support for the prediction that dry forest species should be exposed to higher temperatures throughout
their ranges and that physiological trait space in dry forestcommunitiesshouldbesmallerand less dispersed compared to wet forest communities (Figure 4-2). On average species in the dry forests are not exposed to higher temperatures throughout their ranges than wet forest species and there was no relationship between physiologicaltraitrichnessanddispersionand
rainfall. Also, even though the relationship between mean temperature range of species in the community and rainfall differed significantly, the magnitudeofthedecreasewaslessthanin 0.5◦ C, which might not necessarily represent a strong selective agent. It is possible that the resolution of the environmental layers used to collect the data throughout the ranges of the
species was not high enough to capture the real strength of theenvironmentalfilteringindry forest. First, our data logger captured hourly and daily variability that were not represented in the broad-scale data. The data obtained across the ranges of species were rough estimates of mean maximum annual temperature and monthly temperature ranges. The hypothesis specifically deals with daily temperature in a few hours in a portion of the days of the year were
temperature rises above 40◦ C. Birds can potentially compete for nest resources, which might influence community assembly (Martin, 1988). We found support for this hypothesis as the dispersion in nesting types increased significantly with rainfall. Such patterns further support a shift in the
mechanisms that drive community composition along the gradient. One of the ways that
75 environmental filtering may be operating in the dry forests isthroughhighvariabilityand extreme high temperatures in the dry forests. Such mechanisms would predict lower functional dispersion of nesting types as the nests that better regulatetemperatureshouldbeselectively favored in this type of forest. We provide some evidence that cup and platform nests in
dry forests better regulate temperature than the same types of nest in wet forests, in which temperature extremes may not be great enough to require regulation of the microclimate. Nevertheless, our results indicate that temperature is a potential determinant of species composition and/or behavior. The increase in rainfall was associated with a decrease in temperature variability and maximum temperature. If temperature regulation is not a problem
in wet forests, it opens the possibility of a diversification in nest types to decrease the impact of competition. In dry forests, however, the reduction in nest types could increase competition as it is more critical for species to select for the best placestolocatenestsandavoidhigh temperatures. Thus, environmental filtering may increase competition for a potentially
limiting resource (i.e., nest sites), which could further constrain which species can occur in a community. Our functional trait data do not support the hypothesis that there is stronger competition for resources in the wet forests as there was no difference in the trait space of wet and dry
forests. Alternatively, competition for resources in the dry forest may occur at similar levels in both communities. Many studies have inferred that competition is an important determinant of bird species distributions and abundance (Jankowski et al., 2012), but few of these studies were conducted in dry forests, which have been historically understudied (Oswald et al., 2016). Thus, our data suggest that in addition to the environmental filtering, competition for resources might also influence dry forest communities. However, neither eco-morphological richness nor dispersion was higher than expected by chance inanyofthelocalities.The other potential explanation is that the morphological traits are not related to the niche axes that experience competition (Miles & Ricklefs, 1984; Ricklefs, 2012)orthattherelationship
between ecology and morphology is much more complex than previously thought (Pigot et al.,
76 2016). Thus, it is also possible that that competition happens through other niche axes that we were unable to detect in this study. One hypothesis that remains to be tested is the possibility that predation is stronger in wet forests, influencing community assembly (Martin, 1988; Jankowski et al., 2012). Many of
the most important nest predators were only found in the wet forest. Preliminary data suggest that the three toucan species exclusively found in wet forestduringmystudy(unpublished Data) are strong nest predators in these forests of the Magdalena Valley (G. Londono, unpublished data). In addition to the toucans, the number of forest raptors also increases as well as the richness of primates (Gomez et al unpublished data) in wet forests. While it has
been hypothesized that cavity nests might protect the eggs and nestlings from heavy rainfall (Oniki, 1985), there is more evidence to suggest that this type of nest provides protection against predators (Oniki, 1979, 1985). Thus, increased predation pressure in wet forests might select for the observed increase in cavity nesters and a decrease in cup nesters with rainfall.
Our data support this prediction (Figure A-5)butthemainassumption–thatnestpredationin dry forests is significantly lower than in wet forest– remainstobetested.Thus,ourdataare inconclusive about this hypothesis which we believe might beaninterestingonetotestinthe future.
In our study functional and phylogenetic turnover were lowerthancompositionalturnover. Alowerfunctionalthancompositionalturnoversuggeststhat there are some similar niches to be filled in both wet and dry forests, even though the niches arefilledbydifferent species with the same functional traits. This scenario would support a turnover mediated by interspecific competition (Robinson & Terborgh, 1995). In their work, Robinson & Terborgh (1995)report
that intrageneric replacements along a productivity gradient in lowland Amazonia responded to interspecific aggression between ecologically similar species. The heavier congener almost always actively displaced the smaller congener from the sites with higher productivity. We found several examples of replacement among ecologically similar species along the Magdalena
Valley that fit this scenario. For example, white-bellied antbird (Myrmeciza longipes)inthedry
77 forest is replaced by the chestnut-backed antbird (Myrmeciza exsul)inthewetforest.Both forage in similar habitats, close to the ground and potentially searching for similar insect items. An other example is the replacement of the endemic Apical flycatcher (Myiarchus apicalis) with its close relative dusky-capped flycatcher (Myiarchus tuberculifer). Both of this examples as well as some other ones occur among close relatives most likely in the same genus. Such patterns would lead to lower phylogenetic turnover. The functional and phylogenetic turnover, however, are still higher than expected by chance between forests and lower than expected by chance within forests suggesting a high change in function and evolutionary history of these communities with rainfall (Figure 4-1).
Even though we found a mismatch in the amount of turnover among compositional, functional and phylogenetic metrics, there is a spatial congruence in where the turnover happens (Table 4-2,Figure4-1). The three metrics predict that the community shift happens at the boundary delimiting the Magdalena dry forests and Magdalena-Uraba moist forests ecoregions (Olson et al., 2001). Provided that the ecoregions of northern South America where delimited by vegetation data (Olson et al., 2001), this suggests a spatial match in the turnover of bird and plant communities along the rainfall gradient. Others have found strong associations between the turnover of plant and bird communities (Jankowski et al., 2013), suggesting that vegetation might have a very strong influenceonthestructuringtropicalbird communities. There might be direct and indirect effects of vegetation on bird communities but we hypothesize that in the case of the Magdalena Valley theeffects are direct. The dry forest tree community is mainly deciduous, such that in the dry season, the entire forest loses its canopy over, potentially increasing temperaturesinsidetheforest,atleastduringthe day. In the wet forest, the canopy is more permanent throughout the year, which stabilizes temperature and eliminates the strong filtering by high temperatures. This hypothesis predicts that the limits of the dry and wet forest are associated with a strong change in the proportion of deciduous trees that compose the canopy.
78 In conclusion, we provide evidence that suggests that the mechanisms driving community assembly along the Magdalena Valley in Colombia change with precipitation. In localities with low rainfall (<2400 mm), we found evidence for environmental filtering, whereas in localities above the 2400 mm isocline we found only partial evidence supporting stronger biotic interactions (e. g., predation and nest site use). This changeinmechanismscanpotentially explain the strong compositional, functional and phylogenetic turnover that happens abruptly over a short geographic distance. The Magdalena River has been one of the major centers for development in Colombia since colonial times. The high within-forest community variability might reflect this long history of fragmentation and deforestation (Harrison, 1997; Pardini et al., 2005). We report here that the Magdalena Valley bird communities might be two separate entities with high functional and phylogenetic diversity. Despite its high diversity and high levels of fragmentation and deforestation, there are no protected areas in the region. Our data suggest that the upper and middle Magdalena Valley must be treated separately in conservation strategies.
79 A B Ecomorphological Richness Ecomorphological Dispersion 0 75 150 225 300 1.00 1.25 1.50 1.75 2.00 1500 2500 3500 1500 2500 3500 Rainfall (mm) Rainfall (mm)
C D Nest Richness Nest Dispersion 4567 0.20 0.25 0.30 0.35 1500 2500 3500 1500 2500 3500 Rainfall (mm) Rainfall (mm)
Figure 4-3. Eco-morphological and nest structure of communities along the rainfall gradient of the Magdalena Valley. A and B show no relationship between ecomorphological richness and dispersion and rainfall. C shows a slight but notsignificantincreasein nest richness with rainfall and D shows a significant increaseinnestdispersion throughout the gradient. Grey triangles indicate localities in which functional richness or dispersion was higher (triangles pointing up) orlower(trianglespointing down) than expected by chance
80 CHAPTER 5 CONCLUDING REMARKS The objective of this thesis was to identify the mechanisms driving bird community assembly along a steep environmental gradient. Before making any inferences, I needed to identify methodologies that accurately estimate the abundance of species. Because most of the methods used for inferring mechanisms are based on the shapes of the species abundance distribution, the first step required an investigation of thesamplesizesneededtoestimate densities of bird populations. Surprisingly, I found that using the n-mixture models with count data from fixed-radius point counts requires large sample sizes to estimate the abundance of the many rare species characteristic of neotropical forests In Chapter 2,Idevelopedan extension of the existing models that might allow investigators to get around these limitations in field studies by taking advantage of correlated behaviors among species. All of the published models so far treat detection probability as an intrinsic property of each species, which requires one parameter per species considered in the model. I believe that this is not always necessary because some groups of species might have similar detection probabilities that can potentially be drawn at random from a beta distribution. In this case, I canstillmakemechanistic inferences about detection probabilities because we assume, for example, that different foraging guilds have radically different detection probabilities, but the species within guilds do not. My working hypothesis to explain the turnover of bird communities along the Magdalena Valley was that there is a change in the mechanisms driving such community structure as rainfall patterns change. I hypothesized that environmental filtering determined community structure in the dry forests and that biotic interactions were more important at the wet end. To test this hypothesis I used two different approaches: (1) a modeling approach to estimate migration of species from the metacommunity to the local communities (Chapter 3), and (2) a description of the patterns of community similarity andstructureincomposition,traits and phylogeny along the gradient, which allowed me to make inferences about the likely mechanisms influencing bird distribution and abundance (Chapter 4).
81 Chapter 3 was based on the assumption that the entire rainfallgradientoftheMagdalena Valley is a metacommunity in which localities are connected through immigration with dynamics regulated by deterministic and stochastic processes (Leibold et al., 2004). I used a model that accounts for stochasticity in community assembly to test the hypothesis that species immigrating from the wet forests were not able to recruit into (i. e., they were filtered out of) dry forests. This hypothesis predicts that immigration from the metacommunity to the local communities is lower in the dry forests compared to the wet forests (Jabot et al., 2008). Furthermore, I asked if wet forests also imposed strong restrictions to immigration of dry forest birds, in which case dry and wet forests would behave as independent metacommunities with no immigration. While I found a clear relationship between immigration rates and rainfall, I found stronger support for the hypothesis that dispersal of bird species is restricted in both directions; from dry to wet forests and vice versa. I attribute these restrictions on immigration to the effects of environmental filtering in the dry forests and biotic interactions (e. g., competition and predation) in wet forests. I further provideahypothesistoexplainwhythere are fewer rare species in dry forests. If species immigrate asrarespecies,strongerselection against maladapted immigrants in the dry forest would eliminate entire populations of rare species faster than in the wet forests.
Chapter 3 provided the basis for Chapter 4 because it provided insights about the possible mechanisms explaining community composition and species distribution, which Ifurtherattemptedtotestinchapter4.BecauseIfoundevidence that the Magdalena Valley dry forests are part of a different metacommunity than the wet forests, I predicted that the community turnover would be a non-linear, stepwise function of rainfall in which communities change rapidly with changes in rainfall, regardless of geographic proximity. Studies of coordinated changes in communities along gradients date back to the classic ideas of community assembly by Gleason (1926, 1939)andClements (1916, 1936). While Gleason (1926)hypothesizedthatcommunitieschangegraduallyalonganenvironmental gradient as a product of individualistic species associations with the environment, Clements (1936)
82 hypothesized that communities were strongly determined by the environment and interactions among species and that communities, therefore, would undergo wholesale turnover at various points along the gradient. These seemingly opposite points of view are still the basis of hypotheses that provide alternative, testable predictionsabouttheextenttowhichcommunity turnover is gradual or abrupt along gradients. In Chapter 4,Icomparedtaxonomic,phylogeneticandfunctionalturnover of bird communities on the precipitation gradient to test these predictions. I found strong evidence that community turnover was explained better by rainfall than by geographic distance. Instead of being a linear turnover, we found a turnover point at the 2400 mm isocline where most of the community is replaced. On either side, there was little to no change in the communities above or below this isocline. This turnover is perfectly consistent with the transition between two terrestrial ecoregions defined for this area by WWF (Olson et al., 2001). Based on these parallel changes, I hypothesize that this turnover is strongly correlated with changes in the canopy tree community that transitions from a mainly deciduous to a mainly evergreen species. It is likely that this transition strongly affects temperature variability within forests, especially during the dry season. I collected temperature data in several localities for over two years, and even though a typical dry forest localities have similar mean temperatures to those in wet forests, temperature variation is radically different. Somedaysinthedryforestcanbeashot as 42 − 43◦Cwhilethemaximumtemperatureinsidewetforestsisaround35◦C. Such high temperatures can potentially affect recruitment by killing eggs and nestlings of birds that are not adapted to those temperatures. In Chapter 4,Ifoundpartialsupportforcommunities being environmentally filtered in dry forests, particularlybecauseofharshclimateandhigh temperatures influencing nesting behavior. In wet forest, the mechanisms seem more complicated than a scenario in which community assembly is mainly influenced by competition. We did not find any support for higher morphological diversity in wet forests, which suggests thathistoricalcompetition(niche divergence) might not have had an impact on morphological diversity. Nest diversity, however,
83 did significantly increase with rain, suggesting that nests are restricted in the dry forest, probably by environmental factors, and species increase diversity to avoid competition in wet forests. Competition, however, can influence community assembly on two very different time scales. The first one is associated with the long-standing idea of the ”ghost of competition
past”, in which historical interactions among species promote differentiation of species present in a community to reduce fitness consequences of interspecificcompetition(MacArthur, 1958). The second is more related to ongoing competition that might explain not only local segregation of habitats but potentially abundance of species that are similar in their use of resources and therefore likely to be competing currently (Robinson & Terborgh, 1995; Russo
et al., 2003). One caveat of our analysis is that it is only capable of detecting historical competition, which seems to be a weak factor structuring the communities in the wet forest. Iadvocateforfurthertestsofongoingcompetitionthatcanpotentially determine species abundance distributions.
The high community turnover over a short geographic distances and the high functional turnover associated with the climatic environment providessomesupportforClements’ hypothesis that communities are a collection of coevolving organisms (Clements, 1936). Even though Gleason (1926)wasprobablythestrongestopponentoftheformeridea,several
passages suggested that he believed that environment could be a strong determinant for change in community assembly. For example, in his essay he wrote: ”At the margin of an association, it comes in contact with another, and there is a transition line or belt between them. In many instances, particularly where there is an abrupt change in the environment, this transition line is very narrow and sharply defined, so that a single step may sometimes be sufficient to take the observer from one into another.” (Gleason, 1926:pages11-12). Although in the Magdalena Valley, the change in climatic conditions is gradual, the change in the bird community is abrupt. I argue that one potential explanation for this abrupt change is the sudden change in plant community composition. I do not know of any published
study of the plants in the Magdalena Valley, but it is easy to observe during the dry season
84 that the proportion of deciduous trees in the canopy around the 2400 mm rainfall isocline changes quickly from high to low. The 2400 mm isocline seems high for delimiting dry forests, particularly because they are usually associated with annual rainfall below 1800 mm (Murphy & Lugo, 1986). I hypothesize that because the seasonal patterns of rainfall in the Magdalena
Valley may further limit plant and bird communities. The MV has a bimodal distribution with two extremely dry seasons that might limit evergreen trees towetterconditions.Thispattern of seasonality may further filter bird species that can adapt their life histories to two different breeding seasons annually. Such apparent strong dichotomy between Gleason’s and Clements’ ideas might be some sort of misinterpretation. While one was more deterministically driven, the other believed more in stochastic dispersion, but in the mid point, both of them argued that climate and soil characteristics strongly determined plant communities. Even Gleason’s extremely stochastic point of view might have been a semantic misinterpretation bythemisuseoftheuniform and random distributions to analyze his data (Nicolson et al., 2002). Throughout Gleason’s writings, there is evidence that even though he believed thatspecieswereassembledtogether in a community by pure coincidence, the passage above and many other’s throughout his seminal papers clearly does not reflect the same ideas (Nicolson et al., 2002). This is to say that, with my dissertation, I believe that I provide evidence that both deterministic and stochastic mechanisms operate to create the patterns observed in the communities along the Magdalena Valley. Such Gleasonian versus Clementsian dichotomy is similar to the niche versus neutral theory dichotomy. I personally believe that this is not truly a dichotomy but instead are part of a set of mechanisms that can operate at the same time. The use of mathematical models that incorporate both types of processes can improve our understanding of the overall nature of community assembly and provide powerful tools to predict diversity patterns. They might also be more objective by clearly defining parameters and relationships among them, derived from hypotheses provided by the background theory. Such models can act as a bridge for the reconciliation of niche and neutral models as has happened in other areas of the
85 biological sciences (e.g. population genetics). Such a bridge can also alleviate much of the rivalry among niche and neutral theory supporters, which canhelpustomoveforwardin understanding how biological diversity arises. In conclusion, the pattern-driven studies that I present in this dissertation suggest that precipitation gradients are strong contributors to the determination of community structure and delimitation of species distributions. The analyses of these patterns presented in this thesis suggest which mechanisms underly community structure. Future mechanistic studies will be needed to test the hypotheses provided here. I hope that the hypotheses and methods derived from my study will serves a a foundation for future studies in similar systems.
86 APPENDIX A SUPPLEMENTARY FIGURES This section contains the supplementary figures for Chapter 2 and 4.Thefiguresare referenced in the main text and the labels should be self-explanatory.
87 iueA1 enba nma ubro niiul e 0 ha 100 per individuals of number mean in bias Mean A-1. Figure Number of Point Counts 5 20 5 20 5 20 5 20 5 20 5 20 5 20 5 20 5 20 5 20 5 20 5 20 5 20 5 20 5 20 λ λ λ λ λ λ λ λ λ λ λ λ λ λ λ ======01 20 15 10 5 100 85 75 65 55 40 25 15 10 7 5 4 3 2 1 p =0.1 ubro elcts n reprmtrvle ofrlw m low, for ( to probabilities values detection parameter and true abundances and replicates, of number sabakln nec ftepanels. the th of is each which in 0.1, line select of black We abundance a right. of as the estimation in in (ligh presented bias low is acceptable from scale bias color the The represent (black). panel each in grayscale The 01 20 15 10 5 p =0.2 01 20 15 10 5 p =0.3 01 20 15 10 5 p =0.4 Number ofReplicates 01 20 15 10 5 p =0.5 88 λ 01 20 15 10 5 =7,25,65,100 p =0.6 01 20 15 10 5 p =0.7 λ o ag fpitcounts, point of range for and datrsodfor threshold a ed tgray)tohigh 01 20 15 10 5 dadhigh and id eisoclinepresented p =0.8 p =0.2,0.5,0.8 01 20 15 10 5 p =0.9
). 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 10 bias in λ iueA2 itga fetmtddtcinprobabilitie detection estimated of Histogram A-2. Figure eamdli loson(lc otdln)bsdo h para the on probabi based line) detection dotted (black the shown of α also is fo distribution model dry The beta the of Colombia. birds insectivorous Valley understory 26 of estimates =3.15
^
and Estimated Abundance (λ)
0 2000 4000 6000 0 2000 4000 6000 0 2000 4000 6000 8000 β 04 75 40 10 4 1 =12.7 mxueMdlBetamixtureModel Nmixture Model True Abundance ( 89
0 20 40 60 80 100 140 0 50 100 150 0 50 100 150 200 04 75 40 10 4 1 λ ) sb a s e do nt h eN - m i x t u r em o d e l ie siae ythe by estimated lites eto h Magdalena the of rest
eesetmtdas estimated meters p = 0.75 p = 0.5 p = 0.25 iueA3 apo hwn h itiuinof distribution the showing Barplot A-3. Figure eaNmxuemdl aebe mte o clarity. for omitted been have models N-mixture beta hwn h oaino h revleof value true the of location the showing
^ Estimated Abundance (λ)
0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 5000 0 1000 2000 3000 4000 55 5100 75 50 25 0 mxueMdlBetamixtureModel Nmixture Model True Abundance ( 90 λ ˆ 20 40 60 80 100 20 40 60 80 100 20 40 60 80 100 120 sn -itr n eaNmxuemodels, N-mixture beta and N-mixture using 55 5100 75 50 25 0 λ λ .T h eo u t l i e r sf o rt h eN - m i x t u r ea n d )
p = 0.75 p = 0.5 p = 0.25 Frequency 0246
0.0 0.1 0.2 0.3 0.4 Detection Probability
Figure A-4. Relationship between the mean value of λˆ from the 1000 simulations and the true value of λ.Forreference,weshowtheone-to-onerelationshipline(gray dotted line).
91 Cup Nests Cavity Nests Relative Abundance Relative 0.0 0.2 0.4 0.6 0.8 1.0 1500 2500 3500 Rainfall (mm)
Figure A-5. Relative abundance of Cavity and Cup nests along the rainfall gradient of the Magdalena Valley. Gray and black dotted lines are the fitted lines from a beta regression for the relationship between cup and cavity nestsandrainfall.
92 APPENDIX B SUPPLEMENTARY TABLES This section contains the supplementary tables referenced throughout the text.
93 Table B-1. Population density estimates for species in each of the 13 localities. The estimates are the results from the Zero-Inflated
Poisson model and extrapolated to 100ha. The population of each species in each local community is given in individuals/100ha. Localities are ordered in increasing precipitation from left to right. Barbacoas San Juan Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios
Tinamiformes Tinamidae Tinamusmajor 0 0 0 0 0 0 14 4 3 5 9 15 52
Tinamiformes Tinamidae Crypturellussoui 3 4 5 5 9 10 18 17 15 11 10 86
Tinamiformes Tinamidae Crypturellus erythropus 0 0 0 0 2 2 6 3 1 0 0 0 0 94 Galliformes Cracidae Penelopepurpurascens 0 0 0 0 0 0 3 1 1 1 2 4 22
Galliformes Cracidae Ortaliscolumbiana 10 18 37 37 23 18 3 3 4 8 11 15 30
Galliformes Odontophoridae Colinuscristatus 1 4 9 9 0 0 0 0 0 0 0 0 0
Galliformes Odontophoridae Odontophorusgujanensis 0 0 0 0 0 0 0 0001500
Accipitriformes Accipitridae Accipiterbicolor 0 0 0 3 0 0 0 0 0 0 0 00
Accipitriformes Accipitridae Rupornis magnirostris 0 0 0 0 0 0 2 2 2 2 111
Accipitriformes Accipitridae Buteonitidus 0 11 0 0 0 0 0 0 0 0 0 0 0
Gruiformes Rallidae Aramidescajaneus 1 1 1 1 0 0 0 0 0 1 1 1 2
Gruiformes Rallidae Laterallusalbigularis 0 0 3 0 0 0 0 0 0 0 0 0 0
Charadriiformes Charadriidae Vanellus chilensis 19 12 8 8 6 3 0 0 0 0 000 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Charadriiformes Burhinidae Burhinus bistriatus 0 11 0 0 0 0 0 0 0 0 0 0 0
Columbiformes Columbidae Patagioenasspeciosa 0 0 0 0 0 0 0 2 19 61 31 11 1
Columbiformes Columbidae Patagioenas cayennensis 9 5 5 5 22 2213131416181923
Columbiformes Columbidae Zenaidaauriculata 0 6 16 0 0 0 0 0 0 0 0 0 0
Columbiformes Columbidae Leptotila verreauxi 106 109 112 112 100 94 23 10 7 2 1 1 0
Columbiformes Columbidae Leptotilacassini 0 0 0 0 0 0 0 0 0 0 0 0 10 95 Columbiformes Columbidae Geotrygonmontana 0 0 0 0 0 0 3 3 3 2 2 2 1
Columbiformes Columbidae Columbina talpacoti 5 15 47 49 30 15 0000000
Columbiformes Columbidae Claravis pretiosa 85 86 47 20 58 0 0 0 0 0 000
Cuculiformes Cuculidae Piayacayana 5 5 5 5 7 7 9 9 9 8 8 8 7
Cuculiformes Cuculidae Coccyzusamericanus 0 0 0 0 0 0 0 0 0 0 0 15 0
Cuculiformes Cuculidae Crotophagamajor 0 0 0 0 0 0 0 7 0 0 0 0 0
Cuculiformes Cuculidae Crotophagaani 0 1 5 6 4 2 0 0 0 0 0 0 0
Cuculiformes Cuculidae Crotophagasulcirostris 0 11 46 3 0 0 0 0 0 0 000
Cuculiformes Cuculidae Taperanaevia 0 2 8 8 0 0 0 0 0 0 0 0 0
Cuculiformes Cuculidae Dromococcyxphasianellus 1 1 1 1 0 0 0 0 0 1112 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Trogoniformes Trogonidae Trogon comptus 0 0 0 0 0 0 0 1 6 10 4 1 0
Trogoniformes Trogonidae Trogon melanurus 0 0 0 0 0 0 0 0 0 2 26 62 14
Trogoniformes Trogonidae Trogon chionurus 0 0 0 0 0 1 34 44 41 20 13 9 4
Trogoniformes Trogonidae Trogon caligatus 0 0 0 0 0 1 14 22 25 23 21 1813
Trogoniformes Trogonidae Trogon rufus 0 0 0 0 0 0 0 0 0 0 5 0 0
Coraciiformes Alcedinidae Megaceryletorquata 0 0 6 7 4 0 0 0 0 0 0 0 0 96 Coraciiformes Momotidae Electron platyrhynchum 0 0 0 0 0 0 0 3 15 14 510
Coraciiformes Momotidae Baryphthengusruficapillus 0 0 0 0 0 0 31 2824171272
Coraciiformes Momotidae Momotus subrufescens 4 5 10 10 26 29 30 24211513129
Galbuliformes Galbulidae Galbularuficauda 53 57 61 61 59 57 28 252527282830
Galbuliformes Bucconidae Notharchuspectoralis 0 0 0 0 0 0 12 0 0 0 53810
Galbuliformes Bucconidae Notharchustectus 0 0 0 0 0 0 1 2 3 18 34 600
Galbuliformes Bucconidae Nystalusradiatus 9 9 8 8 7 7 5 4 4 3 3 3 2
Galbuliformes Bucconidae Nonnulafrontalis 0 0 0 0 0 0 0 0 0 0 1 3 17
Piciformes Capitonidae Capitohypoleucus 0 0 0 0 0 0 0 7 0 0 0 0 0
Piciformes Ramphastidae Ramphastosambiguus 0 0 0 0 0 0 6 12 16 33 40 48 62 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Piciformes Ramphastidae Ramphastosvitellinus 0 0 0 0 1 1 81 51 3218181921
Piciformes Ramphastidae Pteroglossus torquatus 0 0 0 0 0 0 57 70 54 26221915
Piciformes Picidae Picumnus olivaceus 12 14 18 18 29 31 3 2 2 5 7 9 9
Piciformes Picidae Melanerpes pulcher 0 0 0 0 0 0 1 2 3 5 6 7 8
Piciformes Picidae Melanerpes rubricapillus 5 6 8 8 13 13 5 2 1 0 0 0 0
Piciformes Picidae Veniliornis kirkii 10 11 12 12 16 16 20 19 19 16 15 14 12 97 Piciformes Picidae Colaptespunctigula 0 0 3 10 4 0 0 0 0 0 0 0 0
Piciformes Picidae Celeus loricatus 0 0 0 0 0 0 92 21 14 16 23 35 93
Piciformes Picidae Dryocopus lineatus 4 3 3 3 3 3 2 2 2 2 2 2 2
Piciformes Picidae Campephilusmelanoleucos 10 4 0 0 0 0 16 2 1 1 2 4 19
Falconiformes Falconidae Herpetotheres cachinnans 12 10 8 7 4 4 2222233
Falconiformes Falconidae Micrasturmirandollei 0 0 0 0 0 0 7 1 1 2 5 11 60
Falconiformes Falconidae Micrastursemitorquatus 1 2 3 3 0 0 0 1 5 30 8 2 0
Falconiformes Falconidae Milvago chimachima 12 11 11 11 10 9 10111113141517
Falconiformes Falconidae Falcorufigularis 0 0 0 0 4 0 0 0 0 0 0 0 0
Psittaciformes Psittacidae Pyrilia pyrilia 0 0 0 0 0 0 0 4 10 28 29 27 17 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Psittaciformes Psittacidae Pionus menstruus 0 1 2 2 10 12 42 37 31 16 11 8 4
Passeriformes Thamnophilidae Cymbilaimuslineatus 3 3 3 3 3 3 4 5588911
Passeriformes Thamnophilidae Tarabamajor 0 0 0 0 0 0 1 1 2 2 2 2 2
Passeriformes Thamnophilidae Thamnophilusdoliatus 15 125 142 142 28 14 0 0 0 0 0 0 0
Passeriformes Thamnophilidae Thamnophilusatrinucha 40 44 51 51 63 64 51 41 34 21 17 14 9
Passeriformes Thamnophilidae Thamnophilusnigriceps 0 0 0 0 0 0 42 15 10 10 13 18 38 98 Passeriformes Thamnophilidae Epinecrophylla fulviventris 0 0 00003213015841
Passeriformes Thamnophilidae Myrmotherulapacifica 1 1 0 0 0 0 6 62 00000
Passeriformes Thamnophilidae Myrmotherula axillaris 0 0 0 0 0 0 16 28 32 28 25 20 14
Passeriformes Thamnophilidae Microrhopiasquixensis 0 0 0 0 0 0 0 0690000
Passeriformes Thamnophilidae Formicivora grisea 41 59 90 91 75 0 0 000000
Passeriformes Thamnophilidae Cercomacra tyrannina 0 0 0 0 0 46 0 21 00000
Passeriformes Thamnophilidae Cercomacra nigricans 12 13 16 16 49 61 8 2 1 0 0 0 0
Passeriformes Thamnophilidae Gymnocichlanudiceps 0 0 0 0 0 0 2 222111
Passeriformes Thamnophilidae Myrmeciza longipes 25 34 58 59 80 68 0 0 0 0 0 0 0
Passeriformes Thamnophilidae Myrmecizaexsul 0 0 0 0 0 0 180 53 38 47 64 85 53 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Passeriformes Thamnophilidae Myrmecizapalliata 0 0 0 0 0 0 0 7 0 0 000
Passeriformes Thamnophilidae Myrmecizaimmaculata 0 0 3 0 0 0 0 0 00000
Passeriformes Thamnophilidae Gymnopithysleucaspis 0 0 0 0 0 0 05107420
Passeriformes Grallariidae Hylopezus perspicillatus 0 0 0 0 0 0 0 0 0010180
Passeriformes Formicariidae Formicarius analis 0 0 0 0 0 0 1 7 12 14 11 73
Passeriformes Furnariidae Sclerurus guatemalensis 0 0 0 0 0 0 0 0 15 8 890 99 Passeriformes Furnariidae Dendrocincla fuliginosa 1 1 1 1 2 3 9 11 1316171718
Passeriformes Furnariidae Glyphorynchus spirurus 0 0 0 0 0 0 5 11 16 27 30 31 31
Passeriformes Furnariidae Dendrocolaptes sanctithomae 0 0 0 0 0 0 810107542
Passeriformes Furnariidae Xiphorhynchus susurrans 49 48 46 46 44 44 48 52 55 64 68 72 79
Passeriformes Furnariidae Xiphorhynchus lachrymosus 0 0 0 0 0 0 0 0 0 5122022
Passeriformes Furnariidae Dendroplex picus 9 4 46 46 30 27 6 4 3 1 1 1 1
Passeriformes Furnariidae Campylorhamphus trochilirostris 9 9 8 8 7 72110000
Passeriformes Furnariidae Lepidocolaptes souleyetii 9 11 14 14 3437333020181511
Passeriformes Furnariidae Xenopsminutus 3 3 4 4 9 10 23 24 23 19 18 1613
Passeriformes Furnariidae Furnarius leucopus 0 0 0 0 0 0 0 0 0 0 0 3 0 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Passeriformes Furnariidae Philydor fuscipenne 0 0 0 0 0 0 0 0 15 85 8 0 0
Passeriformes Furnariidae Automolus ochrolaemus 0 0 0 0 0 0 1 2 3 7 9 1115
Passeriformes Furnariidae Synallaxisalbescens 0 0 0 36 4 1 0 0 0 0 0 00
Passeriformes Tyrannidae Phyllomyias griseiceps 2 2 2 2 4 4 11 14 16 20 22 23 25
Passeriformes Tyrannidae Tyrannuluselatus 2 5 17 18 37 37 37 37 37 38 39 39 40
Passeriformes Tyrannidae Myiopagis gaimardii 32 6 17 20 29 29 24 252630323438 100 Passeriformes Tyrannidae Myiopagisviridicata 30 27 22 21 10 8 0 0 00000
Passeriformes Tyrannidae Elaeniaflavogaster 5 25 98 100 12 4 0 0 0 0000
Passeriformes Tyrannidae Ornithion brunneicapillus 0 0 0 0 0 0 35 15101115231
Passeriformes Tyrannidae Camptostoma obsoletum 15 22 36 36 36 272 223331
Passeriformes Tyrannidae Phaeomyias murina 9 60 81 80 14 9 0 0 0 0 0 0 0
Passeriformes Tyrannidae Capsiempis flaveola 20 63 160 162 8 0 0 0 00000
Passeriformes Tyrannidae Euscarthmus meloryphus 2 22 39 37 0 0 0 0 0 0000
Passeriformes Tyrannidae Zimmerius chrysops 0 0 0 0 0 0 1 8 15 10 5 2 0
Passeriformes Tyrannidae Phylloscartes lanyoni 0 0 0 0 0 0 0 7 0 0 0 0 0
Passeriformes Tyrannidae Mionectes oleagineus 4 5 7 7 12 13 26 29 31 33 33 34 33 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Passeriformes Tyrannidae Leptopogon amaurocephalus 10 11 12 12 16 17 26 28 29 31 31 31 31
Passeriformes Tyrannidae Leptopogon superciliaris 0 0 0 0 0 0 0 36 310 0 0 0
Passeriformes Tyrannidae Oncostomaolivaceum 0 0 0 0 0 0 0 2 6 25 30 33 32
Passeriformes Tyrannidae Lophotriccuspileatus 0 0 0 0 0 0 4 0 0 0 0 0 0
Passeriformes Tyrannidae Atalotriccus pilaris 43 46 51 51 50 48 11 531000
Passeriformes Tyrannidae Hemitriccus margaritaceiventer 25 40 626224140 0 00000 101 Passeriformes Tyrannidae Poecilotriccus sylvia 34 34 35 35 31 30 11 6 5 2 1 1 1
Passeriformes Tyrannidae Todirostrum cinereum 0 0 42 54 35 31 6 4 3 1 1 11
Passeriformes Tyrannidae Todirostrum nigriceps 1 1 2 2 7 9 42 50 51 44 40 36 29
Passeriformes Tyrannidae Cnipodectessubbrunneus 0 0 0 0 0 0 1 5 8 8 752
Passeriformes Tyrannidae Rhynchocyclusolivaceus 20 22 26 26 323328211811985
Passeriformes Tyrannidae Tolmomyias sulphurescens 46 47 49 49 525243373325232117
Passeriformes Tyrannidae Tolmomyiasflaviventris 0 0 0 0 0 0 1 2 3 20 37 68 46
Passeriformes Tyrannidae Myiobiusatricaudus 0 0 0 0 0 0 0 7 15 0 0 0 0
Passeriformes Tyrannidae Aphanotriccusaudax 0 0 0 0 0 0 2 2 2 2 1 1 1
Passeriformes Tyrannidae Cnemotriccusfuscatus 29 4 5 5 24 10 0 0 0 0000 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Passeriformes Tyrannidae Contopuscinereus 0 0 3 0 0 0 0 0 0 0 0 0 0
Passeriformes Tyrannidae Pyrocephalus rubinus 0 0 6 3 0 0 0 0 0 0 0 0 0
Passeriformes Tyrannidae Coloniacolonus 0 0 0 0 0 0 0 7 0 0 0 0 0
Passeriformes Tyrannidae Machetornis rixosa 0 0 3 0 0 0 0 0 0 0 0 0 0
Passeriformes Tyrannidae Legatusleucophaius 5 6 8 8 18 21 47 47 4534302620
Passeriformes Tyrannidae Myiozetetes cayanensis 31 34 39 39 43 43 16 8 6 2 1 1 0 102 Passeriformes Tyrannidae Myiozetetessimilis 0 0 0 33 0 0 0 0 0 0 0 0 0
Passeriformes Tyrannidae Pitangus sulphuratus 3 6 12 12 38 43 9 2 1 0 000
Passeriformes Tyrannidae Pitangus lictor 0 0 0 0 4 0 0 0 0 0 0 0 0
Passeriformes Tyrannidae Myiodynastesmaculatus 0 1 5 5 17 12 2 6 11 2 0 0 0
Passeriformes Tyrannidae Megarynchuspitangua 5 9 20 21 14 11 2 1 1 1122
Passeriformes Tyrannidae Tyrannus melancholicus 12 18 30 30 30 244 344321
Passeriformes Tyrannidae Rhytipterna holerythra 0 0 0 0 0 0 12 14 13 7 542
Passeriformes Tyrannidae Myiarchus tuberculifer 1 1 1 1 2 3 11 14 16 17 17 16 14
Passeriformes Tyrannidae Myiarchuspanamensis 0 0 0 0 0 0 5 6 5 3 2 1 1
Passeriformes Tyrannidae Myiarchusapicalis 9 13 21 21 20 16 0 0 0 0 000 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Passeriformes Tyrannidae Attilaspadiceus 0 0 0 0 1 1 28 38 38 25 20 15 9
Passeriformes Pipridae Chiroxiphia lanceolata 74 81 90 90 78 71 4 1 00000
Passeriformes Pipridae Manacus manacus 5 6 9 9 19 21 51 57 59 56 54 50 43
Passeriformes Pipridae Machaeropterus regulus 0 0 0 0 0 0 6 23 27 10 5 2 0
Passeriformes Pipridae Ceratopipra erythrocephala 0 0 0 0 0 0 7 28 42 53 47 39 24
Passeriformes Tityridae Schiffornis turdina 0 0 0 0 0 0 1 2 3 15 23 35 72 103 Passeriformes Tityridae Laniocera rufescens 0 0 0 0 0 0 0 0 0 0 0 0 10
Passeriformes Tityridae Pachyramphuscinnamomeus 0 1 1 2 4 4 10 11 92100
Passeriformes Tityridae Pachyramphus polychopterus 0 0 1 1 3 3 0 0 0 0000
Passeriformes Tityridae Pachyramphus homochrous 0 0 0 0 0 0 4 13 18 1713105
Passeriformes Incertae Sedis Piprites chloris 0 0 0 0 0 0 0 0 31 0 0 0 0
Passeriformes Vireonidae Cyclarhis gujanensis 8 25 94 98 63 29 0 0 00000
Passeriformes Vireonidae Hylophilus flavipes 69 80 96 97 105 100 11 3 1 0 0 0 0
Passeriformes Vireonidae Hylophilusdecurtatus 0 0 0 0 0 0 0 5 58 23 300
Passeriformes Corvidae Cyanocorax affinis 33 33 34 34 33 32 25 22 20 16141311
Passeriformes Troglodytidae Microcerculus marginatus 0 0 0 0 0 0 42 65 57 17 9 5 1 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Passeriformes Troglodytidae Troglodytes aedon 1 5 38 33 1 22 0 0 0 0 000
Passeriformes Troglodytidae Campylorhynchuszonatus 0 0 0 0 0 0 1 3516223048
Passeriformes Troglodytidae Campylorhynchus griseus 2 3 3 3 3 3 2 1 11000
Passeriformes Troglodytidae Pheugopediusspadix 0 0 0 0 0 0 0 7 0 0 0 00
Passeriformes Troglodytidae Pheugopedius fasciatoventris 0 6 7575625949525463677281
Passeriformes Troglodytidae Cantorchilus nigricapillus 0 0 0 0 0 01038419310 104 Passeriformes Troglodytidae Cantorchilusleucotis 0 0 0 0 0 0 1 1 2 4569
Passeriformes Troglodytidae Henicorhina leucosticta 0 0 26 26 242319181716161515
Passeriformes Troglodytidae Cyphorhinusphaeocephalus 0 0 0 0 0 00080000
Passeriformes Polioptilidae Microbates cinereiventris 0 0 0 0 0 0 092015730
Passeriformes Polioptilidae Ramphocaenusmelanurus 0 0 12 0 0 0 0000000
Passeriformes Polioptilidae Polioptila plumbea 20 25 36 37 28 19 1 1 2 11 17 20 17
Passeriformes Donacobiidae Donacobiusatricapilla 0 0 0 0 0 0 0 0 00030
Passeriformes Turdidae Catharus aurantiirostris 0 0 0 0 0 8 0 0 0 0 0 0 0
Passeriformes Turdidae Catharusustulatus 3 5 7 8 10 10 0 0 0 0 0 0 0
Passeriformes Turdidae Turdus leucomelas 20 23 27 28 38 40 30 20 15 7643 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Passeriformes Turdidae Turdusignobilis 0 0 22 22 20 19 7 4 3 1 1 1 0
Passeriformes Thraupidae Eucometis penicillata 19 18 17 17 14 1412131315161718
Passeriformes Thraupidae Tachyphonusluctuosus 7 8 9 9 12 13 20 212118161411
Passeriformes Thraupidae Tachyphonusdelatrii 0 0 0 0 0 0 0 143 0 0 0 00
Passeriformes Thraupidae Ramphocelusdimidiatus 8 13 24 24 32 288 894210
Passeriformes Thraupidae Thraupisepiscopus 13 14 16 16 19 19 10 6 42110 105 Passeriformes Thraupidae Thraupis palmarum 0 0 1 1 3 4 3 1 0 0 0 0 0
Passeriformes Thraupidae Tangara vitriolina 32 62 83 83 4 1 0 0 0 0 0 0 0
Passeriformes Thraupidae Tangara larvata 0 0 1 1 4 6 23 20 17 7 5 3 1
Passeriformes Thraupidae Tangaracyanicollis 0 1 4 5 5 2 0 0 0 0 0 0 0
Passeriformes Thraupidae Tangara inornata 0 0 0 0 0 0 0 7 0 0 0 0 0
Passeriformes Thraupidae Tangara gyrola 0 0 0 1 2 3 6 4 3 1 0 0 0
Passeriformes Thraupidae Dacnislineata 2 1 1 1 1 1 1 1 1 2 2 3 4
Passeriformes Thraupidae Dacniscayana 0 0 0 0 0 0 0 0 0 0 0 6 0
Passeriformes Thraupidae Cyanerpes caeruleus 0 0 0 0 0 0 0 0 0 0 15 0 0
Passeriformes Thraupidae Cyanerpescyaneus 0 0 0 0 0 0 0 2 3 4 2 1 0 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Passeriformes Thraupidae Hemithraupisflavicollis 0 0 0 0 0 0 0 0 0 0 15 0 0
Passeriformes Thraupidae Conirostrum leucogenys 5 0 0 0 12 0 0 0 0 0 0 00
Passeriformes Thraupidae Sicalisflaveola 5 2 18 22 0 0 0 0 0 0 0 0 0
Passeriformes Thraupidae Volatiniajacarina 0 6 9 0 0 0 0 0 0 0 0 0 0
Passeriformes Thraupidae Sporophila funerea 0 0 0 0 0 0 0 0 0 0 5 0 0
Passeriformes Thraupidae Sporophila crassirostris 0 0 9 0 0 0 0 0 0 0 0 0 0 106 Passeriformes Thraupidae Sporophila americana 0 0 3 0 0 0 0 0 0 0 0 0 0
Passeriformes Thraupidae Sporophila nigricollis 0 3 9 0 0 0 0 0 0 0 0 0 0
Passeriformes Thraupidae Sporophilaschistacea 0 3 12 12 4 0 0 0 0 0 000
Passeriformes Thraupidae Coryphospinguspileatus 3 12 29 29 0 0 0 000000
Passeriformes Thraupidae Coereba flaveola 12 15 23 23 23 17 12 47 50 26 20 15 9
Passeriformes Thraupidae Tiarisbicolor 8 14 23 24 8 4 0 0 0 0 0 0 0
Passeriformes IncertaeSedis Mitrospinguscassinii 0 0 0 0 0 0 0 57 00000
Passeriformes IncertaeSedis Saltatormaximus 0 0 0 0 0 0 2 17 29 20 12 6 1
Passeriformes IncertaeSedis Saltatorcoerulescens 0 0 0 0 1 1 34 3833149 6 3
Passeriformes IncertaeSedis Saltatorgrossus 0 0 0 0 0 0 0 43 69 0 0 0 0 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Passeriformes IncertaeSedis Saltatorstriatipectus 0 1 14 15 8 6 1226400
Passeriformes Emberizidae Ammodramus humeralis 1 1 1 1 2 2 0 0 0 0 0 0 0
Passeriformes Emberizidae Arremonops conirostris 0 0 15 15 10 10 2 1 1 0000
Passeriformes Emberizidae Arremon aurantiirostris 1 3 6 6 14 14 3 1 0 0 0 00
Passeriformes Cardinalidae Habiagutturalis 0 0 0 0 0 0 0 0 15 0 0 0 0
Passeriformes Parulidae Setophagapitiayumi 1 1 1 1 1 1 1 1 1 0 0 0 0 107 Passeriformes Parulidae Myiothlypisfulvicauda 0 1 19 20 10 9 25 92 23 0 0 0 0
Passeriformes Parulidae Basileuterus rufifrons 26 42 74 75 99 85 0 0 0 0000
Passeriformes Icteridae Psarocolius wagleri 0 0 0 0 0 0 0 0 38 2 2 3 0
Passeriformes Icteridae Psarocolius decumanus 0 0 6 6 2 2 1 2 3 12 19 200
Passeriformes Icteridae Psarocolius guatimozinus 0 0 0 0 0 0 0 0 0 0 100 0
Passeriformes Icteridae Icterus auricapillus 0 0 0 0 0 0 0 2 3 4 2 1 0
Passeriformes Icteridae Icterus nigrogularis 0 0 6 0 0 0 0 0 0 0 0 0 0
Passeriformes Fringillidae Euphonia concinna 8 15 32 32 33 25 0 0 00000
Passeriformes Fringillidae Euphonia laniirostris 17 19 22 23 32 3336312718161410
Passeriformes Fringillidae Euphoniafulvicrissa 0 0 0 0 0 0 0 21 0 0 000 Table B-1. continued
Order Family Species Venadillo Bateas ManaDulce Potosi Jabiru Boqueron Rio Manso Rio Claro Maceo La Perla Remedios Barbacoas San Juan
Passeriformes Fringillidae Euphoniaxanthogaster 0 0 0 0 0 1 6 6 5 2110 108 Table B-2. Results of the Maximum likelihood estimation of the parameters for each of the
models tested. We present parametric bootstrap confidence intervals only for the Etienne 2009 model. The results from the Neutrality tests arealsoshown.
Model/Locality θ mI
Full model Venadillo 39.87 0.07 112
Bateas 56.18 0.05 98 ManaDulce 50.8 0.05 156 Potosi 48.72 0.04 122 Jabiru 47.61 0.05 122 Boqueron 48.34 0.05 106
Rio Manso 60.24 0.06 126 Rio Claro 83.33 0.05 124 Maceo 70.83 0.06 142 La Perla 62.02 0.07 136
Remedios 66.92 0.08 151 Barbacoas 70.05 0.06 115 San Juan 55.93 0.05 88 Constant Immigration
One Metacommunity 53.69 0.006 75 Dry Forest 28.59 0.017 241 Wet Forest 37.14 0.015 203 Variable Immigration One Metacommunity
Venadillo 123.94 (86.85,182.07) 0.023 (0.016,0.031) 34 (25,47) Bateas 123.94 (86.85,182.07) 0.022 (0.017,0.031) 43 (32,59)
109 Table B-2. continued
Model/Locality θ mI
ManaDulce 123.94 (86.85,182.07) 0.017 (0.012,0.024) 51 (37,73) Potosi 123.94 (86.85,182.07) 0.015 (0.011,0.021) 45 (34,62) Jabiru 123.94 (86.85,182.07) 0.019 (0.014,0.027) 45 (34,64) Boqueron 123.94 (86.85,182.07) 0.021 (0.016,0.029) 43 (32,61)
Rio Manso 123.94 (86.85,182.07) 0.028 (0.02,0.039) 56 (41,81) Rio Claro 123.94 (86.85,182.07) 0.032 (0.023,0.047) 78 (55,115) Maceo 123.94 (86.85,182.07) 0.032 (0.024,0.047) 74 (54,109) La Perla 123.94 (86.85,182.07) 0.035 (0.026,0.049) 66 (47,94) Remedios 123.94 (86.85,182.07) 0.04 (0.029,0.059) 72 (52,109)
Barbacoas 123.94 (86.85,182.07) 0.033 (0.024,0.048) 61 (44,90) San Juan 123.94 (86.85,182.07) 0.024 (0.018,0.032) 41 (30,56) Two Metacommunities Venadillo 54.05 (32.94,103.49) 0.045 (0.024,0.11) 71 (37,184)
Bateas 54.05 (32.94,103.49) 0.051 (0.024,0.151) 101 (46,333) ManaDulce 54.05 (32.94,103.49) 0.043 (0.019,0.151) 133 (56,525) Potosi 54.05 (32.94,103.49) 0.036 (0.017,0.128) 110 (50,428) Jabiru 54.05 (32.94,103.49) 0.044 (0.021,0.138) 106 (50,371)
Boqueron 54.05 (32.94,103.49) 0.046 (0.023,0.134) 98 (48,314) Rio Manso 95.69 (59.34,170.68) 0.034 (0.022,0.064) 70 (44,134) Rio Claro 95.69 (59.34,170.68) 0.042 (0.024,0.092) 102 (58,239) Maceo 95.69 (59.34,170.68) 0.041 (0.024,0.088) 95 (55,214) La Perla 95.69 (59.34,170.68) 0.044 (0.027,0.082) 83 (50,162)
Remedios 95.69 (59.34,170.68) 0.051 (0.03,0.107) 92 (54,208) Barbacoas 95.69 (59.34,170.68) 0.041 (0.026,0.076) 77 (47,147)
110 Table B-2. continued
Model/Locality θ mI
San Juan 95.69 (59.34,170.68) 0.028 (0.019,0.046) 49 (32,81)
111 Table B-3. Results of the models of the relationship between immigration Rate (m), potential Immigrants (I )andprecipitationanddistancefromthecentroidofthe metacommunity. Terms in bold indicate significance at the 0.05level.
Intercept Precipitation Distance p-value R2 AIC ∆ AIC
One Metacommunity m Full 5.5E-03 8.3E-03 1.43E-05 0.01 0.59 -94.32 1.66 Precipitation 9.1E-03 7.5E-03 <0.01 0.58 -95.98 0 Distance 3.1E-02 -4.01E-05 0.22 0.13 -86.63 9.35 I Full 20.04 14.21 0.02 0.02 0.52 103.47 1.88 Precipitation 24.22 13.32 0.01 0.52 101.59 0 Distance 64.62 -0.08 0.21 0.14 109.17 7.58 Two Metacommunities m Full 4.5E-02 -2.3E-03 2.4E-05 0.60 0.10 -89.20 1.68 Precipitation 4.7E-02 -2.2E-03 0.37 0.07 -90.88 0.00 Distance 4.0E-02 2.0E-05 0.68 0.02 -90.10 0.78 I Full 108.93 -12.30 0.13 0.26 0.24 120.07 1.06 Precipitation 118.05 -11.74 0.15 0.18 119.01 0.00 Distance 82.50 0.11 0.51 0.04 121.05 2.04
112 Table B-4. Results of the models of the relationship between number of rare species and precipitation and total number of species. The coefficients are the result of a Poisson regression in the form: Rare species = eA+β1x1+β2x2 where A is the intercept of the model, x1 and β1 are Precipitation and its estimated coefficient and x2 and β2 are Richness and its coefficient. In this case three levels of rarity where considered: less than 2 individuals/Ha, less than 5 individuals/Ha and less than 10 individuals/Ha. Terms in bold indicate significance at the 0.05 level. Model Intercept Precipitation Richness LogLik AIC ∆AIC 2 ind/Ha Richness 1.51 0.01 -41.6 87.1 12 Precipitation 2.19 0.39 -35.57 75.1 0 Full 2.33 0.41 -0.001 -35.52 77.04 1.94 5 ind/Ha Richness 2.9 0.006 -44.32 92.66 7.7 Precipitation 3.18 0.21 -40.48 84.96 0 Full 3.39 0.25 -0.002 -40.27 86.5 1.54
113 APPENDIX C RCODESFORABUNDANCEESTIMATION Appendix C contains the source codes necessary for estimating abundance using the Beta N-mixture model. It is based on bugs specification of the model, R functions for abundance estimation using N-mixture model and the R code necessary to reproduce the example using real data. The data have been saved in a separate file named UIFcounts.RData. C.1 Bugs Models
Following are the source codes of the Bugs model specificationaswellasthespecification for sampling from the Beta N-mixture model required to do model selection via Information Criterion. C.1.1 Beta N-mixture Model
Jags model specification for the beta N-mixture with data cloning model{
#Priors a˜dunif(0,100) b˜dunif(0,100) for(i in 1:nspp){
lambda[i] ˜ dgamma(0.01,0.01) }
#likelihood
for(i in 1:nspp){ p[ i]˜dbeta(a,b) }
for (k in 1:K){
114 for(i in 1:nspp){
for(j in 1: nsites){ N[i ,j ,k]˜dpois(lambda[i]) } }
}
for(k in 1:K){
for (i in 1:nspp){ for (j in 1: nsites){ for(t in 1: nvisits ){
counts [ j ,t , i ,k]˜
dbin(p[ i ] ,N[ i , j ,k])
}
} } } }
115 C.1.2 Sampling From the Beta N-Mixture model
model{
#Priors for(i in 1:nspp){ p[ i]˜dbeta(3.1482,12.6949) for(j in 1: nsites){ N[i , j]˜dpois(lambda[ i ])T(N.max[i ,j
],) } }
for (i in 1:nspp){ for (j in 1: nsites){ for(t in 1: nvisits ){ counts [ j ,t , i ]˜ dbin
(p[ i ] ,N[ i , j ])
} }
} }
116 C.2 R Functions
Functions for estimating the N-mixture model.
#### Negative log likelihood for the poisson mixture model
#### g u e s s = c ( p , l a m b d a ) #### counts= observed number of counts in n sites and r visits .A matrix of n x r ####K= a value large enough to average over all the possible values that the lambda can take on
pois .mix. nll <− function(guess ,counts ,K){
p <− guess[1]#1/(1+exp(− guess [1]) )
lambda <− guess[2]#exp(guess [2])
if(p<0|p>1|lambda<0){ nll <−9999999} else{
nsites <− nrow( counts ) nvisits <− ncol(counts)
#loglikelihoodofthedetectionprocess
det . proc <− rep(NA, nsites )
for(i in 1: nsites){
ith . site <− counts [ i ,]
max . count <− max( i t h . s i t e ) k.vec <− max . count :K
117 ith .k <− rep(NA, length(k.vec))
for (j in 1:length(k.vec)){
N. j <− k.vec[j]
ith .k[ j ] <− prod(dbinom( ith . site ,N.j ,p))∗ dpois(N. j ,lambda)
}
det . proc [ i ] <− sum( i t h . k )
}
like <− prod(det . proc) nll <−−log( like ) } return ( nll )
}
#####Maximum Likelihood estimation for the poisson mixtur emodel ##### p = d e t e c t i o n p r o b a b i l i t y , v a l u e t o i n i t i a l i z e t h e
optimization ##### lambda = mean number of individuals , value to initiali ze the optimization ##### counts = observed number of counts in n sites and r visit s. Amatrixofnxr
118 #####K= a value large enough to average over all the possible
values that the lambda can take on ##### method = one of the options defined in the optim functio n
pois .mix. est <− function(p,lambda ,counts ,K,method=”Nelder−
Mead”){
guess <− c(p,lambda)
ml . ests <− optim(guess , pois .mix . nll ,method=method , counts=counts ,K=K)
p.hat <− ml . ests$par [1] lambda . hat <− ml . ests$par [2]
negLogLike <− ml . ests$value AIC <− 2∗ length(ml. ests$par)+2∗ negLogLike
results <− c(p=p.hat,lambda=lambda.hat , negLogLike=negLogLike ,AIC=AIC) return ( results ) }
119 C.3 R Code
Code for estimating the abundance 26 species in the dry forests of the Magdalena Valley
in Colombia. The code includes the calculations of AIC valuesformodelselectionbasedon data cloning. It is based on the specification of the model for bugs and the functions listed above. It also uses an additional .RData file with the counts of the species.
library(”rjags”) load(”UIF counts .RData”)
K <− 20
spp . counts <− UIF . counts spp . clons <− array(NA,dim=c(dim(spp. counts) ,K))
nspp <− dim( spp . counts ) [3]
for (j in 1:K){
spp . clons [ , , , j ] <− spp . counts }
max . count <− apply(spp. clons ,c(3 ,1 ,4) ,max)
data list <− list(counts=spp.clons ,nspp=dim(spp .counts)[3],nsites=dim(spp.counts)[1],nvisits=dim(spp . counts ) [2] ,K=K)
inits <− function (){ list(lambda=runif(nspp ,0.1 ,10) ,a=2,b=2,N=max.count)}
120 params <− c(”lambda”,”a”,”b”) jm <− jags .model( file=”Beta Poismix . bugs ”
,data=data list ,n.chains =2,n.adapt=1000,inits= inits) out1 <− coda.samples(model=jm, variable .
names=params ,n. iter =20000,thin=100) hats <− summary(out1)$statistics [ ,1] prop sigma <− 1.96∗ (sqrt(summary(out1)$statistics
[,2]ˆ2∗ 20)) upper <− (hats+prop sigma)∗100/0.78 lower <− (hats−prop sigma)∗100/0.78
out1 .sum <− summary(out1)$statistics nmix . mlest <− matrix(NA,nrow=dim(spp. counts) [3] , ncol=4 ,dimnames=list (
dimnames( spp . counts ) [[3]] , c (”p”,”lambda”,” negloglike”,” AIC”) ) )
121 for(i in 1:nspp){ dat <− spp . counts [ , , i ] nmix . mlest [ i , ] <− pois .mix. est(p=0.5,lambda=1,counts =dat ,K=100)
} names( hats ) [3:28] <− dimnames(spp . counts ) [ [ 3 ] ] mods . mlests <− cbind(p. hat=nmix. mlest [ ,1] , nmixture=nmix. mlest [ ,2] , beta=hats [3:28])
alpha <− hats [1] beta <− hats [2]
ndet <− apply(spp.counts ,3 ,sum)
for . table <− data . frame(ndet=ndet ,p. hat =nmix.mlest[ ,1] ,nmixture=nmix.mlest[,2] ∗ 100/0.78 ,beta=hats [3:28]∗ 100/0.78,
lower= lower [3:28] , upper=
upper [3:28])
#### P e r f o r m i n g m o d e l s e l e c t i o n #### First need to sample from the posterior distribution
122 data list <− list(counts=spp.counts,nspp=dim(spp.counts )[3],nsites=dim(spp.counts)[1] ,nvisits=dim(spp. counts ) [2] ,
lambda=hats [3:28] ,N.max=t ( apply ( spp .counts,c(1,3),
max) ) )
inits <− function (){ list ()}
params <− c(”N”,”p”)
jm <− jags .model( file=”Beta Poismix samp . bugs” ,data=data list ,n.chains =1,n . adapt=10000)
out <− coda . samples (model=jm , variable . names= params ,n. iter =20000, thin=10)
#### O r g a n i z e t h e s a m p l e s
nsamps <− dim( out [ [ 1 ] ] ) [1] npcounts <− dim( spp . counts ) [1] nreps <− dim( spp . counts ) [2] p. lat <− out [[1]][ ,(2470 − 25) :2470] n. lat <− array(NA,dim=c(nsamps,npcounts ,nspp))
123 start <− seq(1,(2470−26) ,by=nspp)
end <− seq(26,(2470−26) ,by=nspp) for(i in 1:npcounts){ ith . start <− start [ i ] ith .end <− end [ i ]
n. lat[,i ,] <− out[[1]][ , ith.start:ith.end] }
#### l i k e l i h o o d f o r t h e n m i x t u r e m o d e l
logpxy .spp <− rep (0,nspp) logpxy . hat1 <− rep (0,nsamps)
for(i in 1:nsamps){
for( ii in 1:nspp){
Yloglikes <− matrix (0, ncol=nreps ,nrow= npcounts )
for(j in 1:nreps){ Yloglikes [ , j ] <− dbinom (UIF . counts [,j,ii],n.lat[i,,ii],mods.mlests[ii ,1],
log=T) }
Y. logL <− sum( Y l o g l i k e s )
124 X. logL <− sum(dpois(n.lat[i ,, ii ],mods.mlests
[ii,2],log=T)) logpxy . spp[ ii ] <− Y. logL+X.logL
}
logpxy . hat1[ i ] <− sum( logpxy . spp )
}
##### l i k e l i h o o d f o r t h e b e t a m o d e l logpxy .spp <− rep (0,nspp) logpxy . hat2 <− rep (0,nsamps)
for(i in 1:nsamps){
for( ii in 1:nspp){
Yloglikes <− matrix (0, ncol=nreps ,nrow= npcounts )
for(j in 1:nreps){
Yloglikes [ , j ] <− dbinom (UIF . counts [,j,ii],n.lat[i,,ii],p.lat[i,ii],log=T) }
Y. logL <− sum( Y l o g l i k e s )
125 X. logL <− sum(dpois(n.lat[i ,, ii ],mods.mlests
[ii,3],log=T))+dbeta(p.lat[i,ii],alpha,beta,log =T ) logpxy . spp[ ii ] <− Y. logL+X.logL
}
logpxy . hat2[ i ] <− sum( logpxy . spp )
} llratio <− (logpxy.hat1−logpxy . hat2) delta .AIC <−−2∗(mean( l l r a t i o ))+2∗(52−28)
126 REFERENCES Ackerly, D.D. & Cornwell, W. (2007) A trait-based approach to community assembly: partitioning of species trait values into within-and among-community components. Ecology letters, 10,135–145. Bacles, C.F., Lowe, A.J. & Ennos, R.A. (2006) Effective seed dispersal across a fragmented landscape. Science, 311,628–628. Birdlife International & NatureServe (2014) Bird species distribution maps of the world.Birdlife International, Cambridge, UK and NatureServe, Arlington, USA. Blake, J.G. (1992) Temporal variation in point counts of birdsinalowlandwetforestincosta rica. Condor,pp.265–275.
Blake, J.G. (2007) Neotropical forest bird communities: a comparison of species richness and composition at local and regional scales. The Condor, 109,237–255. Blake, J.G. & Loiselle, B.A. (2001) Bird assemblages in second-growth and old-growth forests, costa rica: perspectives from mist nets and point counts. The Auk, 118,304–326. Blake, J.G. & Loiselle, B.A. (2009) Species composition of neotropical understory bird communities: Local versus regional perspectives based on capture data. Biotropica, 41, 85–94. Bryant, J.A., Lamanna, C., Morlon, H., Kerkhoff, A.J., Enquist, B.J. & Green, J.L. (2008) Microbes on mountainsides: contrasting elevational patterns of bacterial and plant diversity. Proceedings of the National Academy of Sciences, 105,11505–11511. Buckland, S.T., Anderson, D.R., Burnham, K.P. & Laake, J.L. (2005) Distance sampling. Wiley Online Library.
Cavender-Bares, J., Kozak, K.H., Fine, P.V. & Kembel, S.W. (2009) The merging of community ecology and phylogenetic biology. Ecology letters, 12,693–715. Chao, A., Chazdon, R.L., Colwell, R.K. & Shen, T.J. (2005) A new statistical approach for assessing similarity of species composition with incidenceandabundancedata. Ecology letters, 8,148–159. Chase, J.M. & Myers, J.A. (2011) Disentangling the importance of ecological niches from stochastic processes across scales. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 366,2351–2363. Chisholm, R.A. & Pacala, S.W. (2010) Niche and neutral models predict asymptotically equivalent species abundance distributions in high-diversity ecological communities. Proceed- ings of the National Academy of Sciences, 107,15821–15825. Clark, J.S. (2009) Beyond neutral science. Trends in Ecology & Evolution, 24,8–15.
127 Clark, J.S. (2012) The coherence problem with the unified neutral theory of biodiversity. Trends in ecology & evolution, 27,198–202. Clements, F.E. (1936) Nature and structure of the climax. Journal of ecology, 24,252–284. Clements, F.E. (1916) Plant succession: an analysis of the development of vegetation.242. Carnegie Institution of Washington. Cody, M.L. (1974) Competition and the structure of bird communities,volume7.Princeton University Press.
Cody, M.L. & Diamond, J.M. (1975) Ecology and evolution of communities.HarvardUniversity Press. Commission, I.S.S. et al. (2010) IUCN Red List of threatened species.IUCN,theWorld Conservation Union. Condit, R., Pitman, N., Leigh, E.G., Chave, J., Terborgh, J., Foster, R.B., N´unez, P., Aguilar, S., Valencia, R., Villa, G. et al. (2002) Beta-diversity in tropical forest trees. Science, 295, 666–669. Connor, E.F. & Simberloff,D.(1978)Speciesnumberandcompositional similarity of the galapagos flora and avifauna. Ecological Monographs,pp.219–248.
Cornwell, W.K. & Ackerly, D.D. (2009) Community assembly and shifts in plant trait distributions across an environmental gradient in coastal california. Ecological Monographs, 79,109–126.
Cornwell, W.K. & Ackerly, D.D. (2010) A link between plant traits and abundance: evidence from coastal california woody plants. Journal of Ecology, 98,814–821. Cornwell, W.K., Schwilk, D.W. & Ackerly, D.D. (2006) A trait-based test for habitat filtering: convex hull volume. Ecology, 87,1465–1471. de Valpine, P. (2012) Frequentist analysis of hierarchical models for population dynamics and demographic data. Journal of Ornithology, 152,393–408. Denes, F.V., Silveira, L.F. & Beissinger, S.R. (2015) Estimatingabundanceofunmarkedanimal populations: accounting for imperfect detection and other sources of zero inflation. Methods in Ecology and Evolution, 6,543–556. Dorazio, R.M., Connor, E.F. & Askins, R.A. (2015) Estimating the effects of habitat and biological interactions in an avian community. PloS one, 10,e0135987. Dorazio, R.M., Martin, J. & Edwards, H.H. (2013) Estimating abundance while accounting for rarity, correlated behavior, and other sources of variationincounts. Ecology, 94,1472–1478.
Dorazio, R.M. & Royle, J.A. (2005) Estimating size and composition of biological communities by modeling the occurrence of species. Journal of the American Statistical Association, 100, 389–398.
128 Dunning Jr, J.B. (1992) CRC handbook of avian body masses.CRCpress. Emerson, B.C. & Gillespie, R.G. (2008) Phylogenetic analysis of community assembly and structure over space and time. Trends in ecology & evolution, 23,619–630. Engelbrecht, B.M., Comita, L.S., Condit, R., Kursar, T.A., Tyree, M.T., Turner, B.L. & Hubbell, S.P. (2007) Drought sensitivity shapes species distribution patterns in tropical forests. Nature, 447,80–82. Etienne, R.S. (2005) A new sampling formula for neutral biodiversity. Ecology letters, 8, 253–260. Etienne, R.S. (2007) A neutral sampling formula for multiple samples and an exact test of neutrality. Ecology letters, 10,608–618.
Etienne, R.S. (2009) Improved estimation of neutral model parameters for multiple samples with different degrees of dispersal limitation. Ecology, 90,847–852. Etienne, R.S. & Olff,H.(2004)Howdispersallimitationshapesspecies–bodysizedistributions in local communities. The American Naturalist, 163,69–83. Fisher, R.A., Corbet, A.S. & Williams, C.B. (1943) The relation between the number of species and the number of individuals in a random sample of an animal population. The Journal of Animal Ecology,pp.42–58. Gleason, H.A. (1926) The individualistic concept of the plantassociation. Bulletin of the Torrey Botanical Club,pp.7–26.
Gleason, H.A. (1939) The individualistic concept of the plantassociation. American Midland Naturalist,pp.92–110. Gomez, J.P., Bravo, G.A., Brumfield, R.T., Tello, J.G. & Cadena, C.D. (2010) A phylogenetic approach to disentangling the role of competition and habitat filtering in community assembly of neotropical forest birds. Journal of animal ecology, 79,1181–1192. Gonz´alez-Caro, S., Parra, J.L., Graham, C.H., McGuire, J.A. & Cadena, C.D. (2012) Sensitivity of metrics of phylogenetic structure to scale, source of data and species pool of hummingbird assemblages along elevational gradients. PloS one, 7,e35472. Gotelli, N.J. (2000) Null model analysis of species co-occurrence patterns. Ecology, 81, 2606–2621. Gower, J.C. (1971) A general coefficient of similarity and some of its properties. Biometrics, pp. 857–871. Graham, C.H. & Fine, P.V. (2008) Phylogenetic beta diversity: linking ecological and evolutionary processes across space in time. Ecology letters, 11,1265–1277.
129 Graham, C.H., Parra, J.L., Rahbek, C. & McGuire, J.A. (2009) Phylogenetic structure in tropical hummingbird communities. Proceedings of the National Academy of Sciences, 106, 19673–19678.
Graham, C.H., Parra, J.L., Tinoco, B.A., Stiles, F.G. & McGuire, J.A. (2012) Untangling the influence of ecological and evolutionary factors on traitvariationacrosshummingbird assemblages. Ecology, 93,S99–S111.
Grant, P. & Abbott, I. (1980) Interspecific competition, island biogeography and null hypotheses. Evolution,pp.332–341. Graves, G.R. & Gotelli, N.J. (1993) Assembly of avian mixed-species flocks in amazonia. Proceedings of the National Academy of Sciences, 90,1388–1391. Greenberg, R. & Gradwohl, J. (1986) Constant density and stable territoriality in some tropical insectivorous birds. Oecologia, 69,618–625. Grinnell, J. (1917) The niche-relationships of the california thrasher. The Auk, 34,427–433.
Group, P. et al. (2014) Pari/gp version 2.7.2. Hackett, S.J., Kimball, R.T., Reddy, S., Bowie, R.C., Braun, E.L., Braun, M.J., Chojnowski, J.L., Cox, W.A., Han, K.L., Harshman, J. et al. (2008) A phylogenomic study of birds reveals their evolutionary history. science, 320,1763–1768. Haffer, J. (1967) Zoogeographical notes on the non-forest lowland bird faunas of northwestern south america. Hornero, 10,315–333.
Hardy, O.J. (2008) Testing the spatial phylogenetic structure of local communities: statistical performances of different null models and test statistics on a locally neutral community. Journal of ecology, 96,914–926.
Harpole, S.W. & Tilman, D. (2006) Non-neutral patterns of species abundance in grassland communities. Ecology Letters, 9,15–23. Harrison, S. (1997) How natural habitat patchiness affects the distribution of diversity in californian serpentine chaparral. Ecology, 78,1898–1906. Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G. & Jarvis, A. (2005) Very high resolution interpolated climate surfaces for global land areas. International journal of climatology, 25, 1965–1978. HilleRisLambers, J., Adler, P., Harpole, W., Levine, J. & Mayfield, M. (2012) Rethinking community assembly through the lens of coexistence theory. Annual Review of Ecology, Evolution, and Systematics, 43,227. Hilty, S.L. & Brown, B. (1986) AguidetothebirdsofColombia.PrincetonUniversityPress. Hubbell, S.P. (2001) The unified neutral theory of biodiversity and biogeography (MPB-32), volume 32. Princeton University Press.
130 Hubbell, S.P. (2005) Neutral theory in community ecology and the hypothesis of functional equivalence. Functional ecology, 19,166–172. Hutto, R.L., Pletschet, S.M. & Hendricks, P. (1986) A fixed-radius point count method for nonbreeding and breeding season use. The Auk,pp.593–602. Jabot, F. & Chave, J. (2011) Analyzing tropical forest tree species abundance distributions using a nonneutral model and through approximate bayesian inference. The American Naturalist, 178,E37–E47. Jabot, F., Etienne, R.S. & Chave, J. (2008) Reconciling neutralcommunitymodelsand environmental filtering: theory and an empirical test. Oikos, 117,1308–1320. Jankowski, J.E., Ciecka, A.L., Meyer, N.Y. & Rabenold, K.N. (2009) Beta diversity along environmental gradients: implications of habitat specialization in tropical montane landscapes. Journal of Animal Ecology, 78,315–327. Jankowski, J.E., Graham, C.H., Parra, J.L., Robinson, S.K., Seddon, N., Touchton, J.M. & Tobias, J.A. (2012) The role of competition in structuring tropical bird communities. Ornitol Neotrop, 23,115–124. Jankowski, J.E., Merkord, C.L., Rios, W.F., Cabrera, K.G., Revilla, N.S. & Silman, M.R. (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. Janzen, D.H. (1970) Herbivores and the number of tree species in tropical forests. American naturalist,pp.501–528. Jetz, W., Thomas, G., Joy, J., Hartmann, K. & Mooers, A. (2012) Theglobaldiversityofbirds in space and time. Nature, 491,444–448.
Joseph, L.N., Elkin, C., Martin, T.G. & Possingham, H.P. (2009) Modeling abundance using n-mixture models: the importance of considering ecologicalmechanisms. Ecological Applications, 19,631–642.
Karr, J.R., Robinson, S.K., Blake, J.G., Bierregaard Jr, R.O. & Gentry, A. (1990) Birds of four neotropical forests. Four neotropical rainforests,pp.237–269. Kearney, M. & Porter, W. (2009) Mechanistic niche modelling:combiningphysiologicaland spatial data to predict species’ ranges. Ecology letters, 12,334–350. Keddy, P.A. & Weiher, E. (1999) Ecological assembly rules: perspectives, advances, retreats. Cambridge University Press.
Kembel, S., Cowan, P., Helmus, M., Cornwell, W., Morlon, H., Ackerly, D., Blomberg, S. & Webb, C. (2010) Picante: R tools for integrating phylogeniesandecology. Bioinformatics, 26,1463–1464.
131 Kraft, N.J., Comita, L.S., Chase, J.M., Sanders, N.J., Swenson, N.G., Crist, T.O., Stegen, J.C., Vellend, M., Boyle, B., Anderson, M.J. et al. (2011) Disentangling the drivers of β diversity along latitudinal and elevational gradients. Science, 333,1755–1758.
Kraft, N.J., Godoy, O. & Levine, J.M. (2015) Plant functional traits and the multidimensional nature of species coexistence. Proceedings of the National Academy of Sciences, 112, 797–802.
Kraft, N.J., Valencia, R. & Ackerly, D.D. (2008) Functional traits and niche-based tree community assembly in an amazonian forest. Science, 322,580–582. Lalibert´e, E. & Legendre, P. (2010) A distance-based framework for measuring functional diversity from multiple traits. Ecology, 91,299–305. Lalibert´e, E., Legendre, P. & Shipley, B. (2014) FD: measuring functional diversity from multiple traits, and other tools for functional ecology. R package version 1.0-12. Lebrija-Trejos, E., P´erez-Garc´ıa, E.A., Meave, J.A., Bongers,F.&Poorter,L.(2010) Functional traits and environmental filtering drive community assembly in a species-rich tropical system. Ecology, 91,386–398. Legendre, P., Borcard, D. & Peres-Neto, P.R. (2005) Analyzing beta diversity: partitioning the spatial variation of community composition data. Ecological Monographs, 75,435–450. Leibold, M.A., Holyoak, M., Mouquet, N., Amarasekare, P., Chase, J., Hoopes, M., Holt, R., Shurin, J., Law, R., Tilman, D. et al. (2004) The metacommunity concept: a framework for multi-scale community ecology. Ecology letters, 7,601–613. Lele, S.R., Dennis, B. & Lutscher, F. (2007) Data cloning: easy maximum likelihood estimation for complex ecological models using bayesian markov chain monte carlo methods. Ecology letters, 10,551–563. Lele, S.R., Nadeem, K. & Schmuland, B. (2010) Estimability and likelihood inference for generalized linear mixed models using data cloning. Journal of the American Statistical Association, 105,1617–1625. Lovette, I.J. & Hochachka, W.M. (2006) Simultaneous effects of phylogenetic niche conservatism and competition on avian community structure. Ecology, 87,S14–S28.
MacArthur, R. & Levins, R. (1967) The limiting similarity, convergence, and divergence of coexisting species. American naturalist,pp.377–385. MacArthur, R.H. (1958) Population ecology of some warblers ofnortheasternconiferous forests. Ecology, 39,599–619. MacKenzie, D.I., Nichols, J.D., Lachman, G.B., Droege, S., Andrew Royle, J. & Langtimm, C.A. (2002) Estimating site occupancy rates when detection probabilities are less than one. Ecology, 83,2248–2255.
132 Magurran, A.E. & Henderson, P.A. (2003) Explaining the excess of rare species in natural species abundance distributions. Nature, 422,714–716. Martin, J., Royle, J.A., Mackenzie, D.I., Edwards, H.H., Kery, M. & Gardner, B. (2011) Accounting for non-independent detection when estimating abundance of organisms with a bayesian approach. Methods in Ecology and Evolution, 2,595–601. Martin, T.G., Wintle, B.A., Rhodes, J.R., Kuhnert, P.M., Field, S.A., Low-Choy, S.J., Tyre, A.J. & Possingham, H.P. (2005) Zero tolerance ecology: improving ecological inference by modelling the source of zero observations. Ecology letters, 8,1235–1246. Martin, T.E. (1988) Processes organizing open-nesting bird assemblages: competition or nest predation? Evolutionary Ecology, 2,37–50. Martin, T.E. (1996) Life history evolution in tropical and south temperate birds: what do we really know? Journal of Avian Biology, 27,263–272. Mart´ınez, A.E. & Gomez, J.P. (2013) Are mixed-species bird flocks stable through two decades? The American Naturalist, 181,E53–E59. Matsuoka, S.M., Mahon, C.L., Handel, C.M., S´olymos, P., Bayne, E.M., Fontaine, P.C. & Ralph, C.J. (2014) Reviving common standards in point-count surveys for broad inference across studies. The Condor, 116,599–608. McGill, B.J., Enquist, B.J., Weiher, E. & Westoby, M. (2006) Rebuilding community ecology from functional traits. Trends in ecology & evolution, 21,178–185.
McGill, B.J., Etienne, R.S., Gray, J.S., Alonso, D., Anderson, M.J., Benecha, H.K., Dornelas, M., Enquist, B.J., Green, J.L., He, F. et al. (2007) Species abundance distributions: moving beyond single prediction theories to integration within an ecological framework. Ecology letters, 10,995–1015. McKechnie, A.E. & Wolf, B.O. (2009) Climate change increases the likelihood of catastrophic avian mortality events during extreme heat waves. Biology Letters,p.rsbl20090702.
Meynard, C.N. & Quinn, J. (2008) Bird metacommunities in temperate south american forest: vegetation structure, area, and climate effects. Ecology, 89,981–990. Miles, D.B. & Ricklefs, R.E. (1984) The correlation between ecology and morphology in deciduous forest passerine birds. Ecology,pp.1629–1640. Moore, R., Robinson, W., Lovette, I. & Robinson, T. (2008) Experimental evidence for extreme dispersal limitation in tropical forest birds. Ecology letters, 11,960–968.
Munn, C.A. & Terborgh, J.W. (1979) Multi-species territoriality in neotropical foraging flocks. Condor,pp.338–347. Munoz, F., Couteron, P., Ramesh, B. & Etienne, R.S. (2007) Estimating parameters of neutral communities: from one single large to several small samples. Ecology, 88,2482–2488.
133 Murphy, P.G. & Lugo, A.E. (1986) Ecology of tropical dry forest. Annual review of ecology and systematics,pp.67–88. Nicolson, M., McIntosh, R.P. & Nicholson, M. (2002) H. a. gleason and the individualistic hypothesis revisited. Bulletin of the Ecological Society of America, 83,133–142. Oksanen, J., Blanchet, F.G., Kindt, R., Legendre, P., Minchin, P.R., O’Hara, R.B., Simpson, G.L., Solymos, P., Stevens, M.H.H. & Wagner, H. (2016) vegan: Community Ecology Package. R package version 2.3-3. Olson, D.M., Dinerstein, E., Wikramanayake, E.D., Burgess, N.D., Powell, G.V., Underwood, E.C., D’amico, J.A., Itoua, I., Strand, H.E., Morrison, J.C. et al. (2001) Terrestrial ecoregions of the world: A new map of life on earth a new global map of terrestrial ecoregions provides an innovative tool for conserving biodiversity. BioScience, 51,933–938. Oniki, Y. (1979) Is nesting success of birds low in the tropics? Biotropica,pp.60–69. Oniki, Y. (1985) Why robin eggs are blue and birds build nests:statisticaltestsforamazonian birds. Ornithological Monographs,pp.536–545. Oswald, J.A., Burleigh, J.G., Steadman, D.W., Robinson, S.K. & Kratter, A.W. (2016) Historical climatic variability and geographical barriersasdriversofcommunitycomposition in a biodiversity hotspot. Journal of Biogeography, 43,123–133. Ozkan,¨ K., Svenning, J.C. & Jeppesen, E. (2013) Environmental species sorting dominates forest-bird community assembly across scales. Journal of animal ecology, 82,266–274.
Pardini, R., de Souza, S.M., Braga-Neto, R. & Metzger, J.P. (2005)Theroleofforest structure, fragment size and corridors in maintaining smallmammalabundanceanddiversity in an atlantic forest landscape. Biological conservation, 124,253–266.
Parker III, T., Stotz, D. & Fitzpatrick, J. (1996) Ecological and distributional databases for Neotropical birds.Wiley. Parra, J.L., McGuire, J.A. & Graham, C.H. (2010) Incorporating clade identity in analyses of phylogenetic community structure: an example with hummingbirds. The American Naturalist, 176,573–587. Parra, J.L., Rahbek, C., McGuire, J.A. & Graham, C.H. (2011) Contrasting patterns of phylogenetic assemblage structure along the elevational gradient for major hummingbird clades. Journal of Biogeography, 38,2350–2361. Petchey, O.L. & Gaston, K.J. (2002) Functional diversity (fd),speciesrichnessandcommunity composition. Ecology Letters, 5,402–411. Petchey, O.L. & Gaston, K.J. (2006) Functional diversity: back to basics and looking forward. Ecology letters, 9,741–758.
134 Pigot, A.L., Trisos, C.H. & Tobias, J.A. (2016) Functional traitsrevealtheexpansionand packing of ecological niche space underlying an elevationaldiversitygradientinpasserine birds. Proceedings of the Royal Society of London: B, 283,20152013.
Plummer, M. (2014) rjags: Bayesian graphical models using MCMC.Rpackageversion3-13. Podani, J. & Schmera, D. (2006) On dendrogram-based measuresoffunctionaldiversity. Oikos, 115,179–185. Ponciano, J.M., Burleigh, J.G., Braun, E.L. & Taper, M.L. (2012) Assessing parameter identifiability in phylogenetic models using data cloning. Systematic biology,p.sys055. Ponciano, J.M., Taper, M.L., Dennis, B. & Lele, S.R. (2009) Hierarchical models in ecology: confidence intervals, hypothesis testing, and model selection using data cloning. Ecology, 90, 356–362. Preston, F.W. (1948) The commonness, and rarity, of species. Ecology, 29,254–283. Qian, H. & Ricklefs, R.E. (2007) A latitudinal gradient in large-scale beta diversity for vascular plants in north america. Ecology letters, 10,737–744. RCoreTeam(2013)R: A Language and Environment for Statistical Computing.RFoundation for Statistical Computing, Vienna, Austria.
Ralph, C.J., Geupel, G.R., Pyle, P., Martin, T.E. & DeSante, D.F. (1993) Handbook of field methods for monitoring landbirds. USDA Forest Service/UNL Faculty Publications,p.105. Ralph, C.J., Sauer, J.R. et al. (1995) Monitoring bird populations by point counts. USDA Forest Service General Technical Report PSW-GTR-149,pp.161–175. Ricklefs, R.E. (1969) An analysis of nesting mortality in birds.SmithsonianInstitutionPress. Ricklefs, R.E. (2012) Species richness and morphological diversity of passerine birds. Proceed- ings of the National Academy of Sciences, 109,14482–14487.
Robinson, S.K. & Terborgh, J. (1995) Interspecific aggressionandhabitatselectionby amazonian birds. Journal of Animal Ecology,pp.1–11. Robinson, W.D., Brawn, J.D. & Robinson, S.K. (2000) Forest bird community structure in central panama: influence of spatial scale and biogeography. Ecological Monographs, 70, 209–235. Rodr´ıguez, P. & T Arita, H. (2004) Beta diversity and latitude in north american mammals: testing the hypothesis of covariation. Ecography, 27,547–556. Rosindell, J., Hubbell, S.P., He, F., Harmon, L.J. & Etienne, R.S. (2012) The case for ecological neutral theory. Trends in ecology & evolution, 27,203–208. Royle, J.A. (2004) N-mixture models for estimating population size from spatially replicated counts. Biometrics, 60,108–115.
135 Royle, J.A. & Dorazio, R.M. (2008) Hierarchical modeling and inference in ecology: the analysis of data from populations, metapopulations and communities.AcademicPress. Russo, S.E., Robinson, S.K. & Terborgh, J. (2003) Size-abundance relationships in an amazonian bird community: implications for the energetic equivalence rule. The Ameri- can Naturalist, 161,267–283. Simberloff, D. (2004) Community ecology: is it time to move on? The American Naturalist, 163,787–799. Smith, B.T., McCormack, J.E., Cuervo, A.M., Hickerson, M.J., Aleixo, A., Cadena, C.D., P´erez-Em´an, J., Burney, C.W., Xie, X., Harvey, M.G. et al. (2014) The drivers of tropical speciation. Nature, 515,406–409. Smith, T.B., Wayne, R.K., Girman, D.J. & Bruford, M.W. (1997) A role for ecotones in generating rainforest biodiversity. Science, 276,1855–1857. Sokol, E.R., Benfield, E., Belden, L.K. & Valett, H.M. (2011) The assembly of ecological communities inferred from taxonomic and functional composition. The American Naturalist, 177,630–644. Sollmann, R., Gardner, B., Williams, K.A., Gilbert, A.T. & Veit, R.R. (2015) A hierarchical distance sampling model to estimate abundance and covariateassociationsofspeciesand communities. Methods in Ecology and Evolution. Stegen, J.C. & Hurlbert, A.H. (2011) Inferring ecological processes from taxonomic, phylogenetic and functional trait β-diversity. PloS one, 6,e20906. Stouffer, P.C., Johnson, E.I. & Bierregaard Jr, R.O. (2013) Breeding seasonality in central amazonian rainforest birds. The Auk, 130,529–540.
Strong, D., Whipple, A.V., Child, A. & Dennis, B. (1999) Model selection for a subterranean trophic cascade: root-feeding caterpillars and entomopathogenic nematodes. Ecology, 80, 2750–2761.
Swenson, N.G., Anglada-Cordero, P. & Barone, J.A. (2011) Deterministic tropical tree community turnover: evidence from patterns of functional beta diversity along an elevational gradient. Proceedings of the Royal Society of London B: Biological Sciences, 278,877–884.
Swenson, N.G., Enquist, B.J., Pither, J., Thompson, J. & Zimmerman, J.K. (2006) The problem and promise of scale dependency in community phylogenetics. Ecology, 87, 2418–2424.
Terborgh, J. (1977) Bird species diversity on an andean elevational gradient. Ecology,pp. 1007–1019. Terborgh, J., Robinson, S.K., Parker III, T.A., Munn, C.A. & Pierpont, N. (1990) Structure and organization of an amazonian forest bird community. Ecological Monographs, 60, 213–238.
136 Thiollay, J.M. (1994) Structure, density and rarity in an amazonian rainforest bird community. Journal of Tropical Ecology, 10,449–481. Touchton, J.M. & Smith, J.N. (2011) Species loss, delayed numerical responses, and functional compensation in an antbird guild. Ecology, 92,1126–1136. Tuomisto, H. & Ruokolainen, K. (2006) Analyzing or explaining beta diversity? understanding the targets of different methods of analysis. Ecology, 87,2697–2708. Tuomisto, H., Ruokolainen, K. & Yli-Halla, M. (2003) Dispersal, environment, and floristic variation of western amazonian forests. Science, 299,241–244. Vamosi, S., Heard, S., Vamosi, J. & Webb, C. (2009) Emerging patterns in the comparative analysis of phylogenetic community structure. Molecular ecology, 18,572–592.
Van Houtan, K.S., Pimm, S.L., Halley, J.M., Bierregaard, R.O. & Lovejoy, T.E. (2007) Dispersal of amazonian birds in continuous and fragmented forest. Ecology letters, 10, 219–229.
Vellend, M. (2010) Conceptual synthesis in community ecology. The Quarterly review of biology, 85,183–206. Vill´eger, S., Mason, N.W. & Mouillot, D. (2008) New multidimensional functional diversity indices for a multifaceted framework in functional ecology. Ecology, 89,2290–2301. Webb, C.O., Ackerly, D.D., McPeek, M.A. & Donoghue, M.J. (2002) Phylogenies and community ecology. Annual review of ecology and systematics,pp.475–505.
Webb, D. (1987) Thermal tolerance of avian embryos: a review. Condor, 89,874–898. Weiher, E., Freund, D., Bunton, T., Stefanski, A., Lee, T. & Bentivenga, S. (2011) Advances, challenges and a developing synthesis of ecological community assembly theory. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 366,2403–2413. Weiher, E. & Keddy, P.A. (1995) The assembly of experimental wetland plant communities. Oikos,pp.323–335. Weinstein, B.G., Tinoco, B., Parra, J.L., Brown, L.M., McGuire, J.A., Stiles, F.G. & Graham, C.H. (2014) Taxonomic, phylogenetic, and trait beta diversity in south american hummingbirds. The American Naturalist, 184,211–224. White, E.P. & Hurlbert, A.H. (2010) The combined influence of thelocalenvironmentand regional enrichment on bird species richness. The American Naturalist, 175,E35–E43. Whittaker, R.H. (1960) Vegetation of the siskiyou mountains,oregonandcalifornia.Ecological monographs, 30,279–338.
Willis, E. (1973) The behavior of ocellated antbirds. Smithsonian Contributions to Zoology, 144,1–57.
137 Wilman, H., Belmaker, J., Simpson, J., de la Rosa, C., Rivadeneira, M.M. & Jetz, W. (2014) Eltontraits 1.0: Species-level foraging attributes of the world’s birds and mammals. Ecology, 95,2027–2027.
Yamaura, Y., Andrew Royle, J., Kuboi, K., Tada, T., Ikeno, S. & Makino, S. (2011) Modelling community dynamics based on species-level abundance modelsfromdetection/nondetection data. Journal of applied ecology, 48,67–75.
Zeileis, A., Kleiber, C. & Jackman, S. (2008) Regression models for count data in R. Journal of Statistical Software, 27.
138 BIOGRAPHICAL SKETCH Juan Pablo Gomez is a Colombian evolutionary ecologist that has focused on investigating the main mechanisms driving community assembly of neotropical birds. He holds a bachelor’s and master’s degree from Universidad de los Andes, in Bogota, Colombia, where he grew up.
He worked for CENICAFE (Centro Nacional de Investigaciones de Cafe) in Chinchina Colombia, doing research about the impact of environmentally friendlycertificationtocoffee plantations over tree, butterfly and plant species. During this job, he gotinterestedinbirdcommunity ecology, topic that has been his main research focus since then. After finishing his master’s degree, he worked as a consultant for a Nescent funded group. He also spent several months assisting researchers in the field, mainly working with ecology of bird mixed species flocks. In 2010 he moved into the United States to pursue his PhD in zoology under the advisory of Dr. Scott K. Robinson and Dr. Jose Miguel Ponciano. He received his PhD from the University of Florida in spring 2016.
139