PREDATION AND HABITAT DRIVE REEF FISH COMMUNITY ORGANIZATION

Transcription

1 PREDATION AND HABITAT DRIVE REEF FISH COMMUNITY ORGANIZATION By ADRIAN CRAIG STIER A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA

2 2012 Adrian Craig Stier 2

3 To my family for their extraordinary patience, support, kindness and love 3

4 ACKNOWLEDGMENTS I thank my parents and grandparents for their support, love, and patience and my sister for her encouragement and willingness to listen. My advisor, Craig Osenberg, who has spent countless hours training me to have both an intricate appreciation for the scientific method and a passion for the dissemination of knowledge. His hands off, yet always available approach to advising allowed me the luxury of exploring my own ideas and seeking the limits of my imagination. Committee members Ben Bolker, Colette St. Mary, Todd Palmer, Bob Holt, and Rob Fletcher each provided unique and effective guidance throughout the pursuit of my degree. Ben, exhibited an exceptional willingness and patience in the lab on a daily basis training me in the interface between models and data, statistics, and R; his addiction to helping people learn is both infectious and admirable. Colette encouraged me to think broadly and offered an evolutionary perspective to my work; she also invested in me as a person and has helped me achieve a work-life balance by reviving my appreciation for the humanities. As an earlycareer scientist, Todd has served as a role model throughout my time here and continues to offer perspective on clairvoyance in science. Bob, who continues to improve my work with the breadth of his knowledge, has been essential in promoting me as a professional. Lastly, Rob has been the ideal external committee member, providing extensive feedback on my manuscripts and serving as a kind and dedicated mentor. I also am grateful to my friends and other mentors, Jameal Samhouri, Mark Steele, Pete Edmunds, Chris Stallings, Will White, Mike McCoy, Josh Idjadi, Kate Hanson, Kerry Nichols, Jimmy O Donnell, Ernest Williams, and Pat Reynolds, who out of the goodness of their heart have taken the time to invest in my personal and career 4

5 development. Their patience with my endless reservoir of science questions is impressive. I m particularly grateful to Jameal, who has been my role model, mentor, and friend for nearly a decade. To my closest friends and colleagues during my time in graduate school (Matt Smith, Ashley Seifert, Bret Pasch, James Monaghan, and François Michonneau), our bond and research through the Nexus Biology Group has been inspirational and fulfilling. I thank my friends and family in Gainesville (M. Smith, A. & M. Seifert, J. & S. Monaghan, B. & K. Pasch, F. Michonneau, S. & V. van Montrfans, M. & K. McCoy, and J. Resasco) from the bottom of my heart. The happiness and balance you have brought me has been extraordinarily rewarding. I also thank the friends whom I have lost (Jake Shapiro, Mike Onorato and Marvin Morales) for instilling in me a will to live, achieve, and appreciate every day as a gift. Last, but certainly not least, Caroline Pace, who has worked with me to see beyond stress, anxiety, and fear to love my dear friends and family around me. I am particularly thankful to my coauthors C. Osenberg, B. Bolker, S. Geange, K. Hanson, R. Schmitt, S. Holbrook, M. Kulbicki, and A. Hein for their insight, hard work, and support. My research would not have been possible without field and lab assistance from Z. Boudreau, B. Focht, J. Heinlein, M. Johnson, M. Sogin, B. Sosik, N. Dallin, M. Murray M. Gil, L. Bentley, M. Leray, and M. Beraud, S. Boyer, E. Vuxton, A. Payne, and S. Deeb. My work also benefited from discussion with numerous colleagues including the Nexus Biology Group, J. Shima, the Osenberg Lab, and the St. Mary-Osenberg- Bolker lab. J. Samhouri, R. Fletcher, C. Osenberg, T. Palmer, B. Bolker, M. McCoy, S. Hamilton, W. White, J. Samhouri, C. Stallings, K. Marhaver. Numerous anonymous reviewers also provided helpful comments. Logistical Support was provided by the UF 5

6 shop and administrative assistance and the staff at the Gump Biological Research Station. Funding was provided by NSF (OCE and OCE ), a 3 Seas teaching fellowship, a National Geographic International Ecostations Fellowship, the Ocean Bridges Program funded by the French-American Cultural Exchange, a McLaughlin Fellowship from the University of Florida, and private donations from Mark and Karen Stier and Paul and Marion Morton. To those of you that I have mentioned and perhaps a few that I have forgotten, know that I believe we are all a product of those we are surrounded by and I therefore owe who I am to you. 6

7 TABLE OF CONTENTS page ACKNOWLEDGMENTS... 4 LIST OF TABLES LIST OF FIGURES ABSTRACT INTRODUCTION Theoretical Context Outline Study System PROPAGULE REDIRECTION: HABITAT AVAILABILITY REDUCES COLONIZATION AND INCREASES RECRUITMENT IN REEF FISHES Background Materials and Methods Study Site and Species Data Analysis Projected Recruitment Success Results Settlement Projected Recruitment Success Discussion Propagule Redirection: Connectivity and Temporal and Spatial Scales Propagule Redirection: Unbridled Conjecture PREDATOR DENSITY AND COMPETITION MODIFY THE BENEFITS OF GROUP FORMATION IN A SHOALING REEF FISH Background Methods Study System and Species Fish Handling and Tagging Experiment 1: Effects of Predator and Conspecific Density on Prey Survival.. 45 Experiment 2: Effect of a Heterospecific Damselfish Competitor on Wrasse Survival Estimation of Functional Response Curves Behavioral Analysis Results Experiment One: Effects of Predator and Conspecific Density on Prey Survival

8 Experiment Two: Effect of a Heterospecific Competitor on T. amblycephalum Survival Discussion REDATOR DENSITY AND TIMING OF ARRIVAL AFFECT REEF FISH COMMUNITY ASSEMBLY Background Materials and Methods Study System, Site, and Species Quantifying Temporal Patch Occupancy of Hawkfish Effects of Predator Density, Temporal Variability, and Timing of Arrival Analysis Abundance and Alpha Diversity Rarefied Alpha and Beta Diversity Species Composition Results Predation Intensity Predator Variance Timing of Predator Arrival Discussion Diversity Response Temporal Variance Timing of Arrival INDEPENDENT EFFECTS OF HABITAT CONFIGURATION AND PREDATORS IN REEF FISH COMMUNITY ORGANIZATION Background Materials and Methods Study System Observational Study Experimental Study Data Analysis Results Observational Study Experimental Study Habitat Effects Predation Effects Discussion Habitat Effects Predation Effects BIOGEOGRAPHY DRIVES MARINE FOOD WEB STRUCTURE Background Materials and Methods

9 Results Discussion SUMMARY NESTED HIERARCHY OF MODELS LIST OF REFERENCES BIOGRAPHICAL SKETCH

10 LIST OF TABLES Table page 2-1 Results for each species. Settlement of all recorded species of fish summed over 12 sites and 28 days Description of predator densities across the natural array. Predator community on experimental patch reefs for Experiment Two Functional Responses with Predator Density. Functional response models Recruitment of all recorded species of fish summed over 18 reefs for small and large patches in the absence and presence of predators A-1 Parameters for type II functional response modified to incorporate both depletion and baseline mortality

11 LIST OF FIGURES Figure page 2-1 A standardized habitat unit Effects of habitat availability on patterns of relative settlement for four focal fish species The spatial distribution of fish on the circular array of ten SHUs in the high habitat treatment Extrapolated effects of adjusted habitat availability on settlement and recruitment of Dascyllus flavicaudus The construction of Experiment 1 reef arrays The effect of wrasse density on the per capita mortality of wrasses and the functional response of hawkfish The effect of a competitor, damselfish, and wrasse density on the per capita mortality of wrasses and predator functional response Five experimental predator treatments simulating characteristic patch occupancy of hawkfish observed in surveys Examples of six characteristic patterns of patch occupancy by hawkfish from surveys of 192 reefs in the north lagoon of Moorea from Feb June Summary of surveys of 192 patch reefs on the northern shore of Moorea from Feb-June Effect of hawkfish on prey abundance and alpha diversity Beta diversity of species incidence at 120 days Correlation between habitat availability and the abundance and diversity of predators and prey Effect of habitat configuration and predation on abundance and diversity of prey Composition and beta diversity of fish communities based for two metrics based on species incidence and species relative abundance of communities Main effects of predation and fragmentation on prey community composition based on species incidence and species relative abundance Main effects of fragmentation and predation on the relative abundance

12 6-1 Predators have greater pelagic larval durations relative to prey Positive correlation of prey diversity and predator diversity for coral reef fish from 55 reefs across the South Pacific The total diversity of coral reef fishes as a function of reef size and reef isolation Relationship between predator-prey ratio, reef size and reef isolation Number of predator and prey species across reefs of varying size and isolation Variation in the predator-prey ratio across 55 reefs A-1 Nested models in Experiment 1 examining the effect of a doubling in hawkfish density on attack rate and handling time of hawkfish A-2 Nested models in Experiment 2 examining the effect of damselfish addition on attack rate, handling time, and effective predator density

13 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 PREDATION AND HABITAT DRIVE REEF FISH COMMUNITY ORGANIZATION Chair: Craig W. Osenberg Major: Zoology By Adrian Craig Stier May 2012 Here I examined the ecological drivers of coral reef fish abundance and diversity in the South Pacific. The availability, spatial arrangement, and fragmentation of coral habitat is heterogeneous at local and global scales. The first goal of my research was to document how fish communities vary in abundance, diversity, and composition. The second goal was to quantify how input of larvae shifts across heterogeneous landscapes. The third goal was to describe how competition and predation reshape patterns of abundance and diversity initially established at colonization. A suite of surveys in Moorea, French Polynesia documented substantial spatio-temporal variation in the abundance and diversity of fishes. Using two experiments where I manipulated habitat availability, spatial arrangement, and fragmentation I showed that the supply of larvae can vary substantially across patches of variable configuration. Using a combination of modeling and experiments I showed that patterns initially established at settlement can be substantially modified by competition and predation post-settlement. In subsequent experiments I isolated the effects of predators from habitat configuration. I found that, for a focal predator, the density of predators does not affect each predator s foraging behavior but that the presence of a competitor can substantially augment the 13

14 rate at which predators consume prey. I also showed that at the community level, the effect of predators on prey abundance, diversity, and species composition is a function of both predator density and predator timing of arrival. Lastly, I explored patterns of predator and prey biodiversity at large spatial scales spanning 55 islands in the South Pacific. Specifically, I showed that prey are much more sensitive to shifts in island size and isolation relative to predators, suggesting trophic level dependency of species to biogeographic characteristics. Collectively my findings show that reef fish communities are exceptionally variable at multiple spatial scales, and that this variation is a driven by a suite of ecological processes including dispersal, competition, and predation. The research offers general insight into the interface between landscape ecology and trophic interactions. 14

15 CHAPTER 1 INTRODUCTION Theoretical Context Understanding the contribution of different ecological processes to patterns of abundance and biodiversity is a fundamental goal of community ecology. Historically, the theoretical literature and associated experimental tests have been heavily dominated by the assessment of the interactions between predation and competition (Chase et al. 2002). Less studied, however, are the interactions between other factors. Recently, a number of theoretical studies have begun to narrow this gap, generating explicit predictions by modeling interactions between different ecological factors such as predation and disease (Holt and Roy 2007), predation and abiotic disturbance (Gallet et al. 2007), and predation and species area relationships (Ryberg and Chase 2007). Empiricists are therefore charged with testing this new theory to continue integrating sub-disciplines of ecology. Here I focus on the interface between supply side dynamics (i.e. larval supply), habitat characteristics (availability, configuration, and fragmentation), and species interactions (competition and predation). Using coral reef fish ecosystems as a model, I use a series of observational studies, experiments, and modeling exercises to address the independent and combined effects of larval supply, habitat configuration, competition, and predation in driving the abundance and diversity of fishes. Outline My dissertation contains five data chapters (Chapter 2 Chapter 6). In Chapter 2, I show how the availability and spatial arrangement of habitat affects fish production; I combined field experiments, existing data, and a simulation model to quantify the 15

16 relative importance of two phenomena in driving increased production of habitat dependent organisms when habitat availability is increased: 1) greater input of new colonists to sites with more habitat; and 2) reduced competition for space in areas with more habitat. In Chapter 3, I focus on how predator-predator interactions vary with prey availability. I adapted existing models of predator foraging behavior and fit these mechanistic foraging models to experimental data to determine how the dilution of predators at high predator shifts in the presence of competition between predators and the presence of an additional prey species. Our analysis highlights a model-fitting approach that discriminates amongst multiple hypotheses about predator foraging in a community context and allows us to attribute our findings to a combination of short-term apparent competition and shifts in the time it takes for predators to "handle" their prey. Chapter 4 examines how effects of predators vary with predator density and timing of arrival. Across a landscape of discrete habitat patches, differences in predator colonization, mortality, immigration and emigration can produce a mosaic of predation pressure. Classic studies of predation have emphasized the role of different predator densities in driving in both the mean intensity but also variability in their press/abs and their relative timing of arrival. Most empirical studies of predation use simple experimental approaches to quantify the effects of predators on prey (e.g., using constant densities of predators, such as ambient vs. zero). I expanded on previous studies by comparing the effects of predator density to the role of predator timing of arrival (i.e., early or late relative to prey), and temporal variability in predation pressure on prey abundance, diversity, and composition. My results suggest that the timing of 16

17 predator arrival can be as important as predator density in modifying prey abundance and community composition. In Chapter 5, I examined how habitat fragmentation and size modify top-down effects of predators. Reduced habitat availability and increased efficiency of predator foraging can lead to increased predation pressure in areas where habitat has been fragmented or lost. However, the majority of studies examining the role of habitat in modifying the effects of predators have focused on a single prey species. I used an experiment to document the response of the entire prey community to the simulated loss and fragmentation of coral habitat. Habitat characteristics and predators each played a strong but independent role in affecting prey communities; predators reduced diversity by disproportionately removing rare species. Habitat loss and fragmentation each reduced diversity with even effects across all prey-abundance classes. In Chapter 6, I explore the interface between island biogeography and food web ecology at large-spatial scales to show trophic level dependency in species sensitivity to biogeographic features. Specifically, mining a database of species diversity for 55 different islands across the South Pacific I show that predator diversity is less sensitive to shifts in island size and isolation than prey diversity. I discuss a series of possible mechanisms that may produce this phenomenon including differences in dispersal capacity, diet specialization, and variation in top-down control. These findings provide a novel and unique perspective into the processes governing patterns of coral reef fish biodiversity at large spatial scales. Study System Coral reefs are structurally complex limestone habitats that form in shallow coastal waters of the tropics. Reefs can form near-shore and extend hundreds of kilometers in 17

18 shallow offshore environments. Coral reefs are created by sedentary cnidarians (corals) that accrete calcium carbonate. The majority of the reef structure is dead coral skeleton laid down over millennia, covered by a thin layer of live coral tissue that slowly accretes new limestone. Additionally coral reefs provide a number of ecosystem services to humans including raw materials, coastal protection, maintenance of fisheries, nutrient cycling, and tourism, recreation, education and research (Barbier et al. 2011). Coral communities are extremely diverse. Functionally, species can be loosely divided into primary and late successional species. Primary successional species are massive boulder shaped corals that have dense skeletons and are therefore resistant to disturbance and form the foundation of the reef (e.g., Montastrea in the Atlantic and Porites in the Pacific). These initial colonists are outcompeted by fragile but structurally complex late successional species (e.g. Acropora and Pocillopora), which form much of the habitat occupied by the diverse suite of coral-associated invertebrates and fishes. The benthic composition of reefs (i.e. coral, invertebrate, and fish assemblages), can, however, substantially vary across oceanic basins, longitude, latitude, distance for biodiversity hotspots, and across physical disturbance gradients. Reef community compositioncan additionally be affected ecological processes such as predation, competition, and mutualisms. Thus the community composition of reefs depends on global, regional, and local factors, which interact to produce the wide variety of corals reefs present on earth (Glynn 1976, Pandolfi 2002, Bellwood et al. 2005, Cornell et al. 2007). Below I describe how variation in the abundance, diversity, and composition of reef fish communities is a product of multiple simultaneously operating ecological processes. 18

19 CHAPTER 2 PROPAGULE REDIRECTION: HABITAT AVAILABILITY REDUCES COLONIZATION AND INCREASES RECRUITMENT IN REEF FISHES Background The spatial configuration, quality, and size of habitat patches can determine the distribution and abundance of organisms by affecting colonization, movement, and the strength of species interactions (MacArthur and Wilson 1967, Hanski 1998). Understanding the role of habitat is particularly important in light of tremendous variation in habitat availability driven by both natural (Porter 1972, Gardner et al. 2003, Silliman et al. 2005, Bruno and Selig 2007) and anthropogenic (Barel et al. 1985, Skole and Tucker 1993, Bellwood et al. 2004) causes. Environmental change often involves the loss (or gain) of a biogenic habitat critical to the persistence of other organisms. Thus, many restoration techniques focus on re-establishing the structural attributes of a system (e.g., trees or corals) based on the Field-of-Dreams Hypothesis: if you build it, they will come (a reference to the 1989 movie; Palmer et al. 1997). This concept may be useful in understanding the response of species to restoration of degraded habitat, but it also may provide a valuable foundation for understanding the dynamics of species more generally, especially in response to natural fluctuations in habitat availability. The Field-of-Dreams s proposes that increased habitat availability will lead to a proportionate increase in colonists, with no resulting decrease in density at previously existing sites. Alternatively, new habitat may simply redirect colonists away from other suitable sites (Carr and Hixon 1997, Osenberg et al. 2002a): the Propagule Redirection Reprinted with permission from Ecology. Stier, A. C. and C. W. Osenberg Propagule redirection: Habitat availability reduces colonization and increases recruitment in reef fishes. Ecology 91:

20 Hypothesis. Propagule redirection may create settlement shadows (sensu Jones 1997) and unintentionally contribute to the further degradation of existing sites by removing potential colonists (Osenberg et al. 2002a). However, if post-colonization processes are density-dependent, then redirection may still enhance the population by ameliorating deleterious effects of density in existing habitat. The net effect of redirection will depend on the strength of density dependence, the overall level of colonization, and the magnitude of propagule redirection (Osenberg et al. 2002a). Effects of propagule redirection on the response of populations to altered habitat availability also will depend upon connectivity, temporal scale, and habitat quality (Resetarits 2005, Resetarits and Binckley 2009) (see Discussion for more detail). Studies of larval depletion by predators in which a wall of mouths (predators) reduce settlement to downstream habitats (Gaines and Roughgarden 1987, Hamner et al. 1988, Peterson and Black 1991) provide compelling evidence that biological processes can modify patterns of larval settlement. In a different context, foraging studies of pollinators suggest that the addition of flowers can lead to reductions in local density of the pollinators (reviewed in Mitchell et al. 2009). There are, however, few experimental studies of the effects of habitat on colonization per se. In one of the best examples, Resetarits (2005) examined oviposition by a tree frog (Hyla chrysoscelis) to sets of pools that varied in the proportion of sites with predators. Oviposition was greater in predator-free ponds that had the fewest number of neighboring ponds with predators, demonstrating selective oviposition at both local and landscape scales. However, this work did not directly address redirection of ovipositing females (and their eggs) because the fates of the females (and their eggs) that could have oviposited at 20

21 sites with some predator-containing ponds, but chose not to, were unknown: did they perish or were they redirected to other pools? In contrast, Resetarits and Binckley (2009), working with aquatic beetles, also varied the number of pools, keeping composition fixed. They found a proportionate increase in colonization rates as the number of pools increased, corroborating the Field-of-Dreams Hypothesis. However, this seminal (and to date, unique) study did not examine or extrapolate their results to post-colonization dynamics. Propogule redirection is particularly relevant in species with demographically open local populations in which organisms must find and colonize new sites during each generation, and where colonists have the potential to colonize many possible sites. This phenomenon is particularly relevant to many benthic marine organisms whose propagules are capable of dispersing long distances over short periods of time. For example, the majority of reef fishes have a bipartite life history, with pelagic larvae that disperse tens to hundreds of kilometers from their natal site. Here, we 1) report a field test of the Propagule Redirection and Field-of-Dreams Hypotheses for coral reef fishes, 2) parameterize a model that integrates our observed levels of propagule redirection and previous experimental estimates of density dependence to evaluate the implications of propagule redirection on fish recruitment (i.e., post colonization dynamics), and 3) use the model to assess the relative importance of increased colonization, propagule redirection, and reduced density dependence on fish recruitment. 21

22 Materials and Methods Study Site and Species We conducted fieldwork on the northern shore of Moorea, French Polynesia (17 30 S, W), a shallow lagoon with interspersed sand and reef inside a barrier reef (Galzin and Pointer 1985). Our experiment had two treatments (high habitat vs. low habitat availability) arranged in twelve experimental blocks. To assess the effects of habitat availability on redirection we used a competition design, in which we monitored settlement to focal habitat that either did or did not have neighboring sites competing for larvae. Each location was an open sand flat (at a depth of ~4m) with no other hard bottom habitat, so that each treatment would receive approximately equal larval input within a block without influences from neighboring natural reefs. To make our experimental arrays as uniform as possible, we constructed 168 Standardized Habitat Units (Forrester 1990) by attaching colonies of Pocillopora verrucosa (~15 cm diameter) to cinderblocks (one coral colony per cinderblock) using Z- spar marine epoxy (Splash Zone Compound, Kopcoat, Pittsburg, Pennsylvania, USA). Each of 12 sites (i.e., statistical blocks) contained one replicate of each treatment and consisted of 14 SHUs. Pairs of focal SHUs for each treatment were placed 16 meters apart (and perpendicular to the dominant current direction with at least 15 m to the nearest natural reef). We then surrounded one pair of focal SHUs by 10 single SHUs arranged in a two meter diameter circle (Fig. 2-1). This constituted the high habitat treatment. In the low habitat treatment, the two focal SHUs lacked neighbors. The Field of Dreams and Propagule Redirection Hypotheses differ in their predictions about the input of colonists (settlement) as well as the longer-term effects on 22

23 recruitment (i.e., establishment of older life stages). Because the presence of older life stages (or heterospecifics) can modify settlement patterns (Schmitt and Holbrook 1996), we quantified and removed settlers on a daily basis (see next paragraph). This allowed us to unambiguously evaluate the key prediction that distinguishes the Field-of-Dreams and Propagule Redirection Hypotheses: i.e., does added habitat proportionally increase settlement or simply redistribute settlers? However, this approach precluded us from quantifying subsequent post-settlement recruitment patterns. We therefore used existing data on the strength of post-settlement density-dependent survival to extrapolate the observed settlement effects to their longer-term recruitment consequences. We monitored settlement daily for 28 days from June 15 July 12, Each day between the hours of 0730 and 1130, all fishes that had settled the previous night were removed using handnets and the anesthetic Eugenol (Munday and Wilson 1997). By removing fish daily, we 1) minimized mortality prior to sampling (most fish experience high lev els of mortality during the first 1-2 weeks after settlement thus weekly sampling would be too coarse), 2) reduced the effects of predator aggregation to sites with high prey density, and 3) eliminated potential confounding effects such as densityindependent or density-dependent competition or predation by older conspecifics and heterospecifics. Because recently settled fish are highly site-attached to a given coral, especially during their first 24 hours on the reef (Holbrook and Schmitt 2002), we assumed that new settlers had not moved between experimental arrays. Data Analysis We collected settlers from 16 species and 6 families (Table 2-1), but only four species, which together comprised over 88% of all settlers, were sufficiently abundant to 23

24 permit statistical analysis: Dascyllus aruanus (Pomacentridae), Dascyllus flavicaudus (Pomacentridae), Halichoeres trimaculatus (Labridae), and Paragobiodon lacunicolus (Gobiidae). We quantified settlement for these four species as the cumulative number of individuals that settled to the SHUs during the 28 days of sampling. To test for propagule redirection, we compared settlement of each fish species to the two focal SHUs with vs. without neighbors (i.e., at high vs. low habitat availability). That is, for each block, we calculated ln(s -neighbors /S +neighbors ), where S is the cumulative settlement over 28 days to the focal SHUs that were isolated (no other corals within at least 15 m: - neighbors) or to the two focal SHUs surrounded by 10 neighboring SHUs (+ neighbors). In the absence of propagule redirection, the numbers of settlers should be the same in these two treatments: i.e., ln(s -neighbors /S +neighbors ) = 0. If additional habitat depletes larvae and therefore reduces settlement to nearby habitat, then settlement to the two isolated SHUs should be greater than settlement to the two SHUs surrounded by neighbors: i.e., ln(s -neighbors /S +neighbors ) > 0. To determine whether increased habitat availability affected total settlement to the entire experimental array, we compared the cumulative number of individuals that settled to the twelve SHUs at high habitat availability (i.e., to the two focal SHUs plus the ten neighboring SHUs) versus low habitat availability (i.e., to the two isolated SHUs): ln(s high /S low. Note that S low = S -neighbors, but we use different terminology to make our notation more intuitive for each comparison. The Propagule Redirection Hypothesis (in its extreme) predicts that settlement to the high habitat arrays should be the same as settlement to low habitat arrays (S high /S low = 1), whereas the Field of Dreams hypothesis 24

25 predicts that the relative settlement should be proportional to habitat availability (e.g., S high /S low = 12/2 = 6: but see Results for a modification to this simple expectation). The circular array provided additional information on propagule redirection because the current at our sites was predominantly unidirectional, coming in over the reef crest to the north of the study sites and flowing over the sites before exiting the lagoon through deeper channels (Hench et al. 2008). Thus, we compared cumulative settlement to the SHUs in the upstream (i.e., most northerly) half of the circle to the downstream half. This statistical approach was similar to the previous two, except that we added 1 to the number of settlers due to the presence of zeros at some downstream sectors: ln[(s up +1)/(S down +1)]. For all analyses, we used paired t-tests, comparing ln(s) (or ln(s+1)) for the two treatments (n = 12 blocks). Projected Recruitment Success To assess how effects of habitat availability on fish settlement translate to changes in fish recruitment, we started with the standard formulation of the Beverton-Holt recruitment function (e.g. Osenberg et al. 2002c): D t = e -at D 0 1+ b(1- e-at )D 0 a, (2-1) where D t is the density of recruits to a site (i.e., settlers who survive t days); D 0 is the initial density of settlers to that site, a is the per capita density-independent mortality rate, and b is the per capita density-dependent mortality rate. We modified Equation 2-1 (Osenberg et al. 2002c) to incorporate habitat availability, and parameterized the model using existing field data on density dependence in D. flavicaudus, one of our focal species and a species known to compete for predator-free space within coral heads 25

26 (Schmitt and Holbrook 2007). Because we were interested in the separate effects of habitat (h) and numbers (N), per se, we re-expressed density (D = N/h) to obtain: N t = 1+ e -at N 0 b(1- e -at ) N 0 h a, (2-2) where N t is the number of recruits to a site (i.e., settlers who survive t days); N 0 is the initial number of settlers to that site, a is the per capita density-independent mortality rate, b is the per capita density-dependent mortality rate, and h is the amount of habitat (i.e., N 0 /h is the density of settlers). Notice that the parameters have the same meanings and units in Equation 2-2 as in Equation 2-1 (i.e, a: 1/day, t: day; b: corals / fish / day). We then parameterized the Beverton-Holt model (fitting a and b) using data from previous studies of density dependence conducted on one of the focal species, D. flavicaudus (Holbrook and Schmitt 2002, Schmitt and Holbrook 2007). The corals used by Schmitt and Holbrook (2007) were approximately twice the areal dimension of ours, so we parameterized the model by doubling their coral numbers to reflect the difference in areal cover of corals in the two studies (i.e., h is expressed as the number of corals of the size used in our study). Recruitment patterns of newly settled reef fishes, including D. flavicaudus, are driven by high mortality within the first few days following their arrival to the reef (Almany and Webster 2006, Schmitt and Holbrook 2007). We therefore calculated the expected number of survivors after two days on the reef (i.e., recruitment ). This time scale is the same as the one used to generate the data with which we parameterized the model (Schmitt and Holbrook 2007). We assumed that density dependence was most strong among fish that settled within 2 weeks of one another, and therefore used 26

27 the total number of settlers per coral observed in our 28 day (i.e., + 2 weeks) field study. We then compared the levels of settlement and recruitment in a system with low habitat availability (h = 1) to one with high habitat availability (h = 2.75); this 2.75 fold difference in habitat availability is within the range of habitat losses observed in coral reef systems (Connell et al. 2004, Bruno and Selig 2007). Expected settlement intensities were taken from our field experiment (low vs. high habitat availability) for D. flavicaudus, but we also examined the results under much lower and higher settlement intensities to better generalize the results. Habitat addition can increase recruitment and local density in two ways: 1) increased settlement (because there is more habitat: i.e., Field-of-Dreams); and 2) reduced density dependence (because settlers are spread out among more habitat: i.e., Propagule Redirection). The latter (reduced density dependence) only occurs if the increase in total settlement is less than proportional to the availability of habitat (i.e., if there is propagule redirection). We compared the relative importance of these two pathways by partitioning the potential change in recruitment due to these components. Habitat addition can increase recruitment and local density via two mechanisms: 1) increased settlement (because there is more habitat to attract larvae: i.e., Field-of- Dreams); and 2) reduced density dependence (because settlers are spread out among more habitat: i.e., Propagule Redirection). To partition the potential change in recruitment to these two components, we explored six recruitment levels defined by the two recruitment functions (low vs. high habitat availability: i.e., h = 1 or h = 2.75 in Equation 2-1 or 2-2) and three settlement levels (observed under low habitat availability, observed under high habitat availability, and the expected settlement if settlement was 27

28 proportional to habitat availability). We then used these recruitment levels to define: 1) the potential recruitment effect predicted by the Field-of-Dreams Hypothesis (i.e., the difference in recruitment predicted under high habitat vs. low habitat availability assuming settlement was proportional to habitat availability), 2) the realized recruitment effect (i.e., the difference in predicted recruitment under high vs. low habitat availability given observed settlement responses). We partitioned the realized effect into that attributable to: a) propagule redirection combined with reduced density dependence (i.e., the change in recruitment when habitat is increased but settlement remains at ambient levels); b) increased settlement only (i.e., the change in recruitment when settlement is increased but density dependence is unaffected); and c) the interaction between settlement augmentation and density dependence (arising from non-linearities in the recruitment function). The difference between the potential and the realized effects is the amount of recruitment under the Field-of-Dreams Hypothesis that was not achieved. Results Settlement Settlement was 2- to 4-times greater to the isolated focal SHUs than to the focal SHUs surrounded by neighboring SHUs, demonstrating that propagules were redirected: Fig. 2-2a, D. flavicaudus (t 11 = 11.67; P < ), D. aruanus (t 11 = 12.17, P < ), H. trimaculatus (t 11 = 6.22; P = ), and P. lacunicolus (t 11 = 3.31; P = 0.007). This pattern also was observed for the combined response of all other species that settled (Table 1), although one uncommon species (Acanthurus triostegus) suggested the reverse trend. 28

29 Settlement to the entire experimental array (e.g., the focal SHUs as well as any neighboring SHUs) was 16-66% greater for high vs. low habitat availability (Fig. 2-2b), although this increase was significant for only two of the four species analyzed: H. trimaculatus (t 11 = 3.36; P = ); D. aruanus (t 11 = 3.21; P = ), D. flavicaudus (t 11 = 1.86; P = 0.088), and P. lacunicolus (t 11 = 1.22; P = 0.25; Fig. 2-2). This ~36% increase in settlement (averaged across all four species) demonstrated that added habitat led to increased settlement. However, the response was far less than the predicted 500% (6-fold) increase expected under the simple Field-of-Dreams Hypothesis (Fig. 2-2b). Again, similar results were obtained for the combined response of all other species (with the exception of A. triostegus: Table 1). Within the circular grid of neighboring SHUs, there was a spatial pattern that further suggested propagule redirection (i.e., settlement shadows, sensu Jones 1997). Settlement to the five upcurrent SHUs was % greater than to the five downcurrent SHUs for all four focal species, although this pattern was significant for only two of the four species (Fig. 2-3): D flavicaudus (t 11 = 2.80; P = 0.017), H. trimaculatus (t 11 = 2.65; P = 0.023), D. aruanus (t 11 = 2.06; P = 0.062), and P. lacunicolus (t 11 = 0.44; P = 0.67). Our simple Field of Dreams prediction is based on the 6:1 disparity in number of SHUs in the two treatments. Complicating this prediction was the heterogeneous settlement patterns of fish on the array with neighbors: settlement was greater to upstream (vs. downstream ) neighbors (Fig. 2-4), but also greater to the focal SHUs (on a per SHU basis) relative to the neighbors (Fig. 2-4). For the four focal species, settlement to a neighbor SHU was only 35% of that observed to a focal SHU. The 29

30 magnitude of this effect did not differ among species. One explanation for this result is that fish settled preferentially to SHUs placed side-by-side (i.e., focal SHUs: see Fig. 2-1). We therefore adjusted the expectations under the Field-of-Dreams hypothesis to account for this possible difference in habitat quality. Instead of a 6-fold difference in habitat, we assumed a 2.75-fold difference: i.e., (10SHUs(0.35) + 2SHUs*):2SHUs. The observed settlement to the arrays with neighbors was still demonstrably lower than this expectation for three of the four fishes (Fig. 2-2). Projected Recruitment Success Although total settlement of D. flavicaudus was only 16% greater at high habitat availability, projected recruitment using the Beverton-Holt equation was 125% greater (Fig. 2-4b). Had settlement increased in proportion to habitat, recruitment should have increased 175% (i.e., 2.75-fold). In contrast, if there was complete propagule redirection, so that habitat addition did not increase settlement but only reduced densitydependent mortality, recruitment would have increased by 114% (i.e., most of the observed 125% increase). Thus, the observed increase in recruitment due to habitat enhancement was primarily due to a reduction in density dependence via propagule redirection. The circular array provided additional information on effects of the spatial structure of habitat in the circular array, suggesting patterns of settlement may also be affected by the arrangement of habitat on a local (meter) scale. The increase in recruitment (and the partitioning to different components) was sensitive to the overall level of settlement. For example, if ambient settlement was only 10% of that observed, then the 16% increase in settlement observed for D. flavicaudus to the high habitat sites would have led to an approximately similar increase in recruitment (in this case 17%) because the effects of density would be small at such low 30

31 input levels (Fig. 2-4a). Thus, the main effect of increased habitat availability would be via the increase in settlement (as slight as it was). On the other hand, if overall settlement were an order of magnitude greater than observed, then recruitment to the high habitat arrays would have increased by almost 93% despite the only modest (16%) increase in settlement to those arrays (Fig. 2-4c). If overall settlement increased even more (beyond the 10-fold change examined here), recruitment to the high habitat arrays would have increased in proportion to the change in habitat: i.e., by 175%). Thus, at high settlement, the main effect of habitat addition would be on the amelioration of density dependence rather than the increase in settlement because the available habitat was already saturated. Discussion Changes in habitat availability can affect population dynamics and species interactions by altering colonization rates and/or the strength of density-dependent interactions that arise following colonization. The life-stage responsible for colonizing new habitat varies among species. For example, in many freshwater and terrestrial systems adults colonize habitats, often as sites for oviposition (e.g., frogs (Resetarits 2005), aquatic insects (Stav et al. 1999, Resetarits 2001), and butterflies (Rausher 1979, Renwick and Chew 1994)). In other systems, earlier life stages are dispersive and adults are more sedentary (e.g., most marine organisms and plants). Although our study focused on marine fishes, these insights likely apply to a variety of systems in which local dynamics are relatively open. The simplest expectation is that colonization will be proportional to habitat availability (i.e., the Field-of-Dreams Hypothesis). In our study, increased habitat availability led to a significant increase in the total number of settlers, but the magnitude 31

32 of this increase was small (36% averaged across focal species) relative to the increase in habitat (175 or 500%). This small increase in settlement led to a greater increase in projected recruitment (based on field data and the habitat-modified Beverton-Holt model) because propagule redirection reduced the intensity of density dependence. Thus, increased habitat increased production in two ways: first, by increasing the colonization rate (i.e., more habitat led to more colonists), and second, by decreasing density at other sites and thus reducing the effect of density dependence (because of propagule redirection). This effect is likely general, although the overall effect and relative importance of increased settlement (Field-of-Dreams effect) vs. relaxed density dependence (via the Propagule Redirection effect) will likely depend upon system properties such as propagule supply, connectivity, temporal scale of reproduction (and overlap among cohorts), and the nature of interactions between colonists. Propagule Redirection: Connectivity and Temporal and Spatial Scales Variation in habitat availability, and predictions from different hypotheses (such as Field-of-Dreams and Propagule Redirection), will influence three processes that manifest over different time scales. The first step is initial colonization, which in our system was represented by settlement. The Field-of-Dreams and Propagule Redirection hypotheses differ in their predictions about colonization, but the implications of these hypotheses depend upon subsequent dynamics that play out over longer time scales, both within and between generations. Post-colonization survival translates settlement into recruitment. Our recruitment model incorporated effects of density dependence within cohorts, but over longer terms, as multiple cohorts potentially build-up at a site, effects among cohorts may also become important. Thus, recruitment dynamics will be affected by how many cohorts 32

33 occupy a local site and the patterns of density-dependence among those groups. For example, the Field-of-Dreams hypothesis works well even in the presence of propagule redirection if within-cohort density dependence is sufficiently strong (e.g., Fig. 2-3c). Similarly, if multiple cohorts co-occur and interact with one another, then Field-of- Dreams may also be realized even if there is strong propagule redirection. Accumulation of cohorts will intensify density dependence in a way analogous to our result for high settlement levels (Fig. 2-3c): if the combined effects of settlers and prior residents saturate the system, then there will be little effect of redirection. Conversely, weak density-dependence (within and among cohorts) will cause Field-of-Dreams to fail in the presence of propagule redirection (e.g., Fig. 2-3a). Finally, dynamics across generations will define how recruitment at one time step affects future changes in local population sizes. In relatively closed systems, the dynamics of a newly colonized patch will be largely internally driven (i.e., future colonists will be internally produced). In such a case, local reproduction will increase local abundance and this will eventually lead to a local abundance that is proportional to local habitat availability. Thus, although the initial short-term response may be affected by propagule redirection, the long-term response will not. In open systems (with no selfrecruitment) the within-patch dynamics will always depend on the external supply of propagules. Thus, if local abundance is reduced due to propagule redirection it cannot be regained by future within-patch dynamics. Propagule redirection may be important even if its effects are transient, because it can influence the time scale for recovery of ecosystems following a disturbance or in response to seasonally regenerating habitats. Populations that experience frequent 33

34 disturbance may never reach their full production potential if the projected time to a fully recovered patch through slow natural colonization is long (relative to timing of disturbances: e.g., Robertson 1996). Alternatively, if colonization of newly available habitat is high and does not deplete colonization of other habitats then the overall system may recover from disturbances quickly. Thus, it may not be as important to know whether or not increased habitat availability will eventually reach its full potential, but at what time scale new habitats will fill-up. Propagule Redirection: Unbridled Conjecture Propagule redirection can increase recruitment if it ameliorates negative effects of density (e.g. via predator attraction or competition). In contrast, if density has beneficial effects (e.g. predator dilutionand mate attraction: Sweatman 1985, White et al. 2010), then propagule redirection could slow (or even negate) positive responses to increased habitat availability. This may be pronounced in systems with biogenic habitat if the habitat benefits from animals that occupy the habitat. As in ant-plant or flower-pollinator mutualisms (reviewed in Mitchell et al. 2009), some invertebrates and fishes (including D. flavicaudus) use coral for structure and/or food and also provide positive benefits to the coral (Sweatman 1985, Goldshmid et al. 2004, Stewart et al. 2006). If there is propagule redirection (Fig. 2-2a), then corals effectively compete for colonists of their mutualists. This may have led to the evolution of signaling systems (between corals and their symbionts) that have not yet been fully appreciated by marine scientists. Recent studies demonstrate that many marine organisms can locate and orient to cues produced by their habitat (Lecchini 2004, Simpson et al. 2005). Thus, some marine systems may be more analogous to co-evolved terrestrial plant-pollinators than previously thought. Indeed, because many habitats are biogenic, future studies should 34

35 incorporate propagule redirection and the effects of changing densities on the dynamics of the colonists as well as their habitat. 35

36 Table 2-1. Results for each species. Settlement of all recorded species of fish summed over 12 sites and 28 days. Circle refers to 10 neighboring SHUs on the outer rim of the high habitat availability treatments; Focal refers to the two focal SHUs in each treatment, either surrounded by neighboring corals (high habitat availability) or not (low habitat availability treatment). Total Settlement High Habitat Availability Low Habitat Availability Family (common name) Circle Focal Focal Genus species Holocentridae (Squirrelfishes) Neoniphon sammara Apogonidae (Cardinalfishes) Apogonichthys ocellatus Pomacentridae (Damselfishes) Chromis viridis Dascyllus aruanus Dascyllus flavicaudus Pomacentrus pavo Labridae (Wrasses and Parrotfishes) Halichoeres trimaculatus Pseudocheilinus hexataenia Scarus oviceps Chlorurus sordidus Gobiidae (Gobies) Eviota sp Paragobiodon lacunicolus Paragobiodon modestus Priolepis squamogena Acanthuridae (Surgeonfishes) Acanthurus olivaceus Acanthurus triostegus Total

37 Figure 2-1. A standardized habitat unit (SHU). A) the spatial arrangement of a single replicate block (consisting of 14 SHUs: two alone, two in the center of the circle, ten on the edge of the circle) B), showing the treatments with (left hand side) and without (right hand side) neighboring SHUs; and C) a map of the north shore of Moorea, showing the approximate locations of the 12 experimental blocks. 37

38 Figure 2-2. Effects of habitat availability on patterns of relative settlement for four focal fish species. A) The effect of habitat treatments on the average relative settlement of fish to focal SHUs with vs. without neighboring reefs: S -neighbors /S +neighbors. The dashed line (Field-of-Dreams) represents the expected relationship if neighboring reefs have no effect on settlement to the focal reefs. B) The effects of habitat on the relative total settlement of fish to the entire experimental array (i.e., to 12 SHUs vs. 2 SHUs): S high /S low. The lower dashed line gives the expected result if larvae are fully redirected (i.e., total settlement is the same to both arrays, as in the extreme version of the Propagule Redirection hypothesis). The two upper dashed lines give the expected results if there is no redirection (i.e., Field-of-Dreams hypothesis) based on the 6-fold increase in habitat area or 2.75-fold increase in habitat availability (adjusted for quality). Points and error bars represent back transformed means 95%CI. 38

39 Figure 2-3. The spatial distribution of fish on the circular array of ten SHUs in the high habitat treatment. Wedge size represents the relative magnitude of settlement averaged across the twelve sites. Points give the back-transformed mean (+/- 95%CI) relative settlement to the five upcurrent SHUs vs. the five downcurrent SHUs: (S up +1)/(S down +1), n=12 replicate arrays. 39

40 Figure 2-4. Extrapolated effects of adjusted habitat availability on settlement and recruitment of Dascyllus flavicaudus, under: A) low (10% of observed), B) medium (observed), and C) high (10 times observed) levels of settlement. For each panel, recruitment functions are given for low (lower curve) and high (upper curve) adjusted habitat availability (where h high /h low =2.75). Three settlement intensities (total number of fish to an array) are indicated with vertical dashed lines: settlement at low habitat availability (S Low ), settlement at high habitat availability (S High ), and expected settlement in the absence of redirection (i.e., if settlement was proportional to habitat availability: the Field of-dreams Hypothesis). In all panels, S High =1.17S Low (i.e., the observed response of D. flavicaudus to the increase in habitat: Fig. 2-2b). Solid points give the expected recruitment and settlement under low and high habitat availability with observed settlement, and under high habitat assuming proportional settlement. Open points give the expected recruitment and settlement if only density dependence (or only settlement) was affected by habitat addition. The difference in recruitment between the two most extreme recruitment levels gives the potential effect on recruitment under the Field-of-Dreams Hypothesis (i.e., the expected increase in recruitment if a 2.75x increase in habitat led to a 2.75x increase in settlement). That potential can be divided into four components (histograms on the right), the first three of which comprise the realized effect: Propagule Redirection (PR) which ameliorates density dependent mortality, increased settlement (S), the interaction between density dependence and settlement (I), and the remainder (NA), which represents the potential recruitment that is not achieved 40

41 CHAPTER 3 PREDATOR DENSITY AND COMPETITION MODIFY THE BENEFITS OF GROUP FORMATION IN A SHOALING REEF FISH Background The importance of prey density in modifying predator foraging behavior, and hence community dynamics, has a rich history in ecology (Nicholson and Bailey 1935, Holling 1959). For example, shifts in a predator s foraging behavior can affect the stability of predator-prey dynamics (Deangelis et al. 1975), spatial distribution of predators (Van Der Meer and Ens 1997), food chain length (Schmitz 1992), and the strength of species interactions in complex food webs (Novak and Wootton 2008). The functional response represents mechanisms underlying the prey-predator interaction; thus, quantifying changes in functional response parameters is a natural way to test hypotheses about these mechanisms. Shifts in predator foraging response can occur due to predatorpredator interactions (e.g., intraguild predation, interference competition, or cooperative hunting by predators) (Skalski and Gilliam 2001) as well as common or conflicting behavioral responses of prey to multiple predators (e.g. Sih et al. 1998). Predatorinduced benefits of group living are common for a number of taxa, from caribou (Wittmer et al. 2005) to cliff swallows (Brown and Brown 2003) to queen conch (Ray and Stoner 1994), and active group formation by prey in the presence of predators (e.g., schools of fish and herds of savannah ungulates) or group formation of predators hunting prey can stabilize predator-prey dynamics by modifying the functional response (Fryxell et al. 2007). Below we describe a first experiment where we estimate how two components of a predator s functional response (attack rate and handling time) vary under different predator densities, and link these changes to mechanistic hypotheses. 41

42 Functional responses can also be used to study the effect of multiple prey species on predator foraging behavior (e.g., Murdoch 1973, Golubski and Abrams 2011). For example, an alternative prey species can increase predator attack rates on focal prey by competitively excluding a focal species from refuges. Alternatively, the presence of alternative prey could drive short-term aggregation of mobile predators, again increasing rates of focal prey mortality (Holt and Kotler 1987). Shifts in attack rate and handling have conventionally been examined at fixed predator densities, whereas shortterm aggregative responses of predators are often studied by documenting short-term shifts in predator abundance (e.g. Schmitt 1987, Overholtzer-McLeod 2006). It can be difficult to document short-term aggregative response of highly mobile predators and to distinguish an aggregative response from a shift in attack rate or handling time of sedentary predators. In a second experiment, we quantify mortality rates of a focal prey species in the presence and absence of an alternative prey that is also a likely space competitor. We develop and apply an explicit statistical framework that combines predator functional response curves and short-term aggregative response of predators by estimating shifts in attack rate, handling time, and effective predator density as alternative mechanistic models explaining variation in mortality rates of prey. For each experiment, we take both a "predator-centric" (examining feeding rate of a single predator, as a function of prey density) and "prey-centric" approach (examining per capita prey mortality rate, as a function of prey density). Methods Study System and Species We conducted all fieldwork during the austral summer of 2007 on two arrays of patch reefs within the shallow (3-7 meters deep) northern lagoon of Moorea, French 42

43 Polynesia (17 30 S, W). The array used in Experiment 1 consisted of 12 replicate artificial patch reefs separated by 20 m within a sand flat (Fig. 3-1). The array used in Experiment 2 consisted of 16 reefs randomly selected from 28 patch reefs that had been translocated to a large sand flat. Patch reefs were separated by a minimum of 15 meters. Each patch reef was a monospecific colony of Porites lobata to which we attached one colony of Pocillopora verrucosa to provide additional structural complexity (for specific details on the array see Geange and Stier 2009). Each of our experimental reef arrays closely mimic the size and spatial arrangement of patch reefs in Moorea, where the lagoon is interspersed with patch reefs within a matrix of sand, fine coral rubble and coral pavement with individual patch reefs separated by less than a meter, to tens of meters (Galzin and Pointer 1985). These patch reefs are regularly occupied by a variety of wrasse, damselfish, and hawkfish species. Our studies focused on three species: a focal prey species, the bluntnose wrasse (Thalassoma amblycephalum), an alternative prey species, the sapphire damsel (Pomacentrus pavo), which competes with the bluntnose wrasse for space on the reef (Almany 2004b), and a predator, the arc-eye hawkfish (Paracirrhites arcatus). All three species use Pocillopora spp. (a structurally complex coral) for shelter. The bluntnose wrasse (hereafter wrasse ) is a habitat generalist that feeds diurnally on zooplankton; recently settled individuals shoal above coral colonies in aggregations of approximately 5-15 individuals (McDermott 2006). The sapphire damsel, like the wrasse, is a diurnally feeding planktivore and habitat generalist, often forming loose shoaling aggregations on live coral or coral rubble. The sapphire damsel ( damselfish ) settles at densities from 1-35 individuals per m 2 (Stier and Geange pers. obs), and is known at adult sizes to be 43

44 territorial and aggressive towards heterospecifics inhabiting the same habitat patch, inhibiting the recruitment of newly settled fishes (Almany 2004b). The arc-eye hawkfish ( hawkfish ) is an abundant, diurnally active, territorial predator. Individuals scan the surrounding reef from atop coral colonies of Pocillopora and Acropora spp. for small fishes and invertebrate prey. Hawkfish live in loose harems, with large dominant males accompanied by 4-6 females on neighboring coral heads. Within a given Pocillopora colony, the abundance of hawkfish in Moorea ranges from 0-2 individuals (Kane et al. 2008). Fish Handling and Tagging We collected experimental fishes from the lagoon using hand nets and anesthetic clove oil. Collected fish were transferred to holding tanks for 24 h. After 24 h, fish were randomly assigned to treatments and tagged with subcutaneous Visible Implant Elastomer (VIE) tags (Northwest Marine Technology, Shaw Island, Washington, USA) slightly anterior to the caudal peduncle. VIE tags do not have adverse effects on other fishes (Beukers et al. 1995, Imbert et al. 2007, Simon 2007) and mortality after tagging was less than 2% for 330 wrasses tagged in our study. To allow the assessment of immigration and emigration between reefs, different colored tags were used for each reef. After tagging, we returned fish to aerated aquaria for 6-12 hr before measuring them to the nearest 0.1 mm SL and deploying them in the field. VIE tags were readable through the skin of the fish by observers in the field, so it was not necessary to recapture individuals to determine their identity. We therefore assumed that tagging and handling effects were minimal. Our visual recapture of fishes on experimental reefs and nearby natural reef found no evidence for immigration or emigration of experimental fishes. Therefore the loss of a 44

45 wrasse from a reef was due to predation and not emigration. First, previous experimental studies of T. amblycephalum (wrasse) congeners of similar size on the array used in Experiment 2 detected zero immigration or emigration of tagged individuals between reefs over a seven day experiment (Geange and Stier 2009). Second, another six day study manipulating shoals of recently recruited Thalassoma wrasses on reefs separated by as little as 5 m have found no immigration of tagged individuals (White and Warner 2007). Experiment 1: Effects of Predator and Conspecific Density on Prey Survival We assessed the interacting effects of shoal size of the wrasse (4 levels: 3, 5, 10 or 15 wrasses per reef) and predator density (3 levels: 0, 1, or 2 hawkfish per reef) on the per capita survival of wrasses (12 treatments randomly assigned to 12 reefs). We removed all resident predators (including sandperch, groupers, stonefish, and eels) from the reef array prior to deploying experimental fishes to the reefs. We allowed wrasses to acclimate by deploying hawkfish 1 hr after wrasses. The standard length of wrasses and hawkfish used in the experiment were 13.8 ± 2.0 mm and 72.0 ± 3.5 mm (mean ± SD) respectively. We replicated our experiment in three temporal blocks (beginning May 2, 10, and 21, 2007), with no within-block replication (i.e., n = 3 temporal replicates for each treatment, which was randomized for each time step). For each temporal block, we maintained predator removals (although immigration of new predators was minimal after initial removals, with 2.0 ± 1.0 (mean ± SD) predators removed per temporal block). Each temporal block was terminated after six days because mortality rates of reef fishes are generally greatest in the first hrs after settlement (Almany and Webster 2006). 45

46 To determine whether shoaling behavior was affected by shoal size and/or predator density we quantified the shoaling behavior of wrasses during the second and third temporal blocks. Two divers sequentially visited each reef, allowing ten minutes for fish to acclimate to their presence before beginning observation. Each diver recorded the shoaling behavior of one focal wrasse for five minutes. Shoaling was quantified as the number of seconds out of 300 during which two or more wrasses swam within two body lengths of the focal individual. Experiment 2: Effect of a Heterospecific Damselfish Competitor on Wrasse Survival To test how the presence of an alternative prey species affects the strength of density dependence in wrasses, we crossed wrasse density treatments (3, 5, 8, 12, and 15 individuals per reef) with the presence of damselfish (0 or 20 individuals per reef) on reefs where predators existed at naturally occurring densities and often with heterospecifics (Table 3-1). To improve our estimates of survival we increased the number of replicates at low densities of wrasse: we used 3 replicates for the 3-wrasse treatment; 2 replicates for the 5-wrasse treatment; and 1 replicate each for the 8-, 12- and 15-wrasse treatments. We randomly assigned treatments to reefs, and fish to treatments. We deployed damselfish on the morning, and wrasses on the afternoon of May , and assessed survival of wrasses after six days. Mean sizes of wrasses and damselfish used in the experiment were ± 1.61 mm SD and 15.8 ± 1.16 mm respectively. Fish collections and tagging protocols were identical in Experiments 1 and 2. Although we attempted similar behavior observations of wrasses as were conducted in Experiment 1, we were unsuccessful because damselfish were particularly sensitive 46

47 to the presence and absence of divers making it difficult to monitor interactions between individual wrasses and damselfish. Estimation of Functional Response Curves To test for changes in predator foraging behavior, we examined changes in attack rates (a) and handling times (h) estimated using a modified form of the Rogers random predator (RRP) equation (itself a modified version of a Holling Type II functional response that incorporates prey depletion: Juliano 2001). In our version of the RRP equation, the expected rate of decrease of the prey population at time t depends both on a baseline extrinsic mortality rate and on instantaneous predation mortality, which we assume can be described using a Holling type II functional response: dn j (t) a = -(m + i P i dt 1+ a i h i N j (t) )N (t) j, (3-1) where is baseline mortality, P i is the number of predators, a i is the (per-predator) attack rate when P i predators are present, h i is the handling time (time associated with pursuing, handling, masticating and digesting prey) for P i predators, and N j (t) is the number of prey remaining at time t at wrasse density j. When = 0 Equation 3-1 reduces to the Rogers random predator equation, which can be solved for the total number of individuals eaten by time T either iteratively (Juliano 2001) or using the Lambert W function (McCoy and Bolker 2008); when > 0 we integrated (1) numerically from 0 to T because the equation cannot be solved in closed form. For Experiment 1, subscript i varies with hawkfish density treatment; thus attack rate and handling time are estimated for treatments with one or two hawkfish, and P is fixed at one or two in each respective treatment. For Experiment 2, subscript i varied with damselfish treatment (i.e., control or P. pavo present) rather than predator density. Wrasse mortality did not 47

48 significantly vary among temporal blocks (one-factor ANOVA on model residuals with temporal block as a factor: F = 0.92, p =.40). Notably, we did not observe each component of the functional response (i.e., attack rate and handling time) separately, but rather inferred these processes from the parameters of the fitted models. Using a general-purpose maximum likelihood estimation function assuming binomial responses to obtain parameter estimates for a Hawk and h Hawk, we used the integrated version of Equation 3-1 to address two specific questions about Experiment 1. First, we evaluated if per-predator attack rate a Hawk differed between hawkfish densities by parameterizing a Hawk as a 2 = a 1 da, and testing the null hypothesis that da = 1 (or equivalently log(da) = 0). The parameter da gives the proportional change in perpredator attack rate with two rather than one predators present: da = 1 suggests that the attack rate of a predator was independent of predator density, da > 1 suggests predator synergism, while da < 1 suggests predator interference. Second, we evaluated whether handling time h Hawk differed between the two predator densities by parameterizing h Hawk as h 2 =h 1 dh, and testing the null hypothesis that dh = 1 (or equivalently log (dh) = 0). We tested for changes in per-predator attack rate and handling time (i.e., whether da and dh were significantly different from 1) by using two separate likelihoodratio tests comparing the full model (with both da and dh different from 1) (i) to a reduced model with da=1 and (ii) to a reduced model with dh=1 (Appendix A). The full and reduced models described above are specific nested subsets of existing predator interference models reviewed in Skalski and Gilliam (2001): Beddington-DeAngelis (Beddington 1975), Crowley-Martin (Crowley and Martin 1989), and Hassell-Varley (Hassell and Varley 1969) (Table 3-2). 48

49 Examining shifts in attack rate and handling time in the presence of a competitor allows us to test three different hypotheses about why the number of wrasses eaten can change with the addition of damselfish. (1) The presence of damselfish increases competition for refuges, which decreases wrasses ability to hide. This competition increases encounter rates (i.e., attack rates) with predators, leading to da > 1. (2) Handling time decreases in the presence of damselfish due to decreased wrasse vigilance or exploitative competition for refuges from damselfish, leading to dh < 1). (3) Damselfish attract predators, which then also forage on wrasses (i.e., apparent competition): i.e., P Damsel > P Control. To test these three hypotheses we used a suite of modified functional response models to evaluate prey per capita mortality rates in Experiment two, but fixed the baseline mortality estimate (m from Experiment 1. To test Hypotheses 1 and 2, we parameterized a i, and h i, as a Damsel = a Control da and h Damsel = h Control dh, respectively, where once again da and dh give the proportional change in per predator attack rate and handling time. However, unlike Experiment 1 where predator densities were fixed as an experimental treatment, in Experiment 2 we did not manipulate predators directly. Therefore to test Hypothesis 3, we allowed P to vary with damselfish treatment as well by fitting a third model to ask whether differences in the number of predators could explain observed shifts in a i and h i in the presence of damselfish. We could not allow a i, h i and P i to vary simultaneously (they are jointly unidentifiable); therefore, in the models where we explored variation in a i and h i we fixed P = 1, and in the model where P i was allowed to vary with damselfish treatment, a and h were held at a constant ratio and we parameterized P i as P Damsel = P Control dp. We again used likelihood ratio tests to test our three hypotheses. To test the attack rate (or 49

50 handling time or apparent competition) hypothesis we compared the model parameterized with da 1, dh 1, and dp 1 with the reduced models parameterized with da = 1 (or dh = 1 or dp = 1). The dp = 1 model restricts changes in to be inversely proportional to changes in dh (i.e., da dh = 1); therefore, it is nested within the full model where both da and dh differ from 1. To test the hypothesis that mortality of wrasses in Experiments 1 and 2 was inverse density dependent we used a likelihood ratio test to compare each of the fully parameterized models to reduced models with handling time fixed at zero. Behavioral Analysis Reefs with fewer than three wrasses at the time behavioral observations were conducted were dropped from the analysis (this occurred on a single reef in each of the second and third temporal block). We modeled the proportion of time wrasses spent shoaling using a generalized linear model with a quasibinomial family to account for overdispersion (residual deviance = , residual df = 37) and a logit link function. We included hawkfish density as a categorical fixed factor, wrasse shoal size as a continuous covariate, and used a series of F tests (Venables and Ripley 2002) to test for an interaction between hawkfish density and shoal size and main effects of hawkfish density and shoal size. Because our habitat units were a fixed area, the amount of time a focal fish shoaled should increase with density. To generate a null model, we calculated the expected number of neighbors within a range of shoaling radii (r): 0.05, 0.15, and 0.30 meters as I = p r 2 (N- 1) / A, where A is the area of the reef and N is the number of individuals on the reef. Thus if the neighbors were distributed at random on the reef, the probability of finding > 1 fish in the shoaling radius would be 1 - Poisson(0, l) - Poisson(1, l), where Poisson(n, l) is the Poisson probability of observing 50

51 n individuals at a mean rate of l. All statistical analyses were conducted in R (R Development Core Team 2011). Results Experiment One: Effects of Predator and Conspecific Density on Prey Survival From a prey perspective, mortality of wrasses was inversely density dependent in both Experiment 1 and Experiment 2 (Experiment 1: c 2 3 = , P = 0.002; Experiment 2: c 2 3 = , P = 0.002). From a predator perspective, hawkfish foraged independently from one another. The proportional change in attack rate and handling time did not differ significantly from one (mean and 95% CI: da = 0.66 (0.10, 4.00) prey per predator day: c 2 1 = 0.164, P = 0.686; dh = 0.83 (0.27, 2.82) prey per predator day: c 2 1 = 0.161, P = 0.688, Fig. 3-2c), providing no evidence for risk enhancement or risk reduction with increasing hawkfish density. In other words, the increased wrasse mortality with two hawkfish (relative to one) was entirely explicable by an increase in predator density and not by shifts in attack rate or handling time. Parameter estimates for the full model (mean and 95% CI) were: attack rate (a) = (0.080, infinity) per predator day, handling time (h) = (0.501, 4.101) prey per predator day, and baseline daily mortality rate (m) = (0.009,0.031). The attack rate estimate has an infinite upper bound because of high depletion. Thus, it is not possible to distinguish between a wide range of attack rates, because even predators with arbitrarily large attack rates cannot kill more prey than are available. We observed low (but non-zero) levels of mortality in the absence of hawkfish ((m) = (0.009, 0.031) per day, Fig. 3-2a), which we attribute to predation by transient predators that are known to increase mortality rates of fish (e.g., Carangids and Lethrinids) (Schmitt et al. 2009) and that we occasionally witnessed near experimental reefs (Stier and Geange pers. obs.). 51

52 Shoaling behavior of wrasses was not affected by an interaction between predator density and shoal size (F = 0.027, P = 0.870) or the main effect of predator density (F = 0.050, p = 0.825); however, the incidence of shoaling significantly increased with increasing conspecific density F = , P < 0.001, Fig. 3-3). The observed proportion of time focal individuals spent shoaling greatly exceeded the predictions of the null model for shoaling radii of 0.05 and 0.15 m, but were similar to null expectations for the large 0.3 m (Fig. 3-3). Experiment Two: Effect of a Heterospecific Competitor on T. amblycephalum Survival Of the 20 P. pavo outplanted to each reef, the mean number surviving across all T. amblycephalum of wrasse density (ordinary least-squares regression: t = 0.30, P = 0.77). From a prey perspective, the per capita mortality of T. amblycephalum was inversely densitydependent in the presence of P. pavo (Fig. 3-4). From a predator perspective, the proportional change in attack rate did not significantly differ from zero (da = 0.66 (- infinity, 4.31) per predator day: c 2 1 = 0.330, p = 0.566) with the addition of damselfish. In contrast, the data do suggest increases in either handling time or effective predator density. The proportional change in per-predator handling time significantly differed from zero (dh = 0.23 (-infinity, 0.82) prey per predator day: c 2 1 = 4.888, p = 0.027), and constraining shifts in attack rate and handling time to be proportional did not significantly worsen the model (dp = 2.21 (1.38, 3.78) predators per reef: c 2 1 = 1.730, p = 0.188) implying that shifts in effective predator density or shifts in handling time are equally valid explanations for the increased number of wrasses killed in the presence of damselfish. Parameter estimates for the full model (mean (+ 95% CI, - 95% CI)) were: a 52

53 = 0.52 (0.05,infinity per predator day, h = 2.10 (2.98, infinity) prey per predator day. Estimates for the restricted model with proportional shifts in da and dh were: a =.20 (0.08, 0.48) per predator day, h = 1.40 (0.73, 2.71) prey per predator day. Discussion Despite a rich history in population ecology, the relative strengths of predator and prey density effects on functional responses remain contentious (Arditi and Ginzburg 1989, Abrams and Ginzburg 2000, Fussmann et al. 2005, Kratina et al. 2009). While a variety of predator interference mechanisms have been suggested (e.g., direct interference through group hunting, or indirect interference through common prey antipredator responses) (Abrams and Ginzburg 2000), how much predator interference occurs at the natural densities of predators and prey remains unclear. In our Experiment 1 we found evidence for resource (i.e., prey) dependence in hawkfish consumption rates, but no evidence for predator dependence over naturally occurring hawkfish densities. However, we do not conclude predator dependence is unimportant in all reef fish systems. In Experiment 1 we implemented a simplified community with only a single predator species (other species were removed), mimicking an isolated patch reef with no other resident predators; other studies have documented predator-predator interactions (Hixon and Carr 1997, Bshary et al. 2006, Stallings 2008) that may lead to predator dependence in more diverse predator communities. Traditional evaluations of density dependent mechanisms often use ANCOVA to explore how the per capita probability of prey mortality changes across a range of prey densities in the presence or absence of a specific ecological process (e.g., resource limitation or predation). This approach provides a qualitative answer of whether or not a specific process is operating. In contrast, our model-based approach allowed us to ask 53

54 more specific questions about quantitative changes in attack rates and handling times. The ANCOVA and model-based approaches are generally equally parsimonious (they make differing, but similarly strong, assumptions about the nature of the data) but the parameters estimated in the model-based approach can be more directly interpreted in terms of ecological processes subject, of course, to the usual caveats about the scope and representativeness of the experimental setting and the model used. While modelbased approaches do not mitigate the inherent risk in inferring unobserved processes from observed pattern (which ecologists and especially field ecologists must frequently do), they do provide a well-defined framework for testing among a particular set of alternative hypotheses. Similar studies have used model fitting to evaluate density and size effects on prey mortality rates (Vonesh and Bolker 2005, McCoy and Bolker 2008), suggesting the utility of this approach in extracting information from small empirical datasets. Further, the model fitting approach used here allowed us to test multiple hypotheses regarding the role predator-predator dynamics and competitors play in altering the predator-prey functional responses and/or effective predator density. Prey experienced benefits of intraspecific group living in each of the three predator treatments, and mortality of prey increased with predator density (Fig. 3-2a). The lack of a significant shift in log attack rate or log handling time in Experiment 1 provides no evidence for interference or cooperation among predators, although it is possible that there were interactions between hawkfish and transient predators that were similar in the one and two hawkfish treatments. Predator-prey interactions can change predator foraging efficacy and result in inverse density dependent mortality for prey in three non-mutually exclusive ways: (1) 54

55 the confusion effect, whereby organisms living in herds, flocks, or schools make it difficult for predators to single out individual prey; (2) group vigilance or aggression, whereby alarm calls and predator mobbing reduce predation rates; and (3) predator dilution, whereby predator satiation occurs at high prey densities (Courchamp et al. 1999). Although our behavioral observations focused on wrasse shoaling behavior, casual observations revealed no evidence for either predator mobbing or shifts in prey behavior with changes in the size of wrasse groups (though mobbing has been observed in other reef fishes: Sweatman 1984). Hawkfish consume their prey whole (as do most predatory reef fishes), and therefore spend a limited amount of time masticating prey. We speculate that other processes change, likely gut passage time or the amount of time spent pursuing prey. For example, the confusion effect or group vigilance may have produced inverse density dependence in this study, although more detailed observations of each of these behaviors is necessary to identify a concrete mechanism. The presence of damselfish decreased the survival benefits of intraspecific group formation for wrasses (Fig. 3-4). A competitor-driven shift in the strength of densitydependent mortality matches the findings of other researchers who have shown increased competition at high densities (Carr et al. 2002, Samhouri et al. 2009). Alternatively, it has also been recorded that large adult damselfishes can directly reduce the recruitment of newly settled fishes through aggressive interactions (Sweatman 1985). However, Experiment 2 supported two competing models with different mechanisms. The model with differences in handling time but not attack rate suggests that predators may be decreasing their handling times in response to the introduction of 55

56 an alternative prey. One possible explanation for this decrease in handling time is that exploitative competition by damselfish for refuge space either reduces the vigilance of wrasses or decreases pursuit time by predators (by excluding inferior competitors from refuges). The data also supported a second model where effective predator density varied (i.e., dp 1), suggesting that effective predator densities may be increasing with the addition of damselfish, leading to increased attack rates and handling times (i.e. short-term apparent competition: Holt and Kotler 1987). No ecological studies have tested apparent competition in reef fish, but aggregative response is known to occur in some predators (Anderson 2001a), and short-term apparent competition is widespread in other systems (Holt and Lawton 1994). The parameter estimates for the apparent competition models fall within the confidence limits of the parameter estimates from the handling time model (Figure 3-2); therefore, the data do not discriminate between these two mechanisms. Furthermore, while predator density counts throughout the duration of the study suggest that predator density did not differ between treatments (Table 3-1), apparent competition may still be a plausible explanation because it is difficult to observe aggregative responses of transient predators due to their skittish behavior in the presence of divers. The results of our study suggest that predator and competitor density can affect the benefits that intraspecific group formation confers on prey. We develop and implement a model fitting approach that can be applied to small empirical datasets to distinguish among multiple competing hypotheses about shifts in predator foraging behavior. Furthermore, our findings add to a growing body of literature suggesting that predator-predator interactions (Hixon and Carr 1997, Johnson 2006), predator density 56

57 (Schmitt and Holbrook 2007, White 2007), and competition (Carr et al. 2002, Samhouri et al. 2009) are important in structuring the direction and magnitude of density dependent mortality in benthic marine fishes (White et al. 2010). 57

58 Table 3-1. Description of predator densities across the natural array. Predator community on experimental patch reefs for Experiment Two. The mean (SE) number of individuals for each species is given for each of our wrasse density treatments (3-15 individuals per reef) in the presence (top) and absence of damselfish (bottom). Averages and standard errors are generated from our surveys conducted incrementally on three days during the experiment (May 30, June 1, and June 4, 2007). Total predator densities in bold represent the mean (SE) of the sum of all predators in each treatment. Damselfish Present Family Species Wrasse Density Treatment Aulostomidae Aulostomus chinensis (0.3) Balistidae Balistapus undulatus (0.3) - - (0.5) - - (0.3) Bothidae Bothus pantherinus (0.3) - - (0.3) Carangidae, Caranx melampygus (0.3) - Holocentrida e Neoniphon sammara (0.7) (0.3) 0.7 (0.3) (1.0) 0.7 Labridae Cheilinus trilobatus (0.7) - Halichoeres Labridae trimaculatus (0.6) - (1.3) (0.3) - - (0.7) Muraenidae Gymnothorax javanicus (0.7) - - Pinguipedida Parapercis e millepunctata (0.3) (0.0) Synodontidae Saurida gracilis (0.3) (0.3) (0.3) Total (0.3) (0.3) (1.3) (0.7) (0.6) 1.7 (1.2) (0.6) 0.3 (0.3) 3.7 (0.7) 1.0 (1.0) 5.0 (1.5) 58

59 Table 3-1. Continued. Control Family Wrasse Density Treatment Species Aulostomidae Aulostomus chinensis (0.3) (0.3) Balistidae Balistapus undulatus - (0.3) - (0.3) - - (0.0) - Bothidae Bothus pantherinus Carangidae, Caranx melampygus Holocentridae Neoniphon sammara - (0.3) (0.3) (0.0) (0.3) (0.6) - (0.3) 0.7 Labridae Cheilinus trilobatus - - (0.3) Labridae Halichoeres trimaculatus 1.3 (0.7) (0.6) 0.7 (0.7) 2.0 (0.0) (0.7) 0.3 (0.3) Muraenidae Gymnothorax javanicus 0.3 (0.3) (0.3) - Pinguipedida e Parapercis millepunctata (0.3) (0.3) Synodontidae Saurida gracilis - - (0.3) (0.3) Total (0.3) (0.7) (0.7) (0.7) (0.3) (0.6) (0.9) (0.6) 59

60 Table 3-2. Functional Responses with Predator Density. Functional response models allowing attack rate (a) and handling time (h) to vary with each of the experimental treatments for one and two predators. Our models (far right) are specific versions of existing functional response models (far left) that explore the role of predator density in affecting predator foraging behavior. Moving from left to right shows how the original formulation of these models, can be simplified to be identical to our model under scenarios where there are 1 and 2 predators. Our models are not equivalent to existing models for P > 2. See Skalski and Gilliam (2001) for more details. Model Holling Type II Beddington- DeAngelis Original Formulation f H 2 (N, P) = anp 1+ bn an f BD (N, P) = 1+ bn + c(p -1) 1 Predator f i (N, P 1 ) an 1+ bn an 1+ bn 2 Predator f i (N, P 2 ) 2aN 1+ bn an 1+ bn + c 2 Predator (simplification) 2aN 1+ bn an /1+ c 1+ bn / (1+ c) Our Model anp 1+ ahn (a 1 da)np 1+(a 1 da)hn Crowley- Martin Hassell-Varley an f CM (N, P) = 1+ bn + c(p -1)+ bcn(p -1) f HV (N, P) = an bn + P m an 1+ bn an 1+ bn an 1+ bn + c + bcn an bn + P m a 1+ c N 1+ Nb an 2 m bn 2 +1 m an(p 1 dp) 1+ anh an(p 1 dp) 1+ anh 60

61 Figure 3-1. The construction of Experiment 1 reef arrays. Individual reefs were constructed of 13 cinder blocks (0.5m X 0.25m X 0.25m), placed on a 1m X 1m sheet of corrugated iron. Blocks were stacked with openings facing outward in three tiers forming the shape of a pyramid. Six blocks (3 x 2) formed the base level, four blocks (2 x 2) the second tier, and three blocks (3 x 1) the third tier. On the top of each reef, four Pocillopora verrucosa coral heads (mean surface area m 2, SD = 0.004) attached to small cinder blocks (represented here as grey boxes) were placed in a diamond formation on top of reefs to provide shelter for reef fishes. 61

62 Figure 3-2. The effect of wrasse density on the per capita mortality of wrasses (i.e., proportion lost after 6 days),functional response of hawkfish, and the functional response parameters log(da) and log(dh). In panels A and B, solid ); one ( ); or two ( ) hawkfish. Functional response curves in panel B are adjusted for baseline mortality to isolate the hawkfish effect. Panel C represents maximum likelihood estimates for four models of increasing levels of complexity: 1) null model: no flexibility in attack rate or handling time, 2) : handling time varies but attack rate is fixed, da: attack rate varies but handling time is fixed, and 4) Full model: both attack rate and handling time are allowed to vary. Three grey lines (log dh, da, and dp changes) represent the plane over which each estimate of log(da), and log(dh) change with each of the four models. Note that in order for a point to fall on the grey dp changes line it requires a specific combination of proportional shift in log(da) and log(dh). For the Full model, the 95% CI is plotted for the maximum likelihood combination of log(da) and log(dh), where overlap of this region with the any of the grey lines represents an overlap of the parameter estimate and zero. See results section for significance tests of each of the parameters in the four reported models and Appendix A for details on hypothesis tests using likelihood ratio tests. 62

63 Figure 3-3. Effect of wrasse density and hawkfish treatment (zero ( ); one ( ); and two ( ) hawkfish) on shoaling behavior of wrasses. Gray dashed lines represent null model predicting increases in the probability of shoaling with increases in density for shoaling radii of 0.05, 0.15, and 0.30 m, and solid line represents mean fit from logistic regression for observed increases in shoaling with increases in density. Shaded region around solid line represents 95% CI of model fit. 63

64 Figure 3-4. The effect of a competitor, damselfish, and wrasse density on A) the per capita mortality of wrasses and B) predator functional response. Solid lines and surrounding gray area represent mean + 95% CI for each damselfish treatment estimated from maximum likelihood fit. Per capita mortality rates and functional response curves are adjusted for baseline mortality rate. Panel C represents maximum likelihood estimates for log(da) and log(dh) (points) and 95% CI for full model. Note two models, log(dh) and log(dp), fall within the 95% confidence intervals of the full model. The attack rate estimate has an infinite upper bound because of high depletion. Thus, it is not possible to distinguish between a wide range of attack rates because even predators with arbitrarily large attack rates cannot kill more prey than are available (Figure 3-2 for further details on plot interpretation and Appendix A for further details on hypothesis testing for each model). 64