|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 LOGGING AND HUNTING ALTER PATTERNS OF SEED DISPERSAL AND SEEDLING RECRUITMENT IN AN AFROTROPICAL FOREST By JOHN RANDOLPH POULSEN A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2009
2 2009 John Randolph Poulsen
3 To my wife, Connie Clark, who has explored tropical forests with me, and to my infant son, that the forests will survive so that he will have the opportuni ty to experien ce their splendor
4 ACKNOWLEDGMENTS I thank the government of the Republic of C ongo (particularly the Ministry of Forestry Economy and the Ministry of Scientific Research ), the Wildlife Conservation Society (WCS), and Congolaise Industrielle des Bois (CIB) for their colla boration and support. In particular, I thank B. Curran, J.-C. Dengui, O. Desmet, P. El kan, P. Kama J.M. Mevellec J. Mokoko, P. Ngouemb, D. Paget, P. Telfer, H. Thomas, and L. Vander Walt for making this work possible. Special thanks to B. Curran who first got me to Congo and P. Elkan who later invited me to conduct my research in northern Con go and got the whole thing rolling. Northern Republic of Congo is a bit off the b eaten path and lacks a few of the amenities that make research feasible. My research was conducted in a pl ace and at a scale that would not have been possible without the logi stical support of the Buffer Zone Project (Projet de la gestion des cosystmes pripheriques au parc nati onal de Nouabal-Ndoki, PROGEPP) and the Nouabal-Ndoki National Park project, both of wh ich were managed by the Ministry of Forestry Economy and the Wildlife Conservation Society. Managing logistics and research teams was sometimes a monumental effort and I benefitted from the support and help of R. Aleba, C. Assobam, J. Beck, S. Elkan, M. Gately, A. Niam azock, C. Prevost, and E. Stokes. At one time or another, nearly all the WCS-PROGEPP and WCS-NNNP em ployees (guides, drivers, technicians, staff) played a role in making my research project a success. They all deserve my heart-felt appreciation. I owe a debt of gratitude for the tireless wo rk of an amazingly motivated and effective field team. Special thanks to O. Mbani and Y. Nganga who led the field teams. Thanks to the other capable research technicians, G. Abeya, E. Elenga, I. Loungoumba, M. Mokoke, and U. Sabo, and guides, Ekoume, M. Simba, Mbe, Iyena, P. Ipete, J. Lamba, R. Bokoba. Thank you
5 also to guide teams from Bomassa village that a ccompanied us into the park. All of these people spent countless hours in the forest regardless of sun, rain, sweat bees, or angry elephants. The identification of tree species was comp leted by D. Harris and A. Wortley from the Edinburgh Royal Botanical Gardens and J. M. Moutsembot from the National Herbarium in Congo. V. Medjibe mapped and tagged over 11,000 trees in the plots an d deserves recognition for his diligent and high quality work. I feel extremely fortunate to have done my dissertation in the Department of Zoology/Biology at the University of Florida. The department po ssesses the perfect combination of excellence, commitment, and collaboration. It is hard to imagine that I will find a place where I will work as hard with such friendly peopl e. Thank you to Dr. Colin Chapman who first accepted me as a doctorate student at the University of Florida. Colin helped me refine my ideas and field techniques and encouraged me to pursue my research inte rests. I benefitted greatly from his mentoring and look to him as a role model as someone who balances research with conservation and mentoring. Many, many thanks to my second (and final) ad visor, Dr. Ben Bolker. If the Department of Biology was the perfect place for me, Ben has been the perfect advisor. I would like to thank Ben for his generosity in time, friendship, and in tellectual mentoring. Be n greatly elevated the quality of my work and taught me the analytical t ools that I will need for a career in research. Our work together is not done, and I look forward to learning more from him in the future. I would also like to sincerely acknowledge my committee members, Drs. Doug Levey, Kaoru Kitajima, Todd Palmer, and Wendell Cropper for challenging me to be a better scientist. They all contributed to the ideas and quality of this work. Thank you also to Drs. Craig
6 Osenberg, Emilio Bruna, and Scott Robinson for interesting and useful discussions and comments on my research. The SOB (St. Mary Osenberg Bolker) la b provided me with great feedback and discussion after my return from the field. I hope they appreciated learni ng about trees and seeds as much as I enjoyed learning from them about reef fish ecology and behavior. The Levey and Robinson labs also provided me great feedback during their weekly meetings and taught me a thing or two about birds. Financial support was generously provided by a University of Florida Presidential Fellowship, EPA STAR fellowship, ONeill Disserta tion Fellowship. Field work was funded by two US Fish and Wildlife Service grants to C. Clark and me, the UF Niddrie Travel Award, WCS, and several donors who generously support WCS research and conservation in northern Congo, including the International Tropical Timb er Organization and the Liz Claiborne Art Ortenberg Foundation. I owe this dissertation and my career in ecology to my wife Connie Clark. By refusing to return home after our Peace Corps assignments (mine in Mali, hers in Cameroon), she made me go looking for her in the tropical forest. Af ter a full day in the forest I had changed my career aspirations, giving up what probably would have been more lucrative pursuits for a chance to figure out how tropical forests work and what we can do to conserve them. Since then Connie has been both my best friend and closest colleag ue. Without her support, observational skills, creative ideas and critical pen, th is dissertation would not have b een possible. Without her smile, laugh, sense of adventure and charm, life would be a whole lot less fun. Thanks also to my family and friends for understanding all the holidays, birthdays and social events held late or w ithout us as our work took us away for long field seasons over the
7 years. Special thanks to my parents who have mana ged to visit me in nearly all the places I have lived and worked. Colette St. Mary, Todd Palmer, Rico Holdo, and Dan and Hilary Zarin deserve thanks for being such great friends, especially as our lives became increasingly complicated toward the end of this dissertation.
8 TABLE OF CONTENTS page ACKNOWLEDGMENTS ............................................................................................................... 4LIST OF TABLES ................................................................................................................ .........10LIST OF FIGURES ............................................................................................................... ........11ABSTRACT ...................................................................................................................... .............12 CHAPTER 1 DECOUPLING THE EFFECTS OF LOGGING AND HUNTING ON A AFROTROPICAL ANIMAL COMMUNITY .......................................................................14Introduction .................................................................................................................. ...........15Methods ....................................................................................................................... ...........17Study Area .................................................................................................................... ...17Animal Surveys ...............................................................................................................1 7Environmental Variables .................................................................................................19Data Analysis ................................................................................................................. ..20Results ....................................................................................................................... ..............27Densities of Animal Species ............................................................................................27Relating Species and Guilds to Environmental Variables ...............................................28Decoupling the Effects of Disturbance, Geographic Position, Fo rest Structure and Fruit Abundance on Guild Densities ............................................................................29Discussion .................................................................................................................... ...........312 SEED DISPERSAL PATTERNS DRIV E SEEDLING RECRUITMENT IN AN EXPERIMENTAL MANIPULATI ON OF SEED SHADOWS ............................................45Introduction .................................................................................................................. ...........45Methods ....................................................................................................................... ...........48Study Area .................................................................................................................... ...48Tree and Seed Census Data .............................................................................................49Seed Shadow Models and Parameter Estimation ............................................................50Plant Species Trait Data ..................................................................................................51Results ....................................................................................................................... ..............52Effect of Dispersal Mode .................................................................................................52Effect of Disturbance .......................................................................................................53Discussion .................................................................................................................... ...........53Future of Tropical Forests ...............................................................................................57
9 3 SEED DISPERSAL PATTERNS DRIV E SEEDLING RECRUITMENT IN AN EXPERIMENTAL MANIPULATI ON OF SEED SHADOWS ............................................71Introduction .................................................................................................................. ...........72Methods ....................................................................................................................... ...........75Overview ...................................................................................................................... ...75Study Site and Species .....................................................................................................75Quantification of Seed Shadow .......................................................................................76Experimental Seed Distributions .....................................................................................77Analysis ...................................................................................................................... .....79Results ....................................................................................................................... ..............82Discussion .................................................................................................................... ...........84LIST OF REFERENCES ............................................................................................................ ...94BIOGRAPHICAL SKETCH .......................................................................................................104
10 LIST OF TABLES Table page 1-1 Number of observations, density of in dividuals and 95% conf idence intervals (CI), and the coefficient of variation (CV) for animal species in three types of forest stands: unlogged, unhunted forest; logged, unhunted forest; and logged, hunted forest ...361-2 Results of multivariate analyses .........................................................................................3 82-1 Species, dispersal modes and samples. ..............................................................................592-2 Mean fitted values and their standa rd deviations of model parameters. ............................602-3 Fitted parameters for full (F; all plots), hunted (H+), logged (L+), unhunted (H-) and unlogged (L-) models .........................................................................................................6 13-1 Model comparison for survival analysis of seedlings ........................................................873-2 Parameters [and 95% credible interval s] from generalized linear mixed models (GLMMs) for seed germination, leaf gr owth on seedlings, and leaf damage ...................88
11 LIST OF FIGURES Figure page 1-1 Map of study area in nor thern Republic of Congo. ...........................................................391-2 Densities of animal guilds averaged over all transect during the 24 months from June 2005 to May 2007 .............................................................................................................. 401-3 Ordination based on principal components analysis (PCA) of thirty 1 ha plots showing the first, second, and third axes.. .........................................................................411-4 Display of (A) environmental variables, (B) species traits, and (C) plots along the first two RLQ axes. Each plot is repres ented by a symbol representing the forest type .......................................................................................................................... ...........421-5 Dotplot of predictor va lues for animal guilds. ...................................................................431-6 The effect of hunting, logging, and hunting and logging on guild density. .......................442-1 Location of the tree plots in the Kabo concession and Nouabal-Ndoki National Park in northern Republic of Congo. .........................................................................................672-2 Seed dispersal for Terminalia superba a wind-dispersed tree species, across the 30 tree plots.. .................................................................................................................. .........682-3 Mean dispersal distance (open circle s) for each species within forest type combination grouped by dispersal vector (animal-, bird-, and wind-dispersed species). ..................................................................................................................... .........692-4 Mean degree of dispersion (open circle s) for each species within forest type combination grouped by dispersal vector (animal-, bird-, and wind-dispersed species). ..................................................................................................................... .........703-1 Depiction of the experimental design. ...............................................................................893-2 Fit of the negative exponen tial function to seed trap data from four individuals of Manilkara mabokeensis ....................................................................................................903-3 The density of seedlings surviving with distance from tree and in the different experimental and non-ex perimental wedges ......................................................................913-4 Violin plot of parameters from the seedling survival model .............................................923-5 Number of seedlings that survived ov er 18 months in the three experimental distributions.................................................................................................................. ......93
12 Abstract of Dissertation Pres ented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy LOGGING AND HUNTING ALTER PATTERNS OF SEED DISPERSAL AND SEEDLING RECRUITMENT IN AN AFROTROPICAL FOREST By John Randolph Poulsen August 2009 Chair: Benjamin M. Bolker Major: Zoology Unprecedented rates of logging and hunting in Ce ntral Africa threaten to transform vast tracts of primary forest into a mosaic of degraded forest emptied of its animals. Reductions in abundances of seed-dispersing animals are hypothe sized to alter patterns of seed deposition for many tree species, potentially limiting forest regene ration with long-term consequences for forest structure and composition. In northern Republic of Congo, I set up a la rge scale observational experiment to decouple the effects of l ogging and hunting on animal abundance and seed dispersal patterns by positioning 30 permanent transect s, each linked to a 1 ha tree plot, in forest disturbed by logging and hunting, logging alone, and neither logging nor hunting. I found that the effects of hunting and logging on densities of animal guilds outweighed the effects of local forest structure and fruit resources. Logging and hunting sometimes interacted to reduce guild densities (ape, duiker, monkey, and pig) by as mu ch as 71% and shifted the composition of the animal community away from large mammals toward s squirrels and birds. With seed trap data collected from the tree plots, I used inverse m odeling to quantify seed dispersal patterns of 26 tree species. Hunting reduced mean dispersal dist ances of animal-dispersed species by at least 20%, while hunting and logging both increased mean dispersal dist ances and dispersion of winddispersed species. To determine if changes in seed dispersal patterns reduce seedling recruitment
13 and survival, I experimentally manipulated seed dispersal patterns under several individuals of a monkey-dispersed tree, creating di stributions to mimic scenarios of no dispersal, natural dispersal and good dispersal. After 18 months, good dispersal increased seedling survival by 26% over natural dispersal, whereas no di spersal reduced seedling survival by 78%. Survival of seedlings depended on the density of dispersed seeds but not the distance from the tree. This experiment demonstrated that s eed dispersal patterns do matter for seedling recruitment and survival. Management of hunting is a priority for the conservation of tropical forests because forests emptied of their seed dispersers will have limited regeneration capacity.
14 CHAPTER 1 DECOUPLING THE EFFECTS OF LOGGING AND HUNTING ON A AFROTROPICAL ANIMAL COMMUNITY Throughout tropical forests, logging and hunting are tight ly interconnected: logging opens up frontier forest and hunting quickly follows in its wake. With the expansion of logging in Central Africa, conservation of animal popul ations depends on knowledge of the individual and combined effects of logging and hunting so that management efforts can be appropriately allocated. Our goals were first to decouple the effects of selec tive logging and hunting on densities of forest animal guilds, including apes duikers, monkeys, elephant, pigs, squirrels, and large frugivorous and insectivor ous birds, and second, to compar e the relative importance of these disturbances to the effects of local scale variation in forest structure and fruit abundance. In northern Republic of Congo, we surveyed anim als along 30 permanent transects positioned in forest disturbed by logging and hunting, loggi ng alone, and neither logging nor hunting. Sampling bimonthly for two years, we observed 47,179 animals of 19 species and 8 guilds in 1154 passages (2861 km) over the transects. Dens ities varied by as much as 480% between forest types demonstrating the significant eff ects that human disturbance had on populations of some species. The landscape-level effects of hun ting and logging on densiti es of animal guilds outweighed the effects of variation in local fore st structure and fruit abundance. Logging and hunting sometimes interacted to reduce guild dens ities (e.g., ape, duiker, monkey, and pig) by as much as 71%; for other guilds, logging and hunti ng had opposite effects with positive impacts of one buffering the negative impacts of the othe r (e.g., squirrels, insectivorous and frugivorous birds). Logging and hunting shifted the relative abundance of the animal community away from large mammals towards squirrels and birds. The combination of logging and hunting had the strongest negative impact on tropical species, bu t both logging and hunting alone also negatively affected some species. Therefore the strategy of conserving biodiversit y by managing hunting in
15 degraded forest will not stop reductions in anim al populations by itself. To balance the loss of forest species and the ecological services they provide with th e need for economic development and wild meat in tropical countri es, we suggest that land use pl anning combine large tracts of pristine forest with resource use areas for either logging or hunting. Introduction Habitat loss from conversion of forest for ag riculture, ranching and intensive logging has led to massive extinctions of tropical animals in some regions (Brook et al. 2003, Curran et al. 2004). Models predict that tropica l animals will lose large proporti ons of their curre nt habitat in the coming decades (Laurance 2001, Soares et al. 2006). In short, the extent of remaining old growth forest is likely insufficient to prevent fu rther large-scale loss of species. The future of many animal species might thus depend on their ab ility to persist in secondary and degraded forests. The conservation value of degraded forests has been at the center of the debate over the future of tropical species (L aurance 2007). Wright and Mu ller-Landau (Wright and MullerLandau 2006a, b) state that secondar y forests could rescue species un likely to persist in remnants of primary forest, whereas Brook et al. (2006) argue that this is unlikely because tropical secondary forests support lower biodiversity, a predominance of generalis t species, and act as reproductive sinks. To resolve this debate, we ne ed a better understandin g of the factors that limit the ability of animal speci es and communities to persist in the face of disturbance. Alarmingly, we seem to lack the field data nece ssary to translate the consequences of forest degradation into conservation and manage ment policy (Gardner et al. 2007). Logging is one of the primary drivers of tropica l forest degradation. Most extant tropical forests have been logged or will be in the near future, with only relatively small fragments preserved (Whitmore 1997). Logging has been a si gnificant contributor to the economies of the
16 Congo Basin since the post-colonial period; and at the turn of the century industrial logging expanded into the most remote forests of Gabon, Republic of Congo, and Democratic Republic of Congo (Laporte et al. 2007) Logging concessions now o ccupy 30-45% of all forests, reaching 70% of forests in some countries (Glo bal Forest Watch 2002, Laporte et al. 2007). Predicting the effect of logging on tropical fore st animals is complicated because logging impacts the forest at multiple scales and its e ffects are usually confounde d with other forms of disturbance. Commercial logging modifies landscape-scale fore st structure, local environmental variables and resource abundance (revi ewed in Putz et al. 2001). Fo r example, at the level of the concession, logging can change the si ze, distribution and connectivity of habitat patches. At the level of the forest stand, logging alters canopy openness and light regimes. Changes in canopy openness can release fruit crop trees from competition and increase understory growth, modifying the composition and trophic structure of the stand. The environmental effects of logging alone are diverse, but logging rarely acts by itself. The ne t effect of logging on populations and communities of forest animals is further complicated to the extent that road construction, population growth and settlement, and hunting accompany and interact with timber extraction (Laurance et al. 2006, Blake et al. 2007, Laurance et al. 2008, Clark et al. 2009, Poulsen et al. 2009b). At best, unsustainable levels of hunting in tropical forest reduce populations of large-bodied animals; at worst they lead to local extinctions of species (MilnerGulland et al. 2003). Our goal was to decouple the effects of selective logging and hunting on densities of tropical forest animal guilds. To do this, we established a large scale observational experiment in lowland tropical forest in the Republic of Congo, positioning 30 transects in forest disturbed by logging, logging alone, and neither logging nor hun ting. We then surveyed a broad range of
17 animal species over two years to quantify densitie s of animal species and guilds. Specifically, we sought to (1) quantify the e ffects of logging, hunting, and their combined effects on species and guild density, and (2) compare the rela tive importance of logging and hunting on guild densities to other variab les like spatial heterogeneity, local scale environmental variables like forest structure and light avai lability, and fruit abundance. Methods Study Area We conducted the study in the Nouabal-N doki National Park (NNNP; 400,000 ha) and the Kabo logging concession (267,000 ha) in nort hern Republic of Congo (Clark et al. 2009, Poulsen et al. 2009b). The Kabo concession border s the NNNP to the south, and together they include a mosaic of logged and unlogged forest. Between 20-25 years before the study, the area was selectively logged at low intensity (<2.5 trees ha-1) with four species, Entandophragma cylindricum E. utile Triplochiton scleroxylon and Milicia excelsa comprising 90% of the cutting volume (Congolaise Industrielle des Bois 2006). Approximately 3000 people inhabit the study site, most of whom live in the logging town of Kabo. Reside nts hunt with shotguns, and to a lesser extent with wire snares to supplement their diets and for local trade (Poulsen et al. 2009b). Most hunting originates from the town of Kabo, resulting in a gradient of hunting intensity that deceases with dist ance from it with some variati on with vegetation type (Mockrin 2008). The forests are classified as lowland trop ical forest, and dominant tree families include Meliaceae, Euphorbiaceae, and Annon aceae. Rainfall averages about 1700 mm annually and is seasonal with peaks in May and October. Animal Surveys We established 30 2.5-km transects ove r an area of approximately 3000 km2. We located ten transects and tree plots (see below) in each of three forest types: unlogged and unhunted
18 forest, logged and unhunted forest, and logged and hunted forest. We minimized habitat variation among forest transects by positioning tran sects only in mixed lowland forest, with a buffer of 500 m to the nearest primary road and 100 m to the nearest water source. Using ArcView 3.2 and a 14 class habitat map, we extracted the areas that did not meet those criteria, and then randomly positioned transects on the re maining surface. Transects were oriented perpendicular to water drainage, and were sepa rated by a minimum of 2.5 km. We cut narrow trails along each transect, and marked the trails every 20 meters with flagging tape. Following trail opening, we left all transects undisturbed for one month before sampling for animals. We designed this study as a natural experiment to try to decouple the effects of hunting and logging. The strength of this design is that we can quantify the effects of hunting and logging while controlling for environmental varia tion, something that cannot be done with large scale one-off transect designs (e .g., Clark et al. 2009). The weakness of this design is that it is pseudo-replicated in the sense that we quantify th e relative densities of animal species and guilds at a single site (a 3000 km2 area in northern Congo). This site has particular characteristics, such as its spatial pattern of hunti ng and logging around the village of Kabo, which meant that hunted transects and logged transects were geographically grouped t ogether and which may not be representative of all other sites in the Congo Basin. Surveying forest animals, and in particular, assessing the effects of hunting and logging is complicated by the difficulty of obs erving animals in the forest. To obtain sufficient numbers of observations to estimate species densities, most surveys of tropical animals in the Congo Basin rely on indirect observations of nests for apes and dung for elephant and antelopes. Using indirect observations raises two problems: first, not all animals leave indices that can be counted limiting the number of species th at can be surveyed; and, second, estimates of animal abundance
19 from indirect observations introdu ce a great deal of e rror by assuming constant rates of nest and dung production and decay (Walsh and White 2005). To bypass these difficulties, we surveyed permanent transects that were kept open so th at we could make observations of live animals. Three field teams, each composed of a rese archer and a local guide, surveyed the 30 transects bimonthly over two years for diurnal mammals and a suite of large bird species (Table 1-1). Surveys began between 06:00 and 07:00 when animals were the most active. Observers walked slowly (ca. 1.2 km hr-1), scanning the forest floor a nd canopy for direct and indirect observations of animals. For each observation, we recorded the distance along the transect and estimated the distance from the tr ansect to the individual animal or center of an animal group following standard distance sampling protocol (B uckland et al. 2001). To survey birds, we conducted point counts at 200 m intervals along the transects. Upon arriva l at the point count station, observers waited for two minutes, and then recorded visual observations of birds and the distance from the bird to the observer for tw o minutes. Although we surveyed several bird species, we only present data a nd results for large frugivorous a nd insectivorous birds for which we are confident that our survey methods were r obust. In addition to observations of animals, we also counted signs of hunting encountered al ong transects, such as shotgun shells, wire snares, and camp fires, to asse ss the level of hunting pressure. Environmental Variables We measured canopy openness and light availa bility along transects with hemispherical canopy photographs. Photographs were taken ev ery 200 m along a transect, 30 cm above the ground, in uniformly overcast conditions in the ea rly morning or late afternoon with a leveled Nikon Coolpix P5000 camera body and Nikon FC-E8 Fi sheye converter lens. We analyzed images for the percentage of transmitted diffuse light and canopy openness using Gap Light Analyzer 2.0 (Frazer et al. 2001).
20 We established 30 1-ha tree plot s to examine whether forest structure influences animal abundance and distribution. The plots were posit ioned 50 m to the right or left (randomly selected) of the midpoint (1250 m) of each tran sect. Within each plot, we tagged, measured, mapped and identified to species all trees greater than 10 cm diameter-at-breast-height (dbh) (Harris and Wortley 2008). For each tree, we also recorded the canopy status of the tree (understory, midstory, canopy, and emergent) and the presence/absence of liana s in its crown. In addition, we quantified canopy openness and light availability for each plot, averaging the values of four pictures taken from each quadrant of the plots. Canopy heterogeneity was calculated for each plot and each transect as the standard devi ation of canopy openness for the plot or transect. We quantified fruit availability in each of the 30 plots by collecting fruits and seeds in 21 1m2 traps. Seeds traps were cen tered along three lines at 25, 50, and 75 m from the plot border, with 10 m separating each trap. At the same tim e that transects were wa lked, we collected all seeds and fruits that fell into the traps and counted and identified them to species. When unknown fruits and seeds were collected, we sear ched the canopy over the seed trap for fruit, using the tagged trees to identify the propagule. We quantified fruit a bundance as the monthly density of fruits per m2. Data Analysis To quantify variation in hab itat structure among sites, we summarized the plot data (presence/absence of lianas, numbe r of trees, number of tree speci es, number of secondary forest trees, mean and standard deviation of diamete r-at-breast height, number of understory, midstory, canopy, and emergent trees, and % diffuse li ght) into fewer variables using Principal Components Analysis (PCA). Three axes capture d 95% of the variance and explained the main forest structure gradients (Figure 3). The first ax is (51% of the variance) distinguished a gradient in the number of trees in a plot, with the number of unders tory, emergent, and canopy trees
21 contributing to the number of trees. The second axis (31% of the variance) identified a forest structure gradient in the number of lianas in the plot, with und erstory trees supporting few lianas relative to canopy trees. Finally, the third axis (1 0% of the variation) se parated understory trees from canopy trees. We refer to these variables in subsequent analyses as Number of trees, Number of lianas, and Understory. We estimated animal densities and calculated their associated coefficients of variation and 95% CIs with Distance 5.0 software (Thomas et al. 2006). To ensure robust estimation of detection and an effective strip half-width, the di stance over which the proba bility of detection of an observation is estimated, we calculated densitie s for species for which we gathered at least 40 observations per site and we trunc ated observations farthest from each transect (Buckland et al. 2001). We fitted detection func tions to the data sequentially with half-normal, uniform, and hazard-rate key functions that contained cosi ne, Hermite polynomial, and simple polynomial adjustment terms (Buckland et al. 2001). The best model was selected on the basis of the lowest Akaike information criterion score (AIC). We examined model fit with chi-square goodness-offit tests. Animal species traits were directly linked to environmental variables with a three-table ordination method known as RLQ an alysis (Doledec et al. 1996, Ri bera et al. 2001, Dray et al. 2002). RLQ analysis is an extension of coinertia analysis that rela tes a plot-by-variable table (R) to a species-by-traits table (Q), with a plot-byspecies table (L) serving as a link between R and Q. With RLQ analysis both species traits and environmental conditions affected by disturbance, as well as their interrelationships can be assessed (Cleary et al 2007). To visually assess the relationship between animal species traits and environmental characte ristics of the plots, we used a three-table ordination method kno wn as RLQ analysis (Doledec et al. 1996, Ribera et al. 2001,
22 Dray et al. 2002). RLQ analysis is an extension of coinertia analysis that relates a plot-byvariable table (R) to a species-by-traits table (Q ), with a plot-by-species table (L) serving as a link between R and Q. With RLQ analysis both species traits and environmental conditions affected by disturbance, as well as their interrelationships, can be assessed (Cleary et al. 2007). For animal species traits we included body si ze (small, < 5kg; small-to-medium, < 10 kg; medium, <50 kg; large, >50 kg), feeding guild (brower, frugivore, frugivore-granivore, granivore, grazer, insectivore), and protected stat us (unprotected, partially protected, protected). For environmental traits we included type of disturbance (hunting, logg ing), fruit abundance, number of signs of hunting, mean diamete r-at-breast height of trees > 10 cm, canopy heterogeneity, number of trees, number of trees topped by lianas, and number of understory and canopy trees. Three separate ordinations of the R (envir onmental variables), L (species composition), and Q (species trait) tables were performed prior to the co-inertia analysis. First, the species abundance table containing the number of individua ls of each species occurring at each site was analyzed by correspondence analysis (CA), an ei genanalysis approach that provides a joint scaling of sites and species scores. Only species with five observations or more were included in the analyses due to the sensitivity of corresponde nce analysis to rare species (Lesica and Cooper 1999). We also conducted the CA on the log10 transformed abundances of species, but it did not alter the overall results. The sites a nd species scores were us ed to link the R and Q tables, as sites are shared by the R and L tables and species are shared by the Q and L tables (Ribera et al. 2001, Dray et al. 2002, Hausner et al. 2003, Choler 2005). Ne xt, the relationship between sites and environmental at tributes (i.e., R table) was an alyzed. For the quantitative set of variables, PCA was applied using weights obtained with the corres pondence analysis of
23 species, thereby linking the R to the L table. The final step in this initial procedure was the analysis of the Q table of categorical species traits with row wei ghts obtained with the correspondence analysis of species using multip le correspondence analysis (MCA; Tenenhaus and Young 1985). After these three steps a single inertia analysis was performed on the crossmatrix of R, L, and Q. Co-inertia analysis se lects axes that maximize covariance between the R and Q tables. As a result, the environmental attr ibutes are directly related to species traits. RLQ analysis is a symmetric analysis, in th e sense of finding scores that are optimally related to each table, without emphasizing environmental variables or species traits. Scores are a compromise between maximizing the correlation and explaining the variation in each table. We investigated the significance of the relationshi p between the environmental attributes (R) and species traits (Q) with a Monte Carlo permutatio n test, permuting the R and Q tables 1000 times and comparing the results to observed values (Dol edec et al. 1996). All analyses were carried out using the ADE4 software packag e within R (Dray et al. 2007). We assessed the importance of disturbance (hunting and logging), geographic position, environmental variables, and fruit abundance on animal density by modeling these factors in relation to observations of anim als for each of eight guilds. M odeling the number of animals in a given guild at a particular point in time (mont h) and space (site) is generally done by fitting a generalized linear mixed model (GLMM) us ing a Poisson distribution. However, a distinguishing characteristic of data from surveys is their tend ency to contain a large proportion of zeros. Examination of the frequencies of counts of observations of animal guilds demonstrated that the data incl uded more zeros than can be e xpected by either a Poisson or negative binomial distribution. A high frequency of zeros can aris e in several ways (Kuhnert et al. 2005, Martin et al. 2005a, Martin et al. 2005b). Structural zero s result from a true ecological
24 effect; for example, a strong seasonal or enviro nmental gradient will result in sites with no animals present in one season or under particul ar conditions (e.g., heavy hunting). Random zeros result from observer error and study design: the observer fails to det ect the species or the species occurs but is not presen t in the survey period. These ar e false negative errors and arise when a species is not recorded when it is in fact present. Modeling of da ta with a high frequency of zeros can lead to spurious results if these factors are not taken into account (reviewed in Martin et al. 2005b). Although multiple approaches can be used for modeling high frequencies of zeros (e.g., hurdle model, negative binomial model), when random zeros are present in the data, a zeroinflated mixture modeling approach is required (MacKenzie et al. 2002, Tyre et al. 2003). Our data certainly included random zer os because a species was sometim es not recorded on a transect at time t, even though it had been recorded at times t-1 and t+1. Therefore we employed a zeroinflated Poisson model (ZIP) comprised of a point mass at zero and a Poisson distribution (Lambert 1992). Using this approach, we mode l the probability that a zero comes from a Poisson distribution or alternativ ely as a random zero. The mean number of animals at a site is then estimated given the zeros are modeled in this way. The mixture model is specified as: r = 1, 2, , where
25 In both equations, represents the probabil ity that an observation is modeled as a Poisson distribution. Above, represents the mean expected number of individuals at site and month k and is expressed as a f unction of the explanatory variables, through a log transformation. Similarly, can be expressed as a functi on of the explanatory variables using a logit transformation. Here the parameters and are vectors representing the coefficients estimated for explanatory variables, and (fixed effects), and and are vectors for the grouping variables (random eff ects). To model animal density, we included the area sampled, as an offset. We calculated for the ith species and jth site by multiplying the distance walked along a transect by twice the species-specific effect strip width estimated by Distance 5.0. We did not include a constant intercept term in the models as its inclusion sometimes made para meter optimization difficult. We related the observations of animals of each guild to the following explanatory variables (xijk and zijk): 1) site (logged/hunted, logged/ unhunted, unlogged/unhunted); 2) geographical coordinates (longitude and latitude); 3) number of trees (PCA 1); 4) number of lianas (PCA2); 5) number of understory trees (PCA 3); 6) canopy heterogeneity (standard deviation of transect diffuse light ); and 7) monthly fruit abundan ce. All continuous explanatory variables were transformed into z -scores for analysis (Gelman and Hill 2007). So that fruit abundance would reflect the actual di et of an animal guild, we used diet lists from the literature to categorize fruit species by the types of animal s that consumed them (Gautier-Hion et al. 1985, White et al. 1993, Tutin et al. 1997, Whitney et al. 1998, Clark et al. 2001, Poulsen et al. 2001, Poulsen et al. 2002, Morgan and Sanz 2006). We included the month that the survey was conducted and species as random effects ( and ).
26 To get the estimate of mean guild abundance for each site, we held all other covari ates at their mean values and we multiplied the ZIP mixing pr obability, the probability that the number of individuals at a site ha s a Poisson distribution, by the predicte d density of individuals at a site generated from a Poisson distribution. We then estimated the overall eff ect of logging, hunting, and logging and hunting as log-ratios of guild densities by dividing the posterior mean abundance of a guild at the distur bed site from the posterior mean abundance of the site where that disturbance was not pres ent; for example, the effect size of logging would be: Bayesian inference with Markov Chain Mont e Carlo (MCMC) simulation was used to estimate posterior distributions of model paramete rs and test for significance. Statements of significance are made in the Bayesian context where a significant effect indicates that the 95% credible interval co rresponding to that effect does not incl ude zero, and an estimate that is significantly less than a proposed value has 5% pr obability of being equal to or greater than the proposed value. For the ZIP models, we used w eakly informative, normally distributed priors for fixed effects, half-Cauchy priors for random effects and gamma priors on the precisions of the variance components (Gelman and Hill 2007). The zeros trick rule allowed the specification of the truncated Poisson distributi on (Spiegelhalter et al. 2003, Kuhnert et al. 2005, Martin et al. 2005a, Martin et al. 2005b). We fit our models using the software WinBUGS v. 1.4.1 (Spiegelhalter et al. 2003). For each model, we achieved convergence after 50,000 iterations (the burn-in) and based summary st atistics on an additional 50,000 iterations. We ran three chains to monitor convergence based on variance components of multiple sequences and assessed convergence by visual inspection and with Gelman-Rubin statistics from the R contributed package, coda (Plummer et al. 2005). For point estimates, we extracted the means of
27 the posterior distributions and we derived 95% credible intervals based on the observed quantiles from the MCMC replicates. With the exception of the ZIP models and estimati on of species density, a ll statistical analyses and graphing were performed with the R Langua ge, version 2.7.1 (R Development Core Team 2008). Results Densities of Animal Species We recorded 9811 direct observations of an imal groups (47,179 animals) in 2861 km of transects (n=1154 passages along tr ansects) between June 2005 and May 2007 (Figure 2). We estimated the densities of 19 species at each of the 3 sites (Table 1-1). Several species demonstrated large differences in estimated de nsities among sites. For example, chimpanzee ( Pan troglodytes ) density was higher in the unlogged, unhunted forest than the logged, hunted forest. All four monkey species ( Lophocebus albigena Cercopithecus cephus C. nictitans and C. pogonias ), ape species ( Gorilla gorilla and P. troglodytes ), blue duiker ( Cephalophus monticola ), and medium duikers ( Cephalophus spp.) tended to have th e highest densities in unlogged, unhunted forest and lowest densities in logged, hunted fore st. By contrast, elephant ( Loxodonta africana ) density was highest in logged, hunted forest and logged, unhunted forest. Large frugivorous birds ( Bycanistes albotibialis B. fistulator Ceratogymna atrata Corythaeola cristata and Psittacus erithacus ) tended to have the highest de nsities in logged, hunted forest, whereas insectivorous birds ( Tockus albocristatus T. cameras and T. fasciatus ) had the highest densities in either the logged, unhunted or logge d, hunted forest. Densities of both squirrel species ( Funisciurus lemniscatus and Protoxerus stangeri ) were highest in logged, hunted forest.
28 Relating Species and Guilds to Environmental Variables The RLQ analysis revealed a significant (permutation test : P<0.001) association between environmental variables and species traits. We consider the first two RLQ axes, which together explained 87.7% of variance in the analysis. B ecause the RLQ analysis represents the partial ordination of the environmental ch aracteristics, the species abundan ces, and the species traits, we compared the proportion of variance attributed to each matrix to that resulting from their separate analyses. The first axis of the RLQ analysis accounted for 82.7% of the variance and for 53.5% of the variance in environmental variables and 56 .0% of the variance in species traits (Table 12). The second axis of the RLQ analysis accounted for only 5.0% of the variance and for 56.0% of the variance in environmental variables and 79.2% of the varian ce in species traits. Overall, these results showed that the structure of tabl es R and Q were related to the species abundance gradients in L. Along axis 1, hunted, logged forest with a low abundance of understory trees contained species with a different set of traits than logged, hunted and unlogged, unhunted forest (Figure 4). In addition to having a greater numb er of hunting sign, hunt ed, logged forest was characterized by having a greater nu mber of secondary tree species. These forest sites tended to be occupied by small insectivorous and granivor ous species; whereas unhunted forest tended to contain larger species and more grazers and browsers. Axis 2 explained very little of the variance, but separated plots and species by forest structure. Plots w ith high mean DBH were more heterogeneous and separated from those w ith greater numbers of canopy trees and a more even canopy cover. Abundance of large browsers was higher in the more heterogeneous forest and separated from the rest of the animal community.
29 Decoupling the Effects of Disturbance, Geogr aphic Position, Forest Structure and Fruit Abundance on Guild Densities ZIP models identified the factors that most strongly determine densities of animal guilds. The absence part of the ZIP model, quantifying th e probability of a structural zero (i.e. species truly absent), varied across gu ilds (Appendix 1). Note that ef fects for the absence component presented in Appendix 1 are on the logit scale, wh ere an effect of zero corresponds to a probability of 0.5. Here we focus on the abundance part of the models and the effects of hunting, logging, spatial variation, forest structure, and fr uit abundance on the densitie s of guilds. Effects of abundance are guild densities so that effect s that showed no change would be centered on zero. Because geographic coordinates were stro ngly correlated with logging presence and hunting intensity (see below), we present the results of models configured in two different ways: a full model which includes all factors and a disturbance model which excludes latitude and longitude from analysis. In the full ZIP model, guild densities varied along a latitudinal gr adient. Monkey, pig, duiker, and ape density increased with latitude; whereas frugivorous bird density decreased with latitude. In this model, the effect of disturba nce, both logging and huntin g, did not have a strong effect on most guilds. The logged guild densi ties predicted for each site (logged and hunted, logged and unhunted, and unlogged and unhunted) often had overlapping 95% credible intervals (Figure 5). Moreover, pig, elep hant and monkey densities demons trated differences among sites, but sometimes in an unexpected direction. For example, although Distance analysis found elephant to have a higher densit y in logged, hunted forest (which is in accordance with field observations), the full ZIP model predicted unlogg ed, unhunted forest to have a greater density than logged, hunted forest. This indicates that the effects of disturba nce were confounded by the correlation between geographic coordina tes and hunting indices (latitude: r =-0.66, df=28,
30 p <0.001; longitude: r =-0.53, df=28, p =0.002): the intensity of hun ting and the presence of logged forest both decreased w ith greater distance from the main logging village along a northeastern gradient. With sp atial coordinates absorbing most of the effects of logging and hunting, any differences in guild densities among s ites with different disturbance regimes, like those for pig, elephant, and monkey, were driven by other unmeasured differences among sites. In the full model, the densities of severa l guilds varied according to forest structure and fruit resources. Elephant density was predicted to be higher in forest with a heterogenous canopy with relatively greater numbers of large trees. Ape density was also higher in forest with variable canopy cover and with more lianas. De nsities of frugivorous bird and monkey were higher in less heterogeneous canopy, with frugivo rous bird having higher densities in forest without lianas and monkey having higher densitie s in forest with lianas. Both monkey and frugivorous bird density increased wi th increased fruit abundance. The disturbance model (excluding latitude and longitude) decoupl ed the effects of hunting and logging. After running the ZIP models, we predicted guild density in the three sites keeping all other variables at their mean values and then ca lculated the effects of hunting, logging, and hunting and logging (Figure 6). Hunti ng negatively affected five of the eight guilds: elephant, frugivorous bi rd, and squirrel did not have significantly lower densities. Hunting had the largest negative effect on ape, reducing its de nsity by 61%, and the largest positive effect on frugivorous bird, increasing it s density by 77%. Logging lowered the densities of ape, duiker, monkey, pig, and frugivorous bird, but increased bird and elephant densities. The effect of logging ranged from a density reduction of 44% for pigs to a de nsity increase of 90% for insectivorous birds. The combined effect of logging and hunting negatively affected guild
31 densities of ape, duiker, monkey a nd pig; elephant, bird, and squirrel guilds showed increases in density. Discussion In tropical forest, logging is nearly al ways accompanied by hunting, road construction, and population growth (Robinson et al. 1999, Wilkie et al. 2001, Laporte et al. 2007, Poulsen et al. 2009b). The entangled nature of these disturbances complicates our ability to resolve basic management problems. Can forest degraded by logging sustain communities of tropical forest animals? Can hunting be compatible with the c onservation of forest bi odiversity? To answer these questions the individual effects of distur bances must be decoupled from each other and separated from environmental variation. By comp aring sites differentially affected by hunting and logging, we found the landscape-l evel effects of disturbance on densities of animal guilds to outweigh the effects of variati on in forest structure, canopy cover and fruit abundance. The interaction between logging and hunting was not straightforward. Logging and hunting sometimes worked together to dramatically reduce densities of some guilds (ape, duiker, monkey, and pig). For other gu ilds (elephant, frugivorous and insectivorous bird, squirrel), hunting and logging worked against each other with positive impacts of one type of disturbance buffering negative impacts of the other. T ogether hunting and logging have shifted the composition of the animal community away from large mammals towards squirrels and birds. Selective logging modifies both the large scale architecture of the forest and local scale environmental characteristics. At the landscape level, selective logging primarily alters the forest through the creation of felling gaps durin g timber extraction and the construction of roads and skid trails (Asner et al. 2004). Selective logging in the Congo Basin usually only damages 10-20% of the logged area (Durrieu de Madron et al. 2000, Van Gemerden et al. 2003), but canopy damage can be spatially distributed ove r the logged area, resulting in extensive
32 fragmentation and soft edge effects in contiguous forest (Broadbent et al. 2008). In our study we found the landscape level effects of selective logging to strongly influence densities of animal guilds. In a comparison of logged and unlogged s ites, even logging conducted at low intensity (2.5 trees ha-1) 25 years ago resulted in a detectable de cline in densities of all animal guilds except insectivorous birds and elephants. Altho ugh we positioned transect s so that they would be removed from the immediate influence of roads, logged forest is discontinuous, broken by active and inactive roads, and interrupted by fellin g gaps. These features can strongly influence animal density and distribution. At the local scale, logging opens the canopy and changes forest structure, which has cascading effects on the light re gime, vegetation regeneration a nd fruit production within the felling gap (Putz et al. 2001). Lo cal scale environmental variati on linked to habitat alteration by logging also played a role in anim al density and distribution but to a lesser extent than landscape level characteristics. Canopy cove r and forest structure explained variation in guild densities for several guilds. Notably, elephant and ape pref erred forest with a heterogeneous canopy, whereas frugivorous and insectivorous bi rds and monkey preferred closed canopy forest. Resource abundance also determined densities of forest animals; frugivorous bird and monkey densities responded positively to fruit abundance. The crea tion of gaps released secondary species of fruiting trees like Musanga cecropioides which is consumed by large frugivorous birds, and probably has a large effect on thei r densities. Our resu lts demonstrate that logging by itself can impact densities of guilds and community composition. Even moderate hunting pressure can mark edly alter the stru cture of mammal communities in central Africa (F a et al. 2005, Laurance et al. 2006, Clark et al. 2009). Despite low population pressure (1 person km-2) and active enforcement of hunting laws, hunting
33 reduced the densities of duiker, monkey and pi g by 50, 30, and 31%. These declines in guild densities are consistent with market surveys of bushmeat demonstrating them to be the most commonly hunted animals (Poulsen et al. 2009b). On the other hand, hunting had a large positive effect on large frugivorous and insectiv ous birds and squirrels suggesting that the decline in densities of other guilds might re lease birds and squirrels from competition for resources. Hunting shifts the animal community structure to be composed of higher numbers of birds and small mammals (Fa et al. 2005). Change in the structure of the animal co mmunity may have cascading effects on forest dynamics through the modification or loss of ecol ogical services. Fragmentation and selective logging can indirectly influen ce plant abundance by modifying th e abundance and distribution of rodents and, in turn, seed predation (Asquith et al. 1997). Studies have also found the loss or reduction of large mammals from hunting to reduce seed dispersal rates for some plant species, to favor species dispersed abiotically and by bi rds and small mammals, and to alter seed banks and seedling and sapling recruitment (Wright et al. 2000, Wright et al. 2007a, Wright et al. 2007b, Terborgh et al. 2008). Large birds and monke ys in a Cameroonian fo rest had low dietary overlap and produced different seed shadow patte rns, making it unlikely that the loss of arboreal monkeys from logging and hunting in this study coul d be replaced by the seed dispersal services of birds (Poulsen et al. 2002, Clark et al. 2005). The degree to which di sruptions in ecological services affects tropical forests may depend on the time it takes animal communities to recover. If species composition of animal communities recovers within a few decades, degraded forest may help conserve animal species. On the ot her hand, if animal communities take much longer to recover, the recovery of plant communities and ecosystem processes may be slowed or fundamentally altered.
34 In this study we took advantage of a relativ ely long dataset (2 year s of observations) for Central African forest animals so that our conclusions are not biased by seasonal and phenological changes. In additi on, we only used direct observati ons of animals so that our estimates of species and guild densities do not in clude error introduced th rough estimates of nest and dung decay and production (Walsh and White 2005) We attributed the spatial gradient in guild densities to the effects of logging and hunti ng rather than topographic or floristic changes across the study area. This assumption is justif ied because the topography of this part of northern Congo is flat with little variation. In addition, except for a greater number of secondary trees in vegetation plots in logged forest, we did not detect major changes in tree species composition along the geographic gradient. In our study, hunting pressure and presence of logging decreased with distance from the village. This spatial gradient in disturbance was reflected in a lat itudinal gradient in densities for monkeys, duikers, and apes. The negative relati onship between the abundance of vertebrate communities and hunting pressure or human activity has been observed in other tropical forests (Muchaal and Ngandjui 1999, Peres 2000). This pattern should be exploited in the design of forest management landscapes. To the extent po ssible, industrial towns should be kept out of forests to minimize the human pressure on forest s and animals (Poulsen et al. 2009b). Where the construction of towns and sawmills is necessary in logging concessions, they should be located away from protected areas and positioned to minimize th e area of forest accessibl e to residents. With most of the forests of Central Africa ei ther in the process of being exploited or already logged, degraded forest represents an increasingly important habitat for most tropical animals and needs to be protected from further destruction. Improvement s in tropical logging practices and disincentives fo r illegal hunting are the most immediate ways to conserve
35 functional tropical forests. While our study demonstrates th at even logging alone can significantly modify the density of animal guilds and the structure of the animal community, the conservation value of logged and degraded fore sts might depend largely on the interactions between different forms of disturbance (e.g., agri culture, fire, population pr essure and hunting). To understand and mitigate disturbances, their e ffects need to be separated and quantified. Biodiversity management with in logging concessions should focus on managing for hunting, maintaining large areas of unlogged forest and mitigating the adverse effects of logging on sensitive groups of species.
36 Table 1-1. Number of observations, density of individuals and 95% confidence intervals (CI), and the coefficient of variation ( CV) for animal species in three type s of forest stands: unlogged, unhunted fores t; logged, unhunted forest; and logged, hunted forest. Asterisks indicate a significan t difference in density between A) logged/ hunted forest and logged/unhunted forest and B) logged unhunted forest and unlogged/ unhunted forest. Densities between site s were compared as two-sided t-tests and corrected for multiple comparisons so that statistical significan ce is assessed as p<0.005. Species No. observations per block Density of individuals km-2 (95% CI) CV Logged/Hunted Logged/Unhunted Unlogged/Unhunted Ape Gorilla gorilla 46, 79, 77 1.7 (0.8-3.7) 2.4 (1.6-3.6) 3.1 (2.0-4.7) 19.7-36.7 Pan troglodytes 50, 60, 104 1.7 (0.9-3.2) 1.9 (0.7-5.0) 6.2 (3.2-12.2) 32.5-49.3 Duiker Cephalophus monticolaA,B 101, 163, 210 4.7 (3.2-6.9) 12.3 (6.94-21.84) 22.5 (16.2-31.2) 16.7-29.6 Cephalophus spp. 186, 226, 256 8.9 (5.6-14.3) 10.9 (8.3-14.2) 13.2 (10.3-17.1) 11.9-22.4 Elephant Loxodonta africana 48, 50, 32 1.5 (0.8-3.0) 0.9 (0.5-1.6) 0.4 (0.2-0.9) 27.3-34.5 Large frugivorous bird Bycanistes albotibialisA,B 288, 236, 270 55.9 (41.1-76.1) 32.4 (22.4-46.9) 40.7 (26.3-62.9) 15.1-22.5 Bycanistes fistulatorA, B 61, 63, 90 121.5 (84.0-175.8) 67.6 (39.8-114.7) 25.8 (17.6-37.7) 18.7-27.2 Ceratogymna atrataA, B 197, 172, 179 29.7 (20.6-42.9) 13.6 (10.6-17.5) 19.8 (14.9-26.31) 12.5-19.9 Corythaeola cristataA 198, 90, 80 26.7 (16.3-43.7) 5.4 (3.3-9.0) 11.4 (5.7-22.8) 23.9-34.9 Psittacus erithacusA 338, 280, 271 91.9 (59.3-142.3) 50.0 (33.2-75.4) 31.1 (23.1-42.0) 14.8-22.4 Bird Tockus fasciatusB 110, 89, 62 12.0 (6.23-23.2) 9.4 (5.4-16.4) 1.0 (0.6-1.7) 26.3-33.6 Tockus cameras 104, 126, 56 3.8 (2.5-5.9) 9.0 (6.2-13.1) 2.9 (1.8-4.7) 18.8-25.0 Tockus albocristatusA,B 133, 151, 134 7.1 (3.8-13.5) 17.0 (11.4-25.5) 9.8 (6.0-16.0) 20.3-32.6 Monkey Cercopithecus nictitansA,B 310, 336, 361 37.2 (24.4-56.8) 41.8 (32.8-53.1) 79.7 (62.1-102.3) 11.8-19.9
37 Table 1-1. Continued Cercopithecus pogoniasA,B 227, 252, 259 23.3 (15.9-34.2) 28.4 (20.5-39.4) 34.9 (24.8-49.1) 16.4-19.2 Cercopithecus cephusA 159, 184, 171 28.3 (15.9-50.1) 39.3 (17.2-89.3) 39.7 (27.8-56.6) 17.8-43.3 Lophocebus albigenaA,B 171, 275, 318 14.9 (6.6-33.5) 26.1 (19.8-34.5) 49.5 (34.8-70.3) 13.9-42.4 Squirrel Funisciurus lemniscatusA 75, 80, 83 29.4 (17.4-49.7) 14.6 (10.1-21.1) 12.7 (8.4-19.1) 18.4-25.7 Protoxerus stangeriA 56, 43, 78 26.8 (11.9-60.1) 12.6 (5.6-28.3) 15.5 (9.1-26.5) 26.3-41.9 Density estimated with less than 40 observations. Other species encountered during th e survey with insufficient observa tions for estimation of densities: Panthera pardus (Carnivore), Cephalophus sylvicultor (Duiker), Cercopithecus neglectus, Cercocebus agilis Piliocolobus badius, Colobus guereza (Monkey), Hylochoerus meinertzhageni, Potamochoerus porcus (Pig), Atherurus africanus (Rodent), Syncerus caffer nanus Tragelaphus euryceros (Large ungulate).
38 Table 1-2. Results of multivariate analyses. A) Separate analyses: eigenvalues and percentages of the total inertia accounted for by axes 1 and 2. Ordinations were a principal components analysis (PCA) of R, multiple correspondence analysis (MCA) of Q, and a correspondence analysis (CA) of L. B) Summary of RLQ analys is: eigenvalues and percentages of the total co-i nertia accounted for by RLQ axes 1 and 2, covariance and correlation between the site scores constrai ned by the environmental variables in R and species scores constrained by the traits in Q, projected inertia of table R and Q onto the first two RLQ axes, and percentage of the inertia obtained in the separate analysis of R and Q along the same axes (e.g., RLQ axis 1 accounts for 53.5% of the projected inertia of R along PCA axis 1 [(100 x 1.39)/2.60 = 53.5%]. Statistic Axis1 Axis 2 A. Separate analyses R/PCA 2.60 (28.9%) 2.11 (23.4%) L/CA 0.10 (47.5%) 0.03 (11.9%) Q/MCA 2.39 (23.9%) 1.82 (18.2%) B. RLQ Analysis Eigenvalues (and % tota l inertia) 0.148 (82.7%) 0.007 (5.0%) Covariance 0.39 0.08 Correlation 0.21 0.05 Projected inertia Table R 1.95 (75.0%) 1.49 (70.5%) Table Q 1.79 (74.6%) 1.76 (96.8%)
39 Figure 1-1. Map of study area in northern Republic of Congo. Thir ty transects (2500 m) and 1 ha tree plots (open squares) were positio ned in a random stratified manner in the Kabo concessions and Nouabal-Ndoki National Park. All stations were separated by at least 2.5 km.
40 Figure 1-2. Densities of animal guilds aver aged over all transect during the 24 m onths from June 2005 to May 2007. Shading represents the 95% confidence intervals around the monthly mean densities. Note that we only systematically sampled squirrels for the last 9 months of the study.
41 Figure 1-3. Ordination based on prin cipal components analysis (PCA ) of thirty 1 ha plots showing the first, second, and third a xes. The biplot shows sites and environmental va riables. Transects within sites are de signated with different symbols (logged, hunted forest (open triangles) logged, unhunted forest (filled circles), and unlogged, unhunted fo rest (crosses). Arrows represent the most impo rtant environmental fact ors: under (no. of understory trees), mid (no. of midstory trees), canopy (no. of canopy trees), liana (no. of lianas), disturb (no. of seconda ry species), dist (distance of plot from village). Other variables that were included in the PCA, but did not impor tantly differentiate the plots included the number of trees, number of dead trees, average dbh, standard deviation of th e average dbh, and the % of transmitted light in the plot.
42 Figure 1-4. Display of (A) environmental variable s, (B) species traits, and (C) plots along the first two RLQ axes. Each plot is repres ented by a symbol representing the forest type. Abbreviations for plot characteristics include: Dst (disturbance), Frt (fruit abundance), Hnt (signs of hunting), DBH (mean diameter-at-breast height of trees > 10 cm dbh), Htr (canopy heterogeneity), Trs ( number of trees), Lna (number of trees topped by lianas), Udr (number of understore y trees), Cpy (numbe r of canopy trees). Abbreviations for species traits include si ze (S (small, < 5kg), SM (small-to-medium, < 10 kg), M (medium, <50 kg), L (large, > 50 kg)), feeding guild (B (brower), F (frugivore), FGr (frugivore-gran ivore), Gr (granivore), Gz (grazer), I (insectivore)), and protected status (U ( unprotected), PP (partially protected), P (protected)).
43 Figure 1-5. Dotplot of predictor values for animal guilds. Pa nel A depicts the effects of environmental (PCA1, PCA2, PCA3, and Hete ro), spatial (Lat, Long), and resource (Fruit) variables on guild densities. Pane l B demonstrates the effect of logging and hunting on each guild by depicting the di fference in log(guild density) from the overall log(mean) in logged, hunted forest (L+H+), logged, unhunted forest (L+H-), and unlogged, unhunted forest (L-H-). In both panels, points are the posterior means from the ZIP models and bars are 95% credible intervals. Credible intervals that cross the dotted vertical line suggest that the effect is not different from zero.
44 Figure 1-6. The effect of hunti ng, logging, and hunting and logging on guild density. The effects of disturbances were estimated as log -ratios of guild densities by dividing the posterior mean abundance of a guild at the disturbed site from the posterior mean abundance of the site where th at disturbance was not present; for example, the effect size for logging would be: Bars represent 95% credible intervals. Guilds are pres ented in order of increasing bodyweight and abbreviations are Bird (large insectivorous bird), Squ (squirrel ), Fbird (frugivorous bird), Mky (monkey), Duik (duiker), Pig, Ape, and Ele (elephant).
45 CHAPTER 2 SEED DISPERSAL PATTERNS DRIV E SEEDLING RECRUITMENT IN AN EXPERIMENTAL MANIPULATION OF SEED SHADOWS Unprecedented rates of logging and hunting in Ce ntral Africa threaten to transform vast tracts of primary forest into a mosaic of degraded forest emptied of its animals. Reductions in abundances of seed-dispersing animals are hypothe sized to alter patterns of seed deposition for many tree species, potentially limiting forest regene ration with long-term consequences for forest structure and composition. To determine how lo gging and hunting affect seed dispersal, we measured seed rain in 30 1 ha tree plots in forest disturbed by logging and hunting, logging alone, and neither logging nor hun ting. We used inverse modeling techniques with two years of seed rain data to quantify seed shadows for 26 tree species representing bird-, wind-, and mammal-dispersed species. Mean dispersal distance and degree of dispersion vary with dispersal vector, tree density and size, all of which are modified by hunting and logging. Across all forest types, wind-dispersed species had longer mean dispersal distances and greater dispersion relative to other di spersal modes. Hunting decrea sed mean dispersal distance of mammaland bird-dispersed sp ecies by at least 20%. Loggi ng decreased mean dispersal distance of bird-dispersed species by 37%, in creased mean dispersal distances of mammaldispersed species by 28%, and increased mean dispersal of wind-dispersed species by 49%. Both hunting and logging increase d dispersion of wind-dispersed sp ecies. By altering seed dispersal patterns, human disturbance may provi de an advantage for the regeneration of winddispersed species to the detriment of animal-dispe rsed species, with long-term consequences for forest structure and composition. Introduction Understanding seed dispersal is critical to understanding plant population and community dynamics (Nathan and Muller-Landau 2000, Levine and Murrell 2003). By determining the
46 spatial distribution of seeds, seed dispersa l creates the template through which all other ecological processes that limit plant populations will be filter ed. Many plant populations are limited for lack of seed (Poulsen et al. 2007), thou gh the strength of seed limitation is usually small relative to post-dispersal factors lik e microsite limitation (John et al. 2007), densitydependence (Harms et al. 2000), and predation (Paine and Beck 2007) that kill seeds and emerging seedlings (Clark et al. 2007). Despite the diversity and stre ngth of post-dispersal processes (Wang and Smith 2002), spatial patterns of tropical trees can be correlated with seed dispersal patterns, indicating th at dispersal patterns can have lasting effects on the spatial distribution of trees (Sei dler and Plotkin 2006). In the last decade, our understanding of th e mechanisms that influence seed dispersal processes has advanced considerably. Ecologi sts have employed a vari ety of tools including mechanistic models (Greene and Johnson 1989, Na than et al. 2002, Greene et al. 2008), single source models (Clark et al. 2005), inverse modeli ng (Ribbens et al. 1994, Clark et al. 1999), and genetic sampling (Godoy and Jordano 2001, Hardesty et al. 2006) to quantify dispersal patterns. Mechanistic studies of wind dispersal have demons trated the effects of wind speed, height of release, wind loading, and directionality on pa tterns of seed dispersal (Greene and Johnson 1989, Nathan et al. 2001, Nathan et al. 2002, Greene et al. 2008). Mechanistic models of dispersal by animals have presented a greater challenge b ecause of the complexity of modeling animal behavior, but seed dispersal curves have been pr edicted taking into account seed passage times, movement rates, and some aspects of disp erser behavior (Holbrook and Smith 2000, Westcott and Graham 2000, Levey et al. 2005, Russo et al 2006). Comparisons of seed shadows of multiple tree species have demonstrated detectable differences in the spatial patterns of seed deposition produced by different di spersal vectors (Clark et al. 2005, Muller-Landau et al. 2008).
47 In addition to dispersal mode, subs tantial proportions of variation in interspecific patterns of seed rain can also be explained by pl ant traits like seed mass and tr ee height (Muller-Landau et al. 2008). Landscape traits such as plant aggregat ion and frugivore density also affect seed dispersal patterns, with greater plant aggregation reducing mean dispersal distances and greater frugivore density resulting in kernels with fat ta ils (Morales and Carlo 2006). Together these studies provide an idea of the most important m echanisms drive patterns of seed dispersal. One of the most critical applications of these techniques is understanding how disturbance to forest affects seed dispersal patt erns and the potential consequences for seedling recruitment, forest structure and diversity. Commercial logging is probably the most farreaching anthropogenic disturbances in the tropics, fragmenting millions of square kilometers of tropical forest (Nepstad et al 1999, Asner et al. 2005, Laporte et al. 2007). Selective logging primarily alters forest structure through the cr eation of felling gaps during timber extraction and the construction of roads and skid trails (Asn er et al. 2004). The effects of logging depend on extraction techniques and intens ities (Putz et al. 2001), but canopy damage can be spatially distributed over the logged area, resulting in extensive fragmentat ion and soft edge effects in contiguous forest (Broadbent et al. 2008). Becau se logging operations open remote forests to access by people (Poulsen et al. 2009b), timber e xploitation is usually accompanied by hunting, dramatically reducing the abunda nce of tropical seed-dispersing animals (Wilkie et al. 2001, Peres and Palacios 2007, Poulsen et al. 2009a). The goal of this study is to assess the eff ects of logging and hunting on community-level patterns of seed dispersal in a Central African forest. To deco uple the effects of logging and hunting, we set up a large-scale obs ervational experiment to quantif y seed dispersal patterns in forest stands that were exposed vs. unexposed to logging and hunting. Over two years, we
48 collected seed rain data in thirty 1 ha tree plots that spanned a 55 km gradient capturing the landscape-scale variation in the density of plant species and co mmunity composition in northern Republic of Congo. To quantify seed dispersal on multiple species of trees, we employ inverse modeling, which is ideal for analyzing data from seed trapping studies within multi-species mapped stands (Ribbens et al. 1994, Clark 1998, Clark et al. 1999, Muller-La ndau et al. 2008). We relate observed dispersal parameters to fact ors hypothesized to explain interspecific variation in seed production and dispersal distances incl uding disturbance, disp ersal mode, tree size and density. Methods Study Area We conducted the study in the Nouabal-N doki National Park (NNNP; 400,000 ha) and the Kabo logging concession (267,000 ha) in northe rn Republic of Congo. The Kabo concession borders the NNNP to the south, and together they include a mosaic of logged and unlogged forest. Between 20-25 years before the study, this area of the logging concession was selectively logged at low intensity (<2.5 stem ha-1) with four species, Entandophragma cylindricum E. utile Triplochiton scleroxylon and Milicia excelsa making up 90% of the cutting volume (Congolaise Industrielle des Bois 2006). Approximately 3000 people inhabit the study site, most of who live in the logging town of Kabo. Reside nts hunt with shotguns, and to a lesser extent with wire snares, to suppleme nt their diets and for local tr ade (Poulsen et al. 2009b). Most hunting originates from the town of Kabo creating a gradient of hunting intensity that deceases with distance from it, with some forest types us ed more than others (Mockrin 2008). The forests are classified as lowland tropical forest, and dominant tree families include Meliaceae, Euphorbiaceae, and Annonaceae. Annual rainfall averaged 1518 mm (SD = 96) from 2001 to 2006 and is seasonal with peak s in May and October.
49 Tree and Seed Census Data We established 30 1-ha tree plots ov er an area of approximately 3000 km2 to examine forest dynamics (Figure 1). We located ten pl ots in forest stands that were unlogged and unhunted, logged and unhunted, and logged and hunte d. Plots were only positioned in mixed lowland forest, with a buffer of at least 500 m to the nearest primary road and 100 m to the nearest water source. Using ArcView 3.2 and a 14 class habitat map, we ex tracted the areas that did not meet those criteria, and then randomly po sitioned plots on the remaining surface. Each of the plots was linked to a 2.5 km transect so that animal abunda nce could be related to seed dispersal. Within each plot, we tagged, measur ed, mapped and identified to species all trees greater than 10 cm diameter-at-br east-height (dbh) (species list and descriptions in Harris and Wortley 2008). For each tree, we also recorded the canopy status of the tree (understory, midstory, canopy, and emergent) and the presence of lianas in its crown. In addition, we quantified canopy openness and light availability fo r each plot, averaging the values of four pictures taken from each quarter of a plot. Photographs were taken every 30 cm above the ground, in uniformly overcast conditions in the ea rly morning or late afternoon with a leveled Nikon Coolpix P5000 camera body and Nikon FC-E8 Fi sheye converter lens. We analyzed images for the percentage of transmitted diffuse light using Gap Light Analyzer 2.0 (Frazer et al. 2001). We quantified fruit availability in each of the 30 plots by collecting fruits and seeds in 21 1m2 traps. Seeds traps were cen tered along three lines at 25, 50, and 75 m from the plot border, with 10 m separating each trap so that all traps were at least 20 m from the plot borders. Approximately every two weeks, we collected all seeds and fruits that fell into the traps and counted and identified them to species. Wh en unknown fruits and seeds were collected, we searched the canopy over the seed trap for fruit, us ing the tagged trees to identify the propagule.
50 Seed Shadow Models and Parameter Estimation We used an inverse modeling method that mode ls the density of seed s at a single location (seed trap) as the summed contribution of seeds di spersed from all conspecific adult trees in the plot. We assumed that the contribution of each a dult tree to the seed rain at a location depends only on its individual fecundity and its distance to the seed trap. Our approach is an adaptation of Ribbens et al. (1994) which wa s used to model the dispersion of seedlings, but has also been adapted to modeling seed disper sal (Clark 1998, Sagnard et al 2007, Muller-Landau et al. 2008). The expected number of seeds in seed trap in plot is the sum over each adult of the product of its scaled fecundity by the dispersal kernel which gives the pr obability that a seed is dispersed at a distance from a tree: We modeled adult fecundity by scali ng seed production to tree size, where the number of seeds produced by a tree ( ) is the size (diameter-at-breast height) scaled by which controls the dependence of fec undity on size. We assumed disp ersal was isotropic and modeled the seed shadow as a negative exponential dispersal kernel, Here is a normalization constant that controls overall s eed density by constraini ng the dispersal kernel between 0 and 1 and is the scale of dispersal (a verage dispersal distance). We assumed that the observed number of s eeds in a seed trap followed a negative binomial error distribution with mean equal to and clumping parameter (Hilborn and Mangel 1997, Clark 1998). We calibrated parame ters for the dispersal model by searching numerically for the combination of paramete r values that maximize the likelihood function, using the L-BFSG-B method(R De velopment Core Team 2008).
51 We analyzed interspecific variation in estimat ed mean seed dispersal distance (henceforth referred to as mean dispersal distance), the m ean seed production per adul t diameter (henceforth referred to as mean fecundity), and the parameter of the negative binomial distribution (henceforth referred to as the dispersion paramete r). Smaller values of the dispersion parameter reflect more clumped distributions of seeds acro ss traps, while larger values reflect greater dispersion of seeds among traps. Plant Species Trait Data We classified tree species by their disp ersal mode based on fruit morphology and published and unpublished observations of fruit consumption (Gautier-Hion et al. 1985, Tutin et al. 1997, White and Abernathy 1997, Whitney et al. 1998, Clark et al. 2001, Poulsen et al. 2001, Poulsen et al. 2002, Hawthorne a nd Gyakari 2006, Morgan and Sanz 2006). Despite the use of broad categories, many of the animal-dispersed species were disper sed by both birds and mammals. We assigned a predominant dispersal syndrome of mammal or bird on the basis of fruit and seed traits for plant species with dispersal agents from multiple animal-dispersed categories (Gautier-Hion et al. 1985, Poulsen et al. 2002). In addition to dispersal mode, we also quan tified the mean dbh (cm) and tree density for each species by treatment combination to relate disp ersal parameters to species characteristics. We used the log-transformed values of dbh and de nsity to meet the assumptions of normality. Comparison of dispersal parameters To compare dispersal characteristics with plant species traits, we used linear mixed models to relate dispersal parameters ( ) to dispersal mode (b ird, mammal, and wind), disturbance (hunting, no hunting, logging, and no l ogging), density and dbh of the tree species. Linear mixed models were implemented in th e nlme package in R 2.7.1 (R Development Core Team 2008). For comparisons of dispersal parame ters by dispersal mode and disturbance level
52 significance determined with Tukey multiple compar isons of means (using a z-statistic) in the multcomp package (Hothorn et al. 2008). Note al so that because we cl ass trees by dispersal vector (bird-, mammal-, and wind-dispersed), we have low sample sizes and many of the comparisons of dispersal parameters are not statis tically significant. In spite of the lack of statistical significance, we emphasi ze differences in dispersal patterns where we think they are biologically important (see Caveats). Results Twenty-six tree species met the minimum sample size for analysis. These included seven wind-dispersed species, 10 bird-dispersed specie s, and 11 mammal-dispersed species (Table 21). Wind-dispersed species (41.8 cm dbh) were la rger than species dispersed by other vectors (bird-dispersed species = 29.5 cm dbh, animal-d ispersed species = 26.9 cm dbh), and the logtransformed dbh of wind-dispersed species was significantly larger than that of mammaldispersed species (z = 2.66, p = 0.02) but not bird-dispersed species (z = 01.64, p = 0.23). The densities of tree species did not differ signi ficantly by dispersal mode (bird = 5.1 trees ha-1, mammal = 9.1 trees ha-1, wind = 5.6 trees ha-1). Mean dispersal distances were marginally negati vely correlated with density (r = -0.34, df = 26, p = 0.07), but not dbh (r = 0.17, df = 26, p = 0.38) Dispersion increased with size (dbh) of trees (r = 0.54, df = 26, p = 0.003), but not with density (r = 0.03, df = 26, p = 0.88). Effect of Dispersal Mode Wind-dispersed trees dispersed the farthest, ne arly 1.5 times farther than bird-dispersed species (z = 1.43, p = 0.33) and 2 times farther th an mammal-dispersed trees (z = 1.92, p = 0.13; Table 2-2). Dispersion of wind-dispersed seeds was also 2 tim es greater than for bird(z = 0.14, p=0.18) and 3.5 times greater than for mammal-di spersed species (z = 2.22, p = 0.02). Although bird-dispersed species had the highest fecund ity parameter (1.8), fecundity did not vary
53 significantly among trees dispersed by diffe rent vectors (mammal-dispersed = 1.4, winddispersed = 1.4). Effect of Disturbance Hunting tended to decrease the distance that seeds were dispersed. Seeds in hunted forest dispersed 33% farther than unhunted forest (Table 2-2, Table 2-3) although dispersal distances were not significantly different between the fo rest types. When differentiated by vector, dispersal distances of bird-dispersed trees in unhunted forest were 35% higher than in hunted forest. Mammal-dispersed trees in unhunted forest were dispersed 50% fu rther than in unhunted forest (Figure 2-3). There was only a 3% differe nce in dispersal differences of wind-dispersed species between hunted and hunted forest (Figur e 2-4). Hunting increased the dispersion of seeds for wind-dispersed (72%) and animal-dispersed trees (44%). Overall, mean dispersal distances in logged fo rest were 7% farther than unlogged forest (Table 2-2, Table 2-3). Logging decreased mean dispersal distance of bird-dispersed species by 37%, increased mean dispersal distances of ma mmal-dispersed species by 28%, and increased mean dispersal of wind-dispersed species by 49 % (Figure 2-3). Dispersion rates were 21% higher in logged than unlogged forest, although not statistically diffe rent (Figure 2-4). Dispersion of bird-dispersed trees in logged forest was 63% lower than in unlogged forest, whereas seed dispersal of wind-disper sed trees was 42% more dispersed. Discussion We show that human disturbance of tropical fo rests alters the seed dispersal patterns of trees: hunting reduced mean disp ersal distances of animal-dis persed trees whereas logging increased the dispersal distances and dispersion of wind-dispersed trees. M ean dispersal distance and degree of dispersion vary with dispersal vector, tree density and dbh, all of which are modified by hunting and logging. While our underst anding of both the patterns of seed dispersal
54 and the mechanisms that drive those patterns ha s increased enormously over the last decade, most of the work has focused on intact forest sy stems. Our study is the first to examine seed dispersal patterns over a large, tropical landscape and to compar e forest stands with different types of disturbances. Previous studies have shown that hunting in tropical forests can di srupt seed dispersal mutualisms by reducing the quantity of seeds re moved (Wright et al. 2000, Wright and Duber 2001, Wang et al. 2007, Brodi e et al. 2009) or the distances ov er which they are transported (Chapman and Onderdonk 1998). We show that over a large tract of forest with hunting levels that are similar to or lower than most forests in West and Central Afri ca, hunting altered seed dispersal patterns by reducing di spersal distances. Although the forests of northern Congo still retain their full complement of frugivorous bird s and mammals, these reductions in dispersal distances are a startling trend. Brodie et al. (2009) employed a structured population model to demonstrate that overhunting of tropical frugi vores could create an extinction debt, whereby adults of long-lived trees may remain extant (sometimes even common), but slowly disappear through attrition as they fail to be replaced due to lack of seed di spersal. By selectively reducing populations of larger vertebrate s (Fa et al. 2005, Poulsen et al. 2009a), timber exploitation and hunting likely have a disproportional negative imp act on influence on large-seeded mature forest plant species. The changes in wind dispersal dynamics in our study may be due to the impacts of logging on forest structure. Loggi ng modifies canopy structure at the tree-scale through timber extraction or by causing residual damage to nearby tr ees. Selective logging in particular targets the largest trees, result ing in a more heterogeneous canopy with fewer canopy and emergent trees relative to midstory trees (Asner et al. 2006). Bohrer et al (2008) found that in heterogeneous
55 canopy relatively short trees not overcast by cano py trees are more likel y to encounter strong updrafts and more likely to be ejected above th e canopy. Updrafts with high wind speeds are more likely to cause abscission of seeds from the tree (which requires drag or branch vibration) and to disperse seeds farther (Greene 2005, Schippe rs and Jongejans 2005). Logged forest in our study has a noticeably less even canopy, and ther efore wind dynamics may explain the increase in mean distance and dispersion of wind-dispersed species. The opposite seed dispersal patt erns exhibited by mammala nd bird-dispersed trees in our study might be explained by the differential re sponses of mammals and birds to changes in canopy structure. We show mammal-dispersed speci es to have slightly longer mean dispersal distances in logged forest which may result from the greater distances be tween emergent trees; monkeys, in particular, use emergent trees as sl eeping or resting trees and often deposit large numbers of seeds under them (Russo and Augspur ger 2004). By contrast logging reduced mean dispersal distances of bird-dispersed species and in creased the spatial aggregation of seeds. This may occur if logging alters the distribution a nd abundance of trees th at are preferred by frugivorous birds. At our study si te, logged forest had a higher pr oportion of light-gap specialist trees (e.g., Musanga spp. and Macaranga spp., (Poulsen, Clark and Bolker, unpubl. data) fed on by large frugivorous birds than unlogged fore st (Poulsen and Clark, unpubl. data). In a simulation study, Morales and Carlos (2006) simulate d bird dispersal in artificial landscapes and found that mean dispersal distan ces declined as plan t spatial aggregati on increased. They reasoned that denser landscapes retained simula ted birds that were more likely to spend time foraging or perching, and thus travel shorte r distances before dispersing seeds. In addition to investigating the effects of logging and hunting on seed dispersal patterns, our study highlights the importance of consideri ng seed dispersal mode in conjunction with
56 forest structure. Seed dispersal mode plays a role in determining di spersal patterns, but its role is perhaps less influential than predicted by animal-f ocused seed dispersal studies (Holbrook et al. 2002). Clark et al. (2005) found only small differences in mean dispersal distances between bird-, monkey-, and wind-dispersed species in Cameroon. Muller-Landau et al. (2008) found that seed dispersal mode did not explain significan t variation in seed dispersal distances, but did explain significant variation in clumping with animal-dispe rsed species showing higher aggregations of seed deposition. As our study suggests, the link between dispersal mode and dispersal pattern likely largely de pends on forest structure which can influence the behavior of seed-dispersing animals and the velocity and movement of wind-dispersed seeds. Caveats Through inverse modeling of 26 tree species we have found compelling evidence that dispersal mode, logging and hunting a lter patterns of seed dispersal. Given the relatively low number of tree species per vector, it is perhap s not surprising that we found few statistically significant differences in dispersal parameters in comparisons among different vectors and among comparisons of vector and disturbance type. We only incl uded in our analyses 26 of 428 for which we could model dispersal in all disturbance types (logged, unlogged, hunted, and unhunted). We may be able to add more species to our analys is and increase our statistical power by searching the data more thoroughly for species that can only be modeled in one or two of these disturbance types. In the face of low statistical power, our inferences are based on what we interpret as biological ly important trends. Even so, the results presented in this ch apter are informative albeit preliminary. Although we have worked out the broader fram ework for modeling the data, more intensive modeling is in order. First, we intend to compar e the fits of multiple dispersal kernels rather than
57 just relying on the negative expone ntial model. Other promisi ng models include the Gaussian (Clark 1998), two-dimensional e xponential, Bessel function (Tur chin 1998), full 2DT (Clark et al. 1999), lognormal (Greene et al. 2004), and Weibull functions (Tuf to et al. 1997). Second, we will consider the role of species fecundity in mo re detail. Third, we will attempt to incorporate interannual and among plot variation in seed pr oduction by fitting mixed models across the entire dataset using a Bayesian framewor k. Finally, we would like to e xpand our consideration of plant traits and plot characteristics th at might affect dispersal by inco rporating data on fruit and seed size as well as the forest structure an d animal abundance around each plot. Future of Tropical Forests What do changes in seed dispersal patterns mean for the future of tropical forests? This depends on two pieces of information: 1) the de gree to which seed dispersers are lost and dispersal rates are reduced, and 2) the recruitm ent success of seeds and seedlings under parent plants. At our study site, the densities of some animal species have been reduced by logging and hunting but no seed-dispersing animals have been extirpated to our knowledge (Poulsen et al. 2009a). With the changes in seed dispersal patter ns that we identified we would not expect large changes in forest composition or structure. Howe ver, tropical forests in parts of West Africa, Asia, and the Amazon have been emptied of mo st mediumand large-bodied species (Brook et al. 2003, Fa et al. 2003, Curran et al. 2004, Peres and Palacios 2007), with one study of hunting from 36 sites in West and Central Africa finding that 82% of harvested species were frugivores (Fa et al. 2005). Therefore, seed dispersal rates have and will be further reduced across tropical forests (Webb and Peart 2001, Wang et al. 2007). Wh at is not clear is to what extent frugivores can replace each other as seed dispersers, alt hough low dietary overlap between functionally similar guilds suggests that replacement may be ra re (Poulsen et al. 2002). Furthermore, we
58 need a better understanding of where the threshold lies between a population that can carry out ecological functions like seed dispersal and one that cannot (McConkey and Drake 2006). The loss of seed dispersal function could have a limited effect on tree populations if seeds and seedlings can recruit under parent plants. So far, however, the evidence suggests that species that lose dispersal services have reduced recruitment of s eedlings (Chapman and Onderdonk 1998, Balcomb and Chapman 2003). In experimental tests of recruitment under different seed distributions, Augspurger and Kitajima (1992) a nd Poulsen et al. (2009a) both found lower rates of seed dispersal to lower recr uitment. Lower recruitment was largely due to density-dependent mortality which is pervasive in tropical forests (Harms et al. 2000). Note that negative density dependence does not usually result in the absolute mortality of all seedli ngs (Harms et al. 2000), and so there is the unlikely possibility that ev ery tree could replace itself. In the extreme case where dispersers have been removed from forest s, the loss of dispersal function allowed species dispersed abiotically and by small birds and mammals to substitute for those dispersed by large birds and mammals, at least to the sapling st age (Terborgh et al. 2008). We found both hunting and logging to increase the disper sion of wind-dispersed trees s uggesting that not only do seed dispersal rates of animal-dispersed trees declin e with disturbance, but wind-dispersed species will also gain an advantage of more even dispersal across the landscape.
59 Table 2-1. Species, dispersal modes and samples. Species Dispersal mode Plots Traps Seeds Trees DBH (cm) Stems/ha Albizia gummifera Wind 16 70 561.5 38 31.15 1.3 Angylocalyx pynaertii Mammal 15 52 285 112 27.56 3.7 Caloncoba welwitschii Mammal 11 23 277 46 14.78 1.5 Celtis adolfi-friderici Mammal 26 80 627 164 30.49 5.5 Cleistopholis patens Bird 15 27 71 49 31.18 1.6 Diospyros bipindensis Mammal 17 32 214 382 12.67 12.7 Diospyros canaliculata Mammal 16 31 403 195 14.35 6.5 Entandrophragma cylindricum Wind 18 173 3174.5 104 59.5 3.5 Erythrophleum suaveolens Mammal 25 119 904 56 54.93 1.9 Garcinia punctata Mammal 13 37 980.5 38 20.62 1.3 Greenwayodendron suaveolens Bird 30 315 3717 457 24.83 15.2 Guarea cedrata Bird 18 47 173.5 40 28.59 1.3 Guarea thompsonii Bird 17 49 362 165 22.95 5.5 Lannea welwitschii Bird 19 29 217 20 27.52 0.7 Macaranga barteri Bird 18 30 1790 86 30.81 2.9 Manilkara mabokeensis Mammal 13 36 366 50 26.1 1.7 Nesogordonia kabingaensis Wind 16 34 492 165 27.63 5.5 Petersianthus macrocarpus Wind 30 338 1513 135 44.43 4.5 Pteleopsis hylodendron Wind 15 61 216 30 34.96 1 Pterocarpus soyauxii Wind 24 142 709.5 78 29.6 2.6 Staudtia kamerunensis Bird 20 67 163 13 40.75 0.4 Strombosia nigropunctata Mammal 22 63 582 323 20.37 10.8 Strombosia pustulata Mammal 23 55 254.5 212 27.88 7.1 Strombosiopsis tetrandra Mammal 26 83 480 143 32.27 4.8 Terminalia superba Wind 30 501 14729.5 93 46.57 3.1 Xylopia chrysophylla Bird 20 78 356 40 23.99 1.3 Xylopia hypolampra Bird 21 105 936 17 39.61 0.6 Xylopia phloiodora Bird 15 33 262 49 25.33 1.6
60 Table 2-2. Mean fitted values and their st andard deviations of model parameters. Model SD SD SD SD Bird 18.1 11.3 1.8 1.7 2.9 6.4 0.11 0.08 Monkey 13.4 12.6 1.4 1.5 8.7 24.7 0.19 0.23 Wind 26.9 12.6 1.4 1.4 6.5 14.3 0.39 0.32 Hunt 15.6 12.9 1.7 1.7 5.7 13.2 0.22 0.25 No hunt 20.8 14.2 1.3 1.4 4.5 9.5 0.19 0.21 Log 16.8 11.8 1.4 1.1 4.2 9.5 0.29 0.21 No log 15.7 13.3 2.0 2.0 14.5 34.9 0.24 0.32 Bird, Hunt 15.2 12.5 1.8 2.0 2.0 2.6 0.17 0.11 Bird, No hunt 20.6 12.5 1.6 1.5 3.5 5.8 0.19 0.20 Bird, Log 12.5 9.3 1.6 1.5 5.5 12.3 0.14 0.08 Bird, No log 19.8 12.9 2.8 2.0 1.7 3.1 0.37 0.48 Mammal, Hunt 10.9 8.5 1.8 1.8 11.3 18.9 0.13 0.14 Mammal, No hunt 16.4 13.5 0.8 1.0 2.8 5.8 0.09 0.04 Mammal, Log 14.5 11.4 1.2 1.0 4.9 9.2 0.11 0.08 Mammal, No log 11.3 13.7 1.8 2.1 22.1 47.9 0.11 0.07 Wind, Hunt 28.3 17.2 1.3 0.5 0.2 0.3 0.55 0.42 Wind, No hunt 27.4 16.5 1.5 1.9 8.2 16.1 0.32 0.32 Wind, Log 29.3 9.0 1.1 0.7 0.4 0.5 0.47 0.33 Wind, No log 19.6 12.0 1.4 1.7 15.6 22.0 0.33 0.37
61 Table 2-3. Fitted parameters for full (F ; all plots), hunted (H+), logged (L+), unhunt ed (H-) and unlogged (L-) models. For pa rameter estimates denoted as dashes (-), there were not enough tr ees or trap hits to parameterize the model (Mod). Species Mod SE Z P SE Z P SE Z P SE Z P Albizia gummifera F 38.9 17.83 2.18 0.03 0.8 0.39 2.18 0.03 0.12 1.21 -1.56 0.12 0.12 1.21 -11.29 0.00 H+ 15.7 7.36 2.13 0.03 0.9 0.56 1.61 0.11 0.34 1.29 -0.44 0.66 0.10 1.29 -9.00 0.00 H52.0 33.51 1.55 0.12 0.7 0.44 1.67 0.10 0.15 1.32 -1.20 0.23 0.17 1.32 -6.41 0.00 L+ 29.9 13.94 2.14 0.03 0.5 0.58 0.91 0.36 0.47 1.23 -0.35 0.73 0.10 1.23 -10.96 0.00 LAngylocalyx pynaertii F 35.4 19.96 1.78 0.08 0.6 0.89 0.62 0.53 0.12 1.23 -0.62 0.54 0.08 1.23 -11.97 0.00 H+ 21.2 11.95 1.77 0.08 2.9 1.10 2.63 0.01 0.00 1.53 -2.59 0.01 0.09 1.53 -5.72 0.00 H26.4 17.83 1.48 0.14 0.0 1.01 0 1 1.25 1.27 0.05 0.96 0.09 1.27 -9.71 0.00 L+ 41.2 26.56 1.55 0.12 0.7 0.89 0.74 0.46 0.08 1.25 -0.77 0.44 0.09 1.25 -10.42 0.00 L6.8 0.0 9.05 0.08 Celtis adolfi-friderici F 18.1 3.94 4.61 0.00 2.3 0.67 3.41 0.00 0.00 1.17 -3.02 0.00 0.08 1.17 -15.54 0.00 H+ 23.3 9.53 2.45 0.01 2.8 0.90 3.09 0.00 0.00 1.23 -2.72 0.01 0.10 1.23 -11.35 0.00 H11.6 3.12 3.73 0.00 1.5 1.01 1.44 0.15 0.01 1.32 -1.18 0.24 0.08 1.32 -9.14 0.00 L+ 18.5 4.70 3.93 0.00 2.7 0.79 3.39 0.00 0.00 1.19 -3.00 0.00 0.09 1.19 -13.93 0.00 L16.3 7.41 2.21 0.03 0.8 1.32 0.64 0.52 0.07 1.55 -0.54 0.59 0.06 1.55 -6.43 0.00 Cleistopholis patens F 8.4 2.30 3.65 0.00 0.6 1.08 0.59 0.56 0.44 1.53 -0.20 0.84 0.12 1.53 -4.98 0.00 H+ 4.5 2.47 1.84 0.07 0.8 1.61 0.38 3.03 -0.87 0.38 H9.3 3.06 3.04 0.00 0.7 1.22 0.57 0.57 0.28 1.57 -0.29 0.78 0.11 1.57 -5.00 0.00 L+ 8.4 3.95 2.14 0.03 2.1 1.55 1.33 0.18 0.00 1.98 -1.14 0.26 0.20 1.98 -2.33 0.02 L7.6 2.68 2.82 0.00 0.0 1.59 0.00 1.00 7.80 1.78 0.34 0.74 0.11 1.78 -3.77 0.00 Diospyros bipindensis F 6.4 2.01 3.16 0.00 0.0 4.19 0.00 1.00 3.43 1.29 0.12 0.91 0.04 1.29 -12.32 0.00 H+ 8.6 14.07 0.61 0.54 0.0 17.91 0.00 1.00 2.72 2.53 0.02 0.98 0.04 2.53 -3.60 0.00 H5.7 1.84 3.08 0.00 0.0 4.34 0.00 1.00 4.16 1.31 0.13 0.89 0.04 1.31 -11.62 0.00 L+ 6.1 2.81 2.18 0.03 0.0 7.90 0.00 1.00 5.17 1.52 0.08 0.93 0.04 1.52 -7.48 0.00 L6.2 2.71 2.31 0.02 2.9 5.58 0.52 0.60 0.00 1.41 -0.47 0.64 0.04 1.41 -9.19 0.00
62 Table 2-3. Continued Diospyros canaliculata F 5.4 2.85 1.90 0.06 2.5 3.65 0.67 0.50 0.03 1.36 -0.36 0.72 0.03 1.36 -11.35 0.00 H+ 3.5 4.25 0.82 0.41 0.6 13.12 0.05 0.96 45.65 1.66 0.10 0.92 0.03 1.66 -6.88 0.00 H8.8 4.56 1.93 0.05 0.0 3.79 0.00 1.00 2.43 1.49 0.09 0.93 0.04 1.49 -8.20 0.00 L+ 6.6 3.70 1.78 0.08 1.7 4.96 0.35 0.72 0.11 1.39 -0.16 0.87 0.03 1.39 -10.80 0.00 L0.9 0.67 1.35 0.18 5.6 13.63 0.41 0.68 2.75 2.43 0.03 0.97 0.20 2.43 -1.84 0.07 Entandrophragma cylindricum F 16.2 0.0 35.51 0.17 1.09 -20.32 0.00 H+ 23.1 6.87 3.36 0.00 1.4 0.97 1.45 0.15 0.03 1.23 -0.80 0.43 0.28 1.23 -6.11 0.00 H14.7 0.0 43.00 0.15 1.11 -17.75 0.00 L+ 21.3 5.00 4.25 0.00 0.6 0.75 0.75 0.45 1.11 1.20 0.03 0.98 0.22 1.20 -8.13 0.00 L14.8 0.0 46.93 0.16 1.12 -16.84 0.00 Erythrophleum suaveolens F 11.6 1.27 9.13 0.00 1.6 0.46 3.47 0.00 0.01 1.16 -2.15 0.03 0.24 1.16 -9.93 0.00 H+ 7.4 0.97 7.63 0.00 1.9 0.63 3.05 0.00 0.01 1.29 -1.67 0.09 0.51 1.29 -2.64 0.01 H14.2 2.38 5.97 0.00 1.4 0.62 2.19 0.03 0.03 1.20 -1.36 0.17 0.18 1.20 -9.36 0.00 L+ 10.1 1.18 8.61 0.00 2.3 0.57 4.11 0.00 0.00 1.22 -2.90 0.00 0.31 1.22 -6.06 0.00 L15.5 3.90 3.97 0.00 0.2 0.71 0.24 0.81 2.99 1.26 0.36 0.72 0.20 1.26 -7.06 0.00 Garcinia punctata F 15.7 5.39 2.92 0.00 0.4 1.60 0.23 0.81 4.69 1.25 0.29 0.77 0.06 1.25 -12.78 0.00 H+ 12.7 10.67 1.19 0.23 0.0 4.52 0.00 1.00 46.20 1.40 0.25 0.80 0.06 1.40 -8.40 0.00 H36.5 26.85 1.36 0.17 0.7 1.58 0.42 0.67 0.12 1.43 -0.45 0.65 0.10 1.43 -6.55 0.00 L+ 14.1 5.94 2.37 0.02 0.0 2.12 0.00 1.00 26.08 1.33 0.46 0.65 0.05 1.33 -10.70 0.00 L14.7 5.07 2.90 0.00 2.6 4.37 0.59 0.55 0.00 1.62 -0.54 0.59 0.22 1.62 -3.13 0.00 Greenwayodendron suaveolens F 30.3 4.00 7.58 0.00 3.5 0.58 6.03 0.00 0.00 1.08 -5.63 0.00 0.27 1.08 -17.52 0.00 H+ 25.0 6.96 3.59 0.00 3.5 0.92 3.86 0.00 0.00 1.20 -3.65 0.00 0.29 1.20 -6.97 0.00 H33.1 5.92 5.60 0.00 2.4 0.81 2.98 0.00 0.00 1.09 -2.73 0.01 0.28 1.09 -15.27 0.00 L+ 21.0 3.33 6.31 0.00 3.3 0.64 5.16 0.00 0.00 1.11 -4.61 0.00 0.31 1.11 -11.40 0.00 L
63 Table 2-3. Continued Guarea cedrata F 22.2 10.48 2.12 0.03 0.0 0.67 0.00 1.00 2.93 1.34 0.45 0.65 0.11 1.34 -7.61 0.00 H+ 7.9 2.53 3.12 0.00 0.4 0.83 0.51 0.61 5.07 1.47 0.61 0.54 0.12 1.47 -5.47 0.00 HL+ 7.9 2.53 3.12 0.00 0.4 0.83 0.51 0.61 5.07 1.47 0.61 0.54 0.12 1.47 -5.47 0.00 LGuarea thompsonii F 11.2 3.51 3.18 0.00 1.3 0.58 2.23 0.03 0.04 1.23 -1.69 0.09 0.07 1.23 -13.09 0.00 H+ 22.0 29.20 0.75 0.45 0.0 5.71 0.00 1.00 0.86 1.57 -0.01 0.99 0.09 1.57 -5.39 0.00 H7.9 1.92 4.10 0.00 2.5 0.77 3.21 0.00 0.00 1.31 -2.70 0.01 0.07 1.31 -9.58 0.00 L+ 10.4 3.08 3.37 0.00 0.4 1.17 0.31 0.76 1.01 1.27 0.00 1.00 0.08 1.27 -10.65 0.00 L11.1 6.67 1.66 0.10 3.2 1.53 2.09 0.04 0.00 1.49 -2.01 0.04 0.07 1.49 -6.83 0.00 Lannea welwitschii F 21.1 9.53 2.22 0.03 2.7 2.27 1.18 0.24 0.00 1.50 -1.04 0.30 0.03 1.50 -8.87 0.00 H+ 6.6 3.20 2.05 0.04 2.5 2.06 1.21 0.23 0.00 2.25 -0.78 0.44 0.06 2.25 -3.48 0.00 H28.2 21.69 1.30 0.19 0.0 4.92 0.00 1.00 4.90 1.64 0.08 0.93 0.03 1.64 -7.12 0.00 L+ 5.0 2.21 2.25 0.02 1.6 1.69 0.92 0.36 0.14 2.05 -0.29 0.78 0.06 2.05 -3.98 0.00 LMacaranga barteri F 7.7 2.02 3.79 0.00 3.5 1.97 1.75 0.08 0.00 1.32 -1.19 0.23 0.03 1.32 -12.43 0.00 H+ 2.2 0.79 2.81 0.00 3.0 2.08 1.45 0.15 0.01 1.83 -0.51 0.61 0.07 1.83 -4.46 0.00 H8.1 2.47 3.30 0.00 4.2 2.53 1.65 0.10 0.00 1.40 -1.22 0.22 0.03 1.40 -10.30 0.00 L+ 2.3 0.73 3.22 0.00 4.2 1.70 2.44 0.01 0.00 1.70 -1.39 0.17 0.09 1.70 -4.46 0.00 L8.1 3.26 2.47 0.01 3.5 3.41 1.04 0.30 0.0 1.44 -0.69 0.49 0.03 1.44 -9.70 0.00 Manilkara mabokeensis F 16.3 8.08 2.02 0.04 0.2 1.04 0.19 0.85 3.5 1.38 0.31 0.76 0.08 1.38 -7.99 0.00 H+ 3.5 1.28 2.73 0.01 1.4 1.28 1.07 0.29 2.4 1.61 0.17 0.86 0.23 1.61 -3.10 0.00 H41.2 58.12 0.71 0.48 0.0 1.69 0.00 1.00 0.7 1.67 -0.07 0.94 0.07 1.67 -5.09 0.00 L+ 13.0 5.22 2.50 0.01 0.0 1.02 0.00 1.00 17.1 1.46 0.70 0.48 0.16 1.46 -4.81 0.00 L1.8 1.92 0.91 0.36 0.0 7.76 0.00 1.00 77.1 2.42 0.14 0.89 0.08 2.42 -2.83 0.00
64 Table 2-3. Continued Nesogordonia kabingaensis F 29.0 22.78 1.28 0.20 5.3 1.60 3.34 0.00 0.0 1.25 -3.53 0.00 0.03 1.25 -15.94 0.00 H+ H23.7 15.87 1.49 0.14 5.4 1.72 3.16 0.00 0.0 1.26 -3.31 0.00 0.03 1.26 -14.81 0.00 L+ L15.5 8.28 1.87 0.06 4.1 3.43 1.19 0.24 0.0 1.52 -1.11 0.27 0.03 1.52 -8.34 0.00 Petersianthus macrocarpus F 38.6 6.93 5.57 0.00 1.8 0.20 8.92 0.00 0.0 1.09 -7.90 0.00 0.59 1.09 -5.85 0.00 H+ H28.3 4.04 7.01 0.00 1.9 0.29 6.51 0.00 0.0 1.10 -5.45 0.00 0.61 1.10 -4.97 0.00 L+ 42.3 12.31 3.44 0.00 1.7 0.27 6.32 0.00 0.0 1.14 -5.77 0.00 0.50 1.14 -5.29 0.00 L35.9 8.80 4.09 0.00 1.1 0.58 1.94 0.05 0.0 1.13 -1.43 0.15 0.70 1.13 -2.95 0.00 Pterocarpus soyauxii F 22.9 3.43 6.70 0.00 0.9 0.16 5.36 0.00 0.3 1.15 -2.02 0.04 0.43 1.15 -5.97 0.00 H+ 21.0 3.74 5.61 0.00 0.9 0.16 5.32 0.00 0.5 1.27 -0.91 0.36 0.95 1.27 -0.21 0.84 H45.4 25.65 1.77 0.08 0.3 0.42 0.77 0.44 0.3 1.25 -0.84 0.40 0.37 1.25 -4.45 0.00 L+ 20.4 3.13 6.53 0.00 1.0 0.17 5.95 0.00 0.2 1.19 -2.14 0.03 0.60 1.19 -2.90 0.00 LStaudtia kamerunensis F 59.2 43.50 1.36 0.17 1.3 1.32 1.02 0.31 0.0 1.32 -0.92 0.36 0.21 1.32 -5.71 0.00 H+ H43.0 46.02 0.94 0.35 3.2 2.45 1.30 0.19 0.0 1.72 -1.35 0.18 0.42 1.72 -1.59 0.11 L+ L43.0 46.02 0.94 0.35 3.2 2.45 1.30 0.19 0.0 1.72 -1.35 0.18 0.42 1.72 -1.59 0.11 Strombosia nigropunctata F 3.3 0.45 7.23 0.00 0.9 0.62 1.42 0.16 0.8 1.20 -0.10 0.92 0.10 1.20 -12.34 0.00 H+ 4.9 1.54 3.18 0.00 1.5 1.26 1.17 0.24 0.1 1.43 -0.63 0.53 0.13 1.43 -5.82 0.00 H2.0 0.36 5.60 0.00 0.3 0.66 0.42 0.68 18.9 1.24 1.31 0.19 0.11 1.24 -10.24 0.00 L+ 3.9 0.64 6.05 0.00 1.8 0.85 2.10 0.04 0.0 1.25 -1.22 0.22 0.13 1.25 -9.37 0.00 L1.5 0.53 2.85 0.00 0.0 1.64 0.00 1.00 149.4 1.47 1.11 0.27 0.09 1.47 -6.40 0.00
65 Table 2-3. Continued Strombosia pustulata F 7.5 1.29 5.80 0.00 1.5 0.62 2.44 0.01 0.0 1.28 -1.96 0.05 0.10 1.28 -9.36 0.00 H+ 1.4 0.90 1.58 0.11 0.9 1.60 0.57 0.57 15.5 1.98 0.42 0.67 0.08 1.98 -3.65 0.00 H8.4 1.61 5.20 0.00 2.5 0.84 3.01 0.00 0.0 1.30 -2.73 0.01 0.12 1.30 -7.93 0.00 L+ 6.7 1.34 4.99 0.00 1.4 0.61 2.25 0.02 0.0 1.37 -1.66 0.10 0.11 1.37 -6.97 0.00 L9.2 3.80 2.41 0.02 5.6 3.54 1.58 0.11 0.0 1.54 -1.47 0.14 0.10 1.54 -5.34 0.00 Strombosiopsis tetrandra F 22.5 6.54 3.44 0.00 1.9 0.83 2.25 0.02 0.0 1.18 -2.10 0.04 0.07 1.18 -16.03 0.00 H+ 22.6 29.25 0.77 0.44 5.9 8.48 0.70 0.49 0.0 1.46 -0.68 0.50 0.05 1.46 -7.86 0.00 H9.5 1.84 5.19 0.00 2.1 1.10 1.92 0.05 0.0 1.25 -1.54 0.12 0.13 1.25 -9.24 0.00 L+ 25.0 8.87 2.82 0.00 1.8 0.87 2.13 0.03 0.0 1.21 -2.01 0.04 0.08 1.21 -13.62 0.00 L2.6 0.70 3.76 0.00 2.1 1.85 1.13 0.26 0.1 1.46 -0.34 0.74 0.12 1.46 -5.55 0.00 Terminalia superba F 31.3 2.98 10.48 0.00 0.7 0.14 5.35 0.00 1.7 1.06 0.97 0.33 0.77 1.06 -4.07 0.00 H+ 53.7 13.68 3.92 0.00 2.0 0.26 7.47 0.00 0.0 1.11 -4.47 0.00 0.85 1.11 -1.58 0.11 H22.8 2.02 11.29 0.00 0.3 0.16 1.87 0.06 13.3 1.08 4.04 0.00 0.86 1.08 -1.86 0.06 L+ 32.6 3.32 9.83 0.00 2.1 0.21 9.96 0.00 0.0 1.08 -5.70 0.00 0.93 1.08 -0.94 0.35 L27.0 5.02 5.39 0.00 0.0 0.27 0.00 1.00 31.0 1.13 3.48 0.00 0.77 1.13 -2.07 0.04 Xylopia chrysophylla F 19.9 5.62 3.54 0.00 0.0 0.48 0.00 1.00 4.7 1.20 0.92 0.36 0.18 1.20 -9.58 0.00 H+ 23.8 9.97 2.39 0.02 0.0 0.53 0.00 1.00 3.3 1.22 0.59 0.55 0.21 1.22 -8.05 0.00 H13.6 6.02 2.26 0.02 0.0 2.23 0.00 1.00 12.8 1.60 0.42 0.67 0.10 1.60 -4.90 0.00 L+ 19.5 5.85 3.34 0.00 0.0 0.53 0.00 1.00 5.4 1.21 0.91 0.36 0.17 1.21 -9.10 0.00 L21.8 11.27 1.93 0.05 6.0 16.42 0.36 0.72 0.0 3.31 -0.36 0.72 1.34 3.31 0.24 0.81 Xylopia hypolampra F 26.7 6.21 4.30 0.00 2.5 0.39 6.27 0.00 0.0 1.16 -4.48 0.00 0.22 1.16 -10.04 0.00 H+ 38.3 24.82 1.54 0.12 5.6 3.17 1.76 0.08 0.0 1.32 -1.49 0.14 0.18 1.32 -6.06 0.00 H24.7 4.36 5.68 0.00 1.5 0.34 4.27 0.00 0.0 1.30 -2.83 0.00 0.60 1.30 -1.95 0.05 L+ 30.7 10.49 2.93 0.00 2.4 0.43 5.65 0.00 0.0 1.19 -4.06 0.00 0.19 1.19 -9.43 0.00 L29.3 11.97 2.44 0.01 3.1 1.39 2.24 0.03 0.0 1.43 -1.75 0.08 0.59 1.43 -1.45 0.15
66 Table 2-3. Continued Xylopia phloiodora F 17.2 9.76 1.76 0.08 0.0 1.33 0.00 1.00 8.5 1.33 0.58 0.57 0.05 1.33 -10.56 0.00 H+ 6.3 2.45 2.56 0.01 0.0 2.51 0.00 1.00 7.0 1.77 0.23 0.82 0.11 1.77 -3.90 0.00 H17.2 14.49 1.19 0.24 0.0 2.29 0.00 1.00 13.7 1.45 0.44 0.66 0.06 1.45 -7.69 0.00 L+ 7.0 2.26 3.11 0.00 0.0 1.41 0.00 1.00 37.7 1.41 0.97 0.33 0.07 1.41 -7.67 0.00 L17.6 32.93 0.53 0.59 0.5 4.3 0.02 2.25 -4.59 0.00
67 Figure 2-1. Location of the tree pl ots in the Kabo concession and Nouabal-Ndoki National Park in northern Republic of Congo.
68 Figure 2-2. Seed dispersal for Terminalia superba a wind-dispersed tree species, across the 30 tree plots. Each panel depicts a tree plot Plots 1-10 are in logged, hunted forest; plots 11-20 are in logged, unhunted forest, and plots 21-30 and in forest that is neither logged, nor hunted. Trees (green circles) vary in size with the dbh (cm) of the tree. Seed traps (squares) which cap tured at least 1 seed of Terminalia superba are depicted with the intensity of red increasi ng with greater densities of arriving seeds.
69 Figure 2-3. Mean dispersal distance (open circ les) for each species within forest type combination grouped by dispersal vector (anima l-, bird-, and wind-di spersed species). The closed, black circle is the vector mean dispersal distance over all the tree species.
70 Figure 2-4. Mean degree of dispersion (open ci rcles) for each species within forest type combination grouped by dispersal vector (anima l-, bird-, and wind-di spersed species). The closed, black circle is the vector m ean dispersion parameter over all the tree species.
71 CHAPTER 3 SEED DISPERSAL PATTERNS DRIV E SEEDLING RECRUITMENT IN AN EXPERIMENTAL MANIPULATION OF SEED SHADOWS African forests are being emptied of wildlif e by unsustainable levels of hunting. The extirpation of seed-dispersing animals alters the spatial patterns of seed dispersal for many tree species, with unknown consequences for their regene ration. To test the extent to which spatial distributions of seed deposition dr ive seedling recruitment, we experimentally manipulated seed dispersal patterns under individua ls of a monkey-dispersed tree, Manilkara mabokeensis We created seed distributions with all seeds deposited under the canopy (no dispersal), with density declining (negative exponential curve) as a function of distance from the tree (natural dispersal), and at uniform densities (even dispersal). These distributions mimicked seed dispersal patterns that could occur with the extirpation of monkeys by hunting, low levels of hunting, and high rates of seed dispersal. We monitored seedling emergence and survival for 18 months and recorded the number of leaves and da mage to leaves on all seedlings. Compared to natural dispersal, even dispersal increased seedling survival by 26%, whereas no dispersal reduced seedling survival by 78%. Survival of seedlings depended on the density of dispersed seeds but not the distance from the tree. Th ese results support hypothe ses that invoke densitydependence as a limiting factor to recruitmen t but they fail to s upport the Janzen-Connell hypothesis, which predicts that enhanced distance of seeds from conspecific adults improves survival. Our experiment demonstrates that se ed dispersal patterns strongly influence seedling recruitment and survival. More generally, we concluded that manage ment of hunting is a priority for the conservation of tropical forests be cause forests emptied of their seed dispersers will have limited regeneration capacity.
72 Introduction In Afrotropical forests, where animals disper se the majority of tr ee species, hunting and habitat conversion extirp ate or reduce the abundance of seed-d ispersing animals (Laurance et al. 2006, Peres and Palacios 2007, Poulsen et al. 2009a). In particular the unsustainable harvesting of wildlife is rapidly emptying tropical forest s of large and medium-bodied mammals (MilnerGulland et al. 2003, Fa et al. 2005). The loss of seed-dispersing animals is predicted to dramatically alter patterns of seed deposition with cascading effects for forest regeneration (Wright et al. 2000, Terborgh et al 2008, Brodie et al. 2009). Bu t to predict how the loss of dispersers will affect tropical forests we need to understand the relationship between spatial patterns of seed dispersal and the distribution, abundance, and diversity of tree species (Levine and Murrell 2003). The consequences of hunting and loss of disp ersal services for forest regeneration depend on the extent to which spatial patterns of seed dispersal determine the distribution and abundance of trees. The spatial pa ttern of dispersed seeds of a plant, the seed shadow, is characterized by the distance and density at which seeds are located away from th eir parent. Without dispersal most seeds will fall under or close to the parent plant, accumulating in high densities. Seed dispersal increases the average distance of seed s from their parent and decreases the average density of seeds at any one location; most seed shadows lead to decreasing seed densities as a function of distance from the pa rent plant (Nathan and Muller-La ndau 2000). Seed dispersal is said to lay the template of seeds from which s eeds germinate and seedlings recruit into adults (Schupp and Fuentes 1995). By this logic, tr ee recruitment is a dete rministic process and knowledge of dispersal patterns should allow one to predict the distributi on of trees in space. But several studies report a lack of concordance between patterns of seed deposition and seedling recruitment -demonstrating that post-dispersal processes act to dilute or erase the original
73 spatial distribution of seeds (Herrera et al. 1994, Harms et al. 2000, Balcomb and Chapman 2003). While we are quickly gaining a better grasp of factors that affect local seed dispersal patterns (e.g., Clark et al. 1999), we still do not k now whether these patterns have consequences for plant abundance and diversit y. In theory, by escaping from aggregated dispersal patterns close to the tree, dispersed seeds avoid mortalit y from intraspecific competition and the densityand distance-dependent behavi or of seed and seedling consumers (Janzen 1970, Connell 1971). In this way, seed dispersal increases species recruitment and abundance. However, depending on the scale of favorable and unfavorable sites for regeneration, short-range dispersal may lead to greater rates of recruitment and higher abundance than more distan t dispersal if seeds are more likely to fall into unsuitable mi crosites at greater distances (B olker and Pacala 1999). At the community level, high tree species diversity is pr omoted when seed dispersal is limited because limited dispersal slows rates of competitive di splacement and causes spatial segregation of heterospecific individuals across a landscape (Hubbell et al. 1999, Wright 2002). Thus, even at the local scale, short-range dispersal should in crease diversity and l ong-range dispersal should decrease diversity. The evidence that the spatial distribution of seeds determines juvenile and adult recruitment is mostly indirect and correla tive (Levine and Murrell 2003). Based on the knowledge that heterospecific differences in pa tterns of seed deposition can partially be explained by dispersal mode (Clark et al 2005, Muller-Landau et al. 2008), the spatial aggregation of conspecific saplings and trees has been shown to co rrelate with the mode of seed dispersal in tropical forest (Hubbell 1979, Condit et al. 2000, Seidler and Plotkin 2006). But to directly link the spatia l distribution of seeds to tree abundance and diversity will require
74 experimental approaches that decouple the dist ance and density components of the seed shadow and link them to patterns of seed lings and adult recruitment. In the only experimental study to date, Augspurger and Kitajima (1992) compared distributions of seeds of two wind-dispersed trees to demonstrate that high seed density and proximity to the parent plant lowered seedling recruitment. Their experimental approach allowe d them to directly examine the effects of seed distributions on seedling recruitm ent, but the low level of rep lication limited their ability to quantify the effects of different seed dispersal patterns and to separate distanceand densitydependent effects. Here we build on Augspurger and Kitajimas (1992) experiment to test whether the spatial distribution of dispersed seeds affects th e recruitment of a monkeydispersed tree species, Manilkara mabokeensis and the diversity of the seedling community. Based on seed densities from its natural seed shadow, we sowed seeds of M. mabokeensis under seven trees in three different distributions to mimi c scenarios of no dispersal, natural disper sal, and even dispersal. These distributions mimicked seed dispersal patterns that could occur with the extirpation of monkeys by hunti ng, low levels of hunting, and extraordinarily high rates of seed dispersal. We then monitored the recruiting se edlings for 18 months to test whether seed shadows affect seed and seedling survival and community diversity. We had four specific objectives. First, we examined how different seed shadows affect seedling recruitment and survival. Second, we decoupled the distanceand density-effects. Third, we determined the mechanisms (seed predation, herbivory) that li mit seedling recruitment. Fourth, we tested whether the seed shadow affects co mmunity diversity of seedlings.
75 Methods Overview Our experiment entailed several steps. Firs t we selected a tree species based on the criterion that its primary disperser (arboreal monkeys) was being impacted by hunting. Second, we quantified natural patterns of seed dispersa l around focal trees. Third, with the seed shadow data we calculated the densities of seeds to sow in the three experimental distributions (no dispersal, natural dispersal, and even dispersal). Fourth, we delineated wedges under seven individuals of our study species, M. mabokeensis and sowed seeds into them, according to the experimental distributions. Fifth, we monitored seedling recruitment and survival for 18 months. Finally, at the end of the experiment, we estimated the diversity of the seedling community in the wedges to examine the effects of the seed shadow on species coexistence. Study Site and Species We conducted this study in the Kabo loggi ng concession (220 N, 1625 E) in the northern Republic of Congo from July 2006 thro ugh July 2008. The logging concession is comprised primarily of lowland, semi-deciduous tr opical forest, and had been logged once in the late 1960s at a relatively low intensity (<2.5 tree ha-1). The study site is 20 km from the nearest village and subject to low ra tes of hunting. Human population growth in the concession has dramatically increased hunting pr essure on wildlife, including arbor eal monkeys, to the point that some species may be unsustainably harves ted (Poulsen et al. 2009b). For this study, we randomly selected seven reproductive individuals of M. mabokeensis (Sapotaceae), a canopy emergent tree whose fruits are primarily dispersed by arboreal monkeys. In the study area, M. mabokeensis occurs at a density of 1.67 trees per ha. On average the focal trees were separated by 2061 m, w ith the smallest distance between two trees being 197 m. The trunks measured 121.5 cm (SD = 47.6) diameter at breast height and canopy radius varied among
76 trees (mean = 8.5 m, SD = 2.5). The fruits are pulpy, measuring 3-5 cm in diameter with 1-3 seeds per fruit. Quantification of Seed Shadow To quantify the seed shadow of M. mabokeensis we placed seed traps under and around the canopies of four focal trees pr ior to fruit maturation. Traps were constructed of plastic mesh attached to wood frames and elevated to a he ight of 1-1.5 m above the ground. Using the trunk as the point of origin, traps were placed al ong four lines at 1, 2.5, 5, 10, 20, 40, and 60 m from the tree. To avoid directional bias, we randomly selected a starting angle for the first trap line and then separated the four trap lines by 90 inte rvals. We increased trap sizes with distance from the tree to sample 1% of the area at each trap annulus. We limited seed traps to 60 m from the parent because we were interested in seed ling recruitment in the local neighborhood of the plant, not long-distance dispersa l. Previous studies of animal-dispersed species have demonstrated that as much as 85% of seed cr ops are deposited within 20 m from a tree and only 3% on average are dispersed farther than 60 m (C lark et al. 2005). Fruits and seeds were collected from the traps and counted every two w eeks until all fruits had fallen from the trees. We fitted the negative exponential dispersal kernel, to the seed data, where is the density of seeds as a function of distance from the source, and describes the scale of the function (F igure 1). Given a dispersal kernel is the expected proportion of the total seedfall in an area at distance from the parent tree. Dispersal scale determines how fast seed density falls o ff with distance from the source. If a trees total fecundity is we find the expected number of seeds in a trap at distance by multiplying the dispersal kernel by the fecundity and the trap area and then dividing by the area of the annulus We assumed that observed seed numbers were
77 negative binomially distributed with mean N and dispersion parameter Although we used the negative exponential dispersal kernel, any number of dispersal functions could have fit the data, and some may have fit better than the negati ve exponential function. Our goal was to find a phenomenological model of the seed shadow to inform the seed sowing experiment, not to compare models or model parameters, and the negative exponential pe rformed well for monkeydispersed trees in a previous study (Clark et al. 2005). We calibrated parameters from the dispersa l function by searching numerically for the combination of parameter values that maximi zed the likelihood function using the default Nelder-Mead simplex algorithm in R 2.7.1 (R Development Team 2008). We estimated the scale parameter, to be 0.043, giving a mean disper sal distance of 23.3 m. Fecundity, was 77,135 fruits and the di spersion parameter, was estimated at 0.01 indicating strong overdispersion of seeds. Experimental Seed Distributions To examine the importance of patterns of seed dispersal fo r seedling recruitment, we experimentally planted seeds in three different distributions and monitored seedling recruitment and survival. Before sowing seeds, we first st aked out seven wedges at 45 from each other under seven M. mabokeensis trees (Fig 1). We define d a wedge as a sector (188 m2) of a circle, with its origin at the trunk, a cen tral angle of 6 and a radius of 60 m. Together, the three wedges covered 5% of the area within 60 m of the tree. We delimited the radius of the sector into 5 m sections, and planted seeds at the 12 different distance annuli. In three of the wedges, we monitored seedli ng recruitment derived from the natural seed rain; we did not manipulate seed shadows in these control we dges. In the four remaining wedges, we removed all conspecific seeds from the canopy floor immediately after the fruiting
78 season so that our experimental seed distribut ions were not confounded by previously arriving seeds. One of these served as a seed removal c ontrol and we monitored it to verify that we had effectively removed conspecific seeds. Across a ll seven trees, an average of only 1.3 seedlings germinated in all seed removal control wedges, none of which survived to the end of the experiment; therefore because we successfully removed backgr ound conspecific seeds we do not discuss the seed removal control wedge further. We sowed seeds in differe nt distributions in the remaining three wedges. In the no dispersal wedge, seeds were sown into the wedge in an aggregated distribution with al l seeds dropped within 10 m of the trunk. The no dispersal distribution mimics the scenario in which monkeys are extirpated and disper sal is limited to seed and fruitfall from the canopy. In the natural di spersal wedge, seeds were sown according to the negative exponential dispersal kernel derived from quantifying the seed shadow, where seed densities decrease with distance from the tree. In the even dispersal wedge, seeds were sown in a uniform distribution to simu late high rates of dispersal where seeds are evenly dispersed across the landscape. We used the fitted negativ e exponential dispersal f unction to calculate the density of seeds to sow at each distance annulus for each experimental distribution (Figure 1). By multiplying the density of seeds by the sa mpling area of the experimental wedge (188 m2), we calculated a seed sowing number of 881 seeds for each wedge. For the experimental wedges, we collected mature seeds and seeds from fruits from under the canopies of many different M. mabokeensis trees (~15 trees), mixing all the seeds together before sowing them. We wore plastic gloves wh en sowing the seeds to prevent leaving human odor and scattered the seeds on the ground within a wedge x distance block. We monitored seedling recruitment 1, 3, 6, 9, 12 and 18 months af ter the sowing of seeds, recording the number of seeds in each of the wedges and distance annuli. We also not ed the number of leaves on each
79 seedling and the number of leaves with signs of insect damage and herbivory. To determine whether the spatial distribution of seeds influences the diversity of the seedling community, we counted and identified to morphospecies all th e seedlings less than 50 cm in height in the experimental wedges at the end of the experiment. Although we were able to identify many individuals to species, our primary concern was to correctly classify the nu mber of individuals of different species in each dist ance by wedge combination. Analysis Our main interest was whether the spatial distribution of seeds determined seedling survival. We used survival analysis to anal yze the data and Bayesian inference with Markov Chain Monte Carlo (MCMC) simulation was used to estimate posterior distributions of model parameters and test for significance (see below fo r details). We assume d a Weibull distribution for the survivor function, and included random effects for the replicate trees and the experimental wedges within trees as follows: where is the failure time of an individual at tree and wedge In the data, time starts from the month after the seeds were sown and runs until month 18; we right-censored the data to account for seedlings that survived past the end of the experiment. The shape was allowed to vary by wedge with coefficient The scale was described by a vector of regression coefficients, where is the (log-) linear effect of distance, is the (log-) linear effect of seed density, and represents the effect of th e 3-level factor of treatment, which is the identify of the treatm ent. Because the effects of seed density and distance were modeled on
80 the log scale, their realized eff ects are exponential. The regressi on coefficients and the precision of the random effects were modeled with indepe ndent weak priors, with normally distributed priors for fixed and random effects and uniform priors on the precisions of the variance co mponents. So th at all parameter estimates were interpretable on the same scale, we standardized distance and density (both here and below) by subtracting the m ean and dividing by the standard deviation to yield a Z-score (Gelman and Hill 2007). We used generalized linear mixed models (G LMMs) to gain insight into the spatial mechanisms (distance from tree, density, expe rimental distribution) that drive seedling recruitment and survival by modeling three differe nt indicators of seed ling recruitment. To assess their effects on seedling recruitment, we modeled the proportion of sowed seeds that germinated as a logistic regression using a log it link. To assess their e ffects on seedling health and growth, we modeled the number of leaves pe r seedling using a log link and Poisson error distribution. To assess their effects on herbi vory, we modeled the pr oportion of leaves with herbivory damage as a logistic regression using a logit link. In all three tests we included tree and wedge within tree as random effects. Bayesian inference with Markov Chain Mont e Carlo (MCMC) simulation was used to estimate parameters and test for significance for the survival and GLMM models. We chose weak, proper prior distributions for all parameters (normal distributions with large variance) allowing the observed data to dominate inferen ces. We fit our models using the software WinBUGS v. 1.4.1 (Spiegel halter et al. 2003) run through th e R Language, version 2.7.1 (R Development Team 2008) using the contributed p ackage, R2WinBUGS (Sturtz et al. 2005). We ran three chains to monitor convergence and as sessed convergence by visual inspection and with
81 Gelman-Rubin statistics (i.e., Gelman-Rubin statis tics < 1.2) from the R contributed package, coda (Plummer et al. 2005). For the survival model, we ran 100,000 iterati ons (the burn-in) and based summary statistics on an additiona l 100,000 iterations. For the other models, the chains converged after 25,000 ite rations and we based summary statistics on the next 25,000 iterations. Parameter estimates (means) and their 95% credible intervals were obtained from the quantiles of the posterior dist ribution of model parameters. To assess the importance of parameters to our survival model, we tried to fit all possible candidate models and compare thei r deviance information criterion (DIC). For some models, we had difficulty achieving convergence of at least one parameter. Therefore, we instead ran the full survival model multiple times, each time setti ng a different parameter to 0 (i.e. effectively removing it from the model) to assess its contri bution to the model. We used DIC to compare model fits with lower DIC values indicating a be tter fit of a model to the data. To draw inferences from the GLMM models, we compare pos terior means and 95% credible intervals of the full models. To evaluate whether patterns of seed disper sal affect species diversity of the seedling community, we used Simpsons diversity index which takes the relative abundance of different sp ecies into account, where diversity is a function of the number of individuals of a species and the total number of individuals of all species Specified in this way, a value of 0 is no diversity and a value of 1 is in finite diversity. We calculated Simpsons diversity index for each we dge-by-distance combination, and then tested for differences in mean diversity by wedge to ex amine whether the spatial distribution of seeds affected species diversity. We compared mean diversity among wedges with a permutation test by shuffling the wedge identities ( no dispersal, natural disper sal, even dispersal) within
82 each tree and then calculated the average divers ity for each wedge type across all seven trees. We ran 1000 permutations and compar ed the observed mean diversity, for each wedge to the 2.5 and 97.5% quantiles from the simulated mean diversity, Results Of the 18,501 seeds sown in experimental we dges, 6.5% (1,203) seedlings emerged and 1% (188) survived to 18 months. Seedling morta lity was highest in the area closest to the tree for all seed distribut ions (Figure 3). Seedling survival was determined by the de nsity of seeds sown, with higher seedling survival at lower seed densities (Mean = -1.73, CI = -2.565, -0.9178; Figure 4). Survival increased slightly with distance from the tree, but the effect was indisti nguishable from 0 (Mean = 0.260, CI = -0.352, 0.822). Similarly, the overall eff ects of the experimental treatments did not differ strongly, suggesting that any remaining treatment effects not absorbed by the density and distance parameters were weak (Figure 4). The scale parameters differed among seed distributions (Mean = 391.2, df = 2, p < 2.2e-16), th e instantaneous risk of death or log-hazard (instantaneous rate of change in the log of the number of su rvivors per unit time) decreased rapidly for the no dispersal distribution because mortality occurred early, followed by the natural and good dispersal distributions (Figur e 4). Variation in su rvival was greater among trees than it was among experimental wedges under the same tree. Comparing candidate models (by consecutively setting each parameter to 0 an d comparing DIC values) similarly demonstrated that the distance and treatment pa rameters did not contribute to s eedling survival (Table 3-1). To determine the mechanisms driving differences in seedling survival, we examined patterns of seed germination, leaf production, and leaf damage from herbivores in the experimental wedges. The advantage of low seed density for survival did not accrue before the seedling stage, as seed germination did not str ongly vary with distance from tree, density of
83 sowed seeds, or experimental treatment (Table 3-2). However, seed density did affect the number of leaves per seedling, an indicator of seedling growth a nd health. The number of leaves per seedling decreased with seed density (Table 3-2), suggest ing that leaves are either being removed by vertebrate herbivores or that compet ition for resources with conspecific seedlings leads to a lower production of leaves. Of the leaves on seedlings, the probability of insect damage to leaves decreased by approximately 1% with every 10 m away from the tree (Mean = 0.493, CI = 0.488-0.497). The higher survival that comes from recruiti ng into an area with relatively low seed density means that more uniform dispersal patter ns will lead to higher seedling survival than aggregated patterns. By the end of our expe riment, 4.85 seedlings (boo tstrapped 95% CI = 1.71, 8.57) survived on average in th e good dispersal wedges, 3.86 (bootstrapped 95% CI = 1, 6.86) seedlings survived in natural wedges, a nd 0.86 seedlings (bootstrapped 95% CI = 0.14, 1.71) survived in no dispersal wedges. Scali ng up to the local neighborhood of the tree, good dispersal would result in 79.6 more seedlings than natural dispersal, a nd natural dispersal would result in 240.4 more seedlings than no dispersal. For M. mabokeensis our results suggest that the extirpation of monkeys by hunting would result in a 78% decrease in recruitment at the end of 18 months. More aggregated dispersal resulted in higher seedling dive rsity, although acrosstreatment differences in diversity were not si gnificantly different from random (no dispersal = 0.79 (sd = 0.23), = 0.76 (quantiles = 0.71, 0.81) ; natural dispersal = 0.76 (sd = 0.29), = 0.76 (quantiles = 0.70, 0.81), good dispersal = 0.74 (sd = 0.30), = 0.77 (quantiles = 0.71, 0.81)).
84 Discussion The spatial distribution of disp ersed seeds matters for seedling recruitment and survival. Compared to natural dispersal, good dispersa l increased seedling survival by 26%, whereas no dispersal reduced s eedling survival by 78%. Seed shadow s with more even distributions of seeds resulted in higher numbers of surviving se edlings. Survival of seedlings depended on the density of dispersed seeds but not the distance from the tree. These results support hypotheses that invoke density-dependence as a limiting factor to recruitmen t but fail to support the JanzenConnell hypothesis in its contenti on that enhanced distance of seeds from conspecific adults improves survival (Hyatt et al. 2003). The shape of the seed shadow did not affect seed germination and seedling emergence, but had strong effects on seedling health and surv ival. The benefits of more even dispersal patterns likely arose through escap ing herbivores that responded to seedling density. In a concurrent study, we caged seed addition plots of M. mabokeensis to protect seeds and seedlings from rodents and vertebrate herbivores (Poulsen and Clark unpubl. data). Compared to uncaged plots, seedling survival in caged plots increa sed by 2.6 times after 3 months and increased by 4.9 times after 24 months. Janzen (1970) and Connell (1971) hypothesized that seeds di spersed away from the high densities of seeds aggregated under the parent canopy would avoi d mortality from densityand distance-dependent behavior of seed and seedling predators. Previ ous studies have verified that higher seed or seedling density and greater proximity to the pa rent plant often lower seedling recruitment (Augspurger and Kelly 1984, Clark and Clark 1984, Howe et al. 1985, Augspurger and Kitajima 1992). Here we extended these fi ndings by using a replicated design to decouple and quantify the effects of distance and density on seedling survival and the overall advantage of different dispersal patterns. We demonstrated effects of seed density on seedling recruitment and
85 survival, but found no evidence of distance effect s. The distance component of Janzen and Connells hypothesis has received little support (Hyatt et al. 2003) and likely oversimplifies the interaction between seeds, seedlings and predat ors which depends on the scale of movement of both offspring and their enemies (A dler and Muller-Landau 2005). From a theoretical perspective, we expected that better (more uni form) dispersal would lead to lower community richness by increa sing conspecific seedling establishment to the exclusion of heterospecifics. Connell (1971) argued that if propagule distan ce or density were to impact diversity, it should be seen most strongl y in individual species at the seedling stage. Similarly, differences in diversity should be mo st detectable by comparing seed shadows that represent the extremes of seed dispersal patterns: only dispersal from seeds falling from the tree (no dispersal) and uniform rates of dispersal ( good dispersal). While we did find an inverse trend between dispersal and seed ling diversity, it was not strong enough to conc lude that better dispersal would lead to lower diversity. Dispersal limitation may be less important for maintenance of species diversity than other mech anisms, such as density-dependence and habitat partitioning (Webb and Peart 2001). Alternatively, it is likely that interactions, particularly competition, among seedlings is too weak to driv e patterns of diversity (e.g., Svenning et al. 2008). Our experiment consisted of a single di spersal season with di fferent seed shadows resulting in dramatically di fferent numbers of recruiting and surviving seedlings. We hypothesize that spatial patterns of seedlings that accumulate over several seasons would have a much more profound effect on the di versity of the seedling community. From a conservation perspective, our resu lts from one monkey-dispersed tree species demonstrate that the loss of di spersers could negatively affect the regeneration success of tropical trees. Once protected by their remoteness, 30-45% of Afrotropical forests are now
86 occupied by logging concessions, exposing an ad ditional 29% of forests to increased hunting pressure (Laporte et al. 2007). Compared to our study site, densities of frugivorous monkeys were 35% lower in adjacent forest subjected to moderate (1 person km-2) subsistence hunting (Poulsen et al. 2009b). Hunting markedly alters the structure of mammal communities in central Africa (Fa et al. 2005, Laurance et al. 2006, Clark et al. 2009) and threatens to leave many tree species without reliable dispersers. The loss or re duction of seed dispersal services as a result of widespread hunting will leave so me tree species with limited or no regeneration, thus altering forest composition with time. To maintain func tioning forests, measures need to be taken to protect forest animals.
87 Table 3-1. Model comparison for survival analys is of seedlings. A low deviance information criterion (DIC) value indicates that the model fits the data better than a higher value. To assess the contribution of parameters to the survival model, we ran the model multiple times, each time setting one of the parameters to 0 (i.e. effectively removing it from the model). A parameter improves model fit when its removal from the model causes the DIC value to increas e; the removal of an important parameter should cause the model to get worse. The scale for a ssessing DIC is approximately equal to the scale for the Akaike information criteri on (AIC): <2 = little difference, 2-6 = moderate, 6-10 = substantial, and >10 = overwhelming (pg. 210, Bolker 2008). Model DIC Full model No -2 No +16 No -1 No +2 No +32
88 Table 3-2. Parameters [and 95% credible interv als] from generalized linear mixed models (GLMMs) for seed germination, leaf growth on seedlings, and leaf damage. Density of seeds represents the difference in density from the average natural seed density. Distance from the tree represents the eff ect of moving 10 m farther from the tree. Random effects for tree and the experime ntal distribution are presented as and Parameters Probability of seed germination Log ratio of leaves per seedling Probability of leaf damage Density of seeds 0.522 [0.488, 0.555] -0.127 [-0.215, -0.043] 0.521 [0.468-0.573] Distance from tree 0.502 [0.467, 0.536] 0.051 [-0.006, 0.108] 0.434 [0.392-0.476] Aggregated distribution 0.008 [0.002, 0.042] -0.506 [-0.844, -0.159] 0.054 [0.028-0.101] Natural distribution 0.011 [0.003, 0.055] -0.308 [-0.538, -0.086] 0.058 [0.039-0.086] Uniform distribution 0.010 [0.003, 0.051] -0.330 [-0.547, -0.107] 0.064 [0.042-0.095] 0.657 [0.593, 0.755] 0.182 [0.074, 0.336] 0.592 [0.533-0.668] 0.795 [0.633, 0.956] 0.135 [0.012, 0.375] 0.545 [0.501-0.640]
89 Figure 3-1. Depiction of the expe rimental design. Seven wedges were delineated under each of seven adult individuals of Manilkara mabokeensis We defined a wedge as a sector (188 m2) of a circle, its origin at the trunk with a central an gle of 18 and radii of 60 m. We delimited the radius of the sector into 5 m sections, and planted seeds at the 12 different distance annuli. In three of the wedge s (Non-experimental), we monitored seedling recruitmen t derived from the natural seed rain. In the four remaining wedges, we removed all the s eeds from the canopy floor. One of these served as a seed removal control (Seed removal). We sowed seeds in different distributions in the remaining three: 1) In the No dispersal wedge seeds were sown within 10 m of the trunk (e.g., under the canopy of the tree); 2) In the Natural dispersal wedge seeds were sown with d ecreasing densities from the tree following quantification of the seed sh adow; and 3) In the Even dispersal wedge, seeds were sown at uniform densities with distance from the tree.
90 Figure 3-2. Fit of the negative expo nential function to seed trap da ta from four individuals of Manilkara mabokeensis Closed circles show seed dens ities in traps (on a logarithmic scale) in relation to distance from tree (N=4 ). Mean parameter values for the negative exponential dispersal function were a= 0.44 (scale), f = 77,222 (fecundity), and k = 0.01 (dispersion).
91 Figure 3-3. The density of seedlings surviving with distance from tree and in the different experimental and non-experime ntal wedges. The dark li nes and dark green points represent the density of s eedlings 1 month (top line) af ter seeds were sown and 18 months (bottom line) after sowing. Grey lines and dots indicate seedling densities during the months between the firs t and last seedlings censuses.
92 Figure 3-4. Violin plot of parameters from th e seedling survival model, combining a box plot indicating the median (point) and interquartile range, with the probability density of the posterior distribution of parameters. Note that the distance parameter overlaps zero, whereas the density parameter does not. Similarly, shape parameters are significantly different among tr eatments, but the experimental treatments parameters do not.
93 Figure 3-5. Number of seedlings that survived over 18 months in the three experimental distributions (No dispersal, Natural di spersal, and Good dispersal), the nonexperimental distribution (N on-experimental), and the s eed removal control (Seed removal).
94 LIST OF REFERENCES Adler, F. R. and H. C. Muller-Landau. 2005. When do localized natural enemies increase species richness? Ecology Letters 8 :438-447. Asner, G. P., E. N. Broadbent, P. J. C. Oliveira M. Keller, D. E. Knapp, and J. N. M. Silva. 2006. Condition and fate of logged forests in the Brazilian Amazon. Proceedings of the National Academy of Sciences of the United States of America 103 :12947-12950. Asner, G. P., M. Keller, R. Pereira, J. C. Zweede, and J. N. M. Silva. 2004. Canopy damage and recovery after selective logging in Amazonia: Field and satellite studies. Ecological Applications 14 :S280-S298. Asner, G. P., D. E. Knapp, E. N. Broadbent, P. J. C. Oliveira, M. Keller, and J. N. Silva. 2005. Selective logging in the Brazilian Amazon. Science 310 :480-482. Asquith, N. M., S. J. Wright, and M. J. Clauss. 1997. Does mammal community composition control recruitment in neotropical fo rests? Evidence from Panama. Ecology 78 :941-946. Augspurger, C. K. and C. K. Kelly. 1984. Pat hogen mortality of tropical tree seedlings experimental studies of the effects of di spersal distance, seedling density, and light conditions. Oecologia 61 :211-217. Augspurger, C. K. and K. Kitajima. 1992. Experi mental studies of seedling recruitment from contrasting seed di stributions. Ecology 73 :1270-1284. Balcomb, S. R. and C. A. Chapman. 2003. Bridgi ng the gap: Influence of seed deposition on seedling recruitment in a primate-tree interaction. Ec ological Monographs 73 :625-642. Blake, S., S. Strindberg, P. B oudjan, C. Makombo, I. Bila-Isia, O. Ilambu, F. Grossmann, L. Bene-Bene, B. de Semboli, V. Mbenzo, D. S'Hwa, R. Bayogo, L. Williamson, M. Fay, J. Hart, and F. Maisels. 2007. Forest elepha nt crisis in the Congo Basin. Plos Biology 5 :945953. Bohrer, G., G. G. Katul, R. Nathan, R. L. Walko, and R. Avissar. 2008. Effects of canopy heterogeneity, seed abscission and inertia on wi nd-driven dispersal kernels of tree seeds. Journal of Ecology 96 :569-580. Bolker, B. M. 2008. Ecological models and data in R. Princeton University Press, Princeton, N.J. Bolker, B. M. and S. W. Pacala. 1999. Spa tial moment equations for plant competition: Understanding spatial strategies and the advantag es of short dispersal. American Naturalist 153 :575-602. Broadbent, E. N., G. P. Asner, M. Keller, D. E. Knapp, P. J. C. Oliveira, and J. N. Silva. 2008. Forest fragmentation and edge effects from deforestation and se lective logging in the Brazilian Amazon. Biological Conservation 141 :1745-1757.
95 Brodie, J. F., O. E. Helmy, W. Y. Brocke lman, and J. L. Maron. 2009. Bushmeat poaching reduces seed dispersal and population growth ra te of a mammal disper sed tree. Ecological Applications 19 :854-863. Brook, B. W., C. J. A. Bradshaw, L. P. Koh, and N. S. Sodhi. 2006. Momentum drives the crash: Mass extinction in the tropics. Biotropica 38 :302-305. Brook, B. W., N. S. Sodhi, and P. K. L. Ng. 2003. Catastrophic extinctio ns follow deforestation in Singapore. Nature 424 :420-423. Buckland, S. T., D. R. Anderson, K. P. Burnham, J. L. Laake, D. L. Borchers, and L. Thomas. 2001. Introduction to distance sampling: Estim ating abundance of biological populations. Oxford University Press, Oxford, UK. Chapman, C. A. and D. A. Onderdonk. 1998. Fo rests without primates: Primate/plant codependency. Pages 127-141. Choler, P. 2005. Consistent shifts in Alpine plant traits along a mesotopographical gradient. Arctic Antarctic a nd Alpine Research 37 :444-453. Clark, C. J., J. R. Poulsen, B. M. Bolker, E. F. Connor, and V. T. Parker. 2005. Comparative seed shadows of bird-, monkey-, and wind-dispersed trees. Ecology 86 :2684-2694. Clark, C. J., J. R. Poulsen, D. J. Levey, and C. W. Osenberg. 2007. Are plant populations seed limited? A critique and meta-analysis of seed addition experiments. American Naturalist 170 :128-142. Clark, C. J., J. R. Poulsen, R. Malonga, and P. W. Elkan. 2009. Logging concessions can extend the conservation estate for Central Afri can tropical forests. Conservation Biology. Clark, C. J., J. R. Poulsen, and V. T. Parker. 2001. The role of arboreal seed dispersal groups on the seed rain of a lowland tropical forest. Biotropica 33 :606-620. Clark, D. A. and D. B. Clark. 1984. Spacing dynamics of a tropical rain-for est tree Evaluation of the Janzen-Connell model. American Naturalist 124 :769-788. Clark, J. S. 1998. Why trees migrate so fast: Conf ronting theory with dispersal biology and the paleorecord. American Naturalist 152 :204-224. Clark, J. S., M. Silman, R. Kern, E. Macklin, an d J. HilleRisLambers. 1999. Seed dispersal near and far: Patterns across temperat e and tropical forests. Ecology 80 :1475-1494. Cleary, D. F. R., T. J. B. Boyle, T. Setyawati, C. D. Anggraeni, E. E. Van Loon, and S. B. J. Menken. 2007. Bird species and traits associ ated with logged and unlogged forest in Borneo. Ecological Applications 17 :1184-1197.
96 Condit, R., P. S. Ashton, P. Baker, S. Bunyave jchewin, S. Gunatilleke, N. Gunatilleke, S. P. Hubbell, R. B. Foster, A. Itoh, J. V. LaFranki e, H. S. Lee, E. Losos, N. Manokaran, R. Sukumar, and T. Yamakura. 2000. Spatial patter ns in the distribution of tropical tree species. Science 288 :1414-1418. Congolaise Industrielle des Bois. 2006. Plan damenagement de lunit forestire damnagement de Kabo (20052034). Minist ry of Forest Economy, Brazzaville. Connell, J. H. 1971. On the role of natural enemies in preventing competitive exclusion in some marine animals and rain forest trees. Pages 298-312 in P. J. D. Boer and G. R. Gradwell, editors. Dynamics of populations. Center fo r Agricultural Publica tion and Documentation, Wageningen. Curran, L. M., S. N. Trigg, A. K. McDonald, D. As tiani, Y. M. Hardiono, P. Siregar, I. Caniago, and E. Kasischke. 2004. Lowland forest loss in protected areas of Indonesian Borneo. Science 303 :1000-1003. Doledec, S., D. Chessel, C. J. F. terBraak, and S. Champely. 1996. Matc hing species traits to environmental variables: A new three-ta ble ordination method. Environmental and Ecological Statistics 3 :143-166. Dray, S., A. B. Dufour, and D. Chessel. 2007. The ade4 package-II: Two-table and K-table methods R News 7 :47-52. Dray, S., N. Pettorelli, and D. Chessel. 2002. Ma tching data sets from two different spatial samples. Journal of Vegetation Science 13 :867-874. Durrieu de Madron, L., B. Fontez, and B. Dipapounji. 2000. Dgts d exploitation et de dbardage en fonction de lintensit de xploitation en fort de nse humide d'Afrique Centrale. Bois et Forts des Tropiques 264 :57-60. Fa, J. E., D. Currie, and J. Meeuwig. 2003. Bu shmeat and food security in the Congo Basin: linkages between wildlife and people's future. Environmental Conservation 30 :71-78. Fa, J. E., S. F. Ryan, and D. J. Bell. 2005. H unting vulnerability, ecologi cal characteristics and harvest rates of bushmeat species in afro tropical forests. Bi ological Conservation 121 :167176. Frazer, G. W., R. A. Fournier, J. A. Trofymow, and R. J. Hall. 2001. A comparison of digital and film fisheye photography for an alysis of forest canopy struct ure and gap light transmission. Agricultural and Fo rest Meteorology 109 :249-263. Gardner, T. A., J. Barlow, L. W. Parry, and C. A. Peres. 2007. Predicting the uncertain future of tropical forest species in a data vacuum. Biotropica 39 :25-30.
97 Gautier-Hion, A., J.-M. Duplantier, R. Quris, F. Feer, C. Sourd, J.-P. Decoux, G. Dubost, L. Emmons, C. Erard, P. Hecketsweiler, A. Moungazi, C. Roussilhon, and J.-M. Thiollay. 1985. Fruit characters as a basis of fruit choi ce and seed dispersal in a tropical forest vertebrate community. Oecologia 65 :324-337. Gelman, A. and J. Hill. 2007. Data analysis us ing regression and multilevel/hierarchical models. Cambridge University Press, Cambridge. Godoy, J. A. and P. Jordano. 2001. Seed dispersal by animals: exact identification of source trees with endocarp DNA microsatel lites. Molecular Ecology 10 :2275-2283. Greene, D. F. 2005. The role of abscission in lo ng-distance seed dispersal by the wind. Ecology 86 :3105-3110. Greene, D. F., C. D. Canham, K. D. Coates, an d P. T. Lepage. 2004. An evaluation of alternative dispersal functions for trees. Journal of Ecology 92 :758-766. Greene, D. F. and E. A. Johnson. 1989. A model of wind dispersal of winged or plumed seeds. Ecology 70 :339-347. Greene, D. F., M. Quesada, and C. Calogeropo ulos. 2008. Dispersal of seeds by the tropical sea breeze. Ecology 89 :118-125. Hardesty, B. D., S. P. Hubbell, and E. Berm ingham. 2006. Genetic evidence of frequent longdistance recruitment in a vertebrate-dispersed tree. Ecology Letters 9 :516-525. Harms, K. E., S. J. Wright, O. Calderon, A. Hernandez, and E. A. Herre. 2000. Pervasive density-dependent recruitment enhances seedli ng diversity in a trop ical forest. Nature 404 :493-495. Harris, D. J. and A. H. Wortley. 2008. Sangha tr ees: an illustrated iden tification manual. Royal Botanic Gardens, Edinburgh, Scotland. Hausner, V. H., N. G. Yoccoz, and R. A. Ims. 2003. Selecting indicator tr aits for monitoring land use impacts: Birds in northern coastal birch forests. Ecological Applications 13 :999-1012. Hawthorne, W. and N. Gyakari. 2006. Photoguide fo r the forest trees of Ghana: A tree-spotter's field guide for identifying the largest trees. Oxford Forestry Institute, Oxford. Herrera, C. M., P. Jordano, L. Lopezsoria, and J. A. Amat. 1994. Recruitment of a mast-fruiting, bird-dispersed tree bridging frugivore activ ity and seedling establishment. Ecological Monographs 64 :315-344. Hilborn, R. and M. Mangel. 1997. The ecological detective: confronti ng models with data. Princeton University Press, Princeton, N.J. Holbrook, K. M. and T. B. Smith. 2000. Seed disper sal and movement pattern s in two species of Ceratogymna hornbills in a West African tropical lowland forest. Oecologia 125 :249-257.
98 Holbrook, K. M., T. B. Smith, and B. D. Ha rdesty. 2002. Implications of long-distance movements of frugivorous rain forest hornbills. Ecography 25 :745-749. Hothorn, T., F. Bretz, P. Westfall, and M. Heiberger. 2008. multcomp: Simultaneous Inference in General Parametric Models. Howe, H. F., E. W. Schupp, and L. C. Westley. 1985. Early conseque nces of seed dispersal for a Neotropical tree (Virola-surinamensis). Ecology 66 :781-791. Hubbell, S. P. 1979. Tree dispersion, abundance, a nd diversity in a tropical dry forest. Science 203 :1299-1309. Hubbell, S. P., R. B. Foster, S. T. O'Brien, K. E. Harms, R. Condit, B. Wechsler, S. J. Wright, and S. L. de Lao. 1999. Light-gap disturbances, recruitment limitation, an d tree diversity in a neotropical forest. Science 283 :554-557. Hyatt, L. A., M. S. Rosenberg, T. G. Howard, G. Bole, W. Fang, J. Anastasia, K. Brown, R. Grella, K. Hinman, J. P. Kurdziel, an d J. Gurevitch. 2003. The distance dependence prediction of the Janzen-Connell hypot hesis: a meta-analysis. Oikos 103 :590-602. Janzen, D. H. 1970. Herbivores and number of tree species in tropical forests. American Naturalist 104 :501-&. John, R., J. W. Dalling, K. E. Harms, J. B. Yavitt, R. F. Stallard, M. Mirabello, S. P. Hubbell, R. Valencia, H. Navarrete, M. Vallejo, and R. B. Foster. 2007. Soil nutrien ts influence spatial distributions of tropical tree species. Proceedi ngs of the National Academy of Sciences of the United States of America 104 :864-869. Kuhnert, P. M., T. G. Martin, K. Mengersen, an d H. P. Possingham. 2005. Assessing the impacts of grazing levels on bird density in woodland habitat: a Bayesian approach using expert opinion. Environmetrics 16 :717-747. Lambert, D. 1992. Zero-inflated Poisson regre ssion, with an application to defects in manufacturing. Technometrics 34 :1-14. Laporte, N. T., J. A. Stabach, R. Grosch, T. S. Lin, and S. J. Goetz. 2007. Expansion of industrial logging in Central Africa. Science 316 :1451-1451. Laurance, W. F. 2001. The future of the Brazilian Amazon. Science 291 :988-988. Laurance, W. F. 2007. Have we overstated the tropi cal biodiversity crisis ? Trends in Ecology & Evolution 22 :65-70. Laurance, W. F., B. M. Croes, N. Guiss ouegou, R. Buij, M. Dethier, and A. Alonso. 2008. Impacts of roads, hunting, and habitat al teration on nocturnal mammals in African rainforests. Conservation Biology 22 :721-732.
99 Laurance, W. F., B. M. Croes, L. Tchignoumba, S. A. Lahm, A. Alonso, M. E. Lee, P. Campbell, and C. Ondzeano. 2006. Impacts of roads and hunting on central African rainforest mammals. Conservation Biology 20 :1251-1261. Lesica, P. and S. V. Cooper. 1999. Succession and disturbance in sandhills vegetation: Constructing models for managing bi ological diversity. Conservation Biology 13 :293-302. Levey, D. J., B. M. Bolker, J. J. Tewksbury, S. Sargent, and N. M. Haddad. 2005. Effects of landscape corridors on seed di spersal by birds. Science 309 :146-148. Levine, J. M. and D. J. Murrell. 2003. The comm unity-level consequences of seed dispersal patterns. Annual Review of Ecology Evolution and Systematics 34 :549-574. MacKenzie, D. I., J. D. Nichols, G. B. Lachma n, S. Droege, J. A. Royle, and C. A. Langtimm. 2002. Estimating site occupancy rates when de tection probabilities are less than one. Ecology 83 :2248-2255. Martin, T. G., P. M. Kuhnert, K. Mengersen, a nd H. P. Possingham. 2005a. The power of expert opinion in ecological models using Bayesian methods: Imp act of grazing on birds. Ecological Applications 15 :266-280. Martin, T. G., B. A. Wintle, J. R. Rhodes, P. M. Kuhnert, S. A. Field, S. J. Low-Choy, A. J. Tyre, and H. P. Possingham. 2005b. Zero tolerance ecology: improving ecological inference by modelling the source of zero observations. Ecology Letters 8 :1235-1246. McConkey, K. R. and D. R. Drake. 2006. Flying f oxes cease to function as seed dispersers long before they become rare. Ecology 87 :271-276. Milner-Gulland, E. J., E. L. Bennett, and S. C. B. A. m. W. Meat. 2003. Wild meat: the bigger picture. Trends in Ecology & Evolution 18 :351-357. Mockrin, M. H. 2008. The spatial st ructure and sustainability of subsistence wildlife harvesting in Kabo, Congo. PhD Dissertation. Columbia University, Columbia, New York. Morales, J. M. and T. S. Carlo. 2006. The eff ects of plant distributi on and frugivore density on the scale and shape of di spersal kernels. Ecology 87 :1489-1496. Morgan, D. and C. Sanz. 2006. Chimpanzee feeding ecology and comparisons with sympatric gorillas in the Goualougo Triangl e, Republic of Congo. Pages 97-122 in G. Hohmann, M. Robbins, and C. Boesch, editors. Primate f eeding ecology in apes and other primates: ecological, physiological, and behavioural aspects. Cambridge University Press, Cambridge. Muchaal, P. K. and G. Ngandjui. 1999. Impact of village hunting on wildli fe populations in the western Dia Reserve, Ca meroon. Conservation Biology 13 :385-396.
100 Muller-Landau, H. C., S. J. Wright, O. Ca lderon, R. Condit, and S. P. Hubbell. 2008. Interspecific variation in prim ary seed dispersal in a tropi cal forest. Journal of Ecology 96 :653-667. Nathan, R., G. G. Katul, H. S. Horn, S. M. Thomas R. Oren, R. Avissar, S. W. Pacala, and S. A. Levin. 2002. Mechanisms of long-distan ce dispersal of seeds by wind. Nature 418 :409413. Nathan, R. and H. C. Muller-Landau. 2000. Spatial pa tterns of seed dispersal, their determinants and consequences for recruitment. Trends in Ecology & Evolution 15 :278-285. Nathan, R., U. N. Safriel, and I. Noy-Meir. 2001. Field validation and sensitivity analysis of a mechanistic model for tree s eed dispersal by wind. Ecology 82 :374-388. Nepstad, D. C., A. Verissimo, A. Alencar, C. Nobr e, E. Lima, P. Lefebvre, P. Schlesinger, C. Potter, P. Moutinho, E. Mendoza, M. Co chrane, and V. Brooks. 1999. Large-scale impoverishment of Amazonian fore sts by logging and fire. Nature 398 :505-508. Paine, C. E. T. and H. Beck. 2007. Seed predatio n by neotropical rain forest mammals increases diversity in seedling recruitment. Ecology 88 :3076-3087. Peres, C. A. 2000. Effects of subsistence hun ting on vertebrate community structure in Amazonian forests. Conservation Biology 14 :240-253. Peres, C. A. and E. Palacios. 2007. Basin-wide effects of game harvest on verteb rate population densities in Amazonian forests: Implications for animal-mediated seed dispersal. Pages 304-315. Plummer, M., N. G. Best, N. K. Cowles, a nd S. K. Vines. 2005. coda: output analysis and diagnostics for MCMC. R package, version 0.9-5. Poulsen, J. R., C. J. Clark, and B. M. Bolker. 2009a. Logging and hunting a lter patterns of seed dispersal and seedling recruitment. Un iversity of Florida, Gainesville. Poulsen, J. R., C. J. Clark, E. F. Connor, a nd T. B. Smith. 2002. Differential resource use by primates and hornbills: Implicati ons for seed dispersal. Ecology 83 :228-240. Poulsen, J. R., C. J. Clark, G. Mavah, a nd P. W. Elkan. 2009b. Bushmeat supply and consumption in a tropical logging conce ssion in northern Congo. Conservation Biology. Poulsen, J. R., C. J. Clark, and T. B. Smith. 2001. Seed dispersal by a di urnal primate community in the Dja Reserve, Cameroon. Journal of Tropical Ecology 17 :787-808. Poulsen, J. R., C. W. Osenberg, C. J. Clark, D. J. Levey, and B. M. Bolker. 2007. Plants as reef fish: Fitting the functional form of seed ling recruitment. American Naturalist 170 :167-183.
101 Putz, F. E., G. M. Blate, K. H. Redford, R. Fimbel, and J. Robinson. 2001. Tropical forest management and conservation of biodiver sity: an overview. Conservation Biology 15 :720. R Development Core Team. 2008. R: a language and environment for statistical computing. Version 2.7.1. R Foundation for St atistical Computing, Vienna. Ribbens, E., J. A. Silander, and S. W. Pacala. 1994. Seedling recruitment in forests Calibrating models to predict patterns of tree seedling dispersion. Ecology 75 :1794-1806. Ribera, I., S. Doledec, I. S. Downie, and G. N. Foster. 2001. Effect of la nd disturbance and stress on species traits of ground beetle assemblages. Ecology 82 :1112-1129. Robinson, J. G., K. H. Redford, and E. L. Bennett. 1999. Conservation Wildlife harvest in logged tropical forests. Science 284 :595-596. Russo, S. E. and C. K. Augspurger. 2004. Aggreg ated seed dispersal by spider monkeys limits recruitment to clumped patterns in Virola calophylla. Ecology Letters 7 :1058-1067. Russo, S. E., S. Portnoy, and C. K. Augspurger. 2006. Incorporating animal behavior into seed dispersal models: Implications for seed shadows. Ecology 87 :3160-3174. Sagnard, F., C. Pichot, P. Dreyfus, P. Jorda no, and B. Fady. 2007. Modelling seed dispersal to predict seedling recruitment: Recolonization dynamics in a plantation forest. Ecological Modelling 203 :464-474. Schippers, P. and E. Jongejans. 2005. Release thre sholds strongly determine the range of seed dispersal by wind. Ec ological Modelling 185 :93-103. Schupp, E. W. and M. Fuentes. 1995. Spatial patte rns of seed dispersal and the unification of plant-population ecology. Ecoscience 2 :267-275. Seidler, T. G. and J. B. Plotkin. 2006. Seed disper sal and spatial pattern in tropical trees. Plos Biology 4 :2132-2137. Soares, B. S., D. C. Nepstad, L. M. Curran, G. C. Cerqueira, R. A. Garcia, C. A. Ramos, E. Voll, A. McDonald, P. Lefebvre, and P. Schles inger. 2006. Modelling conservation in the Amazon basin. Nature 440 :520-523. Spiegelhalter, D. J., A. Thoomas, N. G. Best, and D. Lunn. 2003. WinBUGS Version 1.4. Imperial College and MRC Biostatistics Unit: UK. Sturtz, S., U. Ligges, and A. Gelman. 2005. R2WinBUGS: a package for running WinBUGS from R. Journal of Statistical Software 12 :1-16. Svenning, J. C., T. Fabbro, and S. J. Wright. 2008. S eedling interactions in a tropical forest in Panama. Oecologia 155 :143-150.
102 Tenenhaus, M. and F. W. Young. 1985. An analysis and synthesis of multiple correspondance analysis, optimal scaling, dual scaling, ho mogeneity analysis and other methods for quantifying categorical multivariate data. Psychometrika 50 :91-119. Terborgh, J., G. Nunez-Iturri, N. C. A. Pitman, F. H. C. Valverde, P. Alvarez, V. Swamy, E. G. Pringle, and C. E. T. Paine. 2008. Tree recruitment in an empty forest. Ecology 89 :17571768. Thomas, L., J. L. Laake, S. Strindberg, F. F. C. Marques, S. T. Buckland, D. L. Borchers, D. R. Anderson, K. P. Burnham, S. L. Hedley, J. H. Pollard, J. R. B. Bishop, and T. A. Marques. 2006. Distance 5.0. Research Unit for Wildlife P opulation Assessment, University of St. Andrews, St. Andrews, UK. Tufto, J., S. Engen, and K. Hindar. 1997. Stochas tic dispersal processe s in plant populations. Theoretical Population Biology 52 :16-26. Turchin, P. 1998. Quantitative Analysis of M ovement: Measuring and Modeling Population Redistribution in Animals and Pl ants. Sinauer, Sunderland, MA. Tutin, C. E. G., R. M. Ham, L. J. T. White, and M. J. S. Harrison. 1997. The primate community of the Lop Reserve, Gabon: Diets, responses to fruit scarcity, and effects on biomass. American Journal of Primatology 42 :1-24. Tyre, A. J., B. Tenhumberg, S. A. Field, D. Niejalke, K. Parris, and H. P. Possingham. 2003. Improving precision and reducing bi as in biological surveys: Estimating false-negative error rates. Ecological Applications 13 :1790-1801. Van Gemerden, B. S., G. N. Shu, and H. Olff. 2003. Recovery of conservation values in Central African rain forest after l ogging and shifting cultivation. Bi odiversity and Conservation 12 :1553-1570. Walsh, P. D. and L. J. T. White. 2005. Evaluati ng the steady state assump tion: Simulations of gorilla nest decay. Ecological Applications 15 :1342-1350. Wang, B. C. and T. B. Smith. 2002. Closing the seed dispersal loop. Trends in Ecology & Evolution 17 :379-385. Wang, B. C., V. L. Sork, M. T. Leong, and T. B. Smith. 2007. Hunting of mammals reduces seed removal and dispersal of the afrotropical tree Antrocaryon klaineanum (Anacardiaceae). Pages 340-347. Webb, C. O. and D. R. Peart. 2001. High seed di spersal rates in fauna lly intact tropical rainforest: theoretical and conserva tion implications. Ecology Letters 4 :491-499. Westcott, D. A. and D. L. Graham. 2000. Patterns of movement and seed di spersal of a tropical frugivore. Oecologia 122 :249-257.
103 White, L. J. T. and K. Abernathy. 1997. A guide to the vegetation of the Lope Reserve, Gabon. Ecofac Gabon, Libreville. White, L. J. T., C. E. G. Tutin, and M. Fern andez. 1993. Group composition and diet of forest elephants, Loxodonta africana cyclotis matschie 1900, in the Lop Reserve, Gabon. African Journal of Ecology 31 :181-199. Whitney, K. D., M. K. Fogiel, A. M. Lamperti, K. M. Holbrook, D. J. Stauffer, B. D. Hardesty, V. T. Parker, and T. B. Smith. 1998. Seed di spersal by Ceratogymna hornbills in the Dja Reserve, Cameroon. Journal of Tropical Ecology 14 :351-371. Wilkie, D. S., J. G. Sidle, G. C. Boundzan a, S. Blake, and P. Auzel. 2001. Defaunation, not deforestation: commercial l ogging and market hunting in northern Congo. Pages 375-399 in R. A. Fimbel, A. Grajal, and J. G. Robinson, editors. The cutting edge: conserving wildlife in logged tr opical forest. Columbia Univ ersity Press, New York, NY. Wright, S. J. 2002. Plant diversity in tropical forests: a review of mechanisms of species coexistence. Oecologia 130 :1-14. Wright, S. J. and H. C. Duber. 2001. Poachers and fo rest fragmentation alter seed dispersal, seed survival, and seedling recruitment in the palm Attalea butyraceae, with implications for tropical tree diversity. Biotropica 33 :583-595. Wright, S. J., A. Hernandez, and R. Condit. 2007a. The bushmeat harvest alters seedling banks by favoring lianas, large seeds, and seeds dispersed by bats, birds, and wind. Pages 363371. Wright, S. J. and H. C. Muller-Landau. 2006a. The fu ture of tropical forest species. Biotropica 38 :287-301. Wright, S. J. and H. C. Muller-Landau. 2006b. The un certain future of tropical forest species. Biotropica 38 :443-445. Wright, S. J., K. E. Stoner, N. Beckman, R. T. Corlett, R. Dirzo, H. C. Muller-Landau, G. Nunez-Iturri, C. A. Peres, and B. C. Wang. 2007b. The plight of large animals in tropical forests and the consequences for plant regeneration. Pages 289-291. Wright, S. J., H. Zeballos, I. Dominguez, M. M. Gallardo, M. C. Moreno, and R. Ibanez. 2000. Poachers alter mammal abundance, seed dispersa l, and seed predation in a neotropical forest. Conservation Biology 14 :227-239.
104 BIOGRAPHICAL SKETCH John Poulsen was born in Texas. Having su rvived the Army and finished their own Ph.D.s, his parents, Randy and Dorothy, move d him and his twin brother, Chris, to the mountains of Montana at the age of four. In Mo ntana, he eventually learned how to flyfish and spent his childhood running around the mountains with Chris and his little si ster, Melissa. After high school, John left behind small town life to join the Air Force Academy. Finding himself very good at following orders, but not so enthusia stic about killing people, after nine months John transferred to Willamette University wher e he studied internati onal politics. During college, John returned to Montana in the summers to work for the U.S. Forest Service as a forest fire fighter and lear ned important life skills li ke using chainsaws and digg ing fire line that would serve him in his future endeavors. It was in hi s last year at Willamette that John met and fell in love with Connie Clark. Searching for an in ternational experience, John worked for the US Peace Corps as a forestry volunteer in the Republic of Mali. In Mali he learned how to adapt and live in African cultures, learned to speak Fulf ulde and French, and le arned to contract and survive infectious diseases. After two and a ha lf years John traveled to the rainforests of Cameroon where Connie had abandoned him to conduct research on hornbills. Reunited with Connie, he also fell in love w ith the tropical forest and gave up his ambitions to work in international politics to pursue ecology. Three years later John finished his masters degree in conservation biology at San Francisc o State University. He then went to work for the Wildlife Conservation Society in the Re public of Congo as the co-director of the Lac Tl Community Reserve. Enjoying conservation work, but disa ppointed at not seeing Mokel-Mbemb, the dinosaur said to inhabit Lake Tl, John left to start his Ph.D. at the Un iversity of Florida in 2002. He conducted three years of field work in northern Republic of Congo while working as the director of the Buffer Zone Project (P ROGEPP) for the Wildlife Conservation Society.
105 Ironically, his experience in the Air Force serv ed his well as he was largely responsible for managing 30 armed ecoguards. He received his Ph.D. from the University of Florida in 2009. The day after his dissertation defense, John and C onnie flew to Ethiopia to pick up their 3 month old son. With son in tow, they moved to Falmouth, MA to join the scientific staff at the Woods Hole Research Center.
LOGGING AND HUNTING ALTER PATTERNS OF SEED DISPERSAL AND SEEDLING RECRUITMENT IN AN AFROTROPICAL FOREST John Randoph Poulsen Phone: (352) 642-2964; Email: firstname.lastname@example.org Department of Biology Dr. Benjamin M. Bolker Doctorate of Philosophy August 2009 Unprecedented rates of logging and hunting thre aten the existence of the animals that inhabitat tropical forests. Because animals perf orm ecological functions like seed dispersal their extirpation could hinder forest re generation altering the structure and composition of the forest. I quantified the effects of logging and hunting on a tropical animal community, finding that they could reduce the abundance of some types of an imals by as much as 71%. Both logging and hunting changed seed dispersal patterns of 26 tr ee species, reducing the dispersal distance of animal-dispersed trees while increasing dispersal for wind-dispersed trees. In an experimental evaluation of whether different dispersal patter ns matter for seeding recruitment, more even distributions of seed across the landscape dramati cally increased seedling survival. These results demonstrate that the loss of seed-dispersing animal s is likely to inhibit fo rest regeneration. The conservation of tropical forests depends on the co nservation of the animals that inhabit them.