The Importance of Demographic Niches to Tree Diversity Richard Condit,1,2* Peter Ashton,3 Sarayudh Bunyavejchewin,4 H.S. Dattaraja,5 Stuart Davies,3 Shameema Esufali,6 Corneille Ewango,7 Robin Foster,8 I.A.U.N. Gunatilleke,6 C.V.S. Gunatilleke,6 Pamela Hall,9 Kyle E. Harms,10 Terese Hart,11 Consuelo Hernandez,12 Stephen Hubbell,13 Akira Itoh,14 Somboon Kiratiprayoon,15 James LaFrankie,16 Suzanne Loo de Lao,2 Jean-Remy Makana,11 Md. Nur Supardi Noor,17 Abdul Rahman Kassim,17 Sabrina Russo,3 Raman Sukumar,5 Cristi?n Samper,18 Hebbalalu S. Suresh,5 Sylvester Tan,19 Sean Thomas,20 Renato Valencia,12 Martha Vallejo,21 Gorky Villa,12 Tommaso Zillio1,2 1National Center for Ecological Analysis and Synthesis, 735 State St., Santa Barbara, CA 93101, USA. 2Center for Tropical Forest Science, Smithsonian Tropical Research Institute, Unit 0948, APO AA 34002?0948, USA. 3Center for Tropical Forest Science, Asia Program, Arnold Arboretum Asia Program, Harvard University Herbaria, 22 Divinity Avenue, Cambridge, MA 02138, USA. 4Thai National Park Wildlife and Plant Conservation Department, Research Office, 61 Phaholyothin Road, Chatuchak, Bangkok 10900, Thailand. 5Center for Ecological Sciences, Indian Institute of Science, Bangalore 560012, India. 6Department of Botany, Faculty of Science, University of Peradeniya, Peradeniya 20400, Sri Lanka. 7University of Missouri, 8001 Natural Bridge Road, St. Louis, MO 63121?4499, USA. 8Botany Department, The Field Museum, Roosevelt Road at Lake Short Drive, Chicago, IL 60605?2496, USA. 9Department of Biology, Florida State University, 5051 Quail Valley Road, Tallahassee, FL 32309, USA. 10Department of Biological Sciences, Louisiana State University, Baton Rouge, LA 70803, USA. 11Wildlife Conservation Society, 1725 Avenue Monjiba, Chanic Building 2nd floor, Ngalinema, B.P. 240, Kinshasa I, Democratic Republic of Congo. 12Pontif?cia Universidad Cat?lica de Ecuador, P.O. Box 17012184, Quito, Ecuador. 13Department of Plant Sciences, University of Georgia, 2502 Miller Plant Sciences Building, Athens, GA 30602, USA. 14Plant Ecology Lab, Faculty of Science, Osaka City University, Sugimoto 3-3-138, Sumiyoshi-ku, Osaka 558, Japan. 15Faculty of Science and Technology, Thammasat University (Rangsit), Klongluang, Patumtani, 12121, Thailand. 16Center for Tropical Forest Science, Asia Program, National Institute of Education, Nanyang Technological University, 1 Nanyang Walk, Singapore 637616. 17Forest Environment Division, Forest Research Institute Malaysia, Kepong, Kuala Lumpur 52109, Malaysia. 18National Museum of Natural History, Smithsonian Institution, 10th Street and Constitution Ave., NW, Washington, DC 20560, USA. 19Forest Research Center, Sarawak Forest Department, Km10 Jalun Datak Amar Kalong Ningkan, 93250 Kuching, Sarawak, Malaysia. 20University of Toronto, 33 Willcock St., Toronto, Ontario M5S 3B3, Canada. 21Instituto Alexander von Humboldt, Carrera 13 # 28-01, Piso 7, Edifico Palma Real, Bogot?, Colombia. *To whom correspondence should be addressed: condit@ctfs.si.edu Most ecological hypotheses about species coexistence hinge on species differences, but quantifying trait differences across species in diverse communities is often unfeasible. We examined variation of demographic traits using a global tropical forest dataset covering 4500 species in 10 large-scale tree inventories. With a hierarchical Bayesian approach, we quantified the distribution of mortality and growth rates of all tree species at each site. This allowed us to test the prediction that demographic differences facilitate species richness, as suggested by the theory that a tradeoff between high growth and high survival allows species to coexist. Contrary to prediction, the most diverse forests had the least demographic variation. Although demographic differences may foster coexistence, they do not explain any of the 16-fold variation in tree species richness observed across the tropics. Comparative studies of tree demography typically consider the entire community as a unit, ignoring species differences (1), simply because most tree inventories include small samples of many species (2, 3). Comparative studies show that tropical forests typically have higher turnover than temperate forests (4), and that higher tree turnover associates with higher tree diversity (5). These studies cannot, however, test ecological hypotheses about diversity, coexistence, and demography (6?10). A tradeoff between rapid growth and long lifespan permits species coexistence and can foster diversity: species reproducing early in life persist despite poor competitive ability by growing rapidly on disturbed sites where resources are abundant. Long-lived species coexist by out-living the weedy invaders, persisting where resources are scarce. This is a familiar and widely-known tradeoff in plant and animal communities (9?11): the successional-niche hypothesis / www.sciencexpress.org / 8 June 2006 / Page 1 / 10.1126/science.1124712 (7, 12). At a deterministic equilibrium, an indefinite number of species can coexist by this mechanism, each differing from all others along a continuum from short lifespan (with high growth) to long lifespan (and low growth). With stochastic demography, however, there is limiting similarity and the equilibrium species richness is finite (11, 13). This hypothesis is widely quoted as an explanation for tropical forest diversity (14?16). Here we ask whether species differences along a demographic axis explain why some tropical forests have many more species than others. If demographic niches are a key force controlling forest diversity, then more diverse forests have more demographic niches. More niches could come about either by spreading demographic rates over a wider range or packing more in the same range. Here we focus on the first prediction: tropical forests gain diversity by having a wider range of demographic niches, as reflected by the range of mortality and growth rates across species. We provide a direct test by quantifying mortality and growth of 4500 tree species in 10 different forests in America, Asia, and Africa (17). The 10 sites form a large- scale observation program, spanning a wide range of environmental conditions, designed to provide species- specific information for little-known tropical trees (18). At each site, a 20-52 ha tree census was set out in extensive, largely undisturbed forest (table S1). Species richness within the census plots differed by 16-fold, from 73 species per 50 ha in a dry forest at Mudumalai, India, to 1167 species per 50 ha in a wet dipterocarp forest in Sarawak, Malaysia (19). Past studies on the demography of individual tree species were based on direct measures of rate constants. These excluded many rare species because their rate estimates are subject to high error (20, 21). To overcome this limitation, we did not simply record species? rates of mortality and growth, instead we quantified the distribution of demographic rates across the entire community. A hierarchical Bayesian approach accomplishes this with explicit probability models covering both the observations of individual trees within species and the variation among species; all species, including rare ones, are included. For mortality, within- species distributions were modeled with the binomial distribution; for growth, we chose the log-normal based on the tendency for individual growth rates within a species to be highly right-skewed. By separating within-species variation, the hierarchical model allows focus on the question of how species differ (10, 22, 23). At the community level, we need to describe the variation in species? demographic rates across species, and again, right skewness suggested use of the log-normal. Histograms of mortality rate m and growth rate g (24) are fitted well by the log-normal when rare species are excluded (Fig. 1). The log- normal requires two parameters, ? and ?, the mean and standard deviation of the natural logarithm of m (or g). We were able to estimate values of ? and ? that best describe a community?s demography using the Gibbs sampler, simultaneously producing for every species an estimate of mortality and growth rates which are adjusted for abundance. That is, for abundant species, the estimate is barely different from the observed rate, but for rare species, it is guided by the community-wide pattern (25). Fitted log-normal distributions for the Lambir forest in Malaysia are plotted with observed histograms of sapling mortality and growth (Fig. 1). The fit is close for commoner species (filled bars), demonstrating that the large number of zeroes in the mortality histogram are sampling artifacts in rare species. Growth rates are also spread by rare species, though not as conspicuously. Fitted distributions for all the forests can be compared graphically (Fig. 2) or with estimated 2.5th and 97.5th percentiles (Table 1). In Supporting Online Material, tables of mortality and growth rates for all species are provided (25). Most of the forests were dominated by species with sapling mortality rates near 1% y-1 (Fig. 2). Even the high- mortality forests, such as BCI and HKK, had modes close to 1% y-1 and low rates around 0.4% y-1. The main feature separating these high-rate forests and low-rate sites (such as Pasoh in Malaysia) is the long tail reaching 20% y-1 mortality; at Pasoh, nearly all species had mortality rates < 3% y-1. Thus, forests fell broadly in two groups: BCI, HKK, and La Planada had upper sapling mortality rates above 20%, while Sinharaja, Lambir, Pasoh, and Yasuni had upper rates below 8%. The Congo sites had exceedingly low mortality stretching to a modest 10% at the upper end. Distributions of growth rates were similar to distributions of mortality, but growth was about half as variable across species (Fig. 2). Conspicuously, sites with less variation in mortality also had less variation in growth (Fig. 3). These patterns held for larger trees, although mortality and growth rates were lower (table S2 and fig. S1). Examples from individual species help illustrate. At BCI, a fast-growing understory treelet, Palicourea guianensis, had a population of 376 saplings in 1982, and every single one had died by 2005 (nevertheless, the population grew to 851). Although Palicourea?s mortality rate is infinite by direct calculation, the Gibbs sampler produces an estimate of 33% y-1. Alloplectus schultzei, a small, weedy treelet at La Planada, also suffered 33% y-1 mortality, losing 284 of 335 individuals over six years. In contrast, of 1162 species at Lambir, none had mortality of 30% y-1, and just two had rates above 20%; at Pasoh during 1987-1995, the very highest Gibbs-corrected mortality rate among 802 species was 12% y-1, in Macaranga hypoleuca. At the other end of the distribution, Cupania sylvatica, a mid-sized tree of the BCI understory, lost only 10 of 1102 / www.sciencexpress.org / 8 June 2006 / Page 2 / 10.1126/science.1124712 individuals during 1990-1995 (0.23% y-1), and Carapa guianensis at La Planada, a large and valuable timber tree, lost only 9 of 893 (0.27% y-1). In three census intervals at Pasoh, the lowest mortality rate recorded was 0.33% y-1, in Milletia atropurpurea. The Mudumalai forest stood out. Saplings had greatly elevated mortality and growth, with rates stretching much higher than any other site. During 1988-1992, every species at Mudumalai had sapling growth > 6% y-1, and only BCI and HKK had many rates this high. At Lenda and Sinharaja, no species grew by 6% y-1. For larger trees (? 100 mm diameter), however, Mudumalai was in line with other forests, having modes of mortality and growth near 1% y-1 (table S2). In fact, trees at Mudumalai had among the lowest rates as well as the highest: Anogeissus latifolia had 116 deaths out of 2179 trees from 1988 to 2000, while Kydia calycina had 1272 of 1328 trees die over the same interval, many because of elephant herbivory (26). Their rates differ by 50-fold: 27% y-1 in Kydia, 0.46% y-1 in Anogeissus. Three of the sites with long tails of elevated mortality and growth ? BCI, HKK, Mudumalai ? have intense annual dry seasons (table S1). Mudumalai and HKK also burn in some years (26) (other plots do not burn and none suffer large-scale wind damage). It was not surprising that annual drought elevated mortality. Many species at these sites, however, had exceedingly low rates of mortality and evidently did not suffer much from drought. Conversely, forests lacking the tail of high growth and mortality had no or modest annual dry seasons, including the three forests dominated by Dipterocarpaceae (Sinharaja, Pasoh, Lambir). Seasonality, however, was not the only factor predicting high variation in demography: the ever-wet cloud forest at La Planada, Colombia, had a wide spread of growth and mortality, comparable to the seasonally dry sites. Mudumalai and HKK have relatively open canopies compared to all the other sites, a typical feature of dry forests, and many saplings at Mudumalai are sprouts from large root systems. These are likely reasons for elevated sapling growth at the two sites. Both BCI and La Planada have dense canopies and dark understories, though, not obviously differing from the sites that lacked high-growth species. Contrary to the prediction that demographic variability begets species richness, diverse forests had the least variation in demography (Fig. 3 and fig. S1). If anything, the most diverse forests had the fewest demographic niches. At Lambir, high species richness coupled with a low diversity of demographic rates meant that 127 tree species coexisting in close proximity had sapling mortality rates in a narrow window from 0.8% to 1.0% y-1. We do not question that demographic variability plays some role in species coexistence. In American forests, the familiar genus Cecropia is found exclusively in small forest clearings (or outside the forest), where it rapidly colonizes and rapidly dies. The upper end of sapling mortality and growth distributions in America is set by gap specialists: C. obtusifolia at BCI (12% y-1 mortality, 14% y-1 growth); C. sciadophylla at Yasuni (5.0% mortality, 6.3% growth); C. monostachya at La Planada (8.8% mortality, 8.2% y-1). Diverse Southeast Asian forests lacked species with such high rates (27). The most diverse tropical forests are the least diverse demographically. It remains plausible that demographic niches are packed more tightly in some forests than others, but this seems unlikely, since packing should depend only on population size and turnover, which do not vary much. Moreover, the successional-niche hypothesis is not favored by the strong peak in demographic rates near 1-2% y-1: if demographic niches were crucial, then rates ought to be spread evenly over the entire range (28). Instead, the similarity in demography of many species suggests trait convergence (29). We believe that broad diversity differences are due to the source pool of different biogeographic regions, and that demographic differences play a minor role in species coexistence. References and Notes 1. M.D. Swaine, D. Lieberman, F.E. Putz, J. Trop. Ecol. 3, 359 (1987). 2. Most forest censuses cover a single hectare or less and include only larger trees. Except in Europe and North America, such small samples capture less than half the local species, with many represented by a single individual. 3. N. Pitman, J. Terborgh, M.R. Silman, P. N??ez V., Ecology 80, 2651 (1999). 4. N.L. Stephenson, P.J. van Mantgem, Ecol. Lett. 8, 524 (2005). 5. O.L. Phillips, P. Hall, A.H. Gentry, S.A. Sawyer, R. V?squez, Proc. Acad. Nat. Sci. Phila. 91, 2805 (1994). 6. D. Tilman, Plant Strategies and the Dynamics and Structure of Plant Communities (Princeton Univ. Press, Princeton, USA, 1988). 7. S.W. Pacala, M. Rees, Am. Nat. 152, 729 (1998). 8. B.M. Bolker, S.W. Pacala, C. Neuhauser, Am. Nat. 162, 135 (2003). 9. M. Rees, R. Condit, M. Crawley, S. Pacala, D. Tilman, Science 293, 650 (2001). 10. J.S. Clark, J. Mohan, M. Dietze, I. Iba?ez, Ecology 84, 17 (2003). 11. M.B. Bonsall, V.A.A. Jansen, M.P. Hassell, Science 306, 111 (2004). 12. J.S. Clark, S. LaDeau, I. Iba?ez, Ecol. Monog. 74, 415 (2004). 13. D. Tilman, Proc. Acad. Nat. Sci. Phila. 101, 10854 (2004). / www.sciencexpress.org / 8 June 2006 / Page 3 / 10.1126/science.1124712 14. P.J. Grubb, Biol. Rev. Camb. Philos. Soc. 52, 107 (1977). 15. S.J. Wright, Oecologia 130, 1 (2002). 16. J. Silvertown, Trends Ecol. Evol. 19, 605 (2004). 17. Table 1 includes 4956 species, but some occur at > 1 site, and taxonomy has not been fully aligned across sites; we estimate 4500 different species at the 10 sites. 18. R. Condit, Trends Ecol. Evol. 10, 18-22 (1995). 19. The Sarawak plot covers 52 ha; for comparison, the number 1167 was taken from a 50-ha subset. 20. J.K. Vanclay, J. Trop. For. Sci. 4, 15 (1991). 21. R. Condit, S.P. Hubbell, R.B. Foster, Ecol. Monog. 65, 419 (1995). 22. A. Gelman, J.B. Carlin, H.S. Stern, D.B. Rubin, Bayesian Data Analysis (Chapman and Hall, Boca Raton, FL, 1995). 23. M. Liermann, R. Hilborn, Can. J. Fish. Aquat. Sci. 54, 1976 (1997). 24. The rate constant m is the derivative of population size N with respect to time, dt dN , due to mortality alone. It is approximated by 12 12 lnln tt NN ? ? . Relative growth rate is 12 12 lnln tt dbhdbh=g ? ? (dbh = stem diameter-at-breast- height). 25. Methods are presented in Supporting Online Material. 26. R. Sukumar, H.S. Suresh, H.S. Dattaraja, N.V. Joshi, in Forest Biodiversity, Research and Modeling, F. Dallamier, J. Comiskey, Eds. (Smithsonian, Washington, DC, 1998) pp. 495?506. 27. R. Condit et al., Phil. Trans. R. Soc. London Ser. B 354, 1739:1748 (1999). 28. S.J. Wright, H.C. Muller-Landau, R. Condit, S.P. Hubbell, Ecology 84, 3174 (2003). 29. S.P. Hubbell, Ecology, in press (2006). 30. Errors in dbh measurements are 1-2 mm in saplings, and many saplings show slight negative growth (25). We thus cannot measure relative growth rates below about 0.5% yr-1. Where near this figure, the low end of the sapling growth distributions are probably artifacts. In larger trees, errors are smaller relative to dbh, eliminating the problem. 31. Analyses were supported by US National Science Foundation grant DEB-9806828 of the Research Coordination Network Program to the Center for Tropical Forest Science. Data collection was funded by many organizations, principally the US National Science Foundation, Andrew W. Mellon Foundation, Peninsula Community Foundation, Smithsonian Tropical Research Institute, Arnold Arboretum (Harvard), Indian Institute of Science, Forest Research Institute of Malaysia, Royal Thai Forest Department, National Institute of Environmental Studies (Japan), and John. D. and Catherine T. McArthur Foundation. We thank the hundreds of field workers who have measured and mapped trees. Supporting Online Material www.sciencemag.org/cgi/content/full/1124712/DC1 Materials and Methods Tables S1 to S4 Fig. S1 References Database S1 Computer Codes 9 January 2006; accepted 31 May 2006 Published online 8 June 2006; 10.1126/science.1124712 Include this information when citing this paper. Fig. 1. Distribution of sapling demographic rates of all species in the Lambir plot. (A) Annual mortality, m, for all individuals 10-99 mm dbh. Filled bars show the histogram of observed mortality rates for the 951 species with ? 20 individuals; open bars add the 211 species with < 20 individuals. The open bar at m = 0 extends off the graph (162 species had no mortality; 121 of these had < 20 individuals). The horizontal axis is m, expressed as a percent (Fig. 1). The solid line is the fitted log-normal, based on all 1162 species. The dashed vertical line indicates the mean of the logarithm of the fitted distribution (parameter ?, Table 1), which is very close to the median. (B) Annual growth, g, for individuals 10- 49 mm diameter. Filled bars are the histogram for 995 species with ? 10 individuals; open bars for the remaining 155 species. The solid curve and dashed line as in (A). Both histograms are curtailed at 8% to accentuate details where most of the species fall. The number of species above 8% is indicated by arrows. Fig. 2. Comparing the fitted distributions of sapling demography in four forests. (A) Annual mortality rate, m. (B) Annual growth rate, g. The lower end of the growth distribution in saplings is limited by measurement accuracy (30). Fig. 3. Range of sapling demographic rates for tree species within a community versus the number of species at the site. The range is the logarithm of the ratio between the 97.5th and 2.5th percentiles of the fitted distributions (Table 1). The range for mortality is given by filled circles; for growth by open triangles. Sites can be identified by the number of species, that is, Lambir is the most diverse and furthest to the right, etc. Multiple censuses at BCI, Pasoh, and Mudumalai are included, and in each case, fall in a tight group. / www.sciencexpress.org / 8 June 2006 / Page 4 / 10.1126/science.1124712 Table 1. Variation in sapling mortality and growth rates across species in tropical forests. For mortality, all individuals 10-99 mm dbh were included; for growth, 10-49 mm dbh. Species number refers to those with at least one 10-99 mm sapling alive at the outset of a given census interval. Under mortality are percentiles of the distribution of mortality rate parameters (m ) across species: the median plus lower and upper percentiles (2.5th and 97.5th) of the fitted log-normal. Likewise under growth are percentiles for the distribution of growth rates (g ) across species. Rates are expressed as a percentages (100m or 100g), that is 5 = 5% = .05. For each of the percentiles, confidence limits are given, based on the Gibbs sampler (25). BCI = Barro Colorado Island; HKK = Hua Khae Khaeng National Park. Information about the sites is presented in table S1. Annual mortality (%) Relative growth (%) Site Years Median Lower Upper Median Lower Upper BCI 82?85 3.14 ? 0.46 0.38 ? 0.10 26.0 ? 6.9 2.84 ? 0.16 1.35 ? 0.14 6.0 ? 0.7 BCI 85?90 2.56 ? 0.37 0.31 ? 0.08 21.5 ? 5.5 2.41 ? 0.18 0.85 ? 0.12 6.8 ? 1.0 BCI 90?95 2.85 ? 0.43 0.32 ? 0.09 25.3 ? 6.7 2.15 ? 0.13 0.89 ? 0.09 5.2 ? 0.7 BCI 95?00 3.35 ? 0.42 0.48 ? 0.11 23.3 ? 5.7 1.97 ? 0.12 0.81 ? 0.09 4.8 ? 0.6 BCI 00-05 2.91 ? 0.41 0.40 ? 0.10 21.4 ? 5.5 2.10 ? 0.16 0.73 ? 0.10 6.1 ? 0.9 Yasuni 96?03 1.55 ? 0.10 0.31 ? 0.04 7.7 ? 0.9 1.67 ? 0.04 0.83 ? 0.04 3.4 ? 0.2 La Planada 97?03 3.22 ? 0.47 0.45 ? 0.13 22.9 ? 5.9 2.30 ? 0.17 0.93 ? 0.13 5.7 ? 0.8 Pasoh 87?90 1.04 ? 0.06 0.36 ? 0.04 3.0 ? 0.3 2.25 ? 0.05 1.38 ? 0.05 3.7 ? 0.2 Pasoh 90?95 1.35 ? 0.07 0.42 ? 0.05 4.3 ? 0.4 1.59 ? 0.03 1.02 ? 0.04 2.5 ? 0.1 Pasoh 95?00 1.69 ? 0.09 0.47 ? 0.06 6.0 ? 0.6 1.55 ? 0.03 1.07 ? 0.03 2.3 ? 0.1 Lambir 92?97 1.32 ? 0.07 0.34 ? 0.03 5.2 ? 0.4 1.57 ? 0.03 0.96 ? 0.03 2.6 ? 0.1 HKK 93?99 4.38 ? 0.64 0.61 ? 0.20 31.2 ? 9.1 4.83 ? 0.45 1.53 ? 0.26 15.2 ? 2.7 Mudumalai 88?92 13.06 ? 3.48 2.65 ? 1.42 64.4 ? 36 7.87 ? 1.26 4.65 ? 1.52 13.3 ? 3.8 Mudumalai 92?96 17.06 ? 6.43 2.35 ? 1.51 124 ? 113 6.35 ? 1.67 2.57 ? 1.38 15.7 ? 11 Mudumalai 96?00 7.96 ? 2.70 1.73 ? 1.13 36.6 ? 25 5.71 ? 1.53 2.42 ? 1.69 13.4 ? 9.8 Sinharaja 95?01 1.35 ? 0.17 0.30 ? 0.07 6.0 ? 1.3 1.38 ? 0.07 0.75 ? 0.07 2.5 ? 0.2 Edoro 94?00 1.43 ? 0.20 0.21 ? 0.06 9.6 ? 2.5 1.41 ? 0.09 0.57 ? 0.07 3.5 ? 0.5 Lenda 94?00 1.26 ? 0.19 0.18 ? 0.06 8.8 ? 2.2 1.06 ? 0.04 0.66 ? 0.05 1.7 ? 0.1 0 - I I I 1 I I I I I 0 1 2 3 4 5 6 7 8 annual mortality (%) ; L median mortality I I I I I I I I I 0 1 2 3 4 5 6 7 8 annual relative growth (%) 2 3 4 5 6 annual mortality (96) HKK -----..-.--_.._ 2 3 4 5 6 annual relative arowth ( O h ) a growth 0 200 400 600 800 1000 species