Dispersal, Environment, and Floristic Variation of Western Amazonian Forests Hanna Tuomisto, et al.

Science 299, 241 (2003); DOI: 10.1126/science.1078037

This copy is for your personal, non-commercial use only.

If you wish to distribute this article to others, you can order high-quality copies for your colleagues, clients, or customers by clicking here. Permission to republish or repurpose articles or portions of articles can be obtained by following the guidelines here. Downloaded from www.sciencemag.org on September 29, 2011 The following resources related to this article are available online at www.sciencemag.org (this infomation is current as of September 29, 2011 ): Updated information and services, including high-resolution figures, can be found in the online version of this article at: http://www.sciencemag.org/content/299/5604/241.full.html Supporting Online Material can be found at: http://www.sciencemag.org/content/suppl/2003/01/09/299.5604.241.DC1.html This article cites 26 articles, 6 of which can be accessed free: http://www.sciencemag.org/content/299/5604/241.full.html#ref-list-1 This article has been cited by 197 article(s) on the ISI Web of Science This article has been cited by 11 articles hosted by HighWire Press; see: http://www.sciencemag.org/content/299/5604/241.full.html#related-urls This article appears in the following subject collections: Ecology http://www.sciencemag.org/cgi/collection/ecology

Science (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by the American Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. Copyright 2003 by the American Association for the Advancement of Science; all rights reserved. The title Science is a registered trademark of AAAS.

REPORTS
flow makes a substantial contribution to the geomagnetic field at satellite altitude. This has important implications: In broader terms, it encourages future studies to assess the feasibility of monitoring ocean flow from space. A more immediate consequence, however, is that it shows that oceanic signals must be incorporated into geomagnetic field models. Indeed, with recent advances in internal and external field separation (17 ), the ocean flow signal is now the strongest remaining signal in the low-latitude magnetic residuals that has not yet been modeled. Correcting magnetic readings for ocean flow signals could strongly improve the accuracy of lithospheric anomaly maps and greatly raise the detectability of small-scale crustal magnetization.
1. T. B. Sanford, J. Geophys. Res. 76, 3476 (1971). 2. A. D. Chave, D. S. Luther, J. Geophys. Res. 95, 7185 (1990). 3. R. H. Tyler, L. A. Mysak, Geophys. Astrophys. Fluid Dyn. 80, 167 (1995). 4. R. H. Tyler, L. A. Mysak, J. M. Oberhuber, J. Geophys. Res. 102, 5531 (1997). 5. F. E. M. Lilley, J. H. Filloux, P. J. Mulhearn, I. J. Ferguson, J. Geomagn. Geoelectr. 45, 403 (1993). 6. J. C. Larsen, Geophys. J. R. Astron. Soc. 16, 47 (1968). 7. S. R. C. Malin, Geophys. J. R. Astron. Soc. 21, 447 (1970). 8. N. L. Bindoff, F. E. M. Lilley, J. H. Filloux, J. Geomagn. Geoelectr. 40, 1445 (1988). 9. J. D. McKnight, Geophys. J. Int. 122, 889 (1995). 10. R. H. Tyler, T. B. Sanford, J. M. Oberhuber, J. Geomagn. Geoelectr. 49, 1351 (1997). 11. R. H. Tyler, thesis, McGill University (1995). 12. G. D. Egbert, A. F. Bennet, M. G. G. Foreman, J. Geophys. Res. 99, 24821 (1994). 13. R. Holme, N. Olsen, M. Rother, H. Luhr, in First CHAMP Mission Results for Gravity, Magnetic and

References and Notes

Atmospheric Studies, C. Reigber. H. Luhr, P. Schwintzer, Eds. (Springer, Berlin, in press). 14. A. T. Price, Q. J. Mech. Appl. Math. 2, 283 (1949). 15. R. H. Tyler, paper presented at the Third International Conference on Marine Electromagnetics, Stockholm, 11 to 13 July 2001. 16. Materials and methods are available as supporting material on Science Online. 17. S. Maus et al., Geophys. Res. Lett. 29, 10.1029/ 2001GL013685 (2002). 18. CHAMP is supported by Deutsches Zentrum fur Luft und Raumfahrtechnik under contract FKZ 50 EP 9587, data analyses were supported by Deutsche Forschungsgemeinschaft (SPP 1097), and the numerical prediction results are based on work supported by NASA (under award NAG5-10570) and the Ofce of Naval Research (under award N00014-99-1-0407). Supporting Online Material www.sciencemag.org/cgi/content/full/299/5604/239/ DC1 Materials and Methods Fig. S1 Movie S1

Dispersal, Environment, and Floristic Variation of Western Amazonian Forests

Hanna Tuomisto,1* Kalle Ruokolainen,1 Markku Yli-Halla2
The distribution of plant species, the species compositions of different sites, and the factors that affect them in tropical rain forests are not well understood. The main hypotheses are that species composition is either (i) uniform over large areas, (ii) random but spatially autocorrelated because of dispersal limitation, or (iii) patchy and environmentally determined. Here we test these hypotheses, using a large data set from western Amazonia. The uniformity hypothesis gains no support, but the other hypotheses do. Environmental determinism explains a larger proportion of the variation in oristic differences between sites than does dispersal limitation; together, these processes explain 70 to 75% of the variation. Consequently, it is important that management planning for conservation and resource use take into account both habitat heterogeneity and biogeographic differences. Unraveling the relative importance of biological interactions, random variation, dispersal limitation, and environmental determinism in creating differences in species composition among sites (beta diversity) is a central issue in plant ecology (17). The hypothesis that the plant species composition of Amazonian noninundated forests is uniform over large areas emphasizes the role of biological interactions: It is suggested that the forests are dominated by a limited suite of competitively superior tree species (8, 9). The hypothesis that plant species composition fluctuates in a random walk emphasizes dispersal history: The species are competitively equal, and floristic differences are created through random but spatially limited dispersal of species that
1 Department of Biology, University of Turku, FIN20014 Turku, Finland. 2MT T Agrifood Research Finland, FIN-31600 Jokioinen, Finland.

*To whom correspondence should be addressed. Email: hantuo@utu..

evolved in different areas (2, 10, 11). The hypothesis that species distributions are patchy emphasizes environmental determinism: The forests are considered to be a mosaic where plant species composition is determined by edaphic and other environmental site characteristics (12, 13). Which of these hypotheses is accepted as the main explanatory model has important practical implications for biodiversity conservation, forest management, and the planning and interpretation of ecological research. To test how well the distributions of Amazonian plant species conform to the three hypotheses, we inventoried 163 sites in four regions in western Amazonia (Colombia, Ecuador, northern Peru, and southern Peru) using a standard quantitative procedure (500-mby5-m line transects) (14, 15). The inventories included noninundated forests on clay or loam soils (122 sites; henceforth called tierra firme) and on white sand soils (3 sites) and seasonally inundated and swamp forests (38 sites). The topog-

raphy of the transects ranged from flat to hilly (with a difference in elevation of up to 60 m), depending on local terrain. We inventoried two distinct plant groups: pteridophytes (ferns and fern allies) and the Melastomataceae (a family of shrubs and small trees). Because these groups are both phylogenetically remote and dispersed by different agents (wind versus animals), they provide independent test cases for measuring the relative importance of processes related to evolution, dispersal limitation, biological interactions, and niche differentiation. By focusing on these plant groups, we were able to compare voucher specimens and apply a uniform taxonomy over all regions; hence, we did not need to exclude unnamed morphospecies from the analyses [as is usually done in tree inventories (8, 9, 11)]. Our approach also yields relatively high numbers of individuals per species and an indication of leveling-off of the species-area curve within each site (13, 14 ), which should dampen the effect of sampling error. The data set includes 286 species and 297,000 individuals of pteridophytes, and 265 species and 40,300 individuals of Melastomataceae. We analyzed several (usually three) composite soil samples from each site (15) to obtain quantitative data on mean soil chemistry and texture (tables S1 and S2). We first examined the hypothesis that tierra firme forests are uniform over wide areas, especially when only the most abundant species are considered (8, 9). If the forests are uniform, then floristic similarity among sites should be uniformly high and have no identifiable pattern; in particular, the degree of floristic similarity should not depend on the geographic distance between sites or on differences in their environmental conditions. We tested these predictions with Mantel tests, and they were refuted for both pteridophytes and Melastomataceae: Floristic distance between sites showed a significant correlation with both environmental and geographic distance (Table 1). These correlations were

www.sciencemag.org SCIENCE VOL 299 10 JANUARY 2003

241

Downloaded from www.sciencemag.org on September 29, 2011

4 September 2002; accepted 8 November 2002

REPORTS
practically equal whether we considered all species or only the most abundant one-third of the species. Contrary to what would be expected if the most abundant species are generalists that are insensitive to geographic and environmental effects, the correlations were higher for the most abundant one-third of the species than for the intermediately abundant one-third. The actual floristic similarity values also indicated that the forests are not homogeneous: Even sites within the same region could be so different that they did not share a single species of pteridophytes or Melastomataceae, whereas the most similar sites shared more than 70% of their species. The view that dispersal history provides the main explanation for variation in plant species composition is based on Hubbells neutral theory (2), which predicts that floristic similarity among plots decreases with increasing geographic distance because of spatially limited dispersal. The hypothesis of environmental control of species distributions, on the other hand, predicts that floristic similarity decreases with increasing environmental distance. Partial Mantel tests showed that geographic distance was significantly correlated with the floristic distances even after the effect of environmental distance had been taken into account, and vice versa (Table 1). The correlation with environmental distance was practically the same for both plant groups. For pteridophytes, the correlation with geographic distance was smaller than that with environmental distance, whereas for the Melastomataceae, the correlations were about equal. This difference between the plant groups is in agreement with what can be expected on the basis of their dispersal modes: Pteridophytes have wind-dispersed spores, whereas most Melastomataceae have bird-dispersed berries and therefore are likely to be more dispersal-limited (16). In general terms, between-site floristic similarity of both plant groups decreased logarithmically with increasing geographic distance, but there were some intervals of geographic distance within which floristic similarity actually

Table 1. Correlations of oristic distance with environmental and geographic distance for 122 tierra rme sites in western Amazonia as measured with simple and partial Mantel tests. Top one-third and middle one-third are the most abundant two-thirds of species; abundance is the observed number of individuals. The complement of the Jaccard index was used as oristic distance. A simple measure of environmental distance was obtained by using one climate variable (aseasonal versus seasonal) and eight soil variables ( pH; loss on ignition; percentage of clay and silt; and the logarithmically transformed concentration of Ca, K, Mg, Na, and Al). Euclidean distances were computed after all variables were standardized. Mantel tests were based on Spearman correlations between the distance matrices to test for a monotonic but not necessarily linear relationship. All correlations are statistically signicant (P 0.001), as determined with a Monte Carlo permutation test using 999 permutations. Included species Geographic distance Environmental distance Geographic distance with environmental distance partialled out Environmental distance with geographic distance partialled out All Top one-third Middle one-third All Top one-third Middle one-third All Top one-third Middle one-third All Top one-third Middle one-third Pteridophytes (total 269 species) 0.39 0.38 0.34 0.46 0.44 0.42 0.22 0.22 0.18 0.34 0.32 0.31 Melastomataceae (total 249 species) 0.47 0.46 0.38 0.48 0.48 0.41 0.31 0.30 0.23 0.33 0.34 0.28

Fig. 1. Distance decay of oristic and environmental similarity among 122 tierra rme sites; each gray dot represents a pair of sites. (A, B, D, and E) Floristic similarity ( Jaccard index) between sites as measured with pteridophytes [(A) and (D)] and Melastomataceae [(B) and (E)] plotted against geographic [(A) and (B)] and environmental [(D) and (E)] distance between the same sites. (C) Environmental similarity between sites plotted against their geographic distance. Environmental similarity is the complement of the environmental distance shown in (D) and (E), computed as in Table 1. Three regression curves are shown, of which the exible spline function gives the most accurate representation of changes in average similarity with increasing distance.

242

10 JANUARY 2003 VOL 299 SCIENCE www.sciencemag.org

Downloaded from www.sciencemag.org on September 29, 2011

REPORTS
increased with geographic distance (see spline in Fig. 1, A and B). These reversals in the main trend corresponded to the areas that showed increasing environmental similarity with increasing geographic distance (Fig. 1C), illustrating that environmental distance can explain floristic similarities. Floristic similarities showed a clear decreasing trend with increasing environmental distance; these plots show no major trend reversals (Fig. 1, D and E). Geographic distance seemed to have the clearest effect on floristic similarity at distances 80 km, where mean floristic similarity decreased rapidly even though mean environmental similarity remained practically unchanged (Fig. 1, A to C). At greater geographic distances, mean floristic similarity followed the mean environmental similarity more closely. The overall patterns of the two plant groups in Fig. 1 are very similar, and they yielded a very high correlation in the Mantel test (r 0.83, P 0.001, based on the Pearson correlation to test for linear relationship). In Hubbells neutral theory (2), the only force affecting floristic similarity between sites is dispersal limitation, which accounts for the logarithmic trend in Fig. 1, A and B. However, the theory includes no mechanism by which independent plant groups would converge in their similarity patterns beyond the distance effect, so any congruence between plant groups that is independent of distance provides proof of deterministic structures that are not explainable by the neutral theory. To test how well the similarity patterns in the two plant groups correlate after the logarithmically decreasing trend with increasing geographic distance has been taken into account, we fitted the logarithmic models shown in Fig. 1, A and B, to the similarity data of the respective plant groups and correlated the residuals. The Mantel correlation between the two plant groups remained very high (r 0.72, P 0.001), which indicates that their similarity patterns are structured by some common external forces independently of geographic distance. To tease apart the relative importance of the different explanatory factors, we used multiple regression on distance matrices (17) for pteridophytes and Melastomataceae separately. We first analyzed how much of the variation in floristic distances can be explained by the logarithmically transformed geographic distances; we then divided this fraction into two portions, one explained by geographic distance alone and the other jointly explained by environmental distances. Then we used the residual distances obtained after fitting the logarithmic decay model to the similarities of each plant group in turn to test to what degree the residual distances of one plant group can be explained by environmental distances and the residual distances of the other plant group. The use of the residuals in this context makes it possible first to partial out the effect of dispersal limitation (as modeled by logarithmically transformed geographic distance) and then to quantify the additional effect of the external deterministic factors that affect two independent plant groups in the same way. More than half of the residual variation in the pteridophyte distances was explained by the residual Melastomataceae distances and vice versa (Fig. 2). About half of this congruence, in turn, was explained by differences in the measured environmental variables, whereas the other half must be accounted for by differences in variables that were not measured in this study, probably including historical factors such as major disturbances and environmental factors such as drainage, forest structure, and unmeasured soil properties (for example, nitrogen, phosphorus, or micronutrient content). Only about one-quarter of the total variation in floristic distances remained unexplained, which is a small proportion compared with a similar analysis made in Panama (18, 19). The floristic and environmental distinctness of the forests on white sand soils and in seasonally inundated and swamp areas is widely recognized, whereas the possible importance of environmental factors for the species composition of tierra firme forests is much more controversial (810, 13, 2022). Against this background, it is rather surprising that the explained proportion of variation in floristic distances was higher for the tierra firme forests than for the full set of sites (Fig. 2). This result was largely due to the greater explanatory power of geographic distances (and hence dispersal limitation) within tierra firme forests. The proportion of the variation in floristic distances that can be considered deterministic was about the same in tierra firme forests as in the full set of sites. This indicates that floristic differences among tierra firme sites are as readily explainable in terms of niche differentiation as those between inundated and tierra firme forests. The importance of both edaphic and geographic information for understanding floristic patterns in western Amazonian forests is illustrated by ordination of the floristic data (Fig. 3). When all 163 sites were included, most of the inundated sites were concentrated to one edge of the ordination diagram, whereas the noninundated sites showed more heterogeneity and were scattered over a wider area (Fig. 3, A and B). When only the 122 tierra firme sites were included, the sites from the three northern regions (Ecuador, Colombia, and northern Peru) formed a clear gradient where floristic composition was highly related to soil cation content; a similar gradient was apparent in southern Peru (Fig. 3, C and D). The separation between the northern and the southern regions indicates either that the climatic differences between them are important for species composition, that significant geochemical differences other than cation content exist between the regions, or that the long distance between them is sufficient for biogeographic differentiation. All three factors may be contributing simultaneously. Geochemical differences can be expected at

Fig. 2. Relative importance of different factors in explaining variation in oristic distances between sites. In each case, variation in the oristic distances between site pairs (as measured by the response plant group; complement of the Jaccard index) was explained by variation in geographic and environmental distances and residual oristic distances (as measured by the other plant group) by using multiple regression on distance matrices (17, 18). Fractions that do not include geographic distances have been computed by using the residual oristic distances of the response plant group after tting the corresponding logarithmic function from Fig. 1, A or B. Analyses used the same environmental variables as in Table 1, but a separate distance matrix was computed for each variable. For the full set of 163 sites, four additional binary variables indicated whether the sites were seasonally inundated, swamp, white sand, or tierra rme forests. Before analysis, we subjected the environmental variables to backward elimination to retain only the ones with a statistically signicant contribution (P 0.05 after Bonferroni correction). All analyses retained Ca content, percentage of clay and silt, and seasonality. Additional retained variables were K and Mg content and loss on ignition in (A); K and Mg content in (B); Mg content, white sand, and tierra rme in (C); and K content and inundation in (D). The approach is analogous to the variation partitioning of Borcard et al. (30), with the important difference that their method was based on canonical correspondence analysis rather than multiple regression on distance matrices.

www.sciencemag.org SCIENCE VOL 299 10 JANUARY 2003

243

Downloaded from www.sciencemag.org on September 29, 2011

REPORTS
different spatial scales in western Amazonia on the basis of its complex geological history (23 26). Mineralogical composition varies between different geological formations and may be ecologically important because it affects many soil properties, such as buffering capacity, nutrient release potential, and water and nutrient holding properties; however, we did not measure it in this study. In those soil properties that we did measure, the range of variability in our data is very similar to what has been reported in earlier Amazonian soil studies. For example, the content of exchangeable bases (Ca, K, Mg, and Na) in our tierra firme sites ranges from 0.09 to 33.6 cmol( )/kg (median value, 1.2), whereas values between 0.1 and 39 cmol( )/ kg have been reported for Brazil and Peru (27, 28). Important edaphic differences can also be expected because of the occurrence of various soil types (Ultisols, Oxisols, Inceptisols, and Alfisols) in Peruvian Amazonia (28). There is clearly a high degree of floristic differentiation in western Amazonian forests both within and among regions. We suggest that earlier results indicating low beta diversity in western Amazonia (8, 9, 11) were partly due to incomplete sampling (19). Our data confirm the earlier observation that sites in Ecuador and southern Peru can show high floristic similarity. However, our data also make clear that rather than being a typical characteristic of western Amazonian forests in general, high between-site floristic similarity is indicative of higher than average environmental (especially edaphic) similarity (Fig. 3). Our results were obtained with just two plant groups, but these are so different ( phylogenetically remote and dispersed by different agents) that there is no reason for them to behave more similarly than terrestrial plants in general. Earlier studies suggest that pteridophytes and Melastomataceae show practically the same floristic patterns as trees (13, 21, 29). Therefore, we predict that similar spatial and environmental trends would appear with other plants, including trees, if they were analyzed in an equally detailed way. Our findings have implications for both biodiversity conservation and its utilization in Amazonia. Because of habitat heterogeneity, plant species are distributed in a patchy way, and, because of dispersal limitation, there is a gradual turnover in regional species pools between distant areas. These patterns need to be taken into account when delimiting nature reserves and assessing their representativity as well as when estimating the extractable amounts of forest products. Extrapolation from local inventories to areas that have not been field-surveyed should always be done with great caution because real differences exist between areas, even when the forests appear similar.
1. J. H. Brown, Am. Nat. 124, 255 (1984). 2. S. P. Hubbell, The Unied Neutral Theory of Biodiversity and Biogeography (Princeton Univ. Press, Princeton, NJ, 2001). 3. S. P. Hubbell, R. B. Foster, in Community Ecology, J. Diamond, T. J. Case, Eds. (Harper & Row, New York, 1986), pp. 314 329. 4. G. C. Hurtt, S. W. Pacala, J. Theor. Biol. 176, 1 (1995). 5. D. Tilman, Ecology 75, 2 (1994). 6. J. Whiteld, Nature 417, 480 (2002). 7. R. H. Whittaker, S. A. Levin, Theor. Popul. Biol. 12, 117 (1977). 8. N. C. A. Pitman, J. Terborgh, M. R. Silman, P. Nunez V., Ecology 80, 2651 (1999). 9. N. C. A. Pitman et al., Ecology 82, 2101 (2001). 10. R. Condit, Trends Ecol. Evol. 11, 4 (1996). et al., Science 295, 666 (2002). 11. 12. A. H. Gentry, Ann. Mo. Bot. Gard. 75, 1 (1988). 13. H. Tuomisto et al., Science 269, 63 (1995). 14. H. Tuomisto, A. D. Poulsen, J. Biogeogr. 23, 283 (1996). 15. H. Tuomisto et al., Ecol. Appl., in press. 16. J. C. Nekola, P. S. White, J. Biogeogr. 26, 867 (1999). 17. P. Legendre, F. Lapointe, P. Casgrain, Evolution 48, 1487 (1994). 18. J. F. Duivenvoorden, J.-C. Svenning, S. J. Wright, Science 295, 636 (2002). 19. K. Ruokolainen, H. Tuomisto, Science 297, 1439 (2002). 20. J. F. Duivenvoorden, Vegetatio 120, 91 (1995). 21. K. Ruokolainen, A. Linna, H. Tuomisto, J. Trop. Ecol. 13, 243 (1997). 22. J. Terborgh, E. Andresen, J. Trop. Ecol. 14, 645 (1998). 23. C. Hoorn, Palaeogeogr. Palaeoclimatol. Palaeoecol. 105, 267 (1993). 24. M. E. Rasanen, A. M. Linna, J. C. R. Santos, F. R. Negri, Science 269, 386 (1995). 25. M. Rasanen, R. Neller, J. Salo, H. Jungner, Geol. Mag. 129, 293 (1992). 26. M. E. Rasanen, J. S. Salo, R. J. Kalliola, Science 238, 1398 (1987). 27. J. Botschek, J. Ferraz, M. Jahnel, A. Skowronek, Geoderma 72, 119 (1996). 28. P. A. Sanchez, S. W. Buol, Soil Sci. Soc. Am. Proc. 38, 117 (1974). 29. J. Vormisto, O. L. Phillips, K. Ruokolainen, H. Tuomisto, R. Vasquez, Ecography 23, 349 (2000). 30. D. Borcard, P. Legendre, P. Drapeau, Ecology 73, 1045 (1992). 31. We thank numerous colleagues and students from the following institutions for participation in the eldwork and other logistic support: Universidad Nacional de la Amazonia Peruana, Universidad Nacional San Antonio Abad del Cusco (Peru), Ponticia Universidad Catolica del Ecuador, Universidad de Los Andes (Colombia), and University of Aarhus (Denmark). We thank P. Legendre and D. Borcard for inspiring discussions on how to test the various hypotheses and R. Kalliola and L. Schulman for constructive comments on the manuscript. Supported by the Academy of Finland, Commission of the European Union, Ministry for Foreign Affairs of Finland, Suomen kulttuurirahasto, and the Danish Center for Tropical Biodiversity. Supporting Online Material www.sciencemag.org/cgi/content/full/299/5604/241/ DC1 Tables S1 and S2 3 September 2002; accepted 6 November 2002

References and Notes

Fig. 3. Illustration of oristic relationships among western Amazonian sites: Similar sites appear close together, and dissimilar sites are far apart. The principal coordinates analysis is based on oristic similarities ( Jaccard index) as measured with pteridophytes (A and C) and Melastomataceae (B and D). Symbols are used to identify sites from the four different regions [see inset in (A)], and different shades of gray are used to identify sites with different concentrations of soil cations [see inset in (B)]. [(A) and (B)] Analyses with 163 sites, including tierra rme and white sand forests (open symbols) as well as seasonally inundated and swamp forests (solid symbols). [(C) and (D)] Analyses including the 122 tierra rme sites only. The black line separates southern Peruvian sites from sites of the three northern regions.

244

10 JANUARY 2003 VOL 299 SCIENCE www.sciencemag.org

Downloaded from www.sciencemag.org on September 29, 2011