www.sciencemag.org/content/342/6156/1243092/suppl/DC1
Supplementary Materials for
Hyperdominance in the Amazonian Tree Flora
Hans ter Steege,* Nigel C. A. Pitman, Daniel Sabatier, Christopher Baraloto, Rafael P. Salomão,
Juan Ernesto Guevara, Oliver L. Phillips, Carolina V. Castilho, William E. Magnusson, Jean-
François Molino, Abel Monteagudo, Percy Núñez Vargas, Juan Carlos Montero, Ted R.
Feldpausch, Eurídice N. Honorio Coronado, Tim J. Killeen, Bonifacio Mostacedo, Rodolfo
Vasquez, Rafael L. Assis, John Terborgh, Florian Wittmann, Ana Andrade, William F. Laurance,
Susan G. W. Laurance, Beatriz S. Marimon, Ben-Hur Marimon Jr., Ima Célia Guimarães Vieira,
Iêda Leão Amaral, Roel Brienen, Hernán Castellanos, Dairon Cárdenas López, Joost F.
Duivenvoorden, Hugo F. Mogollón, Francisca Dionízia de Almeida Matos, Nállarett Dávila,
Roosevelt García-Villacorta, Pablo Roberto Stevenson Diaz, Flávia Costa, Thaise Emilio,
Carolina Levis, Juliana Schietti, Priscila Souza, Alfonso Alonso, Francisco Dallmeier,Alvaro
Javier Duque Montoya, Maria Teresa Fernandez Piedade, Alejandro Araujo-Murakami, Luzmila
Arroyo, Rogerio Gribel, Paul Fine, Carlos A. Peres, Marisol Toledo, Gerardo A. Aymard C.,
Tim R. Baker, Carlos Cerón, Julien Engel, Terry W. Henkel, Paul Maas, Pascal Petronelli,
Juliana Stropp, Charles Eugene Zartman, Doug Daly, David Neill, Marcos Silveira, Marcos Ríos
Paredes, Jerome Chave, Diógenes de Andrade Lima Filho, Peter Møller Jørgensen, Alfredo
Fuentes, Jochen Schöngart, Fernando Cornejo Valverde, Anthony Di Fiore, Eliana M. Jimenez,
Maria Cristina Peñuela Mora, Juan Fernando Phillips, Gonzalo Rivas, Tinde R. van Andel,
Patricio von Hildebrand, Bruce Hoffman, Eglée L. Zent, Yadvinder Malhi, Adriana Prieto,
Agustín Rudas, Ademir R. Ruschell, Natalino Silva, Vincent Vos, Stanford Zent, Alexandre A.
Oliveira, Angela Cano Schutz, Therany Gonzales, Marcelo Trindade Nascimento, Hirma
Ramirez-Angulo, Rodrigo Sierra, Milton Tirado, María Natalia Umaña Medina, Geertje van der
Heijden, César I. A. Vela, Emilio Vilanova Torre, Corine Vriesendorp, Ophelia Wang, Kenneth
R. Young, Claudia Baider, Henrik Balslev, Cid Ferreira, Italo Mesones, Armando Torres-
Lezama, Ligia Estela Urrego Giraldo, Roderick Zagt, Miguel N. Alexiades, Lionel Hernandez,
Isau Huamantupa-Chuquimaco, William Milliken, Walter Palacios Cuenca, Daniela Pauletto,
Elvis Valderrama Sandovala, Luis Valenzuela Gamarra, Kyle G. Dexter, Ken Feeley, Gabriela
Lopez-Gonzalez, Miles R. Silman
*Corresponding author. E-mail: hans.tersteege@naturalis.nl
Published 18 October 2013, Science 342, 1243092 (2013)
DOI: 10.1126/science.1243092
This PDF file includes:
Supplementary Text
Figs. S1 to S12
Tables S1 to S3
References
Other Supplementary Material for this manuscript includes the following:
available at www.sciencemag.org/content/342/6156/1243092/suppl/DC1
Appendices S1 to S4
3
Supplementary Text
Short
description of our data
The 1170 tree plots used for compositional analyses were distributed between regions and
forest types as shown in Figure 1 and Table S1. The proportion of tree plots in the ATDN dataset
that sample each forest type is roughly equivalent to the proportion of the greater Amazon that
these forest types cover. Várzea and igapó together cover 10% of Amazonia (1, 2) and account
for 19% of our plots. Podzols and Arenosols cover 4.6% of Amazonia (1) and account for 6% of
our plots. Swamps account for 1.8% of our plots, and peatlands are believed to account for
approximately 1.7% of the study area (3).
Main floristic results
We found a total of 4962 valid species, 810 genera, and 131 families in the 1170 tree plots
used for compositional analyses. Fabaceae, not surprisingly, is the most abundant family, with
almost 100,000 individual trees and 119 genera, followed by Arecaceae (52,507; 25),
Lecythidaceae (46,322; 10), Sapotaceae (40,429; 17), Malvaceae (29,424; 36), Burseraceae
(28,762; 7), Chrysobalanaceae (28,597; 7), Moraceae (28,069; 19), Euphorbiaceae (25,955; 42),
and Annonaceae (22,378; 27). Fabaceae are also the most species-rich family, with 795 species,
followed by Lauraceae (311), Annonaceae (289), Rubiaceae (278), Sapotaceae (207),
Chrysobalanaceae (195), Myrtaceae (176), Malvaceae (168), Melastomataceae (168), and
Euphorbiaceae (143). Note that Fabaceae has more than twice as many species as the second
most diverse family. The genera with the largest numbers of individuals were Eschweilera
(31,495), Protium (26,131), Pouteria (21,852), Licania (21,321), Euterpe (14,802), Inga
(14,791), Eperua (10,951), Virola (10,283), Astrocaryum (8973), and Lecythis (8505). The most
species-rich genus was Inga with 134 species, followed by Pouteria (117), Licania (105), Ocotea
(93), Miconia (92), Guatteria (85), Eugenia (76), Protium (69), Swartzia (67), Ficus (59), and
Eschweilera (52).
Finding a fit for the Rank Abundance Distribution (RAD)
We applied a two-step approach to find the best possible RAD for the full Amazonian tree
community. The first step was estimating Amazon-wide population sizes of the 4962 species
occurring in our dataset. The second step was extrapolating the right tail of the resulting RAD to
estimate the number and population sizes of unsampled species (i.e., species that occur in the
Amazon but do not occur in our dataset). This second step required a consideration of the various
different strategies developed over several decades to estimate the total number of species in a
community from a sample (53).
To estimate species richness, statisticians can now choose from a vast array of possible
estimators and distribution models, developed under a theoretical sampling framework (53-57).
This is a daunting challenge (56), “considering the contributions of rare species and the role of
undetected species for a fixed sampling effort” (57). Very sensitive to sample coverage (58),
these methods are mostly designed for local or regional-scale extrapolations. The underlying
distributions are based on some statistical properties of community samples, mostly the
information carried by the number of rare species, first used by Good (59), singletons alone or
4
with other lower abundance taxa, the coverage concept (59, 60) and newly discovered properties
(55, 61).
Community ecologists have long argued over whether the log-series (62), the log-normal
(63), or alternative methods give the best fit for RAD curves. Because extrapolating the total
number of tree species in the Amazon from a very small subsample of the region is a perilous
exercise, we applied at least 11 different extrapolation methods to our data, before concluding
that the log-series was the best. Eighteen estimates from software packages SPECIES (64), and
CatchAll (54) are shown in Table S2.
Sixteen of these 18 methods can be immediately rejected, since they predict the total number of
Amazonian tree species to fall in the range 4015-6412. This is a demonstrably severe
underestimation of the true species richness, since previous estimates of tree species richness in
Amazonia and the Neotropics (10
6
km
2
) based on floras and expert opinion were around 12,500
and 22,500 respectively (1, 65), consistent with our estimate of ca. 16,000 for Amazonia. A new
estimator recently implemented in CatchAll( WLRM_UnTransf) (54, 61) gave an estimated total
richness above 11,000, closer to that calculated with our log-series extrapolation (16,000), but
was not selected by the program as the best estimator. The ACE1_Max Tau estimator gave a
result greatly exceeding the expected richness and its Tau was much higher (9048) than its
recommended value (Tau < 10).
The failure of these models to fit our data is not surprising. Brose et al. (58) noted that
sampling-theoretical methods of estimation require high sampling intensity to avoid what Wang
et al. (55) call the “severe under-estimation observed from popular nonparametric estimators due
to the interplay of inadequate sampling effort, large heterogeneity and skewness”. We believe
that these estimators performed poorly because their assumptions are not met by our system.
Mostly, they measure the expected number of species at a local site, and assume relatively
complete sampling (see eg. (56)). However, we are attempting to estimate the total number of
species across the full Amazon (6 million km
2
) rather than at any one site, and our sampling
intensity is very low (ca. 0.002%).
There is no consensus in the abundant literature on which predictor is the most efficient, nor
regarding the choice of a parametric distribution. There are, however, two important points in
our case study: 1) we were able to estimate the community size, a dimension generally ignored
by estimators and 2) an empirical approach allowed us to build an estimated RAD of the most
abundant species for the total area (i.e. Amazon) - the model of abundances distribution all
methods try to estimate.
The log series fits our data well (Fig. S6) and also fits well the left part of the Amazonian
RAD we obtained by estimating the populations sizes of Amazonian trees (Fig s7). So the
assumption that this is the best form for the RAD of Amazonian trees (1) can actually be met.
Based on the above we concluded that the log series was the best fit for our data and based
our species richness estimate on this distribution.
5
Estimating
species richness with Fisher’s alpha
If species were randomly distributed across Amazonia and we sampled at random
throughout that area, our relative abundance distribution would have the same form and the same
Fisher’s alpha as the Amazonian RAD. Fisher’s alpha would also reach an asymptote after a
sufficiently large sample had been made. Because conspecific trees are clumped at various
spatial scales (due to seed dispersal, preference for soil types) and our sampling was not random,
our RAD differs in some respects from the true Amazon-wide RAD. Specifically, it
underestimates Fisher’s alpha and therefore provides an underestimate of gamma diversity (Fig.
S8).
Hyper-dominant species by plot and forest
The median percentage of individuals that belong to hyper-dominant species within an
individual plot was 40.7% (range = 0-93.9%, Fig. S9). Comparable figures for the five forest
types are: igapó 32.9%, white sand forest 43.6%, swamp 35.9%, várzea 34.7%, and terra firme
30.1%. The median number of hyper-dominant species was 32 per plot (range = 0-78. The 438
plots containing fewer than 20 hyper-dominant species were evenly distributed across Amazonia
but not across forest type. Only 15.1% of all terra firme plots have less than 20 hyper-dominants.
For the other forest types the percentage is: igapó 76.6%, white sand forest 78.9%, swamp
50.0%, várzea 42.8%.
Species richness by country
Our data provide estimates of the number of tree species occurring in each country in the
study area (i.e., in the Amazonian portions of Bolivia, Brazil, Colombia, Ecuador, and Peru; in
the three Guianan countries, which were pooled for this exercise; and in the Guianan and
Amazonian portions of Venezuela; see Fig. 1) by constructing a Rank-Abundance Distribution of
the estimated populations of all species predicted to occur in a country (Fig. S10). Population
sizes were estimated by summing the number of trees of each species in all the 1-degree grids
cells whose centroids were in that country.