Skip to main content

Data for: Resolving whole-plant economics from leaf, stem and root traits of 1467 Amazonian tree species

Cite this dataset

Vleminckx, Jason (2021). Data for: Resolving whole-plant economics from leaf, stem and root traits of 1467 Amazonian tree species [Dataset]. Dryad.


It remains unclear how evolutionary and ecological processes have shaped the wide variety of plant life strategies, especially in highly diverse ecosystems like tropical forests. Some evidence suggests that species have diversified across a gradient of ecological strategies, with different plant tissues converging to optimize resource use across environmental gradients. Alternative hypotheses propose that species have diversified following independent selection on different tissues, resulting in a decoupling of trait syndromes across organs.

To shed light on the subject, we assembled an unprecedented dataset combining 19 leaf, stem and root traits for 1467 tropical tree species inventoried across 71 0.1-ha plots spanning broad environmental gradients in French Guiana.

Nearly 50% of the overall functional heterogeneity was expressed along four orthogonal dimensions, after accounting for phylogenetic dependences among species. The first dimension related to fine root functioning, while the second and third dimensions depicted two decoupled leaf economics spectra, and the fourth dimension encompassed a wood economics spectrum. Traits involved in orthogonal plant functional strategies, five leaf traits in particular but also trunk bark thickness, were consistently associated with a same gradient of soil texture and nutrient availability. Root traits did not show any significant association with edaphic variation, possibly because of the prevailing influence of other factors (mycorrhizal symbiosis, phylogenetic constraints).

Our study emphasises the existence of multiple functional dimensions that allow tropical tree species to optimize their performance in a given environment, bringing new insights into the debate around the presence of a whole plant economic spectrum in tropical forest tree communities. It also emphasizes the key role that soil heterogeneity plays in shaping tree species assembly. The extent to which different organs are decoupled and respond to environmental gradients may help to improve our predictions of species distribution changes in responses to habitat modification and environmental changes.


Study area

Tree species inventories were carried out in ten lowland and lower mountain sites covered by mature tropical moist forests across French Guiana (Fig. 1a), at the eastern edge of the Guiana shield. Mean annual rainfall (calculated over the 2010-2018 period) across inventory sites ranges between 2150 and 3700 mm and is distributed seasonally throughout the year (Gourlet-Fleury et al. 2004; Table 1), with a dry season occurring between August and November (monthly rainfall ≤ 100 mm) and usually more pronounced toward the interior of the continent. Mean annual temperature (2010-2018) oscillates around 25°C, with low seasonal variation, and averages at 22°C in a relatively higher site (> 500 m a.s.l.) located at Mont Itoupé in the centre of French Guiana. Each of the ten inventory sites comprised two to 12 0.1-ha plots (71 plots in total), separated by at least 500 m, and located on contrasted habitats: (i) terra firme, (ii) white-sand and (iii) seasonally flooded forests. Additional details regarding the description of the study area are available in Appendix S1.

Tree inventories

We used a modified version of the Gentry plots proposed by Phillips et al. (2003) and described in Baraloto et al. (2013), consisting of ten parallel transects of 2 × 50 m departing perpendicularly every 20 m from a 200 m central transect, successively oriented in alternate directions, and delimiting an area of 1.9 ha. Within each transect, all stems with a circumference at 1.3 m above soil level > 8 cm (corresponding to a DBH of ca. 2.5 cm) were inventoried. Voucher specimen were collected at least once for each putative distinct species per plot in the field, plus additional vouchers for individuals that were not completely identified in the field. Duplicate vouchers are currently stored in reference collections of UMR EcoFoG (Kourou, French Guiana), UMR AMAP (Montpellier, France), and/or ICTB (Miami, USA). A list of species with their voucher reference is available in Appendix S2. Our inventories resulted in a dataset comprising 13,736 trees belonging to 1,467 species (excluding 25 palm and fern species for which we could not sample whole plant functional traits), 348 genera and 81 families, with an average of 79 species (± SD = 28) per plot (additional details regarding tree diversity and climatic characteristics in each study site are provided in Table 1).

Functional traits

We assembled a dataset of 19 functional traits for the 1467 tree species (excluding ferns and palms) inventoried in our 71 plots, including 11 leaf traits, two stem traits, one coarse root trait and five fine root traits (Table 2). Details on trait measurement protocol are provided in Appendix S1. Trait values were obtained from a dataset comprising measurements from tissue samples collected on 8345 individual trees belonging to 1625 species distributed in 371 genera, 78 families and 26 orders in the Rosidae, Asteridae and early eudicots (Appendix S3). 5735 of these individuals (68.7%) corresponded to trees inventoried in our 71 plots and represented 783 out of the 1467 inventoried species (53.4%). Additional measurements were made on tissue samples collected, respectively, on 1746 and 858 individuals from the Brazilian Amazon (near Manaus) and Peru (representing, respectively, 248 and 541 species, among which 119 and 60 were also observed in French Guiana). We assumed that trait values for the Brazilian and Peruvian samples were representative of the species present in French Guiana (Ackerly 2003, Crisp et al. 2012). We deem this assumption reasonable, based on positive and highly significant correlations (t-test of Pearson’s product moment) of species mean trait values between French Guiana and the two outside regions (Brazil and Peru), for 12 out of 18 traits that could be compared (two traits could not be tested due to sampling limitations; see details in Appendix S1).

The imputations were performed using the matrix of 8345 individuals (belonging to 1625 species) x 19 traits (i.e., the NBA matrix). 779 out of the 1467 species of our French Guiana inventories were represented in this matrix. We therefore added 688 lines (1467-779) of empty trait values, corresponding to the species that were not represented in the NBA matrix, therefore producing a matrix of 9033 individuals and 2313 species on which we performed the BHPMF imputations. The percentage of missing values in this trait matrix ranged between 34.96% for the leaf area up to 87.90% for the coarse root wood-specific gravity (see Appendix S1 for details). Although the latter percentage was high, fine root traits were measured on species covering a wide phylogenetic range (31 families and 9 orders in the Asteridae, Rosidae and Magnoliidae), while previous studies have suggested that phylogenetic trait conservatism represents a major determinant of root functional traits (Valverde-Barrantes et al. 2017). Similarly, woody traits, which showed proportions of missing values that exceeded 65%, have shown strong conservatism (Chave et al. 2006).

To fill the missing trait values, we used the Bayesian Hierarchical Matrix Factorization method (BHMF, Fazayeli et al. 2014), an imputation procedure based on both (i) covariations among traits and (ii) trait information at higher taxonomic levels (species, genus, family, order, class), in a hierarchical way. The BHPMF method has been proved efficient in assessing missing trait values even with a percentage of missing values higher than 90%, providing that there is a good phylogenetic coverage and/or a sufficient number of traits measured among species (Schrodt et al. 2015). This was the case with our traits as they were measured on species covering a wide phylogenetic range, while there were on average 11.3 ± 4.0 (standard deviation) traits measured per species. Prior to the imputations, outliers in the distribution of each trait were eliminated following Zuur et al. (2010) as the BHMF method is sensitive to extreme values, then traits were normalised (Box-Cox transformation) and standardised (z-score transformation). We then detrended each trait with the height of individuals to remove any ontogenetic variation effect, which is rarely taken into account in ecological studies, by regressing each trait on height and using the residuals of each regression as our trait values. Imputations were then calculated using the GapFilling function in the R BHPMF package (Fazayeli et al. 2014), using observed information available at the genus, family, order, subclass and class level to estimate missing trait values. We then extracted the imputed values at the individual level and calculated a matrix of mean trait values at the species level for the 1467 species of our French Guiana inventories. Post Hoc analyses were carried out to evaluate the reliability of our imputations. These analyses showed that the whole correlation structure among traits was well preserved after imputation (Appendix S1).

Environmental data

Soil sampling and analyses were carried out using the protocol described in Baraloto et al. (2011). We collected ten bulked soil cores using 5 cm-diameter auger at 0-10, 10-20 and 20-30 cm depth in each plot, at each intersection between the ten parallel transects and the main central line. For each plot, the ten cores were then combined into a composite 500 g sample which was air-dried then sieved across 2 mm mesh and shipped all together for physicochemical characterisation at CIRAD soil lab (France) using standard soil analysis protocols (Pansu & Gautheyrou 2006). We retrieved nine soil variables, including two physical variables corresponding to soil texture (percentages of sand and clay), and seven chemical variables: soil organic carbon content, C:N ratio, available phosphorus (P), total soil Nitrogen content (TN), and the availability of three base cations (Ca, Mg and K). The other variables used in our analyses comprised the elevation a.s.l. of each plot and three climatic variables calculated using data extracted from, via the R package raster (Hijmans 2019): the mean annual rainfall (mm), the standard deviation of the mean monthly rainfall calculated over 12 months (to quantifies the unevenness of precipitation throughout the year). A Dry Season Index (DSI) was also calculated for each site as the sum (over 12 months) of the ratios between mean monthly temperature and mean monthly rainfall to estimate of the potential water stress accumulated during the dry seasons.