Legacy of Amazonian Dark Earth soils on forest structure and species composition

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. © 2020 The Authors. Global Ecology and Biogeography published by John Wiley & Sons Ltd 1Programa de Pós-Graduação em Biodiversidade e Biotecnologia da Rede BIONORTE, Universidade do Estado de Mato Grosso – UNEMAT, Nova Xavantina, Mato Grosso, Brazil 2Archaeology, College of Humanities, University of Exeter, Exeter, UK 3Ecosystem and Landscape Dynamics/ Institute for Biodiversity and Ecosystem Dynamics, University of Amsterdam, Amsterdam, the Netherlands 4Instituto Federal de Mato Grosso, Primavera do Leste, Mato Grosso, Brazil 5Geography, College of Life and Environmental Sciences, University of Exeter, Exeter, UK 6Divisão de Sensoriamento Remoto, Instituto Nacional de Pesquisas EspaciaisINPE, São José dos Campos, São Paulo, Brazil 7Programa de Pós-Graduação em Biodiversidade e Biotecnologia da Rede BIONORTE, Universidade Federal do AcreUFAC, Rio Branco, Acre, Brazil

One of the most compelling lines of evidence for widespread anthropogenic influence comes from the presence of the anthropogenically made Amazonian Dark Earth (ADE) soils (Glaser et al., 2000;. Previous studies from the eastern Amazon have shown that pre-Columbian crop cultivation and agroforestry altered the modern composition, enriching modern ADE forests in edible plant species (Maezumi et al., 2018).
Following the arrival of European colonialists in the 15th and 16th centuries, the pre-Columbian populations were estimated in the millions (Nevle, Bird, Ruddiman, & Dull, 2011). This population rapidly declined by up to 90% due to introduced diseases, with entire civilizations permanently lost, leaving behind the legacy of enriched ADE soils (Koch, Brierley, Maslin, & Lewis, 2019;Nevle et al., 2011).
Today, ADEs continue to be used by local farmers for planting, given their high fertility (Clement, McCann, & Smith, 2003). ADEs are formed from the anthropic addition of organic matter, household wastes, ceramics, and charcoal (Sombroek et al., 2002). Organic matter and nutrients from household waste are retained in the soil through chemical-physical interactions with pyrogenic carbon, which enhances soil fertility (Kämpf, Woods, Sombroek, Kern, & Cunha, 2003;Lehmann, Pereira da Silva, et al., 2003). This effect is due to the properties of macro-and microscopic pyrolysed carbon formed from incomplete combustion of biomass during burning (Glaser, Haumaier, Guggenberger, & Zech, 2001). This fertilization process contributes to the soil retention/availability of water and nutrients, conferring advantages for agriculture (César et al., 2011) and native forest productivity (Aragão et al., 2009) over Amazonian dystrophic soil types, such as nutrient poor latosols.
As a result of the anthropogenic enrichment of ADE soils, the forests growing on abandoned ADEs may be characterized by different growth and structure (Aragão et al. 2009), such as lower and more closed canopies and more understorey trees (Sombroek et al., 2002).
ADE forests also allocate more carbon to plant biomass gain than non-ADE (NDE) soils (Doughty et al., 2013). ADE sites can therefore be considered a long-term fertilization experiment to test legacy effects on current native vegetation of landscape management | 3 de OLIVeIRA et AL. by ancient human populations (Cook-Patton, Weller, Rick, & Parker, 2014). To date, there is little information to determine which species occur in these environments at a broader scale or whether ADEs can accumulate more species due to their stability, productivity and fertility (Aragão et al., 2009;Cunha et al., 2007;Glaser et al., 2000).
Diameter distributions can reveal patterns of tree species dynamics, e.g., whether a forest is recovering from disturbance (Lima, Bufalino, Alves Júnior, Silva, & Ferreira, 2017), and can be an important tool to detect some legacy of old land-use in forest structure, for the ADEs' land-use legacy can remain recorded in the structure of present-day forests (Junqueira, Shepard, & Clement, 2011;Woods & McCann, 1999).
Additionally, the edaphic changes associated with ADE soils that increase fertility and water retention can create distinct habitats that persist for centuries after abandonment (Glaser et al., 2001).
Some studies suggest that forests growing on ADEs can be compositionally and structurally distinct from surrounding vegetation (Clement et al., 2009;Junqueira, Shepard, & Clement, 2010;Palace et al., 2017) and may contribute to the diverse and heterogeneous tree flora of Amazonia (Aragão et al., 2009;Sombroek, 1966). To date, understanding the mechanisms that determine the composition, structure and diversity of Amazonia forests is a major challenge (Bicudo, Sacek, Almeida, Bates, & Ribas, 2019;Hoorn et al., 2010;Levis et al., 2017). While studies have suggested the importance of past land-use as a factor in increasing the diversity and distribution of current species (Cook-Patton et al., 2014;Levis et al., 2017), it is unclear whether edaphic factors associated with ADE soils also influence species richness in Amazonia.
The aim of this study was to evaluate the potential legacy of ADEs and their contribution to the structure and floristic composition of modern forests with the following questions: (a) Do forests that grow on ADEs have different richness and floristic composition than adjacent forests? (b) Does forest structure and aboveground biomass differ between ADEs and NDEs? (c) Is the diversity of edible species in ADE forests greater than in NDE forests? Our study takes a novel approach by studying ADE and NDE forests in two distant regions, with contrasting environments and occupation history, to evaluate whether differences between ADE and NDE forests that were already identified in local studies are consistent across broader scales. We also focus on mature forests growing on ADEs, which are less studied than secondary forests or agroecosystems growing on ADEs.
The forests in southern Amazonia are seasonal evergreen, a forest type found in seasonal climates with a dry season longer than 120 days (Ivanauskas, Monteiro, & Rodrigues, 2008). These forests have little leaf loss during the dry season, which differs from the semi-deciduous or deciduous seasonal forests of eastern Brazil (Oliveira-Filho & Ratter, 1995). The forests of eastern Amazonia are classified as terra firme dense rain forest (IBAMA, 2004, sensu RADAMBRASIL, 1982. This phytophysiognomic type represents most vegetation of northern Brazil and is characterized by a closed canopy, large individual trees, and high temperature and humidity (Veloso, Rangel-Filho, & Lima, 1991 Information Table S1). All climatic variables were extracted from WorldClim (Hijmans, Cameron, Parra, Jones, & Jarvis, 2005). Some regions of Pará and Mato Grosso have a long history of pre-Columbian occupation. In the Xingu River region (Mato Grosso state), dozens of pre-Columbian settlements have been documented, fortified by ditches and connected by a regional network of roads in a 'galactic' system of regional polities extending over 20,000 km 2 that peaked around c. 750-500 cal bp (Heckenberger et al., 2008).
The Santarem region (Pará state) has one of the earliest occupation histories in Amazonia, with archaeological evidence of Palaeoindian occupation of Pedra Pintada Cave (c. 13,000 cal bp; Roosevelt et al., 1996). Later Archaic occupations in the region are represented by the Taperinha shell-midden, which contains the earliest ceramics in Amazonia (c. 7,900 cal bp; Roosevelt, Housley, Silveira, Maranca, & Johnson, 1991). The Formative Period occurred between c. 4,500 and 1,000 cal bp (Gomes, 2011) followed by the Late pre-Columbian Tapajó Period (LPTP) from c. 1,000 to 400 cal bp (Stenborg, Schaan, & Amaral-Lima, 2012). The LPTP exhibits extensive landscape modifications including the development of ADE sites, a network of ditch and causeway trails connecting sites, and numerous water collecting pools constituted by natural pools that have been artificially ADE patch size in Amazon varies from less than one hectare to several hundred hectares, with most of the recorded sites < 2 ha . ADEs are estimated to cover c. .1-.3% (6,000-18,000 km 2 ) of the forested portion of the Amazon Basin (Woods & Denevan, 2009), although the predictive model used by McMichael et al. (2014) estimated that ADEs may cover as much as 3.2% (approximately 154,063 km 2 ).
ADEs are generally found along major rivers, and in certain regions, such as Santarém  and on the middle Purus and Madeira Rivers (Levis et al., 2012), are also found in interfluvial areas. In the central and lower portion of Amazonia, ADE formation began around or after 2,000 cal bp (Arroyo- Kalin, 2010;Heckenberger & Neves, 2009;Maezumi et al., 2018), although ADEs appear to be older in the Upper Madeira River (Miller, 1992). After the Columbian Encounter around 500 cal bp, indigenous populations began to decline (Denevan, 1992) and indigenous land-use on ADEs was largely abandoned.

| Soil sampling and analyses
ADE soils were characterized according to , being recognized for the high amount of pyrogenic carbon (charcoal, incompletely burned organic matter) and archaeological artifacts, mainly shards of ceramic and lithic material found in soil pits (Costa, Kern, Pinto, & Souza, 2004;Denevan, 1992;Glaser et al., 2001;Kern et al., 2003). In each 100 m × 100 m plot in southern Amazonia, we collected soil samples at 0-10 and 10-20 cm depth from five random points. In the 50 m × 50 m plots of eastern Amazonia, we collected a single soil sample at 0-10 and 10-20 cm depth at the centre of each plot. Samples were air-dried at ambient temperature and analysed for physical

| Vegetation sampling, biomass estimates, useful species
In each study plot, we sampled all live woody plants (including palms, trees and lianas) with diameter (D) ≥ 10 cm at breast height (1.3 m) and measured their heights with a Leica (Heerbrugg, Switzerland) DISTO TM D5 laser measurement device. The lianas were measured at 1.3 m F I G U R E 1 Plot locations in forests in southern Amazonia represented by the state of Mato Grosso (MT) and eastern Amazonia represented by the state of Pará (PA), Brazil. The black line in the inset indicates the boundaries of Brazilian states. The areas in Pará overlap and, due to their proximity, they are represented as a single dot. Archaeological sites -ADE (Amazonian Dark Earth) data sourced from the National Register of Archaeological Sites (CNSA) (IPHAN, 2018) along the stem. We identified species in the field and when necessary collected vouchers for confirmation by specialists. The material was deposited in the NX Herbarium, Nova Xavantina, Mato Grosso State.
We determined the aboveground biomass (B) of each tree based on a pantropical model (Chave et al., 2005) B = .0509 × (ρD 2 H); where D is individual diameter (cm) at 130 cm height or above deformities and buttress roots, H is total tree height (m), and ρ is wood density (g/cm 3 ) compiled from the DRYAD global database Zanne et al., 2009) and summed biomass per hectare (Mg ha -1 ). Where individual tree height estimates were lacking, we estimated tree height using the pantropical allometric equation [H = a(1 − exp(−b D c )) by Feldpausch et al. (2012), where the coefficients a, b and c were determined for each region (e.g., eastern/central and southern Amazonia)].
We classified tree species as 'useful species' (medicinal, food, construction, and fibre) and edible based on the literature (Clement, 1999;Junqueira et al., 2010Junqueira et al., , 2011Levis et al., 2017Levis et al., , 2012López Zent & Zent, 1998;Maezumi et al., 2018;Prance, Balée, Boom, & Carneiro, 1987). Due to the lack of information on which species indigenous people previously used in the Cerrado-Amazonia forest zone of transition, we used the same classification for useful species for both southern and eastern Amazonian sites.

| Statistical analysis
To evaluate the physicochemical properties of the ADE and NDE soils in the two regions, including grain size (clay, silt and sand), We assessed the sampling effort (rarefaction curves) based on the interpolation and extrapolation method in iNEXT (Chao et al., 2014;Hsieh, Ma, & Chao, 2016). This analysis enabled the calculation of the richness of samples by extrapolation of the plot abundance based on plot size. We compared local communities and estimated the richness of each area generating value for estimators incidence coverage-based estimator, abundance coverage-based estimator (ACE), Chao1 Chao2, Jackknife1 and Jackknife2, whose values were obtained from 1,000 randomizations in the program EstimatEs 8.0 (Colwell, 2008). The estimators were selected using abundance data following Hortal, Borges, and Gaspar (2006). These estimators infer the richness when plot size is unequal and/or small, e.g., Chao 1 and ACE are highly precise regarding variation in sample size. However, the most precise estimator was selected using the highest R 2 value from a regression analysis between the observed and estimated values (Brose, Martinez, & Williams, 2003).
We determined tree species diversity using the Shannon index (H') and evenness with the Hurlbert index (probability of interspecific encounter, PIE), in the program Ecosim 7.0 (Gotelli & Entsminger, 2001). To test whether the local diversity varied between ADE and NDE forests, we used a PerMANOVA (Anderson, 2001) based on 1,000 randomizations. We consider the estimated richness (S), Shannon index (H') and Hurlbert evenness index (PIE), highly dependent values, as a measure of local diversity. To minimize sampling bias, we determined the values estimated for S, H' and PIE using the rarefaction method (1,000 randomizations) in the program Ecosim 7.0 (Gotelli & Entsminger, 2001), taking as reference the community with the lowest abundance.
We used one-way ANOVA to compare the total aboveground biomass between ADE and NDE forests at a local scale (southern and eastern Amazonia), and a t-test to compare the biomass between forest types (ADE versus NDE) regardless of the region (Legendre & Legendre, 1998). We tested the normality of residuals and ho- Due to the small number of individual trees in the class > 40 cm in ADE and NDE forests of Pará (eastern Amazonia), we used the three plots together (totaling .75 ha ).
We evaluated dissimilarities in floristic composition and species abundance in ADE and NDE forests using non-metric multidimensional scaling (NMDS; Legendre & Legendre, 1998). The sampling deficit (size of plots) can affect comparisons of richness and floristic composition between areas because the smaller plots harbour only a subset of the regional floristic composition and a reduced number of individuals (Chao & Jost, 2012). Thus, we calculated the dissimilarity expected for a rarefied community considering the smallest number of individuals sampled among all communities (53 living individuals) and applied the extended dissimilarities using an extended version (path = "extended") with the vegan package in the R program (R Development Core Team, 2018). The dissimilarity was also calculated with the Raup-Crick probabilistic estimator considering probability of occurrence greater than zero as presence and equal to zero as absence based on the rarefied matrix. The matrix was ordered using the meta MDS function of the vegan, a nonmetric multidimensional scalin function with stable solutions from random starts, axis scaling and species scores (Oksanen et al., 2018). The result of the analyses showed little or no 'arc effect', while keeping the groupings of the sites homogeneously distributed within the different areas, congruent with the analysis based on the Bray-Curtis dissimilarity. Thus, we maintained NMDS based on the Bray-Curtis dissimilarity matrix.
Our general objective was to compare key abiotic (soil) and biotic (vegetation) attributes between ADEs and NDEs forests, including (a) soil; (b) vegetation richness, diversity and evenness; (c) aboveground biomass; (d) species composition and size classification; (e) useful and edible species. To gain further insights, we conducted these comparisons at two organizational scales: plot and region. Given that we conducted multiple comparisons using the same data, but organized at different scales, we used Bonferroni-adjusted p-values to take account of multiple tests of the PerMANOVA. We divided the level of significance adopted (.05) by the number of statistical tests performed (three) (Dunn, 1961). For these tests, we adopted a p-value of .0167. However, as the p-values were obtained based on randomization, we suggest using both approaches (original and

Bonferroni-adjusted p-values).
All tests where a specific program was not stated were executed in R (R Development Core Team, 2018), with the vegan package (Oksanen et al., 2018).

| Soil and vegetation patterns of ADEs
In general, ADE forests had significantly higher pH and fertility, with P, K, Ca, Mg (magnesium), OM, SB and CEC all higher than in NDE forests. Ca:Mg, Ca:CEC, Mg:CEC ratios were also higher in ADE forests. Only Al and the Al 3 + H:CEC ratio were lower for this type of soil. The soils of eastern Amazonia had higher clay and silt content than the sandier soils of southern Amazonia (see Supporting   Information Table S2 and the PerMANOVA results in Supporting   Information Table S3).  Information Table S4).
After accounting for sampling effort, we recorded a marked increase in rarefaction curves for ADE and NDE-E forests ( Figure 3) and a moderate increase in ADE and NDE-S forests (Figure 3).  (Table 2). At the forest level, in general, we observed low similarity between the first (smallest) and the third (largest) diameter classes (Supporting Information Tables S6 and S7). However, there was no significant difference based on the Bonferroni p-value.
Four distinct groups were formed in the NMDS analysis (see sampling coverage in Supporting Information Figure S1): the forests of eastern Amazonia were clearly separated from the forests of southern Amazonia on axis 1 (r 2 = .35) and ADE and NDE forests of each region individually were grouped on axis 2 (r 2 = .48) ( Figure 5 and Supporting Information Figure S1), with the two axes representing 83.3% of the proportion of the variance in the original distance matrix.
We found a high number of useful and edible species in ADE and NDE forests in both regions. The proportion of edible species tended to be higher in the ADE forests of eastern and southern Amazonia (Table 3). In eastern Amazonia, we found 31% edible species in NDE forests and 33% in ADE forests. In southern Amazonia, we observed 25% edible species in NDE forests and 31% in ADE forests (Table 3 and Supporting Information Tables S8 and S9). Among them, the useful species Hymenaea courbaril (West Indian locust = fruit/food) and    1994). The southern Amazonian vegetation growing on dystrophic soils has been considered hyperdynamic (region with high dynamism from major intra-and inter-year climate variation), with lower species richness (Marimon et al., 2014); in contrast, forests located on higher fertility soils of Andean-Amazon lowland forests are dynamic but have higher species richness. Low species richness may result from more complex trophic interactions in environments with resource limitations, such as water and nutrients (Huston, 1980).

| D ISCUSS I ON
Our study sites in both southern and eastern Amazonia showed neither a positive nor negative soil fertility effect on the richness richness between 165 and 300 per hectare on the more fertile soils of the region. In the case of a negative effect, the fertilization could be favouring the growth of species that better compete on mesotrophic soils, and therefore, decreasing species richness, as predicted in the enrichment paradox (Rosenzweig, 1971), as observed in experiments with artificial fertilization (Tilman & Isbell, 2015), and in field observations in tropical forests (Huston, 1980;Nadeau & Sullivan, 2015). In our case, the assembly of local species may be allowing a Annonaceae families in our study could be related to their adaptability and early appearance in South America (Doyle & Le Thomas, 1997;Doyle & Luckow, 2003;Lavin & Luckow, 1993 Laurance et al., 1999).

TA B L E 2 Permutational multivariate analysis of variance (PerMANOVA) comparing the floristic matrix between diameter classes in Amazonian Dark Earth (ADE) and non-Dark Earth (NDE) forests in southern (ADE-S and NDE-S) and eastern Amazonia (ADE-E and NDE-E), Brazil, including the F-statistic, R 2 and p-value
However, as we observed, there is high variation in biomass values in different regions in Amazonia, in part due to differences in tree height (Feldpausch et al., 2012). Hymenaea courbaril as edible], usually explained by the low similarity between the two regions. We observed a moderate number of useful species at a local scale. The slightly higher number of useful species in the ADE (proportionally) in eastern Amazonian plots may be due to the differences in how ADE and NDE sites were historically managed by pre-Columbian people or by the distance between the studied areas (Maezumi et al., 2018). The relatively small ADE areas were often used for annual crops, while the surrounding area may have been enriched in useful tree species (Paz-Rivera, 2009). There is evidence of enrichment of edible plants in ADE soils (Maezumi et al., 2018). Some factors that could explain the low similarity be- (2019) argued that differences in landuse and socio-political organization may be key to understanding vulnerability versus resilience to environmental stress; by comparing archaeological data with data from palaeoclimate proxies and regional-scale burning, they showed that some Amazonian cultures flourished during periods of climate change, whereas others collapsed. Defining 'useful' and accounting for regional variation in use are challenges since a large number of useful species are used by different indigenous populations in other regions of Amazonia (Boom, 1985;Prance et al., 1987), with up to 82% of the species with different utility levels occurring in only 1 ha.
In the present study, this value varied between 57 and 60%.
Studies combining archaeology, ecology and botany identified variation in early practices that may have resulted in regional variation in the creation, use and subsequent post-abandonment regrowth of ADEs. An interdisciplinary study demonstrated that pre-Columbians enriched the forests c. 2,500 years ago with above 30% increase in edible plants that persist to present (Maezumi et al., 2018). However, we do not yet know which species have distributions influenced by humans (Gordon, Barrance, & Schreckenberg, 2003). A well-known example is the strong relationship between the distribution of the Brazil nut (Bertholletia excelsa) and the range of human populations in the Amazon Basin (Levis et al., 2017;Mori & Prance, 1990;Shepard & Ramirez, 2011). Therefore, the distribution of other species with potential fibre, medicinal and food use, such as

| CON CLUS ION
In ADE forests, the absence of long-term soil fertilization effects on local diversity (species number) points to the importance of the regional species pool in determining the diversity at plot scales.
However, the dissimilarities in floristic composition between ADEs and NDEs indicate a contribution at regional scales to species richness. Moreover, the marked difference in species composition and structure (biomass) between ADE and NDE forests shows that soil fertility can influence other community attributes in Amazonian forests. Therefore, it is important to consider the role of changes in nutrient levels at different scales and ecosystems (e.g., forest burning). In addition, the differences in composition and abundance linked to tree diameter classes suggest a legacy influence of historical land-use and soil enrichment in ADE on the structure of ADE for-