Introduction

The importance of terrestrial ecosystems for the modern carbon (C) cycle has been emphasized in several modelling exercises (Friedlingstein et al. 2006; Yue et al. 2020). Feedback from terrestrial ecosystems on increased atmospheric CO2 are key components in monitoring and mitigating the effects of on-going climate changes (Schimel 1995; Schimel et al. 2015a; Pugnaire et al. 2019). In terrestrial ecosystems, the importance of plants is paramount since the coupled photosynthesis-respiration processes modulate biosphere-atmosphere C exchanges. Recent simulations indicate that the land component of the global C budget has been an increasing C sink, equivalent in 2018 to roughly one-third of the total fossil fuel emissions due to CO2 and N fertilization (Friedlingstein et al. 2019). However, the uncertainty linked to this estimate is still large mainly because the spatiotemporal variability associated with the terrestrial gross primary production (GPP; Anav et al. 2015; Rödig et al. 2018). The largest values of GPP are not surprisingly observed in the tropics due to energy and water availability (Anav et al. 2015). At the same time, the role of the tropics in the modern global C cycle is also important due to the net balance of land-use changes, increasing the land-atmosphere CO2 flux (van der Werf et al. 2017; Duvert et al. 2020; Souza et al. 2020).

It is well recognized that availability of water and energy are important drivers of plant richness at the regional and continental scales (e.g., Currie and Paquin 1987; Field et al. 2005). Most of the Brazilian territory is located within the southern tropical zone between the Equator and the Tropic of Capricorn (23.5°S). Due to Earth’s sphericity, this is a region of the planet that receives year-round abundant solar energy, where the precipitation in most of the country is >1300 mm year-1. As a result, Brazil stands as a megadiverse country regarding its fauna and flora. According to the Brazil Flora Group (2015), more than 30,000 species of angiosperms alone have been cataloged. Amazonia (49% of the country area) and the Atlantic Forest (13%) biomes are well known for luxurious tropical rainforests (Morellato and Haddad 2000). Although in most of the country rainfall is abundant, some regions like the Cerrado (24%) display remarkable seasonality (Eiten 1972), while the Caatinga (10%) is characterized by a negative water balance (evapotranspiration higher than precipitation), being the only truly semiarid region of the country (Oliveira et al. 2019). In the Pantanal (2%), seasonal flooding of lowlands (80% of the area) hinders the development of dense forest formations over broad areas (Evans et al. 2014). The Pampas (2%) has a temperate and subtropical climate, having the lowest temperatures in the country, which together with fires and grazing prevents the development of extensive forests (Overbeck et al. 2007).

In each biome, there is an iconic vegetation type that covers most of the area. Rainforests are the iconic vegetation of Amazonia and Atlantic Forest, while different physiognomies of savannas, dry forests and grasslands are predominant in the Cerrado, Caatinga, Pampas, and Pantanal. However, other vegetation types also occur in each biome, like white-sand forest vegetation, locally called Campina, as well as patches of savanna-like vegetation encroaching into several areas of Amazonia. Savanna-like vegetation is also common in the Atlantic Forest. Even in the Cerrado, there are several savanna physiognomies types: campo limpo (open grasslands), campo sujo, cerrado and cerradão (woody-savanna). In the Pampas, there are the Araucaria forests, which are evergreen temperate forests dominated by the gymnosperm Araucaria angustifolia.

Brazilian biomes are under severe pressure by human-induced land use and land cover changes: in the last 30 years the country lost 71 Mha of natural vegetation (Souza et al. 2020). As a country characterized by an economy largely dependent on the production of raw commodities, Brazil pursues a fragile balance between the exploitation of natural resources and preservation (DeFries et al. 2004). Amazon deforestation is an iconic example that represents the dichotomy between agriculture expansion and conservation (Pereira et al. 2019). Those biomes near the coast (Atlantic Forest and Caatinga), the first to be settled by Europeans, suffered the most extensive changes (Souza et al. 2020). Only 15–20% of the original cover of the Atlantic Forest is left, leaving a highly fragmented landscape with several consequences to the biota and ecosystem functions (Tabarelli et al. 2004; Souza et al. 2020). Chronic land-use and cover changes in the Caatinga has already led to severe biological impoverishment of plant species (Ribeiro et al. 2015); and the Cerrado, considered an agriculture frontier in the 1960s and 1970s, has already lost 40–50% of its original vegetation cover (Sano et al. 2010; Souza et al. 2020).

Stein et al. (2014) have postulated that species richness is driven by “heterogeneity diversity”, and Díaz and Cabido (2001) have shown that life diversity generally translates into plant “functional diversity”, which in turn regulates ecosystem processes. Therefore, the widespread biological impoverishment occurring in Brazilian biomes may have severe consequences on several ecosystem services, including food production (Joly et al. 2019). Townsend et al. (2008) argue that in the tropics, life diversity coupled with environmental conditions generates an extraordinary “biogeochemical diversity”, and to capture such diversity, it is important to understand the functioning of tropical ecosystems across environmental gradients. Considering the high pace of land-use and cover changes in Brazil, a minimum understanding of such biogeochemical diversity is crucial before some Brazilian biomes reach the so-called “tipping point” as defined by Lovejoy and Nobre (2019).

By the same token, the Amazon region is also identified as a “tipping element” (Lenton et al. 2008), defined as a sub-continental area having a disproportional importance on global climate health (Cox et al. 2000; Huntingford et al. 2013; Brienen et al. 2015). Paradoxically, although tropical systems have long been recognized as key areas in modern C cycles, these regions of the globe are still very underrepresented in global climate models (Schimel et al. 2015b). Partially because of the complex task of reaching an adequate survey on tropical areas due to several logistic constraints, and due to their enormous plant biodiversity (Kreft and Jetz 2007; Schimel et al. 2015b).

The natural isotope abundance of N (δ15N) and C (δ13C) offer a powerful way to evaluate the “biogeochemical diversity” of terrestrial ecosystems as well. Over the leaf lifespan, the leaf δ15N integrates the isotopic signature of nitrogen (N) sources and its dynamics and availability in soil (e.g., Houlton et al. 2006; Nardoto et al. 2008; Craine et al. 2009), while leaf δ13C integrates plant C assimilation and transpiration (Farquhar et al. 1989a; Ehleringer et al. 1993). Therefore, these isotopic traits provide a way to integrate the large temporal and spatial variability inherent in point measurements and to scale N and C cycling and plant-water relations.

This study is an exploratory investigation on the “biogeochemical diversity”, considering C and N isotope ratios and associated elemental composition of vegetation types in the main Brazilian biomes using a data set composed of more than six thousand tree leaf samples from more than fifty sites across the country. Although this study is exploratory by design, we anticipated based on well-established trends that: (i) foliar N concentration will be higher in biomes with predominance of deciduous and semi-deciduous trees (Aerts 1996; Aerts and Chapin 1999; Han et al. 2005; Sardans et al. 2016), as a corollary, (ii) we expected that foliar C:N ratios will be also lower in these biomes (Zhao et al. 2016; Zhang et al. 2020). We also expected that (iii) foliar δ13C (Swap et al. 2004; Diefendorf et al. 2010; Kohn 2010; Basu et al. 2019) and δ15N (Schuur and Matson 2001; Nardoto et al. 2008) should be inversely correlated with precipitation.

Material and methods

A total of 6,480 tree leaf samples were collected from 57 sites across the six Brazilian biomes (Fig. 1), encompassing the following main vegetation types according to IBGE (2012) classification: Forested Campinarana (Campina of the Amazon region, forest on sandy soils), Dense Ombrophilous Forest (evergreen non-flooded Amazon forest, evergreen coastal Atlantic Forest), Open Ombrophilous Forest (transition forest in Cerrado-Amazon regions), Mixed Ombrophilous Forests (Araucária forest, evergreen forest of the South region of Brazil with presence of A. angustifolia), Semideciduous Seasonal Forest (inland Atlantic Forest), Savanna (forest savanna of the Cerrado), Forested Savanna-Park of the Pantanal (periodically flooded), Savanna-Steppe (Caatinga of the Northeast region, xeric shrubland and thorny forest), Coastal Marine Vegetation (Restinga, forest formed on sandy, acidic, and nutrient-poor soils near the ocean), Fluvio-Marine Vegetation (Mangrove). Ombrophilous forests are located in areas where the dry season is shorter than four months, while seasonal forests occur in areas with a longer dry season.

Fig. 1
figure 1

Map of Brazil displaying the boundaries of the six biomes found in the country as well as the location of sampling plots with different colors and shapes representing vegetation types used in this study

Full size image

Table 1 summarizes geographic coordinates and climatic data, as well as the type of vegetation in each of the 57 sampling sites. Fig. 1 shows the spatial distribution of these sites within the Brazilian biomes. Although some of the sampling sites are clearly located in transition zones, especially between Amazonia and the Cerrado, they were assigned to only one biome according to the plant community affiliation (forest vs. savanna).

Table 1 Sampling sites including biome, vegetation types according to IBGE (2012) and classification used in this study, geographic coordinates (including elevation), and historical climate data
Full size table

This study analyzed original experimental data together with published ones. Therefore, there were slight differences in sampling protocols of different projects involved, and these are more prominent in the selection of species for sampling, especially in multispecies tropical forests. Several studies followed the abundance criteria, which gives preference to sampling the most common species, genera or family (e.g. Lins et al. 2016), while others selected species based on logistic facilities. The latter was applied in the Large-Scale Biosphere Atmosphere Project in the Amazon (LBA), more specifically, in the Tapajós National Forest, where a catwalk was installed in the middle of the canopy (e.g., Domingues et al. 2018). Most of the studies followed the protocol proposed by Cornelissen et al. (2003); therefore, preference was given to mature, fully expanded, healthy sunlit leaves from branches located in the middle of the canopy. However, it is perhaps naive to assume that almost 6,500 leaf samples were collected following this protocol, especially in tropical forest where access to the top of the canopy is challenging. For instance, the main goal of some samplings was botanical identification and not for elemental analysis or isotope ratio. In these cases, sampling of sunlit leaves was not mandatory. As it is well known, sun exposure strongly influences the C isotopic composition of the leaves (Buchmann et al. 1997; Ometto et al. 2006), which in turn is also affected by the sampling height, not recorded for each leaf sample. In tropical forests, there is a significant difference between δ13C of the leaves sampled from the lower canopy compared with the leaves from the upper canopy, the so-called canopy effect (Medina and Minchin 1980). Such gradient could be as high as 2.5% each 10 m of canopy height (Ometto et al. 2006). To minimize this bias in our analysis, leaf samples from the understory were removed from this study.

All samples included in the present work were analyzed at the Laboratory of Isotope Ecology of the Center for Nuclear Energy in Agriculture (CENA), University of São Paulo (USP) over the last 20 years. From 2 to 3 mg of sub-samples of dry leaves (which were previously oven-dried at 60°C for 72 h) were transfer to tin capsules, which were dry combusted in an elemental analyzer (model CHN1110; Carlo Erba, Milan, Italy) to determine N and C concentrations. The N2 and CO2 generated from the combustion was purified in a gas chromatographic column and passed directly to the inlet of a gas isotope ratio mass spectrometer (model Delta Plus, ThermoFisher Scientific, Bremen, Germany). The isotope ratios of C and N are expressed by the standard equation:

$$ \updelta X (\permil) = \left[\left(\frac{Rsample}{Rstandard}\right)-1\ \right]\times 100 $$

where, δX is the C or N stable isotopic composition and Rsample and Rstandard are the 13C:12C or the 15N:14N ratios of the samples and standards, respectively. The primary standard was Pee Dee Belemnite for C, and atmospheric air for N. The internal organic standard was BBOT (C26H26N2O2S). Experimental precision was measured as the mean ± standard error of the repeatability of BBOT analysis and was found to be equal to 0.2‰ and 0.3‰ for C and N, respectively.

We first presented foliar elemental concentration and isotope values considering all 6,480 leaf samples through statistical parameters like average, standard deviation and coefficient of variation, grouping vegetation types according to the IBGE (2012) classification (Table 2). Then we compared foliar elemental concentration and isotope values among vegetation types. In order to explore such differences, all Ombrophilous Forests were re-grouped as “Evergreen” forests, in contrast with the Seasonal Semideciduous Forest plot, labeled as “Seasonal”. The other vegetation types were kept under the IBGE (2012) classification, but we used local names such as Forested Savanna for Cerrado; Forested Savanna-Steppe for Caatinga; Campinarana for Campina; Forested Savanna-Park for Pantanal; Marine Vegetation for Restinga; Fluvio-Marine Vegetation for Mangrove. Following this re-grouping, we conducted statistical analysis and graphic display by averaging samples by plot, which we called “plot-average”. This was done because our database was not generated specifically for this purpose, and a large proportion of the data was generated from Ombrophilous Forests. Through averaging data by plot, we aimed to reduce a disproportional weight exerted from this type of vegetation within our data set.

Table 2 Statistical parameters of centrality and dispersion of N (mg g-1), C (mg g-1), C:N ratio, and δ13C (‰) and δ15N (‰) of tree leaves from vegetation types found in Brazil
Full size table

Box-plots were used to represent this comparison, sorting vegetation types in the horizontal axis (x-axis) based on increasing order of foliar N concentration. To test the differences statistically, we applied a generalized linear model (GLM, “lme4” package on R platform), according to data distribution that was tested using the “fitdistrplus” package of R. After the GLM, we applied the Tukey post-hoc test to assess differences among pairs of vegetation types. Carbon concentration and C and N isotope ratios followed a normal distribution, while N concentration followed a gamma distribution. Therefore, data regarding foliar N concentration was log-transformed.

We investigated the correlations between plot-average C and N stable isotopic ratios and associated elemental composition across vegetation types through biplots. Finally, we tested the effect of mean annual precipitation (MAP) and mean annual temperature (MAT) on plot-averages elemental and isotopic compositions also using GLM fitting. Climatological data were obtained from Climate-Data (2020), considering the municipality of each sampling site as reference. There is a weak (r2adj = 0.05), but significant direct (P = 0.05) correlation between MAT and MAP (data not showed). Therefore, we did not apply statistical tests using a multiple regression with MAT and MAP in the same model.

Results

The complete data set used in this study can be found at https://doi.org/10.17632/38npddpnts.1. The GLM statistical parameters used in this study are presented in Table S1 of the electronic supplementary material.

Grouping data according to vegetation types revealed significant differences among them (Table 2). The highest foliar N concentrations and δ15N were observed for the Seasonal Forest and Caatinga, significantly higher than the others (Fig. 2a, c). Lower foliar N concentrations and δ15N were observed in the Campina, and Cerrado (Fig. 2a, c). Foliar C:N ratio followed the reverse trend of foliar N concentration, where vegetation types with the highest foliar N concentrations had the lowest foliar C:N ratio and vice-versa (Fig. 2b). Consequently, we found significant correlations between foliar δ15N and foliar N plot-average concentrations (Fig. 3a, r2adj = 0.34, P < 0.01), and between foliar δ15N and foliar C:N plot-average ratios (Fig. 3b, r2adj = 0.40, P < 0.01).

Fig. 2
figure 2

Box-plots of foliar: (a) N, (b) C:N, (c) δ15N, (d) C and (e) δ13C in different vegetation types of Brazil. Horizontal middle lines indicate the median; boxes indicate the first and the third quartiles, and bars indicate the inter-quartile ranges

Full size image
Fig. 3
figure 3

Bi-plots of (a) plot-average foliar N concentration vs. δ15N, (b) plot-average foliar C:N ratio vs. δ15N, and (c) plot-average δ13C vs. δ15N

Full size image

Leaves in the Campina had the highest C concentration, followed by the Cerrado (Fig. 2d). However, foliar C concentration in the Cerrado was not different from Evergreen, Seasonal and Restinga vegetation types. On the other hand, the lowest foliar C concentration was observed in the Mangrove, followed by the Caatinga (Fig. 2d). Finally, the Campina had the most negative foliar δ13C, significantly different from other vegetation types. On the other hand, the least negative foliar δ13C was observed in the Caatinga, followed by the Cerrado (Fig. 2e). We found a weak, but significant positive correlation between foliar δ13C and foliar δ15N plot-average values (Fig. 3c, r2adj = 0.20, P < 0.01).

We investigated if climatic factors such as MAP and MAT could influence plot-average foliar C and N concentrations or their isotopic ratios (Moles et al. 2014). There were no significant correlations between foliar traits and MAT (data not shown). On the other hand, all the investigated parameters showed significant correlations with MAP (Fig. 4) except foliar C plot-average concentration (P = 0.19). Foliar N plot-average concentration had a weak negative correlation with MAP (radj2 = 0.14, P < 0.01) (Fig. 4a). Conversely, foliar C:N plot-average ratio had a weak psotive correlation with MAP (radj2 = 0.13, P < 0.01) (Fig. 4b). Foliar δ13C and δ15N plot-average values were both negatively correlated with MAP (P < 0.01), with higher radj2 compared to foliar N concentration and C:N ratio (0.53 and 0.32, respectively; Fig. 4d and c).

Fig. 4
figure 4

Correlations between MAP vs. plot averages of foliar: N concentration (a), C:N ratio (b), δ15N (c), δ13C (d). Vegetation types indicated by different colors

Full size image

Discussion

We extended the findings of Townsend et al. (2007) beyond foliar N:P ratios, by showing that life diversity is followed by a large variability of C and N elemental and isotopic composition (biogeochemistry diversity). Foliar N concentrations varied from 3 to > 70 mg g-1 [coefficient of variation (CV) = 40%], foliar C:N ratio from 6 to > 80 (CV = 40%), δ15N from −15 to 19‰ (CV = 120%), and δ13C from −40 to −23‰ (CV = 8%) (Table 2). Such variability is only comparable with the diversity found at the global scale (Craine et al. 2009; Diefendorf et al. 2010; Kattge et al. 2011; Zhang et al. 2020). Townsend et al. (2008), argue that such biogeochemistry diversity is a consequence of environmental heterogeneity (Stein et al. 2014), which in turn is the interplay between abiotic factors such as soil age, soil chemistry, landscape dynamics, and the richness of organisms, especially plant species.

The stereotype of tropical soils is that they generally seem to be in geomorphologically stable areas, where intense leaching by high rainfall has produced old deep soils (Porder et al. 2005). However, according to Sanchez and Buol (1974), soil temperature, which is higher in the tropics than in other areas is the major difference between soils of tropical and extra-tropical areas. Other than that, soils in tropical areas are as variable as in any other region of the globe, encompassing soils of different ages with a wide range of weathering states (Richter and Babbar 1991; Sanchez and Logan 1992).

The same variability observed in weathering states of tropical soils is also observed in their nutrient content (Sanchez and Logan 1992). For instance, due to low rainfall, soil pH and base saturation are generally higher in the Caatinga than in any other biome (Arruda et al. 2017). On the other hand, soils of white-sand vegetation like the Amazon Campina and coastal Restingas are poorer than in any other type of vegetation (Arruda et al. 2017; Bonilha et al. 2012; Mendonça et al. 2017). This variability in the soil nutrient pool was reflected in the foliar chemistry composition along gradients of soil fertility among areas of evergreen forests in the Amazon (Fyllas et al. 2009); and in areas of ecotones, such as Campina-Evergreen forests in the Central Amazon region (Mardegan et al. 2009; Mendonça et al. 2017), and Cerrado-Seasonal forests in the southeastern region of the country (Miatto et al. 2016).

On the regional to continental continuum, precipitation and temperature are probably the most important filter to select composition of plant communities (Currie 1991). However, on the local to regional continuum, landscape dynamics plays a key role in species sorting (Porder et al. 2005; Balzotti et al. 2016). For instance, in a toposequence located in the Central Amazon region, topography determine the soil N-availability, which is higher in the top flat areas, decreasing in the middle-slope and reaching a minimum in the lower parts of the terrain, where periodic flooding by small streams increase nutrient limitations (Luizão et al. 2004; Nardoto et al. 2008). Tiessen et al. (1994), have presented a similar example in a toposequence in the upper Negro River in the northern Amazon. In stark contrast, in Kaua’i, Hawai’i; Porder et al. (2005) found that most stable upland areas were nutrient-poor with nutrient availability increasing toward lowland areas.

In areas of high-gradient landscapes, erosion is an important process in reshuffling nutrient distribution by differential landscape stability, which in turn changes soil residence times across steeply dissected terrain (Weintraub et al. 2015). The effects of erosion have been detected not only for rock-derived nutrients, such as phosphorus (Porder and Hilley 2011), but also for N in subtropical (Hilton et al. 2013) and tropical areas (Weintraub et al. 2015). In Brazil, the Atlantic Forest, especially on coastal areas of southeastern Brazil, have steeply dissected terrain along the Serra do Mar rift system. In these areas, shallow soils predominate, with relatively short residence times (Furian et al. 1999; Martins et al. 2015), in sharp contrast with very deep soils that dominate the Amazon region.

All the above-mentioned abiotic factors (soil age and chemistry, and landscape dynamics) have a strong effect on the composition of vegetal species (Stein et al. 2014). In turn, species composition has a high degree of control over foliar N (Fyllas et al. 2009; Asner and Martin 2016; Balzotti et al. 2016). A clear example of this sort of influence is the foliar N higher concentrations found in leaves of Fabaceae specimen, regardless their N-fixing capabilities (Vitousek et al. 2002). Specimens of this family are widespread in the Amazon (Moreira et al. 1992); Atlantic Forest (Joly et al. 2012); Cerrado (Sprent et al. 1996); and Caatinga (Souza et al. 2012).

In our study, the biogeochemistry diversity was amplified by the inclusion of tropical forests in drier regions or regions with a prominent rainfall seasonality. In the Caatinga, MAP varies from 300 to 1,000 mm and drought events are recurrent (Oliveira et al. 2019); in sharp contrast with areas of the Amazon and Evergreen Atlantic Forest where MAP is higher than 2,000 mm. Besides the inclusion of biomes with lower precipitation, this study also includes the Seasonal Atlantic Forest, where MAP averages 1,500 mm, but rainfall seasonality is strong. Drier conditions or seasonality lead to establishing semi-deciduous species in the Seasonal Atlantic Forest (Morellato et al. 2000), and deciduous species in the Caatinga (Machado et al. 1997). As stated before, the highest foliar N concentration (Fig. 2a), and the lowest foliar C:N ratios (Fig. 2b) were observed in these vegetation types. This difference is likely explained by the higher N resorption efficiency in deciduous than in evergreen tree species (Aerts 1996).

Although Moles et al. (2014) showed that on a global level, MAP is a poor predictor of foliar traits, perhaps because MAP is not the best proxy of water availability. We found strong influence of MAP on plot averages δ15N and δ13C values, and to a lesser extent on plot averages foliar N concentration and C:N ratio (Fig. 4).

We found an inverse correlation between foliar δ15N and MAP (Fig. 4c), a commonly found trend (Heaton 1987; Austin and Vitousek 1998; Austin and Sala 1999; Handley et al. 1999; Amundson et al. 2003; Santiago et al. 2004; Houlton et al. 2006; Nardoto et al. 2008; Craine et al. 2009; Nardoto et al. 2014). Higher δ15N would suggest higher N-availability because excess-N not taken up by plants would be prone to volatilization and denitrification, which are highly fractionating processes, leaving an 15N-enriched substrate behind (Högberg 1997; Houlton et al. 2006). Therefore, the observed trend shown in Fig. 4c suggests that N-availability decreases with increasing precipitation (Schuur and Matson 2001; Aranibar et al. 2004; Swap et al. 2004; Pardo et al. 2006).

We also found a weak but significantly negative correlation between MAP and plot average foliar N concentration (Fig. 4a), and a positive weak correlation with foliar C:N ratio (Fig. 4b). These trends would reinforce the hypothesis of lower and higher N-availability in relatively wetter and drier sites, respectively (Austin and Vitousek 1998; Aranibar et al. 2004; Santiago et al. 2004). Additionally, the above trends fit well with the leaf economics spectrum (LES) theory, which states that in wetter areas, longer leaf lifespan, higher leaf mass per area and lower foliar N concentrations are expected (Wright et al. 2004). In turn, in drier sites, plants tend to have higher foliar N concentration as a strategy for maximizing photosynthesis during the shorter growing season (Reich et al. 1997, 1999), such as the Caatinga (Barros et al. 2020).

For tropical and subtropical rainforests, there is enough evidence showing that N availability is lower in wetter areas mainly because soil redox condition decreases decomposition rates of soil detritus, leading to N immobilization by the microorganisms in wetter sites (Schuur and Matson 2001). Nitrification rates also decreased in lowland wetter areas along a soil toposequence in Central Amazon as already mentioned (Luizão et al. 2004). Consequently, with low production of NO3-, chances of 15N-enrichment of the substrate are reduced (Houlton et al. 2006), leading to a soil and plant lower δ15N values in these sites. Houlton et al. (2007), observed that even if NO3- remains in the soil of wetter areas, plants prefer NH4+ which is 15N-depleted compared with this anion. Additionally, N-depleted ecosystems tend to rely strongly on mycorrhizal associations to increase their N content (Michelsen et al. 1996). Ectomycorrhizae is well known for delivering 15N-depleted N to plants (Hobbie et al. 1999), which results in lower δ15N in these N-depleted ecosystems (Hobbie and Högberg 2012). Typical examples of N-poor, and low foliar δ15N vegetation types in Brazil are the Cerrado and Campina (Fig. 2a), where mycorrhizal associations seem to contribute to lower foliar δ15N (Bustamante et al. 2004; Mardegan et al. 2009).

As rainfall decreases, but without causing higher water deficit, mineralization and nitrification rates increase leading to an increase in N-availability to plants and for N-gas losses (Högberg and Johannisson 1993; Austin and Vitousek 1998; Sousa-Neto et al. 2011). The highest foliar δ15N values were observed in the Caatinga, suggesting that N-availability is high in this biome (Freitas et al. 2010a). Certainly, the predominantly eutrophic soil of the Caatinga is one factor that contributes to this (Arruda et al. 2017) since there is a strong correlation between soil fertility and N-availability (Vitousek and Sanford Jr 1986). Although Fabaceae is an important family in Caatinga communities, it seems that most of them are not N-fixing or fix at low rates (Freitas et al. 2010b), which is also an indication of high N availability (Vitousek et al. 2002). On the other hand, Caatinga soil N2O emissions, which are a sensitive indicator of N availability (Davidson et al. 2007), have been reported to be extremely low, similar to the Cerrado (Ribeiro et al. 2016). Alternatively, Freitas et al. (2015), proposed that the high foliar δ15N values in the Caatinga are mostly a consequence of the lag between brief pulses of nitrification-denitrification and low N uptake by plants on the onset of the rainy season.

Precipitation also strongly influenced foliar δ13C (Fig. 4d), in line with several other studies that reported similar trends (Swap et al. 2004; Diefendorf et al. 2010; Kohn 2010; Basu et al. 2019). The most likely explanation for this trend is the decrease in leaf stomatal conductance with reduced water availability, which in turn leads to a decrease of the intercellular pCO2 (pi), if the demand for CO2 is kept higher than the supply of CO2 (Farquhar and Sharkey 1982; Farquhar et al. 1989a, b).

Despite the importance of water availability, we observed that plants in the Cerrado had higher foliar δ13C compared to forests at similar MAP (Fig. 4d). Therefore, in such cases, water availability is not the only factor invoked to explain these variations. The most likely explanation is the so-called canopy effect in tropical forests (Medina and Minchin 1980; Buchmann et al. 1997), which produces 13C-depleted plants and a gradient of foliar δ13C values along the canopy. The reasons behind this are the recycling of 13C-depleted CO2 in the forest (i.e., assimilation of biogenic CO2 derived from forest soil respiration; Van der Merwe and Medina 1991) and the increased fractionation due to photosynthesis taking place under low light conditions. Ometto et al. (2006), assessing four rainforests of the Amazon region, found an average increase of 0.2‰ per meter of canopy height; therefore, in a 30 m-height canopy, foliar δ13C would be increased by 6‰ at the top of the canopy compared to lower height foliage.

Conclusion

The results presented here reinforce earlier findings by Townsend et al. (2008) that high life diversity in the tropics, generated by abundance of water and energy, and enhance by environmental diversity, is followed by high biogeochemistry diversity. Such diversity, however, appears to respond in a synchronized way to changes in precipitation.

Finally, we expect that the database presented here can contribute to a better understanding of the tropics in global models, a fundamental step in helping mitigate effects of land-use and climatic changes that the tropics are facing with global consequences. We also expect that this database can help a broad community of plant ecologists understand regional and global patterns of plant trait variation.