1 Introduction

On the Southern border of the Amazon, there is an extensive ecotone (also known as ‘Zone of Ecological Tension’) approximately 4500 km long between the two largest biomes in South America, the “Cerrado” savanna and the Amazon forest (Marimon et al. 2006). At long centennial and millennial time-scales, the concerted advancement and retreat of the two largest South American biomes occurs due to natural climate changes, with hot and humid periods alternating with cold and dry periods and reflecting to a large degree the movements of the intertropical convergence zone (ITCZ) (Mayle 2000). Because the ITCZ responds to continuously changing climate forcing, it may be expected that forest and savanna in Southern Amazonia also exist in a state of permanent tension and movement (e.g., Ratter 1992). The current Southern Amazon climate is potentially more humid than at any time in the Holocene (Mayle 2000), and potentially as a consequence of this, several studies have suggested that the Amazon forests are naturally advancing into savannas in the “Cerrado” (Ratter et al. 1973; Ackerly et al. 1989; Ratter 1992; Marimon et al. 2006; Morandi et al. 2015). This process potentially parallels recent changes in other savanna–forest transitions in Africa (Cole 1992; Khavhagali and Bond 2008; Veenendaal et al. 2015) and Australia (Cole 1992; Kershaw 1992) and might also help to explain the hyperdynamic condition of the vegetation in the ecotone, with high rates of tree recruitment and mortality (Marimon et al. 2014; Morandi et al. 2015).

However, climatic changes are not the only important influence on the forest–savanna ecotone and other processes may also influence the structure of the vegetation here. This includes changes in the frequency of fire, especially where management may be suppressing “cerrado” fires (Durigan and Ratter 2006; Moreira 2000; Geiger et al. 2011). Additionally, the recent unprecedented rise in atmospheric carbon dioxide is expected to favor C3 plants over C4 grasses, as more CO2 may preferentially increase the water use efficiency of C3 plants (Phillips et al. 2009; Kerbauy 2012). These multiple factors may impact the species composition (Khavhagali and Bond 2008), structural dynamics, density of individuals, basal area, aboveground biomass (Phillips et al. 2009; Marimon et al. 2014; Morandi et al. 2015), occurrence of fire-tolerant plants (Miranda et al. 2002; Henriques 2005) and vegetation encroachment (Khavhagali and Bond 2008), and contribute to the high biodiversity of the region (Marimon et al. 2014; Oliveira et al. 2016).

In some communities, such as savanna areas of typical and open “cerrado” and forest areas, such as the dense woodland, succession usually occurs according to the climax-gradient model (Whittaker 1953; Eiten 1972; Ratter 1992; Henriques 2005). These authors suggest that although vegetation formations can be identified as relatively stable communities adapted to maximize resource use in terms of biological productivity, there is no single, absolute vegetation climax state across large areas. Rather, ‘climax composition’ has meaning only relative to a site’s precise position along multiple environmental gradients, including local topographic, edaphic, and hydrologic factors. Hence, according this model, plant vegetation is likely to change dynamically over time in response to continuous variations in availability of different resources and frequency of disturbances.

There are multiple processes occurring at the ecotone between the Amazon and “Cerrado” biomes with potential to drive changes in vegetation composition and structure. The region has undergone rapid deforestation, which may lead to localized drying and vulnerability of remaining forest fragments resulting from agriculture, grazing and urban activities (Fearnside 2005; Nogueira et al. 2008). Habitat fragmentation can lead to declines in biodiversity, invasion of exotic species, soil erosion, pollution of aquifers, ecosystem degradation, increase in fires, imbalance of the carbon cycle and other elements related to regional climate changes (Fearnside 2005; Klink and Machado 2005; Bonini et al. 2014, 2018), most of which may be expected to degrade remaining forests. Furthermore, the same region has experienced a multi-decadal warming trend and a marked increase in dry season intensity and length (e.g., Marengo et al. 2013). This process may be a consequence of anthropogenic climate change (e.g., Li et al. 2009) and is contributing to increases in Amazon tree mortality (Phillips et al. 2009; Brienen et al. 2015) and reductions in growth (Feldpausch et al. 2016). Thus, several powerful and in some cases opposing factors are influencing vegetation dynamics in this transition zone. Determining which of them are the most influential currently requires careful analysis, including on-the-ground monitoring of vegetation dynamics.

In sum, the factors that determine current changes in the “Cerrado”–Amazon ecotone are many and varied, and poorly known in terms of their relative importance for remaining natural vegetation. Our objective is to assess changes in the structure and composition of tree species in permanent sample plots in savannas (typical and open “cerrado”, known in Brazil as ‘Cerrado Típico’ and ‘Cerrado Ralo’) and forests (dense woodland, known locally as ‘Cerradão’) in the “Cerrado”–Amazon ecotone. We aimed to answer the following questions: (1) Are there detectable changes in structural parameters (density, basal area and aboveground biomass) over time? (2) Is the composition of these systems changing, in terms of an increase in forest tree species in savanna sites? Our working hypotheses are: (1) there is vegetation encroachment, as measured by an increase in structural parameters (density, basal area and aboveground biomass) and (2) there is an increase in the number of forest species through the suppression of savanna species.

2 Materials and methods

Study area and data collection

– The study was carried out in the eastern region of the state of Mato Grosso, in the “Cerrado”–Amazon ecotone. We delimited two dense woodland areas, a forest vegetation known as ‘cerradão’ in Brazil (DW-1 e DW-2), two typical “cerrado” areas (TC-1 and TC-2) and one open “cerrado” area (OC-1). The dense woodland (‘cerradão’) has savanna and forest species with a fairly continuous tree cover that varies between 50 and 90% and mean species height from 8 to 15 m. Typical “cerrado” (‘cerrado típico’) has scattered trees and shrubs with woody cover ranges between 20 and 50% and mean species height from 3 to 6 m, while open “cerrado” (‘cerrado ralo’) is a less dense community with widely scattered woody elements than typical “cerrado” with a tree cover between 5 and 20% and mean species height from 2 to 3 m (Ribeiro and Walter 2008).

The study sites are located at Fazenda Santa Marta, in the municipality of Ribeirão Cascalheira, and in the Bacaba Municipal Park, in Nova Xavantina (Fig. 1). In each area, we set up 1-ha permanent plots and sampled the vegetation in 2008, 2011, 2013 and 2015 as part of the UK-led TROBIT (Tropical Biomes in Transition) and RAINFOR (Red Amazônica de Inventários Florestais) projects and the Brazil-led PELD (Projeto Ecológico de Longa Duração) project, coordinated by UNEMAT-Nova Xavantina. We numbered, identified and measured the height and the diameter at breast height (DBH at 1.3 m) of all individuals with D ≥ 10 cm using standard protocols (e.g., Phillips et al. 2010). Based on the specialized literature (Mendonça et al. 2008) and field experience of some members of the present study, we also classified each species according to the habitat (or physiognomy) of its preferred occurrence: forest, savanna or generalist (when it occurs in both).

Fig. 1
figure 1

Geographic location of the study areas in dense woodland, DW-1 (filled circles) and DW-2 (filled squares), and savannas of typical “cerrado”, TC-1 (open circles) and TC-2 (open squares), and open “cerrado”, OC-1 (open diamonds), and in the “Cerrado”–Amazon ecotone, Mato Grosso, Brazil

Full size image

Data analysis

– We calculated tree density, basal area (g = (π/4).d2), where d = DBH (diameter at breast height of each individual) and total aboveground biomass of each study area, this latter parameter following Scolforo et al. (2008). We calculated the average annual rates of tree recruitment and mortality (Sheil et al. 1995, 2000), the gain and loss in basal area (Guimarães et al. 2008) as well as the turnover rate of individuals (Phillips and Gentry 1994). To compare variations between areas and over time in tree density, basal area and aboveground biomass and habitat-preferences of species, we used repeated measurements ANOVA, followed by a Tukey post hoc tests (Zar 2010). We tested assumptions of homogeneity of variances and normality of residuals with Levene and Shapiro–Wilk tests, respectively. To quantify temporal changes in species composition, we apply NMDS ordination based on Bray–Curtis similarity measure. We also conducted linear regression analyses to examine the temporal relationship in density, basal area and aboveground biomass. We carried out the analyses and graphs with the ‘vegan’ (Oksanen et al. 2016), ‘MASS’ (Venables and Ripley 2002), ‘gridExtra’ (Baptiste Auguie 2016) and ‘ggplot2’ package (Wickham 2009) in the program R (R Core Team 2016). We considered 5% significance level for all analyses.

3 Results

Between the first (2008) and the following inventories (2011, 2013 and 2015), both in dense woodland areas and in typical and open “cerrado” areas, there was a constant or a significant increase in density (F12,360 = 12.57, P < 0.01), basal area (F12,360 = 9.80, P < 0.01) and aboveground biomass (F12,360 = 15.02, P < 0.01), except for one area of dense woodland (DW-1) which tended to decrease in density (Fig. 2, Table S1). When analyzed instead by linear regression, all typical and open “cerrado” areas clearly increase in all parameters (stem density, basal area and aboveground biomass) and dense woodland areas in terms of basal area and aboveground biomass, with R2 values close to or exceeding 0.9 (Table S5). Throughout, the two areas of typical “cerrado” (TC-1 and TC-2) had density, basal area and aboveground biomass values similar to those found in dense woodland DW-2 (Fig. 2, Table S1). Among all the areas studied, we highlight the open “cerrado” (OC-1) as showing a progressive and significant increase in all parameters throughout all inventories, with the greatest relative increases (Fig. 2, Table S1). The DW-1 site had particularly large basal area and aboveground biomass gains after 2011. Across all plots and in all intervals sampled, the recruitment rates and gain in basal area were higher than mortality rates and loss in basal area, except for dense woodland between 2011 and 2015. Finally, both the typical “cerrado” and the open “cerrado” had higher stem turnover rates than those of dense woodland areas (Table S1).

Fig. 2
figure 2

Density (a), basal area (b), and aboveground biomass (c) (average ± standard error) of tree species in dense woodland (DW; green continuous line), typical “cerrado” (TC; red dashed line), and open “cerrado” areas (OC; orange dotted line), in the “Cerrado”–Amazon ecotone. (Color figure online)

Full size image

When we analyzed the data according to the species’ preferred habitat of occurrence, we observed a tendency for savanna species density to be constant or decrease over time, except for TC-1 and OC-1 (F12,360 = 2.40, P < 0.01; Table S2 and Fig. 3). By contrast, there was an increase in basal area (F12,360 = 10.20, P < 0.01) and aboveground biomass (F12,360 = 7.48, P < 0.01) of forest and generalist species (F12,360 = 3.28, P < 0.01; F12,360 = 6.57, P < 0.01, respectively), in particular in the open “cerrado” site (OC-1). As well as the relationship between forest species structural parameters with calendar year was very strong, with an R2 between 0.8 and 0.9, mainly in aboveground biomass (Table S5). In OC-1 site, generalist species increased in terms of their contribution to vegetation structure since 2011, while forest species did so since 2013, although savanna species still dominate the vegetation (Fig. 3, Table S2 and S5). Nevertheless, all areas showed considerable floristic similarity among them and over time (Fig. S1).

Fig. 3
figure 3

Density, basal area, and aboveground biomass (average ± standard error) of tree species in different scale in Y axis. All parameters are classified according to their preferred habitat of occurrence: savanna (a–c), generalist (d–f), and forest (g–i), in dense woodland (DW; green continuous line), typical “cerrado” (TC; red dashed line), and open “cerrado” areas (OC; orange dotted line), in the “Cerrado”–Amazon ecotone. (Color figure online)

Full size image

In general, the recruitment rate and gain in basal area were also higher than the mortality rate and loss in basal area (Table S3). However, for savanna species the mortality rate was higher than recruitment in woodland areas. On the other hand, the rates of recruitment and gain in basal area for forest species were higher than their mortality rates and losses in basal area in all areas (Table S3). Overall, when we analyzed the communities over time in terms of the stem density and basal area of each ecological group, we observed clear increases in forest and generalist species, not only in the “Cerrado” vegetation but also in dense woodland areas (Table S3).

4 Discussion

Our study found an increase in density, basal area and aboveground biomass, both in typical and open “cerrado” as well as in dense woodland. Although the finding that vegetation is becoming denser over time corroborates our first hypothesis, this changes isn’t abrupt and not all the sites are becoming forest-like. However, there is a tendency that all savanna sites, as the parameters assessed in typical “cerrado” (TC-1 and TC-2) have similar values to those of the dense woodland (DW-2), to become a forest-like if the current trends are maintained. Furthermore, other indicators of encroachment such as vegetation dynamics generally showed the higher recruitment than mortality rates and the higher gain than loss in basal area, indicative of a future marked natural change in structure. While other studies on the “Cerrado”–Amazon transition have already detected localized changes at single sites or with semiquantitative assessments (Ratter et al. 1973; Ratter 1992; Marimon et al. 2006; Morandi et al. 2015), this is the first time that quantitative and standardized data corroborated the vegetation encroachment hypothesis over longer temporal scales and considering multiple types of savanna vegetation.

The excess of tree recruitment over mortality can result in an increase in tree stem density, while the excess of growth rates over death rates can result in increasing tree basal area and biomass. Therefore, in these sites at least, vegetation has a tendency to become markedly denser and, consequently, there is a tendency to favor the establishment and growth of forest species in what are savanna habitats. As an example, while we observed that stem density hasn’t changed in dense woodland (DW-1 and DW-2) over the years and nor did it change in typical “cerrado” (TC-2) between 2013 and 2015, in all cases the basal area and aboveground biomass did increase, indicating that the vegetation is tending to become denser. Among the factors that may contribute to the process of vegetation encroachment, we highlight the increase in the availability of resources, such as carbon dioxide (e.g., Lewis et al. 2004; Phillips et al. 2009), changes in the frequency of fires (e.g., Geiger et al. 2011; Durigan and Ratter 2006) in most of the areas, and the increase in rainfall (Gloor et al. 2013; Castanho et al. 2016). These factors can favor the advancement or withdrawal of the forest over the savanna (Ratter et al. 1973; Ratter 1992; Marimon et al. 2014) or speed up the natural dynamics of the vegetation. These changes in dynamics may also favor the establishment and acceleration of individual tree life cycles, especially those of intrinsically fast developing taxa, such as Tachigali vulgaris L.G. Silva & H.C. Lima (Morandi et al. 2015).

While the overall composition of species in savanna systems remains predominantly savanna species, there is still a markedly tendency for savanna species to decline and forest species parameters to increase over time. We observed an increase in basal area and aboveground biomass mainly in generalist and forest species in all areas and a decrease in savanna species in the dense woodland (DW-1 and DW-2) and TC-2. A driver here may be the absence of fire in these areas for at least 15 years, favoring recruitment over mortality rate and the establishment of generalist and forest species. Other studies have reported the replacement of savanna with forest species after fire suppression in the “Cerrado” biome (Moreira 2000; Henriques and Hay 2002; Hoffmann and Moreira 2002; Mews et al. 2011; Durigan and Ratter 2006, 2016). However, apparently our results show that vegetation may recover biomass and stem numbers very quickly after fire. For example, TC-1 showed a decline due to the fire that affected the area soon after the first inventory (2008), but soon afterward we found an increase in density, basal area and aboveground biomass of generalist species (inventory of 2013) as well a further increase in basal area and biomass of forest species in the inventory of 2015. Hence, within a short period after the fire there was a increase in the density of generalist and savanna species, but not of forest species. This can indicate that the vegetation seems to be resilient to the impacts of fire, but that non-forest species dominate post-fire recovery in these areas.

Earlier work by Marimon et al. (2014) revealed dense woodland to be ‘hyperdynamic’ in the sense of having exceptionally fast recruitment and mortality of stems, results which are corroborated here and extended also to open and typical “cerrado”. The marked dynamism of these vegetation types, geographically close to the Amazon (Marimon et al. 2010, 2014) corroborates the hyperdynamism of the whole “Cerrado”–Amazon ecotone (Marimon et al. 2014). The increase in forest taxa is one driver of this hyperdynamism, and at least superficially consistent with a positive relationship between rainfall and biomass accumulation (Costa et al. 2010; Brando et al. 2014; Baker et al. 2014; Veenendaal et al. 2015). Yet, while Amazon rainfall has increased recently in some localities (e.g., Gloor et al. 2013), in our region there have been a sequence of strong droughts. Apparently in the ecotone, biomass has proven resistant to recent climate changes, and this increase is likely driven by other factors such as the increase in atmospheric carbon dioxide (e.g., Phillips et al. 1998), change in fire regimes (Abreu et al. 2017), or both.

Other factors may also be favoring the species studied in the possible encroachment of the vegetation by the increase in canopy cover (Yarranton and Morrison 1974; Silva et al. 2013; Veenendaal et al. 2015) and the accumulation of leaf litter, consequently of soil nutrients (Yarranton and Morrison 1974; Connell and Slatyer 1977; Marimon Junior and Haridasan 2005; Silva et al. 2013; Passos et al. 2014; Oliveira et al. 2016). These factors are related to vegetation establishment and productivity, as well as the accumulation of organic matter and the higher microbial activity in the soil (Haridasan 2001). Such conditions can favor generalist and forest species, as they tolerate low light environments (Silva et al. 2013) and usually require the higher availability of nutrients in the cycling system (Peltzer et al. 2010), as observed in the present study. Our results indicate that transitional habitats, including savannas, deserve greater attention to their ecology and conservation because of the intense interactions between species, which can alter the structure and dynamics of the entire habitat (Miranda et al. 2014).

Indeed, the occurrence of generalist and forest species such as Emmotum nitens (Benth.) Miers, Hirtella glandulosa Spreng., Tachigali vulgaris (Ratter et al. 1973; Ratter 1992; Marimon et al. 2006; Morandi et al. 2015) and Xylopia sericea A.St.-Hil. (Table S4) can be a tendency of changing in vegetation can occur along the time. E. nitens, H. glandulosa and Vochysia haenkeana Mart. have been classified as connectors of riparian forest, dense woodland and transitional vegetation on the Southern edge of the Amazon (Oliveira-Filho and Ratter 1995). Moreover, X. sericea is a species typical of forest habitats and is typically absent from savanna (Ratter et al. 1973; Marimon et al. 2006). Finally, special attention should be given to T. vulgaris, known as a key species at early phases of the successional process on the Southern edge of the Amazon, and which can favor community dynamics and floristic compositional changes (Moreira 2000; Morandi et al. 2015). These dominant species—with high density and basal area in multiple savanna inventories in the transition zone—may be considered keystone species in the process of vegetation encroachment and hence potential ‘ecosystem engineers’ that can favor the increased occurrence of other forest species over time (Ratter et al. 1973; Ratter 1992; Marimon et al. 2006; Vidotto et al. 2007; Morandi et al. 2015) by facilitation processes (Yarranton and Morrison 1974; Connell and Slatyer 1977; Durigan and Ratter 2006) (Table S4).

It is important to point out that our observations have occurred over a relatively short sampling period of 7 years. The encroachment in the structural vegetation recorded in the savanna communities can be clearly favoring carbon accumulation (Ratajczak et al. 2012; Pellegrini et al. 2016). While savanna species remain abundant and species-rich, the changes indicate the potential for afforestation over time. Nevertheless, this process results in a clear dilemma for conservationists, with implications for climate changes and preservation of savanna species, as for example management of fire in “Cerrado” protected and non-protected areas can have large impacts on both carbon storage and biodiversity (Durigan and Ratter 2016). There is an apparent trade-off—let these lands absorb carbon dioxide but lose savanna biodiversity, or actively manage them via increased fire to protect savanna diversity. The long-term consequences and implications of this trade-off, however, remain unknown (Pellegrini et al. 2016) and should be better assessed, especially in this critical ecotone between the two largest biomes in South America.

Our study observed in the savannas sites and dense woodland has a tendency to be similar in structural changes over time. The results showed here suggest that many savannas in the “Cerrado”–Amazon ecotone can be currently undergoing a process of becoming a denser vegetation, even in the face of a regionally warming and drying climate. For example, despite of our results the grass-dominated open “cerrado” (OC-1) in 2008 is already better classified as a typical “cerrado” only 7 years later (2015). Hence, if the current trends are maintained, open and typical “cerrado” can be likely to change into forest-like vegetation, and dense woodland can become a taller and denser forest. Finally, our work suggests the need for further studies to test whether and precisely how a facilitation process is driving the overall encroachment, and specifically the establishment of forest species. Deeper understanding of facilitation here may be critical for improved understanding of the full ecological and conservation processes involved in the complex vegetation dynamics at the transition between the Brazilian savanna and Amazon forest.