Not all trees can make a forest: Tree species composition and competition control forest encroachment in a tropical savanna

Abstract

en

1. Forest encroachment into savannas is a widespread phenomenon, the rate of which may depend on soil conditions, species composition or changes in stand structure. As savanna specialist trees are replaced by generalist species, rates of stand development may increase. Because generalists can persist in forests, they are likely to grow more quickly and survive longer in dense stands, compared to savanna specialists. Furthermore, the faster growth rates of generalists may allow them to overtop and outcompete savanna specialists, causing rapid species turnover.
2. We measured growth and survival of 6,147 individuals of 112 species of savanna and generalist tree species over a period of 10 years in an ecological reserve in Assis, São Paulo State, Brazil. We modelled growth and mortality as a function of soil texture and nutrients, tree size, competitive neighbourhood, and membership in savanna or generalist (species which can persist in forests and savannas) functional groups.
3. Tree growth and survival was strongly influenced by competition, as estimated by the basal area of trees taller than a focal tree. At the stand level, savanna species are unable to contribute basal area growth in closed stands, while generalist species continue to increase in basal area even at high stand basal area. This phenomenon is driven by differences in growth and mortality. Generalists grew faster than savanna species, both in height and diameter. This difference in growth rates led to savanna species becoming suppressed more rapidly than generalists. When suppressed, savanna species were more than twice as likely to die than were generalists. Soils had inconsistent and mostly weak effects which were difficult to separate from gradients of stand structure.
4. Synthesis. We demonstrate that the presence of generalist trees accelerates the rates of basal area accumulation due to their greater growth rates and tolerance of shading. Generalists outcompete savanna trees by growing faster in the open and overtopping savanna specialists. Due to the slow growth and high mortality of savanna species in the shade, they are unable to form closed-canopy stands. Accounting for differences among functional types and development of vegetation structure is critical for modelling forest encroachment.

Resumo

pt

1. O adensamento das savanas vem ocorrendo em larga escala, em taxas que dependem das condições de solo, composição de espécies ou mudanças estruturais das comunidades. À medida que árvores especialistas em savana são substituídas por espécies generalistas, o adensamento é acelerado. As generalistas podem persistir em florestas, podendo crescer mais rápido e sobreviver por mais tempo que as especialistas em savana. As generalistas podem ainda sombrear e suprimir as especialistas de savana, causando rápida substituição de espécies.
2. Nós quantificamos o crescimento e a sobrevivência de 6147 indivíduos, de 112 espécies arbóreas típicas de Cerrado ou generalistas, durante 10 anos, na Estação Ecológica de Assis, SP, Brasil. Modelamos o crescimento e a mortalidade em função da textura e nutrientes do solo, tamanho do indivíduo, competição com os vizinhos e grupo funcional categorizado em especialista de savana ou generalista (ocorre também em florestas).
3. Crescimento e sobrevivência foram fortemente influenciados pela competição, representada pela área basal de árvores maiores que a árvore focal. Nas parcelas mais densas, especialistas em savana não contribuíram para incremento em área basal, enquanto a área basal das generalistas continuou aumentando. As generalistas cresceram mais rápido em diâmetro e altura, rapidamente suprimindo as especialistas em savana, cuja probabilidade de morrer foi duas vezes maior do que entre as generalistas. Os efeitos das propriedades do solo foram fracos e inconsistentes, difíceis de separar do gradiente estrutural da vegetação.
4. Síntese. Nós demonstramos que a entrada de árvores generalistas acelera a acumulação de área basal no Cerrado, devido às maiores taxas de crescimento e tolerância à sombra. As generalistas crescem mais rápido e ultrapassam as especialistas de savana, suprimindo-as. O crescimento lento e a alta mortalidade das espécies típicas de Cerrado nos ambientes sombreados impede que se aglomerem em povoamentos densos. As diferenças entre os dois tipos funcionais e grau de desenvolvimento estrutural da vegetação precisam ser contabilizados na modelagem do processo de adensamento.

1 INTRODUCTION

Encroachment of forests into savannas is a widespread phenomenon. Increases in woody biomass in savannas have been attributed to several sources, including atmospheric CO₂ enrichment (Buitenwerf et al., 2012; Wigley et al., 2010), interannual variation in precipitation (Criado et al., 2020; Lehmann et al., 2008) and especially the suppression of disturbances or herbivory, which reduce tree establishment or survival (Axelsson & Hanan, 2018; Stevens et al., 2016). At the scale of an individual stand, however, more proximate variables such as resource availability and competition drive the individual tree growth and tree mortality that contribute to species turnover and biomass accumulation.

While much attention has justifiably focused on factors with global or regional scope, rates of development of forests are strongly dependent on local abiotic conditions and factors endogenous to the stand itself (Criado et al., 2020). The rate of transformation from savanna to forest depends upon local variables such as soil conditions, species composition and the way that individual trees interact (Archer et al., 2017). The species that drive forest encroachment differ in important ways among regions—in canopy architecture, phenology or nutrient demands—likely explaining some of the global differences in rates of encroachment (Stevens et al., 2017). The degree to which these various biotic and abiotic factors affect rates of growth is important to modelling the dynamics of forests and savannas because they will respond differently to external forcings such as management or atmospheric conditions (Ford et al., 2017).

Abiotic conditions have, regionally, a dominant influence on forest–savanna distributions (Langan et al., 2017; Lloyd & Veenendaal, 2016; Pellegrini, 2016), in part by mediating the effects of disturbance (Hoffmann et al., 2012; Staver, 2018). Soils are especially important within the window of climatic conditions that can support either savannas or forests (Eiten, 1972). Many tropical soils are heavily weathered and leached of nutrient cations, and forests are more common on soils with greater water-holding capacity (Assis et al., 2011) and greater nutrient availability (Cruz Ruggiero et al., 2002; Veenendaal et al., 2015). Specifically, soil nutrients such as phosphorus, calcium and potassium may limit the formation of forests (Lloyd et al., 2015; Silva et al., 2013), or they may play an indirect role by reducing growth rates and thus slowing post-disturbance recovery of woody plants (Bond, 2008; Franco et al., 2014). In the absence of fire, sites with greater availability of water and base cations are likely to be sites of faster forest development, but the relative importance of soils and other factors is seldom characterized.

Abiotic conditions are not the only factor limiting forest development. Density-dependent declines in growth, establishment or survival may lead to decelerating rates of canopy development in denser stands (Stevens et al., 2017). Declining productivity with increasing stand age is a widespread and fundamental phenomenon in forest ecology (Binkley et al., 2002; Gower et al., 1996; Ryan et al., 2004). However, its implications for savanna–forest transitions are seldom explored, though self-thinning has been implicated as a mechanism to maintain open savannas in dry regions (Belay & Moe, 2012; Sea & Hanan, 2012). Light levels are reduced in the understorey as the canopy closes, creating asymmetric, one-sided competition for light (Weiner, 1990) which differentially affects shade-intolerant species. Water may become less available as a greater tree biomass transpires it or intercepts it (Calder, 1998; Honda & Durigan, 2016). On the other hand, the open savanna environment is unsuited for many forest species (Cardoso et al., 2016; Hoffmann et al., 2004): the high vapour pressure deficits and dry soils during seasonal droughts may restrict establishment of the fastest-growing species in savannas. The shifts in the environment that occur with canopy closure are potentially very important for savanna–forest dynamics; it is possible that South American savanna species cannot form closed-canopy stands even after several decades (Geiger et al., 2011), due to their open crowns, poor recruitment in the shade (Franco et al., 2014) or their high resource needs (Silva et al., 2013).

Forest encroachment is often characterized by turnover in species composition, driven by differential responses of species to canopy closure. Savanna species are replaced by forest specialists or generalists (species commonly found in both savannas and in forests) as tree cover increases (Abreu et al., 2017; Flake, Abreu, et al., 2021). Savanna and forest species are functionally distinct, with savanna species adapted to frequent fire and high insolation, and forest species and generalists adapted to rapidly reach forest canopies or to gather light in light-limited forests (Flake, Abreu, et al., 2021). Forest and generalist species grow faster than savanna species (Hoffmann & Franco, 2003; Rossatto et al., 2009), and so the replacement of savanna species is likely to increase the rates of forest growth. Once forests become established, savanna species may grow even more slowly or die off in the shade, further contributing to species turnover. Past work has demonstrated that generalist species do make up the bulk of the new biomass of encroaching forests (Flake, Abreu, et al., 2021; Passos et al., 2018), but it is unclear whether this is due to faster growth rates of generalists, lower mortality rates or differences in other demographic rates.

We explored several questions related to tree growth and mortality using 10 years of tree growth data from permanent plots on the boundary between the Cerrado and Atlantic Forest biomes in Brazil. The study site has experienced dramatic woody encroachment over the past several decades (Pinheiro & Durigan, 2009), paralleling the regional shift from open cerrado to closed-canopy cerradão (Durigan & Ratter, 2006). The study site currently has broad gradients of stand basal area and species composition which allow for comparisons of the roles of species functional type, competitive effects and soils. Specifically, we sought to answer these questions:

1. How do growth and mortality rates vary over a savanna–forest gradient? Are differences driven primarily by stand structure and species composition or underlying soil gradients?
2. What effect does competition have on rates of growth and mortality of trees?
3. Do savanna specialist species and generalist species differ in growth rates? Do they differ in their sensitivity to competition?

2 MATERIALS AND METHODS

2.1 Study site

The study was conducted at Assis Ecological Station (Estação Ecológica de Assis), in the city of Assis, São Paulo State, Brazil (22°33′20″–22°37′41″S, 50°24′4.8″–50°21′27″W). The preserve comprises a mosaic of typical Cerrado vegetation, including open campo, cerrado savanna/woodland and cerradão forest (Pinheiro & Durigan, 2009). There is mild topographic variation, between 510 and 596 m a.s.l. The underlying geology consists of Bauru Group of Adamantina Formation. The climate is subtropical, with a distinct summer rainy season and winter dry season of variable length (Köppen Cfa; Alvares et al., 2013). Mean annual temperature is 21.8°C and mean annual precipitation is 1,400 mm.

2.2 Sampling

We established 30 plots along a gradient of stand structure (Figure 1) from open savanna to closed-canopy forest to monitor the tree community, and inventoried the plots in 2006, 2011 and 2016 during the wet season (January–April). Each plot was 0.1 ha (20 × 50 m), following the recommendation of Felfili et al. (2005) for monitoring Cerrado vegetation. Within each plot, we measured and tagged all trees ≥5-cm diameter at breast height (DBH). For multi-stemmed trees, each stem was measured separately so long as one of the stems was ≥5 cm DBH, and we calculated the equivalent diameter from the sum of the cross-sectional areas of the stems. For each tree, we recorded species identity, DBH to the nearest 0.1 cm with a diameter tape and total tree height measured with a telescoping measuring pole. When identification in the field was not possible, samples were collected and compared to specimens at the D. Bento Pickel Herbarium at the São Paulo State Forest Institute (Instituto Florestal do Estado de São Paulo), or we consulted specialists.

FIGURE 1

Study area map. (a) The study site (orange dot) is located in western Sao Paulo state, Brazil (blue line), at the edge of the Cerrado biome (hashed area). (b) Within the study area (solid black line), plots (orange dots) were established across a gradient from low tree cover to high tree cover. (c) A closed-canopy cerradão forest with ~25 m²/ha of tree basal area. (d) An open-canopy cerrado sensu stricto savanna with ~5 m²/ha of tree basal area. Landsat-8 base layer (b), courtesy U.S. Geological Survey. Images (c) and (d) by Giselda Durigan

Each species was classified as a generalist or a savanna specialist, generally following Durigan et al. (2004). The classification was derived from expert judgement of where the species was most encountered regionally. The classification largely agrees with those made by other methods (Flake, Abreu, et al., 2021). Savanna species occur principally in savannas, whereas generalist species are primarily species of forest environments which can also be found in savannas. There were few forest specialist trees, so we combined them with generalists due to their functional similarity (Flake, Abreu, et al., 2021).

2.3 Soil properties

Soils at the study site are highly weathered, acidic and dystrophic, with high aluminium saturation (>60%) and low base saturation (<15%). They vary from mostly unconsolidated sandy entisols to oxisols with clayey illuvial horizons (Assis et al., 2011). This variation in soil creates a gradient of water and nutrient availability, along with an associated gradient of basal area that existed when the present study began (Assis et al., 2011).

Soil samples were collected in 2009 at depths of 0–20 cm and 60–80 cm for each of the 30 plots. To further reduce the number of soil variables, we considered only soil variables from the 60–80 cm layer, which are less sensitive to short-term changes caused by vegetation (Gray & Bond, 2015). We found the properties of the deeper soil layer to be strongly collinear with measurements of the 0–20 cm layer. A composite soil sample for each plot was made with five subsamples collected at intervals of 10 m along a 50 m line in the centre of the plot. For each sample, we measured fractions of coarse sand, fine sand, silt and clay using the pipette method (Assis et al., 2011). Organic matter, exchangeable cations, labile phosphorus and acidity (H + Al) were measured following Raij et al. (1997). For more details, see Assis et al. (2011). Variables included in our analysis included % sand, %silt, % clay, % organic matter, cation exchange capacity, base saturation, aluminium saturation, and concentrations of Al and nutrient cations: K, P, Ca, Fe, B, Mn and Cu. We used principal components analysis to reduce the dimensionality of correlated soil attributes, and we used the first two axes in our regression models of tree growth and mortality.

2.4 Competition indices

To quantify the intensity of competition, we calculated two alternate measures of the competitive environment, which we refer to as two-sided and one-sided competition, following Weiskittle et al. (2011). The two-sided measure of competitive environment was the total basal area of all trees in a plot (BA_total), and therefore is identical for all trees in the plot. While BA_total is a proxy for total resource use of a stand, and thus resource limitation, metrics of competitive position or one-sided competition is often more strongly predictive of individual tree growth (Sun et al., 2018). As a proxy for one-sided competition, we calculated the total basal of trees taller than the focal tree (BA_above). This variable is calculated for each individual tree, and ranges from zero (for the tallest tree in a plot) to close to BA_total (for the shortest tree in a plot). BA_above accounts for the asymmetry of competition between large trees and small trees.

2.5 Modelling growth and mortality

We modelled tree growth for both time intervals: 2006–2011 and 2011–2016. For each time interval, we included trees which were alive at the beginning and end of the interval (i.e. survivor growth; Vanclay, 1994). To calculate diameter increment, we subtracted the diameter at the beginning of the interval from the diameter at the end of the interval. There were 3,826 trees alive at the beginning of the first interval and 4,564 trees at the beginning of the second interval. We included trees with negative growth, which may occur due to measurement error, natural shrinkage or bark sloughing. However, several trees had large growth declines due to lost stems, and these trees had high leverage in our regression models, which caused biased results. To prevent this problem, we removed trees which grew at <−0.2 cm/year. We removed 101 trees from the first time interval and 100 trees from the second interval. We could not model growth of trees that died, comprising 397 trees in the first interval and 599 trees in the second interval.

We used linear mixed-effects models to investigate the relationship between the average annual diameter increment and several predictor variables: original diameter, BA_above, soil principal components (axes 1 and 2), and savanna or generalist functional type. These variables have sufficiently low pairwise Pearson correlation coefficients (|r| < 0.4; Table S1), and final models had low variance inflation factors (VIF < 1.5) despite these correlations.

To select parsimonious models and compare the influence of predictor variables, we fit several linear mixed-effects models (Table S2) and chose the model with the lowest AICc. We included random intercepts for species to model differences in growth rate not accounted for by their functional group and a random intercept for plot to account for autocorrelation of trees within plots. Including a random intercept in models improved model AIC substantially (by AIC > 20) and reduced heteroscedasticity of residuals. Models were fit in R 4.0.3 (R Core Team, 2020) using package lme4 (Bates et al., 2015). To estimate p-values, we used a Satterthwaite approximation to the denominator degrees of freedom, in package lmerTest (Kuznetsova et al., 2017).

To estimate the probability of mortality, we fit models for two time intervals: one for mortality from 2006 to 2011 and one from 2011 to 2016. We modelled the probability of mortality using logistic regression (generalized linear model with logit link). We modelled mortality as a function of tree diameter, BA_above at the beginning of each time interval, and functional type. Following the same model selection process as for the growth models, we selected the model with the lowest AICc as our final model for each time period. Due to model convergence issues, we did not include any random effects. We assessed model fit using the area under the curve (AUC) of receiver operating characteristic (ROC) plots and using the Hosmer–Lemeshow goodness of fit test (Hosmer Jr. et al., 2013), in r packages pROC (Robin et al., 2011) and ResourceSelection (Lele et al., 2019), respectively.

3 RESULTS

3.1 Stand-level growth patterns

At the plot level, average diameter growth rates declined by more than 50% over the savanna–forest gradient (Figure S1). On a proportional basis, the net increase in basal area (BA) over the decade was lower (or even negative) in stands that began with greater BA_total (Figure 2a). This decline in proportional productivity was much greater for savanna species than generalist species; savanna species underwent a net decline in basal area (i.e. mortality outpaced growth and recruitment) once total basal area exceeded ~17 m²/ha, while the basal area of generalists continued to increase, even in the densest stands (Figure 2a). Much of the decline in savanna species was driven by high mortality at high BA (Figure 2b). If we consider only trees that did not die during the study period (Figure 2c), we found that savanna and generalist proportional productivity was approximately equal. Thus, mortality, rather than declines in growth, plays the predominant role in the productivity declines of savanna species as forest encroaches. These stand-level patterns are supported by individual-level models of growth and mortality.

FIGURE 2

Proportional change in basal area (BA) as a percentage of initial (2006) basal area over the time period 2006–2016. (a) As BA_total increased across the savanna-to-forest gradient, proportional change in basal area decreased substantially, and differed by functional type. (b) This decline in BA increment was driven by a large proportional amount of mortality, with close to 40% of the initial BA lost by savanna species over 10 years in stands with greater basal area, much greater than for generalist species. (c) If declines in BA due to mortality are ignored by only considering trees that did not die during the study period, essentially subtracting panel (b) from panel (a), then savanna and generalist BA patterns are approximately the same

3.2 Individual growth and mortality models

The mean diameter growth rate overall was 0.25 cm/year, with diameter growth of savanna species averaging 0.23 cm/year and generalists slightly faster, 0.26 cm/year (t₁₂₅₀ = 3.98, p < 0.001) before accounting for covariates. Growth varied widely among species. Excluding rare species, the species with the slowest diameter growth were the savanna tree, Leptolobium elegans, and generalist, Tabebuia ochracea, which grew at ~0.7–0.9 times the mean growth rate, while the generalist trees, Maprounea guianensis and Vochysia tucanorum, grew at ~1.2 times the mean rate, after accounting for differences in size, functional type, soils and competitive neighbourhoods (Table S3).

We chose one final model for diameter growth, height growth and mortality in 2006–2011 and 2011–2016 (Table S2). The final model (Model 10) in five of the six cases included log(initial tree diameter), quadratic effects for soil PC1 (associated with soil texture, pH and cation exchange capacity) and PC2 (associated with soil nutrients), functional type (FT), BA_above, and the interaction between BA_above and FT. For mortality in 2006–2011, the same parameters were chosen with the exception of soil PC2 (Model 8). Parameter estimates were very similar between the two time periods (Tables S4–S6), with general agreement on the sign and magnitude of effects.

Model fit was moderately good, with conditional R² (Nakagawa & Schielzeth, 2013) ranging from 0.25 to 0.47 (for full models with random intercepts) and marginal R² ranging from 0.06 to 0.17 (only considering the fixed effects). Our models of growth had heteroskedastic residuals, but bootstrap estimates of the standard errors of coefficients confirmed that our inferences are robust to this heteroscedasticity (Tables S4 and S5). For more details of model fit and model validation, see Supplementary Materials: Model validation.

3.3 Competition reduces growth rates

Competition was associated with dramatically reduced individual tree growth rates, regardless of whether competition is measured total basal area (BA_total; Pearson's r = −0.27; see Figure S1 for plot-level correlation), canopy cover (Pearson's r = −0.20) or BA of trees taller than the focal individual (BA_above; Pearson's r = −0.38; Figure 3). All three of these variables are collinear, however, so we only included BA_above in the model selection process due to its much greater correlation with growth rates and interpretability as an individual-level variable.

FIGURE 3

Growth patterns differ between savanna and forest species for both diameter increment (a) and height increment (b). Both panels show partial effects from Model 10 (Table S2), fit for different response variables. For both time intervals, both savanna species and generalists had large declines in growth at greater BA_above, and generalists grew faster at low BA_above, that is, when they were not shaded by competitors. See Tables S4 and S5 for coefficient estimates

Competitive position (BA_above) was associated with strong declines in diameter growth and height growth (p < 0.001 for three of four models; p = 0.045 for interaction between BA_above and functional type for height growth, 2011–2016; Tables S4 and S5). Generalist species had a stronger decline in diameter growth with increasing BA_above (Figure 3a; Table S4), compared to savanna specialists, due to their much higher growth rates at low BA_above. Though the magnitude of the effect was similar in both time periods, this interaction was only significant for 2011–2016. At high BA_above, that is, when suppressed in the understorey of a forest, generalist species and savanna species had similar diameter growth rates, but generalists maintained greater height growth than savanna species (Figure 3b).

3.4 Savanna species die when overtopped

Though surviving savanna and generalist species grew at similar rates when they were shorter than their neighbours, savanna species died at a much higher rate. Both savanna and generalist species had a small chance of mortality when they were taller than most of their competitors, but mortality rates were much greater for trees suppressed under high BA_above. Savanna species were much more sensitive to BA_above (significant interaction between BA_above and FT; Figure 4; Table S6). For savanna species suppressed by a high basal area of taller competitors (25 m²/ha), our model predicts that over 8% of individual trees will die each year, compared to about 3% of generalists.

FIGURE 4

Probability of mortality as a function of competition differs between savanna and forest functional types. BA_above is an individual-level measure of one-sided competition, with greater values indicating that a greater basal area of trees within the stand have height greater than or equal to that of the focal tree. Probability of mortality was estimated for two 5-year intervals using logistic regression and annualized for presentation (y-axis) using an exponential transformation. See Table S6 for coefficient estimates

3.5 Savanna species become overtopped faster than generalists

BA_above increased over time for most trees, but it increased faster on average for savanna trees compared to generalists (Figure 5). For open sites (with BA_total < 10 m²/ha), savanna species had on average a higher BA_above (8.5 m²/ha, compared to 4.7 m²/ha for generalists; t₇₆₂ = 7.97, p < 0.001), and BA_above increased at a faster rate for savanna species than generalists (0.36 m² ha⁻¹ year⁻¹ vs. 0.25 m² ha⁻¹ year⁻¹, t₂₆₂₀ = −3.96, p < 0.001). The difference was greater for newly recruited trees; for trees that recruited in 2011, both savanna and generalist species recruited at a similar BA_above, but by 2016 savanna species had experienced substantial increases in BA_above as they became overtopped at a rate of 0.40 m² ha⁻¹ year⁻¹. Recruits of generalist species mostly maintained their competitive position; BA_above increased by an average of only 0.14 m² ha⁻¹ year⁻¹ for generalists, more slowly than for savanna species (t₉₀₂ = −3.89, p = 0.001; Figure 5).

FIGURE 5

Change in BA_above over time for each functional type. Data are from plots which were open stands (BA_total < 10 m²/ha) in 2006. BA_above is shown for 100 randomly chosen trees that were alive in 2006 and 2016 (light lines) to demonstrate individual tree patterns. Linear fits are shown for each functional type (heavy lines) for all trees which were added in the initial 2006 inventory (solid line) and for recruits in the 2011 inventory (dashed lines). For each set of lines, the savanna species have a steeper increase in BA_above over time, and the difference in slopes is greater for the 2011 recruits. For savanna species recruited in 2006, BA_above increased by 0.36 m² ha⁻¹ year⁻¹, compared to 0.24 m² ha⁻¹ year⁻¹ for generalists. For 2011 recruits, BA_above increased by 0.40 m² ha⁻¹ year⁻¹ for savanna species and 0.14 m² ha⁻¹ year⁻¹ for generalist species

3.6 Height allometry differs by functional type and environment

For small trees (5 cm ≤ DBH ≤ 7 cm), height:diameter ratios differed between functional types, with generalists generally taller than savanna specialists (F_1,3680 = 60, p ≪ 0.001; Figure 6). Trees in closed stands also had a greater height:diameter ratio than those in open stands (F_1,3680 = 249, p ≪ 0.001).

FIGURE 6

Height:diameter ratios differ among functional types and between environments. Height:diameter ratios are given for savanna specialists (S) and generalists (G) in open savanna environment (BA_total < 10 m²/ha) and woodland or forest environment (BA_total > 10 m²/ha) for small trees (5 cm ≤ DBH ≤ 7 cm) measured in 2011. All pairwise differences between groups are significant (Tukey's HSD, all p < 0.01)

3.7 Soils have minor effects after accounting for competition

The first two principal components of soil variables explained 43% and 31% of the variation in soils (Figure S2). PC1 was primarily associated with soil texture, with lower values having coarser texture, higher pH, and lower cation exchange capacity and aluminium saturation. PC2 was primarily associated with soil nutrients, with positive values of PC2 having higher concentrations of total bases and specific base cations such as K, Ca and P.

Soil principal components were included in all of the selected final models, though the effects were typically small. For diameter growth, soil variables had small effects and were not statistically significant, except for soil PC2 (associated with greater base cation concentrations) in 2006–2011. Mortality was lowest at intermediate values of soil PC1 and soil PC2. Height growth was much higher at positive values of PC1 (soils with finer texture and lower pH) and intermediate values of PC2. For more details, see Supplementary Materials: Soil analysis results.

4 DISCUSSION

4.1 Savanna tree species decline rapidly in closed-canopy stands

Across the gradient from open savanna to closed-canopy forest, average tree growth declined dramatically (Figure S1), and mortality rates increased substantially (Figures 2b and 4). These effects were more acute among savanna species (Figures 3 and 4). Specifically, once total basal area exceeds ~17 m²/ha, savanna species undergo a net decline in basal area (Figure 2a), largely due to increased mortality (Figure 4). Above this BA, losses from mortality exceed gains from growth and recruitment, thereby setting an upper limit on the attainable BA and limiting the ability of savanna species to form a closed-canopy forest. In contrast, generalists continued to accumulate basal area over the full range of stand densities examined, highlighting their important role in forest encroachment. Our individual-level results parallel the stand-level results of Passos et al. (2018), but our results clarify that the loss of savanna trees is driven by high levels of density-dependent mortality of savanna species.

4.2 Generalists grow faster than savanna species

Because generalist species grow faster and experience less shade-induced mortality, the presence of generalist species can rapidly accelerate the rates of canopy closure during stand development. While the role of generalist trees had in the past been inferred from community composition (Flake, Abreu, et al., 2021; Geiger et al., 2011), growth rates (Rossatto et al., 2009) and biomass assessments (Silva et al., 2013), we extend these results to quantify the effects of these differences on stand dynamics. Our results support past findings that savanna and generalist species differ in their growth rates and sensitivity to competition (Hoffmann et al., 2004; Rossatto et al., 2009). Our results indicate that the fast growth of generalists is not due to their ability to grow in the shade, however, but rather because of their capacity to take advantage of available resources in the open. The observation has previously been made that generalist species drive the process of forest encroachment (Geiger et al., 2011); here we demonstrate that this phenomenon is due to their fast growth when released from competition and their lower susceptibility to competition-induced mortality.

4.3 Competition drives growth and mortality

The sensitivity of growth and mortality to competitive position (BA_above; Figures 3 and 4) demonstrates that asymmetric competition is an important process structuring savannas and forests. The consequences of asymmetric competition are well studied in temperate and wet tropical forests and are foundational to silvicultural practices (Binkley et al., 2002; Oliveira et al., 2019; Weiner, 1990). As basal area of canopy trees increases, understorey trees become increasingly suppressed, resulting in lower growth and elevated mortality rates (Binkley et al., 2002). While BA_above accounts for the competitive position of a tree within a 0.1-ha plot, it does not capture finer-scale aspects of competition that may be explained by a distance-dependent metric (Canham et al., 2006; Uriarte et al., 2004; Weiskittel et al., 2011). For example, spatial aggregation of trees or the presence of canopy gaps has large effects on growth, but its influence is absent in our BA_above metric. There is likely more heterogeneity in the competitive environment than modelled here, with potentially important impacts on the spatial arrangement of trees which impacts tree recruitment (Abreu et al., 2021) or effects of fire (see below, Implications for fire regimes).

4.4 Overtopping and the mortality spiral

While the fast growth of open-grown generalist trees may drive the early stages of forest encroachment, in closed-canopy stands diameter growth rates are similar between the functional types (Figure 3). The slow growth of overtopped trees (i.e. those with greater BA_above) implies that they will generally fail to improve their competitive position, compared to taller trees (Figure 5). Savanna species are more likely to become overtopped because they have slower height growth (Figure 3b) and are shorter for a given diameter (Figure 6; Flake, Abreu, et al., 2021). This suppression tends to leave trees susceptible to disease or herbivory that further reduces their ability to compete, creating a positive feedback which leads eventually to tree death (Franklin et al., 1987). Once generalist trees establish in a savanna and begin to overtop savanna trees, the savanna species are likely to enter this ‘mortality spiral’ while generalists, given their lower mortality rates, are more likely to persist in the understorey where they can take advantage of future canopy gaps (Franco et al., 2014).

This overtopping and increased mortality of savanna species appears to be widespread during forest encroachment. Other studies have found that transitional forests experience high rates of mortality as the original cohort dies out (Marimon et al., 2014), leading to rapid species turnover (Abreu et al., 2017; Morandi et al., 2016). Past studies also agree that savanna species are less shade tolerant than generalists (Durigan et al., 2004; Franco et al., 2014), and savanna specialists are unlikely to recolonize closed-canopy stands.

4.5 Mechanisms of competition: Light, water or nutrients?

While productivity of savanna species is especially limited by stand density, probably due to light limitation, several mechanisms could contribute to this pattern. Other mechanisms, such as competition for water or soil nutrients, may be important limits to growth, but our results did not show a clear signal of greater growth on soils with greater available water capacity or base cation concentration. Sites in cerrado and abandoned pasture with finer soil texture support greater tree basal area and are more likely to support forests rather than savannas (Assis et al., 2011; Cava et al., 2019), but we found that soil texture was not a strong predictor of diameter growth or mortality (Figure S3). This lack of effect may be due to past patterns of stand development. There is greater competition on the finer soils, which leads to limited light (Assis et al., 2011), reduced water availability (Honda & Durigan, 2016) and, ultimately, slower growth and greater mortality, as seen in the present study. Especially during the dry season, water limitation caused by interception and greater transpiration is a potentially important mechanism of competition, leading to higher mortality in closed stands, an effect which may contribute to our observed effects of competition.

In contrast to diameter growth, trees on sites with finer soils (greater values of soil PC1) had much faster height growth. This may reflect primarily a change in tree morphology associated with light limitation. Trees in the understorey of a forest (which tend to occur on finer soils) may allocate more resources to height growth, leading to shifts in height:diameter ratios (Figure 6). In savannas, this stem etiolation may contribute to rapid shifts in ecosystem states as shading induces changes in stand structure (Veenendaal et al., 2015). Taken together, our results imply that the apparent importance of soil texture to rates of forest encroachment will depend strongly upon the time since encroachment began, as optimal sites will be filled sooner by forest, inducing changes in growth rates and tree morphology.

Competition and species composition are not the only important endogenous mechanisms that might affect rates of forest growth. Resolving where forests develop faster—and where they are found at any given time—depends on understanding the ecological filters constraining germination and seedling establishment as well as growth and survival of established trees (Pearson et al., 2003). In the seasonal climate that characterizes the Cerrado biome, the upper soil layers are extremely dry over long periods in sandy soils, especially if canopy cover and relative air humidity are low, perhaps limiting forest establishment (Hoffmann et al., 2004). Understanding the overall pattern of forest cover will require analysis of demographic transitions including the regeneration niche.

4.6 Implications for fire regimes

Differences in growth rates of young trees are amplified by fire regimes, which can impose a demographic bottleneck by top-killing small stems (Higgins et al., 2000). The faster the growth rate, the sooner trees may reach fire-resistant sizes (Rodriguez-Cubillo et al., 2021; Trouvé et al., 2021), and the sooner a patch of trees may reach thresholds of canopy leaf area that shade out grassy surface fuels (Hoffmann et al., 2012). Our results make an important prediction for this process: isolated trees, or at least dominant trees, are more likely to grow quickly and reach this threshold, which is of particular importance for thin-barked generalists (Rodriguez-Cubillo et al., 2021). Aggregated trees are likely to be suppressed and reach fire-resistant sizes more slowly (Trouvé et al., 2020), though this is perhaps balanced by the ability for a patch of trees to suppress grassy fuels more readily through their combined leaf area. This variation in growth is an important factor determining which stems are able to reach maturity and which are susceptible to topkill (Hoffmann et al., 2020; Wakeling et al., 2011). Incorporating this source of growth variation, which depends on the spatial arrangement of trees and their neighbours, will likely have large effects on interactions between demographics, fuel and fire regimes, perhaps propagating to landscape-scale effects on tree cover (Li et al., 2019).

5 CONCLUSIONS

The early phases of forest encroachment into savanna, when stands are near thresholds of tree cover that influence fire behaviour (Hoffmann et al., 2012; Newberry et al., 2020), are critically important to understanding the dynamics of forest–savanna ecotones. Tree growth is considerably faster in these early phases, attributable to low competition for light and consequent low mortality and the establishment of young generalist species that grow faster than savanna species. The faster growth and greater height:diameter ratio of generalists lead to rapid overtopping and mortality of savanna species. Growth and mortality were driven predominantly by these endogenous factors rather than soils, and the density dependence of mortality is strong enough that savanna species likely cannot form a forest structure on their own. Modelling the rates of forest formation in savannas requires a recognition of the distinct roles of savanna and generalist species, which inform the effectiveness of fire or other disturbances in maintaining savannas in an open condition.

ACKNOWLEDGEMENTS

We thank all the students and field assistants of Instituto Florestal who contributed to stand inventories. The manuscript was much improved thanks to the input of Lorena Gomez Aparicio, Niall Hanan, Patrick Baker and an anonymous reviewer. S.W.F. and W.A.H. were supported by NSF grant DEB1354943, and S.W.F. by a USGS Southeast Climate Adaptation Science Center (SECASC) Global Change Fellowship. The installation of the permanent plots was supported by Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq, grant #475286/2007-3). NALP was supported by Fundação de Amparo à Pesquisa do Estado de SP (FAPESP, grant #2016/17888-2), and G.D. was funded by CNPq (grant #303179/2016-3).

CONFLICT OF INTEREST

The authors declare no conflicts of interest.

AUTHORS' CONTRIBUTIONS

G.D., W.A.H. and S.W.F. conceived the ideas and designed the methodology; G.D., N.A.L.P. and E.A.H. collected the data; S.W.F. analysed the data; S.W.F. led the writing of the manuscript. All authors contributed critically to the drafts and gave final approval for publication.

Open Research

PEER REVIEW

The peer review history for this article is available at https://publons.com/publon/10.1111/1365-2745.13820.

DATA AVAILABILITY STATEMENT

Data used in this analysis are available on Dryad Digital Repository https://doi.org/10.5061/dryad.vq83bk3tm (Flake, Honda, et al., 2021).