Research articlesInvasive alien acacias rapidly stock carbon, but threaten biodiversity recovery in young second-growth forests
Abstract
Under the UN-Decade of Ecosystem Restoration and Bonn Challenge, second-growth forest is promoted as a global solution to climate change, degradation and associated losses of biodiversity and ecosystem services. Second growth is often invaded by alien tree species and understanding how this impacts carbon stock and biodiversity recovery is key for restoration planning. We assessed carbon stock and tree diversity recovery in second growth invaded by two Acacia species and non-invaded second growth, with associated edge effects, in the Brazilian Atlantic Forest. Carbon stock recovery in non-invaded forests was threefold lower than in invaded forests. Increasingly isolated, fragmented and deforested areas had low carbon stocks when non-invaded, whereas the opposite was true when invaded. Non-invaded forests recovered threefold to sixfold higher taxonomic, phylogenetic and functional diversity than invaded forest. Higher species turnover and lower nestedness in non-invaded than invaded forests underpinned higher abundance of threatened and endemic species in non-invaded forest. Non-invaded forests presented positive relationships between carbon and biodiversity, whereas in the invaded forests we did not detect any relationship, indicating that more carbon does not equal more biodiversity in landscapes with high vulnerability to invasive acacias. To deliver on combined climate change and biodiversity goals, restoration planning and management must consider biological invasion risk.
This article is part of the theme issue ‘Understanding forest landscape restoration: reinforcing scientific foundations for the UN Decade on Ecosystem Restoration’.
1. Introduction
The United Nations Decade on Ecosystem Restoration started in 2021, attracting the attention of multiple stakeholders to the needs and opportunities of reversing degradation at the planetary scale [1–3]. This initiative echoes several recent, ambitious restoration programmes that aim to plant trillions of trees and restore millions of hectares [1]. Many of these initiatives, including the 350-million hectare Bonn Challenge, adopt ‘forest and landscape restoration' (FLR), an approach based on the expansion of tree cover through multiple tree-planting and reforestation schemes, including commercial plantations of exotic species [4]. In fact, 45% of the restoration area committed to the Bonn Challenge consists of large-scale commercial monocultures of questionable environmental benefits [5].
Exotic species may play important roles in FLR due to their rapid recovery of fuelwood and timber, financial benefits to farmers and potential to foster or, at least, not harm forest regeneration [6,7]. Yet some exotics can become invasive alien trees [7–9]. For example, given their high capacity for nitrogen fixation, which can reduce fertilizer use, improve soil quality and promote rapid growth, Acacia mangium and A. auriculiformis (hereafter Acacia spp.) were recently introduced outside their natural Australasian distribution in North and South America, Africa and Asia. But in many regions where they have spread rapidly and widely in both natural and perturbed environments they are now considered invasive alien tree species [10,11].
Invasive alien plant species can (i) change soil properties, nutrient cycling and ecosystem productivity [12,13], (ii) interfere in the establishment of native species [14], compromising the conservation of natural ecosystems, biodiversity, and ecosystem services [13,15,16] and (iii) represent a source of economic harm [17,18]. Invasive plant species thus represent critical barriers for restoring native ecosystems [19], and understanding how alien tree species intersect with carbon stock and biodiversity recovery is key for restoration planning.
Second-growth forest can cost-effectively mitigate climate change and biodiversity loss [2,3,20], particularly under favourable land-use history [21] and landscape characteristics [3]. However, invasive alien tree species can undermine the establishment of second-growth forests [11]. For example, invasive alien tree species can restrict the establishment of native species [22] and promote changes in species composition and community structure [23,24], ultimately with the potential to disturb carbon-biodiversity co-benefits provided by second-growth forest [2,3,20]. Considering the massive funding shortfall for conservation of carbon stocks and biodiversity [2,3], understanding potential trade-offs caused by biological invasion can help conservation practitioners to appropriately plan restoration that contributes to delivering on the ambitious goals of the UN Decade on Ecosystem Restoration and the Bonn Challenge.
Here, we focus on young secondary-growth forests (with up to 20 years of natural regeneration after land abandonment) that have high levels of associated edge effects, spanning those dominated by native species (hereafter ‘non-invaded forests') or by Acacia spp. (hereafter ‘invaded forests') within the globally threatened Brazilian Atlantic Forest [25], a hotspot for tropical rainforest restoration [26]. We answer five main questions: (i) What is the difference in above-ground carbon stocks between non-invaded and invaded forests, and how rapidly does carbon recover over time? (ii) What is the relationship between isolation at the patch scale, and configuration (fragmentation per se) and composition (forest cover and open area) at the landscape scale, with carbon stocks in non-invaded and invaded forests? (iii) What is the difference in taxonomic, phylogenetic and functional diversity (FD) between the non-invaded and invaded forests, and how do they recover over time? (iv) What is the difference in β-diversity and species of high conservation value between the non-invaded and invaded forests, and how do they change over time? (v) How do biodiversity and species composition recover relative to carbon stocks for non-invaded and invaded forests?
2. Material and methods
(a) Study sites
Our study was based in the Atlantic Forest domain [27] in Espírito Santo state, east Brazil (figure 1a). The studied landscape represents a form of coastal forest, termed Floresta de Tabuleiro [3,28], typified by large flat areas rising slowly from 20 to 200 m.a.s.l. Within our study areas, these forests are predominantly seasonal and their physiognomies may be evergreen or semi-deciduous depending on the severity of water deficit in the dry season [29]. The prevailing climate is wet tropical (approx. 1228 mm yr−1; Köppen climate classification), with a dry season (less than 60 mm yr−1) usually from May to September, and with lower temperatures in the dry season (approx. 15.6°C) than in the wet season (approx. 27.4°C) [29]. Electronic supplementary material, appendix S1–table S1 includes additional information about study sites.
Figure 1. Study area and dominance patterns in two communities in the globally threatened Brazilian Atlantic Forest. (a) Study area; (b) proportional importance of the most abundant species by forest types; (c) number of individuals per hectare of native and Acacia spp. species; (d) proportion of sampled plots containing native species and Acacia spp.; and (e) basal area of native species and Acacia spp. Whiskers at the top of the bars indicate standard error and asterisks indicate a significant difference between treatments (p < 0.001).
Forests in our study landscape remained preserved until the 1950s, when large clear-cut logging and charcoal production operations began, followed by agro-pastoral expansion. Currently, the region is highly fragmented [30–32], with old-growth and second-growth forests embedded in a matrix of planted pasturelands, croplands (mainly sugarcane, coffee and papaya) and commercial tree plantations (mainly Eucalyptus spp. and Acacia spp.; [3,33,34]).
(b) Experimental design
Fieldwork was conducted between 2017 and 2021 across two forest types: (i) non-invaded forests—second-growth forests where only native plant communities assembled after abandonment of planted pasturelands. Sampled forests were 5–20 years old and ranged in size from 9 to 203 ha (49 ± 17 s.e.) and in width (up to 810 m; 471 ± 71 s.e.); and (ii) invaded forests—second-growth forests where two invasive alien species (Acacia mangium Willd. and Acacia auriculiformis Cunn. ex Benth) dominated naturally regenerating native plant communities after abandonment of planted pasturelands. Sampled forests were 4–12 years old and ranged in size from 4 to 146 ha (51 ± 20 s.e.) and width (89–623 m; 258 ± 54 s.e.). Yellow Argisol soils dominated sample sites across forest types (electronic supplementary material, appendix S1–table S1 includes description of the study sites and electronic supplementary material; table S2 soil classes).
Acacia mangium and A. auriculiformis have their natural distribution in Australia, Papua New Guinea and Indonesia [35,36]. Both species occur in hot and humid tropical regions, with annual temperatures ranging from 18°C to 28°C, rainfall between 1000 and 4000 mm yr−1 and altitudes ranging from 0 to 700 m (for full details see: https://www.cabi.org/isc/datasheet/2325 and https://www.cabi.org/isc/datasheet/2157). Owing to their varied uses (e.g. production of wood, charcoal, recovery of degraded areas), both species are planted outside their natural area of distribution in Asia, Southern Africa and South America [11,36–40]. In the Brazilian Atlantic Forest, both species were introduced around the 1970s for commercial forestry purposes [11]. In our study landscapes, the first records of A. mangium were in experimental areas in approximately 1984, with the aim of producing timber and recovering degraded areas [37]. There were no records of the introduction of A. auriculiformis in the sampled landscapes.
In this study, both Acacia species were treated as a single species (i.e. Acacia spp.), given their capacity to form natural hybrids [41], which limits morphological separation of species during fieldwork. Invaded forests were characterized by very high species dominance, far higher than non-invaded forests (figure 1b). High species dominance in invaded forests occurs in response to: a great number of individuals of Acacia spp. relative to native species (figure 1c); the absence of native species in over half the plots (figure 1d); and a basal area of Acacia spp. an order of magnitude higher than of native species (figure 1e). Electronic supplementary material, appendix S1 includes description of the invaded second-growth forests, analysis of metrics and the data depicted in figure 1b–e.
(c) Vegetation sampling
For each forest type (figure 1a), we sampled 11 fragments. In each fragment, we sampled 10 plots of 10 m × 10 m (0.1 ha), which were installed systematically at 20 m intervals along each transect. The plots in non-invaded forests were installed between 8 and 583 m (102 ± 49 s.e.) from the forest edge and in invaded forests between 5 and 45 m (22 ± 4.35 s.e.). Thus, aside from the non-invaded plot at 583 m (second largest distance to forest was 114 m), all our forest plots were substantially edge effected [31,32,42,43]. Within each plot, both the shrub and arboreal strata were sampled, including all individuals rooted within the plot with d.b.h. ≥ 4.8 cm (1.30 m above-ground height; following [3]). All plants were identified during fieldwork by botanists. To confirm identifications, vouchers of each morphospecies were compared to botanical collections of the Herbarium of Vale (CVRD) and Herbarium of São Mateus (SAMES).
(d) Characteristics of second-growth forests
Ages of second-growth forests were obtained using open-access satellite images of surface reflectance with 30 m resolution since 1984 (LANDSAT TM4, 5–8; U.S. Geological Survey and NASA). After downloading, raw satellite layers were split using a multiband raster into single bands, which are easier to classify and then manipulate associated in a unique consensus layer. For each generated consensus layer, all regions of interest (ROIs) were determined by polygons. Next, the image was classified with the Semi-Automatic Classification Plugin function in the Geographic Information System-QGIS, version 2.18.4 [44]. Finally, all classifications were manually inspected to assess the chronological sequence of land-cover change to detect the year when each second-growth forest started regrowth.
The relationship between patch and landscape characteristics and carbon stock of second-growth forests was assessed by patch-scale and patch-landscape approaches [45]. For patch-scale, carbon stock was obtained by transect, while two patch characteristics were measured from independent habitat patches: (1) isolation (m)—distance to the nearest neighbouring patch, based on shortest edge-to-edge distance. High values indicate greater inter-patch distance independent of the patch size, and the opposite indicates lower distance; and (2) source distance (km)—based on shortest edge-to-edge distance between the sampled patch to fragments of natural forests of 500 ha or more. High values indicate greater distance between sampled patch and larger fragments (500 ha or more), and the opposite indicates lower distance. For patch-landscape, carbon stock was obtained per transect, while three landscape metrics were measured in an independent circular buffer of 2 km from the middle of each transect: (1) splitting index (without unity) measures the number of patches when dividing the landscape into patches of equal size, independent of total forest cover [46]; a measure of landscape configuration. Splitting index quantifies the level of division of the landscape [46], interpreted as the degree of fragmentation per se [47,48]. High values indicate a highly fragmented landscape, and the opposite indicates a less fragmented landscape; (2) forest cover (%)—the percentage of the landscape coverage by natural forests; and (3) open area (%)—the percentage of the landscape matrix composed of cattle pasture, the most common land use in our landscapes [3]; both are measures of landscape composition. Patch and landscape metrics were measured from the cover and land-use map of MapBiomas (http://mapbiomas.org, collection 4). Electronic supplementary material, appendix S1 includes values obtained for each metric and a full description of patch and landscape analyses.
(e) Above-ground carbon stock
The above-ground biomass (AGB) of each sampled fragment, including Acacia spp. in the invaded second-growth forests, was calculated using an established pantropical algorithm [49]. The value for wood density in dry weight (gcm−3) was obtained from Global Wood Density database–GWD [50]. When a species was identified at the genus level or was not present in the GWD database, we used the average density of wood for all species of the same genus in the database [3,30,51]. Finally, we obtained the value of carbon for each sampled transect assuming that 50% of AGB of each individual is comprised of carbon [52].
(f) Species diversity
Taxonomic diversity (TD) was the rarefied species richness based on sample sizes [53]. This measure was interpreted as α-diversity [3], that is, the total number of species sampled from each forest type, and was obtained using package iNEXT [53] in the R environment [54].
Phylogenetic diversity (PD) was based on Faith's index, characterized by the sum of branch lengths connecting all species in an assemblage, and representing the sum of evolutionary history [55]. To obtain PD, all sampled individuals were classified in their respective species, genus and family in accordance with the APG IV [56]. The above classification was then submitted to ComTreeOpt function to build and optimize community tree topology [57]. Last, phylogenetic hypothesis was calibrated using mean age estimates for internal nodes [58] and the BLADJ algorithm in Phylocom program version 4.2 [59].
FD is the total branch length of a functional dendrogram and it measures the extent of complementarity between species trait values [60]. FD was measured using four steps: (1) trait matrix was obtained for ten functional traits related to food resource for fauna (fruit size (mm), seed size (mm) and fruit type (fleshy or non-fleshy)), fruit dispersal syndrome (animal-dispersed or abiotic-dispersed), tree succession group (early, intermediate and late-successional species) and carbon storage (wood density in dry weight gcm−3). Traits related to food resource and succession group were obtained from specimens in Herbarium CVRD, SAMES and the literature [3,30,33], while carbon storage was obtained from the Global Wood Density database–GWD [49]. Electronic supplementary material appendix S1–table S3 includes descriptions all functional traits. (2) Gower's distance was applied to create a distance matrix from continuous and categorical functional traits [61]. (3) The distance matrix was clustered to produce a dendrogram grouping the species using the methods of UPGMA (unweighted pair-group method using an arithmetic average). To verify the loss of information when we transformed the distance matrix into a dendrogram, we tested Pearson's correlation between the original matrix and the dendrogram cophenetic matrix. We did not find great loss of information (r = 0.934). (4) Finally, the functional dendrogram was transformed into a tree of class phylo using the ‘as.phylo' function available on ape package [62] in the R environment [54], thus allowing the dendrogram to have the same structure as a phylogenetic tree.
PD and FD are usually correlated with species richness [3]. To reduce the influence of species richness on PD and FD, we calculated the standardized effect size of phylogenetic (i.e. sesPD) and functional (i.e. sesFD) diversity [63]. First, we generated random values of PD and FD for each fragment sampled by shuffling the tips of the PD and FD 999 times. Then, the observed PD and FD values were subtracted from the average of simulated values and divided by the standard deviation of simulated PD and FD to calculate sesPD and sesFD, which control for the potential effects of species richness on PD and FD. Communities with sesPD and sesFD values close to 1 (i.e. high quantiles) indicate greater phylogenetic and functional distances among co-occurring species than expected by chance. Communities with ses close to 0 (i.e. low quantiles) indicate smaller phylogenetic and functional distances among co-occurring species than expected by chance. We calculated PD, FD and their standardized effect size of diversity using the picante package [64] in the R environment [54]. All metrics related to species diversity in the invaded second-growth forests were obtained incorporating Acacia spp.
(g) Species composition
To assess the difference in community organization between non-invaded and invaded forests, decomposition of Sørensen dissimilarity-βSOR for all sites was applied using presence–absence data [65]. Sørensen dissimilarity was decomposed into spatial turnover-βSIM and nestedness-resultant-βSNE. βSIM captures only compositional changes due to species turnover, whereas βSNE is derived from species gains and losses or nestedness and represents the difference between βSOR and βSIM [65]. Communities with βSIM values close to 1 indicate a greater species turnover, whereas communities with values close to 0 indicate a lower species turnover, hence a higher similarity. Communities with βSNE values approaching 1 represent a site hosting a subset of species from another site, whereas communities with values approaching 0 represent a unique composition among the sites sampled [65]. We calculated β-diversity using the betapart package [66] in the R environment [54]. All metrics related to species composition in the invaded second-growth forests were obtained incorporating Acacia spp.
(h) Conservation value of species
Conservation value was considered the abundance of threatened and endemic species [3,31]. Threatened species were obtained from the Red List of the International Union for Conservation of Nature (https://www.iucnredlist.org/), whereas endemic species of the Atlantic Forest domain were classified using the Flora of Brazil (https://bit.ly/2G1W2D2).
(i) Statistical analyses
We applied Moran's I test between sites to check spatial autocorrelation of ten response variables of second-growth forests. Significance of spatial autocorrelation test was determined by the Monte–Carlo permutation test (999 permutations), using the R package spdep [67]. We found that sites did not show spatial autocorrelation for any of the response variables used (electronic supplementary material, appendix S1–table S4 includes all Moran's I test results).
To test for differences and recovery over time of carbon stocks (question i), biodiversity (question iii), and species composition and conservation value (question iv) between non-invaded and invaded forests, we applied generalized linear models (GLMs) with forest type as the explanatory variable, age of forest type as a covariate, plus the interaction between them. To determine how biodiversity and composition recover relative to carbon stocks (question v) between non-invaded and invaded forests, our GLMs included forest type and carbon stock as explanatory variables, and the interaction between them. For these four questions, we applied the ‘dredge' function in the package MuMIn [68] to determine whether interactions were present in our most parsimonious model (i.e. ΔAICc = 0). Finally, to determine recovery over time of carbon stocks (question i), biodiversity (question iii), and species composition and conservation value (question iv), and biodiversity and composition recovery relative to carbon stocks (v) for non-invaded and invaded forests individually, we constructed separate GLMs.
To investigate the relationship between patch-scale and landscape-scale metrics with carbon stocks (question ii), GLMs were constructed separately using the carbon stock of non-invaded and invaded forests (i.e. only the carbon stock of Acacia spp.) as the response variable, and patch-scale and patch-landscape variables as the explanatory variable. We used multi-model inference and an information-theoretic approach to assess the effect of exploratory variables on carbon stock [69]. Before constructing the models, any pair of explanatory variables that have a correlation r ≥ 0.5 were included in separate models (electronic supplementary material, appendix S1–figure S2 includes all correlation results). For the non-invaded and invaded forests, we built 10 models representing the combination of all uncorrelated exploratory variables and the null model. Subsequently, we rank the models from best to worst based on the Akaike information criterion corrected for small samples (AICc). The set of models with ΔAICc ≤ 5 were considered to have good explanatory power and biological relevance [70]. Model-averaged parameter estimates were calculated using the mean of regression coefficients weighted by the AICc weight (wi). In addition, for each model that was averaged, the relative importance of the exploratory variable was obtained by the sum of AICc weights (wi) from all models where each variable appears [69].
GLMs with Gaussian distributions and an identity link were applied to all questions, since residuals were normally distributed (confirmed by the Shapiro–Wilk test). The exception was the count data for conservation value (questions: iv and v), for which we applied negative binomial distributions with log link functions, having rejected the Poisson model. GLMs are implemented in the ‘glm/glm.nb' function from the MASS package (see https://cran.r-project.org/web/packages/MASS/MASS.pdf) and the comparison of the log-likelihood of the negative binomial regression model with the Poisson regression model was performed using the odTest function from the pscl package [71]. For all the above models, we assessed the goodness of-fit as: (adjusted D2 = (null deviance of model – residual deviance of model)/null deviance of model) [72]. All analyses were performed in the R environment [54].
3. Results
Across all forest types, 2173 individuals of 173 species were recorded. For non-invaded forests, we sampled 605 individuals of 147 species, and for invaded forests 1568 individuals of 26 species.
(a) Carbon stocks of second-growth forests
Carbon stocks of invaded forests were approximately three times higher than in non-invaded forests (D2 = 0.79, t = −8.67, p < 0.001, figure 2a), driven by the rapid increase in carbon stocks of Acacia spp., which represented 91% of the total carbon stock (53.34 Mg ha−1, native species = 5.14 Mg ha−1). There was a positive relationship between forest age and carbon stock (t = 4.72, p = 0.001), but the interaction between age and second-growth type was absent from our best model (ΔAICc = 0). Invaded forests presented a higher rate of carbon stock increment over time (D2 0.56, t = 3.17, p = 0.011) than non-invaded forests (D2 0.52, t = 3.38 p = 0.008; figure 2b).
Figure 2. (a) Impact of non-invaded and invaded forests on carbon stocks, (b) forest age, (c) relationship between patch-scale and patch-landscape attributes in non-invaded forest with the carbon stock carbon, and (d) relationship between patch-scale and patch-landscape attributes in invaded forest with the carbon stocks. Asterisks indicate significance at p ≤ 0.001, and error bars in (a) represent standard error. Solid (invaded) and dashed (non-invaded) lines represent changes in carbon stocks over time. The positive or negative position of the bars represents an effect of the predictive variable on carbon stocks, and errors bars in (c–d) represent the confidence interval obtained after analysis of average models. FC, forest cover; IS, isolation; OA, open areas; SD, source distance; SP, splitting index.
For non-invaded forests, the goodness of fit of the model for the relationship between patch and landscape characteristics and carbon stock was D2 = 0.80, while for invaded forests it was D2 = 0.90. In non-invaded forests, at the patch-scale, isolation (β = −11.54 ± 3.01 s.e., z = 3.82, p = 0.0001) and source distance (β = −10.46 ± 3.55 s.e., z = 2.94, p = 0.0003; figure 2c) had a negative relationship with carbon stock. For invaded forests, isolation (β = 11.40 ± 2.96 s.e., z = 3.84, p < 0.001) and source distance (β = 7.56 ± 2.99 s.e., z = 2.53, p = 0.01; figure 2d) had a positive relationship with the carbon stock. At the patch-landscape level, for configuration we found a negative relationship of splitting index (fragmentation per se) on carbon stock in non-invaded forests (β = −10.59 ± 3.49 s.e., z = 3.03, p < 0.002; figure 2c), but a positive relationship in invaded forests (β = 11.47 ± 3.60 s.e., z = 3.17, p < 0.001; figure 2d).
For landscape composition, percentage of forest cover had a positive relationship with carbon stock in non-invaded forests (β = 10.56 ± 3.50 s.e., z = 3.01, p < 0.02; figure 2c), but a negative relationship in invaded forests (β = −12.07 ± 3.44 s.e., z = 3.50, p = 0.0004; figure 2d). Percentage of open area had no relationship with carbon stock in non-invaded forests (β = −1.64 ± 3.22 s.e., z = 0.51, p = 0.60; figure 2c), but a positive relationship with invaded forests (β = 11.08 ± 2.69 s.e., z = 4.12, p < 0.0001; figure 2d). In order of independent contribution, the predictive variables best explaining relationships with carbon stock in non-invaded forests were isolation (IS), then splitting index (SP), forest cover (FC), source distance (SD) and open areas (OA), whereas in the invaded forests they were SD, followed by IS, OA, FC and SP. Electronic supplementary material, appendix S2, table S5 includes all model selections, electronic supplementary material, table S6 the equations of the models, and electronic supplementary material, figure S3 the contribution of the exploratory variables of the average models.
(b) Diversity of second-growth forests
Taxonomic diversity (TD) in non-invaded forests was six times higher than in invaded forests (D2 = 0.76, t = 4.55, p < 0.001, figure 3a). Non-invaded forests had four times higher phylogenetic diversity (PD) than invaded forests (D2 = 0.79, t = 5.96, p < 0.001; figure 3b). Lastly, non-invaded forests had three times higher functional diversity (FD) than invaded forests (D2 = 0.75, t = 4.88, p < 0.001; figure 3c). Forest age had a positive effect on TD (t = 3.20, p < 0.01), PD (t = 2.22, p = 0.03) and FD (t = 2.57, p < 0.01). Interaction between age and type of second-growth forest was absent from best models for TD, PD and FD. Additionally, for non-invaded forests we found an increase over time of TD (D2 = 0.43, t = 2.61, p = 0.02), PD (D2 = 0.42, t = 2.56, p = 0.03) and FD (D2 = 0.46, t = 2.78, p = 0.02), but no effect in invaded forests to TD (D2 = 0.04, t = 0.65 p = 0.52; figure 3d), PD (D2 = 0.05, t = 0.74, p = 0.47; figure 3e) and FD (D2 = 0.01, t = 0.39 p = 0.70; figure 3f).
Figure 3. Biodiversity within non-invaded and invaded second-growth forest. Impact of second-growth forest type on (a) taxonomic diversity (TD), (b) phylogenetic diversity (PD) and (c) functional diversity (FD); (d–f) impact of forest age on TD, PD and FD; (g–h) impact of second-growth forest type on standardized effect size of PD (sesPD) and FD (sesFD); and (i) impact of age on standardized effect size of sesPD. Asterisks indicate significance at p ≤ 0.001 (a–c, g), p ≤ 0.05 (h); error bars represent standard error (a–c, g–h) and dashed (non-invaded) lines represent changes in biodiversity over time (d–f, i).
After correcting the relationship between diversity and species richness, standardized effect size (ses) was related to second-growth forest type to PD (D2 = 0.65, t = 4.24, p < 0.001) and FD (D2 = 0.25, t = 2.50, p = 0.02), with significantly higher phylogenetic and functional dispersion for non-invaded than for invaded forests (figure 3g,h). Considering only the best model selected on the basis of their AICc values, sesPD at the transect level was significantly influenced by the interaction between age and type of second-growth forest (t = −2.49, p = 0.02), but the interaction was absent from sesFD. Forest age had no effect on sesFD (t = −1.00, p < 0.32). Additionally, for non-invaded forests we found a negative effect on sesPD over time (D2 0.42, t = −2.56, p = 0.03), but no effect in invaded forests (D2 0.17, t = 1.40, p < 0.19; figure 3i). Electronic supplementary material, appendix S2, table S7 includes all model selections and electronic supplementary material, table S8 the equations of the models.
(c) Patterns of β-diversity and species of conservation value
Compositional changes due to species turnover (i.e. βSIM) in non-invaded forests was four-fold higher than in invaded forests (D2 = 0.83, t = 8.31, p < 0.001; figure 4a). In addition, compositional change derived from species gains and losses or nestedness (i.e. βSNE) was eight times higher in invaded forests than in non-invaded forests (D2 = 0.79, t = −7.03, p < 0.001; figure 4b). Forest age had no effect on βSIM (t = 0.28, p = 0.78) and βSNE (t = −0.36, p = 0.72), while the interaction between age and type of second-growth forest was absent in our best model for both metrics.
Figure 4. Species composition and conservation value between non-invaded and invaded second-growth forest. (a) Impact of Brazilian Atlantic second-growth forest type on compositional changes due to species turnover (i.e. βSIM); (b) component derived from species gain and loss or nestedness (i.e. βSNE); (c) abundance of threatened tree species; and (d) effect on abundance of Atlantic Forest endemic species. Asterisks indicate significance at p ≤ 0.001 (a,b), p ≤ 0.01 (d) and p ≤ 0.05 (c) and errors bars represent standard error.
Non-invaded forests had three times higher abundance of threatened species than invaded forests (D2 = 0.07, z = 1.96, p < 0.04, figure 4c). Endemic species abundance was six times higher in non-invaded than invaded forests (D2 = 0.31, z = 3.20, p = 0.001, figure 4d). Forest age had no effect on threatened species abundance (z = −0.37, p = 0.71) and endemic species abundance (z = 1.20, p = 0.22), while the interaction between age and type of second-growth forest was absent from our best model for IUCN Red-listed and endemic species. Electronic supplementary material, appendix S2, table S9 includes all model selections and electronic supplementary material, table S10 gives the equations of the models.
(d) Co-benefits between carbon stock and biodiversity
Considering our best model (ΔAICc = 0), we found an interaction between above-ground carbon stock recovery and the type of second-growth forest for TD (D2 = 0.82, t = 2.11, p = 0.04; figure 5a), PD (D2 = 0.87, t = 4.13, p = 0.0006; figure 5b), FD (D2 = 0.84, t = 4.71, p < 0.001; figure 5c), and sesPD (D2 = 0.69, t = −2.76, p = 0.01; figure 5d), but not for sesFD. For non-invaded forests, there was a positive impact of carbon stock on TD (D2 = 0.60, t = 3.87, p = 0. 003; figure 5a), PD (D2 = 0.69, t = 4.50, p = 0.001; figure 5b) and FD (D2 = 0.70, t = 4.56, p = 0.001; figure 5c). However, for invaded forests, there was no effect of carbon stock on TD (D2 = 0.01, t = 0.30, p = 0.76; figure 5a), PD (D2 = 0.12, t = 1.11, p = 0.29; figure 5b) and FD (D2 = 0.03, t = 0.53, p = 0.60; figure 5c). Increasing carbon stocks in non-invaded forests reduced average values of sesPD (D2 = 0.37, t = −2.34, p = 0.04), whereas there was no effect in invaded forests (D2 = 0.31, t = 2.01, p < 0.07; figure 5d), while increasing carbon stock reduced sesFD independent of type of second-growth forest (D2 = 0.68, t = −2.94, p < 0.008; figure 5e).
Figure 5. Biodiversity and species composition recovery relative to carbon stocks for non-invaded and invaded forests in the Brazilian Atlantic Forest. (a) Carbon stock against taxonomic diversity (TD); (b) phylogenetic diversity in millions of years (PD); (c) functional diversity (FD); (d) standard effect size of phylogenetic diversity (sesPD); (e) standard effect size of FD (sesFD); and (f) abundance of Atlantic Forest endemic species. Solid (invaded), dashed (non-invaded) and dashed-dotted lines (overall) represent relationship between standard effect size of functional diversity (sesFD) and carbon stock.
An interaction between carbon stock and second-growth forest for βSIM, βSNE and threatened species was absent from our best models, but present in our best models for abundance of endemic species (D2 = 0.91, z = 2.04 p = 0.04). Additionally, there was no effect of average carbon stocks on βSIM (D2 = 0.84, t = 0.43, p = 0.67), βSNE (D2 = 0.79, t = 0.09, p = 0.92) and abundance of threatened species (D2 = 0.08, z = −0.68, p = 0.49). For abundance of endemic species, non-invaded forests revealed a positive impact of carbon stock (D2 = 0.91, z = 4.42, p < 0.001), but no effect of carbon stock for invaded forests (D2 = 0.01, z = −0.56, p = 0.57; figure 5f). Electronic supplementary material, appendix S2, table S11 includes all model selections and electronic supplementary material, and table S12 gives the equations of the models.
4. Discussion
Invasive alien trees can dominate tropical landscapes, and a key question is how such species impact the carbon and biodiversity outcomes of restoration programmes focusing on natural regeneration in second-growth forests. Here, we demonstrated that second-growth forests invaded by Acacia spp. accrue carbon stocks rapidly, but they undermine biodiversity recovery relative to non-invaded second-growth forests. In landscapes with high levels of fragmentation, edge effects [30,32] and prevalence of invasive alien trees [11,73], our results suggest that optimizing the combination of carbon accumulation rate with biodiversity recovery requires control of regenerating Acacia spp., plus planting early successional natives to occupy the ecological niche of light-demanding Acacia species.
(a) Carbon stock of second-growth forests
Invaded forests had approximately three-times higher carbon stock than non-invaded forests, likely due to Acacia spp.'s strong competitive ability. Acacia spp. form symbiotic relationships with N2-fixing microorganisms in their root systems [11,12,36] and they are able to tolerate acid soils, low nutrient levels and moderate levels of aluminium [74], driving gradual recovery of the soil fertility [36,38]. This generates higher rates of growth, survival and biomass allocation [23]. These favourable productive characteristics of Acacia spp. have promoted their use in monocultures (e.g. most wood pulp produced in Southeast Asia) and in mixed plantations with eucalypts, the predominant commercial tree in the study region [3,75]. Acacia spp. can also outcompete native pioneers for essential resources, including light, water and nutrients [23,36,40,76], which is potentially a key factor in our study landscape given that soils vary from sand to sandy clay and nutrient-poor, and with potential for droughts [29,33,77]. This competitive ability promotes greater vegetative growth and establishment, especially in degraded areas where landscape permeability caused by fragmentation promotes invasive species [73].
There are also indirect effects of anthropogenic-induced landscape changes, with increasing isolation and fragmentation having a positive relationship with Acacia spp. carbon stocks. First, isolation and fragmentation prevent seed dispersal of native species [78], which compete with Acacia spp. for resources and reduce their biomass accumulation. Second, Acacia spp. have high dispersal capacity, mediated by bird and ants [79]; for example, A. mangium can be dispersed 900 m from the edge of commercial plantations in the Brazilian Amazon [80], whereas dispersal of native pioneer trees in the Atlantic Forest drastically declines over 50 m from forest remnants [78]. In addition, increasing prevalence of open areas leads to increased Acacia spp. carbon stocks, and the opposite for native forests, likely due to shade intolerance of Acacia spp. versus shade tolerance of most native tree species. The quality of the intervening matrix as a dispersal conduit between native patches [7,81,82] thus also facilitates the dominance of carbon stocks in invasive alien tree species.
(b) Diversity and composition of second-growth forests
Invaded forests had only 16% of taxonomic, 24% of phylogenetic and 30% of FD found in non-invaded forests, while tree diversity increased with stand age in non-invaded forests but not in invaded forests. Previous research supports the increasing recovery of TD, PD and FD across a range of taxa as native second-growth forests mature [2,3,20,51]. The minimal recovery of diversity in invaded forests mirrors other studies revealing the potential of invasive trees species to transform ecosystems [11,12,23], resulting in high dominance and diversity changes [14,22] and ultimately compromising ecosystem services [23]. In fragmented and isolated landscapes with high prevalence of invasive alien trees [11,73], the potential for diversity recovery is questionable.
Invaded forests presented only 36% of the IUCN Red-listed species and 14% of endemic species abundance of non-invaded secondary forest, with no enhancement with forest age, suggesting arrested recovery of conservation value in non-invaded forests. Thus plant communities assembled in the presence of invasive tree species have increased floristic homogeneity, while non-invaded forests increase floristic differentiation (increased β-diversity) [83]. Invaded forests had low compositional changes due to species turnover (spatial turnover-βSIM) and high nestedness (βSNE), with floristic homogenization driven by the dominance of Acacia spp. Limited recruitment and growth of native species is likely guided by key Acacia spp. traits, such as nitrogen fixation, greater competitive capacity, dispersal capacity and formation of large and persistent seed banks [11,36,40,76]. This is a particularly negative finding for highly degraded landscapes, such as the Brazilian Atlantic Forest, where fragmentation and intensive land use drive long-term conservation losses [3,31] and potentially cascading extinctions of endemic animal species [84].
(c) Carbon and biodiversity co-benefits
Our results again reveal positive relationships between carbon stock and diversity in non-invaded forests [2,20], reinforcing the potential for carbon-driven policies to provide free biodiversity protection under well-targeted REDD+ projects or ambitious FLR programmes [85]. However, in invaded forests, rapid carbon sequestration trades off with biodiversity recovery, breaking the potential of REDD+ or FLR to also deliver on global biodiversity targets. Maximizing co-benefits of carbon-focused funding depends not only on climate change [86], stand age and landscape characteristics [3], but also the risk of biological invasion, including planting non-invasive species in tree plantations and their active control and management in invaded areas.
Increasing carbon led to functional clustering independently of forest types, indicating a process of convergence of community trait composition. In the non-invaded forests, this possibly results from increasing abundance of species with similar functional characteristics, suggesting high resilience in the provision of ecosystem functions. By contrast, in invaded forests, this occurs in response to the hyper-abundance of Acacia spp. and low species turnover, undermining ecosystem resilience [23].
(d) Study caveats
Our study has four main caveats. First, while we focused on invasive alien acacias in the Atlantic Forest, other exotic trees sometimes have invasive potential, such as Artocarpus heterophyllus Lam., Hovenia dulcis Thunb. ([87–89]; but see [90]) and Piper aduncum, which is found next to roads in selectively logged Bornean forests [91], while Acacia is also invasive in southeast Asia, potentially threatening forest successional dynamics and high biodiversity values [92–94]. Whether these other invasive trees similarly trade-off with biodiversity goals is a key unknown, although given their impacts on diversity, structure and species composition [87,89], this seems likely. Second, we only focused upon the impacts of invasive acacias on tree diversity, thus overlooking potential impacts on faunal recovery in Acacia-dominated regeneration. Studies suggest that Acacia plantations have low faunal value [93,94], pointing towards low biodiversity value, whether in plantations or regenerating habitat. Third, while invasive acacias rapidly stock carbon, which is important given the urgency of the climate crisis, long-term carbon stocks are likely to be similar or lower than non-invaded forests where many late-successional trees will have larger stature and higher wood density than early successional species of invasive acacias [95]. Thus, from a policy perspective, there is likely a trade-off between early carbon wins against longer-term carbon losses. Fourth, our study focused on highly fragmented landscapes with high edge effects [30–32], which have the potential to affect the recovery of biodiversity and carbon stocks [3]. Given that non-invaded second-growth forests adjacent to old-growth forests present a higher recovery rate of biodiversity and carbon stocks [2], in this sense, future studies should be conducted in landscapes with different degrees of fragmentation and edge effects, and thus vulnerability to the biological invasion process.
5. Conclusion and recommendations
Given rapid conversion of rainforest and associated release of carbon dioxide, second-growth forest has been promoted as a possible mitigation strategy. Our results reinforce that non-invaded secondary forests provide a low-cost mechanism that produces positive carbon-biodiversity co-benefits, whereas in invaded forests, at least in the short-term, carbon recovery and biodiversity conservation goals do not align. Thus, in human-modified tropical landscapes, including the globally threatened Brazilian Atlantic Forest, land abandonment for regeneration may be insufficient for restoring a highly diverse and carbon-rich forest.
Our study suggests that incorporating invasion potential in large-scale restoration planning, including FLR under the Bonn Challenge, is critical for restoration practioners seeking to balance competing societal and environmental needs [7]. When firewood or timber are the primary restoration goals, focusing on regions dominated by Acacia or other rapidly growing invasive species could speed their provision, while saving the need for costly investment in seedlings, planting and maintenance. However, where longer-term carbon stocking, biodiversity recovery or hydrological benefits are restoration goals, planning should seek to either avoid invasion-dominated regions or to incorporate management plans to reduce the dominance of invasive species. In young forests, assisted regeneration by controlling colonizing individuals of invasive trees and planting native trees to occupy areas more efficiently is important. In more mature forests, invasive trees can be harvested to release the forest from an arrested succession, maximizing long-term biodiversity recovery and carbon sequestration. Indeed, use of Acacia spp. as firewood or timber could make control cost effective while increasing energy security or meeting wood needs of local communities [96]. Importantly, invasive exotic tree species should not be planted in forestry and large-scale restoration programmes [7], given that their invasion can result in severe damage to biodiversity and ecosystem services.
Data accessibility
The data are provided in electronic supplementary material, along with additional methods, tables, figures and references supporting this article [97].
Authors' contributions
F.A.R.M. conceived the idea; F.A.R.M., D.P.E., L.F.S.M., G.H., A.V.N. and J.A.A.M.-N designed the methodology; F.A.R.M., L.F.S.M., G.H., N.V.H.S., M.P.S., and M.S. collected field data; F.A.R.M. performed the statistical analysis; F.A.R.M. conducted the literature review and led the writing, with D.P.E. and J.A.A.M.-N. co-leading the writing; L.F.S.M., G.H., A.V.N., T.B., R.D.Z., L.F.T.M., F.Z.S., C.E.G.R.S., N.V.H.S., M.P.S., M.S., S.V.M. and P.H.S.B. helped to structure and review the manuscript.
Conflict of interest declaration
We declare we have no competing interests.
Funding
This study is supported by FAPEMIG, CAPES (grant no. 6537-14-6), CNPq (grant nos. 316211/2020-6, 306335/2020-4, 308575/2019-9 and 316010/2020-0), FAPESP (grant no. 2018/18416-2), European Union's Seventh Framework (grant no. 269206) and FAPES (grant no. 41657.648.37258.06022020). G.H. is supported by Coordenação de Aperfeiçoamento de Pessoal de Nível Superior – Brasil (CAPES).
Acknowledgements
We thank Reserva Natural Vale, Reserva Biológica de Sooretama, Reserva Biológica Córrego do Veado, Flona do Rio Preto, Reserva Biológica do Córrego Grande, and IBAMA for the work permit granted in federal conservation units (license numbers 42532 and 78393-2 to F.A.R.M.) and Suzano Papel, Cellulose and Fibria for permission to access their forest fragments. We would also like to thank two anonymous referees for comments that greatly improved the manuscript and Jessica Erland for an admirable final review. A.V.N. would like to thank Capes-Print for the grant – Finance code 001.
