Long-Term Impact of Forest Fragmentation on Tree Functional Diversity, Trait Composition and Aboveground Biomass.

Background: Community composition on fragmented forest remnants has been shown to diverge from their natural successional path, revealing an alternative state which has been known as retrogressive succession. Here we show the long-term monitoring of a small forest fragment concerning community structure, species richness, functional diversity and composition throughout 17 years to offer a means to generalize community performance under the impacts of forest fragmentation. Results: The adult tree community showed shifts on its functional composition: reduction in species’ maximum height and percentage of simple leaves, indicating an investment in resource-acquisitive and disturbance adapted traits. However, we also found a gradual increase in wood density throughout the years for the adult community. For the overall community, functional composition analysis indicated a gradual reduction in the percentage of simple leaves and a signicant increase in aboveground biomass. The decrease in Hmax and simple leaves, especially for the adult tree community, are mostly related do microenvironmental conditions caused by edge effects, as desiccation and tree canopy damage. Conclusion: Our results show that natural regeneration is being negatively affected, except for aboveground biomass. Although these ndings could have resulted from a transient dynamic, they constitute a warning to future conservation policies around the ecological integrity of small forest fragments.

Consequently, fragmentation promotes the reduction of forested areas, isolation of remnants and creation of edges (Kupfer et al. 2006;Matos et al. 2016), and threatens the long-term persistence of species (Turner and Corlett 1996;Pereira et al. 2010; Rands et al. 2010), as well as the goods and services provided by those ecosystems (Foley et al. 2007; Zambrano et al. 2020). Small and isolated fragments are exposed to a variety of edge effects (hotter and drier microclimate, increased light intensity, wind turbulence) and immigration of alien species (Murcia 1995 Community composition from fragmented forests has been shown to diverge from their natural successional path, revealing an alternative state which has been known as retrogressive succession Tabarelli et al. 2008aTabarelli et al. , 2012Magnago et al. 2014;Rocha-santos et al. 2016;Ewers et al. 2017). This altered successional trajectory is characterized by increased rates of mortality (Lindenmayer et al. 2012), species-poor communities with a proliferation of disturbance-adapted tree species (Laurance et  Habitat fragmentation can lead to major changes on species community composition, yet little is known about its long term impacts, especially on functional traits composition and aboveground biomass (Laurance et al. 2006b;Santos et al. 2008;Magnano et al. 2014). Old-growth forest fragments surrounded by human-modi ed landscapes offer an excellent opportunity to examine the long-term effects of habitat fragmentation on diversity, functional composition and ecosystem services potential of such small forest remnants. Thus, understanding how these effects in uence forest regeneration and successional processes over time is of crucial importance to ensure forest persistence in fragmented landscapes.
Here we investigated how species richness, functional diversity and composition and aboveground biomass change over 17 years in a small tropical forest remnant surrounded by human-modi ed landscapes. We hypothesized that the tree community will show altered succession trajectory signs such as i) high mortality rate (superior to the recruitment rate); ii) decrease in species diversity and aboveground biomass throughout time and iii) shifts in functional composition toward a prevalence of species with highly colonization ability and early-successional traits (abiotic dispersion, small seed size, lower stature and wood density and compound leaves) throughout time. Finally, we also tested for shifts in functional composition within the adult and juvenile trees, as different forest stratum face different environmental constraints.

Study area
The study was conducted in the southeast region of Minas Gerais, Brazil (21°13'11" S − 44°58'15" W), on a remnant of the endangered Brazilian Atlantic Rain Forest, classi ed as Semideciduous Seasonal Forests (IBGE 2012). The mean altitude is 884 meters. The regional climate is classi ed as Cwb (Mesothermic climate of Köppen), de ned by dry winters and mild summers. Mean annual rainfall and temperature are 1493 mm and 19.3ºC, respectively (Vilela and Ramalho 1979). Soil types in the region are primarily argisols and nitosols (Machado and Oliveira-Filho 2010). Fragment area is currently around 4.0 hectares (ha) and it is surrounded by a low biomass landscape matrix, composed mostly by coffee plantations, pastures and a small dawn drainage canal.

Demographic metrics
The changes in the tree community over time were determined for each plot by calculating the following mean annual abundance rates: Mortality (M = (1 -[(N0 -Nm) / N0] 1 / t) * 100) and recruitment (R = (1 -(1 -Nr / Nt) 1 / t)*100, whereas biomass dynamics were described by basal area loss (P = (1 -((AB0 -(ABm + ABd)) / AB0) 1 / t) *100) and gain (G = (1-(1 -(ABr + ABg) / ABt) 1 / t) * 100), where: t is the time elapsed between censuses; N0 and Nt the initial and nal tree counts; Nm and Nr are dead trees and recruit counts; AB0 and ABt are the initial and nal tree basal area; ABm is the basal area of dead trees; ABr is basal area of recruits; and ABd and ABg are the increment and decrement in the basal area from the surviving trees (Sheil et al. 2000;Sheil and May, 1996). Species maximum height (Hmax, m) is an indicator of the adult stature of species, potentially related to the species longevity and life-history strategy (King et al. 2006), and was calculated as the 95th-percentile height of all trees of the species. Species wood density (WD, g.cm-3) represents biomass per wood volume constructed and was obtained from the Global Wood Density Database ( ltered by Tropical South America, Zanne et al. 2009). For the species with WD not available, we used mean values for the WD of the genus or family. The leaf type (LT, categorical data) re ects the species heat balance. In other words, leaf size profoundly affects a variety of biological water and energy processes and has particularly important implications for understanding the adaption strategy of plants to environmental changes (Wang et al. 2019). All species were categorized: compound and simple leaves. Seed size (SS, categorical data), although usually related to the competitive vigour of the seedlings (Kitagima 2007), is also an important life-history trait for trees, correlated to a suite of morphological and physiological traits of pioneer species (small seeds) and shade-tolerant species (large seeds) (Poorter and Rose 2005; Osuri and Sankaran 2016). Qualitative data for species SS were obtained from herbarium specimens, and the species were classi ed as small seeds species (seed length < 1.5 cm) and large seeds species (seed length > 1.6 cm), following Tabarelli and Peres (2002)  and the parameters, DBH (cm), species wood density (WD, g cm − 3 ) and E, which is a measure of environmental stress (retrieved from Rejou-Mechain and Chave, 2014).

Statistical analysis
We used generalized linear mixed models (GLMM) to assess the differences in species richness, functional diversity and composition (SR, FRic, CWM Hmax, CWM WD, CWM LT, CWM SS and CWM DM) and aboveground biomass among the four forest inventories (2000, 2005, 2011 and 2017). Measures from the plot from the different years of monitoring were considered as a random factor, to account the lack of independence within the plots among the four inventories (Bates et al. 2014a).
When necessary, data were log10 or square root transformed prior to analysis to meet the assumptions of normality and homoscedasticity. Tukey post-hoc tests were used to assess the differences between the forest surveys (2000, 2005, 2011 and 2017).
To test for potential spatial nonindependence of plots, we assessed the degree of spatial autocorrelation in our linear mixed model residuals using the Moran's I test. A p-value < 0.05 would indicate that the model residuals show spatial autocorrelation, which was not found for any of our models (Kissling and Carl 2008) (Table S1).

Results
We found that the annual mortality in the tree community was higher than recruitment for all forest inventories, except for the last one, resulting in a net decrease in total tree abundance (Table 1). In 2000, 1275 individuals were recorded consisting of 151 species within 46 families (Table S2). In 2005, 1234 individuals were recorded, a reduction caused by the higher mortality rate (3.28% year − 1 ) in comparison with the recruitment rate (2.65% year − 1 ) for the survey interval. A similar pattern was found for 2011 when the lower recruitment rate (2.86% year − 1 ) and high mortality rate (4.00% year − 1 ) resulted in a reduction of tree abundance (1150). In 2017 there was a small increase in abundance (1153 individuals), a result from the slightly higher recruitment rate (2.48% year − 1 ) in comparison with the mortality rate (2.43% year − 1 ). Overall, from 2000 and 2017, we found a decrease in 9.56% in tree abundance ( Table 1).  (Table 1). Indeed, we found an increase in AGB for the last survey, resulting in a net gain in 18.61% in biomass (Table 2).

Page 7/18
The small reduction found in species richness was not signi cant ( Table 2, Table S3), as also suggested by the rarefaction curves as there is an overlap of the standard deviation of the four richness curves (Fig. 1). Species functional richness showed a small but not signi cant reduction (Table 2).  For the overall community, functional composition analysis indicated a gradual reduction in the percentage of simple leaves (Fig. 2, Table S4). The adult tree community showed a reduction in Hmax and the percentage of simple leaves, indicating an investment in resource-acquisitive and disturbance adapted traits. However, we also found a gradual increase in wood density through the years for the adult community (Fig. 2). The juvenile tree community did not show any shifts in functional trait composition.

Discussion
This study has provided relevant insights into the successional trajectory effects of fragmentation on the tree community of a small forest remnant surrounded by human-modi ed landscapes. Under 17 years of community monitoring, the dynamics' analysis revealed a continuous reduction in the overall tree abundance and shifts in trait composition favouring adult species with resource-acquisitive, pioneer lifestrategies (reduction in tree maximum height and in species with simple leaves and a slight increase in the percentage of small seeds). However, for the adult tree community (DBH > 10 cm), we found a signi cant increase in wood density. Indeed, the last monitoring interval (2011-2017) showed a strong increase in total basal area and a signi cant increase in aboveground biomass (reaching higher values than found for the rst survey). Based on these ndings, our hypothesis was partly accepted, as natural succession is being negatively affected, except for aboveground biomass.
Other studies in Neotropical small forests have shown tree communities dominated by a small set generalist species in forest fragments with one to 10 hectares (Laurance 2001; Laurance et al. 2006b).
After analyzing the long-term dynamics of a forest fragment with 4 ha, we found a decrease in Hmax and simple leaves for the adult tree community and a decrease in simple leaves for the overall tree community. During succession, pioneer species are expected to be replaced by shade-tolerant ones, with the community gradually accumulating species diversity and ecological functions (Guariguata and Ostertag 2001). This trajectory is changed when strong environmental lters lead to the selection of species that share the adaptation strategies required to colonize and survive in a changed postdisturbance habitat (Tabarelli et al. 2008).
In disturbed habitats, the establishment of late-successional species is compromised until the stressful conditions had reduced to an acceptable state (Lebrija-Trejos et al. 2010). Edge creation, e.g., promotes microclimatic and structural changes within the forest with severe consequences to tree communities (Murcia 1995), such as desiccation which act as a climate lter favouring mostly generalist pioneer species (Pierce et al. 2017). Leaf type re ects adaptations to reduced water availability and therefore a gradual increase in species with simple leaves is expected in the course of succession towards more mature and undisturbed forests (Wright et al. 2017;Gei et al. 2018). Consequently, we found a gradual increase of trees with compound leaves, a strategy to prevent the excessive water loss as plants with pinnate or bipinnate leaves can release individual lea ets (and not complete leaves) during severe water stress, increasing heat dissipation and regulating temperature (Wright et al. 2017;Gei et al. 2018).
When considering the adult tree community, we also found a decrease in species Proximity to edge and small fragment size favours a turnover with trees typical from late-successional stages being replaced by pioneer and especially the short-lived species (Nascimento  Availability of data and materials The datasets generated during and/or analysed during the current study are available in the ForestPlots repository, www. orestplots.net.

Competing interests
The authors declare that they have no competing interests.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. supplementary.doc