Dynamics of tree stems and biomass in old‐growth and secondary forests along gradients in liana dominance, elevation and soil

Lianas, or woody vines, are key components of many tropical forests and can have substantial impacts on the dynamics and functioning of these important ecosystems. Their competition with trees for resources, in particular light, can hamper the recovery of forests from disturbances. Yet, it is unclear how forest disturbance interacts with liana–tree ratio (LTR), topography and soil properties to shape tree dynamics and the trajectories of forest succession. Using temporal data from the Kilombero Valley and the Udzungwa Mountains of Tanzania, we demonstrate how the dynamics of tree stems and biomass vary between secondary and old‐growth forests with changes in the dominance of lianas and environmental gradients. Greater tree recruitment and mortality in secondary forests compared with old‐growth forests suggested rapid regeneration processes and faster turnover. However, no significant differences were found in the net annual changes in the number or biomass of trees between secondary and old‐growth forests. Our findings also showed that higher LTRs were positively associated with stem mortality but also with tree biomass growth, indicating a nuanced ecological role of lianas in forest ecosystems, which warrants further investigation to fully understand the causal factors at play. Net changes in tree stem numbers decreased significantly with elevation, implying climatic constraints on forest regeneration at higher elevations. Soil cation exchange capacity and organic carbon were found to significantly influence tree stem recruitment and net change in abundance, although their effects on biomass remained unclear. Synthesis: Our findings indicate that the recovery of tropical forests from disturbance in terms of the number and biomass of tree stems may be predictable along environmental gradients. These insights have the potential to broaden our capacity to develop more nuanced strategies that identify when and where tropical forests may require restoration interventions, with a focus on structural recovery.


| INTRODUC TI ON
Tropical forests play a crucial role in sequestering carbon, supporting biodiversity and maintaining ecological balance (Bonan, 2008;Pan et al., 2011).Their capacity to fulfil these functions is regulated at least partially by the dynamics of the forest stand that encompasses tree growth, recruitment, mortality and the net change in productivity, which govern both the structure and the ecosystem function of forests (Johnson et al., 2016;McDowell et al., 2018;Sheil, 1995).Both natural and anthropogenic disturbances have remarkable effects on tropical forest function and services (Makana et al., 2011;Matsuo et al., 2021;Poorter et al., 2016).Subsequently, secondary and old-growth forests exhibit variations in stand dynamics due to age, structure, species diversity and environmental conditions (Chazdon, 2003;Chazdon et al., 2016;Letcher & Chazdon, 2009).
In both old-growth and secondary tropical forests, interactions between climate, soil, trees and other plant growth forms, especially lianas, shape forest stand attributes and dynamics (Martin et al., 2004;Norden et al., 2009;Phillips et al., 2002).Forest disturbances create openings in the canopy, allowing more light to reach the forest floor and stimulating the growth of new trees (Schnitzer et al., 2000), while climate factors and the availability of nutrients and water are associated with topography (Körner, 2007;Schnitzer & Bongers, 2011).
Although numerous studies have focused on the abundance of lianas and trees, including in disturbed forests, the impact of lianas on forest succession, biomass recovery and carbon storage has less frequently been examined.Despite lianas being major components of tropical forests (Schnitzer, 2018), only a few studies have rigorously explored and quantified these patterns and their implications for forest dynamics (Finegan, 1996;Lai et al., 2017;Martin et al., 2013;Poorter et al., 2021;Rüger et al., 2020;van der Sande et al., 2023).
Lianas also cause mechanical damage and increase the risk of tree mortality (Ingwell et al., 2010), leading to an interruption of forest recovery processes (Estrada-Villegas et al., 2020;Lai et al., 2017;Schnitzer & Carson, 2010).Then again, lianas can benefit forest ecosystem function by promoting diversity and canopy connectivity, boosting soil fertility and nutrient cycling through rapid leaf litter turnover (Estrada-Villegas & Schnitzer, 2018;Roeder et al., 2022;Tang et al., 2012), maintaining microclimates (Meunier et al., 2021), and protecting recovering forests from further disturbances (Marshall et al., 2020).Interactions between lianas and trees can be complex and are context-dependent (Schnitzer, 2018), with the feedback of liana competitive success over trees on forest recovery varying based on forest disturbance and other environmental conditions, such as climate and topography (Ngute, Schoeman, et al., 2024).
Elevational gradients, which generally correlate with temperature, moisture, slope, soil fertility and the availability of water, nutrients and light (Körner, 2007), influence tree growth both directly and indirectly (Johnson et al., 2016) and can also be useful predictors of tree stem and biomass recovery (Norden et al., 2009).However, climate change, with altered temperature and precipitation patterns, can significantly impact the number and biomass of tree stems along elevation gradients (Bauman et al., 2022;Bennett et al., 2021).As temperatures increase, tree species that were once limited by cold temperatures could start to colonise higher elevations, potentially leading to an increase in stem number and biomass (Cuni-Sanchez et al., 2024).Additionally, increased temperatures can exacerbate drought conditions, particularly on sunny slopes, which could offset the gains of thermal relief by reducing moisture availability, further complicating predictions of biomass and stem number dynamics (Yin et al., 2023).
The effects of soil, climate and topography on tree recovery are further shaped by forest disturbance legacies (Chazdon, 2003).These impacts are more pronounced in younger (secondary) forests than in old-growth forests (Chazdon, 2008), since tree communities in secondary forests in less fertile soils exhibit slower growth rates (Chazdon et al., 2016).
In view of these complex interactions, our understanding of how forest disturbance, liana abundance and environmental gradients influence the dynamics of tree stems and biomass in tropical forests remains limited.Previous studies have largely focused on individual factors, offering a fragmented view (di Porcia e Brugnera et al., 2019;Laurans et al., 2014).Holistic insights into how these factors interact and shape the number of trees and their biomass over time are crucial for predicting how tropical forests will respond to environmental pressures such as land use change (Marshall et al., 2020) and for understanding whether these factors can have long-lasting effects on forest structure and function (Phillips et al., 2004).This study aimed to quantify temporal variations in tree stems and above-ground biomass (hereafter, 'biomass') across old-growth and secondary tropical forests, taking into account gradients in liana dominance, topography and soil properties.To fulfil this aim, our specific objectives are to: (1) measure and establish temporal variations in tree stem numbers and biomass between old-growth and secondary tropical forests, (2) assess the influence of liana dominance on the dynamics of tree stems and biomass and (3) examine how environmental gradients, especially topography and soil properties, shape patterns in tree stem numbers and their biomass.

| Study site
The study took advantage of a network of permanent sampling plots spread across 17 sites within forested landscapes of the Kilombero Valley and the Udzungwa block of the Eastern Arc Mountains of Tanzania (Figure 1).The area is a mosaic of mature and secondary forests, agricultural land and human settlements.The area has a network of protected areas, including the Udzungwa Mountains National Park and three Nature Reserves (Kilombero, Magombera, and Uzungwa Scarp).Recognised as a global conservation priority (Burgess et al., 2007), this region is located within biodiversity hotspots of the Eastern Afromontane and the Coastal Forests of Eastern Africa (Myers et al., 2000).Forest vegetation at the sites predominantly comprises semideciduous lowland Zanzibar-Inhambane forest, submontane and montane evergreen moist forests.These were classified into two disturbance categories: old-growth forests (i.e.remnants of mature and primary forests with some understorey or mid-strata tree cutting but with the canopy remaining largely intact) or secondary forests (i.e.heavily disturbed forest patches of various ages with more than 80% canopy loss, mostly consisting of stunted or damaged trees, with occasional isolated, tall trees).locations.These locations were stratified by elevation (270-2300 m; Figure 1).

| Study design
We used a stratified random approach to select plot locations within elevation ranges to represent the primary environmental gradients in the region (Marshall et al., 2012).Different locations were sampled across elevations, to capture as much regional variation as possible within the limits of logistical constraints, including time and access.To ensure comparability while minimising potential edge effects, the old-growth plots were strategically placed as close to the secondary plots as possible, ensuring a minimum distance of 100 m.The distances between the neighbouring secondary and oldgrowth forest plots ranged from a minimum of 110 m to a maximum of 980 m.Where feasible, the elevation difference between paired plots was also limited to <100 m in elevation.To minimise the risk of pseudoreplication, we ensured that no two plots of the same forest category (old-growth or secondary forest) within the same elevation band were less than 500 m apart.

| Stem and biomass data
Between July 2015 and July 2018, we measured and tagged all living lianas and trees with a diameter at breast height (DBH; measured 1.3 m from the ground) of ≥10 cm within the entire 0.4 ha area of each sample plot, following standard protocols (Marthews et al., 2014;Schnitzer et al., 2008).We also measured all liana and tree saplings (1 ≤ DBH < 10 cm) located within the central 4 m × 100 m (0.04 ha) cross-sectional strip of each plot (Figure S2).Following Marshall et al. (2012 and2017), we recorded all tree heights.All lianas and trees were remeasured between July 2021 and February 2022, mirroring the initial measurement methods, noting mortalities and adding any recruits.The time gap between the initial census and the subsequent recensus ranged from 3.6 to 6.2 years across the sample plots.
We identified trees at the species level whenever possible, following taxonomic guidelines from World Flora Online (<www.world flora online.org>).For trees that remained unidentified, we collected voucher specimens for further identification at the National Herbarium of Tanzania (NHT; Thiers, 2018).We also noted the local name (in the Hehe language) of all the unidentified lianas and trees and recorded identification features to distinguish them from others.Fewer than 21% of the tree individuals sampled in both censuses could not be identified to species level.
To estimate the density of the mass of the wood (WMD), we collected 1 to 3 samples of wood from each tree species common to each plot (cumulatively comprising up to 50% of the stems in a plot), from sites of similar forest type adjacent to each plot.We then measured the density of each wood sample using the water displacement method (Chave et al., 2006;Williamson & Wiemann, 2010).
Tree WMD values were assigned to each species (or nearest taxonomic unit) using records from the Global Wood Density Database (Zanne et al., 2009), averaged with locally measured data where applicable while adhering to the method of Lewis et al. (2009).For stems that did not have taxon-specific data represented in either database (~4%), we used plot-based WMD averages, following Lewis et al. (2013).
Tree measurements were converted into biomass (above-ground biomass) using a pantropical allometric model (Chave et al., 2005(Chave et al., , 2014)).This model, which incorporates tree diameter, height, and WMD, was selected due to its previous successful applications on sapling stems (van der Heijden et al., 2019) including in our study region (Marshall et al., 2017).Alternative models were disregarded as they were either too generic (Vieilledent et al., 2012) or did not account for tree height (Mugasha et al., 2016;Ngomanda et al., 2014), a notable drawback considering the proven variance in forest stature with topography and climate (Feldpausch et al., 2012;Marshall et al., 2012).
In instances where tree height data could not be obtained, this information was extrapolated from diameter values using plot-specific and stem size-specific height-diameter allometric regression models (Marshall et al., 2012).These calculations were performed using loglinear transformations in the case of old-growth forests, and loglog transformations for secondary forests, following Niklas (1995) and Chave et al. (2014).Although species-specific models offer more precise insights under conditions of ample data (Feldpausch et al., 2012), our study did not explore separate models for individual species or functional groups due to the limited sample sizes available, which would potentially undermine the statistical robustness of such models.
For liana biomass estimation, we used locally developed allometric equations, tailored for each forest category based on destructive sampling and stem diameter measurements of liana species commonly found in each forest type (Ngute, Pfeifer, et al., 2024).These multi-species equations, specifically designed for each forest category, did not require any taxonomic attributes or wood density data to effectively estimate biomass.This is because they could leverage the robust correlation between stem diameter and biomass across multiple species within our study region.
We used functions from the 'BiomasaFP' R package (Sullivan et al., 2023) to calculate and annualise stand-level stem mortality (the number of stems that died), stem recruitment (the number of new stems that reached the minimum diameter threshold), net change in stem numbers, biomass mortality (the biomass of stems that died), biomass growth (tree biomass gain as a result of growth), biomass recruitment (the biomass of new stems that reached the minimum diameter threshold) and net change in biomass over the census period, following Kohyama et al. (2019).
The Kohyama approach (Kohyama et al., 2019) includes corrections for unobserved metrics and accounts for census-interval bias.
These adjustments for unobserved metrics consider stems that were recruited (unobserved recruitment), grew (unobserved growth), and subsequently died (unobserved mortality) within the census interval.
Before statistical modelling, all these annualised stand-level metrics (stem mortality, stem recruitment, net change in stem numbers, biomass mortality, biomass growth, biomass recruitment and net change in biomass) were expressed as proportions to corresponding data from the initial census.LTRs were also calculated, using data from the initial census, for both stem numbers and biomass and were subsequently used in statistical models to quantify the dominance of lianas and their competitive success relative to trees (Marshall et al., 2020).These ratios were defined as the number of stems of lianas per tree stems and the total biomass of lianas relative to the total biomass of trees, calculated using the following formulae:

| Soil data
Soil samples (approximately 250 g each) were taken from 0 to 30 cm depth at three equidistant points located on the central line of each plot (Figure S2) using a 2-cm-diameter root corer, after sweeping aside leaf litter and loose topsoil.These samples were pooled for each plot, placed in plastic bags and dried for later analysis at the Analytical Soil and Plant Laboratory of the International Institute of Tropical Agriculture in Dar es Salaam (Tanzania).
Soil pH was determined in water using a calibrated potentiometer and pH meter, after mixing soil samples with deionised water in a 1:2.5 mass/volume soil-to-water ratio, following Henderson and Bui (2002).Soil particle size was determined using the hydrometer technique on soil samples that had been oven-dried and sifted following Gee and Bauder (1979).Specifically, the proportion of sand-sized particles (53-2000 μm) was measured by sieving, while the proportions of silt-sized (2-53 μm) and clay-sized (<2 μm) particles were evaluated by solution density after a settlement period of 6 h at 20°C.The ratio of silt to clay (SCR) was calculated for each sample and used as a measure of soil age and erodibility (Sullivan et al., 2022).Soils with higher SCR values are considered younger and more susceptible to erosion and nutrient leaching (Bruijnzeel, 2004).
Total nitrogen content in the soil was estimated using the Kjeldahl method (ISO5315) following Nelson and Sommers (1972).
Total organic carbon (SOC) proportions were estimated in each soil sample using Walkley-Black analytical procedures (Walkley & Black, 1934).Subsequently, the carbon-to-nitrogen ratio (CNR) was calculated from total organic carbon and nitrogen in the soil.CNR has been identified as a key indicator of organic matter decomposability and macronutrient availability in forest soils, with lower CNR values denoting a higher decay rate and greater immediate availability of nutrients for use by plants (Ostrowska & Porębska, 2015).
Subsequently, the phosphorus content was colourimetrically quantified from the extracts, following Murphy and Riley (1962).
Exchangeable bases (Ca, Mg, K and Na) were extracted and quantified using a neutral solution (pH = 7) of ammonium acetate at a concentration of 100 cmol L −1 , as per Schollenberger and Simon (1945).The total cation exchange capacity (CEC) was then calculated for each sample as the sum of the exchange capacities of all base cations (i.e.Ca 2+ , Mg 2+ , K + and Na + ).Soil CEC served as an indicator of soil fertility for each sample plot, as it has previously been shown to represent the available soil micronutrients for plant growth in the topsoil (Ali et al., 2019).

| Data analysis
All analyses were performed in the R statistical software, version 4.3-1 (R Core Team, 2023).All numeric covariates were standardised by centring and scaling them to a mean of zero and a unit standard deviation, using the built-in 'scale' R function (R Core Team, 2023).
We conducted a principal component analysis using the 'vegan' package (Oksanen et al., 2022) to explore the relationship between topographic measures (elevation, slope and aspect) and soil property metrics across sample plots (Figure S3).These data exploration informed the inclusion of predictors for subsequent modelling.These predictors included elevation, slope, aspect, pH and soil contents of silt, clay and sand, as well as SOC, nitrogen, phosphorus, CEC, CNR, SCR, forest categories and LTRs.
Before modelling, we applied natural logarithm transformation, where required, to ensure that all response metrics met assumptions of normality and homoscedasticity.We used the 'lme4' package (Bates et al., 2015) to fit linear mixed-effects models using Maximum Likelihood.For each response variable, two models were fitted: one with all data and another using a subset comprising sapling data only.
In all models, we included study site locations as a discrete random effect.This served to delineate pairs of sample plots and account for any ecological differences between sites while mitigating any potential bias arising from an unbalanced distribution of sample plots across the study sites.Forest categories-designated as 'oldgrowth forests' and 'secondary forests'-were incorporated into all models as a two-level factor to define types of forest disturbance.
Furthermore, to fully assess the interaction effects of forest disturbance on response metrics, we included interaction terms between forest categories and each of the environmental gradients-that is liana dominance, topography and soil.This approach allowed us to explore how these relationships vary under different forest disturbance conditions.
To minimise multicollinearity, we retained predictor variables in each model in a stepwise manner, following Peng and Lu (2012), and based on their standardised variance inflation factor (VIF) values computed using the 'car' package (Fox & Weisberg, 2019).Given our initial focus on the combined effects of forest disturbance, liana dominance (LTR), soil and topography, the selection process ensured that at least one key variable from each of these groups was retained in the final models to guarantee complete coverage of each gradient.Specifically, the variables retained included forest disturbance categories (old-growth and secondary forests), LTRs We also assessed the decay in both the marginal R-squared (R m 2 ) and conditional R-squared (R c 2 ), estimated based on the methods of Nakagawa et al. (2017).Predictors with the VIF >4 and the lowest contribution to R m 2 were identified and removed.This process was repeated until all remaining predictors had a VIF ≤4.Subsequently, we used the 'bootMer' function (Bates et al., 2015) with 1000 iterations to bootstrap the parameters of the best models refitted by Restricted Maximum Likelihood.
To quantify the effect sizes of the retained predictors on response metrics, we used multimodel inference and averaging (Zhang et al., 2016).This involved generating all potential predictor combinations from the best models fitted by Maximum Likelihood, then limiting the model subsets to a 95% confidence set, ensuring that the sum of the Akaike Information Criterion (AIC) weights was 0.95.This approach excluded improbable models and averaged predictor coefficients based on AIC weights from each model subset.
This inference and averaging were performed using the 'dredge' and 'model.avg'functions of the 'MuMIn' package (Barton, 2023).
This method allows predictors with limited support to shrink towards zero and considers the importance of each predictor across all model subsets.It also accounts for uncertainties and coefficient variability among the top models (Dormann et al., 2018).

F I G U R E 3
Annual stem mortality (a-c), stem recruitment (d-f) and net change in stem numbers (g-h) predicted in old-growth forests (green) and secondary forests (yellow) forests for covariates using best-fitted mixed-effect models.Black points and error bars denote means ± bootstrapped 95% confidence intervals (1000 iterations).Lines and ribbons show fitted slopes (±95% bootstrapped confidence intervals, 1000 iterations).

| Stem dynamics
Sapling stem mortality showed a statistically significant positive relationship (p = 0.05) with increasing LTRs (Figure 2a).Specifically, each 0.1 increment in LTR was associated with an increase in the annual mortality of saplings (Figure 3c) by 4.4 stems ha −1 year −1 (0.1-8.8).However, no significant change in stem mortality was associated with LTRs when testing only large individuals (R m 2 = 0.20; p = 0.79; Figure 2a).Furthermore, LTRs did not show significant effects on either the annual stem recruitment (p = 0.12-0.84; Figure 2b; Table S1) or the net change in stem numbers (p = 0.24-0.94; Figure 2c; Table S1).

| Stem dynamics
The net annual change in the number of sapling stems decreased significantly (p = 0.001) with elevation (Figure 2c).Specifically, each 100-m elevation increase was associated with a decrease in the net number of sapling stems by 36.3 stems ha −1 year −1 (18.1-63.5, Figure 3g).However, did not show a statistically significant relationship with elevation if only large individuals were tested (R m 2 = 0.11; p = 0.38; Figure 2c).Furthermore, neither annual stem mortality (p = 0.54-0.96; Figure 2a) nor recruitment (p = 0.11-0.98; Figure 2b) were significantly related to elevation (Table S1).

| Stem dynamics
The net change in the number of sapling stems (Figure 2c) decreased significantly with increasing soil CEC (p = 0.02).
increment in CEC (Figure 3h).Although CEC was retained in the best-fitted models, it did not show significant effects on the annual net changes in the numbers of large trees (p = 0.99; Figure 2c) or stem mortality of either size class (p = 0.48-0.91; Figure 2a; Table S1).
Sapling stem recruitment decreased significantly with increasing SOC (p = 0.01; Figure 2b).Each 1% increase in SOC was associated with a reduction in annual sapling stem recruitment (Figure 3f) of 13.5 stems ha −1 year −1 (3.7-23.3).However, SOC did not show significant effects on large tree stem recruitment (p = 0.50; Figure 2b) and was not retained in the best-fitted models for other stem metrics.

F I G U R E 5
Annual biomass mortality (a), growth (b-e) and recruitment (f-g) predicted in old-growth (green) and secondary (yellow) forests for covariates using best-fitted mixed-effect models.Black points and error bars denote means ± bootstrapped 95% confidence intervals (1000 iterations).Lines and ribbons show fitted slopes (±95% bootstrapped confidence intervals, 1000 iterations).
This study has unveiled the multifaceted and variable influences of forest disturbance, liana dominance, elevation and soil attributes on the dynamics of tree stems and biomass.Forest disturbance emerged as a critical factor, explaining most of the observed patterns (Figures S4-S7), with secondary forests exhibiting higher tree turnover rates compared with old-growth forests, yet showing similar changes in stem and biomass over time.While forest disturbance was a predominant driver of changes in tree stems and biomass, liana dominance also showed significant effects on tree dynamics but explained relatively less variation across other environmental gradients.This variation was particularly evident when considering the broader context of each forest category-secondary forests generally had higher ratios of lianas relative to trees (Figures S4a-S7a).
However, the influence of lianas was context-dependent, exhibiting positive or negative effects on tree dynamics, which aligns with the fluctuating LTRs observed across different environmental gradients (Figures S4-S7).Furthermore, elevation was found to negatively impact tree dynamics, while soil properties demonstrated mixed effects.

| Forest disturbance and tree dynamics
Secondary forests, compared with old-growth forests, exhibited higher stem mortality and recruitment rates, indicating faster turnover and heightened sensitivity to environmental changes.
The rapid stem turnover in secondary forests, which did not lead to a net increase in stem numbers, might be explained by the swift replacement of older stems by recruits (van Breugel et al., 2013).
This observation is a piece of evidence of the high dynamism of secondary forests in the early stages following disturbances (Broadbent et al., 2008;Enquist & Enquist, 2011;Letcher, 2010;Poorter et al., 2021).Despite this, there was no significant net change in stem numbers between secondary and old-growth forests.This overall stability in stem numbers indicates that losses in tree stems are offset by gains from recruitment, resulting in no net change in the structural composition of the forests (Chazdon, 2014;Marshall et al., 2020).
Regarding biomass dynamics, secondary forests showed higher annual biomass mortality and accumulation rates than old-growth forests, possibly reflecting intense resource competition (Brown & Lugo, 1990;Chazdon, 2008;Corlett, 1995;McDowell et al., 2018;Rüger et al., 2020).The substantial difference in growth rates between large trees and saplings in secondary and old-growth forests underscores the ecological significance of secondary forests for regeneration processes and carbon sequestration (Chazdon et al., 2016;Heinrich et al., 2023;Poorter et al., 2021).However, despite these differences, the net annual change in tree biomass was similar between secondary and old-growth forests, indicating that the intensified rates of biomass accumulation and loss counterbalance each other (Chazdon, 2014;van Breugel et al., 2007).This observation suggests that the rapid rates of both biomass accumulation and loss in secondary forests-characteristic of their dynamic nature-are effectively balanced, leading to stability in net biomass change over time (Martin et al., 2004).
The similarity in net biomass change observed between secondary and old-growth forests does not necessarily indicate high turnover alone but demonstrates the equilibrium between growth and mortality rates, which is crucial for maintaining the overall biomass balance in these ecosystems (Holdaway et al., 2017).Thus, while the high turnover within secondary forests contributes to their dynamic biomass profile, the comparable net change observed with old-growth forests underscores the resilience and sustainability of biomass under varying ecological dynamics (Norden et al., 2009;Poorter et al., 2016;van Breugel et al., 2013).

| Liana dominance
Our findings indicate a substantial increase in stem mortality with higher liana-tree stem ratios.These results support earlier observations that increases in liana abundance can inhibit tree recruitment and even lead to increased mortality, thus limiting forest regeneration (di Porcia e Brugnera et al., 2019;Estrada-Villegas et al., 2022;Marshall et al., 2017;Schnitzer & Bongers, 2002;Schnitzer & Carson, 2010).
In contrast, liana-tree biomass ratios were positively associated with the annual increment of tree biomass, challenging the generally negative perceptions of lianas (Schnitzer, 2018).This suggests that lianas, through their unique ecological roles, could potentially enhance the productivity and resilience of forest ecosystems (Campbell et al., 2015;Schnitzer et al., 2015;Schnitzer & Bongers, 2002), especially in conditions where light is not limiting (Schnitzer & Bongers, 2011;Schnitzer & Carson, 2001), for example in heavily disturbed (secondary) forests (Marshall et al., 2020).This observation is also contrary to previous research in the region showing that biomass accumulation among saplings was negligible until lianas were removed (Marshall et al., 2017).
It is worth noting that the observed overall positive relationship between liana-tree biomass ratios and the annual increment of sapling and large tree biomass appears to be driven by plots that are predominantly occupied by fast-growing, light-demanding pioneer species, such as Bridelia micrantha, Macaranga capensis, Psychotria spp., Tabernaemontana pachysiphon and Trema spp.
These species are characteristic of disturbed areas and are known for their rapid growth and capacity to quickly colonise such sites (Finegan, 1996;Martínez-Ramos et al., 2021).They play a crucial role in the early stages of forest regeneration, thriving on poor soils and providing shade and protection to saplings of other forest trees (Chazdon, 2014;Denslow, 1980).However, the growth rates of these pioneer species in secondary forests can be substantially influenced by environmental factors such as light intensity, and water and nutrient availability, which underlines the complexity of their relationship with lianas and warrants further investigation into how these dynamics affect forest recovery processes (Uriarte et al., 2004).

| Elevational variations in tree dynamics
Net changes in stem numbers were negatively associated with elevation.The decline in stem turnover at higher elevations suggests that environmental factors that vary with elevation may constrain tree recruitment, thus slowing the pace of forest regeneration (Körner, 2007;Norden et al., 2009).This reduced tree turnover at higher elevations might be attributed to cooler temperatures and a shorter growing season (Kitayama, 1992;Moser et al., 2007).

| Soil properties and tree dynamics
Our results showed that soil properties, especially CEC and SOC, were pivotal in forest dynamics.This underscores the need to consider soil micro-and macronutrients in assessments of forest regeneration and carbon sequestration potential (Jobbagy & Jackson, 2000;Paul et al., 2002).Soil properties can influence the availability of nutrients and water for plants, which, in turn, affects their growth, survival and reproduction (Guo & Gifford, 2002;Kattge et al., 2020).The effects of soil properties on forest dynamics vary depending on the context and the metric used (Chazdon et al., 2016;Clark et al., 1999;Rozendaal et al., 2019).
Higher CEC values, indicating richer nutrient availability, were associated with lower net stem changes.This might suggest that higher CEC values intensify intraspecies competition, leading to increased stem mortality (Chazdon et al., 2016;Clark et al., 1999).
Alternatively, an increase in CEC could promote a stable mature forest state with lower stem turnover, since soil can retain and supply nutrients to plants more effectively (Jobbagy & Jackson, 2000).
Interestingly, higher SOC values, indicating higher soil quality and carbon storage, were associated with lower stem recruitment.
This could suggest that higher proportions of SOC might lead to conditions less favourable for the establishment of new nitrogenfixing stems (Silver et al., 2000;van der Sande et al., 2023).Another possible explanation is that higher SOC values reflect older, more closed forests, which have lower recruitment rates due to reduced light penetration (Grubb, 1977;Guo & Gifford, 2002;Paul et al., 2002).Furthermore, we observed that SOC levels indeed increased with elevation (Figure S8).This elevation-related accumulation of SOC, potentially due to slower decomposition rates at higher altitudes (Körner, 2007), might also explain the observed patterns in stem recruitment.The cooler temperatures and increased moisture content typical of higher elevations can lead to a build-up of organic matter, as decomposition processes become less efficient (Jobbagy & Jackson, 2000).Thus, the variations in recruitment may not only reflect the direct effects of SOC on seedling dynamics but also broader ecological shifts associated with altitude (Körner, 2007).
The lack of significant associations between soil metrics and biomass dynamics might indicate that other factors, such as disturbance legacies and forest structure, have a stronger effect on tree biomass dynamics within our study area (Rozendaal et al., 2019;Waring et al., 2019).It might also indicate that soil properties are not very heterogeneous in our study area, limiting their explanatory power.

| CON CLUS IONS
We investigated how the dynamics of tree stems and biomass in tropical forests are related to disturbance, liana dominance, elevation and soil attributes.Our findings indicate that secondary forests have higher tree turnover rates than old-growth forests suggesting that these forests can preserve their carbon storage and ecological functions despite disturbances.Additionally, lianas were found to have context-dependent negative or positive impacts on tree dynamics, highlighting the need for further research on the mechanisms and outcomes of tree-liana interactions.Moreover, we found that elevation negatively influences tree dynamics, while soil properties had mixed effects, indicating the importance of abiotic factors for forest regeneration and productivity.Our study offers new insights into the role of environmental and biotic factors in forest regeneration, highlighting the complex interplay between lianas and trees, underpinned by the influence of elevation and soil properties.However, it does not establish causality, and further research is needed to identify the causal factors driving the patterns observed in the dynamics of tree stems and biomass, as well as the nuanced relationships between tree biomass uptake and co-occurring lianas.
To fully understand these interactions, further studies specifically focusing on host-liana relationships are essential.Such research could provide deeper insights into how lianas influence their host trees, encompassing aspects like nutrient competition, physical support dynamics, and light interception.It is also important to note that our study did not measure light availability, a crucial variable in both old-growth and secondary forests that strongly affects growth and recruitment.The absence of this data constitutes a limitation of our study.Future research should incorporate measurements of light our capacity to develop more nuanced strategies that identify when and where tropical forests may require restoration interventions, with a focus on structural recovery.K E Y W O R D S carbon, disturbance, landscape ecology, plant-climate interactions, plant-plant interactions, plant-soil interactions, restoration, topography, tropical forests, vines Although previous studies have assessed the recovery of forest trees from disturbances across different environmental gradients and the resilience of tropical forests in the face of global changes, to our knowledge, none have combined forest disturbance, topography, liana dominance-defined as the competitive success of lianas over trees, quantified through liana-tree ratios (LTRs)-as well as soil properties.However, Ngute, Schoeman, et al. (2024) have also recently underscored the importance of considering the influences of disturbance, climatic and topographic factors when examining forest responses to the competitive success of lianas over trees, as they unveiled a global increase in liana dominance catalysed by forest disturbance, climate and topography.
Elevations across study plots ranged from approximately 270 m to 2300 m above sea level.The topography and soil composition change along this elevational gradient, varying between flat terrains and floodplains with sandy loam soils, to steep slopes with sandyclay soil types.The sites experience a long dry season from May to October with negligible or no rainfall, and a shorter wet season exhibiting bimodal precipitation spanning from November until December and March until May, with a mean annual rainfall range of approximately 1200-2400 mm, while the average annual temperature ranges from roughly 18 to 30°C.

A
total of 42 permanent 100 m × 40 m (0.4 ha) plots were set in oldgrowth (13 plots) and secondary (29 plots) forests, spread across 17 F I G U R E 1 Map of sampling plot locations across protected landscapes and elevation gradients of the Kilombero Valley and Udzungwa Mountains in South-Central Tanzania (see Figure S1, for close-up views).Google Satellite Imagery ©2023 TerraMetrics.

F
I G U R E 2 Effect sizes for annual stem mortality (a), stem recruitment (b), and net change in stem number (c) across 13 old-growth and 29 secondary forest plots.Points indicate coefficients (±95% confidence intervals) calculated from best-fitted linear mixed-effects models using multimodel inference for large trees (filled symbols) and saplings (open symbols).Coefficients are standardised to show the change in response variable per unit of standard deviation change in each predictor.Red asterisks indicate significant effects (*p < 0.05, **p < 0.01, ***p < 0.001).The coefficient values are provided in Table Figure4d; TableS2).

F
I G U R E 4 Effect sizes for annual tree biomass mortality (a), biomass growth (b), biomass recruitment, and (c) net change in biomass (d) across 13 old-growth and 29 secondary forest plots.Points indicate coefficients (±95% confidence intervals) calculated from bestfitted linear mixed-effects models using multimodel inference for large trees (filled symbols) and saplings (open symbols).Coefficients are standardised to show the change in response variable per unit of standard deviation change in each predictor.Red asterisks indicate significant effects (*p < 0.05, ***p < 0.001).The coefficient values are provided in Table