Decomposition and stabilization of the organic matter in an old-growth tropical riparian forest: effects of soil properties and vegetation structure

Background: Nutrient cycling in tropical forests has large importance for primary productivity, and decomposition of litterfall is a major process inuencing nutrient balance in forest soils. Although large-scale factors strongly inuence decomposition patterns, small-scale factors can have major inuences, especially in old-growth forests that have high structural complexity and strong plant-soil correlations. We evaluated decomposition rates and stabilization of soil organic matter using the Tea Bag Index in an old-growth riparian forest in southeastern Brazil to evaluate the effects of forest structure and soil properties on decomposition processes. These data sets were described separately using Principal Components Analysis (PCA). The main axes for each analysis, together with soil physical properties (clay content and soil moisture), were used to construct different structural equations models that evaluated the different parameters of the TBI, decomposition rates and stabilization factor. The best model was selected using Akaike’s criterion. Results: Forest structure and soil physical and chemical properties presented large variation among plots within the studied forest. Clay content was strongly correlated with soil moisture and the rst PCA axis of soil chemical properties, and model selection indicated that clay content was a better predictor than this axis. Decomposition rates presented a large variation among tea bags (0.009 and 0.098 g g -1 day -1 ) and were positively related with forest structure, as characterized by higher basal area, larger trees, and tree density. The stabilization factor varied between 0.211 – 0.426 and was related to forest stratication and soil clay content. Conclusions: The old-growth forest studied presented high heterogeneity in both forest structure and soil properties at small spatial scales, that inuenced decomposition processes and probably contributed to small-scale variation in nutrient cycling. Decomposition rates were only inuenced by forest structure, whereas the stabilization factor was inuenced by both forest structure and soil properties. Heterogeneity in ecological contribute to the resilience of old-growth forests, highlighting the importance of restoration focused on the spatial of

matter can strongly in uence local and global biogeochemical cycles (Benbow et al. 2019;Sayer et al. 2020) .
The energy source used in the microbial activity during decomposition consists of organic carbon, so that part of the nutrients is released and made available for plant use, and part is immobilized during microbial growth (Berg and McClaugherty 2014;Wachendorf et al. 2020). However, not all litter residue is necessarily mineralized, and can stabilize or decompose at very slow rates, contributing for the soil organic matter (Berg 2018). The fraction effectively decomposed and the speed of the decomposition process generally increase with temperature and humidity, stimulating microbial activity (Bradford et al. 2016). On the other hand, litter quality is strongly related to the amount of lignin and other recalcitrant compounds (Duddigan et al. 2020). High-quality litter is more easily decomposed due to the lower C:N ratio, whereas the decomposition of low-quality litter, with higher amounts of lignin, is slower and may need specialized organisms to be decomposed (Berg 2014), or higher availability of soil nutrients that can be transferred to leaf litter (Bonanomi et al. 2017).
The stabilization of soil organic matter (SOM) is a complex process, and can be in uenced by the available nutrients, soil texture and aggregation, and biological activity (Lajtha et al. 2018;Wiesmeier et al. 2019). The recalcitrant fraction of leaf litter that is not decomposed can contribute to SOM stocks (Berg and McClaugherty 2014). On the other hand, part of the labile fraction can be incorporated in the microbial biomass or their byproducts, and can be stabilized in SOM (Cotrufo et al. 2013). The stabilization of SOM is also favored by higher clay surfaces, oxides of iron and aluminum, and by the formation of aggregates (Wiesmeier et al. 2019).
Several factors in uence litter decomposition processes, especially local climatic conditions, litter quality, and the composition of microbial communities (Aerts 1997;Bradford et al. 2016Bradford et al. , 2017Djukic et al. 2018), which are strongly in uenced by the vegetation structure and soil physical and chemical properties. Soil fertility present large variation among ecosystems, and even at small spatial scales, since fertility is strongly related with soil texture and the structure and composition of plant communities (Spielvogel et al. 2016;Metzger et al. 2017). The structure of the vegetation directly in uences the microclimate conditions, hydrological movements, and soil properties (Krishna and Mohan 2017;Bélanger et al. 2019). For example, higher forest strati cation can provide higher radiation input, increasing temperatures and reducing soil moisture (Yeong et al. 2016), which are positively related with decomposition rates (Petraglia et al. 2019). Further, forests with more developed canopies and higher tree density produce more leaf litter, and this greater input can promote local microbial activity (Nunes and Pinto 2007;Silva-Sánchez et al. 2019). This process can be more complex in unmanaged old-growth forests due to the higher structural complexity on the soil surface, litter spatial distribution, stronger relationships between plants and soil, and responses of soil properties to this variation (Austin et al. 2014;Lajtha et al. 2018;Soares et al. 2020).
The evaluation of environmental factors that in uence the decomposition and stabilization of the organic matter can be carried out by using leaf litter with different qualities, since different factors can in uence the decomposition of the labile and recalcitrant fractions of the organic matter (Manzoni et al. 2012).
With these considerations, Keuskamp et al. (2013) proposed the Tea Bag Index method (TBI), which uses standard leaf litter with higher (green tea) and lower (rooibos tea) quality to obtain estimates of decomposition and stabilization of the organic matter, using an asymptotic model of mass loss (Wider and Lang 1982). The usage of standard material enables the evaluation of environmental factors on the decomposition process, independently of leaf litter quality (Keuskamp et al. 2013). Also, the TBI material (green and rooibos tea) is representative of the leaf litter found in natural ecosystems (Duddigan et al. 2020).
Riparian forests have large importance for the environmental quality of watersheds, contributing for important ecosystem functions such as the maintenance of biodiversity and water quality, cycling of nutrients and carbon, and contributing to regional climate (Naiman et al. 2005). These forests may present large spatial variation, related to the distance from watercourses, topography, geomorphology, and internal processes (Rot et al. 2000;Naiman et al. 2005). In response to this heterogeneity, there may be large spatial variation in soil properties and composition of plant communities, which then contribute with spatial variation in processes such as nutrient and carbon cycling (Woodward et al. 2015). Due to their ecological importance and strong deforesting pressure, the restoration of riparian forests has been stimulated worldwide (Hjältén et al. 2016;Dybala et al. 2019). However, the recovery of canopy cover may be not enough for ecosystem recovery, and establishing references is necessary to evaluate the recovery of ecosystem functions (Boudell et al. 2015;Matzek et al. 2016).
Therefore, an understanding of the functioning of preserved ecosystems is necessary to select adequate indicators for the monitoring of restored ecosystems, such as the processes of organic matter decomposition (Pollock et al. 2012;Dey and Schweitzer 2014). Some studies found that decomposition processes can be very different in old-growth and degraded forests (Borders et al. 2006;Yeong et al. 2016). However, comparative studies may not allow the evaluation of the factors that have most local in uence on decomposition rates, and more detailed studies on old-growth riparian forests are necessary to understand the decomposition processes in these preserved systems (Bradford et al. 2016). For example, Saint-Laurent and Arsenault-Boucher (2020) did not nd effects of environmental variables or soil properties on the decomposition and stabilization of the organic matter in Canadian riparian forests. On the other hand, Oliveira et al. (2019) found both soil and plant functional diversity effects on leaf litter decomposition of a tropical Atlantic Forest in Brazil, suggesting that factors with effects at small spatial scales can be important. Further, the structure of old-growth Atlantic forests can be very complex, contributing for higher microclimate heterogeneity (Ottermanns et al. 2011), as well as in uencing patterns of litterfall, nutrient cycling and carbon xation (Teixeira et al. 2020).
In this study, we evaluated the decomposition rates and stabilization of the organic matter in a tropical old-growth riparian forest in southeastern Brazil. We evaluated whether forest structure and soil properties could in uence these processes using the standard TBI methods (Keuskamp et al. 2013). In this way, we aimed to identify which factors at small spatial scales in uence the decomposition processes in old-growth forests, contributing for the identi cation of indicators for the monitoring of restored forests.

Study area
This study was carried out in a high preserved remnant riparian forest that belongs to the Air Force Base of Pirassununga (FAYS) in central São Paulo state (21°59'39.98" S, 47°20'12.73" W), Southeastern Brazil. FAYS includes a total forest area of 2,608 ha, where about 45% is composed by semideciduous seasonal forest and transition with riparian forest. The studied forest is adjacent to the Mogi-Guaçu River, in the upper Paraná River watershed, and is part of a 140 ha forest fragment at 620 meters above sea level (masl). Geologically, the region is in the residual plateaus of Franca/Batatais and is composed by intrusive Serra Geral, with deep distroferric red latosols, very deep, and a wavy relief (Rossi 2017).
Climate is Cwa following Köppen's classi cation, with wet summer and dry winter (Rolim et al. 2007). Between 1976 and 2008, the mean minimum and maximum temperatures recorded at FAYS were 10.6 and 32.8 ºC, respectively, and mean annual rainfall was 1,290 mm (Ferrari et al. 2012). The experiment was carried out in the early dry season (24 April to 24 June 2019). Mean temperatures and monthly rainfall recorded in May were 23.7 ºC and 39.4 mm, and in June 21.2 ºC and 16.2 mm, respectively, in the meteorological station of University of São Paulo, campus Pirassununga, located about 15 km from FAYS.

Experimental design
We evaluated the effects of spatial variation in soil attributes and forest structure on the decomposition and stabilization of the organic matter by establishing ten 10 × 10 m plots in the riparian forest, ve plots distant 5 m from the Mogi-Guaçu River (R plots) and ve plots at a 30 m distance from the river (I plots). Within each distance, each plot was 30 m distant from each other.
Soil sampling and description of forest structure were carried out just before the experiment began. Soil samples from each plot were obtained by randomly collecting three 0-20 cm depth subsamples with an auger, which were then composed to form a single sample. In the laboratory, a subsample was obtained from each compound sample and put up in containers. We determined soil moisture by initially determining the subsample wet mass and after drying at 65 ºC in an oven until dry mass stabilized. Soil moisture was then obtained as h = (wet mass -dry mass)/dry mass × 100. Soil chemical analyses were carried out following Embrapa (1997) and Raij et al. (2001): available phosphorus (P), potassium (K), calcium (Ca), and magnesium (Mg) were determined with the anion exchange method; soil organic matter was determined with the Walkley-Black method, pH was determined with CaCl 2 solution at 0.01 mol L -1 ; nitrogen (N) was determined with the Kjeldahl method, and potential acidity (H + Al) was determined with a buffered solution of calcium acetate at pH = 7. Cation exchange capacity (CEC) was obtained from the sum K + Ca + Mg + (H + Al), and soil base saturation (V%) was calculated by dividing the sum of bases (K + Ca + Mg) by CEC.
Forest structure characterization was adapted from Souza et al. (2013) by measuring all trees in each plot with circumference at breast height (CBH) > 10 cm, which was determined with a measuring tape. For each plot we determined mean diameter at breast height (DBH), tree strati cation (calculated as the coe cient of variation of DBH), tree density (individuals ha -1 ), and plot basal area (m² ha -1 ).
We estimated the decomposition rates and stabilization of the organic matter using the TBI method (Keuskamp et al. 2013). In this method, a pair of tea bags is buried at 8 cm depth, and each tea bag differs in litter quality: green tea (sencha tea, C:N ratio = 12.2) and red tea (rooibos tea, C:N = 42.9), as manufactured by Lipton ® (Keuskamp et al. 2013). The tea bags are made of polypropylene, have a tetrahedral shape with 5 cm sides, and contain about 2.0 g of tea. Within each pair, each tea bag type is buried 15 cm from each other. Five pairs were randomly buried in each plot at a minimum distance of 2.0 m from the plot sides, for a total of 100 tea bags buried in 24 April 2019. One day before the experiment was set up, 55 bags from each tea type were weighed in the lab with a spring (precision = 0.0001 g). The ve extra bags from each tea type were carried to the eld but were not buried and were used as manipulation controls. After returning from the eld, they were weighed, dried in an oven at 60 ºC for 72 h, and weighed again. We combined both measures to obtain a single correction factor to account for mass losses during transportation and humidity losses: 0.9433 for the green tea, and 0.9318 for the red tea. All initial mass values from the 100 tea bags were multiplied by the corresponding correction factor to obtain initial dry mass values for each bag.
After two months (24 June 2019), the tea bags were recovered and taken to the lab, where they were dried at 60 ºC in an oven for 72 h. Each bag was then carefully cleaned from soil particles and plant roots using a brush and weighed. The nal tea dry mass was obtained by subtracting 0.2424 g for the red tea and 0.2449 for the green tea, which was the mass related to the bag, threads, and label. We left the tea bags for only two months instead of three months as suggested by Keuskamp et al. (2013) because in another experiment carried out in 2018 in the same region, we lost many tea bags due to excessive decomposition (Soares et al. 2020). In these cases, Keuskamp et al. (2013) recommend a reduced incubation time because decomposition rates can be underestimated. The method assumes that the red tea is in the rst phase of the decomposition, so that if the red tea decomposes excessively, entering the second phase of decomposition, it is not possible to calculate k (http://www.teatime4science.org/faq/). Even though the tea bags stayed for only 60 days, we still lost seven tea pairs due to the high decomposition of the red tea.

Data analysis
We used a Principal Components Analysis (PCA) to reduce the dimensionality of soil chemical variables. Concentrations of K and P were ln-transformed to obtain normal distributions, which were veri ed using Shapiro-Wilk tests. All variables were then standardized for zero means and unity variance. We also used PCA to evaluate forest structure variables; forest strati cation (coe cient of variation of DBH) was transformed using the Box-Cox method to obtain a normal distribution. For both PCA analyses we used the Kaiser criterion and selected axes with eigenvalues > 1 (Legendre and Legendre 2012).
The PCA soil and vegetation axes were then used in a Structural Equations Model (SEM) as exogenous variables to evaluate their direct effects on decomposition rates (k) and stabilization factors (S) (Grace 2006). Considering that soil and vegetation covary, we also included correlations between soil and vegetation variables. Model t was evaluated considering differences between the observed and predicted covariance structure, the Comparative Fit Index (CFI), and the Root Mean Square Error of Approximation (RMSEA) following (Hooper et al. 2008). These analyses were carried out using the lavaan package (Rosseel 2012) in R (R Core Team 2018).
We found a strong effect of PCA Soil axis 1 on the stabilization factor S (see Results), and previous studies also found an effect of soil base saturation in natural remnants of riparian forest (Soares et al. 2020). Therefore, we evaluated the following variables as predictors of mean S per plot: soil organic matter, soil base saturation, clay content, soil moisture, and PCA Soil axis 1. Different linear regression models were t for each predictor variable, and we selected the model with the lowest Akaike Information Criterion calculated for small samples (Burnham and Anderson 2002). Analyses were carried out using Systat 13 software, except SEM analyses as noted above, with signi cance levels considering a = 0.05.

Results
The two PCA axes related to soil chemical attributes explained 85.4% of the variance. The rst axis (Soil 1) explained 60.2% of the variation (eigenvalue = 5.42) and was positively related with CEC, OM, V%, pH, N, and P (Fig. 1). The second axis (Soil 2) explained 25.2% of the variation (eigenvalue = 2.27) and was positively related with K and N:P ratios, and negatively related with CN ratios (Fig. 1). The ordination indicates a soil fertility gradient, with plots nearer the river less fertile than those in the forest interior, but with intermediate fertility plots in both distances (Fig. 1). The second PCA axis contributed to variation at the same distance from the river, with no apparent pattern (Fig. 1).

Fig. 1. Ordination of plots near the river (squares) and in the forest interior (circles) by Principal
Components Analysis in relation to soil chemistry: cation exchange capacity (CEC), soil base saturation (V), soil organic matter (OM), nitrogen (N), phosphorus (P), potassium (K), N:P and C:N ratios. Soil fertility as indicated by the rst PCA axis (Soil 1) was strongly correlated with clay content and soil moisture, as well as with soil base saturation and soil organic matter (Table 1). In fact, all these variables were correlated with each other (Table 1). Table 1.
Pearson's correlation coe cients between soil predictor variables. ** P < 0.01, *** P < 0.001. The PCA on forest structure variables explained 82.7% of the variation in the rst two axes. The rst axis (Veg 1) explained 51.2% of the variance (eigenvalue = 2.07) and was positively correlated with mean DBH, tree density, and basal area (Fig. 2). The second axis (Veg 2) explained 30.8% of the variation (eigenvalue = 1.23) and was positively correlated with forest strati cation and, to a lesser degree, negatively correlated with mean DBH and tree density (Fig. 2). Both axes contributed to separate plots at different distances from the river, so that plots nearer the river presented higher forest strati cation and basal area than those in the forest interior (Fig. 2). Tree density and mean tree sizes contributed to variation in forest structure among plots located at the same distance from the river (Fig. 2). The structural equations model showed a good t to the covariance matrix (c² = 0.021, df = 2, P = 0.990; CFI = 1.000; RMSEA = 0.000, P = 0.991). Veg 1 (which was positively related with basal area and tree sizes) was weakly correlated with Soil 1 (P = 0.057), whereas Veg 2 (which was more related with forest strati cation) was negatively correlated with Soil 1, indicating that higher forest strati cation was found nearer the watercourse, in less fertile plots (Fig. 3). Decomposition rates estimated from each pair of tea bags varied tenfold, between 0.009 and 0.098 g g -1 day -1 (CV = 0.60). Decomposition rates were related with Veg 1, indicating higher rates in plots with higher basal areas and larger trees (Fig. 3), although this relationship explained relatively few (19%) of the variation.
The stabilization factor S presented lower variation (0.211 -0.426, CV = 0.15) than decomposition rates. The model explained 53% of the variation in the stabilization factor. The stabilization of the organic matter was strongly in uenced by soil chemistry, with positive effects of soil fertility, as well as K concentrations and N:P ratios (Fig. 3). There was also a trend for a negative effect of Veg 2 (P = 0.054), suggesting lower stabilization of the organic matter in areas with higher forest strati cation (Fig. 3). Fig. 3. Structural equation model depicting the effects of forest structure (Veg 1, Veg 2) and soil chemical attributes (Soil 1, Soil 2) on decomposition rates (k) and stabilization factor (S) as calculated by the Tea Bag Index. Arrow widths are proportional to the standardized coe cients, which are also indicated next to the line. Dashed arrows indicate that P = 0.054 (Veg 2 ® S) and P = 0.057 (Veg 1 « Soil 1). ** P < 0.01, *** P < 0.001. Green arrows indicate positive effects, red arrows negative effects, and grey lines indicate nonsigni cant effects (P > 0.25).
Considering the mean values per plot, these ve soil variables that were proposed as predictors were positively related with the stabilization factor S (Table 2). However, the best predictor was soil clay content, with the lowest AICc value, explaining 79% of the variation in the stabilization factor; the probability that this model was the best explanation for the variation in S was almost three times higher than for the second best model (Table 2).

Discussion
Drivers of leaf litter decomposition at small spatial scales can in uence the decomposition and stabilization of the organic matter, sometimes with effects as large as those of drivers that in uence decomposition at large spatial scales (Bradford et al. 2016(Bradford et al. , 2017. In the present study, we found a large variation in leaf litter decomposition rates, spanning almost an order of magnitude even at a small spatial scale, which was related with the structure of an old-growth riparian forest. The stabilization of the organic matter was less variable and was strongly related with soil properties. Although forest structure and soil properties were correlated, the structural equations model allowed to separate the effects of each driver on the processes involved in the decomposition of the organic matter. Forest structure was weakly correlated with soil fertility as represented by the rst PCA axis (Soil 1). The correlation between Soil 1 and Veg 1 can be due to higher production of leaf litter in areas with higher basal area, larger trees, and higher tree density, as observed in other studies on semideciduous seasonal forests (Werneck et al. 2001;Nunes and Pinto 2007;Pinto et al. 2008). Higher leaf litter input can promote microbial activity and in uence SOM stocks (Lajtha et al. 2018;Silva-Sánchez et al. 2019), whereas litter mineralization release important nutrients for tree growth. On the other hand, soil fertility was negatively correlated with forest strati cation, which was related with distance from the river. Plots near the river presented higher forest strati cation, but also lower clay content and soil moisture. Clay content, soil moisture, and soil fertility were highly correlated with each other. Also, several nutrients are correlated with clay content, including major cations, CEC, N, and P (Schoenholtz et al. 2000;Woodward et al. 2015;Aprile and Lorandi 2019).
Decomposition rates of the organic matter were correlated with the rst axis of forest structure (Veg 1), which was represented by areas with higher tree density, mean DBH, and basal area. Forests with more developed canopies and higher tree densities produce higher amounts of leaf litter, which can positively in uence microbial activity (Nunes and Pinto 2007;Silva-Sánchez et al. 2019), increasing decomposition rates (Lajtha et al. 2018). Further, some studies suggest that soil microbial respiration is highly related with photosynthetic rates of the plant community, because with increased photosynthesis and primary production, the availability of root substrates increase and promote microbial activity, since the highest microbial biomass is found in the rhizosphere (Ryan and Law 2005;Tang et al. 2005).
Although signi cant, the correlation with Veg 1 explained few of the variation of decomposition rates (19%) in relation to pure error, since our data indicate a high variation in the individual estimates of k, varying almost an order of magnitude. High individual variation in k estimates was also found by Saint-Laurent and Arsenault-Boucher (2020) in their study on riparian forests, but decomposition rates were not related with soil properties or other environmental variables. Our data showed that decomposition rates were related with forest structure, suggesting that at this spatial scale the variation in vegetation is more important than other environmental factors such as differences in soil properties. The composition of oldgrowth forests may have more in uence on soil communities, with the development of more specialized microbial communities depending on the leaf litter traits of different plant species, contributing to the heterogeneity in decomposition rates at this spatial scale (e.g., Austin et al., 2014).
On the other hand, the SEM model explained 53% of the variation in the stabilization factor S, which was strongly in uenced by soil properties. The strong correlation with Soil 1 indicates that areas that presented higher S values were those with higher SOM, CEC, and soil base saturation, variables that were strongly correlated with each other and with clay content. Clay surfaces, as well as oxides of iron and aluminum, tend to stabilize the organic matter (Schmidt et al. 2011). Since the tea was not in direct contact with clay surfaces, clays may have not directly in uenced SOM formation, but may have had indirect effects by in uencing the microbial community. The decomposition of the labile fraction enables the incorporation of this substrate in microbial biomass and in byproducts, which may be a large part of stabilized soil organic matter (Cotrufo et al. 2013). Considering that the stabilization factor is directly related with the transformation of the labile fraction into recalcitrant fraction (Keuskamp et al. 2013), these results suggest that this mechanism can contribute strongly with the stabilization of the organic matter and carbon xation in the soil.
The variation in the stabilization factor S was not signi cantly in uenced by Veg 2 although a marginal value was found (P = 0.054), suggesting that forest strati cation can negatively in uence S. This is an expected result, since forest strati cation can provide higher radiation input, increasing soil temperature and reducing humidity (Yeong et al. 2016). Our results showed that soil moisture was positively related with the decomposition process, although the model considering clay contents had a higher probability of explaining variation in S. Therefore, both soil properties and microbial activity can contribute for the stabilization of SOM.
The drivers of decomposition in forests can be related with vegetation structure, which in uences microclimate conditions and local hydrological movements, and therefore forest soil properties and decomposition rates (Krishna and Mohan 2017;Bélanger et al. 2019). The contribution of soil properties for the stabilization of the organic matter can be lower in managed forests when compared to natural forests (Lukumbuzya et al. 1994;Berkelmann et al. 2018). In a study carried out in the same region, Soares et al. (2020) found a strong relationship between S and soil base saturation in a riparian forest remnant, but did not nd this relationship in a riparian forest under restoration. Cations such as Ca 2+ and Mg 2+ can reduce the organic matter mass loss (Lukumbuzya et al. 1994), as well as contribute for aggregate formation and consequent stabilization of SOM (Powers and Salute 2011). Our study suggests that the stabilization of SOM can be strongly in uenced by microbial composition, which should differ according to soil clay content, but future experiments are necessary to directly test this hypothesis.

Conclusions
The factors that contribute to the variation on decomposition processes within old-growth forests can differ from those in degraded forests or those under restoration (Borders et al., 2006;Yeong et al., 2016;Zhou et al., 2018), pointing to the importance of understanding these processes at small spatial scales (Bradford et al. 2016). This heterogeneity at small spatial scales can contribute to the resilience of oldgrowth forests (e.g., Feit et al., 2019), strengthening ecosystem functions such as nutrient cycling and carbon xation, and highlighting the importance of restoration strategies focused in the recovery of ecosystem processes. Studies comparing the stabilization factor in soils within preserved and degraded forests could test these hypotheses and contribute on restoration techniques that aim to increase carbon xation in the soil. Figure 1 Ordination of plots near the river (squares) and in the forest interior (circles) by Principal Components Analysis in relation to soil chemistry: cation exchange capacity (CEC), soil base saturation (V), soil organic matter (OM), nitrogen (N), phosphorus (P), potassium (K), N:P and C:N ratios.

Figure 2
Ordination of plots near the river (squares) and in the forest interior (circles) by Principal Components Analysis in relation to forest structure: basal area, tree density, mean DBH, and forest strati cation (CV DBH).

Figure 3
Structural equation model depicting the effects of forest structure (Veg 1, Veg 2) and soil chemical attributes (Soil 1, Soil 2) on decomposition rates (k) and stabilization factor (S) as calculated by the Tea Bag Index. Arrow widths are proportional to the standardized coe cients, which are also indicated next to the line. Dashed arrows indicate that P = 0.054 (Veg 2 S) and P = 0.057 (Veg 1 Soil 1). ** P < 0.01, *** P < 0.001. Green arrows indicate positive effects, red arrows negative effects, and grey lines indicate nonsigni cant effects (P > 0.25).