Attributes of stand-age-dependent forest determine technosol fertility of Atlantic forest re-growing on mining tailings in Mariana, Brazil

Pedro Manuel Villa · Sebastião Venâncio Martins · Aline Pilocelli · Gabriel Correa Kruschewsky · Andreia Aparecida Dias · Fabio Haruki Nabeta

Abstract Understanding how soil fertility changes due to environmental conditions and stand-age-dependent forest attributes is important for local-scale forest restoration.We evaluated the effects of stand-age-dependent forest attributes (plant community composition and litter stock) on soil and technosol fertility across two second-growth Atlantic forests (SGF) after the deposition of mining tailings in Mariana, southeastern Brazil.We hypothesized that technosol fertility in the SGF tailings is positively affected by plant community composition variability, stand age, and litter stock.We used total exchangeable bases and organic matter as fertility indicators for technosol and soil, and species composition and litter stock as stand-age-dependent forest attributes.Our results showed significant differences in the stand-age-dependent forest attributes and soil chemical properties between the two forest patches (SGF tailing and SGF non-tailing) evaluated.Thus, there was a marked gradient of litter storage and fertility between soil and technosol that can be important forest recovery indicators for the affected plant communities.Furthermore, according to the tested models, we corroborated the hypothesis that technosol fertility is positively affected by stand age, plant community composition variability, and litter stock, which may contribute considerably to forest recovery on tailings.Our results demonstrate that the fertility predictors analyzed to explain the forest recovery on tailings can also be considered as ecological indicators for assessing forest restoration in areas impacted by mining tailings in Mariana.

Keywords Ecological indicators · Forest recovery · Litter stock · Nutrient recovery · Technosol

Introduction

Tropical forest restoration is an important global objective (Chazdon et al.2017; Crouzeilles et al.2017) because they provide multiple ecosystem services, such as global carbon cycling and climate change mitigation (Lewis et al.2015; Chazdon et al.2017; Villa et al.2020a).Therefore, understanding the processes related to natural forest recovery is essential to improve management criteria for different restoration methods (Crouzeilles et al.2017; Martins 2018; Holl et al.2020).Passive restoration, which removes the source of disturbance to promote natural regeneration (Holl and Aide 2011; Villa et al.2019; Martins et al.2020) has been demonstrated as the most cost-effective method for forest recovery (Chazdon et al.2017; Crouzeilles et al.2017).

Second-growth forest (SGF) that re-grows after a disturbance still represents an important reservoir of biodiversity (Chazdon 2014; Rozendaal et al.2019).Most of the previous studies on secondary succession in tropical forests have compared changes in tree species richness and composition in old-growth forest (OGF) or reference forests (i.e., late successional stages) without disturbances (Guariguata and Ostertag 2001; Chazdon et al.2014; Martins 2018).Studies comparing SGF and OGF have suggested that areas that regrow after a disturbance may harbor higher variability in the plant species composition due to the maximized coexistence of common light-demanding pioneer species and shade-tolerant species that are gradually recruited (Guariguata and Ostertag 2001; Mwampamba and Schwartz 2011; Poorter et al.2019; Villa et al.2019, 2020b).The common fast-growing and light-demanding pioneer species that colonize immediately after disturbance have short life cycles that can be completed between 10 and 15 years (Chazdon et al.2014; Villa et al.2018, 2019).When the long-lived species become dominant in the forest canopy, shade-tolerant species are recruited in late-successional stages and old-growth forest (Chazdon et al.2014; Poorter et al.2019).These forest recovery processes are shaped by multiple environmental factors, which may largely determine the successional trajectories (Arroyo-Rodríguez et al.2015; Rozendaal et al.2019; Mukul et al.2020).For example, soil properties and initial conditions influence successional trajectories and shape plant community composition of regenerating tropical forests (Fernandes-Neto et al.2019; van Breugel et al.2019).Thus, soil properties determine soil resource availability and can filter species composition as forests mature and regulate the speed of forest succession (Fernandes-Neto et al.2019; Safar et al.2019; van Breugel et al.2019).

Soil fertility recovery of second-growth forests that regrow after a disturbance depends on multiple drivers at different spatiotemporal scales (Becknell and Powers 2014; Poorter et al.2017).At a regional scale, the climate (e.g., temperature and rainfall) can have positive effects on the recovery of different forest attributes (e.g., Ali et al.2019a, b).On a local scale, where the climate does not change spatially, stand-age-dependent forest attributes (i.e., tree community composition and litter stock) are expected to increase these positive effects on second-growth forest recovery (Schilling et al.2016; Villa et al.2018; Rozendaal et al.2019; Mukul et al.2020).Stand age is also an important driver of the dynamics of forest attributes that have effects on soil fertility (Chazdon 2014; Becknell and Powers 2014; Poorter et al.2019).Therefore, most studies assess the effects of soil proprieties on forest attributes, such as tree community diversity and structure, and aboveground biomass (Poorter et al.2017; Villa et al.2018, 2020b; Ali et al.2019a).However, the true nature of the causal relationship between plants and soil fertility is bidirectional (van der Putten et al.2013; Laughlin et al.2015).Despite this relationship, few studies analyze the reverse effect (forest attributes on soil fertility) in second-growth tropical forests after disturbance by mining, for example, how forest attributes dependent on stand age such as litter and species composition variability affect soil fertility.

The litter stock is one of the most important ecological indicators for evaluating and monitoring forest recovery during restoration (Schilling et al.2016; Silva et al.2018; Rocha-Martins et al.2018, 2020).Litter, the layer of dead plant material that accumulates on the soil surface of forest ecosystems (Krishna and Mohan 2017), has a key role in ecosystem functioning as a determinant of nutrient cycling and forest dynamics (Krishna and Mohan 2017; Veen et al.2019).Thus, litter is the main pathway for nutrient transfers from plants to the soil in a degraded forest and thus provides available resources for plant growth and forest recovery (Laughlin et al.2015; Veen et al.2019).The litter stock is also related to stand-age-dependent forest attributes (i.e., plant community composition and structure) during succession after anthropogenic disturbances (Schilling et al.2016; Rosenfield and Müller 2019) and thus influences soil fertility recovery in a degraded forest (Rocha-Martins et al.2020).Therefore, litter stock and plant species composition can simultaneously explain chemical soil properties and fertility variabilities in the tropical forest, such as organic matter and nutrient availability (Laughlin et al.2015; Oliveira et al.2019).For example, exchangeable bases have been used as an indicator of soil fertility that better represent the available soil nutrients for plant growth in tropical forests (Poorter et al.2017; Ali et al.2019b), and exchangeable bases and soil nutrients can both be used as soil fertility indicators due to their strong positive correlations in most conditions (Huston 2012).In this sense, most studies of tropical forest recovery have evaluated mainly the effects of soils on plant composition and the effects of soil properties on changes in tree community diversity and structure and in ecosystem functioning (Villa et al.2018; Rodrigues et al.2019).However, the inverse relation, the effect of plant species composition on soil properties in Atlantic second-growth forests re-growing on mining tailing is poorly understood.

The Brazilian Atlantic forest is one of the most speciesrich biomes on the planet, but less than 12% of the original forest remains (Scarano and Ceotto 2015).The forests are found mainly as second growth (i.e., forests regenerating following an anthropogenic disturbance) in small remnant fragments with different stand ages (Scarano and Ceotto 2015).In this context, forest restoration activities after the rupture of the Fundão tailings dam in Mariana (November 2015) have been a high priority (Martins et al.2018).About 34 million m3of tailings were released (80% of the total volume contained) and deposited on the secondgrowth patches and areas that had a long agricultural history, mainly on the river banks (Prado et al.2019).Thus, after the tailings deposition, technosols have been forming, constituted by a complex tailing (sand, silt, clay and organic remains), from the mining residues and their treatment (Rossiter 2007; Schaefer et al.2017; Campanharo et al.2020, 2021).Accordingly, it has been important to evaluate the forest recovery processes on these technosols using different ecological indicators.

In this study, we therefore evaluated how forest attributes (plant community composition and litter stock) dependent on stand age affect soil fertility across second-growth Atlantic forest that is re-growing after the deposition of mining tailings in Mariana, Minas Gerais state, southeastern Brazil.We used total exchangeable bases and organic matter as indicators of soil fertility and plant community composition as stand-age-dependent forest attributes and of forest recovery at a local scale.We hypothesized that soil fertility is positively affected by stand age, plant community composition variability, and litter stock.

Materials and methods

Study area

Two second-growth Atlantic seasonal semideciduous forests of different stand ages were selected: a 3-year-old stand of SGF re-growing on tailings (SGF-tailing) and a 30-year-old stand of native reference forest not affected by tailings (SGFnon-tailing).Both SGFs are located in the municipality of Mariana (20°15′11″S and 48°22′25″ W, Minas Gerais, Brazil (Fig.1) within the limits of Area I (Sect.13) where the Renova Foundation operates in the Rio Doce basin, a major drainage of the Southeastern Atlantic hydrogeographic region.This region provides key ecosystem services to part of the country′s most populous and industrialized region.The Rio Doce is also one of the main rivers that supply water and nutrients to the endangered Atlantic Forest (Neves et al.2016).

Fig.1 Location of study area in Brazil (a), Mina Gerais state (b), Mariana municipality (c), and in the second-growth forest (d).The second-growth forest study site influenced by tailings (SGF tailing) is outlined in red, and second-growth reference forest without tailings (SGF non-tailing) is outlined in green

The climate of the study area, which is 712 m above sea level, is a moderate humid tropical system, with a dry season from May to September and a wet season between December and March.The mean annual relative humidity is ca.80%, mean annual air temperature is 19 °C, and mean annual precipitation is 1340 mm.Precipitation is highest in December, January, and February (Martins et al.2020).

The total collapse of the Fundão tailings dam on 05 November 2015 deposited to depths of ca 50-100 cm on flat, homogeneous topography tailings in our study area along the river (Campanharo et al.2020, 2021).Thus, secondgrowth forest re-growing on tailings is located at the margin of the river bed.The second-growth reference forest without tailings is approximately 150 m away from the river bank (Fig.1).

Sampling and quantification of litter stock

For litter sampling, 30 permanent 2 m × 2 m plots were systematically allocated in each SGF to form two transects of 100 m with 15 plots per transect with 5 m between plots.The accumulated litter was quantified in June and July 2018, approximately 3 years after the dam burst.In a 0.5 m × 0.5 m (0.25 m2) subplot in the center of each plot, all the undecomposed organic material (leaves, branches, fruits, and flowers) was collected, packed in plastic bags, labeled, and transported to the Forest Restoration Laboratory (LARF) of the Federal University of Viçosa (UFV), where it was transferred to paper bags and oven-dried at 70 °C for 72 h.The dry mass of litter was then weighed using a precision analytical balance.The amount of accumulated litter in each subplot (g/0.25 m2) was then converted into megagrams per hectare (e.g., Silva et al.2018).

Samples for evaluating soil properties

Five samples each of top-technosol and topsoil (0-10 cm deep) were collected in each SGF and systematically distributed within the plot (2 m × 2 m) to obtain a single composite sample for chemical analysis.Regular protocols (EMBRAPA 2017) were followed at the soil laboratory of the Federal University of Viçosa to measure pH (H2O), available phosphorus (P), potassium (K+), calcium (Ca2+), magnesium (Mg2+), effective cation exchange capacity (TCE), exchangeable acidity (H + Al), organic matter (OM).We used total exchangeable bases (SB; the sum of base cations Ca2+, Mg2+, K+, and Na+, in cmolc/dm3) as an indicator of soil fertility (Huston 2012; Poorter et al.2017; Ali et al.2019b).

Data analyses

All analyses were run in R 3.6.0 (R Core Team, 2019).Graphs were drawn using the ggplot2 package (Hadley 2015), except for those for the two-way cluster analysis which was done using PC ORD software (Villa et al.2019).

Plant community composition variability and dissimilarity

To test the normal distribution of all variables, we used the Shapiro-Wilk test and the Q-Q graph, and the homogeneity of the variations was evaluated by the Bartlett test using the dplyr package (Crawley 2013).To compare litter stock (expressed in Mg ha-1) and soil properties (non-normally distributed) between second-growth forest patches (SGFtailing and SGF non-tailing), we used the Mann-Whitney-Wilcoxon test (Crawley 2013).

Variation in plant community composition across secondgrowth forest (SGF-tailing and SGF non-tailing) was compared using a non-metric multidimensional scaling (NMDS) analysis and the metaMDS function based on Jaccard dissimilarities (Clarke 1993; Oksanen et al.2018).After checking the stress generated by the NMDS, we corroborated the nonmetric fit based on stress using linear regression (Fig.S1 from Supplementary Material, SM).Then, a permutational multivariate analysis of variance (PERMANOVA, 9999 permutations) was used to test differences in species composition by using the adonis function.Furthermore, we used the MDSrotate function, which rotates an external environmental variable (litter stock) to be parallel to the first multidimensional scaling dimension (Oksanen et al.2018; Villa et al.2020b).Finally, we extracted the scores on frequency-weighted NMDS axis 1 as a proxy for plant community composition variability (Villa et al.2018, 2020b).All the functions for the NMDS are available in the vegan package for R (Oksanen et al.2018).In addition, a two-way cluster analysis (also known as biclustering based on Jaccard dissimilarity) was used to assess the level of similarities and differences in species composition between second-growth forests (Villa et al.2019).Using the result from the two-way cluster analysis, we compared the plant community composition according to the regeneration strategies (i.e., pioneers, early and mid successional species) between the two secondgrowth forests to infer possible effects on soil and technosoil fertility indicators.

Gradient analysis: soil properties and litter stock across second‑growth forests

Principal component analysis (PCA) was used in the correlation matrix to reduce the number of redundant soil properties after data were log-transformed (Villa et al.2018).Spearman′s correlations between soil properties and PCA axes were also calculated using the FactoMineR package (Husson et al.2017).After the correlation between soil properties and the PCA was evaluated, the variables that showed the highest correlation and represented the best soil fertility indicator were chosen (Schmitz et al.2020), for example, the total exchangeable bases and organic matter in the soil (Fig.S2 and S3, from SM).These two variables were chosen as indicators because they have the opposite patterns in the disturbance gradient between SGF-tailing and SGF non-tailing and because the correlation between them is low (Fig.S2 from SM).

Generalized linear mixed models

The main effects of stand-age-dependent forest attributes, such as NMDS axes 1 (plant community composition variability) and litter (and their intereactions with stand age), on the properties related to soil and technosol fertility (i.e., total exchangeable bases and organic matter) were tested using different generalized linear models of mixed effects (GLMMs, with random and fixed effects).The GLMMs with negative binomial error distribution were tested, then the distributions of residuals were visually checked, and the most suitable distribution and link function (i.e., normality was confirmed by the Q-Q graph) was evaluated (Fig.S4).We corroborated a high correlation between total exchangeable bases and organic matter with other soil fertility indicators, such as nutrients (Figs.S2 and S3).Thus, the predictors with fixed effects were grouped into three categories: (i) the properties of the soil and technosol fertility, i.e., total exchangeable bases and organic matter (continuous explanatory variables), (ii) litter stock (continuous explanatory variable), (iii) NMDS axis 1 (continuous explanatory variables) and stand age of second-growth forest (discrete explanatory variable with two levels).We assessed collinearity between selected predictor variables using Spearman correlation analysis; when two variables were strongly correlated (r≥ 0.7), they were included in separate models (Fig.S2).We tested alternative models with individual effects of predictors and different combinations of predictors with low correlation, and the patches and plots were considered as a random effect (1│patch: plot).All models were calculated using the R package lme4 (Bates et al.2019).

Finally, we have compared the most parsimonious model using the information-theoretical approach based on the Akaike information criterion (AIC), considering all models (GLMMs) tested with AIC ＜ 2.0 as equally plausible (Burnham and Anderson 2002; Burnham et al.2011).For this procedure, we applied a multi-model inference approach using the Dredge function in the MuMIn package (Barton 2017).We also used the estimates of the predictor coefficients to interpret parameter estimates on a comparable scale using the jtools package (Long 2020).

Results

Plant community composition variability and floristic dissimilarity

Plant community composition showed significant differences between second-growth forests (Permanova:F1, 53=5.78;p＜ 0.001) forming two groups on two NMDS axis.When the two second-growth forest patches were compared, the SGF non-tailing had higher species composition variability, while SGF-tailing was nested with lower variability (Fig.2).

Fig.2 Nonmetric multidimensional scaling based on species composition along secondgrowth forest influenced by tailings (SGF tailing) or secondgrowth reference forest without tailings (SGF non-tailing).Litter stock (Mg ha-1) variability along SGFs is indicated

The two-way cluster analyses broadly divided the plant species into two groups based on forest patches with different stand age, which can be clearly seen by the two main dendrogram branches in both the SGF-tailing and SGF non-tailing patches (Fig.3).Thus, there was a higher proportion of pioneer species in the SGF-tailing patch (～ 80% of the total community), while only 20% corresponded to early and mid successional species.In contrast, the SGF non-tailing had a lower proportion ofpioneer species (＜ 50%) and a higher proportion of early and mid successional species (～ 50%) (Table S1).

Fig.3 Distribution of 86 species within 60 sampled plots along second-growth forest influenced by tailings (SGF-tailing) or second-growth reference forest without tailings (SGF non-tailing) using twoway cluster dendrogram based on the Jaccard similarity metric.Species are listed in the supplementary material (Table S1 and S2)

Litter stock across second-growth forests

The average value of litter stock on the tailings in the regeneration forest (SGF tailing) was 20.2 Mg ha-1, which differed significantly (p＜ 0.001) from 10.7 Mg ha-1in the SG reference forest (SGF non-tailing) (Fig.4).

Fig.4 Litter stock in secondgrowth forest influenced by tailings (SGF tailing) and secondgrowth refence forest without tailings (SGF non-tailing).The red point and range indicates mean values

Soil and technosol properties

When the averages were compared between soil and technosol properties, only P and Mg did not differ signifciantly between the two second-growth forests (Table 1).In the SGF tailing, the pH values varied between 5.35 (medium acidity) and 6.42 (weak acidity), and the exchangeable aluminum content was 0.00 cmolc dm-3, indicating that the tailings are not considered acid and do not present aluminum toxicity (Table 1).As for base saturation (V%), overall, the values were consideredhigh.The organic matter (OM) content was classified as low (0.13-1.39 dag kg-1).

)%( V 73.0 ± 56.4*12.0 ± 8.55 03 MO 1-)g k gad(898.0 ± 66.6*23.0 ± 46.0 5.0 CEC 3-)m d lomc(c*95.0 ± 69.0 3.0 226.0 ± 46.9 1 elbaT)gniliat-non FGS( sgniliat tuohtiw dna )gniliat FGS( sgniliat yb decnelufni tserof htworg-dnoces ni lios fo seitreporp lacimehC BS 3-)m d lomc(c 765.0 ± 44.0*33.0 ± 79.0 2.0 lA + H 3-)m d lomc(c 565.0 ± 02.9*33.0 ± 99.0 1.0+3 lA 3-)m d lomc(c 00.0 ± 10.2*20.0 ± 01.0 90.0+2 gM 3-)m d lomc(c 30.0 ± 11.0 sn 90.0 ± 31.0 1.0 10.0 ± 81.0 7.0 42.4 ± 00.16 6.21+2 aC 3-)m d lomc(c 3-3-)m d gM( K )m d gM( P 3.8+)H( Hp 33.2 ± 58.3 30.0 ± 00.4 sn *32.0 ± 77.0 *00.31 ± 02.4295.1 ± 45.4 *42.0 ± 19.5 7-non FGS gniliat gniliat FGS ecnerefeR gniliat noxocliW-nnaM-yentihW* .ytidica elbaegnahcxe :lA + H ,xedni noitarutas esab :V ;yticapac egnahcxe noitac evitceffe :CEC ;sesab elbaegnahcxe latot :BS ;rettam cinagro :MO :snoitaiverbbA detacidni era ecnabrutsid eht retfa htnom 1 seitreporp sgniliat fo seulav ecnerefeR .secnereffid tnaicfingis on=sn ,)50.0 ＜ p( sFGS neewteb secnereffid tnaicfingis dewohs tset

The results showed that SGF non-tailing soil was highly acidic (pH 3.98 and 4.02) corroborating the high levels of exchangeable aluminum (2.01 cmolc dm-3); consequently, this result can be evidenced by the low values of the sums of bases (0.40 and 0.48 cmolc dm-3) and base saturation index (V%), but high organic matter (6.02 and 7.29 dag kg-1).Conversely, the SGF-tailing presented high bases saturation index (＞ 50%), which is in the fertile soil category.

Litter stock and chemical properties across second-growth forests

The first two axes of the PCA explained ～ 70.1% of the variation in soil properties (Fig.5).The first axis (PCA1) explained 50% of the variation in soil data and was positively correlated with litter (R=0.56,p＜ 0.05) and with the variability of soil properties related to fertility, i.e., total exchangeable bases (R=0.89,p＜ 0.05) and pH (R=0.95,p＜ 0.05) and negatively correlated with organic matter (R=-0.96,p＜ 0.05).The second axis (PCA2) explained 19.8% of the variation in soil variables and was positively correlated with Mg (R=0.89,p＜ 0.05) (Fig.5, Fig.S3).

Fig.5 Principal component analysis (PCA) for litter stock properties between secondgrowth forests patches (SGF non-tailing and SGF-tailing).After the selection of parameters based on the Spearman correlation, the following variables were chosen: available Mg, pH, organic matter (OM), total exchangeable bases (SB) and litter.Cos2 means the relative contribution of the variables represented by the vectors

Effects of stand age, litter stock and plant community composition on soil and technosol fertility

The total exchangeable bases (SB) varied significantly with the positive effects of stand age (est.= 0.03,t=5.10,p＜ 0.001), litter stock (est.= 0.013,t=3.14,p＜ 0.001) and plant community composition (est.= 0.18,t=5.48,p＜ 0.001) (Table 2, Fig.6).However, when analyzing the effect of the interaction between litter and stand age on the SB (Model 4 from Table 2), a negative effect was found on the SGF non-tailing (est.=-0.026,t=-4.52,p＜ 0.001) and a positive effect on the SGF tailing (est.= 0.06,t=2.12,p＜ 0.01).On the other hand, there was also a significant relationship between organic matter and interaction between litter-stand age (Model 8 from Table 2), where the SGF non-tailing had a positive relationship (est.= 0.35,t=9.20,p＜ 0.001).However, we found a negative influence of the SGF tailing (est.=-0.082,t=-4.09,p＜ 0.01) and variation in the plant community composition (est.=-2.56, t =-7.5,p＜ 0.01) on organic matter (Table 2).

Fig.6 Main effect of litter stock (a) and NMDS axis 1 (b) on the total exchangeable bases (SB) as the main ecological indicator of soil fertility according with GLMM approach.Colored circles indicate data per treatments.Solid lines represent the fitted values (prediction) of the models, and the gray area is the 95% confidence interval of the predicted values for each model.The pH (a) and organic matter (b) variability along SGF-tailing and SGF non-tailing gradient are indicated

Table 2 Subset of models predicting (generalized linear mixed effect model) the main effect of litter stock, stand age, NMDS axis 1 (plant community composition variability) and their interactions on the total exchangeable bases (SB) and organic matter (OM) as the main ecological indicator acros SGF-tailing and SGF nontailing gradient.

Discussion

Our results showed significant differences in stand-agedependent forest attributes (litter stock and plant community composition) and soil chemical properties between the two forest patches (SGF-tailing and SGF non-tailing) evaluated.Thus, there is a marked litter storage and fertility gradient between soil and technosol that can be important recovery indicators for plant communities affected by the tailings in Mariana.Although SGF-tailing is in the early successional stage, with only 3 years of regeneration after the disturbance, the litter stock and its interaction with stand age have a positive relationship with total exchangeable bases (SB) as the main fertility indicator of technosol, due to its high correlation with essential nutrients for plant growth (P, K, Ca2+, Mg2+).Furthermore, according to the tested models, the plant community composition variability had significant positive effects on the SB in the second-growth forests studied and, as well demonstrated in previous studies, that the vegetation also shaped soil fertility (Laughling et al.2015; Schilling et al.2016; Olveira et al.2019).Higher soil organic matter content was observed in the SGF patch with mid successional stage, which was dominated by early and mid successional species, but with negative effects on SGFtailing where the lowest values were present.The highest SB values on technosol (SGF-tailing) were observed after the disturbance due to the high litter supply, probably related to the high proportion of pioneer species (Machado et al.2015; Rocha-Martins et al.2018).However, our study tested the direct effect of plant community composition on litter stock.These results corroborated our hypothesis that technosol fertility is positively affected by stand age, plant community composition variability, and litter stock, which may be contributing considerably to forest recovery on the tailings.

Stand-age-dependent forest attributes across second-growth forests

Our results also corroborated marked changes in the species composition across second-growth forests and showed the highest proportion of pioneer species in SGF-tailing at an early successional stage.Most research on tropical forest succession has been based on the analysis of changes in the community structure using a chronosequence approach (Chazdon 2014; Rozendaal et al.2019; Poorter et al.2019).In addition, a marked pattern of changes in the species composition and functional groups has been identifeid along secondary successions (Rozendaal et al.2019; Poorter et al.2019; Villa et al.2019, 2020b).For example, the dynamics of pioneering species after disturbance correspond to the early successional stage, followed by mid successional species, and finally shade-tolerant species in late-successional stages (Chazdon 2014; Poorter et al.2019; Villa et al.2019).Moreover, these changes in functional groups along the succession affect different ecosystem processes, such as litter production and nutrient dynamics (Laughling et al.2015; Schilling et al.2016; Olveira et al.2019), which are determining ecological indicators for the tropical forest recovery after a mining disturbance (Martins et al.2020; Rocha-Martins et al.2020).Thus, we assume that both the independent and joint effects of litter stock and stand age on the fertility indicators studied (SB and OM) are the main predictors of forest recovery on the tailings in the study area from Mariana.

In this context, the forests re-growing on tailings had a litter stock twice as large as that in the reference forest (SGFnon-tailing) and also higher than reported for other tropical forests (4-20 kg ha-1; O′Connell and Sankaran 1997), including forests recovering after a mining disturbance (Rocha-Martins et al.2018; Silva et al.2018).However, the values reported for tropical forests are close to those we found for the reference forest (～ 10 kg ha-1).On the other hand, we highlighted that the SGF on tailings was in an early-sucessional stage compared to other studies in the region where the forests were at different mid and late successional stages after disturbance; for example, 45 years (Pinto et al.2008), 40 years (Miranda-Neto et al.2014), and 5 years (Silva et al.2018) of forest since the disturbance, with 4.60, 3.40 and 6.30 kg ha-1of litter stock, respectively.Other studies in the region have also recorded similar values; for example, a dense ombrophilous forest after 23 years of regeneration had 3.17 kg ha-1(Correia et al.2016), and a semideciduous seasonal forest after 18 years had an average litter stock of 5.6 kg ha-1(Sperandio et al.2012).Furthermore, during the early successional stageof dense ombrophilous forests, litter sock ranged from 4.5 to 9.1 Mg ha-1(Caldeira et al.2008; Klippel et al.2016).Thus, despite the SGF on tailing being in an initial successional stage compared to the cited studies (including areas degraded by mining), litter stock was high, highlighting the relevance of this potential indicator of resilience.

Relationships of stand age, community composition variability, and litter stock

Most studies have reported that during the early successional stage of second-growth forests (stand age of 3-15 years old) there is a tendency toward greater litter production (Martins and Rodrigues 1999; Silva et al.2018), resulting from the dominance of common light-demanding pioneer species (i.e.,Cecropia glazioviiandCecropia hololeuca), due to their fast growth and short life cycle (Guariguata and Ostertag 2001; Chazdon 2014).These species produce high biomass in a short time, being early in reproductive phenophases with a higher production of flowers and fruits and in the production and renewal of leaves (Martins and Rodrigues 1999; Woiciechowisky and Marques 2017).Thus, our study reveals that the positive relationship between species composition and SB is due to the high proportion of common light-demanding pioneer species in SGF on tailing.A high proportion of common light-demanding pioneer species are present in SGF-tailing, which is common in postdisturbance tropical forests (Martins and Engel 2007; Maza-Villalobos et al.2011; Sousa et al 2017), due to their greater quantity of viable seeds in the soil seed bank (Martins 2009, 2018), which simultaneously may contribute to the litter stock during natural regeneration.Therefore, we assume that these species are shaded by the upper canopy strata species (Guariguata and Ostertag 2001; Chazdon 2014).Also at the end of their phenological cycle, these species can become part of the technosol litter, which decompose and contribute to fertility and forest recovery, even on the technosol.

Previous studies report that litter storage can also increase as secondary succession progresses (Woiciechowisky and Marques 2017), thus returning more nutrients to the soil (Caldeira et al.2013; Sloboda et al.2017) and promoting greater microbial activity (Prado et al.2019).Prado et al.(2019), studying areas impacted by the tailings in Mariana, demonstrated a positive relationship between natural regeneration and an increase in microbial communities.Thus, when the averages for soil properties were compared in the present study, only P and Mg did not differ between the two forest patches (SGF-tailing and SGF-non-tailing) studied.We assume that along forest succession the litter stock and the nutrients turnover will increase, contributing to the structure and fertility of technosol.The litter stock is expected to stabilize during late-succesional stages and oldgrowth forest, when the dominance of shade-tolerant species increases (Grugiki et al.2017; Gomes-Júnior et al.2019).We presumed that the changes in both stand age and plant community composition will induce spatial heterogeneity in the forest floor litter stock at a local scale and consequently promote soil recovery (Feng et al.2019; Souza et al.2019).

Most previous studies determining the effects of plant community composition on litter (litterfall, stock and decomposition) have been made with highly contrasting forest types and site conditions (Laughling et al.2015; Schilling et al.2016; Olveira et al.2019).We recognize that it is still necessary to evaluate different ecosystem processes (plant to soil and soil to plant feedback) in our study area.However, this study highlights the relevance of ecological indicators of forest restoration, which for this case study several predictors show positive effects on technosol fertility.We suggest that forest type (i.e.decidual and semidecidual, dry and wet forests) and stand age (i.e.successional stage) be evaluated for their effects on litter stock that may result from drivers of litter quality and quantity, regulation of soil processes, and modification of soil properties.

Our results showed that SGF on tailing have higher total exchangeable bases and base saturation index, indicating higher fertility (V ＞ 50%), because the technosol has a more advanced mineral phase, due to the litter being in the initial decomposition process, confirmed by the low organic matter content.The low organic matter content may be due to erosion caused by the tailings, which causes the loss of the superficial horizon, that is rich in organic matter (Schaefer et al.2017).In contrast, the high Al3+values (～ 2.01 cmolc dm-3) and low SB values (0.44 cmolc dm-3) in the reference forest (SGF-non-tailing) may be related to lower litter production and stock and reduced decomposition (Martius et al.2004; Caldeira et al.2008).In this context, the SGF-non-tailing forest had higher acidity, which can be attributed to its higher organic matter accumulation because the litter was at an advanced stage of mineralization, which can increase the pH (McBride 1994).Thus, the pH values tend to be relatively high, due to the approach of the zero charge point of iron oxides (close to neutrality), which predominate in these substrates (Schaefer et al.2017).The ecological indicators evaluated in this study based on stand-age-dependent forest attributes and their effects on soil properties related to fertility showed an increase in total exchangeable bases in the technosol and that these growth conditions on tailings are not limiting regeneration of the natural forest.

Conclusions

This study revealed that stand-age-dependent forest attributes (litter stock and community composition) have a significant positive relationship with total exchangeable bases, and litter and stand age were the best predictors that explain the increase in technosol fertility.The litter stock in the secondgrowth forest re-growing on tailing (early successional stand age of 3 years) was two times higher (～ 20 Mg ha-1) than in the second-growth reference forest (SGF-non-tailing) at a mid successional stage.Our results demonstrate that the predictors of fertility analyzed to explain forest resilience on tailings can be considered as ecological indicators for assessing forest restoration in areas impacted by mining tailings in Mariana.