Effect of Amazonian tree species on soil and pasture quality in silvopastoral systems

With the expansion of livestock in the Amazon region, a high percentage of pasture areas are degraded and unproductive. Novel strategies are needed, including the use of native tree species, to simultaneously achieve economic and ecosystem benefits. This study aimed at assessing the effects of five multipurpose native tree species on soil fertility and forage quality of Urochloa brizantha pastures in the southern Amazon. Soil and forage samples were collected under the crown and adjacent to 25 isolated trees belonging to five species during a dry and a rainy season. The presence of native trees positively affected the level of potassium, calcium and manganese in the soil, as well as the mineral matter and crude protein of the forage, especially in the dry season, suggesting a protective effect against the seasonal drought. The tree species had variable effects on soil fertility and forage quality. Soil under Apeiba tibourbou had higher potassium levels, while the forage under Handroanthus serratifolius had higher protein and fiber content. Our results indicate that it is important to diversify silvopastoral systems in the Amazon through the use of native tree species, contributing to the design of novel silvopastoral strategies in the region. Common multipurpose tree species with widespread natural distribution could be used as a complementary aspect of pasture management to provide a protective effect against drought, contribute to enhanced nutrient cycling and even increase forage quality.


INTRODUCTION
Agriculture-dominated landscapes cover approximately 50% of tropical biomes, with pastures being an ultimate cause of two thirds of deforested land in the neotropics (Lerner 2015). In Latin America, there is a strong link between deforestation and livestock expansion, either for establishing pastures or for producing grains for animal feed (FAO 2012). Livestock expansion throughout forested regions has brought short-term benefits, considering the sheer financial volume involved with this activity. However, livestock farming has disadvantages in the long term, especially associated with biodiversity loss, carbon emissions, and reduction or disruption of ecosystem services such as species habitat, conservation of soil and water resources, and carbon stock and sequestration (Barret et al. 2013). Since worldwide food demand is increasing and food production is directly linked with biodiversity and ecosystem services, it is of paramount importance to develop and consolidate novel agricultural systems that could increase productivity while mitigating climate change (IPCC 2019).
Silvopastoral systems are one of the possible solutions towards a more efficient production system, consisting of an intentional combination of various trees and livestock using different arrangements through time and space (Almeida et al. 2013). These systems produce extensive benefits that vary according to local traits and designs, generally including the establishment of biological corridors (Harvey et al. 2005;Vergne et al. 2016), and improvements to the biological, chemical and physical structure of the soil (Nair 2007;Pinho et al. 2012;Martinez et al. 2014;), forage quality (Bernardi et al. 2016), and thermal comfort of animals (West 2003;Carvalho et al. 2018).
Several studies on the effects of the presence of native trees in pastures, especially in the tropics, demonstrated benefits such as increased animal productivity, improved quality of forage plants and improved physical, chemical and biological parameters of the soil (Franke 1999;Delgado et al. 2014;Muroe and Isaac 2014;Santos et al. 2016;Camero-Rey and Diaz 2017;Ledesma et al. 2017;England et al. 2020;Alvarez et al. 2021). There is solid evidence that different tree species affect soil fertility in temperate environments by minimizing erosion and leaching while increasing nutrient inputs, as well as enhancing soil structure and biological activity (Binkley and Giarina 1998;Schroth and Sinclair 2003;Devendra 2014;Chará et al. 2019;Sarvade et al. 2019;Rozek et al. 2020).
These effects are related to the optimization of nutrient cycling and the establishment of a microenvironment that improves forage quality (Schroth and Sinclair 2003;Lambers et al. 2008;Sileshi et al. 2014;Bernardi et al. 2016;Araujo et al. 2020). In Brazil, the most widely recommended systems are based on the incorporation of exotic tree species such as eucalyptus (Eucalyptus spp.) and teak (Tectona grandis L.f.), mainly to increase financial profitability land use. However, in northern Mato Grosso state (Brazil), scattered native trees in pastures are used to provide shade and nutritional supplement (from fruits and leaves) for the herd, especially in the dry season (pers. obs. by the authors). These are usually the main reasons for farmers to keep trees in pastures (Harvey and Haber 1999;Lasco et al. 2016;Oliveira and Carvalhaes 2016;Olival et al. 2020).
As the action mechanisms of native Amazonan tree species on pastures are still little known, we assessed the forage quality and soil fertility in pasture under the influence of five native tree species during the rainy and dry season in southern Amazonia. Our hypothesis was that the presence of native trees on pastures positively influences soil fertility as well as forage quality and that these effects depend on the characteristics of the tree species and also on the pasture management.

Study area
The study was carried out in two farms in the municipality of Nova Canaã do Norte (10°36'42.840"S, 55°2'35.880"W; 10°48'59.654"S, 55°41'54.350"W) and two farms in Nova Guarita (10°12'21.030"S, 55°14'49.369"W; 10°13'8.279"S, 55°23'18.122"W), in northern Mato Grosso state, Brazil, all within the watershed of the Teles Pires River, a major tributary of the Tapajós River in the Amazon basin. Local landscapes are characterized by large expanses of pasture interspersed with soy and cotton crops and native vegetation, mainly represented by open tropical forest fragments (IBGE 2012). Soil is dystrophic red-yellow podzolic (Mato Grosso 2000). The climate is tropical monsoon (type Am by the Köppen system), with total annual precipitation around 3000 mm, mainly from November to March, and average annual temperature above 26 °C (Alvarez et al. 2013).
The areas are located in the "arc of deforestation" of the Amazon region, where large areas of forest have been transformed into pastures over the last 30 years. In many pastures, remnant trees persist or are allowed to grow, especially in family farms, which account for more than 80% of rural properties in northern Mato Grosso (IBGE 2019). The four rural properties sampled in the study participate in forest-restoration and agroforestry projects since 2010 in partnership with local organizations.

Species selection
For the selection of trees to test the effect on soil fertility and nutritional properties of the forage, we mapped all trees scattered across 237.9 ha in pastures of 39 family farms across the study area. Based on this survey, three workshops were held with the farm owners to decide, based on their perception, which of the naturally occurring tree species on their lands had the highest potential to bring benefits for their productive systems. The main benefits identified by the farmers were the VOL. 51(4) 2021: 281 -290 ACTA AMAZONICA provision of thermal comfort to livestock through shade and the improvement of soil conditions. We selected five tree species considered by the farmers as having the highest potential to bring benefits: Apeiba tibourbou Aubl. (Malvaceae), Handroanthus serratifolius (Vahl.) S. Grose (Bignoniaceae), Maclura tinctoria (L.) D.Don ex Steud (Moraceae), Platymiscium floribundum Vogel (Fabaceae) and Samanea tubulosa (Benth.) Barneby & J.W.Grimes (Fabaceae). These are multipurpose species with a widespread natural occurrence in Brazil and adjacent countries, with spontaneous colonization and persistence in local pastures. All provide high quality timber, H. serratifolius being currently the most valued Brazilian timber species (Brancalion et al. 2018). Platymiscium floribundum and S. tubulosa are N-fixing species (de Souza 2010). Leaves of A. tibourbou and M. tinctoria, and fruits of S. tubulosa provide alternative forage for cattle, as indicated by the farmers and by Carvalho (2007). See Supplementary Material (Table S1) for further information on functional and silvicultural characteristics of the species.

Sampling design
For each selected tree species, we identified one pasture containing five isolated individuals in the same paddock. Trees had to be mature, older than 10 years and isolated from other trees by at least 50 m, in order to avoid confusing the effects of focal trees with those of neighboring trees. Out of the 39 mapped farms, only four met all these criteria. Three contained sets of one species each, and one contained sets of two species (H. serratifolius and S. tubulosa). The pastures in all four farms were of palisade grass (Urochloa brizantha (A. Rich) R.D.Webster), the most common forage grass used in the southern Amazon.
Pastures containing H. serratifolius and S. tubulosa covered 3.41 ha and had 4.7% of total tree cover, including trees of other species, and the highest cattle-stocking rate (6.3 AU ha -1 ), with 14 paddocks for rotational grazing and grassfed supplementation during the dry season. The pasture containing A. tibourbou covered 5.29 ha with 13.1% tree cover. This was the only pasture where cattle received no feed supplementation in the dry season, and had the lowest cattle-stocking rate (0.6 AU ha -1 ) and grazing divided in 11 paddocks. The pasture containing M. tinctoria covered 8.17 ha with 9.8% tree cover, a cattle-stocking rate of 1.32 AU ha -1 , 23 paddocks and grass supplementation in the dry season. The pasture containing P. floribundum covered 1.93 ha, with the highest proportion of tree cover (31.6%), cattle-stocking rate of 0.97 AU ha -1 , and five paddocks. Resting time of the pastures varied around 30 days, except for the A. tibourbou pasture, which rested for 45 days in the rainy season, and 120 days in the dry season.
For each tree, we measured diameter at breast height (DBH) and crown area by measuring eight crown-radius lengths (distance from trunk to crown projection limits using measuring tapes) in order to form eight triangles. The crown area was then calculated from the sum of the triangle areas (Gomes et al. 2003) (Supplementary Material, Table S2). We collected forage and soil samples in the crown area of each tree (considered to be directly influenced by the tree presence) and in areas 5 m outside the crown area, considered not to be influenced by the trees, asthe benefits of trees on soil decrease with increasing distance from the tree (Oliveira et al. 2012). Thus, although there were other trees in the pastures, each focal tree was assumed to be isolated from the effect of other trees of the same or other species. Forage and soil samples were collected during one dry (August 2017) and one rainy season (January 2018). In each season, we randomly collected four forage and five soil samples within the crown area of each focal tree, and four and five samples, respectively, outside the crown area. For each forage sample, we placed a hollow square (1 m 2 ) and collected the forage mass (cut at 2/3 of blade height) inside the square area. The soil samples were collected at 0-10 cm depth. The samples for each tree and season were pooled to form one composite sample of forage and one of soil under the canopy (within the crown area), and one sample each off the canopy (outside the crown area). Therefore, we had five replicates per species and season for each treatment (presence or absence of the tree).
The fresh mass of the forage samples was weighed in the field and then transported to the laboratory, where they were oven dried at 65°C for 72 h to obtain the dry mass. The percentage of mineral materials (MM), ethereal extract (EE) and crude protein (CP) were quantified according to AOAC (2012). Neutral detergent fiber (NDF), acid detergent fiber (ADF) and lignin content (LC) were determined following Van Soest et al. (1991). The soil samples were analyzed for organic matter (OM), calcium (Ca), magnesium (Mg), potassium (K), phosphorus (P) following Embrapa (2009). Forage samples were analyzed at the Laboratory of Animal Nutrition of Universidade Federal de São Carlos, and soil samples were analyzed at the Laboratory of Soil and Leaf Analysis of Universidade so Estado de Mato Grosso.
All farms showed some sign of pasture degradation, mainly due to management problems. We found a high incidence of invasive plants and sudden-death syndrome, as well as signs of poor quality of forage plants, especially in the dry season. Forage yield was not the focus of this study, but, for the sake of charcterization, we observed that average forage availability varied from around 300 kg dry mass ha -1 off the canopy of M. tinctoria in the dry season to almost 5,000 kg ha -1 off the canopy of H. serratifolius in the rainy season. There was a two to three-fold reduction in pasture dry mass in the dry season, when yield varied from 300 to almost 2,000 kg ha -1 . In the rainy season it varied from 2,000 to almost 5,000 kg ha -1 . Average yield difference between forage grown under and off canopy varied from less than 20 kg in the dry season A. tibourbou pasture to almost 1,000 kg ha -1 in the M. tinctoria pasture. VOL. 51(4) 2021: 281 -290 ACTA AMAZONICA

Data analysis
As forage quality proxies, we used the proportion of MM, CP, NDF, ADF and LC, and as proxies for soil fertility we used P, K, Ca, Mg and OM. The adherence of the variables to different theoretical distribution functions was tested through skewness and kustosis comparisons using graphical analysis of the package 'fitdistrplus' (Delignette-Muller and Dutang 2015), in the R environment (R Core Team 2019). Then we used the information-theoretic approach for model selection (Burnhan and Anderson 2002) by fitting generalized linear mixed models (GLMM) to the data of each of the above dependent variables and using tree presence/absence, season and species as categorical fixed effects. To account for repeated measures of individual trees, we fitted random intercept models, with varying intercept among individuals. The beta distribution family with a logit link was used to fit forage quality models, while the gamma family with a log link was used to fit soil fertility models. The models were fitted with the aid of 'glmmTMB' and 'lme4' packages (Bates et al. 2015;Brooks et al. 2017). Summary statistics such as the coefficient of determination (R 2 ) and intra-class correlation coefficients (ICC) were computed according to Nakagawa et al. (2017). Finally, Wald Type II chi-square tests on the fixed effects and Tukey post hoc tests were performed for every selected model (Fox and Weisberg 2019). Confidence intervals for model estimates were derived from Wald approximation.

RESULTS
The best-fitting GLMM models for soil fertility included a significant effect of tree presence for all variables (except P) as well as variation between seasons and among species (Supplementary Material, Table S3). There was a general tendency of decrease in soil fertility in the dry season ( Figure  1). The selected models explained between 65% (P) and 90% (K) of the observed variance in soil fertility, as indicated by the conditional coefficient of determination (R 2 GLMM(c) ) ( Table  1). The presence of trees accounted for 10 to 28% of observed variation, as indicated by the adjusted ICC values (Table 1).  Figure S1). K content was significantly higher in both seasons under A. tibourbou (rainy: z ratio = -5.672, p < 0.001; dry: z ratio = -3.849, p = 0.0001) and (almost twice as high under the trees), H. serratifolius (rainy: z ratio = -4.024, p = 0.0001; dry: z ratio = -2.972, p = 0.0030), and P. floribundum in the rainy season (z ratio = -2.892, p = 0.0038) (Figure 1b).
The best-fitting models for forage quality (except for LC) included the interaction effects of tree presence, season and species (Table 1; Supplementary Material, Table S4). Contrary to the soil fertily models, there was no significant influence of random effects of individual trees on forage-quality attributes (Supplementary Material, Figure S1b and S2b). The best-fitting LC model included seasonality and species (a proxy for sampling sites) (Figure 2e), but not tree presence (Supplementary Material, Table S4).
As expected, MM and CP content in the forage decreased in the dry season in all cases, while fiber content increased in most cases ( Figure 2). Tree presence tended to increase MM and CP in forage, with positive effects of varying magnitude in all species (Figure 2). MM was significantly higher under A. tibourbou in the dry season (t ratio = -4.787, p < 0.0001), M. tinctoria in both seasons (dry: t ratio = -2.335, p = 0.0221; rainy: t ratio = -4.969, p < 0.0001), and P. floribundum in the rainy season (t ratio = -2.347, p = 0.0215) (Figure 2a). CP was significantly higher under H. serratifolius in both the dry (t ratio = -3579, p = 0.0006) and rainy season (t ratio = -2.593, p = 0.0114), and S. tubulosa in the rainy season (t ratio = -2.060, p = 0.0428).

DISCUSSION
Our results showed that native Amazonian tree species can have neutral or positive eefects on forage quality and soil fertility in agricultural landscapes that interact with seasonal and site-specific effects. Quantitative and qualitative responses of forage depend on the architecture of forage plants, the density and shading of tree canopies, and environmental factors such as soil fertility and moisture, (Jose et al. 2019). Several studies have demonstrated the benefits of natural or artificial shade for forage yield and quality (Andrade et al. 2004;Sousa et al. 2010;Pang et al. 2019a;Pang et al. 2019b;Pezzopane et al. 2019). Some forage species, including grasses and legumes commonly used in pastures across the Amazon basin, are shade tolerant, usually adjusting to decreased sunlight influx by devoting more primary production to leaf area than root matter (Andrade et al. 2004).
Our results suggest that the presence of native tree species could potentially mitigate forage quality decline in the dry season, as evidenced by the significantly higher contents of MM and CP in dry season forage under A. tibourbou and M. tinctoria. Handroanthus serratifolius in particular stood out as the forage under its canopy had higher levels of CP in both seasons, in addition to a reduced fiber content in the dry season, and overall higher K levels in the soil under the canopy, which was also found for A. tibourbou. Direct effects of trees on the microclimatic and edaphic factors depends primarily on the tree`s ability of shading, which means covering a large area and allowing enough light simultaneously (Andrade et al. 2012). Therefore, allowing some level of tree shading in pastures as a drought mitigation practice, and also as a key source of protective shade, is widely desirable for sustainable tropical livestock systems (Chará et al. 2019;Gomes et al. 2019;Olival et al. 2020).
Overall, our best-fitting models suggested that most soil and forage variables responded to some extent to tree presence and seasonality, and also varied among the species. Yet, although we were able to observe some effects of tree presence on soil fertility and forage quality within species, our sampling design did not allow the identification of interspecific effects. All pastures were located in the same microregion, and were subject to the same climate, overall soil type and stocking rate, yet, as all focal trees of each species were aggregated in different pastures, the species effect was confounded with site effects such as different management practices, stocking rates and herd productivity, as well as soil and terrain specificities. Thus the differencial occurrence and magnitude of the effects of these native tree species on pastures should be elucidated in further studies using an appropriately replicated sampling design.
Species-specific traits generate distinct root exudation, microenvironments and litter quality that strongly influence the soil nutrient supply (Lambers et al. 2008). Higher contents of polyphenols can have a strong effect on litter quality (Lambers et al. 2008), which may be the case with A. tibourbou, as its leaves have high content of rosmarinic acid (Souza et al. 2012) and it had the largest effect on soil fertility, especially on K content. Apeiba tibourbou is commonly found in secondary forests across the Amazon, and probably prefers soils with higher fertility (Tavares et al. 2019). Hence, because of its nutrient requirement, it may recycle nutrients from deeper layers (Sileshi et al. 2014).
The effect of tree presence on forage quality may vary depending not only on tree characteristics but also on the forage species. Many grasses, such as Urochloa spp., can tolerate a reduction in light intensity up to 30 or 40% without Table 1. Summary statistics of selected GLMM models for additive (+) or interactive (*) effects of tree presence, season and species on soil and forage attributes in northern Mato Grosso state, Brazil. GLMM adapted coefficient of determination (R 2 ) of marginal (m) and conditional (c) terms and adjusted intra-class correlation (ICC adj ) are shown. Model calculation for NDF and lignin was not possible due to some variance components being equal to zero. Forage under M. tinctoria was the only with MM significantly higher than off canopy in both seasons. Litter from this species has an accelerated decomposition time and high potential for rapid nutrient incorporation into the soil, especially N, P, K, Ca and Mg (Silva 2016). Constant and abundant leaf production leads to greater nutrient cycling, offering optimal conditions for forage development, making pioneer species such as M. tinctoria good candidates for further studies on the strengthening of ecological processes for the improvement of silvopastoral systems (Parsons and Congdon 2008). The presence of H. serratifolius and S. tubulosa was associated with significantly higher CP in forage in both seasonsand the rainy season, respectively. Trees of Samanea, particularly S. saman and S. tubulosa, provide protective shade in pastures, produce palatable pods that are suitable as a dryseason feed supplementation and enhance grass production beneath their crown (Durr 2001; Andrade et al. 2012). We showed that forage under S. tubulosa has higher CP content, possibly due to higher nitrogen content in the ryzosphere and litter fall, because of its ability to symbiotically fix atmospheric nitrogen or unlock mineral nutrients through enhanced microbiome activities (Epihov et al. 2021).

Selected
The higher CP content in forage under H. serratifolius and S. tubulosa could be related to the larger size of forage-leaf cells and the higher amount of intracellular nutrients, including proteins, resulting from the greater cell volume as a response to higher moisture, stable temperature and light reduction (Sousa et al. 2010;Moreira et al. 2013). Also, forage plants under trees may be physiologically younger, since they have a higher proportion of green leaves, thereby extending the juvenile vegetative state and allowing the maintenance of higher metabolic levels for a longer period (Sousa et al. 2010). In addition, nitrogen-rich organic matter from litter of N-fixing species, such as S. tubulosa, could explain a higher content of CP in forage under these trees (Ledesma et al. 2017). It is noteworthy, however, that, although forage CP under H. serratifolius and S. tubulosa was higher than off the canopy, the protein levels were still less than the minimum required for cattle (Valente et al. 2013), pointing to the importance of complementary measures to manage the livestock system.
The seasonality observed in some positive effects of the presence tree was probably related to the intensity and frequency of rainfall, as the effect of trees on forage quality and soil depends mainly on the weather conditions, which influence the decomposer biota and the quality of the organic matter offered by the trees, such as the C/N ratio and lignin content (Prause and Lopes 2012;Machado et al. 2017). Deciduousness patterns of trees along the year may also affect forage quality and yield during dry seasons, which is a major challenge for the sustainability of livestock systems, as it implies supplementing animal diet at a high cost in order to spare pasture. The long-term effect of native tree species in minimizing forage loss during the dry season, as well as providers of supplemental fodder from their leaves and fruits is a promising line for further studies.
Tree and crown size might be an important factor determining forage and soil characteristics beneath canopies (Treydte et al. 2009) and also the chemical composition of leaves (Binkley and Giardina 1998). Considering the wide spectrum of tree functional traits, such as leaf lifespan, leaf composition, crown dimensions, vegetative and reproductive phenological patterns, further studies should also address diversified set of tree species in the pasture environment for functional complementarity.
The absence of significantly negative effects of the presence of trees in almost all cases is noteworthy, since cutting trees is a culturally widespread practice among cattle ranchers across the Amazon, which has increased the negative impacts of livestock, especially on soil fertility (Machado et al. 2017;Steingraber et al. 2018). The history of recent colonization of agricultural frontiers at the fringes of the Amazon biome is marked by high deforestation encouraged by governments and colonization companies, followed by livestock occupation (Godar et al. 2012). The extensive livestock systems have been established on the basis of slash and burn practices, and subsequent pasture degradation and mismanagement lead to productivity reduction and, eventually, to abandonment of the land. This is a very common condition in Brazil, where more than half of livestock production takes place on degraded pastures (Silva et al. 2017). Reintroducing the tree element in pastures in Brazil's arc of deforestation has the potential to contribute as a sustainable practice to restoring degraded pastures and improving soil quality and the quality and amount of forage, in addition to the adjustment of stocking rates to the actual carrying capacity of the pasture (Boval and Dixon 2012). The implementation of these types of silvopastoral systems is advancing across the Amazon region and can potentially support livestock intensification while benefitting ecosystem services (Gomes et al. 2020).

CONCLUSIONS
Our study complements the empirical knowledge of cattle farmers in the southwestern Brazilian Amazon about the silvopastoral benefits of five native Amazonian tree species. VOL. 51(4) 2021: 281 -290 ACTA AMAZONICA We observed some effects of the presence of common multipurpose tree species widely used in pastures in our study area regarding protecton against drought effects, enhancement of nutrient cycling and improvement of forage quality through higher mineral and protein content, as well as lower fiber content. The species with more significant positive effects were Apeiba tibourbou, Handroanthus serratifolius and Maclura tinctoria. Our results suggest that tree cover is a beneficial complementary strategy of pasture management. Contrary to cultural beliefs, our results showed practically no signs that the presence of trees impairs forage quality. Further studies using larger samples and site-independent designs should further assess the influence of the presence of these species on the pasture environment and their benefits for sustainable silvopastoral systems.

ACKNOWLEDGMENTS
We are very thankful to the farmers participating in our research program and to the editor and three anonymous reviewers for their bright contributions. We thank the students of the Universidade do Estado de Mato Grosso and the technicians of Instituto Ouro Verde. This work was supported by Fundo Amazônia, through the Banco Nacional de Desenvolvimento Econômico e Social (BNDES).     The rows in bold indicate the best-fitting models with the lowest AIC value or ΔAIC < 2. df = degrees of freedom; AIC: = Akaike's Information Criterion; ΔAIC = difference in AIC between the given model and the model with the lowest AIC score; Weight = Akaike weight showing the relative support for each model; Likelihood = the likelihood of the model, an indicator of fit; R 2 GLMM(m) = marginal coefficient of determination; R 2 GLMM(c) = conditional coefficient of determination. The rows in bold indicate the best-fitting models with the lowest AIC value. df = degrees of freedom; AIC = Akaike's Information Criterion; ΔAIC = difference in AIC between the given model and the model with the lowest AIC score; Weight = Akaike weight showing the relative support for each model; Likelihood = the likelihood of the model, an indicator of fit; Pseudo R 2 = a generalized measure of the model coefficient of determination.