Introduction

Only a small part of the huge tropical plant biodiversity is used in agro-ecosystems. Those plants that are used have been systematically overlooked by science (Leakey et al. 2005). For these reasons, agro-ecosystems lose important ecosystem services associated with biodiversity, and they do not contribute as much as possible to environmental preservation. For instance, in the Atlantic Rainforest biome, which is one of the five biodiversity hotspots of the world (Myers et al. 2000), the agricultural matrix of coffee (Coffea arabica L.) in full sun and monoculture pasture (Cardoso et al. 2001) does not contribute to preservation of the numerous forest fragments that still exist in the biome. A more diversified agricultural matrix using agroforestry where native trees are intercropped with coffee and pasture has been regarded as one of the best options for conservation of this biome, whose natural habitats have been largely converted for agricultural or other use (Perfecto and Vandermeer 2008).

Atlantic Rainforest tree species are not widely used in agro-ecosystems, and little is known about their functions. Despite the importance of intrinsic biodiversity values, the trees by themselves do not justify monetary and political efforts in the study and conservation of biodiversity in most cases (Moonen and Bárberi 2008). These efforts are only fruitful if biodiversity has a role in the functioning of agro-ecosystems and if these systems directly or indirectly benefit human society. Such benefits are known as ecosystem services (Costanza et al. 1997).

However, the trees of the Atlantic Rainforest can be used in agroforestry systems, where they benefit nature conservation and provide ecosystem services. For instance, agroforestry systems can serve as corridors, linking forest fragments, which contribute to their preservation (Perfecto and Vandermeer 2008) and the use of native tree species in agroforestry systems can improve soil quality, an important ecosystem service (Souza et al. 2012). The organic material (above- and below-ground litter) produced by trees in agroforestry systems protects the soil against erosion, serves as food for soil organisms, improves soil structure, and contributes to nutrient cycling, thus increasing nutrient availability to plants (Beddy et al. 2010).

The use of native trees in agroforestry systems in the tropics could be prioritised. These trees usually have efficient mechanisms to deal with exchangeable acidity and low levels of available nutrients, especially N and P (Kanmegne et al. 1999). Species of the Fabaceae (legumes) are of particular importance because many of them are associated with N2-fixing bacteria, which contribute to the continuous incorporation of N in the system, thereby stimulating cycling of other nutrients (Sá and Vargas 1997). Positive interactions between N2-fixing and non-fixing tree species have been widely reported in diverse cropping systems, including coffee systems (Nyfeler et al. 2011).

Decomposition and nutrient cycling rates depend both on environmental conditions and quality of the plant material, which is determined by its (bio-) chemical composition. These characteristics are often used to predict decomposition rates and their effects on soil nutrient availability (Mendonça and Stott 2003; Silva et al. 2008). Nitrogen and lignin (LG) mass fractions, C/N ratios, polyphenol (PP)/N ratios, and (LG+PP)/N ratios are considered to be the best predictors of N release by species in agroforestry systems (Palm and Sanchez 1991).

Material with high contents of compounds with slow decomposition (LG and PP) and low content of N hinder action of microorganisms, leading to a lower rate of decomposition and nutrient release. Materials with a fraction of N lower than 1.74 % (Palm 1995), hence a C/N ratio larger than 28.7 (assuming 50 % C), typically have low decomposition rates. Decomposition rates also decrease as a function of LG+PP content (Palm 1995; Hadas et al. 2004). According to Palm (1995), N mineralisation rate decreases when N is smaller than 1.74 %, or when LG and PP are larger than 15 and 3 % respectively, and the (LG+PP)/N ratio is larger than 10. Plant materials with PP values lower than 0.5 % are considered of high quality and can be decomposed more rapidly (Tian et al. 1992). According to Silva et al. (2008), materials that have LG/N ratios larger than 5 (different from Palm 1995, LG/N >8.5) have low decomposition rates. Even though slow litter decomposition hinders nutrient release, litter accumulation can contribute to better soil protection, higher moisture content, and less soil loss through erosion (Silva et al. 2008).

For native tropical species that are currently used or have the potential for use in agroforestry systems, information about decomposition and nutrient release is scarce, which hampers to implement and manage agroforestry systems. To provide this information, the current study had the following aims: (i) to assess leaf decomposition of seven tree species used in agroforestry systems in the Atlantic Rainforest (ii) to test if the decomposition of the leaves of those trees is correlated with their (bio-) chemical quality (iii) to assess nutrient release from the leaves of three species of the Fabaceae and (iv) to quantify biological nitrogen fixation (BNF) of these three species.

Materials and methods

Experimental area

The plant material used in this study was collected from agroforestry systems in the municipalities of Araponga (20°48′S and 42°32′W) and Divino (20°33′S and 42°11′W). Both municipalities are located in Zona da Mata (Minas Gerais, Brazil) in the Atlantic Rainforest biome.

The average temperature of the region is 18 °C, and the annual rainfall varies from 1,200 to 1,800 mm with a dry period of 2–4 months. The terrain is mountainous with slopes ranging from 20° to 45° (Golfari 1975). Soils are predominantly oxisols, which are deep and well drained. These soils are acidic and have low nutrient availability, particularly phosphorus (P). The area is mainly used for pasture and full sun coffee (cash crop) production. Coffee is usually intercropped with beans, corn, and other subsistence crops (Cardoso et al. 2001).

In this region, subsistence family farmers are predominant, and they are considered of vital importance, especially with regard to food and coffee production. To improve soil quality, farmers began experimenting agroforestry coffee systems in 1994, with the support of the Center for Alternative Technologies of Zona da Mata (NGO), the Federal University of Viçosa and organisations composed of family farmers, especially rural worker unions (Cardoso et al. 2001). Along the experimenting process, the farmers selected various tree species to be used in the agroforestry coffee and pasture systems consisting of native species of the Atlantic Rainforest and some exotic species, which are adapted to the region and are often fruit-bearers (Souza et al. 2010, 2012).

Species studied

The species selected for this study are among the most widely used in agroforestry coffee systems in Araponga and Divino: Aegiphila sellowiana Cham. (Verbenaceae), Inga subnuda subsp. luschnathiana (Benth.) T. D. Penn. (Fabaceae), Erythrina verna Vell (Fabaceae), Luehea grandiflora Mart. (Malvaceae), Senna macranthera (DC. ex Collad.) HS Irwin & Barneby (Fabaceae), Zeyheria tuberculosa (Vell.) Bur. (Bignoniaceae), and Persea americana Mill. (Lauraceae). Except P. americana, all species are native to the Atlantic Rainforest.

Four randomly selected individuals (considered replicates) of each species were selected from four agroforestry systems located in the properties of family farmers. According to the farmers, the individually sampled trees were introduced as seedlings or spontaneously established on the farms in 1994, with the exception of E. verna, which was introduced to the system in 1998. Only A. sellowiana individuals were drastically pruned during the collection of material.

Leaf sampling

For the carbon dioxide fluxes, chemical and biochemical characterisation of plant material and decomposition and nutrient release, physiologically mature leaves of the four individuals (considered replicates), of each of the seven tree species were sampled. Physiologically mature leaves were characterised by a fully expanded limb and no signs of senescence or necrosis.

In October 2005, fresh leaf samples were collected at the middle canopy position from points located at the North, South, East, and West coordinates (~20 leaves from each coordinate). Most authors dealing with residue decomposition in agroforestry systems consider litterfall as basis for decomposition studies and there is a lack of information about the characteristics and decomposition dynamics of pruned material, a normal practice in agroforestry systems. Therefore, we studied freshly picked leaves more similar to pruned material than litterfall.

The collected material from each individual was pooled, forming a composite sample, dried in a forced-air circulation oven (72 h at 65 °C) and ground for further analysis. For the Fabaceae species, part of the fresh material was separated for remaining dry matter and released nutrients study.

Carbon dioxide fluxes from leaves

Carbon dioxide fluxes from leaves were analysed by respirometry in which CO2–C evolution was measured using a Stotzky respirometer as previously described by Curl and Rodriguez-Kabana (1972). The composed leaf samples of each individual of the seven species were used to determine total C and N by the dry combustion method using an elemental analyser (Perkin Elmer CHNS/O 2400).

In glass bottles (500 ml), plant material corresponding to 2 g of C from each composite sample was mixed with 100 cm3 of B horizon soil (Oxisol), and soil moisture content was increased to 70 % of field capacity. The soil was acidic (pH 4.5), had low available P (1.0 mg dm−3), K (13 mg dm−3), Ca and Mg (0.0 cmolc dm−3) and high exchangeable Al (1.4 cmolc dm−3). The carbon content was 20.9 g kg−1 and field capacity 0.32 kg kg−1 (Embrapa 1999).

The glass bottles containing soil and plant material were connected to the elemental analyser and subjected to continuous flow of free-air CO2. Evolved CO2 was then captured by a NaOH solution and quantified by 0.25 M HCl titration with a phenolphthalein indicator. Controls included four empty bottles and four bottles containing soil without addition of leaf litter. The bottles were prepared in duplicate, and randomized in the respirometer (a total of 72 bottles: four replicates of seven species and two controls, each replicate in duplicate). Fifteen CO2 measurements were taken for each flask over a period of 37 days. The total amount of CO2 produced was calculated from the sum of the values obtained during each sampling. The CO2 (mg C g−1 soil) calculation was performed using the formula:

$$ {\text{CO}}_{2} \left( {\text{mg}} \right) = \left( {{\text{B}} - {\text{V}}} \right) \times {\text{M}} \times 6 \times \left( {{\text{V}}_{1} /{\text{V}}_{2} } \right) $$

where B is the volume of HCl in an empty vial (ml); V is the volume of HCl used in the sample (ml); M is the concentration of HCl; 6 is the atomic mass of carbon (12) divided by moles of CO2 that react with NaOH (2); V1 is the volume of NaOH used to capture CO2 (ml); and V2 is the volume of NaOH used in titration (ml).

Leaf chemical and biochemical characterisation

Chemical and biochemical characterisations were performed on composite samples (without replications) of each species in duplicate. N was determined with the Kjeldahl method after sulphuric acid digestion (Tedesco et al. 1995). P was measured with the colorimetric method in which the molybdate-phosphate complex turns blue in the presence of ascorbic acid (Braga and Defelipo 1974). For calculating C/N and C/P, we assumed that plant dry matter consists of 50 % carbon (Pribyl 2010).

Soluble polyphenols (PPs) were extracted with methanol (50 %) and determined by colorimetry using the Folin–Denis reagent (Anderson and Ingram 1993). Lignin (LG), cellulose (CL), and hemicellulose (HC) were determined with the neutral detergent fibre (NDF) and acid detergent fibre (ADF) methods, which are referred to as the van Soest method (Goering and van Soest 1970; van Soest et al. 1991). The HC values were determined by subtracting ADF from NDF. Lignin contents were obtained from the ADF fraction. CL contents were obtained by subtracting lignin from ADF. Cumulative CO2 flux during a period of 37 days was correlated with mass fractions of LG, CH, CL, PP, N, and P, and also with C/N, LG/N, LG/PP, PP/N, and (LG+PP)/N ratios.

Decomposition and released nutrients

Remaining dry matter and released nutrients were assessed for the three legume species I. subnuda, S. macranthera, and E. verna. Fresh leaves (75 g; in duplicate) from four trees (replicates) of each species were placed in litter bags (20 × 20 cm) with mesh size 2 × 6 mm, allowing most animals of mesofauna (and partly macrofauna) to pass through. Dry matter weight (dried at 65 °C in an oven) was determined in samples equivalent to 75 g of fresh material. The litter bags were randomly distributed in a shaded area located on the campus of the Federal University of Viçosa. The campus is situated in the same region, with the same class of soil and climate as the municipalities where the leaves were collected, thus, we assumed to have similar decomposer biota. Temperature and monthly rainfall data in Viçosa during the experiment are shown in Fig. 1.

Fig. 1
figure 1

Average rainfall and temperature in Viçosa (Minas Gerais, Brazil), from October 2005 to September 2006

Full size image

Two litter bags of four replicates of each tree species were collected at 5, 15, 30, 60, 90, 120, and 150 days (24 litter bags each time). After collection, litter was cleaned and impurities, such as soil, insects, roots, and other materials, removed using brushes and tweezers. The material was then dried in an oven (65 °C) until constant weight and remaining dry matter was calculated.

To analyse decomposition and nutrient mineralisation rate, the one-component exponential decomposition model was applied:

$$ {\text{X}} = {\text{X}}_{0} {\text{e}}^{ - {\text{kt}}} $$

where X is the amount of remaining dry matter or nutrients after a period of time (t; in days); X0 is the amount of dry matter or nutrients at t = 0; k is the decomposition constant (day−1).

Half-life was calculated as t1/2 = ln (2)/k, where t1/2 is half-life (days) for dry matter or nutrients, time required for half of these residues to disappear.

Total N content in the residues was determined using the Kjeldahl method after sulphuric acid digestion. After nitric/perchloric digestion, the P and K contents were determined by flame photometry, and the Ca and Mg contents were determined by atomic absorption spectrophotometry (Braga and Defelipo 1974).

Atmospheric N2 fixation

To assess N2 fixation of the legumes, the 15N natural abundance technique was used (Peoples et al. Peoples et al. 1989). The percentage of N2 fixed by legumes (BNF %) was estimated with the following formula:

$$ \% \,{\text{BNF}} = \frac{{\delta {}^{15}{N}_{reference} - \delta {}^{15}{N}_{legume} }}{{\delta {}^{15}{N}_{reference}} - {\text{B}}} $$

where δ 15 N reference is the natural enrichment of the non-N2-fixing reference species; δ 15 N legume is the natural enrichment of the legume evaluated in the system; and B is the 15N natural abundance (δ 15 N) of the legume grown under exclusive dependence on N2. The B value used was −1.3  ‰ based on the shrub Prosopis glandulosa (Shearer et al. 1983); this value was also used by Teixeira et al. (2006). Despite having more shallow root systems compared to the legume trees, two herbaceous plants, Bidens pilosa (Asteraceae) and Brachiaria plantaginea (Poaceae), were used as non-N2-fixing reference plants. Mature leaves of five plants of B. pilosa and entire shoots of B. plantaginea were collected nearby the trees. The samples were grouped to form one composite sample per reference species.

In January 2006, physiologically mature leaves (fourth pair of leaves on the branch; leaf + petiole) were collected from four individuals (replicates) of each species of the Fabaceae (I. subnuda, S. macranthera, and E. verna). Sampling was performed at the middle canopy position at points located at the North, South, East, and West coordinates (replicates) and pooled to form a composite sample per individual.

Materials from legumes and reference plants were placed in paper bags and subsequently dried in a forced-air circulation oven at 65 °C for 72 h. The materials were then milled and ground in a ball mill followed by 15N analysis with a mass spectrometer (SERCON; model ANCA GLS 20.20).

Statistical analysis

Analyses of variance were performed for respirometry and decomposition followed by planned comparisons to test differences between means. Regression analyses were also performed between decomposition and time; we do not present parameter estimates when these regressions were not significant. Statistica (Statsoft 1997) was used for the statistical analyses.

Results

CO2 fluxes

The average and the standard errors (n = 4) of CO2 fluxes from leaf litter of the seven tree species over a period of 37 days are shown in Fig. 2. At the end of this period, there were differences among species (F 7,21 = 359.0; P < 0.001). Two groups of species that significantly (P < 0.05) differed in the amount of CO2 produced could be distinguished: species that produced more CO2, including S. macranthera, E. verna, Z. tuberculosa, and A. sellowiana; and species that produced less CO2, including P. americana, I. subnuda, and L. grandiflora.

Fig. 2
figure 2

Average and standard errors (n = 4) of CO2–C content in leaf material of tree species used in agroforestry systems in Zona da Mata (Minas Gerais, Brazil) over 888 h of incubation in soil, in the laboratory

Full size image

Leaf chemical and biochemical characterisation

Table 1 summarises data on leaf (bio-)chemistry. LG ranged from 7.7 % (E. verna) to 27.3 % (I. subnuda); HC ranged from 13.4 % (P. americana) to 41.6 % (E. verna); CL ranged from 11.3 % (S. macranthera) to 21.3 % (I. subnuda); PP ranged from 4.4 % (Z. tuberculosa) to 8.3 % (L. grandiflora); N ranged from 2.0 % (L. grandiflora) to 3.82 % (A. sellowiana); and P ranged from 0.11 % (Z. tuberculosa) to 0.19 % (S. macranthera and A. sellowiana).

Table 1 Lignin (LG), hemicellulose (HC), cellulose (CL), polyphenol (PP), N (nitrogen), and P (phosphorus) mass fractions of leaves of trees used in agroforestry coffee systems in the region of Zona da Mata, Minas Gerais, Brazil
Full size table

The three species with the lowest CO2 fluxes (Fig. 2) had the highest (LG+PP)/N ratio, and these species were P. americana (13.8), L. grandiflora (10.9), and I. subnuda (10.1). The species with the highest LG/N ratio were: P. americana (10.6) and I. subnuda (8.6). The highest CL contents were obtained for I. subnuda (21.3 %) and L. grandiflora (17.2 %). L. grandiflora, Z. tuberculosa, and P. americana had C/N ratios greater than 23, and A. sellowiana, and the three legumes had C/N ratios <16. Leaf C/P ratios were larger than 300 in L. grandiflora, Z. tuberculosa, P. americana and I. subnuda, and below 300 in the other species. Throughout the studied period, the CO2 flux was significantly negatively correlated with LG/N and (LG+PP)/N. At the end of the measurements, the cumulative CO2 flux was significantly negatively correlated with the LG/N ratio (r = −0.76; P < 0.05) and (LG+PP)/N ratio (r = −0.79; P < 0.05).

Decomposition and nutrient release

The percentages and standard errors (n = 4) of remaining dry matter from legume leaf litter during the 150 days of analysis are shown in Fig. 3. At 15 days, the remaining dry matter differed among species (F 2,63 = 56.9, P < 0.0001) and with time (F 6,63 = 65.02, P < 0.0001). The differences among species became more pronounced over time (interaction between time and species, F 12,63 = 3.42; P < 0.001). At 150 days, weight loss of I. subnuda was 31 %, S. macranthera 74 % and E. verna 84 %.

Fig. 3
figure 3

Average and standard errors (n = 4) of remaining dry matter from leguminous tree species used in agroforestry systems (Zona da Mata of Minas Gerais, Brazil) over 150 days of incubation in the field, using litter bags

Full size image

Table 2 shows the k values and half-life values (t 1/2) for dry matter (carbon), N, P, K, Ca, and Mg for leaves of the three legume species. For dry matter, k ranged from k = 0.012 day−1 (E. verna), followed by S. macranthera (0.008 day−1) and I. subnuda (0.003 day−1). K values for N and P were similar with the highest k values for E. verna (k N = 0.014 and k P = 0.012 day−1) and lowest k values for S. macranthera (k N = k P = 0.007 day−1) and I. subnuda (k N = 0.004 and k P = 0.007 day−1). After 150 days, approximately 80 % of N and P was released in E. verna litter and 60 % in S. macranthera. For I. subnuda, 35 % N and 50 % P was released.

Table 2 Estimated parameters from the X = X0 e−kt equation adjusted to dry material, N, P, K, Ca, and Mg values and respective half-life times (t1/2 = ln (2)/k) for leaves of trees placed on the soil surface during 150 days and coefficient of determination (R 2) of the equations
Full size table

K release occurred more rapidly in E. verna litter (k K = 0.027 day−1) than in I. subnuda litter (k K = 0.011 day−1). After 150 days, ~99 % of K was released from E. verna litter and 82 % from S. macranthera and I. subnuda litter. Ca and Mg release was slower than K release with KCa = 0.007 day−1 and KMg = 0.010 day−1 for E. verna (values corresponding to 75 % released Ca and 80 % released Mg) and KCa values of 0.004 day−1and KMg of 0.007 day−1 for S. macranthera (values corresponding to 35 % released Ca and 50 % released Mg). For litter of I. subnuda, Ca and Mg release did not follow the exponential model; the p value was not significant.

Biological nitrogen fixation

Table 3 shows the average and standard errors (n = 4) of δ 15N of legume (E. verna, S. macranthera, and I. subnuda) and reference species (B. pilosa and B. plantaginea). The difference between reference species and legume was less than 5 ‰. The highest percentage of BNF was found for E. verna (22.6 %) followed by I. subnuda (20.6 %) and S. macranthera (16.6 %), but these differences were not significantly different (P < 0.05).

Table 3 Average and standard error (n = 4) of natural enrichment of 15N and biological nitrogen fixation (BNF) of legume species (Erythrina verna, Senna macranthera and Inga subnuda) used in agroforestry coffee systems of Zona da Mata, Minas Gerais, Brazil and reference species (Bidens pilosa and Brachiaria plantaginea) for δ 15N
Full size table

Discussion

CO2 fluxes and leaf chemical and biochemical composition

The results of the present study indicated that the differences in C fluxes among species were due to the different compounds in the analysed leaves in which the LG+PP/N and LG/N ratios negatively correlated with CO2–C evolution, which explain the lower decomposition rates of L. grandiflora, I. subnuda, and P. americana leaves than those of A. sellowiana, E. verna, S. macranthera and Z. tuberculosa.

All species had N mass fraction larger than 1.74 % and C/N ratios smaller than 28.7, which could lead to a high decomposition rate, according to Palm (1995). The three species (L. grandiflora, I. subnuda, and P. americana, Table 1) with low decomposition rates (Fig. 2) had LG/N ratios in leaves greater than 6.8 (more in agreement with Silva et al. 2008, than Palm 1995). A. sellowiana, E. verna and S. macranthera with high decomposition rates (Fig. 2) had LG/N ratios smaller than 5; together with Z. tuberculosa (also with high decomposition rate) these species had (LG+PP)/N smaller than 10, thus indicating more rapid nutrient release (Palm 1995) with subsequent rapid nutrient recycling. PP mass fractions (Table 1) of all species were higher than 4.4 %, therefore only PP content did not explain the differences in decomposition rate (<0.5 % easy decomposition, Tian et al. 1992) among the species (Fig. 2), in disagreement with Mendonça and Stott (2003) who studied four native trees from the same region and found that high PP and low P contents controlled the decomposition rates of the pruned materials.

As a general rule, decomposition of leaf litter with C/P ratios >300 (L. grandiflora, Z. tuberculosa, P. americana and I. subnuda in the present study; Table 1) is P-limited according to Stevenson and Cole (1999). According to the same authors, the ratio of C to cations (K, Ca, Mg) is not a reliable guide for predicting litter decomposition rate.

According to Duarte (2007), senescent materials of L. grandiflora, I. subnuda, P. americana and S. macranthera had the highest content of nutrients. However, the three first species produced the most recalcitrant litter (highest (LG+PP)/N and LG/N ratios) and low decomposition rate. Pleguezuelo et al. (2009) also found a low decomposition rate for Persea americana, and Palm and Sanches (1990) showed that the decomposition rate of leaves of Inga edulis was lower than that of Erythrina sp. Fresh leaves of Senna spectabilis were rated as high quality litter and Partey et al. (2009) suggested their use as green manure.

We expect (as shown for I. subnuda, Table 2) that nutrients from L. grandiflora, I. subnuda and P. americana are not rapidly released because of the low decomposition rate. This can be a problem for annual crops (especially N-limited cereals), due to a lack of synchrony between N and P release and plant demand (Palm 1995), essential to ensure optimum crop yield and to avoid negative environmental impacts (Grant et al. 2002). However, in the case of perennial crops such as coffee, these nutrients can be slowly released and still used by the crops. There is also a continuous nutrient release due to a continuous supply of material by the trees. Moreover, slow litter decomposition can contribute to greater soil protection against erosion (Silva et al. 2008), which may be high in sloping soils, such as soils at Zona da Mata. Increased soil cover by litter will also reduce nutrient losses by erosion (Palm et al. 2001), leaching, and volatilisation. Therefore, faster decomposition is not always better.

Decomposition and nutrient release

Weight loss data obtained from legumes (Fig. 3) were similar to the CO2 fluxes (Fig. 2). Both data showed high decomposability of litter of E. verna and S. macranthera and low decomposability of litter of I. subnuda (Fig. 2). Similar results were found for Erythrina sp. and Inga edulis by Palm and Sanchez (Palm and Sanches 1990), in Peru. Other authors have also found that Inga sp. has lower decomposition rate and, consequently, lower nutrient release than other species used in tropical agroforestry systems (Leblanc et al. 2006; Schwendener et al. 2007).

Initially, there are greater amounts of easily decomposable materials, such as sugars, amino acids, and proteins, and there is a predominance of recalcitrant materials, such as LG and PP, as the process progresses (Hadas et al. 2004; Matos et al. 2008). Therefore, decomposition rates decline over time Materials with LG contents greater than 15 % and PP contents greater than 3 %, which are associated with N contents less than 2.5 % (Palm et al. 2001), indicate N immobilisation by the soil microbial biomass (Leblanc et al. 2006) during decomposition and a slower N release. Of the three legumes studied, N immobilisation and a slower N release may particularly be occurring with the I. subnuda residues. In agronomic terms, slower N release to the soil can be advantageous, especially in the case of perennial crops, as there is a supply of N throughout the crop cycle. Contrary, in annual crops that are N-limited, such as maize, temporary N immobilisation has a negative impact on crop yield.

The results of the present study indicated that the dynamics of P release were similar to that of N, which was expected because the behaviour of N and P are controlled by similar processes. However, K release was faster, which is due to the non-participation of K (unlike N, P, Ca and Mg) in plant organic compounds. K is an active element in the plant. It exists in an ionic form and is easily released. Calcium and magnesium help make up the middle lamella of the plant cell wall, thus becoming one of the most recalcitrant compounds in plant cells (Vitti et al. 2006; Marschner 2008). Perhaps this explains why the release of Ca and Mg did not follow the negative exponential model.

Biological nitrogen fixation

Overall, percentages of nitrogen fixation were low (17–23 %). This low estimate for N fixation is a consequence of small differences in δ 15N values between the legumes and the reference plant species. As Gehring and Vlek (2004) noted, in cases where the δ 15N values between legumes and reference plants are small, quantification of N2-fixation is difficult. Högberg (1997) suggested that the difference should at least differ by 5 ‰ in order to estimate biological nitrogen fixation with some accuracy. Thus, BNF may have been underestimated in the present study.

Moreover, the choice of reference plants is important for accuracy. It has been reported that δ 15N values vary with soil depth with lower values in the layers between 0–10 cm and higher values in deeper layers (Ledgard et al. 1984; Morais et al. 2006). Both reference species used in the present study are herbaceous and have more shallow root systems compared to the legume trees that have deeper root systems. It was observed in the field that fine roots of S. macranthera are mainly found below 60 cm. Thus, the difference in root systems could also contribute to underestimating the BNF.

According to the literature, species of Senna do not fix nitrogen (Giller 2001) and S. macranthera does not form nodules (De Faria et al. 1984; Barberi et al. 1998). If we treat S. macranthera as non-fixing, the δ 15N values of the reference plants would even be lower, hence even further lowering our estimate of BNF in these agroforestry systems. Using only S. macranthera as reference would have resulted in the conclusion that BNF is not taking place. However, the δ 15N values of the two other reference plants suggest BNF by S. macranthera. This species also had high N mass fractions, both in green and senescent leaves (Duarte 2007; Jaramillo-Botero et al. 2008), which were not different from those of the two other legume species (Table 1) for green leaves; for senescent leaves, for the three Fabaceae, N mass fractions varied between 21.4 and 24.6 mg g−1 (Duarte 2007). One explanation for high leaf N is the presence of different mechanism for N acquisition in S. macranthera (Bryan et al. 1996), for example N scavenging, such as is described for S. spectabilis, an African species (Livesley et al. 2002). Such scavenging might explain the high leaf N mass fractions of S. macranthera, but so far no mechanism for scavenging is known.

Hoewever, Senna, together with the genera Cassia, Melanoxylon and Chamaecrista, belongs to the monophyletic Cassia clade (Bruneau et al. 2008). The Brazilian Melanoxylon and several Amazonian species of Chamaecrista are known to nodulate and fix nitrogen (Naisbitt et al. 1992; De Faria et al. 2010) and it may be noteworthy that Roggy and Prévost (1999) mentioned the occurrence of nodule-like structures in S. quinquangulata. That latter species is very closely related to S. macranthera (Marazzi et al. 2006) and it would therefore be important to assess the capacity for BNF in Brazilian species of Senna.

When considering the annual litter production by these legumes, their contribution to the N cycle (even at low percentages of BNF) can be substantial, especially for S. macranthera and I. subnuda. According to Duarte (2007), the annual leaf litter production per tree of E. verna, S. macranthera, and I. subnuda is 4.0, 17.7, and 32.8 kg, respectively. Considering their N mass fractions, each S. macranthera tree and I. subnuda tree would contribute ~60 and 140 g N year−1, respectively, due to BNF. Family farmers apply ~40 g N year−1 plant−1 of coffee with a NPK formulation of 20-05-20.

Conclusion

Diversification of tree species in agroforestry systems enables input of diversified organic material. Leaf litter of different species has different characteristics of decomposition rate and nutrient release. Leaves of I. subnuda, L. grandiflora, and P. americana are more difficult to decompose than those of A. sellowiana, E. verna, S. macranthera, and Z. tuberculosa. These differences in decomposition correlate with and are likely caused by higher lignin/N ratios and higher (lignin+polyphenol)/N ratios of leaves of the first group of species. Among the legumes, there was greater nutrient release in leaves of E. verna and S. macranthera than leaves of I. subnuda. There were only small differences in BNF among the legumes with greater N fixation in E. verna and I. subnuda than in S. macranthera.

This diversification can contribute to the maintenance of different ecosystem services provided by litter and soil organic matter, resulting in potential improvements in physical, chemical, and biological soil quality and, consequently, in ecological interactions.

Knowledge of decomposition and nutrient release dynamics of species can contribute to decision making by farmer families about agroforestry system design as it allows for spatial arrangements where species that produce slower decomposing litter (important for better soil protection and organic matter formation) can be combined with species that produce litter of rapid decomposition (which is important for faster nutrient release).