1. Introduction

With economic growth, urbanization and changing diets, world demand for plant-derived oils and their derivates is soaring. Oils are used inter alia in the food, feed and cosmetics industry, and increasingly as a biofuel. Soybean is one of the major booming oil crops. In addition to oil, it offers a very valuable by-product, i.e. protein-rich soy meal, a raw material for animal feed. Increasing global demand for animal products puts pressure on arable land, given the fact that at least two but often more kilograms of grains are needed to produce one kilogram of meat, i.e. the so-called feed meat ratio (Keyzer et al., 2005), and that an array of environmental problems is looming due to the expanding livestock sector (FAO, 2006). The demand surge largely stems from China. Table 1 shows a ninefold increase in soy imports in 10 years, versus an insignificant home production. This was triggered by China's 2002 WTO membership, which put an end to border tariffs and boosted trade (Van Berkum et al., 2006). Increasing demand has been met with a supply response that is particularly strong in Brazil and Argentina. Country statistics show that in 2004, Brazilian production exceeded 50 million Mg, twice the amount realized in 1997. Area and production increases were particularly strong in the period 2001–2005, following a favorable devaluation of the national currency, the real. Soybean exports in 2004 earned Brazil over 10 billion US $, against 4.2 billion US $ in 2000.

During the 1960s, soybean cultivation was concentrated in the three states of the Southern Region of Brazil: Rio Grande do Sul, Santa Catarina and Paraná. Later on, expansion to the Central Cerrado Region took place once varieties had been developed that were adapted to low latitudes. Currently, 60% of soybean production is concentrated in the Cerrado Region. The southern area is characterized by smallholder farmers, mostly organized in cooperatives, whereas the central area is characterized by large holdings, with very high levels of mechanization, and mostly organized in large private groups. In addition, climate is more stable in Central Brazil. The latest soybean expansion has taken place in the northern Legal Amazon. Soybean production in the North represents only a small fraction of the currently planted area, but is expanding.

Recognizing that soybean (i) is a highly important commodity for the Brazilian economy, (ii) contributes to the conversion of forest and savanna land, and (iii) travels long distances, ending up in many different foods and feeds worldwide, there is a need to better understand the dynamics of the soybean sector, and how economic and environmental targets can be realized simultaneously. In this study, nitrogen (N) is used as a marker. The important role of N for life, for growth of crops and as an essential element in aminoacids/proteins in our food is obvious. The final judgment is about the trade offs between overloading and loss of N versus the benefits in our food of plant and animal origin. Many studies have looked into the N cycle and the flow of N in the food system, at the global scale ( [Galloway and Cowling, 2002] , [Galloway et al., 2004] , [Galloway et al., 2007] and [Smil, 2002] ), or the national scale (e.g., Antikainen et al., 2005). In this paper we look more specifically into the role of N from soybean in the food system. Soybean is a leguminous crop, capable of fixing atmospheric N through a symbiosis with Rhizobium bacteria in the plant's root nodules. Soybean crops therefore receive no or very small amounts of N fertilizer as a starter.

The objective of this study is to quantify N stocks and flows in and between the compartments of the Brazilian soybean chain, and to find out where in the process N either remains inside or disappears from the food chain. These compartments include forest and savanna conversion, soybean cultivation, transport and processing, animal and human consumption, and waste disposal (Fig. 1).

2. Materials and methods

2.1. Spatio-temporal database

In Brazil, soybean is grown in large and small holdings, and both as a monoculture and in annual rotations with maize, wheat and other crops. For characterizing the soybean farming systems, spatially explicit information was used. To show trends and avoid outliers, average data were taken for the years 1993–1995, 1998–2000, and 2003–2005. Harvested area and production data were obtained from municipality statistics (IBGE, 2006) and related to the country-wide spatial municipality database (n = 5564 for 2003–2005). Besides production data, several spatial data sets were necessary for the N balance calculation. Soil properties were taken from 1:5-million-scale SOTER and WISE databases (FAO et al., 1998; [7] and [8] ). Texture, bulk density, total N and total carbon content of the 0–20 cm soil layer were included for calculation of the N balance. Rainfall data of the CRU TS 2.10 database were used, at a resolution of 0.5 degrees (Mitchell and Jones, 2005). For the calculation of erosion, the slope gradient was required, which was derived from the GTOPO30 digital elevation model (USGS, 1996), with a resolution of 30″ (approximately 1 km). The maps were overlaid in a Geographic Information System, and the mean soil properties, rainfall and slope were calculated for each municipality. The datasets allowed the mapping of soybean land for each time period (Fig. 2a–c), as well as the particular system, i.e. soybean monoculture, soybean/maize and soybean/wheat (Fig. 3a–c).

2.2. Conversion

The expansion of soybean land happens at the expense of pastures, which in turn farmers developed from Cerrado savanna and now increasingly from more densely wooded savanna. Data on decreasing forest and savanna area cannot, however, be fully explained by soybean expansion. To calculate N flows under conversion regimes, a literature review was made. Removals of N are in timber, firewood, in slash and burn of remaining above-ground biomass, and in soil and litter. Land clearing methods range from selective cutting to the destructive savanna clearing system, where a correntao, a giant chain is attached between two bulldozers that move jointly, uprooting not only all vegetation but also topsoil and litter. The totality of this material is then burned (Van Gelder and Dros, 2005).

Mielniczuk et al. (2003) compared data on soil carbon inputs and stocks in long-term experiments in Sweden (Paustian et al., 1992) and in the South of Brazil (Lovato, 2001). In equilibrium, the decomposition rate was 1.05% year⁻¹ in the temperate soil and 3.2% year⁻¹ in the subtropical soil. The carbon to nitrogen ratio of organic matter of Southern Brazilian soils is on average 12.5:1 (Jantalia et al., 2006). A woodland savanna (Cerrado stricto sensu) stores 229 t of C ha⁻¹ in biomass plus soil (1 m depth; Roscoe, 2005). Texture is also an important factor determining carbon stocks in Cerrados soils. Tognon et al. (1998) found a strong linear relationship between texture and carbon stored in soil, varying from 75–87 t of C ha⁻¹ in sandy soil to 182–188 t of C ha⁻¹ in clayey soils. Considering a C:N ratio of 12:1, this would represent an N stock of about 7 t ha⁻¹ in sandy soils and about 15 t ha⁻¹ in clayey soils. Pastures normally develop higher soil N contents than forest and savanna. Neill and Davidson (1999) showed that, from 29 studies reviewed, 19 forest soils had increased their organic matter content in surface when converted into pastures. Other literature data revealed that values for carbon in the upper 1 m of Brazilian forest and savanna soils range from 100 to 200 t ha⁻¹ ( [Matson et al., 1987] , [Cerri et al., 1991] and [Da Silva, 2004] ). Carbon in vegetation ranged from 30–80 t ha⁻¹ in savanna to 140 t ha⁻¹ in forest (Graca et al., 1999). Total N losses in the forest system following clearcutting, based on review data by Matson et al. (1987) are in the order of 1000 kg N ha⁻¹, but data from Cameroon (Kanmegne, 2004) and Madagascar (Brand and Pfund, 1998) show losses as high as 3000 and 5000 kg ha⁻¹ for forest. Bouwman (1995) found in La Selva, Costa Rica, soil C and N stocks of 81,000 and 7600 kg ha⁻¹, respectively. Soil N losses due to decomposition of soil organic matter following forest conversion to pasture were 24% after 3 years, and after 5 years 40%. Schlesinger (1986) gives values of 21% loss of soil C when converting land from tropical rainforest to agriculture (based on 19 sites). Based on this survey, it is assumed that forest and savanna have a ‘starting N stock’ in vegetation and soil, whereas the situation prior to soybean cultivation is the ‘final N stock’, which is the input value for the ‘Cultivation’ phase (Fig. 1). Pasture is considered an intermediate phase. Loss rates are then estimated on the basis of actual deforestation data, and on the percentage loss that can be attributed to soybean expansion. Based on the literature study, and assuming a soil C:N ratio of 12:1, and a relevant topsoil thickness of 20 cm, an initial N stock in forest and savanna systems of 8000–12,000 kg N ha⁻¹ seems plausible, savanna being at the lower end and forest on the upper end. Moreover, it is assumed that 25–50% of forest/savanna system N is lost in the process of conversion to pasture and then to soybean.

2.3. Cultivation

2.3.1. Nutrient balance

The soybean system was analysed, largely using the NUTMON nutrient balance approach ( [Stoorvogel et al., 1993] , [FAO, 2004a] and [Lesschen et al., 2007] ). Table 2 lists the relevant flows. The nutrient balance calculation involved a full cropping cycle of soybean (monoculture) or soybean + maize or wheat (rotation), and the combined balance was thus calculated using the planted areas of each soybean system.

2.3.2. Tillage systems

Agriculture in the 1960s and 1970s in the South was heavily mechanized, leading to strong soil degradation ( [Mielniczuk et al., 2003] and [Jantalia et al., 2006] ). There was a widespread concept among farmers that well-managed soils would have completely clean and smooth surfaces, with residue incorporation and pulverized soils (Jantalia et al., 2006). Comparing the dynamics of soil organic matter in a long-term experiment in the South of Brazil described by Lovato (2001), Mielniczuk et al. (2003) estimated that no-tillage reduced the decomposition rate from 3.2% year⁻¹ to 1.7% year⁻¹. In the Cerrado region of Bahia State, Silva et al. (1994) sampled 220 topsoils (0–15 cm) from soybean fields under conventional tillage and monoculture. The authors observed severe losses from 41% (clayey soils with >30% clay) to 80% (sandy soils with <15% clay) of the initial soil organic matter contents after 5 years of cultivation. However, Freitas et al. (2000) and Roscoe and Buurman (2003) did not observe changes in the SOM stocks (0–40 cm) of a clayey Dark Red Latosol after 25 and 30 years of maize–bean successions with conventional tillage. Llilienfein and Wielcke (2003) reported no significant changes in C content of a clayey Oxisol after 12 years of maize–soybean rotation under conventional tillage. The presence of a high-residue crop such as maize in these experiments may partly explain the persistence of organic matter in these soils. However, the most important factor seems to be the high protection of organic matter in clayey Cerrado soils, which are rich in iron and aluminum hydroxides (Resende et al., 1997).

2.3.3. IN1 Mineral fertilizer

The fertilizer use by crop study for Brazil (FAO, 2004b) estimated fertilizer consumption values per crop and per region (Table 3). These values were estimated for 2002 and were multiplied with a correction factor, based on the relative difference between the fertilizer consumption of Table 3 and the IFA national fertilizer statistics for the calculation years 1993–1995, 1998–2000 and 2003–2005. Total N consumption for these periods was derived from the FAO fertilizer statistics, the values being 1,148,600 Mg N for 1993–1995, 1,738,100 Mg N for 1998–2000 and 2,415,400 Mg N for 2003–2004.

2.3.4. IN2 Organic inputs

Although Brazil has the largest livestock herd in the world, as well as large numbers of pigs, sheep and poultry, application of organic manures is limited to horticulture and perennial crops located close to the production sites. In the case of grain crops, the use of organic fertilizers is not very common (FAO, 2004b). Therefore, the organic inputs for soybeans, maize and wheat were set at zero.

2.3.5. IN3 Wet and dry deposition

Trebs et al. (2005) measured dry and wet N deposition for a remote pasture site in the Amazon Basin, and found 7.3–9.8 kg ha⁻¹ year⁻¹. Krusche et al. (2003) also estimated an N deposition rate of about 9 kg ha⁻¹ year⁻¹. According to two chemistry transport models ( [Collins et al., 1997] and [Dentener et al., 2006] ), N deposition for Brazil amounts to 5–10 kg ha⁻¹ year⁻¹. Based on this, input through N deposition was set at 8 kg ha⁻¹ year⁻¹.

2.3.6. IN4 Biological N fixation

Biological N fixation (IN4, kg N ha⁻¹ year⁻¹) is the most important input for a leguminous crop such as soybean. Based on studies of Zotarelli et al. (1998), [Alves et al., 1999] and [Alves et al., 2005] , Hungria et al. (2006) and Araújo et al. (2006), Rhizobium-based N fixation by soybeans was set at 80% of the N in the above-ground plant parts. For non-symbiotic N fixation (based on the input by free living bacteria), the following equation applies (FAO, 2004a):IN4=0.5+0.1×√rainfall

2.3.7. OUT1 Harvested crop parts

Production and harvested areas for soybean, maize and wheat were derived from the PAM (Agricultural Municipalities Production) database (IBGE, 2006). The N content of the harvested crop parts for soybeans was set at 58 g kg⁻¹ harvest product, based on studies by Zotarelli et al. (1998), [Alves et al., 1999] and [Alves et al., 2005] , Hungria et al. (2006) and Araújo et al. (2006). For maize and wheat, the values of FAO (2004a) were used, i.e. 17 and 22 g kg⁻¹, respectively.

2.3.8. OUT2 Removed crop residues

The removal of crop residues from the field is not a common practice in soybean farming systems in Brazil, as almost no alternative uses for them, such as stall-feeding, occur. Therefore, we assumed that all crop residues remain on the field and, therefore, OUT2 was set at zero.

2.3.9. OUT3 Leaching

For N leaching (OUT3, N in kg ha⁻¹ year⁻¹), the regression model of De Willigen (2000) has been used:where P is the annual precipitation (mm); C the clay (%); L the layer thickness, i.e. rooting depth (m); F the mineral and organic fertilizer nitrogen (kg N ha⁻¹ year⁻¹); D the decomposition rate (% year⁻¹); NOM the amount of nitrogen in soil organic matter (kg N ha⁻¹); U the uptake by crop (kg N ha⁻¹ year⁻¹).

The first part of the regression equation determines which fraction of mobile N will be leached, and the second part determines how much mobile N is available. To prevent overestimation of leaching, the first part of the equation was maximized at 1. In the case of soybeans, only the uptake from the soil is taken into account (20%) and not the part which is derived from N fixation. For rooting depth, values of Allen et al. (1998) have been used, which are 0.6 m for soybeans, 0.9 m for maize, and 1.0 m for wheat. Decomposition rate was set at 3.2% year⁻¹ for conventional tillage and 1.7% year⁻¹ for no-tillage (Mielniczuk et al., 2003).

2.3.10. OUT4 Gaseous losses

The regression model developed by FAO (2004a) has been used with an additional factor for crop residues. NH₃ volatilisation from field crop residues is set at 4%, which is the volatilisation rate of some ammonium-based fertilizers (Bouwman et al., 2002).OUT4=(0.025+0.000855×P+0.01725×F+0.117×O)+0.113×F+0.04×Rwhere P is the annual precipitation (mm year⁻¹); F the mineral and organic fertilizer nitrogen (kg N ha⁻¹ year⁻¹); O the organic carbon content (%); R the nitrogen in remaining crop residues (kg N ha⁻¹ year⁻¹).

2.3.11. OUT5 Erosion

The RUSLE model was used (Renard et al., 1997), in combination with the spatial rainfall and soil data discussed in Section 2.1. The RUSLE is expressed as:A=R×K×LS×C×Pwhere R is the rainfall erosivity factor (−); K the soil erodibility factor (kg m⁻²); LS the slope length and slope steepness factor (−); C the crop management factor (−) and P the erosion control practice factor (−).

The R-value was based on the annual erosivity map for Brazil by Da Silva (2004). The K-factor was based on the nomographs of Wischmeier and Smith (1978), but a simplified version was used because of a lack of soil permeability and soil-structure data. The LS factor was calculated using the slope gradient and a default slope length of 22.13 m. According to Morgan (1995), the C-factor for soybeans ranges from 0.20 to 0.50. As the C-factor is significantly influenced by tillage, a distinction was made between conventional tillage and no-tillage. Erosion on no-tillage fields is on average only 25% of erosion under conventional tillage (Roscoe, unpublished data). As a consequence, a C-factor of 0.40 was used for conventional tillage and 0.10 for no-tillage. Table 4 gives the percentages of agricultural area under no-tillage (Roscoe, unpublished data). Besides no-tillage management, no other erosion control practices are used on a large-scale, hence the P-factor is 1. Finally, the loss of N was calculated by multiplying the annual soil loss by the soil N content and an enrichment factor, which was set at 2.3 (FAO, 2004a).

2.4. Transport and processing

Once harvested, the soybeans move in different directions (Fig. 4). The part undergoing processing inside Brazil (56%) is crushed to produce soy oil and the protein-rich by-product soy meal. Upon crushing, the chain is only continued for soy meal, soy oil not containing N anymore. Then, 13% of the soy meal is used as feedstuff in Brazil, whereas 30% is exported. Part of Brazilian animal produce is exported (4%). Further assumptions necessary for the calculations of N destinations are given in Table 5.

2.5. Consumption

For the (animal and human) consumption phase, a series of assumptions was made relating to feed conversion efficiency in animals, slaughterhouse efficiency, and human consumption characteristics, including wasting of food. The N losses are in fact not all ‘losses’. Considerable percentages leave the ‘useful product’ chain but can be recycled. This potential re-use is, however, often far away from the site where soybeans are produced, hence causing an accumulation of N at consumption sites. The assumptions are summarized in Table 6.

2.6. Human waste disposal

The final part of the chain looks into the human excreta. In our calculations we consider all N in human waste as potential emission to the environment, excluding the N removed in wastewater treatment systems. This is because there is no quantitative information on the fate of the N for other systems such as latrines, septic tanks and other less institutionalized human waste disposal systems. For example in China, many of the human excreta have been recycled, and used to fertilize agricultural systems until the late 1970s (FAO, 1977). However, it is not well known if this practice is still important at present (Zhu and Chen, 2002). Hence, by ignoring recycling of human excreta, we may overestimate the potential emission of human N to the environment.

Depending on the country, part of the population is connected to sewerage systems, and part of the sewage water is treated in wastewater treatment plants. The N removal in treatment installations depends on the type of installation. Assumptions are summarized in Table 7.

2.7. Uncertainty analysis

An uncertainty analysis was applied to gain insight in the accuracies and possible ranges of the results. For the cultivation phase, we followed the procedure of Lesschen et al. (2007), in which uncertainties were first calculated or estimated for each flow at municipality level (Table 8). Since the results were aggregated to regional and national level the uncertainties have to be adjusted for spatial correlation. Spatial correlation occurs because part of the input data is spatially dependent, e.g., soil and rainfall data. The spatial correlation coefficient was estimated for each nutrient flow and ranged between 0 and 1.

For Phase 3 (transporting and processing) and Phase 4 (consumption), a simple sensitivity analysis was done by varying a number of parameters by 25% and calculating the effect on the soybean-derived human N intake.

3. Results

3.1. Conversion

Fig. 5 shows the cumulative area of deforestation, taken from the National Institute of Space Research (INPE), and the total soybean area. The area increased from 11.3 million ha in 1993–1995, to 20.1 million ha in 2003–2005. This increase occurred mainly in the central and northern regions. Until 2000, there is no relation between deforested area and soybean area, but between 2000 and 2004, loss of forest is approximately equal to expansion in soybean area. In reality, the conversion rate of forest/savanna to pasture may, in these years, be approximately equal to the conversion rate between pasture and soybean land. However, these data do not fully explain whether most of the new soybean areas are savanna-derived or forest-derived conversions. Also, soybean expansion has always taken place in land under other arable crops as well.

Taking the rather constant deforestation rate of 2 million ha year⁻¹ (Fig. 5), and attributing 50% of deforestation to soybean cultivation, N losses during Phase 1 are in the order of 2000–6000 million kg year⁻¹ on the basis of the N losses presented in Section 2.2, with 1993–1995 at the lower and 2003–2005 at the higher end of the range. It is difficult to attribute exactly the conversion to vegetation types and time periods, but it is obvious that with the advance of soybean into the central and northern parts of the country, soybean expansion takes place increasingly at the expense of forest.

3.2. Cultivation

Soybean production data have been summarized in Table 9. OUT1 is given for the overall soybean-growing area, also including outputs in maize and wheat. The final row gives the total N production through harvested soybeans, assuming N content in beans of 5.8%. The uncertainty of the total amount of N from soybean production is only 7%, based on the 22% uncertainty for OUT1 and a spatial correlation factor of 0.3.

N flows were calculated for the different regions and different soybean farming systems portrayed in Fig. 2 and Fig. 3. Fig. 6 shows that surface N balance deficits decrease with time and become surpluses, primarily because of the increased importance of no-tillage, reducing leaching and erosion. Average uncertainty of the N balance is ±9 kg N ha⁻¹, ranging between ±6 kg N ha⁻¹ for the Southeast region and ±16 kg N ha⁻¹ for the North region as calculated for the 1998–2000 period. Soybean is estimated to fix 80% of its needs, and since the crop residues remain in the field, the values of OUT1 and IN4 are about equal (Fig. 7). Based on the total soybean area and N balances calculated for the three time periods in the “Cultivation” phase, there is a total loss of 182 million kg N year⁻¹ in 1993/1995, a loss of 15 million kg N year⁻¹ in 1998/2000, and a gain of 62 million kg N year⁻¹ in 2003/2005.

3.3. Transport, processing and consumption

Based on the assumptions listed in Table 5 and Table 6, flows were calculated during processing, transport and animal and human consumption. Fig. 8 shows the values and flows representing the relevant processes for the three periods. Starting stocks correspond with values given in Table 9, i.e. the output of ‘Cultivation’. About 20% of total N entering Phases 3 and 4 ends up as animal products that are consumed by humans. This percentage decreased slightly between 1993–1995 and 2003–2005 (from 21.1 to 19.8%). In addition, Table 10 shows total N ingested by humans per country/region. Adding up the rows gives totals of 300, 377 and 583 million kg N year⁻¹ for the three periods.

Table 11 provides a sensitivity ranking for Phases 3 and 4, i.e., on final soybean-derived N intake from meat due to variation of the different input parameters by +25%. It is clear that the feed N conversion in livestock production is the major uncertainty in Phases 3 and 4. This is related to the large fraction of soy meal that goes into livestock production, and the generally low efficiency of feed conversion to meat.

3.4. Waste disposal

Table 12 shows the fate of consumed N. Total consumption (583 million kg N year⁻¹) is the output value of Phase 4 for 2003–2005. European Union (EU) households have by far the largest percentage of sewerage connections (84%). However, even with advanced wastewater treatment in Europe with 51.5% of N in influents being removed, 48.5% of the N entering sewage systems still reaches surface waters. If we consider all households in the EU including those lacking sewage connection, 57% of the soybean N consumed is not removed in wastewater treatment.

In Brazil and China less than 10% of the soybean N is removed from sewage water, hence more than 90% is not treated and is a potential emission to the environment, since far fewer households have a sewerage connection than in the EU. The fate of N in non-sewerage outlets, ranging from latrines to septic tanks, has not been considered explicitly due to lack of data. Human excreta that are collected may be used to fertilize agricultural fields, particularly in China. Hence, the potential N emission to the environment may be an overestimation.

4. Discussion

4.1. Limitations of the study

Given the booming market, expansion of the soybean area in Brazil seems economically viable, but it has a social and environmental price. Many negative externalities of the soybean sector are not covered in this study, including plant and animal biodiversity loss, carbon emissions, changing air circulation and increased drought occurrence, unlawful land acquisition, forced migration of forest inhabitants, rural poverty and low employment rates.

On ‘Conversion’, studies exist that look at vegetation carbon, and at soil carbon available in forest and savanna. Fewer look at N, let alone both vegetation N and soil N. Some studies take soil N down to a depth of 1 m, others just consider 0–20 cm. Land can be cleared in many different ways. The fully fledged bulldozer-assisted clearing of cerrado mentioned by Van Gelder and Dros (2005) is much more destructive than the selective cutting that takes place elsewhere. For this study, ‘clearcutting’ is considered to take place, although the actual process may include periods of partial clearing as well as periods under pasture, which often show increases in soil C and N over time, particularly when managed well.

On ‘Cultivation’, N is clearly not a limiting factor as soybean is a leguminous crop, but other nutrients may well be or become limiting. FAO (2004b) shows that soybean in Brazil normally does receive P and other fertilizers, next to tiny amounts of N. These other nutrients are not considered in this study. No attention is paid either to the fact that, currently, most soybean varieties are genetically modified (Brazilian Ministry of Agriculture estimates are 70% of the soybean area under GM soybean in 2005/2006), and based on ‘Round-up ready’ herbicide use, which are imperative for adopting minimum tillage practices. A social externality not taken into account is that the labour/land ratio of large-scale soybean production in central Brazil is very low (1 on 400 ha) compared with small-scale agriculture in the south (80 on 400 ha) (Van Berkum et al., 2006).

Another issue outside the scope of the current study is the increased use of soy oil as a biodiesel. A new market was opened to soybean oil in 2006/2007 as a consequence of the Brazilian National Biodiesel Programme. The programme has allowed the inclusion of 2% of biodiesel in the diesel from petroleum since January 2006. This proportion will become compulsory in 2008 and will be increased to 5% in 2013. The demand for vegetable oil in 2008 is estimated about 1 million t and, in 2013, 2.5 million Mg. According to Petrobrás projections, in 2008, the demand for vegetable oil in this process will be about 500,000 Mg. Therefore, the energy sector will absorb about 1.5–3.0 million Mg of vegetable oil in the coming 2–5 years.

On ‘Transport’, a recently opened port in Santarem and the completion of the road between Cuiaba and Santarem opens up new opportunities for agricultural development, but it may equally pose substantive social, environmental and production risks. Van Berkum et al. (2006) show that export and import figures can differ markedly. For example, ITC/WTO data show that on soy meal trade between Latin America and The Netherlands, Brazil exported 4 million Mg of soy meal, but The Netherlands only imported 1.3 million Mg. Similarly, Argentina export data show 2.3 million Mg shipped to The Netherlands, but the Dutch statistics show imported soy meal from Argentina to have amounted to just over 1.1 million Mg. Although ships do at times change destination (e.g. to France, Belgium or Germany) while moving overseas, the differences are significant.

On ‘Consumption’, there are several causes of uncertainty. The most uncertain parts of the consumption chain are the feed N conversion in livestock production systems and food spoilage and waste. The process that determines a large part of the N loss from the soybean chain is the feed conversion. A small variation causes a large deviation in the end results (Table 11). Estimates for feed N conversion are from large-scale studies, and differences between the estimates for Brazil and China (21% efficiency for pigs) and Europe (30%) indicate that management has an important effect on N lost from the soybean chain. The results shown here show greater efficiency of N capture by animal production than reported previously ( [Smil, 1999] and [Smil, 2001] ). In the previous analysis, the effect of N recycling within the system was not considered, and older data were used to estimate the efficiency of feed N utilization by animals. Since livestock production systems are rapidly changing, particularly in developing countries, where modern units with high levels of management replace existing ones, efficiencies may also increase rapidly in countries like Brazil and China.

In our study roughly 20% of the N in the soybeans is actually consumed as meat by humans. Since we consider only pork and poultry, our estimate exceeds the 14% estimated by Smil (2002) for all livestock including milk and beef cattle.

Estimates of food availability for human consumption from food balance sheets are generally used as an approximate level of food actually consumed. This may work reasonably well in developing countries, but in developed countries this approach can overstate the level of consumption because the amount of food spoilage and waste in catering establishments is rather high. The amount of food actually consumed may be lower than the quantity shown in the food balance sheet depending on the degree of losses of edible food and nutrients in the household, e.g. during storage, in preparation and cooking (which affect vitamins and minerals to a greater extent than they do calories, protein and fat), as plate-waste or quantities fed to domestic animals and pets, or thrown away. Household survey data may prove useful in such instances for adjusting the waste component in the food balance sheet.

On ‘Waste Disposal’, other defecation destinations than sewerage and water treatment are not explicitly dealt with in this study. They may be equally or more constraining to public health, but can be considered as potential emissions of N compounds into the environment.

Quantification of the fate of the N that is lost from the chain and assessment of the environmental impact is outside the scope of this paper. This would require a completely different approach like life cycle analysis (LCA). In actual fact, results of our study may serve as a basis for LCA of the soybean food chain. In this respect, the intention is also not to provide an exhaustive overview of N stocks and flows as was done for Finland by Antikainen et al. (2005).

4.2. Opportunities to make the chain more sustainable

Major reductions in N losses are possible in Phases 1, 4 and 5 (Fig. 1). Although no research was done on solutions, some opportunities stand out that may contribute to striking a more sustainable balance between economic and environmental goals.

On ‘Conversion’, improvements and solutions are more in the field of forest management policies, law enforcement, and human rights protection than in agronomy. Payment for environmental services along the lines of Kyoto Protocol mechanisms may be a way to harmonize economic, environmental and social goals at national and local scale.

As to ‘Cultivation’, Brazil's National Agricultural Research Institute (Embrapa) has indicated that considerable strides can still be made to raise productivity on existing soybean land, to further spread the practice of no-tillage, to restore and cultivate degraded pastures, and to rotate pastures and soybean. Estimates on ‘degraded’ land that could be converted to soybean are between 10 and 30 million ha.

Although trade generates economic benefits, soybeans and soy meal face ‘Transport’ over very wide distances before entering the consumption cycle. Atmospherically fixed N from Brazil ends up in dung piles in Western Europe and China. In the Netherlands, for example, this has for some decades now been regarded as a major environmental problem.

The consumption chain in the EU (soy meal–animal feed sector–farms–slaughterhouse–supermarkets) is increasingly vertically integrated. This makes the sector more efficient, but also a more powerful force vis-à-vis producers and consumers. Reduced consumption of animal products is a way to reduce the tension on the soybean chain. Although not desirable from an economic viewpoint, livestock production with a 20–30% N conversion is quite inefficient when converting plant N to proteins in meat, milk or eggs. Price regulations and campaigns to influence consumer behavior represent a possible way to change human diets.

On ‘human waste disposal’, research and dissemination of good hygienic practices can help in keeping excess N out of surface waters.

On a more general note, the instrument of round tables is used with some degree of success, both for oil palm as well as for soybean. The Round Table on Responsible Soy includes private and public sector stakeholders from Brazil and export countries, social and environmental NGOs, and forest inhabitants. In The Netherlands, a ‘Soy Coalition’ has managed to table the adverse social and environmental aspects of soybean production in parliament. Some major food industries have decided to track and trace soy meal using more specific and measurable indicators to ascertain they are selling products grown in a socially and environmentally benign way.

5. Conclusions

• 1.

  Brazil is a major soybean producer and exporter and is expected to become even more powerful. In 1993–1995, 1998–2000, 2003–2005, soybean production in Brazil was 24.4, 31.8 and 50.8 million Mg, respectively. About 75% is exported. Soybean exports cover 10% of the country's total export income; EU and China are major importers.

• 2.

  Soybean is grown in monocultures on very large farms (>1000 ha) in the Central parts of the country, and on family farms in the South, in rotation with maize and wheat. In the North, soybean cultivation is relatively new. The Central part now is the major producer, having overtaken the Southern part in the past 5 years.

• 3.

  Amazon forest and Cerrado savanna has most often been replaced by pastures. These are increasingly replaced by soybean farms. Some parts of the Amazon are converted into soybean fields directly. Soybean has also replaced other arable crops. Its growth can partially be linked to forest and savanna removal.

• 4.

  Estimates (Phase 1) of N losses in forest and savanna that can be attributed to soybean range from 2000 to 6000 million kg of N year−1. This comes on top of biodiversity loss, destruction of living areas, and the creation of increasingly dry conditions due to reduced air circulation on the continent.

• 5.

  Estimates (Phase 2) of N losses in soybean fields were 182 million kg of N year−1 in 1993–1995, but in 2003–2005, there was a gain of 62 million kg of N, due to generally improved practices of conservation agriculture, and due to the N-fixing nature of (the leguminous species) soybean. Other nutrients may be limiting, and/or have negative nutrient balances (P, K and trace elements), but this was not investigated. In addition, the issues of genetically modified soybean (currently estimated at 43%) and the massive use of pesticides and associated health risks have not been dealt with here.

• 6.	Of the total soybean production, 75% is turned into soy meal, 20% into soy oil, and about 5% is lost during processing and transport. As soy oil contains no N, it is no longer relevant for the chain studied here. The soy meal is largely fed to animals, mainly to pigs and chickens.

• 7.	During animal production and human consumption, a series of N losses come to the fore, production of animal manure being the main (recyclable) loss. The N remaining in consumed meat for humans is about 20% of the N available in the harvested soybeans.

• 8.	Of the N excreted by humans into sewerage systems, 79% is emitted into the environment, whereas the complement is removed from wastewater by treatment.

Acknowledgements