We present an approach for providing quantitative insight into the production-ecological sustainability of biofuel feedstock production systems. The approach is based on a simple crop-soil model and was used for assessing feedstock from current and improved production systems of cassava for bioethanol. Assessments were performed for a study area in Mozambique, a country considered promising for biomass production. Our focus is on the potential role of smallholders in the production of feedstock for biofuels. We take cassava as the crop for this purpose and compare it with feedstock production on plantations using sugarcane, sweet sorghum and cassava as benchmarks. Production-ecological sustainability was defined by seven indicators related to resource-use efficiency, soil quality, net energy production and greenhouse gas (GHG) emissions. Results indicate that of the assessed systems, sugarcane performed better than cassava, although it requires substantial water for irrigation. Targeted use of nutrient inputs improved sustainability of smallholder cassava. Cassava production systems on more fertile soils were more sustainable than those on less fertile soils; the latter required more external inputs for achieving the same output, affecting most indicators negatively and reducing the feasibility for smallholders. Cassava and sweet sorghum performed similarly. Cassava production requires much more labour per hectare than production of sugarcane or sweet sorghum. Production of bioethanol feedstock on cultivated lands was more sustainable and had potential for carbon sequestration, avoiding GHG emissions from clearing natural vegetation if new land is opened.
The importance of biofuels is increasing rapidly, mainly due to renewable energy targets in transport set by several governments (e.g. US Congress, 2005; European Parliament, 2009). One country considered promising for biomass production is Mozambique, due to its relative abundance of land resources, favourable environmental conditions and low population density (Batidzirai et al., 2006); only one-sixth of its 30 million hectares of arable land are currently cultivated (Arndt et al., 2009). Although it is one of the fastest growing economies in sub-Saharan Africa and poverty rates are dropping, Mozambique is still among the world's poorest countries (Schut et al., 2010). Biofuels investment may help enhancing growth and poverty reduction, although food prices may also increase due to competition for land and labour (Arndt et al., 2009). Mozambique is listed as a country where there is high risk of deteriorating food security due to increased food prices (FAO, 2008). Apart from pressing policy questions that these issues raise, the environmental implications of large-scale biofuels production also matter for the sustainability of an emerging biofuels sector and of Mozambican agriculture in general. Changes in relative or absolute acreage of different biomass crops may impact production-ecological sustainability in different ways. Two examples: first, cultivation of sugarcane (Saccharum officinarum L.) relies on irrigation in Mozambique, and there are competing demands for the available water resources (Cheesman, 2004). On the other hand, sugarcane with so-called green cane trash blanketing contributes to carbon (C) sequestration and improved soil quality through increase in soil organic C (SOC) content (Razafimbelo et al., 2006; Robertson & Thorburn, 2007). Secondly, cultivation on slopes of cassava (Manihot esculenta Crantz), a starch crop that may be used for producing bioethanol (cf. Dai et al., 2006), generally causes more erosion on an annual basis than other crops grown under the same circumstances (Howeler et al., 2000). On the other hand, it is a low-cost feedstock and has a pro-poor profile (Econergy et al., 2008; Arndt et al., 2010).
In this study, we examine the production-ecological sustainability of current and improved cropping and management options in Mozambique. Our focus is on the potential role of smallholders in the production of feedstock for biofuels. We take cassava as the crop for this purpose and compare it with feedstock production on plantations using sugarcane, sweet sorghum and cassava. Cassava was chosen because of its pro-poor profile (Arndt et al., 2010), its low cost (Econergy et al., 2008) and because it is a crop that is familiar to Mozambican smallholders (Worldbank, 2006). Sugarcane and sweet sorghum were selected due to existing interest from investors (Schut et al., 2010) and scientists, and they are among the officially approved biofuel crops in Mozambique (Conselho do Ministros, 2009).
Cassava is mostly produced by smallholders (Howeler et al., 2000); in Mozambique, smallholders make up 95% of the country's agricultural gross domestic product and few use any fertilizers or other inputs (ca. 4%; Worldbank, 2006). Yields are mostly low (on average 6.5 Mg ha−1 over 2003–2007; FAO, 2010) or, where they are higher, soil fertility is effectively mined (Folmer et al., 1998).
Sugarcane, contrary to cassava, is currently mostly grown on large-scale plantations with a high degree of mechanization where substantial amounts of inputs are normally used, particularly water and fertilizers (Cheesman, 2004). Sweet sorghum has been researched as a sugarcane ‘off-crop’ for sugarcane plantations, since it is often able to produce sugar at times when sugarcane does not accumulate enough sugar for harvesting due to its photoperiod sensitive nature (Woods, 2000). It then presents an opportunity to utilize otherwise idle equipment (e.g. the sugar mill).
Using methods of De Vries et al. (2010), we concentrated on sustainability indicators relating to energy use and the quality of soil and water resources, and their ability to sustain agricultural production. The indicators that we used are relevant within the ‘Principles and Criteria for Sustainable Biofuel Production’ set by RSB (2009) and comprise greenhouse gas (GHG) emissions (Principle 3, GHG emissions), soil organic matter (Criterion 8.a.1.), water productivity (Criterion 9.c.), nitrogen (N) leaching (Criterion 9.d.), soil erosion (Criterion 8.a.1.), employment creation (Criterion 5.a.), net energy yield and N-use efficiency (NUE).
The assessment was conducted for three locations in Mozambique. For smallholder cassava we selected Gafaria (15°43′S, 37°38′E, 600 m a.s.l., sandy clay loam) and Nacuaca (15°41′S, 37°49′E, 500 m a.s.l., loamy sand), two villages in Alto Molocue district, Zambezia province which is an important cassava producing region. For plantation-scale cultivation of sugarcane, sweet sorghum and cassava we selected Dombe village, Sussundenga district, Manica province (19°58′S, 33°23′E, 150 m a.s.l., sand). A new extensive sugarcane plantation has been established here that specifically aims at the production of bioethanol (Principle Energy, 2009). In all locations, soil samples were collected and analysed in the soil laboratory of Eduardo Mondlane University, Maputo for pH, SOC, total N, exchangeable potassium (K), CEC, phosphorus (P) Olsen and soil texture (Table 1).
Table 1. Rainfall and soil characteristics for the three study sites (standard deviations in parentheses)
Sandy clay loam
pH H2O (–)
C (g kg−1)
N (g kg−1)
K (cmol+ kg−1)
H++Al3+ (cmol+ kg−1)
ECEC (cmol+ kg−1)
P (mg kg−1)
Assessed cropping systems and crop management options
For smallholder cassava production, we assessed a current production system without fertilization and improved systems with fertilization and/or residue mulching (Table 2). Introduction of irrigation and agricultural machinery were not considered feasible for smallholders. All fertilizer rates for cassava in Table 2 are sufficient to overcome yield limitations of macronutrients and were formulated based on a simulation exercise carried out before the actual assessment. We estimated that 70% of the cassava residues were available for mulching and the remainder required as planting material.
Table 2. Crop management of the assessed production systems of cassava, sugarcane and sweet sorghum
Crop yield (Mg ha−1)
Fertilization (kg ha−1 of N:P:K)
Mulching of residues
Other agricultural operations
0 : 0 : 0
40 : 150 : 0
0 : 0 : 0
100 : 150 : 200
200 : 300 : 400
470 mm drip
750 mm surface
120 : 40 : 200
170 : 60 : 285
650 mm furrow
380 mm pivot
340 mm drip
650 mm furrow
380 mm pivot
340 mm drip
90 : 60 : 60
In plantations, cassava is usually grown with substantial inputs (cf. Dai et al., 2006). We assessed a production system (Table 2) where sufficient fertilizers are applied for removing yield limitations by macronutrients and included two types of irrigation, drip and surface irrigation. Agricultural operations were assumed to be carried out mechanically, except harvesting.
For sugarcane (Table 2), we assessed two fertilizer rates, three types of irrigation (drip, pivot and furrow) and two residue management strategies (residue mulching, commonly called ‘green cane trash retention’ and residue burning). Mulching is normally combined with mechanical harvesting (Wood, 1991) and residue or trash burning with manual harvesting. Agricultural operations other than harvesting were assumed to be mechanized.
Only one sweet sorghum production system was assessed due to limited data: fertilized, rainfed cultivation (Table 2). Rainfall during the rainy season is normally sufficient to attain a good sweet sorghum yield and if irrigation is available, sugarcane is the preferred crop. In Mozambique, the optimal time for sweet sorghum harvesting coincides with the end of the rainy season and the soil generally is too wet to allow any machinery in-field, hence we assumed manual harvesting. In addition, at that time the leaves are still green and not burnt. Therefore they must be removed before transporting the stalks to the mill (Woods, 2000) and we assumed they are left in the field as mulch.
Response of cassava production to crop management was simulated by a simulation model of the soil-crop system with a seasonal time step: ‘FIELD’ (Field-scale Interactions, use Efficiencies and Long-Term soil fertility Development; Tittonell et al., 2010). This model satisfactorily simulated maize response to application of manure and fertilizers on smallholder farms in Kenya (Tittonell et al., 2008) and has also been applied to e.g. seed cotton and sweetpotato (Tittonell et al., 2010).
A simplified representation of the structure of the crop-soil model and the way it is integrated with the calculation of the sustainability indicators (discussed in detail further on) is provided in Fig. 1. FIELD consists of submodels for crop (‘CROPSIM’) and soil (‘SOILSIM’) that can be either used separately or combined. In this study, SOILSIM was used for all three crops, while we ran CROPSIM only for cassava, due to limited data availability for sweet sorghum and sugarcane. CROPSIM calculates crop production based on seasonal availability of light, water, N, P and K and their interactions. Potential yields are estimated through a radiation-use efficiency approach, while the integrated effect of the relative availability of the other resources on crop productivity is calculated according to a methodology developed by Janssen et al. (1990) for N, P and K. Solar radiation data were derived from Jones et al. (2002). Resource availability in the soil is kept track of by a seasonal bookkeeping approach in SOILSIM. Crop available N, P and K are estimated from soil parameters using functions developed by Janssen et al. (1990); N availability is strongly linked to SOC content which is simulated as the net effect of annual SOC decomposition and the addition of organic resources like crop residues. Crop available N may be further improved by addition of residues and/or fertilizers. To account for the residual effect of P fertilizers, we applied the approach described by Wolf et al. (1987) and Janssen et al. (1987). In their model, a labile and a stable P pool are distinguished; the model calculates the P transfers between the pools, the uptake of P by the crop, and the resulting pool sizes. Regarding soil K, we assumed that after addition of fertilizers, exchangeable K returns to its previous ‘base level’ within 1 year after application, similar to findings of e.g. Cox & Uribe (1992).
Model parameters specific for cassava, e.g. minimum and maximum N, P and K contents of roots and crop residues were derived from the literature (Table S1, Supporting Information). We assumed that roots are harvested after 12 months, average practice in sub-Saharan Africa (Fermont, 2009). Using data from Fermont (2009), we minimized the root mean squared error (RMSE) between measured and simulated dry total cassava biomass by inverse modelling. The ‘tuning’ parameters comprised adjustment factors to the estimated supply of N, P and K from the soil (Janssen et al., 1990).
Simulated cassava yields for the no input systems (see ‘Results’) were in the same range as actual yields reported for Gafaria and Nacuaca by Van den Dungen (2010). However, the yields that our model simulated are nutrient- and water-limited yields in absence of pests and diseases. In the study area, estimated yield losses due to among others cassava brown streak, cassava mosaic virus and termites are between 21% and 33% (Van den Dungen, 2010). In this light, simulated yields seem somewhat low.
For sugarcane we lacked site-specific data for calibrating FIELD, hence instead of using the model for simulating fertilizer response, we used fixed combinations of yields and fertilizer applications (76 Mg ha−1 of fresh cane with 120 : 40 : 200 kg ha−1 of N : P : K and 100 Mg ha−1 fresh cane with 170 : 60 : 285 kg ha−1 of N : P : K; Table 2), based on Tongaat Hulett (2010); Lewis (1984); Ndlovu (2000); Principle Energy (2009). The same approach was applied for sweet sorghum production, using data from Woods (2000): crop yield was set at 46 Mg ha−1 of fresh stems with fertilizer application of 90 : 60 : 60 kg ha−1 of N : P : K.
The model FIELD (Fig. 1), in particular the SOILSIM component, was also used for simulating SOC dynamics; changes in SOC are the net effect of addition of C in crop residues and the ongoing decomposition of added residues and SOC. Tittonell et al. (2007) successfully used it for simulating long-term SOC dynamics in agricultural soils in Zimbabwe. We used the calibration of Tittonell et al. (2007), based on chronosequence data for similar soils in Zimbabwe (Zingore et al., 2005).
In our sustainability assessments, we used two sets of initial conditions: SOC contents that we actually measured in the field (Table 1) and simulated equilibrium SOC contents 20 years after clearing. In all simulations, FIELD was run for 25 years within which new equilibria were established in all cases.
Calculation methods for the sustainability indicators are provided below. System boundaries and N, P, K and C flows of the systems assessed are displayed in Fig. 2. For additional details and values of coefficients, see Table S1 (Supporting Information).
The net energy yield per hectare is calculated by subtracting the energy requirements for producing and transporting fertilizers, agricultural operations, pumping irrigation water, harvesting and transporting feedstock to the mill from the gross energy yield (the energy present in the produced ethanol). For conversion of feedstock into ethanol, fixed efficiencies were used: 137, 90 and 60 L ethanol Mg−1 of fresh product for cassava, sugarcane and sweet sorghum, respectively (Table S1, Supporting Information). It was assumed that for sugarcane and sweet sorghum, energy requirement for processing can fully be covered by burning bagasse, i.e. the biomass remaining after stalks are crushed to extract their juice (cf. Macedo et al., 2004; Gnansounou et al., 2005). With cassava, only part of the processing energy requirements is met by biogas produced from the distilled mash. Taking this into account, the energy consumption for processing cassava is about 6.7 MJ L−1 ethanol (Nguyen et al., 2007). The effect of this extra energy requirement for cassava is demonstrated by calculating net energy with and without taking into account processing energy.
The GHG emissions indicator was calculated on a gross energy basis (kg CO2 eq. GJ−1 gross energy), similar to e.g. Farrell et al. (2006). We took into account emissions from production and transport of fertilizers, N2O emissions from the soil, emissions from production and combustion of fossil fuel, and simulated net emission or sequestration of C by the soil (simulated by the FIELD model). Changes in aboveground C stocks were not taken into account. We used the annual average GHG emission over 25 years as indicator value. For cassava, processing requires additional fossil energy hence emits additional GHGs. Nguyen et al. (2007) estimated emissions from converting cassava feedstock into bioethanol at 23.5 kg CO2 eq. GJ−1 ethanol.
Soil erosion (Mg soil loss ha−1 yr−1) was estimated by implementation of the Revised Universal Soil Loss Equation (Renard et al., 1996; Table S1, Supporting Information). For sweet sorghum and sugarcane, we used annual average crop factors from literature (Table S1, Supporting Information). Cassava yielded poorly in some simulations for smallholder systems hence standard values are not applicable. Roose (1977) indicated that crop factors for cassava vary between 0.2 and 0.8. We assumed this factor was proportional to biomass yield, where the seasonal crop factor was estimated at 0.8 for root yields of ≥25 tons and above and 0.2 for yields of ≤2 tons; in between these values the C factor is interpolated linearly.
The change in soil organic matter (Mg SOC ha−1) over 25 years was simulated by the FIELD model (Tittonell et al., 2007) as explained above.
NUE was calculated as
Where NUE the N-use efficiency (GJ net energy kg−1 available N); Enet the net energy yield (GJ ha−1); Navailable the crop available N (kg N ha−1), comprising applied fertilizer N and N mineralized from soil organic matter and crop residues calculated by the FIELD model.
N leaching (kg N ha−1 yr−1) is estimated as
where Nleached the quantity of N lost through leaching (kg N ha−1); Fleached the fraction of mineral N lost by leaching, estimated from soil texture and rainfall by transfer functions derived by Smaling et al. (1993) (Supporting Information).
Water productivity (MJ net energy m−3) was calculated as:
where Pwater is the water productivity of the biofuel (MJ net energy m−3) and Wavailable the volume of water potentially available to the crop (m3 yr−1), hence before e.g. conveyance losses and runoff occur. It includes both precipitation and supplied irrigation water.
Irrigation water requirements were calculated as the water required to bridge the gap between rainfed and target yield; we used fixed water-use efficiencies for rain and the types of different irrigation (see Table S1, Supporting Information).
Labour demand for agricultural operations and harvesting was estimated from literature data (Table S2, Supporting Information). Labour requirements for harvesting were assumed to be proportional to yield (manual harvesting) or acreage (mechanized harvesting): 35 man hours Mg−1 cassava (manual); 2.7 man hours Mg−1 of fresh sorghum stems (manual) and 1.6 man hours Mg−1 of fresh cane (manual) or 2 man hours ha−1 of cane (mechanized). Labour requirements of all other agricultural operations were assumed to be constant at 1226 man hours ha−1 (cassava, manual), 100 man hours ha−1 (sweet sorghum, partly mechanized) and 14 man hours ha−1 (sugarcane, mechanized).
Simulated SOC dynamics depend strongly on the initial SOC content of the soil: at equilibrium or recently cleared. For plantation systems of sugarcane, sweet sorghum and cassava with residue mulching on recently cleared soils, SOC declined, while in the cultivated soils that started at base equilibrium, SOC content increased (Fig. 3a). After about 15 years, simulations for both situations reached the same equilibrium. The largest equilibrium SOC content was obtained with sugarcane, due to the largest input of organic material (10 Mg DM ha−1), followed by cassava and sweet sorghum with residue inputs of 8.9 and 6.0 Mg DM ha−1, respectively.
Even though fertilized cassava in Nacuaca yielded the same quantity of residue mulch as in Gafaria (see next section), simulated equilibrium SOC contents in Gafaria were higher due to a higher clay content (Fig. 3b). In Nacuaca, unfertilized cassava yielded little residues for mulching, hence simulated SOC of this system with residue mulching was similar to the ‘no inputs’ treatment. Fertilization strongly increased the amount of mulch, hence simulated SOC content (100 : 150 : 200, mulch) was higher than the ‘no inputs’ treatment with mulch. In Gafaria, residues of unfertilized cassava had a relatively strong effect due to more fertile soil, hence more residues and slower decomposition of SOC due to higher clay content. Fertilization resulted in a further increase of (tuber and) residue yield, hence a further improvement in SOC content. In Dombe, irrigation almost doubled the amount of mulch available in Gafaria and Nacuaca. However, the effect of these residues on SOC is limited, due to rapid decomposition in the very sandy soils.
Simulated cassava root yields
Simulated potential cassava yields (in absence of nutrient and water shortages) were 76.8 Mg ha−1 of fresh roots for all locations; water-limited yields (in absence of nutrient shortages) were 28.6 Mg ha−1 of fresh roots for Gafaria and Nacuaca and 25.7 Mg ha−1 for Dombe.
Taking into account soil fertility, for unfertilized cassava in Gafaria we simulated a resource-limited yield of 19.0 Mg ha−1 fresh roots for recently cleared soils (Fig. 4a). This yield starts declining slightly in year 12, when soil N supply becomes limiting due to declining N availability. N mineralization, which is proportional to SOC content, decreases due to lack of organic inputs. In the new equilibrium, reached after about 20 years, N limitation is only moderate and yields remain rather good. The soil has a relatively high clay content hence protects a large SOC content that sustains N supply. Nevertheless, the yield decline may be prevented by applying 70% of the available crop residues [containing 62 kg N, 6 kg of P and 62 kg of K (nutrient contents in roots and residues are simulated by CROPSIM)] or 40 kg ha−1 yr−1 of N fertilizer (Fig. 4a); larger N applications do not result in further yield increase (data not shown). Yield improvement can only be achieved by P fertilization (Fig. 4a, 40 : 150 : 0). Owing to the very low P status of the soil (Table 1), annual application of 150 kg P ha−1 initially leads to a yield increase and build-up of the stable soil P pool. After 4 years sufficient P is exchanged with the labile (crop available) P pool, P limitations are removed and the water-limited yield of 28.6 Mg ha−1 is attained. Combining P application with residue mulch (70% of the residues left on the soil, containing 68 kg N, 17 kg P and 89 kg of K (nutrient contents in roots and residues are simulated by CROPSIM) led to a somewhat faster yield increase due to the additional nutrients supplied by the crop residues. After reaching the yield plateau, yields could be maintained with less P fertilizer (data not shown). Smaller P application rates lead to a longer period of P build up in the soil before the water-limited yield is achieved. Owing to the high K status of the soil in Gafaria (Table 1), fertilization with this nutrient is not immediately required.
In Nacuaca unfertilized resource-limited yield on recently cleared soils was 9.5 Mg ha−1 of fresh roots (Fig. 4b), substantially less than in Gafaria due to poorer soil fertility. Over time, root yield decreases further to about half its initial value; the decline starts after year 7, when soil N supply starts to fall below crop requirements; the mechanism is similar to that described above for Gafaria. Owing to the small amount of residues produced, mulch contributed little to cassava yield. Simulations indicated that 150 kg N ha−1 (150 : 0 : 0, Fig. 4b) was required to prevent this decline, while higher applications did not further improve yields (not shown). The crop also responded to P applications up to 150 kg ha−1 and K applications up to 200 kg ha−1 [0 : 150 : 0 and 0 : 0 : 200 respectively, Fig. 4b (These fertilizer applications only serve for demonstration and are not part of the sustainability assessment)], hence both P and K severely limited crop growth (see cf. Table 1). However, after some years P-only or K-only fertilized yields are affected by the declining availability of N, hence illustrating the need for balanced fertilization. The use efficiency of the macronutrients was much enhanced when applied jointly: 100 : 150 : 200 was sufficient for achieving the water-limited yield of 28.6 Mg ha−1 (Fig. 4b). Reaching this yield took a number of years, depending on the amount of P applied and whether or not mulch was applied. Again, after reaching the yield plateau, yields could be maintained with less P (data not shown).
For Dombe, we investigated commercial cassava production with fertilization, mulching and irrigation. For reaching a yield of 50 Mg ha−1 of fresh roots, the maximum yield recorded in East Africa (Fermont, 2009), 200 : 300 :400 kg ha−1 of N : P : K and 470 mm of drip irrigation or 750 mm of surface irrigation were required (not shown).
Net energy yields
Cassava and sweet sorghum systems yielded less net energy per hectare than sugarcane systems (Fig. 5a). Fertilization, although also consuming energy, improved net energy yield of cassava systems, especially in Nacuaca where it more than tripled. In Dombe, where soils were the least fertile, combining fertilization with 470 mm of drip irrigation resulted in a better net energy yield (116 GJ ha−1) than was obtained on more fertile soils in Gafaria and Nacuaca without irrigation. With surface irrigation nearly the same net energy yield was obtained (not shown); it required more water (750 mm) but less energy per volume of water due to the lower pressure requirement hence the net result was similar. Taking into account the processing energy consumption (Nguyen et al., 2007) reduced the net energy yield of cassava (Fig. 5a) in comparison with the other crops since with this crop additional fossil energy is required.
Irrigated sugarcane systems perform best for net energy yield (Fig. 5a). The differences among the irrigated sugarcane systems were due to different energy requirements for irrigation of 4.1 (drip), 7.0 (surface, not displayed) or 13.9 GJ ha−1 (pivot), and for harvesting: 1.1 (manual) or 3.8 GJ ha−1 (mechanical); they remain relatively small compared with the differences with the other systems however. Rainfed sugarcane produces less net energy than irrigated sugarcane, due to the smaller crop yield.
Sweet sorghum performs similar to cassava: it produced less biomass and contains less sugar than sugarcane while still requiring substantial N fertilization. Transport energy requirements were higher than for cassava; the latter crop can be transported as (sun-)dried chips while sweet sorghum stalks have to be transported fresh; this also applies to sugarcane.
For recently cleared soils, GHG emissions on a gross energy basis of the no-inputs treatment in Nacuaca (Fig. 5b) were much larger than in Gafaria due to the very low net energy yield in Nacuaca. Balanced fertilization improved GHG performance since the increase in cassava yield was greater than the additional emissions from fertilization. Emissions of fertilized treatments in Gafaria were somewhat lower than in Nacuaca where, due to poorer soil fertility, more fertilizers were needed to achieve the same yield. The largest share of the emissions was caused by SOC decomposition, as becomes evident from comparison with emissions from cultivated soils (Fig. 5b). Application of crop residues had a beneficial effect due to higher equilibrium SOC levels hence reduced net emissions.
Emissions of sugarcane systems where generally slightly lower than those of the cassava systems. Sweet sorghum on recently cleared soils produced rather high emissions: net energy yield is equal to, e.g. cassava with no inputs in Gafaria, however, in sweet sorghum substantial inputs (e.g. fertilizers) are used.
Emissions of all treatments, except no-input cassava on recently cleared soils in Nacuaca were lower than those from production and combustion of conventional gasoline (Fig. 5b).
Of the three crops, sugarcane had the best NUE, followed by cassava and sweet sorghum, respectively (Fig. 5c). Owing to the very low availability of P from unfertilized soil in Nacuaca, N efficiency for the no input system was the poorest. Mulching combined with fertilization had a poorer NUE than fertilization only; apparently, the extra N mineralizing from the residues was used less efficiently by the crop.
For zero-input systems, nutrient-limited yields fell below water-limited yields, hence only a fraction of the effective rainfall was utilized, resulting in poor Pwater (Fig. 5d). Fertilization was such that water-limited yields were attained, hence the crop fully utilized the effective rainfall. For sugarcane, target yields (Table 2) were greater than the water-limited yields, hence the effective rainfall is fully utilized by this crop. The effect of irrigation on overall Pwater depends on the method; sugarcane with drip irrigation had the best Pwater (14.3 MJ m−3), followed by pivot (13.0 MJ m−3) and furrow irrigation (11.0 MJ m−3, not shown), respectively. Sweet sorghum performed similarly to the fertilized cassava systems.
Since rainfall in Gafaria and Nacuaca is similar, differences in N leaching between these two villages depended on crop growth and management and soil texture. Therefore, the larger N fertilizer rates in Nacuaca led to more N leaching, as did application of crop residues also led to increased leaching (Fig. 6a). Sugarcane with mulching gave more leaching than when sugarcane residue was burned due to the additional N mineralized from the trash.
Soil erosion was strongly related to crop yield and soil cover (Fig. 6b). Erosion was greatest for zero-input cassava in Nacuaca, the poorest yielding system with least soil cover. Fertilization increased crop growth and thereby reduced erosion due to better soil cover. Residue mulching further reduced soil erosion. Sugarcane and sweet sorghum performed better than cassava, although the differences largely disappeared for high-input cassava systems.
Cassava systems required much more labour than systems based on sugarcane or sweet sorghum (Fig. 6c); cassava required around one person year ha−1. In contrast, for sugarcane and sweet sorghum that are harvested manually, requirements are 4–5 person weeks ha−1, while for mechanically harvesting it is <0.5 person week ha−1. Labour requirement for harvesting and agricultural operations were in the same order of magnitude if harvesting was manual.
Synthesis of results
We indexed the obtained values for each indicator except GHG emissions in percentages relative to the most sustainable value found for that indicator across all crops and locations, which was set to 100. Only GHG emission reduction (in percentages) was calculated relative to that of the replaced gasoline. Results for selected systems are displayed in Fig. 7(a)–(i); since not all systems are displayed, the ‘100%’ value for the best indicator cannot always be found in the figure. The GHG indicator is displayed for cultivated soils; SOC dynamics were captured in a single figure by taking the difference between soil C stocks at year 25 and at year 1 for cultivated soils.
Comparison of crops
Smallholder cassava systems (Fig. 7a–d) were outperformed by sugarcane in plantations (Fig. 7f–h) with respect to net energy yield, SOC build-up if residues of both crops are mulched, NUE, soil erosion and water productivity. Only for N leaching cassava performed better, due to lower N fertilizer rates. The improved smallholder cassava systems performed similarly or somewhat better than sweet sorghum in plantations (Fig. 7b,d vs. 7i).
Taking into account the energy requirements of cassava processing changes the energy- and GHG-related indicators in favour of sugarcane and sweet sorghum, as these crops generate sufficient energy for this purpose by burning bagasse. Although sugarcane performed better than cassava for production-ecological sustainability, Arndt et al. (2010) found that ethanol production from cassava grown by smallholders has stronger economic growth effects than sugarcane ethanol. On the other hand, while cane sugar is mostly an export product, cassava is a vital domestic food security crop in Mozambique and increased demand could easily lead to price hikes. In formulating policies, a balance should be struck between environmental aspects, poverty reduction goals and food security considerations.
We analysed systems in which one crop is cultivated continuously. Sustainability could also be improved by, for example, rotating biofuel crops with crops that have low harvest indices and high residue yields for maintaining SOC at favourable levels, managing GHG emissions and even fixing N (e.g. pigeonpea). Pigeonpea is currently a common crop among smallholders in the study area.
Cassava cultivation in Gafaria performed better than in Nacuaca due to better soil fertility. Fertilization and residue mulching (Fig. 7b and d) for both locations resulted in a substantial improvement compared with the zero-input system (Fig. 7a and c). Net energy yield, SOC sequestration, NUE and Pwater increase, while soil erosion is strongly reduced.
A practical problem with the improved systems is that mulching large quantities of cassava tops, especially stems, may obstruct preparation of the field for the next crop: this was mentioned by smallholders in a rapid farm survey in the study area (Van den Dungen, 2010). A shredding device could be the solution, but getting such an implement adopted among smallholders may be difficult or expensive compared with just removing or burning residues.
Cassava yields can be boosted further by introducing irrigation and increasing fertilizer rates (Fig. 7e). The result is that net energy yield and SOC sequestration improve, while GHG reduction and N leaching are less favourable due to increased input use. This type of system is more feasible for plantations than poor smallholders.
Sugarcane and sweet sorghum
Sugarcane cultivation with residue mulching, drip irrigation and mechanical harvesting yielding 100 Mg ha−1 emerged as the most sustainable production system in our comparison; it only performed suboptimal for N leaching. Burning sugarcane residues (Fig. 7g) negatively affected GHG performance and virtually eliminated SOC build up (Garside, 1997), but reduced N leaching.
We did not reduce N applications to compensate for the simulated extra N mineralized from residues. Robertson & Thorburn (2007) found that fertilizer N application should not be reduced in the first 6 years after adoption of residue mulching in sugarcane because of immobilization, and that small reductions may only be possible in the longer term (>15 years).
Water requirements of furrow irrigation were highest, followed by pivot and drip, respectively. Lower yielding rainfed cane with residue burning (Fig. 7h) performed rather similar to irrigated, high yielding sugarcane with burning (Fig. 7g); the main difference was that N leaching was reduced due to lower fertilizer application and that net energy yields were slightly reduced due to lower crop yield. Since it fully utilized the effective rainfall, Pwater was somewhat better. Owing to uneven rainfall distribution, the majority of the Mozambican sugar estates are irrigated, however (Instituto Nacional do Açúcar, 2000).
Regarding sweet sorghum, Fig. 7i displays the performance of mechanized cultivation on plantations, where fossil fuel use contributes to GHG emissions and reduces net energy yield. Sweet sorghum cultivation in smallholders systems with predominantly manual labour may perform better. Also, the N fertilization rate in sweet sorghum (Woods, 2000) was based on US practices and could well be too high, contributing to a low NUE; often, response to N is absent in sweet sorghum (cf. Barbanti et al., 2006). Yields could also be increased by ratooning; two crops a year may then be obtained, provided there is enough water. Owing to much lower labour requirements of sweet sorghum as compared with cassava (Fig. 6c), this crop might be more suitable for smallholders in Gafaria and Nacuaca than cassava.
Cultivated vs. recently cleared soils
On cultivated soils, all analysed systems met the EU standard of 35% GHG reduction (European Parliament, 2009) compared with the emissions of conventional gasoline of 86 kg CO2 eq. GJ−1 (Fig. 5b, Punter et al., 2004), even when cassava processing was taken into account. On recently cleared soils, cassava with no inputs in Nacuaca does not meet the standard, due to the small net energy yield. If emissions from processing were taken into account, GHG emissions in Fig. 5b would increase by 23.5 kg CO2 eq. GJ−1 (not shown); in that case only cassava with fertilization and residue mulching in Gafaria and irrigated cassava in Dombe would meet the 35% reduction target.
When land is allotted to biofuel production in the tropics, it will often be somewhere on the trajectory between clearing and equilibrium; clearing natural vegetation for biofuel production is not preferred, among others to avoid creation of a ‘C debt’ (Fargione et al., 2008) and to preserve biodiversity. For indicators that differed strongly between cultivated and recently cleared soils (GHG emissions, SOC dynamics), actual values will be between the two simulated extremes.
The approach presented provides quantitative insight into the production-ecological sustainability of biofuel feedstock production systems with different crops and crop management, on different soil types and for different land-use histories. For our set of production-ecological sustainability indicators, cassava for bioethanol both in smallholder and plantation systems performed more poorly than plantation-style cultivation of sugarcane. The latter system requires substantial volumes of surface water for irrigation, however, which in the future may also be needed for other purposes. Also, it has a less pronounced pro-poor effect and generates less labour than smallholder cassava production. Cassava and sweet sorghum performed similarly.
For smallholder cassava, increased but targeted use of inputs improved sustainability, e.g. through greater net energy yield, greater production of crop residues and better erosion control. With increased N fertilization, N leaching increased however. If smallholders are to be involved in production of feedstock for biofuels, sweet sorghum has the advantage of requiring far less labour than cassava. In the study area, labour shortage of rural households often limits the acreage cropped and hence, agricultural output.
Cassava production systems on more fertile soils (Gafaria) were more sustainable than those on less fertile soils; the latter required more agricultural inputs, affecting most indicators negatively. Increased input requirements on less fertile soils also reduce the financial feasibility of achieving yield improvement for smallholders. However, instead of producing biofuel feedstock, it is often preferred to use the more fertile (clayey) soils for production of food crops. In contrast, relatively sustainable sugarcane systems can be achieved on poor sandy soils if irrigation is available. Production of bioethanol feedstock on cultivated lands is more sustainable than on newly cleared land as large GHG emissions after clearing natural vegetation are avoided and instead there is a potential for C sequestration, which can be realized through suitable management of crop residues. New SOC equilibrium levels are established within 15 years. After this time, further sequestration is only possible by increasing C inputs.
Overall, sugarcane systems performed better than cassava and sweet sorghum systems for our set of seven production-ecological sustainability indicators.
We thank Sanne van den Dungen for the field work in Mozambique, Anneke Fermont for letting us use the cassava dataset, Bert Janssen for discussions and practical advice and Mark van Wijk for help with inverse modeling. We acknowledge Shell Global Solutions for partly funding the work.