Can intensification reduce emission intensity of biofuel through optimized fertilizer use? Theory and the case of oil palm in Indonesia

Closing yield gaps through higher fertilizer use increases direct greenhouse gas emissions but shares the burden over a larger production volume. Net greenhouse gas (GHG) footprints per unit product under agricultural intensification vary depending on the context, scale and accounting method. Life cycle analysis of footprints includes attributable emissions due to (i) land conversion (‘fixed cost’); (ii) external inputs used (‘variable cost’); (iii) crop production (‘agronomic efficiency’); and (iv) postharvest transport and processing (‘proportional’ cost). The interplay between fixed and variable costs results in a nuanced opportunity for intermediate levels of intensification to minimize footprints. The fertilizer level that minimizes the footprint may differ from the economic optimum. The optimization problem can be solved algebraically for quadratic crop fertilizer response equations. We applied this theory to data of palm oil production and fertilizer use from 23 plantations across the Indonesian production range. The current EU threshold requiring at least 35% emission saving for biofuel use can never be achieved by palm oil if produced: (i) on peat soils, or (ii) on mineral soils where the C debt due to conversion is larger than 20 Mg C ha−1, if the footprint is calculated using an emission ratio of N2O–N/N fertilizer of 4%. At current fertilizer price levels in Indonesia, the economically optimized N fertilizer rate is 344–394 kg N ha−1, while the reported mean N fertilizer rate is 141 kg N ha−1 yr−1 and rates of 74–277 kg N ha−1 would minimize footprints, for a N2O–N/N fertilizer ratio of 4–1%, respectively. At a C debt of 30 Mg C ha−1, these values are 200–310 kg N ha−1. Sustainable weighting of ecology and economics would require a higher fertilizer/yield price ratio, depending on C debt. Increasing production by higher fertilizer use from current 67% to 80% of attainable yields would not decrease footprints in current production conditions.


Introduction
The Borlaug or 'land sparing' hypothesis (Sanchez, 1994;Rudel et al., 2009) states that intensifying agriculture will have net positive effects on the environment, regardless of any directly negative environmental impacts of 'green revolution' production technology, as it reduces the land base needed to meet world market demand for agricultural products (food, fibre and fuel) and thus reduces biodiversity loss (Green et al., 2005) as well as emissions from deforestation and forest degradation. The 'land sharing' or ecological agriculture hypothesis that forms its counterpart suggests that a careful balancing of productivity and environmental services in integrated production systems can contribute to multifunctionality of integrated landscapes that is superior to the 'segregated' agriculture-plus-forest perspective of the Borlaug hypothesis ( van Noordwijk et al., 1995;Tomich et al., 1998;Angelsen & Kaimowitz, 2001;Lee & Barrett, 2001). Choices for an optimal level of intensification may depend on location ('theory of place') ( , type of environmental services considered (Grau et al., 2013) and scale . As currently framed , the sparing plus sharing debate considers the wider policy context that is needed to turn a 'necessary' to a 'sufficient' condition: environmental issues and deforestation cannot be resolved without an increase in yield levels that exceeds the growth in global demand for food, fibre and energy. However, it is naive to expect markets to directly effectuate environmental benefits through a pathway of reducing the profitability of less-efficient production modes. Both 'sparing' and 'sharing' approaches will only achieve environmental benefits if the opportunity for such benefits is utilized in active 'caring' approaches (Jackson et al., 2012). As earlier assumptions about direct links between yield and the efficiency gap are not supported by evidence ( van Noordwijk & Cadisch, 2002; van Noordwijk & Brussaard, 2014), there is space for intermediate intensity solutions to be optimal from a societal perspective. Regulation of land use, however, cannot easily incorporate the fine-tuning needed to minimize environmental effects of land use change (Lambin et al., 2014). Pricing of input costs deserves further analysis as possible finetuning policy instrument. We here provide a quantitative analysis of intermediate optimum intensification levels, applicable to biofuels as costs and benefits can both be expressed in terms of net greenhouse gas emissions. For nonbiofuel crops, a similar analysis will require further steps to bring alternative options onto a single denominator.
In the biofuel debate, the interest has shifted from single characteristics of feedstock types (e.g. comparing soybean, palm and Jatropha oil), to recognition of the management swing potential (Davis et al., 2013;Creutzig et al., 2015) where the footprint of any feedstock depends on where and how it is produced as much as on what crop it is. The widest swing potential, according to current data, exists for palm oil, with both the best and worst emission intensities per unit product. Oil palm (Elaeis guineensis Jacq.) (Corley & Tinker, 2016) expansion is a 'Pandora's box' example (Tomich et al., 1998) of intensified tree crop production that attracts new activities in the tropical forest margins and increases forest conversion rather than reducing it. In public debate, oil palm expansion is held responsible for much of the loss of biodiversity and flagship species, but also as a cause of increased greenhouse gas emissions (Sheil et al., 2009). The yield gap in oil palm production is considerable for large-scale plantations and even larger for smallholder production systems (L.S Woittiez 2016), indicating a land equivalent ratio of below 1.0. Existing self-regulation in the industry is based on recommended 'good agricultural practice' without quantification of existing yield and efficiency gaps (von Geibler, 2013). There is little clarity on the level of fertilizer use that is considered good practice, from both a farm-level profitability and an environmental perspective.
The irony of biofuel use increasing rather than decreasing net anthropogenic greenhouse gas emissions in the 'biofuel boom' of the 2000s has led to a rapid regulatory response. A review of recent life cycle assessment (LCA) studies (van der Voet et al., 2010) showed considerable variation in outcomes, due to real-world differences, data uncertainties and methodological choices. If fossil fuel is partially substituted by 'biofuel', there are costs as well as benefits in terms of greenhouse gas emissions (Wicke et al., 2008). European regulation of the minimum emission reduction factors compared to the use of fossil fuel (emission saving) due to biofuel use in the Renewable Energy Directive of 2008 (European Communities Commission, 2008) has drawn attention to three types of emission costs (Hoefnagels et al., 2010): (i) the carbon debt due to land conversion from higher to lower time-averaged C stock, (ii) emissions associated with the production phase of the biofuel, part of which are on-site and part in the industry producing the inputs used, and (iii) emissions due to processing and transport. For emissions associated with the production phase, the issue of 'optimum intensification' levels is relevant (Jackson et al., 2012): Is there any merit in not fully utilizing the biophysical production opportunity of land that is already in agricultural use by moderating the use of fertilizer and similar inputs? Are current costs of fertilizer and other inputs sufficient to induce their wise and efficient use and low enough to allow 'environmentally optimum' levels of intensification?
The relationship between agricultural yields, environmental impacts of production and optimized use of inputs has been debated since the 1980s or earlier (van Noordwijk & de Willigen, 1986;de Wit, 1992;Zoebl, 1996). Increased yields increase the denominator of an efficiency (output/input) metric, but increased environmental impacts increase the numerator; the outcome depends on the shape of the yield and environmental impact response curves. The shape of these response curves themselves depends on factors such as the within-field spatial variability (Cassman & Plant, 1992;van Noordwijk & Wadman, 1992) and the degree of 'precision farming' adjustment of input levels to patchlevel production conditions (Heege, 2015). de Wit (1992), showed that in the presence of multiple yieldlimiting factors, the overall response of yield to aggregated input levels (or their associated environmental consequences) can be multiphasic. Neither of the extreme positions in the agriculture-environment debate ('Optimize yields economically and environmental impacts per unit product will be minimal' or 'Minimize inputs to maximize efficiency and minimize environmental impacts per unit product') are tenable as generalizations (Wicke et al., 2008). High yield levels can be achieved in combination with low and high efficiency; high efficiency can be coupled to low and high yields ( van Noordwijk & de Willigen, 1987). The relationship between efficiency and yield depends on the finer details of the yield and environmental impact responses to input use, requiring empirical study for each crop and its specific physiology and agronomy (Corley et al., 1971).
Fertilizer subsidies have a long history in developing countries as a part of policies to intensify agriculture and maintain affordable staple food provisioning to urban people. Attempts to segment the markets and subsidize fertilizer only for certain crops or types of farmers are hard to implement, as regional or local fertilizer markets function well. If fertilizer prices are (too) low, efficiency enhancement is not economical (van Noordwijk & Scholten, 1994); if they are (too) high, the economical optimum solution may well be the nearcomplete mining of the soil ( van Noordwijk, 1999). The relevance of shifting net fertilizer subsidies towards net taxation has been debated as a measure to reduce negative environmental effects of agricultural production through groundwater and surface water pollution or emission of N 2 O, a powerful greenhouse gas. van Noordwijk & Wadman (1992) defined an environmentally optimum fertilizer level by reference to tolerated levels of nitrate enrichment of ground and surface water in the Netherlands; for potential biofuels, this target can be replaced by minimizing the emission footprint.
We will here focus on the relationships between N fertilizer use and the greenhouse gas emissions per unit biofuel use, relative to the fossil fuel use it can substitute for, using oil palm production in Indonesia as case study. Data from 23 plantations across the Indonesian production conditions (Khasanah et al., 2012(Khasanah et al., , 2015a provided insights into what is currently considered good agronomic practice, as participation in the survey was voluntary. We will provide an algebraic analysis of the problem in generic terms and then review the available quantitative data. Key policy-relevant questions are as follows: 1. Is there an 'environmental optimum' production level at which net emission savings per unit biofuel use are maximized? 2. At what fertilizer/product price ratio is the 'economic optimum' fertilization rate equal to the 'environmental optimum' that minimizes attributable emissions? 3. How do the answers to questions 1 and 2 depend on the overall emissions from the life cycle analysis: (i) C debt due to initial land conversion, (ii) CO 2 emissions due to fertilizer production, (iii) the N 2 O emission factor per unit of fertilizer used, and (vi) the technical coefficients for emissions due to transport and processing? 4. Are current policies for fertilizer subsidies and taxation aligned with environmental efficiency?

Theory
Four production phases contribute to emission estimates of biofuel production in a life cycle analysis ( Fig. 1): (i) carbon debt (positive in all cases where the preceding vegetation had higher C stock than the oil palm plantations themselves) and additional emissions due to conversion (e.g. use of fire on peat soils), (ii) production of external inputs, such as inorganic fertilizer, (iii) feedstock production that determines the yield per ha that relates area-based terms to product-based accounting, but that may also lead to a change in belowground C stocks, recurrent GHG emissions related to drainage and/or N 2 O emissions due to fertilizer use, and (iv) transport and processing stages before the product reaches the end users. A detailed scheme to estimate biofuel production's net emissions and emission savings is provided in the supporting information [The Biofuel Emission Reduction Estimator Scheme (BERES)], with some key parameter values that are considered 'defaults' based on measurement and literature review.
The shape of the response curve describing yield as function of fertilizer input has been much debated in the literature, with many empirical results converging on a Mitscherlich curve with asymptotic approach of a maximum yield. de Wit (1992) posed that the diminishing returns interpretation of Mitscherlich curves disappears when multiple constraints are addressed simultaneously. van Noordwijk & Wadman (1992) explored how empirical Mitscherlich-type curves can be interpreted as the result of spatial variability at field-level and patch-level responses by the crop that can be described by a quadratic equation, with a maximum that can be obtained or exceeded in practice. Quadratic models represent the most optimistic perspective on nutrient use efficiency at crop level with minimum field-scale variability. We use them here and will revert to the validity of this assumption in the Discussion section.
A quadratic fertilizer (N) yield (Y) response curve (Y = Y 0 + f N + c N 2 ) has three parameters (Y 0 , f and c), corresponding to the yield without fertilizer use, the initial efficiency of fertilizer use and a parameter that combines f and the maximum attainable yield, respectively. Net annual emissions per unit crop yield are E 0 + e 0 + e f N + e Y with the parameter E 0 or annualized attribution of the C debt representing phase I, e 0 (emissions independent of fertilizer use) phase III, e f (proportional to fertilizer use) phases II and III and e (proportional to yield) phase IV.
The emissions per unit production have a local minimum (hence, the emission savings compared to fossil fuels use a local maximum in case of a biofuel crop) when the N fertilizer rate equals (as derived in Appendix S1): where R and B are intermediate terms, f is initial marginal yield increment per unit of fertilizer used, E 0 is attributable CO 2 eq emissions per ha per year due to initial land conversion, e 0 is attributable CO 2 eq emissions per ha per year in the production stage at zero fertilizer use, e f is attributable CO 2 eq emissions per ha per year per unit fertilizer use in the production stage, Y 0 is the yield level in the absence of fertilizer use and Y max is the maximum attainable yield under current circumstances beyond fertilizer use. The dimensionless B grouping is the ratio of the 'fixed cost' emissions E 0 + e 0 and the maximum of fertilizer-related emissions, (Y max À Y 0 ) (e f /f). If p is the price ratio of yield products and fertilizer inputs, the economic optimum N fertilizer rate equals the fertilizer rate that minimizes emissions per unit yield, if p = p SWEET (SWEET = 'Sustainable Weighting of Ecology Economics Tradeoffs') (see Appendix S1): While this provides a generic answer to question 2, questions 1, 3 and 4 require parametrization for specific combinations of crop, attributable emissions from C debt and fertilizer use. Please note that the postharvest emissions (phase IV) represented in the term e are not influencing the fertilizer rate that minimizes net attributable emissions and the outcome of the sparing vs. sharing debate. Phase IV emissions, however, can be an important determinant of the absolute level of emission attribution per unit final product and whether or not the overall footprint meets standards set.

Sampling design
The IPOC/ICRAF survey of Indonesian palm oil production in 2010 was designed to estimate greenhouse gas emissions due to palm oil production across the major stratifying production factors in Indonesia (Khasanah et al., 2015a,b). The three primary stratifiers of the survey were defined at national level as mineral vs. peat soils, plantations directly derived from forest or other land cover types and three levels of the prevalence of oil palm at provincial level (<1%, 1-5% and >5%). These act as indicators of the fact that the areas that first developed oil palm are probably most suited to it climatically and have the most advanced input and output markets. Not all 12 factorial combinations are important in practice, as oil palm on peat has mostly been directly derived from forest. The sampling design followed a stepwise cluster approach, soliciting self-nomination of companies to involve in learning the method while involving in data collection (Khasanah et al., 2012(Khasanah et al., , 2015a. Candidate companies were asked to describe land history, soil type and the scale of management (plantations, outgrowers, independent smallholders). A total of 23 plantations were selected for study, representing nine of the 12 clusters (Table 1). Figure 2 presents the spatial distribution of the selected samples by relative oil palm density in a province.

Data collection and analysis
In the 23 selected oil palm plantations, we collected the main parameters needed for the Biofuel Emission Reduction Estimator Scheme (BERES) (van Noordwijk et al., 2013; see Appendix S2) to calculate the net emission of biofuel production and emission savings using a life cycle approach. The scheme is aligned with the way the EU RED policy requires life cycle data on the biofuel value chain. In the application, however, we did not use the 2008 'grandfather' rule that ignores C debts for land converted before the rules were made. It requires data for (i) C stock (Mg C ha À1 ) of land cover preceding oil palm plantation (with the concept of time-averaged C stock applicable to rotations, and the current one to land cover types that are supposed to be in equilibrium), (ii) time-averaged C stock of the oil palm plantation, Mg C ha À1 , (iii) nitrogen (N) fertilizer level, kg N ha À1 and production level of fresh fruit bunches (FFB), Mg ha À1 yr À1 , (iv) oil extraction rate (OER) of crude palm oil (CPO) and kernel extraction rate, (v) soil CO 2 loss, (vi) emission factors due to fertilizer production and application and (vii) emissions due to postharvest commodity transport and processing before the product reaches the end-user (Germer & Sauerborn, 2008;Kamahara et al., 2010;Wicke et al., 2008;Alkabbashi et al., 2009).
Biomass C stock of land cover preceding oil palm plantation. The 'time-averaged aboveground C stock' is the sum of the average over a production cycle of C pools (aboveground tree biomass, understorey vegetation and surface necromass). The belowground part of biomass is usually considered to be a proportion of aboveground biomass, with land cover-specific data hard to obtain. Data from the survey were used to establish estimates of the time-averaged aboveground C stock of oil palm plantation of around 40 Mg C ha À1 , as described in Khasanah et al. (2015b). The time-averaged aboveground C stock of forests and other preceding land cover types were assessed following the rapid carbon stock assessment (RaCSA) methodology and technical manuals (Hairiah et al., 2011). Root biomass   was estimated as 25% of aboveground biomass for all land cover types. Identification of land cover type preceding oil palm used the analysis of land use and cover trajectory (ALUCT) protocols (Dewi & Ekadinata, 2013). Changes in timeaveraged soil carbon stock when forest or other land cover types were converted into oil palm were analysed by Khasanah et al. (2015a). Khasanah et al., 2015a,b discussed how life cycle inferences could be made for soil C org and the oil palm biomass, respectively, despite the incomplete data for certain age classes in the various clusters (Table 1).

FFB and CPO production in relation to N Fertilizer level.
The companies participating in the study provided time series data of their fertilizer use and production level of FFB across the age range of plantations under their management control. For each company, we developed a quadratic equation of FFB production (Y, Mg ha À1 yr À1 ) as a function of age (years after planting; T, year) to estimate (by integration) the time-averaged FFB production level over the life cycle: Y = a + b T + c T 2 . Total N input was calculated across the various fertilizer types reported. The data of N fertilizer application did not show clear correlations with the age of oil palm. Therefore, an average rate of N fertilizer application over the whole life cycle was calculated and used. Time-averaged yield (Y) was related to this average fertilizer rate by regression analysis for a quadratic response model [Y = Y 0 + f N + c N 2 , with c = Àf 2 /(4 (Y max À Y 0 )); see equation [5] in Appendix S1]. While a range of fertilizer types was reported, we focussed on the N content as basis for expected yield response, but used the most commonly used compound fertilizer (15-15-15) as a basis for fertilizer costs. The companies also provided data on their CPO and kernel extraction rate. As variation in these two parameters was limited, an average value of CPO and kernel extraction rate extraction rates was calculated and used in the subsequent analysis.
Emission factors due to postharvest transport and processing. Emission factors due to postharvest transport and processing were based on fossil fuel use and technical design of the mills, and processing steps before the product reaches the end-user (Demirbas, 2007;Kamahara et al., 2008;Wicke et al., 2008;Alkabbashi et al., 2009).

Sensitivity analysis. To understand the responses of emission
saving to changes of carbon debt, N fertilizer application and ratio of N 2 O-N/N fertilizer, a sensitivity analysis was carried out. Five carbon debts, 0, 20, 30, 40 and 60 Mg C ha À1 , were combined with N fertilizer applications in the range of 0-550 kg N ha À1 , with an interval of 5 kg N ha À1 . The Intergovernmental Panel on Climate Change's National Greenhouse Gas Inventory Guidelines suggest that the ratio of N 2 O-N/N fertilizer is 1% (IPCC, 2006). Other literature suggests this can be 4% (Crutzen et al., 2008). In the absence of site-specific measurements, both assumptions were compared for impact on the end result. The IPCC national greenhouse gas inventory framework (IPCC 2007;Bentrup & Palli ere, 2008) includes the CO 2 emissions involved in fertilizer production under industrial processes, rather than land use sections. With a default value of 3.5 kg CO 2eq per kg N fertilizer, the net effect of these CO 2 costs of fertilizer production is less than the 4.65 kg CO 2eq per kg due to the associated N 2 O emissions in land use for an N 2 O-N/N fertilizer ratio of 0.01, with global warming effect of a molecule of N 2 O calculated as 296 times that of a molecule of CO 2 .

Time-averaged aboveground C stock of land cover preceding oil palm
We found 21 types of land use systems surrounding the 23 oil palm plantations, which were further classified into three larger categories 'forest', 'tree-based systems' and 'non-tree-based systems' (Khasanah et al., 2012). The range of time-averaged aboveground C stock values was 150-250 Mg C ha À1 for 'forest', 50-150 Mg C ha À1 for 'tree-based systems' and less than 50 Mg C ha À1 for non-tree-based systems (Fig. 3). These figures were derived from 924 measured plots, 800 of which came from the ICRAF database of earlier studies in Indonesia.

Level of N Fertilizer and production of FFB and CPO
Based on a survey of 23 plantations throughout the oil palm production domain in Indonesia (Khasanah et al., 2012), we found an average fresh fruit bunch (FFB) yield of 18.8 Mg ha À1 yr À1 and an average fertilizer use of 141 kg N ha À1 yr À1 (Fig. 4 and Table 2). FFB yield and N fertilizer use were closely associated, with an apparent fertilizer response curve of Y(FFB) = 8.23 + 0.0889N À 0.0001N 2 , with 74.2% of variance accounted for (Fig. 5a), suggesting Y 0 = 8.23 and Y max = 27.98 Mg ha À1 yr À1 . This apparent fertilizer response is derived from survey data, rather than randomized experiments. While we used the N fertilizer rate as basis for the regression, we can assume that other plant nutrients were provided in proportion and/or that residual variation in Fig. 5a is due to such factors. For oil yield (CPO) per ha, the relation was Y(oil) = 1.47 + 0.0298N À 5E-05N 2 , with 70.2% of variance accounted for (Fig. 5b). With the average N fertilizer level reported, the yield is expected to be 67% of the maximum FFB yield that can, apparently, be obtained with existing germplasm and plantation management represented in the data set.

Emission saving and sensitivity analysis
A default estimate of 40 Mg C ha À1 (Khasanah et al., 2015b) of aboveground C stock and no mineral soil loss (Khasanah et al., 2015a) was used to estimate emission saving. When the preceding vegetation had a higher C stock (and conversion took place after the cut-off date of applicable standards, e.g. 2008 for the EU RED), the plantation started with a 'carbon debt'. If preceding C stock was less, the calculation can reflect a net emission saving for the first production cycle. Rather than a single 'typical' value, the IPOC/ ICRAF data set shows wide variation in C debt (phase I), yield levels and N fertilizer use (phases II and III). Our data support the conclusion that peatland emissions are off the scale and preclude attainment of the emission saving standards (Fig. 6) (Couwenberg et al., 2010;Maswar, 2011).
The curvature of the relationship between the level of N fertilizer and production of FFB plus the effect of a 'fixed cost' of C debt lead to an interesting shift in the shape and positions of the curves that relate the emission savings to the N fertilizer level in the production stage (Fig. 7). A net emission saving target of 35% cannot be achieved if C debt is more than 20 Mg C ha À1 for a N 2 O/N fertilizer loss rate 4% and when C debt is more than 40 Mg C ha À1 for a N 2 O/N fertilizer loss rate of 1%.
For many parameter combinations cases, there is a weakly defined 'optimum' N fertilizer level that maximizes the emission savings, within a rather broad range where emission savings vary less than 5% Fig. 4 Relationship between the age of the oil palm and fresh fruit bunch (FFB) production level as derived for the plantations that were part of the IPOC/ICRAF survey. (differences that may be below experimental error); in some cases, the optimum is outside the 0-500 kg N ha À1 yr À1 , and zero fertilizer use would give the highest emission reduction rate per unit biofuel derived from the production system (Table 3).

Discussion
Our analysis showed that under parameter conditions that apply to relevant subsets of palm oil production on mineral soil in Indonesia, there is an 'environmental optimum' production level at which net emission savings per unit biofuel use are maximized. The net emission savings decrease strongly with increasing C debt, but the N fertilizer rate that maximizes emission savings increases with C debt. For the production systems represented in the survey (which may not represent the real average of Indonesian palm oil production across all production conditions as the sampling design included self-nomination of companies), the reported N fertilizer rate of 141 kg N ha À1 yr À1 was substantially below the 'economic optimum' rate and likely achieving only 67% of attainable yield (as defined by the empirical Y max parameter), but using much less fertilizer than would be needed to achieve the maximum (444 kg N ha À1 ). However, the economic optimum estimate will be lowered if further risks (physical production, prices) are included in the model. A safety margin of a factor 6 on p has to be inferred to explain the average fertilizer level observed. The N fertilizer level used at each of the plantations might reflect the actual economic optimum for the type of planting material and local circumstances, which may not be the same across all plantations in the data set. Figure 5 is not the result of a controlled fertilizer experiment, but a summary of current fertilizer use and yields, where assessment of the life cycle average yield required extrapolation beyond the data (Fig. 4) and as such has some  uncertainty built in. Our assumption of a quadratic response model represents the most efficient side of the spectrum, with field-scale variability likely shifting towards Mitscherlich-type response curves with higher economic optimum fertilizer rates, and lower environmentally acceptable ones ( van Noordwijk & Wadman, 1992). Figure 8 shows the effects of spatial variability in the three parameters of the fertilizer-yield response. Variability in Y max (e.g. through variability in plant characteristics) has a stronger depressing effect on the response curve than variation in Y 0 or f; however the effect is strongest when all three are variable. Interestingly, effects become noticeable at fertilizer rates above 200 kg N ha À1 and relative yield levels at 80% of Y max . There may be space to increase yields from 67% to 80% of Y max before negative effects on the emissions footprint emerge. The recorded fertilizer rate is in the environmental optimum range that maximizes emission savings per unit of biofuel use if a 4% N 2 O-N/N fertilizer ratio is used, and below this level if a 1% N 2 O emission factor applies. As default we assumed no change in CH 4 sink strength between forests and wellmanaged plantations, which may not be true at high N fertilizer rates (Tate, 2015). The C debt (phase I) and N 2 O emission per unit N fertilizer use (phase III) are the two dominant parameters in the calculation. The first factor had been recognized before (Agus et al., 2013), the second not yet explicitly. Under the EU RED policy, conversion to oil palm before 2008 is not considered, so older plantations have zero C debt. This grandfather clause was not included in the construction of Fig. 6. Details of soil and crop management in phase III may influence results for N 2 O emissions. The realistic value of N 2 O-N/N fertilizer ratio may well be between the 1% estimate of IPCC (2006) and the 4% value proposed by Crutzen et al. (2008). Uncertainty about the true value of this parameter needs to be considered in the application of biofuel standards, but may apply across all crops. Further measurements of this ratio are a priority for research   (Reijnders & Huijbregts, 2008;Reijnders, 2011). Richards et al. (2016) compared a number of existing models and calculation schemes for N 2 O fluxes from tropical agricultural soils and found that the substantial variation in both space and time in measured fluxes is not adequately accounted for by any current model. The IPCC emission factors are at least calibrated to global average emission data, but there is opportunity to improve on both practice and accounting method. Phase IV of the life cycletransport and processingcan have a strong impact on the absolute levels of emission savings (Choo et al., 2011), but does not influence the environmentally optimum N fertilizer rate in phase II, as phase III is expressed per unit product. Methane capture at the mill, an established technology that is not yet widely used, can increase emission savings by about 10% across the board. Full utilization of biomass residues and by-products for static energy production, rather than focus on biofuel, can increase emission savings (Kamahara et al., 2008), but may interfere with the recycling of organic residues to the plantations, affecting the maintenance of soil organic matter. Further analysis of the IPOC/ICRAF data will clarify which current management practices risk a declining C org content, while overall levels can be just about maintained (Khasanah et al., 2015a). Maintaining C org content of forest-derived mineral soils is probably possible (Powers et al., 2011).
In the 100-400 kg N ha À1 yr À1 range that includes virtually all data points, the emission savings per unit biofuel respond weakly positive or weakly negative to the N fertilizer level. Given the uncertainties around the data, there is no strong argument for modifying fertilizer price policies as a measure to reduce emissions. The N fertilizer rates currently used are slightly below what would be 'environmental' optimum in most conditions. Overall, the data indicate that intensification through increases of fertilizer rates above current practice could increase yields from the current 67% to 80% of attainable yields without negative effects on the footprint of Indonesian palm oil. Our analysis showed that there is an intermediate level of intensification of palm oil production system, achieving between 67% and 80% of its potential, that maximizes the possible net emission savings when palm oil is used as biofuel. The C debt (phase I) and processing/transport parameters (phase IV) have an overriding impact on the net emission savings attributed to palm oil use, but phase III does not influence the level of intensification that minimizes emissions. Behavioural studies on fine-tuning management decisions matter for achieving sustainability goals in the oil palm industry (Choong & McKay, 2014), as well as elsewhere. In between the Borlaug hypothesis and ecological agriculture, intermediate levels of intensification need to be fine-tuned to match the emerging public policy standards. Fertilizer price instruments cannot, in this situation, be expected to secure environmental policy outcomes beyond what land use policies and market-based accountability for footprints can achieve.