Quantifying global greenhouse gas emissions from land-use change for crop production

Soil C stock

Changes in soil C stocks were calculated as follows:

The emissions calculated for changes in soil organic carbon (SOC) were allocated over 20 years, as explained below. Reference soil C stock values were the default C levels under native vegetation in the top 30 cm of the soil profile in t ha⁻¹, taken from IPCC (2006) for six different soil types across nine climate zones. LU stock change factors modify C stocks up or down, and account for the impact of management (e.g. tillage regimes) and inputs (crop residue or manure additions) where relevant, as well as LU type (native or managed forest and grassland, tropical shifting cultivation short or mature fallow, set aside, annual crops and permanent crops), and were also taken from IPCC (2006). The resulting C stock changes were converted to CO₂-eq ha⁻¹ yr⁻¹ for comparison, with a negative value representing an increase in soil C, that is, the soil acts as a C sink. The exception to this methodology was organic soils; where they are mentioned, annual C losses per-ha were not calculated from reference C stocks but rather taken directly from IPCC (2006) and again converted to CO₂-eq.

N₂O emissions from SOM loss

Where a decrease in the soil C stock occurred, the following calculation was used to convert the loss of soil C (as a measure of SOM loss) to N₂O emissions (as per IPCC, 2003), assuming a 15 : 1 C : N ratio and 1% emission factor for the proportion of N lost as N₂O (equivalent to losing 1 kg per 100 kg of N added used for N fertilizer additions):

where both soil C and N₂O-N are in t ha⁻¹ from the top 30 cm of the soil profile.

This value was again converted to annual CO₂-eq for comparison, using a global warming potential of 296 for N₂O (IPCC, 2001), which is the value used for national inventory assessments (to ensure consistency in estimates at different times), despite a more recent estimate of 310 (IPCC, 2007). These emissions are directly proportional to soil C emissions, making up 8% of total soil LUC emissions where SOM is lost as a result of land conversion and are therefore not considered separately in the results.

Biomass C stocks

Changes in biomass C stocks were calculated as follows:

The results were also allocated over 20 years as with soil C (see explanation below). Biomass C stock values representing total above and below ground, living and dead matter C stock, averaged over a production cycle where applicable, were taken from the EU RED guidelines (2010). These are, in turn, based on biomass dry matter stocks, C fraction, and root to shoot ratio data given by the IPCC (2006). As data were not available for all ecological zones some substitutions were made; for temperate steppe, temperate grassland values were used as the nearest vegetation type, and all tropical and subtropical mountain systems were treated as tropical mountain systems. For Australia, available biomass C values for each vegetation class from other regions were averaged, and for European subtropical dry forests, continental Asian values were used. In desert regions, it was assumed some level of reclamation would be necessary before crop production was possible so these ecological zones were excluded from the analysis of conversions from natural vegetation. For soy and oilseed rape, which are not covered in the EU RED report, the IPCC (2006) default value of 5 t C ha⁻¹ was used. As with soil C stock changes, a positive value shows a loss of C on conversion and a negative value indicates a C sink.

20-Year equilibrium rule

The IPCC methodology uses a 20-year period for soil C to equilibrate to management or LUCs. As explained in the Revised 1996 Guidelines (IPCC, 1996), this represents a compromise as systems vary in their response times. Tropical systems reach a new equilibrium faster than temperate ones, as do systems where soil C is being degraded in comparison with those where there is a build-up of soil C in response to land abandonment or increased residue inputs. In general, while a longer inventory period would mean systems were closer to equilibrium, the most rapid changes in soil C occur during the first 10–20 years following a significant change in management practices or land use. Therefore, 20 years is deemed an appropriate time period for the inclusion of most of the change in soil C stocks resulting from land conversions to agriculture and management-induced changes. At the same time, it limits the historical LU data requirements and the amount of bias introduced by the assumption of linear change over longer inventory periods (IPCC, 1996).

Biomass C changes are inventoried in the year they occur as the IPCC methodology is designed to calculate annual inventories of GHG. In contrast, product assessment methods, such as PCF or life cycle assessment (LCA), aim to ascribe the impacts from LUC to the products obtained following a principle of causality (whoever causes the LUC bears the burden). Thus, when land is converted to provide agricultural products, the impacts from LUC should be allocated to such products even if this class of emissions occur mainly in the first year. An allocation problem arises because there is usually no certainty over how many years the land will be used for that purpose. There is no scientific justification for choosing one allocation period or another, but as both PAS 2050 (BSI, 2008, 2011) and EU RED (EU, 2009, 2010) suggest allocating all LUC emissions (or fixation) over the 20 years following LUC, including both those derived from SOC degradation (or build-up) and biomass C stock loss (or gain), we have followed the same allocation principle.

Choice of crops and conditions

Three oil crops were selected – oil palm and soybean as examples of perennial and annual crops, respectively, which are currently implicated in high levels of LUC, potentially threatening biodiversity across Asia and South America (Koh & Wilcove, 2007), and oilseed rape as a temperate annual crop with similar food, biofuel and other uses for comparison. Further crop types (cocoa, tea, sugar, sunflowers and arable fruit and vegetables) are given in the Appendix S1. For each crop, combinations of climate, soil type and ecological zone were selected to reflect conditions within the FAO top 20 producing countries. All ecological zones except deserts were included provided they covered at least 1% of the total area of suitable ecological zones. Oil palm was assumed to be grown only in tropical regions and oilseed rape in temperate ones (except for in India, where temperate zones cover only 2% of the land area), whereas soy was assumed to grow across both tropical and temperate regions, and only boreal regions were excluded. No further considerations of suitability for crop production were taken into account. Previous land uses covered were natural vegetation/forest (where referred to as forest, shrubland and other nonforest ecological zones are excluded), improved grassland, and for tropical forest zones, short fallow shifting cultivation where vegetation has regenerated as far as grassland, and mature fallow where scrub has developed, with set aside grassland as a comparison for temperate regions. A range of management options were tested for each crop. Typical management was taken to be medium inputs and full tillage for the two annual crops as this is the default baseline for the IPCC methodology. The other two scenarios were full tillage and low inputs (crop residues removed and nutrients not replaced with fertilizer or N-fixing crops) (Scenario A), and reduced tillage (because no-till is not suitable for all crops or soil types) and high inputs (crop residues returned and significant additional organic matter via use of green manure, cover crops, etc. but not animal manure, as it is unclear how widely this is used) (Scenario B). For oil palm, no-till management was assumed for all scenarios (Wahid et al., 2005) but input levels were the same as for annual crops (low inputs for Scenario A and high inputs for Scenario B). These management scenarios reflect a range of likely organic matter inputs and disturbance affecting soil C stocks, not a full range of management options which may affect yield levels. Therefore, only per-ha emissions under typical management were used to assess per-yield emissions.

Per-yield/product-based emissions

Per-ha emissions were converted to per-tonne-of-product levels to reflect product-based climate impact. This was performed at country scale using average emission levels under typical management and crop oil yields. Two sets of calculations were performed; one using emissions for conversions from natural vegetation, and one using emissions for conversions from agricultural or formerly agricultural (set aside or fallow) land. For conversions from natural vegetation, per-ha emission levels under typical management were averaged for each country, using a proportional weighting based on the coverage of each vegetation type analysed. Ecological zones deemed climatically unsuitable for crop production were discounted such that 100% coverage represented all suitable areas rather than 100% of the land area of the country. Where multiple soil types occurred within the ecological zones, these were assumed to be equally distributed, as were crops and different types of agricultural or previously agricultural land. This is because a multilayered, high-resolution spatial database would be required to account for these distributions; the aim of this study was to produce a matrix which could be easily manipulated in a simple spreadsheet form. It means that these emissions have no spatial resolution below country level, which is in keeping with the use of a single yield value for each country, rather than multiple yield values taking into account the effects of climate and soil type within each country.

FAO cropping area and oil production data for 2008 were used to calculate oil yields for oil palm and oilseed rape (because yield values are only given for crops, not their oil). For soy, rather than assume all cropping area is used for oil production, FAO national production figures for soybeans were used and converted to oil production assuming 1 t of soybeans produces 0.19 t of vegetable oil (Brandão, 2011). FAO data were used for consistency as these are the most recent data available which cover all the countries considered. However, these yield levels are not always those expected from commercial growers, so it is recommended that grower data be used for finer scale applications of the methodology. This would also allow the effect of management scenarios on per-yield emissions to be properly assessed, as discussed below.

Country-scale per-product emissions were then averaged using a weighting according to the percentage of production from each of the top 20 producing countries to give two estimates of global impact, one based on the assumption that all LUC occurs on previously natural vegetation, and one assuming all new crop production occurs on land which is currently, or has recently been, under agricultural management. Both of these estimates assume that LUC is distributed according to national production levels, that is, that a country which produces 20% of the crop also contributes 20% of the LUC emissions. This is because the alternative is to use scenarios of LUC distribution which need to be based on complex agrieconomic modelling to be realistic. These final estimates, therefore, have a very high degree of associated uncertainty (see below), and are presented here to illustrate the purpose of the methodology rather than to provide an accurate estimate of global LUC emissions.

Area-based LUC emissions

Conversion of natural vegetation to oil palm under typical crop management on mineral soils results in the loss of −4.5 to 29.4 t CO₂-eq ha⁻¹ yr⁻¹ over 20 years, with tropical shrubland conversions in Africa and Central & South America acting as a C sink. The highest emissions are from tropical rainforests in the countries of insular Asia (Fig. 1a). This maximum emission level is within the range calculated by Fargione et al. (2008) for Malaysian and Indonesian rainforests not on peat soils (equivalent to 35 ± 10 t CO₂ ha⁻¹ yr⁻¹ if taken over 20 years). Note that because conversions to oil palm do not reduce soil C stocks under any of the conditions tested, these emissions do not include any N₂O. Conversion of natural vegetation to soy cropping under typical management emits between 1.2 and 47.5 t CO₂-eq ha⁻¹ yr⁻¹; with the lowest emissions from temperate dry steppe conversions in Europe and South America and the highest from tropical rainforests in the countries of insular Asia (Fig. 1b). Emissions from Brazilian rainforest converted to soy were calculated as 41.6 t CO₂ ha⁻¹ yr⁻¹, which is similar to previously published estimates for this land-use transition (Fargione et al., 2008; Reijnders & Huijbregts, 2008b). Conversion to oilseed rape production under typical management results in LUC emissions of 1.2–39.2 t CO₂-eq ha⁻¹ yr⁻¹, with the lowest emissions from warm temperate dry steppe conversions in Europe. The highest emissions are from tropical rainforest conversions in India, although this covers only 6% of the land area, and warm temperate moist subtropical humid forest conversions in the United States have the next highest emissions of 27.2 t CO₂-eq ha⁻¹ yr⁻¹ (Fig. 1c). Emissions ranges for all the previous LU types selected are given in Table 1 (note that for natural vegetation, all classes are included on the emission maps [forest, shrubland, natural grassland, etc.] but for simplicity, only forests are included in the table).

Emissions are higher from the two annual crops because postconversion biomass C stocks are much lower than for oil palm which is a long-term tree crop, and because continuous cultivation reduces soil C stocks for all the scenarios investigated, whereas no-till permanent crops are deemed to increase soil C stocks. This second point is an area of some contention as the IPCC acknowledge the methodology for tropical systems is based on fewer data points than that for temperate ones (IPCC, 2006), and more recent studies suggest it may overestimate the C sink strength of plantation soils. For example, measurements of SOC concentrations under plantations of different ages indicate high levels of spatial variation and no constant directional change over time (Smiley & Kroschel, 2008). Meanwhile, comparisons of soil C concentrations in top soil under primary and secondary forest and plantations have indicated LUC reduces soil C stocks at least at the surface (van Noordwijk et al., 1997; Schroth et al., 2000, 2002). A meta-analysis of SOC concentrations under different land uses in Brazil suggested that nonintensive cropping systems (including perennial crops and plantations) had the same SOC stocks as natural vegetation in general, but that coarse-textured soils under this management lost ca. 20% of their stored C (Zinn et al., 2005). Reijnders & Huijbregts (2008a) used direct CO₂ measurements made by Ishizuka et al. (2005) to infer a loss of 1.2 t CO₂-eq ha⁻¹ yr⁻¹ for South Asian forests converted to oil palm plantations in the first 20 years after conversion. Inclusion of N₂O emissions from this level of SOM loss would give total soil emissions of 1.4 t CO₂-eq ha⁻¹ yr⁻¹, which in turn would increase the LUC emissions calculated here for this region from a range of 5.6–29.4 to 8.9–32.6 t CO₂-eq ha⁻¹ yr⁻¹, assuming conversions from natural forest to typical management.

Previous land use

As shown in Table 1, previous land use can make a big difference to resulting C losses when land is converted. For oil palm, LUC emission levels do not overlap between categories of previous land use, even when comparing across different world regions. Only forest conversions to palm oil produce net LUC emissions; conversions of all the former agricultural land uses to palm oil produce net sinks of C. Conversion of grassland which is fallow after shifting cultivation provides the biggest sink of C, followed by conversion of improved grassland with, on average, 65% of the C sink strength of fallow grassland conversions (comparing emissions for different previous land uses from the same region, soil type and climate zone). Conversion of mature fallow scrubland provides the smallest sink, with a sink strength of 41% that of converted fallow grassland on average. Across all the conditions tested, forest conversion to oil palm cropping loses an average of 12.9 ± 9.0 t CO₂-eq ha⁻¹ yr⁻¹ (mean ± SD), while mature fallow conversions to palm oil are a sink of −6.4 ± 1.5 t CO₂-eq ha⁻¹ yr⁻¹. Conversions of improved grasslands to palm oil are a sink of −10.0 ± 0.1 t CO₂-eq ha⁻¹ yr⁻¹, and conversions of short fallow grasslands are a sink of −15.7 ± 1.6 t CO₂-eq ha⁻¹ yr⁻¹.

For the two annual crops, the picture is more complex; all the conversion scenarios tested result in a net loss of C and in general, forest conversions give the highest emissions, then mature fallow scrubland (where applicable) losing 62% less CO₂-eq from LUC on average (when compared with the same soil, climate and region), then improved grassland with LUC emissions 73% lower on average than forest conversions. The lowest emissions are from short fallow for tropical regions (95% lower than forest conversions), and set aside grassland for temperate regions (91% lower than forest conversions). This fits with previous work which showed conversion of abandoned agricultural land greatly reduces carbon losses upon conversion to biofuels (Fargione et al., 2008). There is some overlap in LUC emissions from different previous LU categories when compared across all the soil, climate and regions tested, however; forest conversion to annual arable cropping loses 24.8 ± 8.0 t CO₂-eq ha⁻¹ yr⁻¹, mature fallow 10.4 ± 1.3 t CO₂-eq ha⁻¹ yr⁻¹, improved grassland 6.5 ± 2.8 t CO₂-eq ha⁻¹ yr⁻¹, set aside grassland 2.1 ± 1.2 t CO₂-eq ha⁻¹ yr⁻¹, and short fallow 1.6 ± 1.0 t CO₂-eq ha⁻¹ yr⁻¹.

Crop management decisions

As shown in Fig. 2a, crop management decisions can have a strong impact on loss of soil C and N₂O from SOM under annual arable crops, with scenario B management (reduced tillage and high inputs) significantly reducing emissions upon conversion from natural vegetation. For oil palm, the relationship is even stronger with scenario B management producing a more than double strength of C sink, despite there being less difference between the two management strategies as both feature no tillage. However, as noted above, this is based on a methodology which may underestimate the impact of conversions to plantations on soil C stocks and assume they are more positive than more recent field data may suggest (see, e.g. Hertel et al., 2009; Li et al., 2011; Potvin et al., 2011).

When total LUC emissions are considered, the effect of management strategies is far smaller for transitions from natural vegetation with high standing above ground C stocks (forests) because the biomass component of LUC emissions generally outweighs the soil component (Fig. 2b). This is also reflected in Reijnders & Huijbregts (2008b), where tillage vs. no tillage management has little impact on land use and LUC emissions associated with soybean production on former Brazilian rainforest sites. However, for land-use transitions from grasslands to arable crops, soil emissions play a much bigger role in total LUC emissions, and overall crop management can be considered a key factor when estimating GHG emissions from LUC (Kim et al., 2009). Ideally, if yield levels were available for different management scenarios, for example, at finer spatial scales, it would be preferable to assess how these management scenarios affect emissions per tonne of oil production. This would, for example, allow land managers to assess tradeoffs in terms of possible yield loss if they switched to a reduced tillage regime, or supplied a greater proportion of the crop's N requirement with organic matter rather than mineral N fertilizers. However, without management-specific yields, these assessments are not possible.

Soil type

Soil type does not, in general, have a clear impact on LUC emission levels because initial soil C stocks are also dependent on climate zone and prior land use and management (see Fig. 3, which shows soil LUC emissions from conversions to oilseed rape and soybean cropping only, as oil palm is either C neutral or a sink for C when only soil LUC emissions are taken into account). Soils in dry regions tend to lose less C than those in wetter areas. Therefore, sandy soils, which tend to occur in dry regions, tend to have lower LUC emissions than spodic soils, which only occur in cool temperate moist zones. However, the most common soil types, high and low activity clays (HAC and LAC), occur across a wider variety of climate zones, so it is not possible to make any further generalizations about relative emission levels based on soil type. The exception to this is organic soils, which always lose the most C on conversion. These were not included in the analysis as they tend to occur in very localized pockets and, in the absence of fine resolution spatial data on coverage, would greatly skew the results. However, as the IPCC considers that cropland on organic soils, such as peat, loses 5–20 t C ha⁻¹ yr⁻¹, depending on climate zone, converting these emissions to CO₂-eq and including N₂O emissions from SOM loss, gives LUC emission levels just for soils of 20–80 t CO₂-eq ha⁻¹ yr⁻¹ (assuming LUC from undrained peat to cropping, as opposed to forest or grassland on already drained peat to cropping). This means that growing palm oil on a peat rainforest in Malaysia or Indonesia would lose 111 t CO₂-eq ha⁻¹ yr⁻¹, a total of 2223 t CO₂-eq ha⁻¹ over 20 years, which is at the higher end of the range of 1294 ± 2158 t CO₂ ha⁻¹ calculated by Fargione et al. (2008).

Emissions per tonne of production for top 20 producing countries

Using FAO yield data and averaging emissions across soil, climate and ecological zones, gives an emission level per tonne of vegetable oil production by country, assuming either conversion from natural vegetation and typical management. These values can then be averaged using a weighting based on country production levels to give an estimated global contribution to LUC emissions. These estimates assume LUC is distributed according to national production levels, that is, that if a country grows 20% of global oil palm, it contributes 20% of global LUC emissions. For palm oil on mineral soils, national emissions vary from 1.0 to 105.6 t CO₂-eq t oil production⁻¹ (see Table 2), although this highest value is for Guinea, with a subsistence farming yield level. This gives a weighted global average of 7.0 t CO₂-eq t oil production⁻¹ for oil produced on land converted from natural vegetation. In comparison, assuming the previous land use is agricultural (fallow after shifting cultivation or improved grassland) gives a weighted global average of −3.1 t CO₂-eq t oil production⁻¹, demonstrating that these land-use transitions can act as a C sink, and can offset emissions from conversion of forest land. Reijnders & Huijbregts (2008a) calculated an approximate value of 5.8 t CO₂-eq t oil production⁻¹ for South Asian oil palm plantations on former rainforest on mineral soils, which is lower than those calculated here for insular Asia, but only because they used a higher yield value and divided emissions over 25 years, representing the lifespan of the plantation.

Land-use change emissions for soybean oil vary from 29.2 to 149.5 t CO₂-eq t oil production⁻¹ on land converted from natural vegetation on a national basis, giving a weighted global average emission of 50.6 t CO₂-eq t oil production⁻¹ (see Table 3). The national emission levels for soybean oil have a greater uncertainty around them than for the other two crops because a single conversion factor was used for oil production per t of soybeans to avoid the assumption that all the crop area is used for oil production. Considering only land-use transitions from improved grassland and either shifting cultivation fallow for tropical regions, or set aside grassland for temperate regions, reduces this global average emission to 11.9 t CO₂-eq t oil production⁻¹. For rapeseed oil, national LUC emissions vary from 10.2 to 457.3 t CO₂-eq t oil production⁻¹, giving a weighted global average of 31.4 t CO₂-eq t oil production⁻¹ when natural vegetation is converted (see Table 4). When land-use transitions from improved or set aside grassland are considered, this weighted global average drops to 7.7 t CO₂-eq t oil production⁻¹.

Sources of uncertainty

Use of the IPCC default methodology (recommended when country-specific data are unavailable) means there is a high level of uncertainty associated with the results of this study, but because a consistent data set was used for all conditions, specific bias was avoided. The variation in biomass C stock estimates in natural vegetation differs depending on the ecological zone and geographical region, while the default biomass C stock of annual crops has a 75% error associated with it (IPCC, 2006). Default reference soil C stocks have a nominal estimate of 90% error associated with them (IPCC, 2006). The soil stock change factors for land use, tillage and inputs generally have error levels of 4–14% associated with them but tropical annual cropping, shifting cultivation fallow, and all activities in tropical montane regions have a much higher error of 46–61% (IPCC, 2006). Further uncertainty is added to per-tonne-of-production emission levels by using FAO cropping area and production data to calculate yields, especially for soybean where crop yields were converted to oil yields using a single factor for all countries to avoid the assumption of all cropping area being used for oil production.

These error levels are largely estimates based on expert knowledge and not suitable for conversion to a single error value for the results presented here. However, a sensitivity analysis to investigate the impact of assuming equal crop distribution within countries, and of using default soil and biomass C input data is described below.

Crop distribution

The impact of assuming equal crop distribution within countries was investigated by comparing global average LUC emissions per tonne of oil production calculated using the lowest and highest per ha LUC emissions for each country. This gives a minimum and maximum level of GHG impact for natural vegetation conversions to crop production. For palm oil, this gives a range of 2.4 to 8.0 t CO₂-eq t oil production⁻¹ for conversions from natural vegetation, representing 34–114% of the value shown in Table 2. This reflects the fact that much of the main producing countries is covered by ecological zones with high biomass C stocks, such as tropical rainforest, and therefore the equal distribution scenario is much closer to the worst-case scenario than the minimum emissions scenario. For soybean, the range is 17.3–78.2 t CO₂-eq t oil production⁻¹, representing 34–156% of the value shown in Table 3. For oilseed rape, the range is 12.2–49.0 t CO₂-eq t oil production⁻¹, representing 39–156% of the value shown in Table 4. These values indicate that assuming LUC is equally distributed within countries could be overestimating LUC emissions by as much as 66% in some cases but is by no means the worst-case scenario, especially for the annual crops.

The assumption that LUC is distributed across countries according to their current level of crop production is the single biggest source of uncertainty in the global LUC emissions per-product estimates. However, complex agrieconomic modelling would be required to improve on this, and the values are included here illustrate the possibilities based on the range of assumptions outlined.

Biomass C input data

This study used regional defaults for broad classes of vegetation but local knowledge of biomass C stocks could greatly improve the estimates made using the same methodology. For example, the National GHG Inventory Report for Brazil (Brazil, 2010) gives a range of biomass C stocks for 6 different forest classes within the Amazon rainforest area (including submontane but not montane forest or scrubland) which is covered by a single ecological zone in this methodology. Substituting the minimum and maximum values of these ranges for the default biomass C value, gives per ha emissions of 19.2–56.1 t CO₂-eq ha⁻¹ yr⁻¹ upon conversion of Amazon rainforest to soybean production using typical management, in comparison with 41.6 t CO₂-eq ha⁻¹ yr⁻¹ calculated here. This single change makes Brazil's weighted average LUC emissions vary between 44.6 and 78.4 t CO₂-eq t oil production⁻¹ for natural vegetation conversions, giving a range of 46.4–53.0 t CO₂-eq t oil production⁻¹ for the global average. This represents 92–105% of the average calculated here (shown in Table 3). The large impact of these biomass C stocks is reflected in Hiederer et al. (2010) as they found the removal of biomass contributed ca. 80% of the LUC emissions calculated for a scenario, where most of the additional cropland was assigned to Brazil.

Soil C input data

In the background material to the recent European Commission study (Carré et al., 2010; Hiederer et al., 2010), default soil C stocks under native vegetation used by the IPCC methodology and this study are updated using the latest complete global dataset of soil parameters – the Harmonized World Soil Database. Substituting these new values into the calculation of LUC emissions for conversion from natural vegetation to oilseed rape production under typical management gives a range of per ha emissions from 0.7 to 38.0 t CO₂-eq ha⁻¹ yr⁻¹ in comparison with the values shown in Fig. 1c (the minimum emission level for forest conversions as shown in Table 1 is 15.1 t CO₂-eq ha⁻¹ yr⁻¹). Table 5 shows how these changes reduce the national average per ha emissions by 3–15% and the global average LUC emissions for this LU transition by 9%. This indicates that improving soil C stock change factors would probably have a greater impact in terms of reducing the uncertainty around soil LUC emissions than using different soil C stocks prior to conversion.

Implications for carbon footprinting

In 2009, at least 13 different methodologies for calculating carbon footprints were in use or development (Plassmann et al., 2010), many with differing boundaries and assumptions to account for LUC impacts, which can result in significantly disparate carbon footprint results with differing degrees of variability. The PCFs which omit LUC impacts may fail to account for a substantial portion of the product's true contribution to climate change. This is most pronounced for products containing tropically produced agricultural materials from developing countries, where recent deforestation for oil and food crops is widespread (Fargione et al., 2008; Gibbs et al., 2010), and has occurred within the last 20 years. For example, the Ecoinvent database suggests a value of ~1.7 t CO₂-eq t palm oil⁻¹ (Ecoinvent, 2007) over the whole life cycle of oil production excluding any considerations of LUC. Adding the global average emissions from LUC calculated in this study would therefore increase the PCF of palm oil by more than fivefold if all oil came from recently cleared forest land. On the other hand, palm oil would become a net C sink if all the plantations were previously agricultural or fallow land. This also means that tropically produced agricultural raw materials will often have an inequitable emissions burden compared to agricultural materials from developed countries where LUC has occurred well over 20 years ago (Brenton et al., 2010; Cederberg et al., 2011). Moreover, limited data from developing countries make accurate LUC accounting more contentious, as illustrated by the higher level of error associated with soil C changes under agricultural management discussed above. The first version of PAS 2050 (BSI, 2008) methodology advocated that highest tier (IPCC, 2006) available data be used where possible and if this is unknown, a ‘worst in class’ approach should be taken, represented by the conversion of tropical forest to annual cropland in Malaysia. In this instance PCFs can be unfairly overestimated, by up to 1900% (Plassmann et al., 2010). Recent revisions to PAS 2050 (BSI, 2011) recognize this as ‘overly severe’ and replace the worst-case default value with a tiered approach using land-use factors based on country of sourcing or countries of global production.

This has important implications for companies who are increasingly seeking to quantify, and in many cases communicate, the GHG impact of their products. Nilsson et al. (2010) conducted an LCA comparing butter and margarine and demonstrated the significance of including LUC for palm oil when accounting for the GHG emissions associated with margarine. The carbon footprint of margarines containing a high proportion of palm oil and palm kernel oil (ca. 50%) was found to decrease by at least 25% if LUC was not considered when compared to palm oil coming solely from former tropical forest land. No LUC was considered for the other oils used to produce margarine. However, this study suggests that should LUC be taking place for the other oils, then significant additional C emissions could occur. Should a proportion of palm oil be sourced from land already under agricultural management or formerly so, then there may be no LUC emissions or even negative emissions (i.e. sequestration) associated with this palm oil (Table 1). LUC scenarios for butter, which would largely be associated with land required for feed production for cows, were also not considered in Nilsson et al. (2010). This could result in a significant underestimation of the PCF for butter where, for example, soy-based feeds are used (Table 3; see also Cederberg et al., 2011). Brenton et al. (2010) also highlighted the importance of using country-specific data and the potential for huge inflation of the GHG estimate when this was unknown due to the severe default LUC values. They present the sensitivity of carbon footprints to a number of parameters, including loss of soil C from management practices and electricity emission factor used, but none are more significant than the factors relating to LUC (Brenton et al., 2010).

We use readily available methods and data sources to develop a framework to estimate the LUC components of PCFs. We illustrate the utility and limitations of this framework by assessing the LUC emissions from three oil crops globally. We show that the framework can be used in all global regions, and also highlight where finer resolution data and information (particularly on land management and yield) could improve reliability of the estimates. Frameworks operating at higher tiers (region specific soil C change factors or process-based models, and high-resolution, spatial data) are desirable to reduce the uncertainties identified using this approach, but the framework presented represents an important step forward for including LUC emissions in PCFs.