for agricultural commodities applications to Brazilian ...

Global Change Biology Global Change Biology (2014), doi: 10.1111/gcb.12635

A method for calculating a land-use change carbon footprint (LUC-CFP) for agricultural commodities – applications to Brazilian beef and soy, Indonesian palm oil U . M A R T I N P E R S S O N 1 , S A B I N E H E N D E R S 2 and C H R I S T E L C E D E R B E R G 1 1 Physical Resource Theory, Chalmers University of Technology, G€oteborg 412 96, Sweden, 2Centre for Climate Science and Policy Research (CSPR), Link€oping University, Link€oping 581 83, Sweden

Abstract The world’s agricultural system has come under increasing scrutiny recently as an important driver of global climate change, creating a demand for indicators that estimate the climatic impacts of agricultural commodities. Such carbon footprints, however, have in most cases excluded emissions from land-use change and the proposed methodologies for including this significant emissions source suffer from different shortcomings. Here, we propose a new methodology for calculating land-use change carbon footprints for agricultural commodities and illustrate this methodology by applying it to three of the most prominent agricultural commodities driving tropical deforestation: Brazilian beef and soybeans, and Indonesian palm oil. We estimate land-use change carbon footprints in 2010 to be 66 tCO2/t meat (carcass weight) for Brazilian beef, 0.89 tCO2/t for Brazilian soybeans, and 7.5 tCO2/t for Indonesian palm oil, using a 10 year amortization period. The main advantage of the proposed methodology is its flexibility: it can be applied in a tiered approach, using detailed data where it is available while still allowing for estimation of footprints for a broad set of countries and agricultural commodities; it can be applied at different scales, estimating both national and subnational footprints; it can be adopted to account both for direct (proximate) and indirect drivers of land-use change. It is argued that with an increasing commercialization and globalization of the drivers of land-use change, the proposed carbon footprint methodology could help leverage the power needed to alter environmentally destructive land-use practices within the global agricultural system by providing a tool for assessing the environmental impacts of production, thereby informing consumers about the impacts of consumption and incentivizing producers to become more environmentally responsible. Keywords: beef, Brazil, carbon footprint, deforestation, Indonesia, land-use change, palm oil, soybeans Received 13 February 2014; revised version received 2 May 2014 and accepted 12 May 2014

Introduction The world’s agricultural system has received increasing attention over the last few years as a significant driver of global environmental change, including climate change, biodiversity loss, and degradation of land and freshwater (Foley et al., 2011). A large share of the negative impact from agriculture arises from its contribution to land-use change (LUC), primarily tropical deforestation. Approximately 10 million hectares (Mha) of tropical forests are lost annually (Lindquist et al., 2012; Hansen et al., 2013), leading to carbon dioxide (CO2) emissions of 3.0 1.1 GtCO2 yr1 (Harris et al., 2012a) and constituting the single largest threat to terrestrial ecosystems and biodiversity (Millenium Ecosystem Correspondence: U. Martin Persson, tel. +46-31 772 2148, fax +46-31 772 3150, e-mail [email protected]

© 2014 John Wiley & Sons Ltd

Assessment, 2005). Over 70 per cent of this forest loss is driven by expanding agriculture (Hosonuma et al., 2012; Houghton, 2012). There is a growing literature on the importance of distant drivers of LUC, primarily facilitated by rapidly increasing international trade of agricultural commodities (Meyfroidt et al., 2013). The globalization and commercialization of deforestation drivers offers new opportunities for nature conservation, as actors in global supply chains become increasingly sensitive to the environmental concerns of consumers and pressures from conservation NGOs (Rudel et al., 2009) and there are already examples of how consumer preferences have affected land-use practices in the tropics (e.g. ecocertification of coffee (Rueda & Lambin, 2013) and the Brazilian Soy Moratorium (Rudorff et al., 2011)). This development has spawned an interest in estimating carbon footprints of products (CFPs) for 1

2 U . M A R T I N P E R S S O N et al. different agricultural commodities. CFPs provide producers and consumers with information about the contribution of a given product to greenhouse gas (GHG) emissions and allow the identification of mitigation options; i.e. substitution between different products for consumers and improved production practices for producers. However, at present there is no established consensus methodology for calculating GHG emissions from LUC in CFPs and consequently

most CFPs of agricultural commodities have so far not included these effects. Results from a number of recent studies with different approaches to estimate these emissions vary widely due to differences in the methodological approach, see Table 1. A distinction can be made between studies that take a direct vs. an indirect approach to allocating emissions from LUC to products. Direct LUC studies (e.g. Reijnders & Huijbregts, 2008a,b; Cederberg et al., 2011;

Table 1 Summary of reported carbon footprints for a selection of agricultural commodities linked to land-use change. Results differ widely, depending on the type of methodology (direct vs. indirect) adopted and assumptions regarding the land use being replaced Carbon footprint (tCO2/t product) Product/Region Soybean meal Argentina Brazil

Non-EU Soybeans Argentina Brazil

Non-EU Beef Amazon Brazil Palm oil Indonesia Malysia South Asia

LUC

0.93* 7.69§ 0.63k 0.17–0.18** 3.05 0.40 0.40–8.67 6.65 4.14‡‡ 12.37§§ 0.78 0.55 3.77‡‡ 0.37–7.91 726 44 24 8.00§§ 3.22kk 6.10§§ 1.16kk 7.25***

Non-LUC

Methodology

Reference

0.72† 0.48††

Direct‡ Direct‡ Direct‡ Direct‡ Indirect (national) Indirect (global) Indirect (national)

(Opio et al., 2013) (Opio et al., 2013) (Meul et al., 2012) (Middelaar et al., 2012) (Middelaar et al., 2012) (Middelaar et al., 2012) (Leip et al., 2010)

0.64†

Direct‡ Indirect (national) Direct‡ Direct‡ Indirect (global) Indirect (national) Indirect (national)

(Flynn et al., 2012) (Ponsioen & Blonk, 2012) (Flynn et al., 2012) (Meul et al., 2012) (Meul et al., 2012) (Ponsioen & Blonk, 2012) (Leip et al., 2010)

Direct‡ Direct‡

(Cederberg et al., 2011) (Cederberg et al., 2011) (Opio et al., 2013)

Direct‡ Indirect (national) Direct‡ Indirect (national) Direct‡

(Flynn et al., 2012) (Ponsioen & Blonk, 2012) (Flynn et al., 2012) (Ponsioen & Blonk, 2012) (Reijnders & Huijbregts, 2008b)

0.32–0.36††

28 32.4 2.2†††

*Assumes that 22% of soybean expansion occurs on forested land, 31% on shrubland, and 44% on cropland. †(Dalgaard et al., 2008). ‡20 year amortization. §Assumes that all soybean expansion occurs in natural forest areas. kAssumes that 3.2% of soybean expansion occurs on forested land, 5.2% on shrubland. **Assumes that 1% of soybean expansion occurs on forested land, 3.4% on shrubland. ††(Prudêncio Da Silva et al., 2010). ‡‡Converted from the reported carbon footprint per ha using the average soybean yield in 2010 from FAOSTAT. §§Conversion from ‘natural vegetation’ (a weighted average over all vegetation types suitable for soybean cultivation). kkConverted from the reported carbon footprint per ha using the average soybean yield 2010 from FAOSTAT and allocating all emissions to palm oil. ***Based on conversion of forest to oil palm plantations and recalculated from 25 to 20 years amortization. †††(Schmidt, 2010).

© 2014 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12635

A LUC-CFP METHOD FOR AGRICULTURAL COMMODITIES 3 Flynn et al., 2012) take a bottom-up approach, attributing the GHG emissions from land clearing directly to the commodities produced on the cleared land over a given time period (amortization time), possibly accounting for yield and land-use dynamics following clearing. The indirect approach is more top-down. It allocates LUC to different products based on their relative contribution to total cropland expansion or total cropland area (e.g. Ponsioen & Blonk, 2012; Weiss & Leip, 2012), not necessarily to the commodities that are produced on the cleared land. There are two main concerns with the direct approach, the first being that results are highly sensitive to the amortization time, the choice of which is ultimately arbitrary and will have to be determined in a political context (Cederberg et al., 2011; Flynn et al., 2012; Ponsioen & Blonk, 2012). Secondly, the estimated CFP only applies to products originating on recently cleared land, and consumers will in most cases have no information about the exact origins of a product, although some studies calculate country-averaged LUC-CFPs based on the share of expansion of a given crop causing LUC. Also, land-use linkages and displacement effects imply that it is the total demand for a product that determines its contribution to LUC; e.g. it can be argued that it is the increased demand for Brazilian beef that drives the expansion of pastures into the Amazon, regardless of whether or not a given country or consumer sources his or her beef from the Amazon region (Cederberg et al., 2011). Accounting for these land-use linkages is the rationale behind the indirect approach to estimating LUC-CFPs. The advantage of the indirect approach is that it sidesteps the issue of amortization period (by relating LUC in a given time period to the production in the same period), it is based on easily available data, and aims to account for land use displacement effects. However, there are also severe disadvantages. First, the circumvention of the amortization issue hides the issue of how the time-lag between LUC and production on the cleared land affects the estimated CFP. Second, most of the proposed methodologies do not utilize any real data on the clearing of natural vegetation, but rely on simplistic assumptions (e.g. cropland expansion is randomly spread over existing natural vegetation types; Ponsioen & Blonk, 2012) or scenarios (Weiss & Leip, 2012) that are inconsistent with empirical data on the proximate drivers of LUC. Third, the implicit assumption underlying the indirect approach to estimating CFPs is that global or national agricultural commodity markets are perfectly integrated and that different agricultural crops are perfect substitutes both on the supply and demand side. As a general assumption this cannot be theoretically or

empirically defended; with agricultural land for crops and pasture being highly heterogeneous, a change in the demand for one product will not necessarily lead to changes in land availability for another. Nor do consumers treat agricultural commodities from different countries as perfect substitutes, implying different LUC impacts depending on where a change in demand for land-based products occurs (Kløverpris et al., 2010). Given that the LUC emissions can constitute a major share of the CFPs of commodities that are linked to LUC (see Table 1), developing a robust and transparent methodology for estimating LUC-CFPs is crucial. Here, we propose a new LUC-CFP methodology that is less sensitive to an arbitrary amortization period, is based on our current understanding of LUC processes (including the possibility to weigh direct vs. indirect drivers of LUC), and can be applied in a tiered approach (i.e. adapting the level of detail to the availability of data across countries and commodities). This enables the estimation of LUC-CFPs for a wide range of agricultural commodities and countries using data that is readily available without significant loss of precision. We illustrate this methodology by applying it to three of the main agricultural commodities associated with forest clearing in the tropics: beef and soy in the Brazilian Amazon and Cerrado and palm oil in Indonesia.

Materials and methods This section presents a methodology for estimating LUC-CFPs for agricultural (and other land-based) products associated with clearing of natural vegetation in two steps: (i) by calculating the LUC carbon footprint for production that occurs on cleared land, and (ii) by accounting for the share of production in a region originating from cleared land. Possibilities for simplifying the methodology so that it can be implemented in data-scarce environments (which are often the case in the tropics) are then explored, before turning to the data used for applying the methodology.

The carbon footprint of products originating from cleared land The carbon footprint of agricultural products originating from cleared land is calculated using a generalized adaptation of the direct LUC methodology presented by Cederberg et al. (2011). For a given biome n, region i (e.g. a country, state, or municipality) and product j, the carbon footprint (in tC t1 product) at year s is calculated as follows: DCn;i;j di;j ELUC n;i;j;s ¼ PsþT s¼s ai;j;ss yi;j;s

ð1Þ

Here, DCn,i,j is the carbon loss associated with clearing native vegetation in biome n in region i for agricultural land producing product j (in tC ha1); di,j is a factor that allocates


4 U . M A R T I N P E R S S O N et al. the carbon emissions between different products generated from the cleared land (e.g. between different agricultural products produced from the same crop or in intercropping systems) based on the revenues received for the different products; ai,j,s[s 2(0,1,. . ..,T]) is a crop and region specific factor that accounts for land use and yield dynamics over time (e.g. fallow requirements, land degradation, and abandonment processes, or yield variation over the rotation period for perennial crops) following forest clearing for product j; yi,j,s is the average yield of product j on cleared land in region i at time s (in t ha1 yr1), and T is the amortization period (in years). Put in words, Eqn (1) calculates the carbon footprint of product j produced from cleared land by distributing carbon stock changes due to LUC evenly over the total production over the T years following clearing. For example, if clearing tropical forests for pasture releases 400 tCO2 ha1, on average only half of the cleared land remains in productive pasture over time (i.e. ai,j,s = 0.5), and the average beef yield is 0.04 t ha1 yr1, the LUC-CFP will be 1000 tCO2 t1 beef if evaluated over a 20 year amortization period.

An average LUC-CFP for land-based products from a region As noted in the introduction, the LUC-CFP calculated using Eqn (1) will be highly sensitive to the choice of amortization period, T. Moreover, Eqn (1) calculates the carbon footprint solely of production originating from land deforested in the last T years and in most cases consumers (e.g. individuals, retailers, and companies) will not be able to acquire information about whether a given product has been sourced from recently deforested land or not. A more reasonable approach is therefore to estimate the average LUC-CFP for a given agricultural commodity and region (e.g. beef from Brazil or palm oil from Indonesia, or for individual states or municipalities in these countries). This is done by adjusting the LUC-CFP in Eqn (1) by a factor that accounts for the share of production of commodity j in region i that originates from land cleared in the last T years in the following way: P Ps LUC n s¼sT Dn;i;s rn;i;j;s ai;j;ss yi;j;s En;i;j;s av ð2Þ Ei;j;s ¼ Pi;j;s Here, Dn,i,j,s, rn,i,j,s and Pi,j,s are, respectively, the annual rate of clearing in biome n (in Mha), the share of cleared land that is dedicated to the production of product j, and total production (in tons) of product j, in region i in year s. Note that rn,i,j,s can be based both on data of the proximate drivers of deforestation (i.e. the land uses replacing natural vegetation), estimates of the indirect drivers of LUC in a region, or a mix of the two. As can be seen in Eqn (2), if there are different types of vegetation with differing carbon stocks cleared for the same product – e.g. the clearing of both humid forests and savannah vegetation for beef and soy production in Brazil – a compound LUC-CFP is calculated by estimating the numerator in Eqn (2)

for both types of vegetation, adding them and dividing by total production. Equation (2) basically calculates the total LUC emissions associated with production of j at time s that comes from land cleared 1, 2, . . ., to T years ago, and then distributes this over total production of j in region i at time s. In other words, if we make the simplifying assumption that the carbon footprint (ELUC n;i;j;s ) is stable over time we can express Eqn (2) as X Eav ELUC an;i;j ð3Þ i;j;s ¼ n n;i;j where an,i,j is the share of total production that comes from land cleared in biome n in the last T years. Following up on the beef example from above, if the deforestation rate has been a steady 1.25 Mha the last 20 years, 80% of this is due to expanding pastures and total national beef production is 10 million tons, the average LUC-CFP is 40 tCO2 t1 beef (i.e. 4% of beef production originates from land cleared in the last 20 years, this having a direct carbon footprint of 1000 tCO2 t1 beef). An important property of the expression in Eqn (2) is that while a larger T will imply a smaller carbon footprint for production occurring on deforested land (ELUC n;i;j;s ), it will also in most cases imply that more of production occurs on recently cleared land, since the total amount of land cleared in the last T years increases with T. These two effects counteract each other and consequently the resulting LUC-CFP will tend to be less sensitive to assumptions regarding the amortization period than the direct LUC-CFP expressed in Eqn (1). Note also that by simply removing the changes in carbon stocks (DCn,i,j) in the above LUC-CFP expressions [Eqns (1-2)] one can also calculate a LUC footprint (LUC-FP) that can be used to relate production of a given agricultural commodity to other impacts of LUC than just carbon emissions, such as biodiversity loss.

Simplified expression for the proposed average LUC-CFP Estimating LUC-CFPs using the expression in Eqn (2) requires a lot of data input. Especially detailed data on yields on cleared land (that may differ substantially from national averages) and land use and yield dynamics over time may not be available for many tropical countries experiencing rapid LUC. We therefore derive a simpler expression approximating the full LUC-CFP specification proposed above which relies mainly on data that is readily available for a broad set of countries and commodities: P ESi;j;s ¼

n ½DCn;i;j

di;j

Ps

s¼sT

Pi;j;s

Dn;i;s rn;i;j;s =T

ð4Þ

It can be shown (see derivations in Supporting Information) that this expression and the full specification [Eqn (2)] are equivalent if land clearing for a given commodity or average yield and land use/yield dynamics (ai,j,s) are constant over time. The more LUC and land use/yield dynamics vary, the poorer this approximation will be of the full specification.


A LUC-CFP METHOD FOR AGRICULTURAL COMMODITIES 5

Materials – applications to Brazilian beef and soybeans, Indonesian palm oil We illustrate the proposed LUC-CFP methodology by applying it to three of the main agricultural commodities linked to deforestation in the tropics: beef and soybeans from Brazil and palm oil from Indonesia. These two countries together incurred 55% of total emissions from tropical deforestation between 2000–2005 (Harris et al., 2012b). The analysis of the LUC-CFP for Brazilian beef and soy also includes the emissions associated with clearing of Brazilian savannah (Cerrado) vegetation for pastures and cropland. Clearing rates in the Cerrado have historically rivalled those of the Brazilian Amazon and soybean cultivation primarily has expanded in the Cerrado biome. However, the carbon content of Cerrado vegetation is substantially lower than that of Amazon forests and there are larger uncertainties in estimates of clearing rates and, more importantly, quantification of the direct drivers of LUC. The data sources used to calculate the LUC-CFPs according to Eqns (1), (2) and (4) are summarized here but a detailed description and the full data set can be found in the Supporting Information. Deforestation data for the Brazilian Amazon was obtained from the PRODES programme that has provided reliable remotely sensed estimates of forest loss in the Amazon on an annual basis since the late 1980s (Inpe, 2013). Clearing rates in the Cerrado biome are based on Klink & Moreira (2002), Machado et al. (2004), and Bustamante et al. (2012). For Indonesia, we base the deforestation data on a mix of recent remote sensing studies (Hansen et al., 2009; Miettinen et al., 2011); and a synthesis of national and international statistics (FAO, 2010; Wicke et al., 2011). Primarily we adopt a direct LUC approach here, based on remote sensing analyses of the extent to which natural vegetation is being replaced by pastures and soybean cultivation in Brazil (Brown et al., 2005; Morton et al., 2006; Galford et al., 2010; Rudorff et al., 2011; Bustamante et al., 2012) and oil palm plantations in Indonesia (Carlson et al., 2012; Lee et al., 2014), supplemented by additional data sources due to incomplete spatial and temporal coverage of the remote sensing data. The deforestation rates and direct (proximate) driver (rdirect n;i;j;s ) assumptions are displayed in Fig. 1.

(a)

To be able to analyse how sensitive estimated LUC-CFPs are to the choice of approach (direct vs. indirect) to attribute LUC and the associated carbon emissions to agricultural production, we also estimate an indirect LUC allocation factor (rindirect n;i;j;s ) based on the approach proposed by Weiss & Leip (2012). This indirect allocation factor is calculated by dividing annual changes in area devoted to produce a given commodity (i.e. pastures for beef, cropland for soy in Brazil, and oil palm plantations in Indonesia) by the total change in agricultural area for all commodities exhibiting an increase in area. We estimate above and below ground biomass in natural vegetation, deducting the carbon lost due to logging prior to clearing, arriving at an average carbon density of 125 tC ha1 for the Brazilian Amazon (based on Aguiar et al., 2012), 35 tC ha1 for the Cerrado (Batlle-Bayer et al., 2010), and 192 tC ha1 for Indonesian forests (see SI for sources). We also account for changes in soil organic carbon (SOC), as well as the carbon losses from drainage of peatlands for oil palm cultivation in Indonesia (Lee et al., 2014). For the land uses replacing natural vegetation, we assume average carbon densities of zero for cropland, 13 tC ha1 for the landscape following clearing for cattle ranching (being a mix of productive and abandoned pastures, the latter with biomass regeneration) (Cederberg et al., 2011), and 48 tC ha1 for palm oil plantations (averaged over a 25 year rotation period) (Reijnders & Huijbregts, 2008b). For Brazil carbon emissions due to large-scale cropping are allocated between soybean and corn production (the dominating crop rotation) based on the prevalence of double cropping and the gross revenues from each crop. As a result of increased double cropping in recent years (Arvor et al., 2012) the share of emissions allocated to soybeans decreases from 91% to 77% over the 2000–2010 period. Similarly, Indonesian carbon emissions due to oil palm expansion are allocated between palm oil and palm kernel oil and expellers based on relative yields and market prices, with a resulting 84–86% of emissions being allocated to palm oil. Land use and yield dynamics following deforestation are based on Fearnside (1997), Brown et al. (2013, 2005);) and Persson & Azar (2010), for cattle ranching, soybean cultivation, and oil palm plantations, respectively. Apart from pasture degradation and abandonment (reflected in the average

(b)

Fig. 1 Annual deforestation rates in Brazil (Amazon and Cerrado biomes) and Indonesia, divided between clearing for cattle ranching, soybean cultivation, and other land uses (primarily small-holder agriculture) in Brazil and between oil palm plantations and other land uses in Indonesia. See Supporting Information for details and sources. © 2014 John Wiley & Sons Ltd, Global Change Biology, doi: 10.1111/gcb.12635

6 U . M A R T I N P E R S S O N et al. biomass stock for forest land cleared for cattle) we do not account for land-use successions following deforestation here. For instance, conversion of pastures to soy plantations (e.g. in the frontier areas of Mato Grosso, Brazil) could be accounted for by adjusting the allocation factor (di,j) for beef, resulting in a larger share of LUC allocated to soy. Although we do not have information on the average time for this succession across the Amazon, we argue that at least for shorter amortization periods (5–10 years) this omission does not have a large effect on the results presented for Brazilian beef and soy. Also, we exclude LUC between agricultural uses. Again, soy expanding into pastures can lead to losses in soil carbon, implying a slight underestimation of the LUC-CFP for soy here. For palm oil replacing rubber plantations or agroforestry systems the effect is most likely small, though, as these agricultural systems have similar carbon densities. Data on yields are taken from Cederberg et al. (2011) for Brazilian beef, Instituto Brasileiro de Geografica Estatıstica (IBGE; http://seriesestatisticas.ibge.gov.br) for Brazilian soybeans, and FAOSTAT (http://faostat3.fao.org) for Indonesian palm oil. Production data for cattle meat [in carcass weight (CW) equivalents], soybeans and palm oil are also taken from FAOSTAT.

(a)

(b)

Results Figures 2,3 display the calculated land-use change carbon footprints for Brazilian beef and soybean and Indonesian palm oil for the year 2010. As expected, the LUC-CFP for produce originating from cleared land (ELUC) is highly sensitive to the choice of amortization period (Fig. 2). While the LUC-CFP proposed here (Eav) still display some variability with respect to the amortization time, the reason for this dynamics differs from that of the direct approach and relates to the changes in deforestation rates and proximate drivers over time. For Brazilian beef, where the share of clearing of natural vegetation for pasture expansion has been relatively stable historically (see Fig. 1), the estimated Eav is quite insensitive to the choice of amortization period, hovering around 70 tCO2/tCW; only if focusing on the more recent past (i.e. through a amortization period of 1– 7 years) does the striking reduction in deforestation rates cause the LUC-CFP to drop significantly from this number. Similarly, for palm oil a shorter amortization period puts more emphasis on the recent past where clearing rates for oil palm plantations have been higher (see Fig. 1), resulting in a higher Eav. For soybeans both short and a long amortization periods reduce the Eav as more emphasis is put on the times when there has been no direct rainforest clearing of forests for soybeans. Figure 3 (also showing results expressed in energy terms, per kcal) specifies the relative contribution of Amazon and Cerrado clearing to the LUC-CFP for beef and soybean, the relative contribution of living biomass

(c)

Fig. 2 Land-use change carbon footprints (LUC-CFPs) in the year 2010 as a function of the amortization period, for (a) Brazilian beef, (b) Brazilian soybean, and (c) Indonesian palm oil, for three different LUC-CFP formulations: E-LUC is the carbon footprint for production originating on cleared land (ELUC, differentiated between clearing of Amazon forest and Cerrado for beef and soy in Brazil), E-av is a nationally averaged carbon footprint (Eav) for all production, E-S (solid markers, no line) represent LUC-CFPs calculated using the simplified expression (ES). Note the logarithmic scale.


A LUC-CFP METHOD FOR AGRICULTURAL COMMODITIES 7 (a)

(b)

(c)

(d)

Fig. 3 Footprints for land-use change and associated carbon emissions for Brazilian beef and soy and Indonesian palm oil in 2010, estimated using a ten amortization period. Panel a display the results expressed in energy terms, comparing across all three products. Carbon footprints (LUC-CFP, left axes, b–d) are given in tons of CO2 emissions per ton of product and LUC footprints (LUC-FP, right axes, b–d) in hectares per thousand tons of product (note the differences in scales across the panels b–d). Results are divided between the contributions from LUC in the Amazon and Cerrado biomes for Brazil, and between emissions from living biomass (above and below ground) and peat oxidation for Indonesia.

and peatland emissions to the palm oil LUC-CFP, as well as the LUC footprint for all three products. As expected, emissions from Cerrado clearing contribute only little to the LUC-CFP of Brazilian beef, but account for roughly half of the soy LUC-CFP. For both products, however, Cerrado clearing constitutes a large share of the estimated LUC area footprints, illustrating the need to account for LUC outside of the Amazon if one is concerned with LUC impacts other than carbon emissions, such as biodiversity. In Fig. 4 we vary the factor attributing LUC to different drivers between a purely direct measure (rdirect n;i;j;s Fig. 1 and Table S1) and a purely indirect measure (rindirect see Table S1). The fact that 80% of forests n;i;j;s cleared in the Amazon is replaced by pastures, but that total pasture area in Brazil has been stable over the last decade, implies that shifting from a direct to an indirect approach drastically reduces the estimated LUC-CFP for Brazilian beef, with a zero LUC-CFP in the purely indirect case. Conversely, for soy the indirect approach increases the LUC-CFP by more than four times compared to the direct approach, since soy has been the most rapidly expanding crop in Brazil but there has been relatively little direct conversion of forests for soy cropping. Table 2 summarizes the difference in LUC-CFPs for the three products using the full specification (Eav) and the simplified expression (ES) that is less demanding in terms of data inputs (results for both specifications are also displayed in Fig. 2). As can be seen, differences are in most cases small (1.4–3%) to moderate (10–15%). The exception is the LUC-CFP for soybeans in year 2000, since in the full specification there is little

Fig. 4 The estimated land-use change carbon footprints (LUCCFPs) for Brazilian beef (left axis) and Brazilian soy and Indonesian palm oil (right axis), as a function of the relative (linear) weight given to direct vs. indirect estimates of the extent to which these commodities are driving land-use change in the respective country.

production of soy on cleared forest land as Amazon deforestation for soy has just picked up and according to the assumed land use dynamics (ai,j,t) soy is not planted the first year after clearing due to remaining debris. This dynamics is not accounted for in the simplified LUC-CFP specification. The deviance is smallest for Brazilian beef, averaging just over one per cent in relative terms. The differences in soy are also small in 2010, whereas for palm oil the deviation is moderate. This is expected because, as


8 U . M A R T I N P E R S S O N et al. Table 2 Average differences between the full land-use change carbon footprint (LUC-CFP) formulation (Eav) and the simplified formulation (ES) that require less data inputs, evaluated over a 1–20 years amortization period for the years 2000 and 2010. Standard deviations in parenthesis 2000

Beef, Brazil Soybeans, Brazil Palm oil, Indonesia

2010

Absolute diff. (tCO2/t product)

Relative diff.

Absolute diff. (tCO2/t product)

Relative diff.

1.53 (1.62) 0.15 (0.07) 0.77 (0.71)

1.5% (1.6%) 43% (11%) 9.8% (5.5%)

0.96 (1.05) 0.02 (0.01) 1.2 (0.38)

1.4% (1.4%) 3.0% (2.4%) 15% (7.2%)

noted above, the simplified expression will yield more accurate results in cases where either clearing rates or yield and land-use dynamics shows small variations over time. Since soybeans and palm oil both have more pronounced yield dynamics (with zero production in the first and first three years, respectively) and show larger relative variability in clearing rates overtime than beef, the deviations from the full specification are larger.

Discussion We have proposed a footprinting methodology that allows for linking land-use change processes to final consumers over large distances. Comparing the LUCCFPs estimated here with previous estimates (Table 1) illustrates how diverging results from different studies stem both from difference in method and data input. For example, the LUC-CFP for soybeans as calculated here is lower than studies that assume that all soy expansion occurs of forest land (e.g. Opio et al., 2013) and more similar to studies positing – consistent with empirical data – a smaller share of soy expansion occurring on forest and shrubland (e.g. Meul et al., 2012; Middelaar et al., 2012). The soy LUC-CFP is also lower than studies that take an indirect approach to allocating emissions from LUC (e.g. Ponsioen & Blonk, 2012), as that implies that more of the burden falls on soy (illustrated in Fig. 4). An advantage of the methodology proposed here, over previous direct approaches, is the difference in the sensitivity to and, more importantly, representation of the amortization period (T). Here, changing T affects how quickly producers are rewarded for reductions (or punished for increases) in clearing rates for a given commodity. This implies that the choice of T, while still a political decision, has different implications than if a higher/lower T simply leads to a lower/higher footprint, and it might therefore be easier to agree on its value in a political setting. We also note that the simplified expression proposed here bears similarities to the indirect LUC-CFPs

proposed by Ponsioen & Blonk (2012) and Weiss & Leip (2012). The main advantages of the expression in Eqn (4) over the previous specifications is that it utilizes empirical data on deforestation rates and drivers and makes explicit how the choice of amortization period affects the estimated LUC-CFP. We argue that the main advantage of the proposed methodology is its flexibility. First, it can be adjusted to the level of information available, in a tiered approach. At the lowest tier, one can use the simplified LUC-CFP specification [Eqn (4)] and an indirect measure of LUC drivers (rindirect n;i;j;s ) to calculate LUC-CFPs for a wide set of commodities and countries, using country level data that is readily available today: national deforestation rates can be sourced either from global assessments such as the FAO Forest Resources Assessment (FAO, 2010) or remote sensing analyses (Hansen et al., 2013), or regional or national assessments (e.g. Hansen et al., 2009; Miettinen et al., 2011; Inpe, 2013); default forest carbon stocks (DCn,i,j) data for different biomes are available (e.g. in the IPCC Guidelines); country level production data for crops needed to calculate rindirect n;i;j;s can be taken from the FAOSTAT database. At a second tier, the full LUC-CFP specifications [Eqns (1) and (2)] will be used together with data on the proximate drivers of LUC (rdirect n;i;j;s ), but still with regionally averaged (e.g. national) data on carbon stocks (ΛCn,i,j), yields (yi,j,t) and yield dynamics. At the highest tier, spatially explicit data on carbon stocks (Aguiar et al., 2012; Baccini et al., 2012; Harris et al., 2012b) and yields on cleared land would be used, as well as detailed data on land use and yield dynamics following LUC (ai,j,s). The main limiting factor in calculating detailed LUCCFP for a broad set of agricultural commodities is the lack of quantitative data on the proximate drivers of LUC. A recent attempt to compile quantified information on deforestation drivers (Hosonuma et al., 2012) illustrates this; only for 11 of 100 tropical countries were quantitative estimates of the direct drivers of deforestation available, although the study only attributed clearing to broad land-use classes (e.g. subsistence


A LUC-CFP METHOD FOR AGRICULTURAL COMMODITIES 9 vs. commercial agriculture), rather than to individual agricultural commodities. Analyses of the soy expansion in Brazil (Rudorff et al., 2011; Arvor et al., 2012; Macedo et al., 2012; Brown et al., 2013) have shown how recent advances in remote sensing enables identification of land uses replacing natural vegetation over large scales. The lack of reliable data quantifying the amount of LUC for different commodities, together with uncertainties in biomass carbon stock estimates, imply that there are still considerable uncertainty regarding exact LUC-CFP values. We perform a simple Monte Carlo analysis, randomly varying the assumed values for LUC, drivers and carbon stocks (see Supporting information for details), estimating that the uncertainty in the average LUC-CFPs for Brazilian beef and Indonesian palm oil are 27–28%, while for Brazilian soy it is 16% (95% confidence intervals). Second, while the LUC-CFPs presented here for beef, soybeans, and palm oil have been evaluated at the national level there is no scale inherit in the proposed methodology. Given the availability of data on deforestation rates and proximate drivers, production, etc. at a subnational scale, one could evaluate regional or local LUC-CFP. Although initiatives such as the Brazilian Soy Moratorium have shown that consumer power to affect land-use practices can be leveraged at a national scale, estimating LUC-CFPs at a lower scale would provide individual producers with larger incentives to alter land-use change patterns. However, this also requires the ability to track products from the producer to consumer. Third, the methodology easily allows for shifting between the direct and indirect approach to allocating LUC and associated carbon emissions to different agricultural commodities. As clearly illustrated by the Brazilian beef and soy cases, this choice can have large implications for the estimated LUC-CFPs, and consequently there is a need for models and empirical analyses that can help guide the extent to which emissions should be allocated between agricultural commodities that directly and indirectly drive LUC. For example, Barona et al. (2010) and Arima et al. (2011), running spatial regressions models to assess the impact of soy expansion on deforestation through displacement of pastures to deforestation frontiers, provide empirical evidence on the relative importance of soy and cattle in driving deforestation in the Brazilian Amazon. However, different uses of LUC-CFPs may also call for different emphasis of direct vs. indirect drivers. For instance, if used for consumer labelling, a zero LUCCFP for Brazilian beef may be seen as misleading, given that 80 per cent of the deforestation in the Brazilian

Amazon is due to expansion of pastures. Also, under an indirect approach producers (e.g. Brazilian soy farmers) cannot affect the LUC-CFP in other ways than stopping the total expansion of land for a given crop (regardless of which type of land they are expanding on to). However, in an analytical framework where one aims to link changes in overall demand of agricultural products to LUC and associated emissions, an indirect approach may be more reasonable. Apart from informing consumers and incentivizing producers, the proposed LUC-CFPs could also be used in analyses tracking the environmental impacts of LUC from producers to final consumers through international trade. So far these analyses have considered land use impacts, but few studies have tried to link consumption of agricultural commodities to land-use changes and associated carbon emissions in producing countries (Meyfroidt et al., 2013). Quantifying LUC and carbon emissions embodied in trade can help illuminate global ‘teleconnections’, linking land-use change processes to final consumers over large distances, and harness the opportunity presented by the increasing globalization of deforestation drivers.

Acknowledgements The preparation of this article has been supported by the Swedish Energy Agency (STEM) and by the Norden Top-level Research Initiative subprogramme ‘Effect Studies and Adaptation to Climate Change’ through the Nordic Centre of Excellence for Strategic Adaptation Research (NORD-STAR). We are thankful for valuable comments from three anonymous reviewers, as well as from Patrick Meyfroidt and colleagues at Environmental Systems Analysis at Chalmers.

References Aguiar APD, Ometto JP, Nobre C et al. (2012) Modeling the spatial and temporal heterogeneity of deforestation-driven carbon emissions: the INPE-EM framework applied to the Brazilian Amazon. Global Change Biology, 18, 3346–3366. Arima EY, Richards P, Walker R, Caldas MM (2011) Statistical confirmation of indirect land use change in the Brazilian Amazon. Environmental Research Letters, 6, 024010. Arvor D, Meirelles M, Dubreuil V, Begue A, Shimabukuro YE (2012) Analyzing the agricultural transition in Mato Grosso, Brazil, using satellite-derived indices. Applied Geography, 32, 702–713. Baccini A, Goetz SJ, Walker WS et al. (2012) Estimated carbon dioxide emissions from tropical deforestation improved by carbon-density maps. Nature Climate Change, 2, 182–185. Barona E, Ramankutty N, Hyman G, Coomes OT (2010) The role of pasture and soybean in deforestation of the Brazilian Amazon. Environmental Research Letters, 5, 024002. Batlle-Bayer L, Batjes NH, Bindraban PS (2010) Changes in organic carbon stocks upon land use conversion in the Brazilian Cerrado: a review. Agriculture, Ecosystems & Environment, 137, 47–58. Brown JC, Koeppe M, Coles B, Price KP (2005) Soybean production and conversion of tropical forest in the Brazilian Amazon: the case of Vilhena, Rond^ onia. AMBIO: A Journal of the Human Environment, 34, 462–469. Brown JC, Kastens JH, Coutinho AC, Victoria DDC, Bishop CR (2013) Classifying multiyear agricultural land use data from Mato Grosso using time-series MODIS vegetation index data. Remote Sensing of Environment, 130, 39–50.


10 U . M A R T I N P E R S S O N et al. Bustamante MC, Nobre C, Smeraldi R et al. (2012) Estimating greenhouse gas emissions from cattle raising in Brazil. Climatic Change, 115, 559–577.

Meul M, Ginneberge C, van Middelaar CE, de Boer IJM, Fremaut D, Haesaert G (2012) Carbon footprint of five pig diets using three land use change accounting

Carlson KM, Curran LM, Asner GP, Pittman AM, Trigg SN, Marion AJ (2012) Carbon emissions from forest conversion by Kalimantan oil palm plantations. Nature Climate Change, 3, 283–287. Cederberg C, Persson UM, Neovius K, Molander S, Clift R (2011) Including carbon emissions from deforestation in the carbon footprint of Brazilian Beef. Environmental Science and Technology, 45, 1773–1779. Dalgaard R, Schmidt J, Halberg N, Christensen P, Thrane M, Pengue WA (2008) LCA

methods. Livestock Science, 149, 215–223. Meyfroidt P, Lambin EF, Erb K-H, Hertel TW (2013) Globalization of land use: distant drivers of land change and geographic displacement of land use. Current Opinion in Environmental Sustainability, 5, 438–444. Middelaar C, Cederberg C, Vellinga T, Werf HG, Boer IM (2012) Exploring variability in methods and data sensitivity in carbon footprints of feed ingredients. The International Journal of Life Cycle Assessment, 18, 768–782.

of soybean meal. The International Journal of Life Cycle Assessment, 13, 240–254. FAO (2010) Global Forest Resources Assessment 2010. Food and Agriculture Organization of the United Nations (FAO), Rome, Italy. Fearnside P (1997) Greenhouse gases from deforestation in Brazilian Amazonia: net committed emissions. Climatic Change, 35, 321–360. Flynn HC, Canals LMI, Keller E et al. (2012) Quantifying global greenhouse gas

Miettinen J, Shi C, Liew SC (2011) Deforestation rates in insular Southeast Asia between 2000 and 2010. Global Change Biology, 17, 2261–2270. Millenium Ecosystem Assessment (2005) Ecosystems and Human Well-Being: Biodiversity Synthesis. World Resources Institute (WRI), Washington, DC, USA. Morton DC, Defries RS, Shimabukuro YE et al. (2006) Cropland expansion changes deforestation dynamics in the southern Brazilian Amazon. Proceedings of the

emissions from land-use change for crop production. Global Change Biology, 18, 1622–1635. Foley JA, Ramankutty N, Brauman KA et al. (2011) Solutions for a cultivated planet. Nature, 478, 337–342. Galford GL, Melillo J, Mustard JF, Cerri CEP, Cerri CC (2010) The amazon frontier of land-use change: croplands and consequences for greenhouse gas emissions. Earth Interactions, 14, 1–24.

National Academy of Sciences of the United States of America, 103, 14637–14641. Opio C, Gerber P, Mottet A et al. (2013) Greenhouse gas Emissions From Ruminant Supply Chains – A Global Life Cycle Assessment. (ed. Fao, FaaOOTUN). Food and Agriculture Organization of the United Nations (FAO), Rome. Persson UM, Azar C (2010) Preserving the world’s tropical forests–a price on carbon may not do. Environmental Science & Technology, 44, 210–215. Ponsioen TC, Blonk TJ (2012) Calculating land use change in carbon footprints of

Hansen MC, Stehman SV, Potapov PV, Arunarwati B, Stolle F, Pittman K (2009) Quantifying changes in the rates of forest clearing in Indonesia from 1990 to 2005 using remotely sensed data sets. Environmental Research Letters, 4, 034001. Hansen MC, Potapov PV, Moore R et al. (2013) High-resolution global maps of 21stcentury forest cover change. Science, 342, 850–853. Harris NL, Brown S, Hagen SC, Baccini A, Houghton RA. (2012a) Progress Towards a

agricultural products as an impact of current land use. Journal of Cleaner Production, 28, 120–126. Prudêncio Da Silva V, van der Werf HM, Spies A, Soares SR (2010) Variability in environmental impacts of Brazilian soybean according to crop production and transport scenarios. Journal of Environmental Management, 91, 1831–1839. Reijnders L, Huijbregts MAJ (2008a) Biogenic greenhouse gas emissions linked to the

Consencus on Carbon Emissions From Tropical Deforestation. Winrock International and Woods Hole Research Center, Falmouth, MA. Harris NL, Brown S, Hagen SC et al. (2012b) Baseline map of carbon emissions from deforestation in tropical regions. Science, 336, 1573–1576. Hosonuma N, Herold M, Sy VD et al. (2012) An assessment of deforestation and forest degradation drivers in developing countries. Environmental Research Letters, 7, 044009.

life cycles of biodiesel derived from European rapeseed and Brazilian soybeans. Journal of Cleaner Production, 16, 1943–1948. Reijnders L, Huijbregts MAJ (2008b) Palm oil and the emission of carbon-based greenhouse gases. Journal of Cleaner Production, 16, 477–482. Rudel TK, Defries R, Asner GP, Laurance WF (2009) Changing drivers of deforestation and new opportunities for conservation. Conservation Biology, 23, 1396– 1405.

Houghton RA (2012) Carbon emissions and the drivers of deforestation and forest degradation in the tropics. Current Opinion in Environmental Sustainability, 4, 597–603. Inpe (2013) Projeto PRODES: Monitoramento da Floresta Amaônica Brasileira por Satelite. (ed. Inpe, INDPE), S~ao Jose dos Campos. Available at: http://www.obt.inpe.br/ prodes/index.php (accessed 7 March 2014)

Rudorff BFT, Adami M, Aguiar DA et al. (2011) The soy moratorium in the amazon biome monitored by remote sensing images. Remote Sensing, 3, 185–202. Rueda X, Lambin EF (2013) Linking globalization to local land uses: how eco-consumers and gourmands are changing the colombian coffee landscapes. World Development, 41, 286–301. Schmidt JH (2010) Comparative life cycle assessment of rapeseed oil and palm oil.

Klink CA, Moreira AG (2002) Past and current human occupation, and land use. In: The Cerrados of Brazil: Ecology and Natural History of a Neotropical Savanna. (eds Oliveira PS, Marquis RJ) pp. 69–88. Columbia University Press, New York, NY, USA. Kløverpris JH, Baltzer K, Nielsen P (2010) Life cycle inventory modelling of land use induced by crop consumption: Part 2: Example of wheat consumption in Brazil, China, Denmark and the USA. The International Journal of Life Cycle Assessment, 15,

The International Journal of Life Cycle Assessment, 15, 183–197. Weiss F, Leip A (2012) Greenhouse gas emissions from the EU livestock sector: A life cycle assessment carried out with the CAPRI model. Agriculture, Ecosystems & Environment, 149, 124–134. Wicke B, Sikkema R, Dornburg V, Faaij A (2011) Exploring land use changes and the role of palm oil production in Indonesia and Malaysia. Land Use Policy, 28, 193–206.

90–103. Lee JSH, Abood S, Ghazoul J, Barus B, Obidzinski K, Koh LP (2014) Environmental impacts of large-scale oil palm enterprises exceed that of smallholdings in Indonesia. Conservation Letters, 7, 25–33. Leip A, Weiss F, Wassenaar T. et al. (2010) Evaluation of the Livestock Sector’s Contribution to the EU Greenhouse gas Emissions (GGELS). Joint Research Centre (JRC), European Commission, Ispra. Lindquist EJ, Aannunzio RD, Gerrand A. et al. (2012) Global Forest Land-use Change 1990–2005. Food and Agriculture Organization of the United Nations (FAO) and European Commission Joint Research Centre (JRC), Rome, Italy. Macedo MN, Defries RS, Morton DC, Stickler CM, Galford GL, Shimabukuro YE (2012) Decoupling of deforestation and soy production in the southern Amazon during the late 2000s. Proceedings of the National Academy of Sciences, 109, 1341–1346. Machado RB, Neto MBR, Pereira PGP. et al. (2004) Estimativas de Perda da Area do Cerrado Brasileiro. Relatorio Tecnico n~ao Publicado. (ed. Internacional C), Conservacß~ao Internacional, Brasılia.

Supporting Information Additional Supporting Information may be found in the online version of this article: Data S1. A derivation of the simplified LUC-CFP expression, as well as a detailed account for the data (sources, justifications, and full data set) used to calculate the LUC-CFPs for Brazilian beef and soybeans and Indonesian palm oil. Data S2. The full calculations of the LUC-CFPs presented here.
