The (environmental) price of FAME

Fatty acid methyl ester (FAME), or biodiesel as it is more commonly known, is a diesel substitute produced by reacting methanol with vegetable oil, and is the second most widely used bio-additive in America fuels (behind ethanol added to gasoline). As shown in Figure 1, U.S. biodiesel consumption has increased more than a hundredfold since 2001[1]. The main driver of this rapid expansion of the biodiesel industry has been the federal Renewable Fuel Standard (RFS), under which biodiesel produced from soy oil canola oil, yellow grease and tallow qualifies as an advanced renewable fuel.

Figure 1 Rising biodiesel consumption in the U.S.

Source: EIA Monthly Energy Review, Table 10.4 https://www.eia.gov/totalenergy/data/monthly/index.php#renewable

For many people, the word biodiesel conjures up an image of restaurant grease being collected and recycled as fuel. In fact, about two thirds of U.S. biodiesel is produced from food-grade unused vegetable oils, and only an eighth is actually produced from recycled restaurant oil. Most of the food-grade vegetable oil turned into biodiesel in the U.S. is soy oil (about 55% of total biodiesel), and the rest is largely canola oil (about 10% of total biodiesel production). According to the EPA’s analysis for the RFS, soy biodiesel use has associated lifecycle greenhouse gas emissions of 40 grams of carbon dioxide equivalent released for every megajoule of energy in the biodiesel (gCO2e/MJ). More recent analysis[2], which provides the basis of values included in Argonne National Laboratory’s GREET 2017 model, estimated an even lower value of 29 gCO2e/MJ, even when including emissions from potential land use changes in the U.S. and internationally (indirect land use change, ILUC).


Figure 2. GHG intensity of soy biodiesel, as reported by Chen et al. (2018)

This combination of rapid increase in renewable fuel use and relatively low lifecycle greenhouse gas emissions appears to be a good news story for the climate. However, dig a little deeper into the analysis underneath the numbers and the picture gets much murkier. In fact, it turns out there’s a very real possibility that U.S. government agencies have systematically overstated the environmental benefit of using soy biodiesel, and that soy biodiesel should not qualify as an advanced fuel under the RFS.

Soy in the EPA’s analysis for RFS

This article will mostly focus on some issues in the soy biodiesel analysis from GREET, but before we get into that it’s worth taking a moment to revise how soy biodiesel qualified as an advanced biofuel in the first place. The magic number to be counted as an advanced biofuel in the RFS was 46 gCO2e/MJ. Provided soy biodiesel’s assessed greenhouse gas intensity was below that number it would qualify for support as an advanced fuel alongside biodiesel from waste and residual oils and imported sugarcane ethanol.

While we are now used to the idea that soy biodiesel counts as an advanced biodiesel, this result was anything but inevitable ten years ago. Indeed, the preliminary lifecycle analysis undertaken by the EPA for the RFA concluded that when land use change emissions were included soy biodiesel had a larger climate impact than fossil fuels, assessed at 95 gCO2e/MJ. In the final analysis, the EPA presented results for three years: 2012; 2017 and 2022. In general, we would expect that over time processes would become more efficient and yield will increase, and indeed we see that the lifecycle emissions results for soy biodiesel in 2012 and 2017 were both above the threshold value – only by 2022 was soy biodiesel expected to achieve the requisite climate benefit. This is shown in Figure 2. The reduction in expected emission intensity from 2012 to 2022 is rather significant – a 17 gCO2e/MJ reduction.

Figure 3 Evolution of soy biodiesel GHG intensity results in RFS analysis

So, how did the negative preliminary result turn into the positive (for soy biodiesel, at least) Unpacking the results reveals a number of interesting features. Figure 3 shows the results from the final RFS analysis decomposed into constituent emissions. There are three features of the analysis that are particularly interesting. Firstly, there are credits of 8 gCO2e/MJ given in 2022 for domestic land use change and for domestic rice methane, and a credit of 6 gCO2e/MJ from international livestock emissions. Without any one of these credits (assuming the rest of the analysis remained the same), soy biodiesel would not have qualified as an advanced biofuel under the RFS. Secondly, the international land use change emissions fall by 30% from 2017 to 2022, avoiding 19 gCO2e/MJ. Without this reduction, soy biodiesel would also have failed to qualify as an advanced biofuel.

Figure 4 Decomposition of RFS lifecycle results by type of emissions
Source: https://www.regulations.gov/document?D=EPA-HQ-OAR-2005-0161-3173

 

The three large emissions credits is each somewhat surprising. The rice methane credit is based on a prediction from the FASOM model that 8% of increased soy acreage in 2022 would replace U.S. rice paddies.[3] This is about the only place in the biofuel discussion that you’ll see anyone claim that biofuels strongly impact rice production. Indeed, the modeling with FAPRI undertaken for the RFS final analysis shows no significant reduction in paddy area, and during discussions of the 2008/09 food price crisis it was repeatedly stated that rice markets could not be strongly impacted by biofuel demand.

The domestic land use change emissions credit is arguably even more surprising. In general in the land use change discussion it is accepted that increased agricultural area will generally result in losses of carbon from biomass and soils in the converted landscapes. It is, to say the least, counter-intuitive to find a result in which a large expansion of soybean area apparently increases total carbon storage in U.S. landscapes rather than reducing them.

The third credit, for a reduction in international livestock methane emissions, is also somewhat surprising. A significant reduction in beef production is predicted in Latin America by the model, with soybean area in Brazil expanding at the expense of pastureland. While this seems a plausible prediction, the co-product of soy oil production, soy meal, is widely used as a livestock feed. It is therefore not clear that increased area devoted to soybeans, even at the expense of pastureland, should result in reduced livestock raising. Indeed, some biofuel producers have sought to characterize the biofuel industry as producing ‘fuel and feed’ because increased oilseed crushing can support increased availability of high protein animal feed. One way in which increased vegetable oil production could be delivered without producing large quantities of animal feed would be to use palm oil instead of soy oil to meet increased demand – but the EPA modeling does not predict a large palm oil response.

Finally, why are the international land use change emissions calculated for 2022 so much lower than for the other years given? It turns out that the reduction is primarily due to a large assumed reabsorption of carbon in the Amazon in the modeling, in the period 2023 to 2051, which is absent in the modeling of 2012 and 2017[4]. It’s unclear why the carbon change dynamics are so different in the 2022 results than the earlier results, but it suggests that the model may be predicting an unrealistic carbon sequestration dynamic in the 2022 results.

Of course, the points we’ve identified here are only in the areas in which it seems that the EPA modeling may have underestimated emissions, and it would be possible to identify places in the model where one assumption or another may have tended to overestimate soy biodiesel emissions. Still, there are enough questionable outcomes documented in the modeling that it’s worth at least asking whether the results may have been unduly favorable to soy biodiesel, and whether soy biodiesel was rather lucky to make it into the advanced biofuel category.

Soy in GREET

While the oddities in the EPA lifecycle analysis are interesting, the EPA numbers have not been updated since the final RFS rule was adopted, whereas the lifecycle values bundled in the GREET lifecycle analysis tool from Argonne National Laboratory are regularly updated. As we noted above, these numbers are now rather more favorable to soy biodiesel even than the EPA numbers were, having been reduced to 29 gCO2e/MJ in the analysis by Chen et al. (2018)  – but it turns out that these values may also be giving a misleadingly positive impression.

Allocations

The first issue in the GREET analysis relates to the way that emissions are divided between the oil and the meal that are produced when soybeans are crushed. In the analysis by the EPA, this step is not necessary – the availability of soy meal is fed back into the models used in the emissions assessment, and allows for reductions in the total amount of additional crop cultivation required in the assessment. The results in GREET deals the co-products in the same way for the land use change analysis through the GTAP model, but for agricultural emissions GREET uses a system of allocation. 

The purpose of allocation is to fairly divide the emissions associated with a process between different outputs of that process. The three most common allocation approaches in lifecycle analysis are:

  1. Market value: emissions are allocated between co-products in proportion to the amount of each co-product produced multiplied by a typical price for that material. The advantage of market value is that it allocates more emissions to higher value co-products, and thus best corresponds to the economic decision making of producers. The main disadvantage is that prices are variable, and so there is no single correct answer as to which set of prices should be used, meaning that results could change over time.
  2. Energy content: emissions are allocated between co-products based on chemical energy content. This approach is used by the European Union for its lifecycle analysis. It has the advantage that energy content is often a reasonable proxy for the value of a co-product (especially where materials are used for food or feed or directly for energy), but the disadvantage that it is not appropriate for co-products whose use does not require any chemical energy content.
  3. Mass: emissions are allocated between co-products based on mass. This is the approach taken in GREET. It has the advantage that it can be applied very widely to any material co-products (it is not appropriate for power or heat as co-products) but the disadvantage that the mass of a material is often not strongly correlated to its value. For instance, it would not seem appropriate to allocate emissions to residues with very little value based on their mass.   

GREET uses mass allocation for oilseed crushing, giving the justification in Chen et al. (2018) that, “meals are not an energy product and mass is not subject to market value vacillation”. This rationale is fine insofar as it goes, but it ignores the fundamental weakness of using a mass-based allocation, which is that it results in a disproportionate allocation of emissions to heavy, low value products. While soy meal is clearly not normally burned for energy, the energy in the meal is metabolized by animals when it is used in feed rations, and so the argument against using energy allocation is disingenuous at best. Indeed, in the European Union it is standard to use energy allocation in these calculations.

As can be seen in Figure 4, using the mass allocation for soybean crushing results in allocating only half as many emissions to the oil as would be allocated on a value basis. Given that soy meal is sold for animal feed, and the animal feed sector does not generally use lifecycle accounting as part of its decision making, the result of this is that a large chunk of the emissions associated with soybean production simply disappears from the GREET lifecycle analysis. Using mass allocation gives a misleading impression of the balance of the meal and oil in the value of the soybean, and reduces the stated GHG intensity of soy biodiesel by about 10 gCO2e/MJ. This issue is particularly important in the case of soy biodiesel because the soybean has a much larger fraction of its mass go to the co-product than any other major biofuel crop. The use of mass allocation therefore also skews any comparison of the environmental credentials of biofuels in favor of soy biodiesel.

Figure 5 Comparing the results of different allocations between soy oil and meal

Source: GREET 2017

 

Indirect land use change

The use of mass allocation has been a feature of the GREET lifecycle analysis for many years. In contrast, the land use change emissions estimates included in GREET regularly change, and the model used to estimate ILUC emissions for GREET (called GTAP) is regularly updated.

In 2009, the GTAP framework was used to assess the ILUC emissions from soy biodiesel for the regulatory analysis under the California Low Carbon Fuel Standard (LCFS). At that time, the estimate for ILUC emissions for soy biodiesel was 62 gCO2e/MJ, higher than those estimated by EPA. By 2018, however, an ILUC value of 6.3 gCO2e/MJ from Chen et al.[5] was integrated into GREET 2017. The downward trajectory of soy biodiesel ILUC estimates from GREET over this period is shown in Figure 6. What accounts for this 90% drop in estimated emissions, and are the new values more credible than the older California estimates?

Figure 6 Point ILUC factor estimates for US soy biodiesel obtained with variants of the GTAP-BIO model

Part of the answer is that a series of amendments to the GTAP modeling framework has been made since 2009 in a way that in several cases was effectively designed to reduce the output ILUC values. A forthcoming paper[6] shows in detail that a number of innovations have been introduced to the GTAP framework in the interim in ways that inevitably resulted in lower estimated ILUC emissions, but that were not well supported by available evidence.

As an example, a 2017 paper introduced a regionally differentiated approach to set the response of crop yields to increased demand, replacing a uniform global parameter that had been used previously. This paper provided no evidence that the previous response had been too weak on average, and provided no new econometric evidence for the regional values adopted. This adjustment could have been implemented in a way that added sophistication to the model but avoided changing the average global yield response. Instead, the new parameters were raised for more than half the regions considered, including the regions most important to the analysis. This model change therefore resulted in lower ILUC estimates, despite presenting no evidence that the ILUC estimates actually ought to be lower.

Similarly, in 2017 a new model feature was introduced allowing cropping intensity to increase (i.e. allowing the model to assumed that one piece of land may be harvested two or more times in a year). Making this change would obviously result in reduced ILUC estimates, but the evidentiary basis for setting the new parameters was exceedingly weak. Previously, some commentators had argued that this cropping intensity response was already accounted for implicitly in the model in the yield response, but no consideration was given to whether the yield parameters should therefore be adjusted when the new feature was added. The picture that emerges is of a willingness to edit the model in ways that reduced the output values, and a corresponding resistance to making any amendments that would increase the ILUC estimates. It is therefore very hard to have confidence that the reductions in expected land use change between the older modeling and the latest modeling represent an improved understanding, rather than simply reflecting the underlying opinion of the modeling teams doing the work.

Alongside changes in the GTAP economic model itself, the Chen et al. (2018) ILUC assessment reflects the use of a new set of emissions assumptions for the amount of carbon released following any given change in land use. The result in Chen et al. (2018) that has been made the default ILUC value for soy biodiesel in GREET (6.3 gCO2e/MJ) is based on emission factors from a model called ‘CCLUB’. If instead the emission factors developed by the California Air Resources Board are used, the results is quite different, 18.3 gCO2e/MJ. It turns out that the CCLUB model uses some very questionable assumptions about soil carbon in particular that explain much of this difference. The use of an incorrect land use history in soil carbon modelling for conversion of ‘cropland pasture’ to permanent cropping results in large carbon credits that are almost certainly not justified. The model also systematically understates the importance of peat drainage emissions when palm oil expands in Indonesia and Malaysia, justified by an inappropriate use of analysis from the EPA’s final RFS rule.

The choice of emission factors alone cuts 12 gCO2e/MJ from the lifecycle estimate in GREET. It’s hard to make a numerical assessment of the overall impact of poorly justified model changes in GTAP, but these likely shave further tens of grams of emissions from the reported numbers.

The palm oil connection

Above, we noted that the CCLUB emissions model has resulted in a systematic underestimation of peat drainage emissions associated with palm oil expansion. In the current GREET modeling, the expansion of oil palm in Southeast Asia accounts for only a small fraction of the global increase in vegetable oil production to meet increasing U.S. demand for soy oil for biodiesel – the model assumes that the principle response is for soybean area and hence soybean oil production to increase. Even so, the peat emissions are an important term, and one of the major differences between results using different sets of emission factors.

There is evidence, however, that GTAP may significantly underestimate the link between U.S. soy oil demand and palm oil production elsewhere in the world. The GTAP model is calibrated based on known trade patterns. There has been relatively little import of palm oil to the United States in the past, and therefore the model assumes that there will be relatively little in future. However, there is evidence that the link between the soy and palm oil markets is stronger than this, and that palm oil exports are actually quite responsive to the use of soy oil in the U.S. The basic idea is that because soybeans are grown in large part in order to produce soy meal, soybean area may be unresponsive to increases in demand for vegetable oils alone. In contrast, the oil palm industry produces very little meal, and therefore oil palm area responds only to vegetable oil demand. The hypothesis is therefore that the palm oil supply will increase even if other vegetable oils are used for biodiesel. In this case, as a limited supply of soy oil is increasingly diverted to biodiesel production, palm oil will be imported to the U.S. to meet supply shortfalls in other markets).

A recent paper presenting an econometric assessment of the linkage between soy oil demand and palm oil imports[7] provided support for this hypothesis, suggesting that the connection between soy oil consumption and palm oil production may be much stronger than GTAP has previously been calibrated to assume. If so, this would likely imply that ILUC emissions from soy biodiesel are larger than previously believed, because the oil palm industry is very strongly associated with tropical deforestation and peat loss[8].

Conclusion

Soy biodiesel is currently treated as an advanced biofuel under the RFS, and recent lifecycle analysis of soy oil by Chen et al. (2018) (corresponding to the default soy biodiesel emissions given in GREET 2017) concluded that, “soy biodiesel could achieve … 66–72% reduction in overall GHG emissions, relative to its petroleum counterpart.” This article explains that the picture for soy biodiesel is in fact much murkier than these results suggest. Both the EPA modeling and the modeling for GREET include a number of modeling assumptions that seem very likely to be unduly favorable, and result in published lifecycle GHG intensity values that are misleadingly low.

In the case of GREET 2017, we have shown that the use of unduly favorable mass based allocation saves about 10 gCO2e/MJ, the use of a flawed emission factor model saves another 10 gCO2e/MJ, and a series of optimistic modeling assumptions in GTAP likely account for published ILUC values being at least a further 10-20 gCO2e/MJ below a reasonable estimate. Recalibrating the modeling to more accurately reflect the link between soy and palm oil markets would further increase the estimated GHG intensity. Taken together, we can conclude that the value of 29 gCO2e/MJ for soy biodiesel given by Chen et al. (2018) could easily be 40 gCO2e/MJ too low, and therefore that soy biodiesel may well deliver much less than a 50% GHG emissions reduction compared to fossil diesel.

Acknowledgement

This article was supported by the National Wildlife Federation.


[1] There is a regular annual drop off in biodiesel blending during the winter months as biodiesel solidifies at higher temperatures than fossil diesel, and thus biodiesel blending must be reduced in cold weather.

[2] Henceforth referred to as “Chen et al. (2018)”. R. Chen, Z. Qin, J. Han, M. Q. Wang, F. Taheripour, W. E. Tyner, D. O’Connor, and J. Duffield, “Life cycle energy and greenhouse gas emission effects of biodiesel in the United States with induced land use change impacts,” Bioresour. Technol., vol. 251, pp. 249–258, 2018.

[3] https://www.regulations.gov/document?D=EPA-HQ-OAR-2005-0161-3179

[4] https://www.regulations.gov/document?D=EPA-HQ-OAR-2005-0161-3153

[5] From the “GTAP 2011” ILUC scenario.

[6] Malins, C., Plevin, R., Edwards, E. (2019). Accentuating the positive – how robust are reductions in modeled estimates of the indirect land use change from conventional biofuels?

[7] https://www.sciencedirect.com/science/article/pii/S0301421518307924

[8] Cf. http://www.cerulogy.com/palm-oil/driving-deforestation/