Comparing GTAP ILUC results to observations of ethanol related land use change

For over ten years, indirect land use change modeling has been an important part of assessing the environmental impact of U.S. biofuel policy. While several models have been developed to undertake these assessments, notably the FAPRI-FASOM model used by the Environmental Protection Agency (EPA) in determining the lifecycle emissions of fuels supported by the Renewable Fuel Standard (RFS) (U.S. Environmental Protection Agency, 2010), the most prolific use of a single model has been the GTAP computational general equilibrium model, used for regulatory analysis by the California Air Resources Board (California Air Resources Board, 2014), in the Argonne National Laboratory’s GREET model (Argonne National Laboratory, 2017) and in a sequence of independent analyses by researchers. The GTAP model itself does not include land use change emission factors, and thus areal land use change results output from GREET must be coupled to emission factor models to produce ILUC results in terms of carbon dioxide emissions per unit of energy produced.

One notable feature of the ILUC factor results reported using the GTAP model over the past decade is that they have shown a tendency to reduce over time, as shown for in the figure below for corn ethanol.

Point ILUC factor estimates for US corn ethanol obtained with variants of the GTAP-BIO model

A forthcoming academic paper[1] will discuss in detail the underlying analytical reasons for these reductions, and queries whether the changes to the model and emissions factors that have driven these changes have been adequately justified. While assessing modeling choices is an interesting exercise, it is also enlightening to compare reported model results to observed changes in U.S. agriculture. One source for observed data on U.S. land use changes that may be associated with the growth of the corn ethanol industry is a series of papers by researchers at the University of Wisconsin (including Lark, Salmon, & Gibbs, 2015; Wright et al., 2017), which use data from the USDA cropland data layer (CDL). Here, we compare ILUC results for corn that are documented by Qin, Dunn, Kwon, Mueller, & Wander (2016) against the observed land use changes in the U.S. in the period 2008-2012 documented in Wright et al. (2017).

Qin et al. (2016) give one of the lowest corn ILUC numbers available in the literature, as low as 2.1 gCO2e/MJ at the bottom end of the interval given (the range reflects different yield and tillage assumptions), but are these results consistent with observed land use changes? Below, the reported model outputs from the ‘Corn2’ scenario in  Qin et al. (2016)[2] are considered, and compared to agricultural statistics and the results of the analysis presented in (Wright et al., 2017) for the years 2008-2012, during which ethanol productions increased by 4.2 billion gallons. It is important to note up front that it is always difficult to compare observed land use changes to the results of indirect land use change models. It is generally not possible to definitively identify whether a given land use change would have occurred in the absence of biofuel demand. In contrast, in ILUC modeling we know by hypothesis that all land use changes reported are due to biofuel demand. Apparent inconsistencies between observed land use data and ILUC models may have several reasonable explanations. The results noted below cannot refute the results of the modeling, but do perhaps suggest that some model assumptions ought to be reexamined. 

Qin et al. (2016) consider an increase of biofuel demand by 11.59 billion gallons (the ‘shock’), which is about equivalent to the increase in ethanol production in the U.S. between 2004 and 2015 (based on data from the EIA[3]), and assess how much land use change was likely to have been caused by that increase. Given average corn production[4] and ethanol refinery[5] yields for the period 2011-2015, producing a billion gallons of ethanol requires about a million hectares of corn. The net cropland expansion in the U.S. modeled by Qin et al. (2016) is much less than this however, only 160 thousand hectares per billion gallons of additional corn ethanol demand. Almost all of the U.S. land use change (154 thousand hectares per billion gallons) was modeled coming from a category of land identified as ‘cropland pasture’ – defined by the USDA as land that at some previous point was cropped, and is still suitable for cropping, but since then has been use as pasture. A further 112 thousand hectares of land use change per billion gallons are predicted outside the U.S. That the net cropland expansion modeled should be less than the absolute cropland requirement is not surprising to anyone familiar with ILUC modeling. There are several effects that reduce net requirements for additional land – notably these include the use of co-products from ethanol production (distillers’ grains) as animal feed, reducing demand for food[6], and the possibility that agricultural productivity increases (Malins, Searle, & Baral, 2014). While it is not surprising that the net land requirement is less than the gross requirement, it is clear that GTAP is modelling a strong role of responses other than land use change in meeting additional corn demand, given that the net land demand is only about a quarter of the gross. 

Wright et al. (2017), looking at satellite data, identify 1.1 million hectares of land within 50 miles of ethanol refineries being converted from non-crop status to cropping in the period 2008-2012, with most of this land (700 thousand hectares) converted to corn or soybean cropping. A further 600 thousand hectares of land from 50 to 100 miles from ethanol refineries was converted from non-crop statuses, though over half of this was planted with other crops. Further than 100 miles from corn ethanol refineries only 20% of newly converted land was planted with corn or soybeans[7]. Wright et al. (2017) argue that depending on the mix of rotations adopted (how much land remains in corn-soy rotation and how much is given over to continuous corn) then their analysis would be consistent with between 500,000 and 1 million hectares of previously uncropped land being converted for corn production within 100 miles of ethanol refineries over the period considered. If all of this additional corn area could indeed be attributed to the corn ethanol mandate, that represents somewhere between 130 thousand and 260 thousand hectares of land use change to corn production for every billion gallons of additional ethanol demand in that period, comparable to the modelled values. Data from the AFDC[8] shows that during this period use of corn for non-fuel applications reduced, so it is not unreasonable to assume that most conversion of uncropped land to corn production in this period was a response to the corn ethanol mandate. Preliminary analysis by the same group (Lark et al., 2019) estimates that 1.1 million hectares of land conversion since 2005 is associated with the ethanol mandate of the RFS2, about 200 thousand hectares per billion gallons.

Another comparison point comes from related work supported by the National Wildlife Federation (Hendricks, 2018), which aims to provide an economic assessment of the impact of the RFS2 on land use changes in the period 2008-2016  using data from the National Resources Inventory (NRI). This assessment attributes 1.2 million hectares of net new cropland in the period 2009-2016 to the RFS, predominantly coming from conversion of former CRP land. If all of this was associated with the corn mandate, it would be equivalent to about 190 thousand hectare per additional billion gallons of ethanol production – part of this expansion ought to be associated to the biodiesel mandate, and so the overall conclusion is quite comparable to the U.S. land use change modeled by (Qin et al., 2016).

While the realized rates of land use change for corn ethanol production identified in these studies are comparable to those modeled by Qin et al. (2016), there is evidence that the role of the cropland pasture land category as a source of cropland may be dramatically overstated by the GTAP modeling. For example, using the same dataset as considered by Wright et al. (2017), this study finds that 22% of newly cropped land in the period 2008-2012 was converted from “long-term (20 + year) unimproved grasslands”. This compares to the (Qin et al., 2016) result in which at most 5% of converted grasslands could have possibly have fallen into the long-term unimproved category[9]. The cropland pasture category in GTAP does not directly correspond to any single land category in the University of Wisconsin assessments, but the observations certainly suggest that assuming that over 90% of new cropland associated with ethanol demand comes from cropland pasture is likely to be a considerable over-estimate. The economic analysis by (Hendricks, 2018) also gives a very different result to the GTAP modeling. This economic analysis suggests that over 95% of grassland converted to cropping due to RFS was former CRP land. As CRP land is not pastured, it does not fall under the definition of cropland pasture in the agricultural census, and indeed GTAP has a separate category for CRP land introduced at the same time as the cropland pasture category.   

Lark et al. (2015) also document rates of forest land conversion to cropland that appear to be larger than anticipated by Qin et al. (2016)[10], although it is even more difficult for these relatively small forest area changes to draw a firm conclusion about whether this should be attributed to the ethanol mandate.

These discrepancies regarding the source of new land are potentially important in the ILUC analysis, because different land use changes have very different assumed carbon losses in the modeling. In particular, when using the CCLUB emission factors included in the GREET model the conversion of cropland pasture to corn is assumed to result in an increase in carbon sequestration, whereas Lark et al. (2015) argue that the conversion to cropland they identify is likely to be associated with very significant carbon dioxide emissions. This fundamentally different interpretation seems to arise primarily due to the use of a rather questionable modeling assumption in the development of the CCLUB emission factors. In CCLUB, soil carbon under the cropland pasture land use is modeled using DAYCENT and treating cropland pasture as if it has uniformly been cropped before 1951, then used as pasture for 25 years before spending the last 35 years being farmed for a generic crop. This land use history is almost the opposite of the cropland pasture definition (previously cropped but then used as pasture for several years. Combining this difficult to justify modeling choice with the high rates of cropland pasture conversion modeled with GTAP by Qin et al. (2016) leads to a dramatic reduction in the predicted ILUC factor. 

The importance of these emission factor assumptions can be further illustrated by comparing the average carbon loss per hectare assumed for conversion of grassland to cropland in Qin et al. (2016) (for the Corn2 scenario with CCLUB emission factors) with average values estimated by Spawn, Lark, & Gibbs (2012) for grassland conversion in the Lark et al., (2015) results. Spawn et al. (2012) estimate an average loss of about 50 tonnes of carbon per hectare for conversion of new land to corn agriculture. Qin et al. (2016) in contrast assume an average increase by 13 tonnes of carbon stored per hectare.

Summary

Recent analysis with the GTAP model has suggested that ILUC emissions from corn ethanol may be much lower than was calculated in either the initial or revised ILUC analysis for the California Air Resources Board, or by the US Environmental Protection Agency. The results from Qin et al. (2016) using GTAP with the CCLUB emission factor model have even been integrated into the GREET lifecycle analysis model. Comparing these GTAP results to land use changes identified by other sources, however, suggests that total U.S. land use changes may be slightly underestimated, and strongly suggests that the role of cropland pasture as a land source may be significantly exaggerated. These features of recent GTAP modeling, combined with highly questionable assumptions about carbon sequestration when cropland pasture is converted to corn cropping have likely led to significant underestimates of likely ILUC emissions.

References

Argonne National Laboratory. (2017). The Greenhouse gases, Regulated Emissions, and Energy use in Transportation Model. Retrieved August 2, 2018, from https://greet.es.anl.gov/

California Air Resources Board. (2014). Appendix I – Detailed analysis for indirect land use change. Sacramento, CA. Retrieved from https://www.arb.ca.gov/regact/2015/lcfs2015/lcfs15appi.pdf

Hendricks, N. P. (2018). The Impact of the Renewable Fuels Standard on Cropland Transitions.

Lark, T. J., Hendricks, N. P., Pates, N., Smith, A., Spawn, S. A., Bougie, M., … Gibbs, H. K. (2019). IMPACTS OF THE RENEWABLE FUEL STANDARD ON AMERICA ’ S LAND AND WATER. In American Academy for the Advancement of Science (AAAS) Annual Meeting (pp. 1–13). Washington DC.

Lark, T. J., Salmon, J. M., & Gibbs, H. K. (2015). Cropland expansion outpaces agricultural and biofuel policies in the United States. Environmental Research Letters, 10, 44003. Retrieved from http://stacks.iop.org/1748-9326/10/i=4/a=044003

Malins, C., Searle, S. Y., & Baral, A. (2014). A Guide for the Perplexed to the Indirect Effects of Biofuels Production. International Council on Clean Transportation. Retrieved from http://www.theicct.org/guide-perplexed-indirect-effects-biofuels-production

Qin, Z., Dunn, J. B., Kwon, H., Mueller, S., & Wander, M. M. (2016). Influence of spatially dependent, modeled soil carbon emission factors on life-cycle greenhouse gas emissions of corn and cellulosic ethanol. GCB Bioenergy, 8(6), 1136–1149. http://doi.org/10.1111/gcbb.12333

Spawn, S. A., Lark, T. J., & Gibbs, H. K. (2012). U.S. cropland expansion released 115 million tons of carbon (2008-2012). In America’s Grasslands Conference 11/15/2017, Fort Worth, TX. Fort Worth. Retrieved from http://www.gibbs-lab.com/wp-content/uploads/2017/11/spawn_Summary.pdf

U.S. Environmental Protection Agency. (2010). Renewable Fuel Standard Program (RFS2) Regulatory Impact Analysis, 2010(February 2010), 1109. http://doi.org/EPA-420-R-10-006., February 2010

Wright, C. K., Larson, B., Lark, T. J., Gibbs, H. K., Salmon, J. M., Gibbs, H. K., … Gibbs, H. K. (2017). Recent grassland losses are concentrated around U. S. ethanol refineries. Environmental Research Letters, 12(4), 044001. http://doi.org/10.1088/1748-9326/aa6446

Acknowledgement

This article was supported by the National Wildlife Federation.


[1] 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?

[2] Referred to in the CCLUB module of GREET as the ‘Corn 2013’ scenario.

[3] https://www.eia.gov/dnav/pet/hist/LeafHandler.ashx?n=pet&s=m_epooxe_ynp_nus_mbbl&f=a  and http://www.eia.gov/totalenergy/data/annual/showtext.cfm?t=ptb1003

[4] USDA NASS

[5] Calculated from AFDC data on total corn use for ethanol https://afdc.energy.gov/data/10339

[6] In the case of feed corn, this would primarily mean reducing demand for animal products and therefore reducing use of corn as animal feed.

[7] Corn and soy are often grown in rotation.

[8] https://afdc.energy.gov/data/10339

[9] Of a total 1.9 million acres of conversion of cropland pasture plus other grassland in GTAP-CCLUB, 95% was from cropland pasture.

[10] In 2008-12 nineteen thousand hectare of forest conversion for every additional billion gallons of ethanol demand, about half of it identified by (Wright et al., 2017) as within 100 miles of ethanol refineries  compared to 6 thousand hectares per billion gallons in (Qin et al., 2016).