Model Comparison Exercise
Technical Document
£% United States
Environmental Protect
Agency
-------
Model Comparison Exercise
Technical Document
Transportation and Climate Division
Office of Transportation and Air Quality
U.S. Environmental Protection Agency
NOTICE
This technical report does not necessarily represent final EPA decisions
or positions. It is intended to present technical analysis of issues using
data that are currently available. The purpose in the release of such
reports is to facilitate the exchange of technical information and to
inform the public of technical developments.
United States
Environmental Protection
^1 Agency
EPA-420-R-23-017
June 2023
-------
Executive Summary
A primary policy goal of the Renewable Fuel Standard (RFS) program is to reduce
greenhouse gas (GHG) emissions by increasing the use of renewable fuels, such as ethanol and
biodiesel. In the Energy Independence and Security Act (EISA), Congress required that biofuels
used to meet the RFS obligations achieve certain lifecycle GHG reductions. To qualify as a
renewable fuel under the RFS program, a fuel must, among other requirements, be produced
from qualifying feedstocks and have lifecycle GHG emissions that are at least 20 percent less
than the baseline petroleum-based gasoline and diesel fuels.1 To determine whether fuels meet
the lifecycle GHG emissions threshold requirement, EPA developed a methodology to evaluate
the lifecycle GHG emissions of renewable fuels. EISA also provided a definition of "lifecycle
greenhouse gas emissions" to guide this methodology.2
In the March 2010 RFS2 rule, EPA used lifecycle analysis (LCA) to estimate the GHG
emissions associated with several biofuel production pathways, i.e., the emissions associated
with the production and use of each biofuel, including significant indirect emissions, on a per-
unit energy basis. At the time of the analysis for the 2010 RFS2 rule, there were no models
available "off the shelf' that could perform the type of lifecycle analysis required by EISA.
Several supply chain LCA tools existed at the time, e.g., the Greenhouse Gases, Regulated
Emissions, and Energy Use in Technologies Model (GREET). However, EPA determined in the
final RFS2 rule that these tools, when used on their own, lacked the ability to consider significant
indirect emissions, one of the core statutory requirements of the EISA definition of lifecycle
greenhouse gas emissions. EPA thus developed a new modeling framework to perform the
required analysis. The framework EPA developed and ultimately used in the 2010 RFS2 rule
included multiple models and data sources, including the Forest and Agricultural Sector
Optimization Model with Greenhouse Gases model (FASOM), the Food and Agricultural Policy
Research Institute international model developed at the Center for Agriculture and Rural
Development at Iowa State University (the FAPRI-CARD model, or, more simply, FAPRI), and
the GREET model.3
Since the development of EPA's 2010 LCA methodology, multiple researchers and
analytical teams have further studied and assessed the lifecycle GHG emissions associated with
transportation fuels in general and crop-based biofuels in particular. New models have been
developed to evaluate the GHG emissions associated with biofuel production and use, and more
models developed for other purposes have been modified and expanded to evaluate biofuels as
well. We now have over a decade of historic observations to compare with model results and
parameters and to use in model calibration. There has also been rapid growth in available data on
land use, farming practices, crude oil extraction and many other relevant factors. While the
1 See 42 USC 7545(o)(l), (2)(A)(i).
2 EISA defines lifecycle greenhouse gas emissions as "the aggregate quantity of greenhouse gas emissions
(including direct emissions and significant indirect emissions such as significant emissions from land use changes),
as determined by the Administrator, related to the full fuel lifecycle, including all stages of fuel and feedstock
production and distribution, from feedstock generation or extraction through the distribution and delivery and use of
the finished fuel to the ultimate consumer, where the mass values for all greenhouse gases are adjusted to account
for their relative global warming potential." CAA 21 l(o)(l)(H).
3 EPA (2010). Renewable fuel standard program (RFS2) regulatory impact analysis. Washington, DC, US
Environmental Protection Agency Office of Transportation Air Quality. EPA-420-R-10-006. Chapter 2.4.
1
-------
results from our 2010 LCA methodology for the RFS program remain within the range of more
recent estimates from the literature, we acknowledge that our previous framework is
comparatively old, and that a better understanding of these newer models and data is needed. In
consultation with our interagency partners at USD A and DOE, EPA hosted a virtual public
workshop on biofuel GHG modeling on February 28 and March 1, 2022.4 At this workshop,
speakers within and outside of the federal government presented on available data, models,
methods, and uncertainties related to the assessment of GHG impacts of land-based biofuels.
The workshop presentations and public input clarified that there continues to be
substantial uncertainty and a wide range of estimates on the climate effects of biofuels,
especially regarding biofuel-induced land use change emissions. Uncertainties in land use change
emissions estimates stem from both economic modeling of market-mediated effects as well as
biophysical modeling of soil carbon and other biological systems and processes. The workshop
proceedings, including the workshop presentations and the comments submitted to the workshop
docket, discussed a broad and complex set of topics. A general theme that emerged from this
process is that, in support of a better understanding of the lifecycle GHG impacts of biofuels, it
would be helpful to compare available models, identify how and why the model estimates differ,
and evaluate which models and estimates align best with available science and data. Recognizing
this need, we have conducted a model comparison exercise (MCE) to better understand these
scientific questions.
While we are presenting the results of this MCE along with the RFS "Set" final
rulemaking, the MCE does not model or otherwise inform the GHG impacts of the Set final
volumes. Although this MCE produced GHG emission and carbon intensity results5 from a range
of models under different assumptions, we do not use these values in the context of RFS program
implementation. For example, we do not use the MCE to determine whether or not fuel pathways
meet the lifecycle GHG threshold requirements of the CAA. Rather, the MCE has three main
goals:
1. Advance the science in the area of analyzing the lifecycle greenhouse gas emissions
impacts from increasing use of biofuels.
2. Identify and understand differences in scope, coverage, and key assumptions in each
model, and, to the extent possible, the impact that those differences have on the
appropriateness of using a given model to evaluate the GHG impacts of biofuels.
3. Understand how differences between models and data sources lead to varying results.
We conducted this model comparison exercise with five models: the Greenhouse Gases,
Regulated Emissions, and Energy Use in Technologies Model (GREET), Global Biosphere
Management Model (GLOBIOM), Global Change Analysis Model (GCAM), Global Trade
4 For more information see the Federal Register Notice, "Announcing Upcoming Virtual Meeting on Biofuel
Greenhouse Gas Modeling." 86 FR 73756. December 28, 2021. More information is also available on the workshop
webpage: https://www.epa.gov/renewable-fiiel-standard-program/workshop-biofiiel-green.honse-gas-modeling.
5 In general, a carbon intensity, or CI, is a measure of greenhouse gas emissions per unit of fuel. Assumptions
related to the estimation of emissions or changes in volumes of fuel may differ between studies which define CI with
different scopes or for different purposes.
2
-------
Project (GTAP) model, and Applied Dynamic Analysis of the Global Economy (ADAGE)
model. To facilitate appropriate comparisons of these models, we ran common scenarios through
each framework: a reference case, a corn ethanol scenario (also referred to as the "corn ethanol
shock"), and a soybean oil biodiesel scenario (also referred to as the "soybean oil biodiesel
shock").
Given the complex nature of these models, and the scope and scale of the analysis
involved, drawing firm conclusions from a comparison of these models and their results — and
presenting them for interested stakeholders — presents several challenges. We discuss these
challenges in detail throughout this document. However, despite the challenges inherent in such
a comparison, we have drawn several broad conclusions from this exercise, including the
following:
• Supply chain LCA6 models, such as GREET, produce a fundamentally different
analysis than economic models, such as ADAGE, GCAM, GLOBIOM, and GTAP.
Supply chain LCA models evaluate the GHG emissions emanating from a particular
supply chain, whereas economic models evaluate the GHG impacts of a change in biofuel
consumption.7
• Estimates of land use change (LUC) vary significantly among the models used in
this study. Drivers of variation in these estimates include differences in assumptions
related to trade, the substitutability of food and feed products, and land conversion, as
well as structural differences in how models represent land categories. The variability of
LUC estimates significantly influences variability in overall biofuel GHG estimates.
• Economic modeling of the energy sector may be required to avoid overestimating
the emissions reduction from fossil fuel consumption. Economic models that include
energy market impacts (ADAGE, GCAM, GTAP) estimate a global refined oil
displacement that is less than the increase in biofuel consumption on an energy basis.
• Model trade structure and assumed flexibility influence the modeled emissions
results. There is general agreement among the economic models that these trade-driven
impacts will occur to some degree. However, these models show different degrees of
trade responsiveness, which impacts trade flows at differing magnitudes across model
results.
• Explicit modeling of the global livestock sector, and especially of the impact of biofuel
feed coproducts on global feed markets, is an important capability for estimating the
emissions associated with an increase in biofuel consumption.
• The degree to which other vegetable oils replace soybean oil diverted to fuel
production from other markets can impact GHG emissions associated with soybean
6 Many terms are used in the LCA literature to describe this type of analysis, such as attributional LCA, lifecycle
inventory analysis, or process-based LCA. We use the term "supply chain LCA" as we believe it is descriptive of
what this type of modeling considers.
7 As discussed more in Section 1, different types of LCA approaches are appropriate for different applications. In
this exercise, we are not evaluating which approaches could be appropriate for RFS program implementation.
3
-------
oil biodiesel. Results in this exercise from economic models (ADAGE, GCAM,
GLOBIOM, and GTAP) align in estimating commodity substitution as a significant part
of their scenario solutions.
• The ability to endogenously consider tradeoffs between intensification and
extensification is an important capability for estimating the emissions associated
with an increase in biofuel consumption. Both intensification and extensification of
corn and soybean feedstock production occur across economic model results (ADAGE,
GCAM, GLOBIOM, and GTAP) in response to changing commodity prices.8
• Models included in the MCE produced a wider range of LCA GHG estimates for
soybean oil biodiesel than corn ethanol. The models show much greater diversity in
feedstock sourcing strategies for soybean oil biodiesel than they do for corn ethanol, and
this wider range of options contributes to greater variability in the GHG results.
• Differences in model assumptions, parameters, and structure impact the results from each
of the models. Sensitivity analysis, which considers uncertainty within a given model,
can help identify which parameters influence model results. However, pinpointing the
direct causes of why one estimate differs from another would require additional research.
This document describes EPA's biofuel lifecycle GHG emissions model comparison
exercise in detail. In the first section, we describe our goals and scope for the exercise. Following
this we describe the models included in the comparison and their key characteristics. We then
describe the core scenarios evaluated for this project and the model estimates from those
scenarios. After that, we describe alternative scenarios and sensitivity analyses we conducted to
further improve understanding of these models. Finally, we summarize our findings and discuss
areas of future research and next steps.
EPA is interested to hear from stakeholders and researchers working in this field about
the results of our MCE, and we intend to engage with stakeholders to discuss this analysis. As
we describe throughout the document, this MCE has helped EPA to identify important
characteristics of existing models, areas for future data collection, and areas for additional
research. As we engage with stakeholders, EPA will be interested to hear perspectives on the
state of science and models in light of the findings of this exercise. As we engage in these
conversations, we will also seek areas to collaborate with stakeholders on the priority areas for
further research identified below, such as collecting new data, leveraging existing data sets,
conducting economic and statistical studies, and running additional model scenarios. Ultimately,
EPA hopes that the examination of models and understanding that flow from the exercise will
lend itself to informing the scientific discussion on which and to what extent biofuels contribute
to reduced environmental harm in comparison to consuming petroleum-based fuels.
8 We define intensification as an increase in the amount of crop production on a given area of land, and
extensification as an increase in the total area used to grow the crop of interest. Where we use the term
extensification, we are including both non-cropland that was converted to cropland and shifting of cropland from
one type of crop to another. However, our discussion of the results shows cropland shifting and land conversion to
cropland separately.
4
-------
Model Comparison Exercise Goals and Scope
1 Goals of Model Comparison
We conducted a model comparison exercise (MCE) with five models: the Greenhouse
Gases, Regulated Emissions, and Energy Use in Technologies Model (GREET), Global
Biosphere Management Model (GLOBIOM), Global Change Analysis Model (GCAM), Global
Trade Project (GTAP) model, and Applied Dynamic Analysis of the Global Economy (ADAGE)
model. As mentioned above, this MCE had three main goals:
1) Advance the science in the area of analyzing the lifecycle greenhouse gas emissions
impacts from increasing use of biofuel.
2) Identify and understand differences in scope, coverage, and key assumptions in each
model, and, to the extent possible, the impact that those differences have on the
appropriateness of using a given model to evaluate the GHG impacts of biofuels.
3) Understand how differences between models and data sources lead to varying results.
This effort is consistent with some of the conclusions and recommendations in the
National Academies of Sciences, Engineering, and Medicine (NASEM) report titled "Current
Methods for Life Cycle Analyses of Low-Carbon Transportation Fuels in the United States."9
For example, NASEM recommended that "[cjurrent and future LCFS [low carbon fuel standard]
policies should strive to reduce model uncertainties and compare results across multiple
economic modeling approaches and transparently communicate uncertainties," (recommendation
4-2) and "LCA studies used to inform policy should explicitly consider parameter uncertainty,
scenario uncertainty, and model uncertainty" (recommendation 4-3).
LCA plays several diverse roles in the context of the RFS program. For example, LCA is
used for rulemaking impact analysis as well as to determine whether an individual pathway
meets the lifecycle GHG emissions reduction requirements. Different LCA tools may be
appropriate for different purposes. The NASEM report concluded that, "[t]he approach to LCA
needs to be guided on the basis of the question the analysis is trying to answer. Different types of
LCA are better suited for answering different questions or achieving different objectives, from
fine tuning a well-defined supply chain to reduce emissions, to understanding the global,
economy-level effect of a technology or policy change" (conclusion 2-2).10
9 National Academies of Sciences, Engineering, and Medicine ("NAS") (2022). Current Methods for Life Cycle
Analyses of Low-Carbon Transportation Fuels in the United States. Washington, DC: The National Academies
Press, https://doi.org/10.17226/26402.
10 The NASEM report provided the following recommendations related to LCA approaches: "When emissions are to
be assigned to products or processes based on modeling choices including functional unit, method of allocating
emissions among co-products, and system boundary, ALCA [attributional lifecycle analysis] is appropriate.
Modelers should provide transparency, justification, and sensitivity or robustness analysis for modeling choices"
(Recommendation 2-1). "When a decision-maker wishes to understand the consequences of a proposed decision or
action on net GHG emissions, CLCA [consequential lifecycle analysis] is appropriate. Modelers should provide
transparency, justification, and sensitivity or robustness analysis for modeling choices for the scenarios modeled
with and without the proposed decision or action" (Recommendation 2-2).
5
-------
This document includes multiple sections:
• Section 2 introduces and summarizes the models considered in this exercise.
• Section 3 compares model characteristics, input parameters, and input data.
• Section 4 describes the common scenarios that were run across all the models for
purposes of this analysis.
• Section 5 provides details on the reference case used.
• Section 6 compares the results of the modeling work related to corn ethanol.
• Section 7 compares the results of the modeling work related to soybean oil biodiesel.
• Section 8 describes the scenarios run as part of our alternative volume sensitivity
analysis.
• Section 9 describes parameter sensitivity analyses.
• Section 10 summarizes the findings of this exercise and discusses future research.
2 Models Considered
Numerous factors influence biofuel GHG estimates, including model framework choice,
data inputs and assumptions, and other methodological decisions. In this section we discuss the
models considered in this MCE: GREET, GLOBIOM, GCAM, GTAP,11 and ADAGE.12 This
selection of models provides a broad cross-section of the most common types of modeling
frameworks used to assess biofuels, as discussed in this section. We chose to use these models
based on discussions with our partners at USDA and DOE and our experience reviewing
scientific literature on the lifecycle GHG emissions of biofuels, including for our 2022 biofuel
LCA workshop discussed above. In addition, our choice to use these particular models is also
informed by the statutory definition of lifecycle greenhouse gas emissions in Section
21 l(o)(l)(H) of the Clean Air Act, which includes significant indirect emissions, including
indirect land use change emissions.13 Furthermore, in the 2010 RFS2 rule EPA interpreted this
11 There are multiple GTAP models. The version used for this model comparison exercise is the GTAP-BIO model.
For brevity we refer to it throughout this report as "GTAP" or the "GTAP model", except for instances where we are
describing the distinctions between GTAP-BIO and other GTAP models.
12 The model runs for this exercise were conducted by members of the modeling teams at Argonne National
Laboratory, IIASA, PNNL, Purdue University, and RTI International. The final contents of this document do not
necessarily represent the views of the modeling teams involved or the organizations they represent. All statements in
this document are ultimately those of EPA.
13 The full text of CAA 21 l(o)(l)(H) is "The term "lifecycle greenhouse gas emissions" means the aggregate
quantity of greenhouse gas emissions (including direct emissions and significant indirect emissions such as
significant emissions from land use changes), as determined by the Administrator, related to the full fuel lifecycle,
including all stages of fuel and feedstock production and distribution, from feedstock generation or extraction
through the distribution and delivery and use of the finished fuel to the ultimate consumer, where the mass values
for all greenhouse gases are adjusted to account for their relative global warming potential."
6
-------
definition as including significant indirect emissions14 occuring anywhere in the world (i.e.,
international impacts), as GHG emission impacts are global.15
In this exercise, we did not include FASOM or the FAPRI-CARD model, which we used
for the 2010 RFS2 rule. Given time and resource constraints, we chose to focus on models with
global scope. FASOM is not a global model, and instead covers the continental USA. The
FAPRI-CARD model is no longer maintained at the same level as it was in 2010; for example,
most of its projections still end in the 2022/2023 marketing year. There is another FAPRI model
maintained by the University of Missouri that projects further into the future, but this model
covers only the USA in detail and does not include GHG emissions. This exercise was not meant
to include every possible model that could be used to estimate biofuel GHG emissions, and
omission of a model from this exercise does not preclude its use in the future.
We provide a summary of each model included in this exercise, including its history,
sectoral representation, spatial coverage and resolution, temporal representation, and GHG
emissions representation. We then compare the characteristics of these models and describe
previously published literature which may assist the reader in understanding which factors may
contribute to variation in the biofuel GHG estimates these models produce. Our goal in this
section is not to provide a comprehensive accounting of any one of these models. Rather, our
objective is to summarize each model at a high level and highlight important similarities and
differences between models that we explore further when discussing MCE modeling results in
Sections 5-9.
There are four types of models commonly used for biofuel GHG analysis: supply chain
LCA models, partial equilibrium (PE) models, computable general equilibrium (CGE) models
and integrated assessment models (IAM). Supply chain LCA models, also known as attributional
LCA (ALCA) models, such as GREET, are designed to estimate the inputs and outputs of a
particular product supply chain in detail, using rule-based methods (e.g., allocation or
displacement) to account for coproducts.16 PE models, such as GLOBIOM,17 equate supply and
demand in one or more selected markets such that prices stabilize at their equilibrium level. PE
models focus on representing one or a few sectors of the economy, such as the agricultural
sector, but lack linkages to other sectors of the economy. In contrast, CGE models, such as
GTAP and ADAGE, are comprehensive in their representation of the economy, reflecting
feedback effects among all economic sectors and factors of production, such as land, capital,
14 When using the terms "direct" and "indirect" to refer to emissions, impacts or effects, NAS (2022) recommends
carefully defining these terms, or avoiding their use altogether (Recommendation 4-1). Given that the CAA
21 l(o)(l)(H) definition of lifecycle emissions uses the terms direct and indirect emissions, we believe it is
appropriate to use the direct/indirect terminology in this document. As a general matter, when we use the term
"direct emissions" in this document we are referring to emissions from the fuel supply chain itself, whereas "indirect
emissions" refers to emissions that results from market-mediated impacts induced by a change in biofuel
consumption. The same distinction holds for direct/indirect impacts or effects.
15 EPA. 2010. RFS2 Final Rule, 75 FR 14670 (March 26, 2010), https://www.gpo.gov/fdsvs/pkg/FR-2010-Q3-
26/pdf/20.1.0-385.1. .pdf. See in particular Section V, pages 14764-14799.
16 Supply chain LCA models such as GREET can also be supplemented with results from economic models to
consider indirect effects such as land use changes; however, doing so "can complicate the interpretation" of the
results (NAS 2022, p. 45).
17 The FASOM and FAPRI models EPA used for the March 2010 RFS2 rule biofuel GHG analysis are also
categorized as PE models.
7
-------
labor and resources. IAMs, such as GCAM, integrate knowledge from several disciplines, for
example, biogeochemistry, economics, engineering, and atmospheric science, to evaluate how
changes in any of these areas affect the others. While it is hard to state the specific criteria for
identifying an IAM, we might distinguish them from PE and CGE models by their deeper
integration of human economic systems with Earth (biosphere and atmosphere) systems and
GHG emissions into one modelling framework.
PE, CGE and IAM models can all be called economic models since their model solutions
include achievement of a partial or general economic equilibrium. Supply chain LCA models are
categorically different from the other three model types as they do not simulate economic
equilibria, behavior, or prices. Instead, supply chain LCA models inventory the emissions that
occur along each stage of a supply chain and assign or attribute the emissions to a functional
unit, such as a volume or energy unit of fuel.18 In contrast, the other types of models (PE, IAM,
CGE) can be used for a consequential lifecycle analysis, which looks at how the emissions or
impacts, including market-mediated impacts, will change in response to a decision or action,
such as a change in the level of biofuel consumption.19 All of these models have strengths and
weaknesses, as well as uncertainties and limitations. Thus, there are often tradeoffs to consider
when selecting between models for a particular analysis. For example, there may be tradeoffs
between sectoral and temporal scope on the one hand, versus supply chain and technological
resolution on the other. The potential tradeoffs between scope and detail most relevant to this
MCE are discussed in more detail in Section 3. As discussed above, when considering these
tradeoffs, the NASEM report says that analysts need to be guided on the basis of the question
their analysis is trying to answer.20
2.1 The Greenhouse Gases, Regulated Emissions, and Energy Use in
Technologies (GREET) Model
The Greenhouse gases, Regulated Emissions, and Energy use in Technologies (GREET)
Model is a lifecycle analysis model based on supply chains of technologies and products. It
provides lifecycle energy, water, GHG, and other air emissions results intended to evaluate the
impacts of various vehicle and fuel combinations, as well as chemicals, products, and materials
that crosscut major economic sectors. The developer is Argonne National Laboratory (ANL), and
the project is sponsored by the U.S. Department of Energy (DOE). Initially made available in
1995, it was developed with the purpose of evaluating the energy and environmental (e.g., GHG
emissions, criteria air pollutant emissions, and water consumption) impacts of new fuels and
vehicles for use in the transportation sector.21
18 NAS (2022) lists many definitions of an attributional lifecycle analysis without prescribing one particular
definition. This sentence is adapted from the first sentence under the heading "Attributional Life-Cycle Assessment
on page 22 of NAS (2022).
19 NAS (2022) lists many definitions of a consequential lifecycle analysis without prescribing one particular
definition. This sentence is adapted from the first sentence under the heading "Consequential Life-Cycle Assessment
on page 26 of NAS (2022).
20 NAS (2022), conclusion 2-2.
21 Elgowainy, A. and Wang, M. (2019) 'Overview of Life Cycle Analysis (LCA) with the GREET Model', p. 21.
https://greet.es.anl.gov/files/workshop 20.1.9 overview.
8
-------
GREET includes a suite of models and tools. For the transportation sector, it includes a
fuel cycle model of vehicle technologies and transportation fuels (GREET1) and a vehicle
manufacturing model of vehicle technologies (GREET2). Given that our focus is on renewable
fuels, we are primarily concerned with GREET1. GREET is available in two platforms, a large
Excel workbook and a ".net" version. The Excel version of GREET provides transparency while
the .net version offers a modular user interface with a structured database. There are several
derivates of the core GREET model, such as CA-GREET developed with the California Air
Resources Board (CARB) and used in support of the California Low Carbon Fuels Standard
(CA-LCFS), and ICAO-GREET developed with the International Civil Aviation Organization in
support of the Carbon Offsetting and Reduction Scheme for International Aviation (CORSIA).
New versions of GREET are normally released in October of each year, with the latest version as
of the time of this writing being GREET-2022. GREET includes more than 100 fuel production
pathways including fuels used in road, air, rail, and marine transportation. It also examines more
than 80 on-road vehicle/fuel systems for both light and heavy-duty vehicles. The model reports
lifecycle energy use, air pollutants, GHGs and water consumption. It includes detailed
representations of the petroleum, electric, natural gas, hydrogen, and renewable energy sectors.
The GREET modeling framework is largely a process-based LCA approach (sometimes
referred to as attributional LCA).22 GREET can be used to estimate the carbon intensity (CI)23 of
individual supply chains and the benefits of specific supply chain adjustments, such as reducing
fertilizer application rates or switching to more efficient fuel distribution modes. Fundamentally,
GREET is most closely related to other supply chain LCA frameworks such as SimaPro, GaBi,
and OpenLCA, though GREET differs in that it comes with predeveloped fuel pathways and pre-
populated data and assumptions developed by ANL. In general, GREET evaluates production of
a fuel commodity by considering the activities from the associated supply chain. In the context of
GREET, the data on the activities controlled within a fuel commodity supply chain are called the
"foreground" data. GREET accounts for important biofuel coproducts such as distillers grains
and soybean meal through allocation or displacement rules. Figure 2.1-1 provides a schematic
overview of how the biofuel lifecycle is represented in GREET. GREET can be used to estimate
the CI of individual supply chains and the benefits of specific supply chain adjustments, such as
reducing fertilizer application rates or switching to more efficient fuel distribution modes. The
model can also consider technology improvements at the process- or site-specific level for
biofuels.
22 Wang, M. (2022). "Biofuel Life-cycle Analysis with the GREET Model." Presentation at the EPA Biofuel
Modeling Workshop. Argonne National Laboratory. March I, 2022.
https://www.epa.gov/svstem/files/documents/2022-03/biofuel-ghg-model-workshop-biofuel-lifecvcle-analvsis-
greet-modei-2022-03 -01 .pdf. Slide 5.
23 Carbon intensity is a measure of greenhouse gas emissions per unit of fuel.
9
-------
Figure 2.1-1: Schematic of Biofuel Supply Chain Representation in GREET24
•« Fuel Production (Well to Pump) »
Feedstock
Energy
Farm input
manufacturing
• On-farm energy
consumption
Feedstock
production
CO} wniwlorw fiom
lime/urea application
Land use change
emission*
Total emissions
per ton biomass
Conversion
Fuel combusted In
vehicle}
Energy consumed in
pr«-p recasting
Energy
Process chemkals
Feedstock
storage and
transportation
Feedstock
conversion
Co-product (e.g., animal feed)
Displacement of conventional
products
Fuel combusted In
vehicles
Biofuel
transportation
and distribution
Total emissions
per gallon fuel
Energy /material
inputs
IXA ct^ja
Emissions
_ Fuel Combustion _
(Pump to Wheels)
Biofuel
combustion
GREET primarily estimates default fuel CIs using data for average resource and energy
production in the United States. In the context of GREET, these data on resource and energy
production are referred to as the "background data." For example, GREET by default models
electricity based on data for average U.S. electricity generation. However, GREET includes
some pathways representing foreign fuel production (e.g., Brazilian sugarcane ethanol) and in
some cases users can choose to model som e supply chains located in particular regions of the
U.S. (e.g., states or electricity grid regions). A user with enough data on their supply chain could,
in certain cases, cu stomize the background data in GREET to estimate the CI of their fuel
considering regional details and particular suppliers of energy and material inputs.
GREET is not a dynamic model as it does not make projections whereby future time
periods depend on the simulation of prior time periods. However, it does include projected
background data, using projections from sources such as the U.S. Energy Information
Administration (EIA). GREET users can select a target year, between 1990-2050, to estimate
lifecycle emissions for their supply chain given background data assumptions for the selected
year. Thus, it can be used to show how the estimated CI of a fuel changes over time based on
changes in technological efficiency and other factors. For example, Lee et al. (2021) used data on
U.S. ethanol production efficiencies and corn yields to estimate the CI of U.S. corn ethanol each
year from 2005 to 2019.23
Although GREET does not endogenously estimate indirect emissions such as those
resulting from direct and indirect land use change, GREET incorporates a static module called
the Carbon Calculator for Land Use Change from Biofuels Production (CCLUB) to account for
241 Copied from Wang (2022), slide 9.
25 Lee, U„ et al. (2021). "Retrospective analysis of the US com ethanol industry for 2005-2019: implications for
greenhouse gas emission reductions." Biofuels, Bioproducts and Biorefining.
10
-------
land use change emissions.26 CCLUB relies on a set of estimated induced land use changes for
various biofuel pathways obtained from GTAP studies conducted between 2011-2018 (see Table
2.1-1), combined with emissions factors estimated with a parametrized CENTURY model and
derived from various data sources to estimate land use change GHG emissions per unit of biofuel
production.27 Thus, the well-to-wheel emissions for crop-based pathways are estimated as the
process-based emissions plus the induced land use change estimates from CCLUB. The data
sources and calculations in CCLUB are summarized in Figure 2.1-2, reproduced from the
CCLUB user manual.
Figure 2.1-2: Schematic of Data Sources and Calculations in CCLUB28
Models & Data
Sources
Output
CCLUB
Calculations
Domestic &
International
Domestic
Only
GTAP
WinrockS
Woods Hole
Dntascts
Parameterized
CENTURY
Biofuel production . . , . . .
r Land transition) by area and
scenarios
* type
Abovcgroiind carbon
stocks (for forest and/or
grassland)
Bclowground or soil carbon
stocks
LUC
scenarios
LMC scenarios
Adljust US forest area baseline
with US Forest Service data
Carbon emission factors
IPCC N*0 emission factors
IPCC CH* amission factors
Belowground or toil carbon
stocks
Soil carbon emission factor*
I
NnO emission factors
Carbon Online
^ Aboveground carbon
Carbon emission factors of
Estimator
stocks (only for forest)
Irarvested wood product
GHG emissions (g CO^MJ)
by combining land area
chafes v*iih emissions
factors and applying
assumptions
Soil carbon emissions
{g C03e/tij)
by applying assumptions
Spatial coverage
County
AEZ
CountrWBiomc
CCLUB includes land use change area estimates from nine different GTAP scenarios:
four soybean oil biodiesel shocks, two corn ethanol shocks, and one shock each for ethanol from
corn stover, miscanthus and switchgrass. The corn ethanol and soybean oil biodiesel scenarios
included in CCLUB are described in Table 2.1-1. The two corn ethanol scenarios are similar
except that the "Corn Ethanol 2013" estimate was produced with a version of GTAP with
regionally differentiated land transformation elasticities and a modified land nesting structure
that makes it more costly within the model to convert forest to cropland relative to converting
pasture to cropland.
26 Kwon, Hoyoung, et al. (2021). Carbon calculator for land use change from biofuels production (CCLUB) users'
manual and technical documentation, Argonne National Lab, Argonne, IL. https://greet.es.anl.gov/publication-
cclub-manual-r7-2021
27 Hoyoung Kwon and Uisung Lee (2019) 'Life Cycle Analysis (LCA) of Biofuels and Land Use Change with the
GREET Model', https://greet.es.anl.gov/files/workshop 2019 biofuel luc.
28 Kwon, Hoyoung, Liu, Xinyu, Dunn, Jennifer B„ Mueller, Steffen, Wander, Michelle M„ and Wang, Michael.
(2020). Carbon Calculator for Land Use and Land Management Change from Biofuels Production (CCLUB). United
States: N. p., 2020. Web. doi:10.2172/1670706. Copy of Figure 1.
11
-------
Table 2.1-1: Corn Starch and Soybean Oil Based Biofuel Scenarios Available in CCLUB29
Case Description
Shock Size
(Billion Gallons)
Source
"Corn Ethanol 2011." An increase in corn ethanol
production from its 2004 level (3.41 billion gallons
[BGD to 15 BG
11.59
Taheripour et al.
(2011)30
"Corn Ethanol 2013." An increase in corn ethanol
production from its 2004 level (3.41 billion gallons
[BGD to 15 BG
11.59
Taheripour and
Tyner (2013)31
Increase in soybean oil biodiesel production by
0.812 BG(CARB case 8)
0.812
Chen et al.
(2018)32
Increase in soybean oil biodiesel production by
0.812 BG (CARB average proxy)
0.812
Chen et al. (2018)
Increase in soybean oil biodiesel production by 0.8
BG (GTAP 2004)
0.8
Taheripour et al.
(2017)33
Increase in soybean oil biodiesel production by 0.5
BG (GTAP 2011)
0.5
Taheripour et al.
(2017)
For each case, the estimates CCLUB uses from GTAP are the area of changes in
cropland, forest, pasture in each agro-ecological zone (AEZ) and region, and cropland pasture in
the U.S., Brazil, and Canada. Land use change GHG emissions are estimated based on these land
conversion areas using data from a few different sources. Based upon user selections, CCLUB
ultimately combines a given GTAP scenario's estimated land use change impacts with sets of
user-selected emission factor data34 to provide domestic and international land use change GHG
emissions per functional unit of biofuel. By default, for corn ethanol and soybean oil biodiesel,
among other crop-based fuels, GREET adds the LUC GHG estimates from CCLUB to the rest of
the supply chain LCA estimates to produce a CI score for each fuel pathway.
A module called the Feedstock Carbon Intensity Calculator (FD-CIC) was more recently
added to GREET.35 FD-CIC is designed to examine CI variations of different corn, soybean,
sorghum, and rice farming practices at the farm level. The FD-CIC uses county level data and
allows users to input their own farm level data on energy and chemical farming inputs, tillage,
cover cropping and other crop management practices. Based on these input data, the FD-CIC
29 Adapted from Table 1 in Dunn, J. B., et al. (2017). Carbon calculator for land use change from biofuels
production (CCLUB) users' manual and technical documentation, Argonne National Lab. (ANL), Argonne, IL
(United States).
30 Taheripour, F., et al. (2011). Global land use change due to the U.S. cellulosic biofuels program simulated with
the GTAP model, Argonne National Laboratory: 47.
31 Taheripour, F. and W. E. Tyner (2013). "Biofuels and land use change: Applying recent evidence to model
estimates." Applied Sciences 3(1): 14-38.
32 Chen, R., et al. (2018). "Life cycle energy and greenhouse gas emission effects of biodiesel in the United States
with induced land use change impacts." Bioresource Technology 251: 249-258.
33 Taheripour, F., et al. (2017). "The impact of considering land intensification and updated data on biofuels land use
change and emissions estimates." Biotechnology for Biofuels 10(1): 191.
34 For this model comparison exercise, we use the default emissions factor data used by GREET, which are from the
parameterized CENTURY model and Winrock. See Kwon, Hoyoung, et al. (2021) for details.
35 Liu, X., et al. (2020). "Shifting agricultural practices to produce sustainable, low carbon intensity feedstocks for
biofuel production." Environmental Research Letters 15(8): 084014.
12
-------
estimates the farm level emissions from energy, fertilizers, herbicide, and insecticide, as well as
effects on soil organic carbon relative to the baseline assumptions in GREET. The FD-CIC may
be useful to estimate the soil carbon benefits of reduced tillage and cover cropping, and to
examine regional differences or farm-level differences in feedstock CI.
While GREET accounts for indirect land use change emissions, it does not consider other
indirect effects associated with a change in biofuel demand, such as through market-mediated
impacts on the agriculture, livestock, or energy sectors.
GREET is used by a variety of academic, commercial, and government entities.
California's Low Carbon Fuel Standard (LCFS) program relies in part on a customized version
of GREET called CA-GREET to provide state-specific fuel pathways and CI values.36 Oregon
uses a similar approach for their LCFS program.37 The International Civil Aviation Organization
(ICAO) uses GREET among several models to provide carbon intensities for specific aviation
fuel pathways.38 Most of these programs (with the exception of Oregon) use the non-land use
change GHG estimates from GREET and add their own land use change estimates in specific
market and policy contexts instead of those derived from CCLUB to calculate biofuel carbon
intensities. Among other applications, EPA has used GREET since the inception of the RFS
program to provide data for rulemakings and biofuel pathway support as part of our suite of tools
in addition to FASOM and FAPRI.
2.2 The Global Biosphere Management Model (GLOBIOM)
The Global Biosphere Management Model (GLOBIOM) was developed and continues to
be managed by the International Institute for Applied Systems Analysis (IIASA). The model was
developed in the late 2000s originally to conduct impact assessments of climate change
mitigation policies of biofuels and other land-based efforts.39 It was developed on the basis of the
U.S. Forest and Agricultural Sector Optimization Model (FASOM model).40 There are several
model versions of GLOBIOM available for different applications and contexts. A sample of
GLOBIOM code is available to the public, and an open-source version is under development.41
36 California Air Resources Board. LCFS Life Cycle Analysis Models and Documentation.
https://ww2.arb.ca.gov/resonrces/docnments/lcfs-life-cvcle-analvsis-models-and-docnmentation.
37 Oregon Department of Environmental Quality. Carbon Intensity Values: Oregon Clean Fuels Program.
https://www.oiegon.gov/dea/ghgp/cfp/Pages/Clean-Fiiel-Pathwavs.aspx. This version is based on a previous version
of Argonne GREET.
38 ICAO. Models and Databases, https://www.ieao.int/environmentat-proteetion/pages/modetting-and-
databases.aspx.
39 International Institute for Applied Systems Analysis, "GLOBIOM," https://iiasa.ae.at/models~toots~data/gtobiom.
40 Frank, Stefan, et al. "The Global Biosphere Management Model,"
https://www.epa.gov/svstem/fites/docnments/2022-03/biofiiel-ghg-model-workshop-gtobat-biosphere-mgmt-model-
2022-03-01.pdf. See also, Valin. Hugo et al. The Land Use Change Impact of Biofuels Consumed in the EU:
Quantification of Area Greenhouse Gas Impacts. August 27, 2015, pg. 128.
41 See, GLOBIOM, "Model Code," https://iiasa.githnb.io/GLOBIOM/model code.html.
13
-------
Figure 2.2-1: GLOBIOM Regional Mapping
42
1 -4
GLOBIOM is a PE model that captures the agricultural, forest, and bioenergy sectors.
The model solves recursively dynamic using an economic equilibrium modeling approach with
detailed grid cell land representation.43 The model finds market equilibria that maximize the sum
of producer and consumer surplus subject to resource, technological, demand and policy
constraints at a country/regional level. Producer surplus is defined as the difference between
market prices at a regional level and the product's supply curve at the regional level. The supply
curve accounts for labor, land, capital and other purchased input. Consumer surplus is based on
the level of consumption of each market and is arri ved at by integrating the difference between
the demand function of a good and its market price. The model uses linear programming to
solve, although it also contains some non-linear functions that have been linearized using
stepwise approximation.44 GLOBIOM features global coverage with 37 regions (see Figure 2.2-
1) and simulates for the years 2000-2100 using ten-year time steps. As a PE model, GLOBIOM
does not have feedback from labor, capital, or other parts of the economy. However, the model
can be linked to other models, such as IIASA's energy sector model MESSAGE.
42IIASA. (2020). "GLOBIOM regional and country level modeling." SUPREMA GLOBIOM-MAGNET Training.
December 4, 2020. https://iiasa.github.io/GLOBIOM/training material/GLOBIOM/GLOBIOM-
Topie RegionalApplications APalazzo Nov2020.pdf.
43 In models with recursive dynamic solution algorithms, the model solves at each time step before moving forward
to the next time step. In contrast, forward looking optimization models solve for all time periods at once.
44 IIASA, "GLOBIOM Documentation_20180604.pdf,"
https://iiasa.github.io/GLOBIOM/GLOBIOM Documentation 20180604.pdf.
14
-------
Figure 2.2-2: Schematic Overview of GLOBIOM45
Population, GDP, consumer preferences
¦d
c
l/>
D
~TD
C
ro
Worldwide:
18 crop* (FAO ~ SPAM)
Management systems
low/high input & rrigated
EU28:
9 additional crops,
crop rotations.
Management opfeons
fertilizer, irrigation & tillage
RUMINANT
Digestibility model
-> Feed intake
-> Animal production
GH6 emissions
7 animals
(FAO • Orldded livestock)
Cattle & Buffalo
Sheep & Goat
*1
Poultry
8 different systems
BIOFNFRGY
Processing
•wj
jlBL
-> MJ twofuel
MJ bioelectric
Coproducts
Perennial crops
Short rotation coppke
Conversion technologies
Firat ceneraton biofues
Second generation
biofuei
Bio mass power p ants
G4M
Global Forest model
Harvestable wood
Harvesting costs
Downscaled FAO FRA at
grid level
Area
Carbon stock.
Age
Tree M#
Species
Rotation time
Thinning
>
o
u
C
n>
The detailed grid cell-level spatial coverage for GLOBIOM includes more than 10,000
spatial units worldwide. The model represents 18 crops globally (and nine additional crops in
Europe) using FAOSTAT as the primary database for crop statistics. Area of other crops that are
not represented dynamically (e.g., fruits and vegetables) are kept constant. Crop modeling
includes differentiation in management systems and multi-cropping.
GLOBIOM also features highly detailed livestock representation, based on FAOSTAT
data. The model includes 7 animal products, which can be produced in differentiated production
systems. For ruminants there are 8 production system possibilities, including grazing systems in
different climatic locations such as arid and humid, mixed crop-livestock systems, and others.
Pigs and poultry are classified under either small holder or industrial systems. Based on the
production system, animal species, and region, GLOBIOM differentiates diets, yields, and GITG
emissions. For instance, dairy and meat herds are modeled separately, and their diets are
differentiated. Poultry in industrial systems is split into laying hens and broilers, again with
different dietary needs.
;ation
Gridded representation of world land use
Cropland
Managed forest
Natural
forest
'Other
natural land
45IIASA. GLOBIOM Online Documentation. https://iiasa.github.io/GLOBIOM/introduction.htinl.
15
-------
For ruminants, livestock production is modeled spatially in GLOBIOM's gridded cell
structure. At the cell level, animal yields for bovine and small ruminants are estimated using the
GLOBIOM module, RUMINANT. RUMINANT calculates a production yield that matches
plausible feed rations and checks this against regional-level data of livestock production. Feed
for animals is also differentiated in the RUMINANT model and can be composed of feed crops,
grass, stover, and other feed. Monogastric productivities are calculated based on FAOSTAT and
assumptions of potential productivities of smallholder and industrial systems. Livestock
production is allowed to intensify or extensify, thereby altering the amount of feed or grass
consumed.46 Since for ruminants this is modeled spatially, any changes in grassland consumed
due to changes in production systems, animal type, yield, and GHGs is captured in the spatially-
relevant areas. Each final livestock product is considered a homogenous good with its own
specific market (apart from bovine and small ruminant milk).
Forestry in GLOBIOM is captured through the G4M module47 and includes detailed
representation of the sector and its supply chain and a differentiation between managed and
unmanaged forest areas. GLOBIOM includes bilateral trade for agricultural and wood products.
These products are assumed to be homogenous and traded based on least expensive production
costs though transportation costs and tariffs are also included.
The model also includes a bioenergy sector with first and second generation biofuels and
biomass power plants. Perennial crops and short-rotation coppice are included as inputs to the
bioenergy sector. GLOBIOM represents biofuel coproducts including distillers grains, oilseed
meals, and sugar beet fibers. These coproducts can be traded either in their processed or whole
forms. Coproducts that can be used for livestock feed are incorporated into the livestock
RUMINANT module and can substitute other forms of feed depending on protein and
metabolizable energy content.48
There are nine land cover types in GLOBIOM, and 6 of these are modeled dynamically:
cropland, grassland, short rotation plantations, managed forests, unmanaged forests, and other
natural vegetation land. The other three land cover categories are represented in the model but
kept constant, they include other agricultural land, wetlands, and not relevant (ice, water bodies
etc.). Greenhouse gas emission coverage includes 12 sources of emissions that cover crop
cultivation, livestock, above and below-ground biomass, soil-organic carbon, and peatland.
Although GLOBIOM does not track terrestrial carbon stocks dynamically, carbon fluxes from
land use change are calculated with equations, following IPCC guidelines, that estimate changes
over time and allocate the average annual emissions to the time period in which the land use
change occurs.
46 Intensifying involves increasing livestock output without expanding the area of pasture land by grazing more
livestock per area of land, increasing feed relative to grazing, or using feedlots. Extensifying is the opposite - it
involves expanding pasture area in order to increase livestock production.
47 International Institute for Applied Systems Analysis, "Global Forest Model (G4M)", https://iiasa.ac.at/models-
and-data/global-forest-model.
48 Valin, Hugo, et al., September 17, 2014, "Improvements to GLOBIOM for Modelling of Biofuels Indirect Land
Use Change," http://www.globiom-ilnc.en/wp-content/nploads/2014/12/GLOBIOM All improvements Septl4.pdf.
pg. 38.
16
-------
Land use in GLOBIOM allows for both intensification and extensification. When land is
converted, this is endogenously determined in the model based on conversion costs, and the
profitability of primary products, coproducts, and final products. Costs increase as the area
converted expands. Additionally, there are biophysical land suitability and production potential
restrictions. Land use change is determined at the grid cell level.49 There is a land transition
matrix that sets the options for land conversion for each cell and is based on land conversion
patterns specific to that region and conversion costs depending on the type of land converted.50
In the USA and EU regions, GLOBIOM, by default, does not allow forest conversion and
restricts natural land conversion though these assumptions can be changed.
In policy settings, GLOBIOM is used for both modeling the European Union's biofuel
mandates and for estimating induced land use change impacts of biofuels for the International
Civil Aviation Organization's Carbon Offsetting and Reduction Scheme for Civil Aviation
(CORSIA). In research contexts, the model has regularly participated in AgMIP, an agricultural
model intercomparison and improvement project.51 One result of this project was an article on
the key determinants of global land use projections.52 GCAM, discussed in Section 2.3, was also
part of the AgMIP study. GLOBIOM has been used to assess other topics in the academic
literature, publishing work on topics such as reducing greenhouse gas emissions from the
agricultural sector, food security, and climate mitigation of livestock system transitions.
2.3 The Global Change Analysis Model (GCAM)
The Global Change Analysis Model (GCAM) is a partial equilibrium, integrated
assessment modeling framework which explores human and earth dynamics. The model includes
representation of energy, economy, land, water, and physical earth systems and interactions
between these systems within a fully integrated computational system. The model includes all
human systems and economic sectors which produce or consume energy, or which emit GHGs.
GCAM operates as a recursive dynamic framework, generally in 5-year time steps. In practice,
the model is often run from a base year in the recent past through the years 2050 or 2100.
However, time step and scenario length are flexible input assumptions to GCAM, and the
framework can support scenario analysis across a wide range of time scales. By default and for
the purposes of this model comparison exercise, the model base year is currently 2015. But other
historical base periods may be specified. For each modeled time period, GCAM iterates until it
finds a vector of prices that clears all markets and satisfies all consistency conditions. The model
49 GLOBIOM represents most land in the world using a 5 arcminutes by 5 arcminutes grid. At the equator, this is
roughly 9km by 9km.
50IIASA, "Spatial Resolution and Land Use Representation,"
fattps://iiasa.gitfanb.io/GLOBIOMMoenmentation.fatint#spatiat~resotntion~and~tand~nse~representation.
51 Several studies have estimated water use and availability impacts associated with future scenarios of increased
cellulosic biofuel production. These studies often project future land use/management for different scenarios of
increased production of cellulosic crops, and then estimate impacts on water use and changes in streamflow for
specific watersheds. See for example: Cibin, R., Trybula, E., Chaubey, I., Brouder, S. M., & Volenec, J. J. (2016).
Watershed-scale impacts of bioenergy crops on hydrology and water quality using improved SWAT model. Gcb
Bioenergy, 8(4), 837-848 orLe, P. V., Kumar, P., & Drewry, D. T. (2011). Implications for the hydrologic cycle
under climate change due to the expansion of bioenergy crops in the Midwestern United States. Proceedings of the
National Academy of Sciences, 108(37), 15085-15090.
52 Stehfest, E., vanZeist, WJ., Valin, H. et al. Key determinants of global land-use projections. Nat Commun 10,
2166 (2019). https://doi.org/10.1038/s41467-019-09945-w
17
-------
is designed to explore different "what-if' scenarios, assessing the implications of different
futures on a wide range of outcomes, such as energy supplies and demands, land allocation, or
commodity prices.
The core GCAM is developed and maintained at the Joint Global Change Research
Institute, a partnership between Pacific Northwest National Lab (PNNL) and the University of
Maryland (UMD) in College Park, Maryland. PNNL is the primary steward of the model, though
members of a larger GCAM Community also contribute to development of the framework.53
GCAM was originally developed in the early 1980s to assess the magnitude of GHG emissions
from fossil fuel CO2 through the mid-21st Century. Over time, the model has expanded in scope
to serve a wide set of scientific modeling applications. The model has now been in continuous
development for over 40 years and has been applied in several studies and model inter-
comparison activities, including the IPCC's Representative Concentration Pathways54 and
Shared Socioeconomic Pathways.55 GCAM is an open-source community model that can be
downloaded from a public repository.56 The model documentation is also publicly available57
and includes a partial list of GCAM publications.58
Economic systems in GCAM are divided into sectors and, within each sector, specific
technologies. Figure 2.3-1 provides an overview of the sectors represented in GCAM, along with
the inputs and outputs of the model. As shown in the figure, there are exogenous natural resource
supply, land, economy, and demand inputs to the model. These exogenous inputs include global
population and GDP. Each sector of GCAM is structured with a multi-level nesting approach
that allows competition between different nodes at each level, and any number of levels. This
nested competition follows a discrete logit59 or modified logit model60, depending on the object.
The market share of each discrete technology is determined by a) a share-weight parameter that
reflects the specific preferences for a particular choice, b) the cost, which includes fuel and non-
fuel costs, and c) an exogenous logit exponent that determines the price responsiveness of the
competition. In most cases the share-weights are derived from base-year calibration when market
shares are known. Technologies that are introduced in future time periods are assigned
exogenous share-weights in each model time period. The market shares are therefore influenced
by a number of endogenous and exogenous parameters, including fuel and non-fuel costs,
efficiency or input-output coefficients, share-weights, and logit exponents. These parameters are
documented and can be consulted in online repository.61
53 For more information, see fattps://gei iris. pmil, gov/eo minimi tv.
54 Thomson AM, Calvin KV, Smith SJ, Kyle GP, Volke A, Patel P, et al. RCP4. 5: a pathway for stabilization of
radiative forcing by 2100. Clim Change 2011;109:77.
55 Calvin K, Bond-Lamberty B, Clarke L, Edmonds J, Eom J, Hartin C, et al. The SSP4: A world of deepening
inequality. Glob Environ Change 2017;42:284-96.
56 See https://githiib.com/JGCRI/gcam-core.
57 See http://igcri.githnb.io/gcain-doc/index.html.
58 See more specifically http://igcri.githnb.io/gcam-doc/references.htinL
59 McFaddenD. Conditional logit analysis of qualitative choice behavior 1973.
60 Clarke JF, Edmonds JA. Modelling energy technologies in a competitive market. Energy Econ 1993;15:123—9.
61 See Calvin et al. 2019. GCAM v5.1: Representing the linkages between energy, water, land, climate, and
economic systems. Geoscientific Model Development 12, 1-22. See also the online documentation
(https://github.com/JGCRI/gcam~doc/blob/gh-pages/ssp.md') for the specific quantification of the inputs and
parameters to the model.
18
-------
International trade of commodities in GCAM is specified using one of two methods.
Agricultural, livestock, and forestry primary goods are traded through regionally-differentiated
markets following an Armington-style approach.62 In the version of GCAM used for this
exercise, all other commodities are traded through homogenous global markets following the
Heckscher-Ohlin theorem. 63 These approaches are described in detail in GCAM's online
documentation.64
Figure 2.3-1: GCAM diagram of model inputs, sectors, and outputs65
INPUTS GCAM
OUTPUTS
Supply
Supply
• Resource bases
• Conversion
k.
technologies
V
• Agriculture
technologies
Land
• Baseline
land
productivity
• Baseline
carbon
density
• Land value
Economy
• Population
•Labor force
• Labor
productivity
Demand
• Demand
technologies
• Behavioral
assumptions
Land
• Land use & land
cover
¦ Carbon storage
Energy
• Coal, Gas, Oil
• Renewables
• Electricity
• Hydrogen
• Fertilizer
Water
• Renewable
• Groundwater
¦ Desalinated
Food, forestry,
etc.
•Crops
• Livestock
• Forest
• Bioenergy
•Fish
Economy
• Regional GDP
¦ Regional
population
Marketplace (prices and trade)
• Fossil fuel • Bioenergy • Water
• Electricity • Crops • Emissions
•Liquids 'Livestock -Fish
• Hydrogen • Forest
1
Demand
Energy
Water
Food, forestry, etc.
• Coal, Gas, Oil
• Irrigation
• Crops
• Renewables
• Municipal
• Livestock
• Bioenergy
• Industry
• Forest
• Electricity
• Livestock
• Aquacuiture &
• Hydrogen
• Electricity
Fish
• Primary
• Fertilizer
Quantity
• Energy production
•Energy consumption
•Agriculture production
• Agriculture consumption
•Water withdrawals
• Water consumption
• Water supply
Prices
•Energy
• Agriculture & Forestry
• Water
•Fish
Trade
•Energy
•Agricultures Forestry
• Water
•Fish
Land
• Land use
• Land cover
•Carbon fluxes
Emissions
• Greenhouse gases
(GHG)
• Non-GHG emissions
GCAM includes detailed representations of the energy sector, inclusive of liquid biofuels,
and the agriculture and land sectors. The energy sector module in GCAM consists of depletable
and renewable resources66, energy transformation and distribution sectors (electricity, refining,
61 The Armington approach to modeling international trade is based on the premise that products traded
internationally are differentiated by country of origin. This is in contrast to models that assume perfect substitution
between products produced in different countries. Armington, P. S. (1969). A Theory of Demand for Products
Distinguished by Place of Production. IMF Staff Papers, 1969 (001).
63 Note that the most recent public version of GCAM trades all energy' goods through the Annington-like approach,
rather than through homogenous markets. This version of the model was not released in time for inclusion in this
exercise.
64 See ttp://igcri.github.io/gcam-doc/details trade.html
65 See ttp://igcri.github.io/gcam-doc/index.html.
68 Depletable resources are based on graded supply curves for coal, oil. gas and uranium. Renewable resources
include annual flows of wind, solar, geothermal, hydropower, and biomass.
19
-------
gas processing, hydrogen production, and district services), and final energy demand sectors
(buildings, industry, and transportation).67 For transportation biofuels specifically (referred to in
the GCAM documentation as "biomass liquids"), by default the model includes a total of 11
biofuel production technologies. These include four "first generation" technologies, representing
ethanols and biodiesels produced from agricultural commodity crops, and seven "second
generation" technologies representing fuels produced from a variety of feedstocks, including
energy crops and residues. By default, the technology assumptions for second generation
represent the inputs and outputs of cellulosic ethanol and Fischer-Tropsch fuels. However, the
input assumptions for these technologies can be modified to represent other fuel production
pathways. Secondary outputs such as dried distillers grains (DDG) and electricity produced from
lignin can be considered, as can the potential for carbon capture and storage. Further description
of these technological representations is available in the online GCAM documentation.68
The agriculture and land use module differentiates 384 land use regions globally,
generated as the intersection of 32 socioeconomic regions with 235 water basins (see Figure 2-
2). Within each land use region, up to 25 land use types compete for land share based on the
relative profitability of each use, using a nested land allocator tree structure.69 The conversion of
land from one type to another is determined in part by the logit structure of the model and the
land nesting structure.70 GCAM land categories are structured in sub-nests, with easier
conversion between land types within a sub-nest than across sub-nests. Land use types include
exogenous land types (tundra, desert, urban), commercial and non-commercial pasture and forest
lands, grasslands and shrublands, and a detailed set of agricultural crop commodities, including
bioenergy crops, classified by irrigation type and fertilizer use.71
Within this nesting structure, the allocations of land to each land use type are calibrated
in the model base year, and in the future, changes from the base-year allocations are driven by
changes in the relative profitability of each land use type, including both commercial and natural
lands. Profitability of lands in agricultural and forestry production changes over time as a
function of future commodity prices, yields, and costs of production (including endogenous costs
of fertilizer, fuel, and irrigation water). The intrinsic profitability or value of natural lands is
inferred from the base year profitability of proximate land used for agriculture and forestry in
each region. The logit competition for land is non-linear and exhibits diminishing marginal
67 More detailed information on the GCAM energy system can be found in online documentation, see
http://igcri.githnb.io/gcain-doc/index.htinl and also in previous studies (see Clarke L, Eom J, Marten EH, Horowitz
R, Kyle P, Link R, et al. Effects of long-term climate change on global building energy expenditures. Energy Econ
2018;72:667-77; Muratori M, Ledna C, McJeon H, Kyle P, Patel P, Kim SH, et al. Cost of power or power of cost:
A US modeling perspective. Renew Sustain Energy Rev 2017;77:861-74.)
68 See http://igcri.githnb.io/gcain-doc/snpplv energy.html.
69 See Wise M, Calvin K, Kyle P, Luckow P, Edmonds J. Economic and physical modeling of land use in GCAM
3.0 and an application to agricultural productivity, land, and terrestrial carbon. Clim Change Econ 2014;5:1450003,
and Zhao X, Calvin KV, Wise MA. The critical role of conversion cost and comparative advantage in modeling
agricultural land use change. Clim Change Econ 2020; 11.
70 See http://igcri.githnb.io/gcain-doc/details land.html
71 A complete description of the land use module can be found in the online documentation (see
http://igcri.gitlinb.io/gcain-doc/toc.htmi') and in Kyle GP, Luckow P, Calvin KV, Emanuel WR, Nathan M, Zhou Y.
GCAM 3.0 agriculture and land use: data sources and methods. Pacific Northwest National Lab.(PNNL), Richland,
WA (United States); 2011.
20
-------
returns to expansion of each use as well as non-constant elasticities.72 This nonlinear nature
allows the land shares to be solved based on equal value at the margin without need the explicit
constraints used in linear models.
GCAM also uses land suitability and land protection assumptions to determine what land
is available for expansion. All versions of GCAM divide land into arable and non-arable
categories and, by default, protect some portion of the arable land from conversion to agricultural
or silvicultural use. In the version of GCAM used for this exercise, GCAM-T, other assumptions
limit the suitability of arable lands for crop production based on biophysical limitations (e.g.,
slope, annual rainfall) and human-imposed limitations such as land protection policies. The latter
are parameterized using the International Union for Conservation of Nature's (IUCN) World
Database of Protected Areas.7j
Terrestrial carbon stocks and flows are modeled for each land type in each water basin.14
The agricultural sector of the model primarily relies on input data from the UN Food and
Agriculture Organization (FAO) historical data sets, and includes all crops for which FAO
reports area and production data for the model base year of 2015.75 Major global commodity
crops, such as corn, rice, soybeans and wheat are modeled individually, while all other crops are
modeled as a series of thematic aggregations.
figure 2.3-2: GCAM Regional Mapping76
Energy/Socioeconomics: 32 regions
yw- V,
Land: 384 regions
1J
y * ,
Water 235 regions
Climate: 1 global 'egion
In addition to the core GCAM described in this section, there exist several other
subversions and downscaling tools which can be used to examine regions and systems at a finer
grain of resolution. These include, among others, GCAM-USA77, which models each U.S. state
72 See Wise et al (2020).
For more information, see documentation provide at . ttps://gitlmb .com/gcamt/gcam-core/tree/GC AM-T-2020.
" Input assumptions related to terrestrial carbon and land transitions are documented at http://i gcri. github.io/gcam-
doc/land.html.
73 See ttp://i gcri. github.io/gcam-doc/inputs land.html for further data on land inputs to the model.
See ttp://i gcri. github.io/gcam-doc/overview.html.
77 See ttp://igcri.github.io/gcam-doc/gcam-usa.html.
21
-------
as an individual region, Tethys78, which allows for the downscaling of modeled GCAM water
impacts, and Demeter79, which allows for the downscaling of modeled land allocation impacts.
Numerous additional tools are in various stages of development at JGCRI and other research
groups which participate in the GCAM Community.80
One of these, GCAM-T, was used in a recent study of corn ethanol impacts by Plevin et
al. The results of that study are discussed in greater detail later in this chapter.81 GCAM-T is also
the version of the model used for the present model comparison exercise. This version of the
model includes greater detail in several sectors relevant to the modeling of transportation energy
technologies, including biofuels. The version of GCAM-T used for the Plevin et al paper,
GCAM-T 2020.0, is publicly documented.82 Additional documentation for the version of
GCAM-T used for this model comparison exercise, GCAM-T 2022.0, is included as a
memorandum to the docket.83 GCAM-T 2022.0 is referred to simply as "GCAM" for the
remainder of this RIA discussion and in the preamble of this final rulemaking.
In addition to biofuel modeling,84 GCAM is used for diverse purposes across a wide
range of stakeholders, including federal, state, and local U.S. government, foreign governments
and international governance bodies, academia, private industry, and non-governmental
organizations. As noted above, GCAM is used on an ongoing basis by the IPCC in the
development of socioeconomic and climatic projections via the Representative Concentration
Pathways85 and Shared Socioeconomic Pathways.86 Another notable recent application was the
use of GCAM to produce scenario analysis for the Long-Terms Strategy of the United States,
submitted to the United Nations under the Paris Agreement by the U.S. State Department and
Executive Office of the President.87 Numerous other research papers associated with GCAM are
accessible via PNNL's publications page for the model.88
2.4 The Global Trade Analysis Project (GTAP) Model
The GTAP-BIO model is an extension of the standard Global Trade Analysis Project
(GTAP) model which has been developed at the GTAP center of the Department of Agricultural
Economics at Purdue University to study the economic and environmental impacts of biofuel
production and policy.
78 https://githnb.com/JGCRI/tethvs.
79 https://githnb.com/JGCRI/demeter.
80 For more information, see littps://gei his . pmil, gov/eo minimi ty.
81 Plevin, R. J., et al. (2022). "Choices in land representation materially affect modeled biofuel carbon intensity
estimates." Journal of Cleaner Production: 131477.
82 See https://githnb.com/gcamt/gcam-core/tree/GCAM-T-2020 and https://zenodo.org/record/4705472.
83 See "GCAM-T 2022.0 Documentation" in the docket.
84 See for example, Mignone, B. K., Huster, J. E., Torkamani, S., O'Rourke, P., & Wise, M. (2022). Changes in
Global Land Use and CO2 Emissions from US Bioethanol Production: What Drives Differences in Estimates
between Corn and Cellulosic Ethanol?. Climate Change Economics, 13(04), 2250008.
85 Thomson AM, Calvin KV, Smith SJ, Kyle GP, Volke A, Patel P, et al. RCP4. 5: a pathway for stabilization of
radiative forcing by 2100. Clim Change 2011;109:77.
86 Calvin K, Bond-Lamberty B, Clarke L, Edmonds J, Eom J, Hartin C, et al. The SSP4: A world of deepening
inequality. Glob Environ Change 2017;42:284-96.
87 See lit tps ://unfccc. i nt/dociime nts/308.1.00
88 See https://gcims.pn.nl.gov/gcims-pnblications
22
-------
The GTAP center is the focal point of a global network of more than 27 thousand
researchers, scholars, academic institutions, and policy research entities that are conducting
quantitative analysis of a wide range of policy issues related to trade, energy, agriculture, and
climate change. The members of this network provide and share various databases, develop
modeling ideas and codes, conduct research, and disseminate their research findings. The GTAP
center facilitates these activities by providing various databases and modeling tools. In particular
this center assembles databases that support modeling practices around the world for various
modeling approaches. The standard GTAP database is centerpiece of these activities. The most
recent versions of this database include Input-output (I-O) tables for 160 regions converting the
whole world economic activities; bilateral trade data at global scale; production, consumption,
and trade of energy products; data on various types of GHG and non-GHG emissions generated
around the world; land use and land cover data; and several other items. The GTAP database is
particularly supports CGE modeling activities. However, it has been used by many other
modeling practices around the world. To various extents, several of the models participated in
this modeling comparison exercise rely on the GTAP database. The latest available version of
this standard database represents the global economy in 2017.
In addition to providing data, the GTAP center develops standard modeling platforms as
well. The standard GTAP model is the core of these platforms. This model has been originally
developed in 1999 and documented in Hertel (1999).89 This model and its extensions have been
used in many research activities and thousands of publications. Corong et al. (2017) has
introduced the latest version of this standard model and its capabilities and extensions, with
detailed discussion on the theory and derivation of the behavioral and equations in the model.90
The standard GTAP is a global, comparative static, multi-commodity, and multi-regional
Computable General Equilibrium model that traces production, consumption, and trade of all
good and service produced across the world. This model assumes perfect competition in all
markets with price adjustments to ensure that all markets are simultaneously in equilibrium.
Some GTAP versions deviate from the perfect competition assumption.
As shown in Figure 2.4-1, in each region of this model a regional household collects all
the income in its region and spends it over three expenditure types: private household
(representing all consumers), government, and savings, as governed by a utility function. A
representative firm maximizes profits subject to a production function that combines primary
factors of production including labor, land, capital, and resources and intermediate inputs to
produce a final good or service. Firms pay wages/rental rates to the regional household in return
for their uses of primary inputs. Firms also sell their output to other firms (as intermediate
inputs), private households, government, and investment. Since this is a global model, firms also
export the tradable commodities and import the intermediate inputs from other regions. These
goods or services are assumed to be differentiated by region and thus the model is able to track
bilateral trade flows. The model follows Armington assumptions for bilateral trade, to account
for product heterogeneity among outputs produced in different regions. Taxes are paid to the
89 Hertel, T.W., ed. 1997. Global Trade Analysis: Modeling and Applications. New York,
NY: Cambridge University Press.
90 Corong, E. L., Hertel, T. W., McDougall, R., Tsigas, M. E., & Van Der Mensbrugghe, D. (2017). The standard
GTAP model, version 7. Journal of Global Economic Analysis, 2(1), 1-119.
23
-------
regional household. The rest of the world receives revenues by exporting to the private
household, firms, and government. These revenues are spent on export taxes and import tariffs,
which eventually go to the regional household. The rest of world represents other regions of the
model.
As noted above, the standard GTAP model is a comparative static model. Hence, as noted
by Corong et al. (2017) "a GTAP simulation presents not changes through time, but differences
between possible states of the global economy - a base case and a policy case - at a fixed point
in time, or with respect to two points in time (base period vs. a future projection period)."91 The
version of GTAP used for this exercise is based on the 2014 database; thus, we can say that the
biofuel simulations for this exercise with GTAP estimate changes in the 2014 economy due to a
change in biofuel consumption. A typical comparative static simulation isolates the impacts of a
phenomenon or changes in one or a set of variables that may affect the global economy from
many other factors that vary over time.
Figure 2.4-1: Standard GTAP Model Analytical Framework92
Our model comparison exercise includes the GTAP-BIO model. While this comparative
static model is the most widely used GTAP model for biofuel analysis, we recognize there are
other GTAP models available that could potentially be used for this purpose. For example,
GDyn-BIO and GTAP-DEPS are recursive-dynamic versions of GTAP that have been used to
91 Ibid.
92 An updated version of the depiction first developed in Brockmeier M. (2011) "A graphical exposition of the
GTAP Model", GTAP Technical paper No. 08.
24
-------
model U.S. corn ethanol impacts.93 ENVISAGE is another dynamic model complemented by an
emissions and climate module that links changes in temperature to impacts on economic
variables such as agricultural yields.94 While we did not have the ability to include more than
one GTAP model in our current model comparison exercise, exploring and comparing the
capabilities of other GTAP models for biofuel analysis is a potential area for future research.
Such an exploration and comparison may consider multiple factors. For example, other GTAP
models do not currently carry all the modifications incorporated in the GTAP-BIO model to
show the role and importance of various factors that could affect the economic and
environmental impacts of biofuel production and policy. Assessing induced land use changes due
to biofuels has been the core of many of these GTAP-BIO modifications, and it has also been
used to evaluate the consequences of climate change, water scarcity, and environmental
policies.95 Another factor to consider are the trade-offs between using a historical comparative
static framework like GTAP-BIO, versus using a model that projects into the future. Projecting
changes in the global economy over time is helpful to answer certain analytical questions, and
requires making projections on many factors with associated uncertainties.
Over time, various modifications have been made in the standard GTAP databases to
study the economic and environmental impacts of biofuel production and policy. The standard
GTAP databases do not explicitly represent production, consumption, and trade of biofuels, their
byproducts and coproducts. They also lack proper sectoral disaggregation to support biofuel
studies. The GTAP-BIO databases have been generated to remove these barriers. These
databases explicitly represent traditional biofuels (grain-based ethanol, ethanol produced from
sugar crops and biodiesel produced from oilseeds) that are produced and consumed across the
world. Some GTAP-BIO databases represent more advance biofuel technologies that produce
road and aviation fuels from traditional feedstocks and lignocellulosic materials. These
databases, depending on the application, provide more disaggregated crops, and further
disaggregate some standard GTAP sectors to facilitate biofuel studies. For example, the
substitution between biofuels and fossil fuels occurs in a newly introduced sector that blends
fossil fuels and biofuels.
For analyzing land use change, the GTAP-BIO databases follow the GTAP-AEZ land
databases and divide the land rents and land areas of each country into 18 Agro-Ecological
93 Golub, A. A., et al. (2017). Global Land Use Impacts of U.S. Ethanol: Revised Analysis Using GDyn-BIO
Framework. Handbook of Bioenergy Economics and Policy: Volume II: Modeling Land Use and Greenhouse Gas
Implications. M. Khanna and D. Zilberman. New York, NY, Springer New York: 183-212.; Oladosu, Gbadebo, and
Keith Kline. "A dynamic simulation of the ILUC effects of biofuel use in the USA." Energy policy 61 (2013): 1127-
1139.
94 Van der Mensbrugghe, Dominique. "The environmental impact and sustainability applied general equilibrium
(ENVISAGE) model." The World Bank, January (2008): 334934-1193838209522.
95 A few examples are: Taheripour F., Hertel, T. W., & Ramankutty, N. (2019). "Market-mediated responses
confound policies to limit deforestation from oil palm expansion in Malaysia and Indonesia," Proceedings of the
National Academy of Sciences, 116 (38), 19193-19199; Pena-Levano, L. M., Taheripour, F., and Tyner, W. E.
(2019). "Climate change interactions with agriculture, forestry sequestration, and food security," Environmental and
Resource Economics, 74, 653-675; Yao G., Hertel T., and Taheripour F. (2018). "Economic drivers of telecoupling
and terrestrial carbon fluxes in the global soybean complex," Global Environmental Change, 5: 190-200; Liu J.,
Hertel T., Taheripour F., Zhu T., and Rigal C. (2014). "International trade buffers the impact of future irrigation
shortfalls," Global Environmental Change, Vol. 29, 22-31.
25
-------
Zones.96 The AEZs represent 18 relativity homogeneous groups of lands based on length of
growing days, moisture regions, and climate zones. The GTAP-BIO databases trace land cover
items (forest, pasture and cropland), harvested areas, and crops produced at AEZ level. While the
GTAP databases represent managed and unmanaged lands, in modeling induced land use
changes due to biofuels only managed lands are represented in GTAP-BIO for various reasons.97
Figure
1
2
The most recent version of GTAP-BIO available in time for our model comparison
exercise uses GTAP-BIO database version 10, representing the global economy in 2014.99 The
geographical aggregation of this this data is presented in Figure 2.4-3. Researchers at Purdue
have the ability to project a database forward in time based on macro-economic projections in
96 Hertel et al. (2009) described the original GTAP land use data. Baldos and Corong (2020) documented the recent
GTAP land use databases up to 2014. Hertel, T.W., S. Rose, and R. Tol. 2009. "Land use in computable general
equilibrium models: An overview." In Economic Analysis of Land Use in Global Climate Change Policy. United
Kingdom: Routledge. Routledge Explorations in Environmental Economics; Baldos U. and E. Corong (2020)
Development of GTAP 10 Land Use and Land Cover Data Base for years 2004, 2007, 2011, 2014. GTAP Research
Memorandum No. 36.
-------
order to simulate future time periods.100 EPA and Purdue explored the possibility of creating a
version of GTAP-BIO with a projected 2030 database to align better with the scenarios modeled
with the dynamic models in our model comparison. Unfortunately, we were unable to complete
this work in time for the model comparison exercise.
Figure 2.4-3: Economic regions represented in GTAP
i United States
3 European Union 27
I Brazil
I Canada
¦ Japan
] China and Hong Kong
I India
¦ Central and Caribbean Americas
n South and Other Americas
¦ East Asia
D Malaysia and Indonesia
¦ Rest of South East Asia
~ Rest of South Asia
¦ Russia
mother East Europe and Rest of Former Soviet
nRest of European Countries
~ Middle Eastern and North Africa
¦ Sub Saharan Africa
¦ Oceania Countries
GTAP-BIO has been updated multiple times to add features that are relevant for biofuel
GHG modeling. Tyner et al. (2010) included marginal lands and productivity estimates for
potential new cropland based on a biophysical model."" Taheripour et al. (2012) used a
biophysical model (TEM) and estimated a set of extensification parameters which represent
productivity of new cropland versus the existing land by AEZ region.102 Taheripour and Tyner
(2013) used a tuning process to differentiate land transformation elasticities by region based on
FAO data.103 Taheripour and Tyner (2013) modified the land supply tree putting cropland
pasture and dedicated energy crops (e.g., switchgrass) in one nest and all other crops in another
nest, "to make greater use of cropland pasture (a representative for marginal land) to produce
dedicated energy crops."104 Taheripour et al. (2016) altered the land use module of GTAP-BIO
100 Yao G., Hertel T., and Taheripour F. (2018). "Economic drivers of telecoupling and terrestrial carbon fluxes in
the global soybean complex." Global Environmental Change, 5: 190-200
101 Tyner, W. E., Taheripour. F., Zliuang, Q.. Birur, D.. & Baldos, U. (2010). Land use changes and consequent CO2
emissions due to US corn ethanol production: A comprehensive analysis. Department of Agricultural Economics,
Purdue University, 1-90.
102 Taheripour, F„ et al. (2012). "Biofuels, cropland expansion, and the extensive margin." Energy, Sustainability
and Society 2(1): 25.
103 Taheripour, F. and W. E. Tyner (2013). "Biofuels and land use change: Applying recent evidence to model
estimates." Applied Sciences 3(1): 14-38.
104 Taheripour, F. and W. E. Tyner (2013). "Induced Land Use Emissions due to First and Second Generation
Biofuels and Uncertainty in Land Use Emission Factors." Economics Research International 2013: 12.
27
-------
to include cropland intensification due to multiple cropping or returning idled cropland
production, defined a new set of regional intensification parameters and determined, and defined
regional yield responses to price based on analysis of regional changes in crop yields.105
Taheripour et al. (2017) brought all of these modifications into one version of GTAP-BIO using
the GTAP database representing 2011.106 The version of GTAP-BIO used in this exercise
includes the above developments and adds cropland pasture as a land category in all regions
using the FAO land use database, whereas the previous version included cropland pasture in only
the United States, Brazil and Canada.
GTAP estimates areas and types of land use change by region in response to a biofuel
shock. Given that this model does not endogenously estimate land use change GHG emissions,
land use change areas are translated to GHG emissions using either the AEZ-EF model107 or the
CCLUB module of GREET, which produce significantly different estimates.108 These tools
make assumptions about how land use changes will occur in the future. To calculate a land use
change CI metric, the land use change emissions are annualized (e.g., over 20-30 years,
depending on the policy context) and divided by the energy content of the simulated biofuel
shock. For this model comparison exercise, land use change areas estimated with GTAP are
converted to land use change GHG emissions with AEZ-EF, version 52, and annualized over 30
years.
In general, the GTAP-based models are able to evaluate changes in GHG emission due to
changes in economic activities. While the GTAP-BIO model has been used mainly to assess
induced land use change emissions, this model can also estimate changes in GHG and non-GHG
emissions due to changes in economic activities. For this model comparison exercise, we are
interested in broadly evaluating the capabilities of each model. Thus, we also consider GTAP
estimates for all global economic sectors such as energy, livestock and forestry. These estimates
include changes in CO2 and non-C02 emissions due to biofuel induced changes.109 While, this
report provides these results, the results could be further studied for potential improvements in
model parameters that govern changes in these emissions.
GTAP-BIO is used widely for biofuel land use change analysis. As discussed above, the
GREET model incorporates land use change estimates from this model through the CCLUB
module. The GTAP-BIO results are used to estimate induced land use change GHG emissions
for the California, Oregon, and Washington low carbon fuel standard programs. GTAP-BIO is
also one of two models, along with GLOBIOM, used to estimate induced land use change
emissions for the International Civil Aviation Organization (ICAO) Carbon Offsetting and
Reduction Scheme for International Aviation (CORSIA). Furthermore, GTAP-BIO has been
105 Taheripour, F., et al. (2016). An Exploration of Agricultural Land Use Change at Intensive and Extensive
Margins. Bioenergy and Land Use Change: 19-37.
106 Taheripour, F., et al. (2017). "The impact of considering land intensification and updated data onbiofuels land
use change and emissions estimates." Biotechnology forBiofuels 10(1): 191.
107 Plevin, R., Gibbs, H., Duffy, J., Yui, S and Yeh, S. (2014). Agro-ecological Zone Emission Factor (AEZ-EF)
Model (v52).
108 Chen, R., et al. (2018). "Life cycle energy and greenhouse gas emission effects of biodiesel in the United States
with induced land use change impacts." Bioresource Technology 251: 249-258. Figure 4.
109 Chepeliev, M. (2020). Development of the Non-CCh GHG Emissions Database for the GTAP Data Base Version
10A (No. 5993). Center for Global Trade Analysis, Department of Agricultural Economics, Purdue University
28
-------
used to estimate biofuel induced land use change emissions for numerous journal articles (see for
example the articles cited above).
2.5 The Applied Dynamic Analysis of the Global Economy (ADAGE)
Model
The Applied Dynamic Analysis of the Global Economy (ADAGE) model is a multi-
region, multi-sector computable general equilibrium (CGE) model developed and maintained by
RTI International.110 The original ADAGE model was a forward-looking model.111 It was
originally developed to examine impacts of climate change mitigation policies and was used, for
example, to analyze economy-wide impacts of various legislative proposals, including the
American Clean Energy and Security Act of 2009. More recently, the ADAGE model has been
developed to have additional sectoral detail, particularly in agriculture, bioenergy, and
transportation.112 This version of the ADAGE model (hereinafter referred to as "ADAGE" or
"the ADAGE model") is global, rather than national, and is recursive-dynamic, which means that
decisions about production, consumption, savings, and investment are based on previous and
current economic conditions.
ADAGE represents the entire economy, including private and public consumption,
production, trade, and investment, and follows the classical Arrow-Debreu general equilibrium
framework.113 The model uses nested constant elasticity of substitution (CES) production
functions. As illustrated in Figure 2.5-1, ADAGE includes representative households and firms,
and economic flows among households, firms, and government are considered. Bilateral trade is
represented using an Armington aggregation approach.114 Dynamics in ADAGE are represented
by 1) growth in the available effective labor supply from population growth and changes in labor
productivity; 2) capital accumulation through savings and investment; 3) changes in stocks of
natural resources; and 4) technological change from improvements in manufacturing, energy
efficiency and land productivity, and advanced technologies that become cost competitive over
time.
110 The ADAGE model is available at https://githnb.com/RTIIiiteniational/ADAGE.
111 Ross, M. 2009. Documentation of the Applied Dynamic Analysis of the Global Economy (ADAGE) Model.
Working paper 09 01. Research Triangle Park, NC: RTI International.
112 Cai Y., Beach R., Woollacott J., Daenzer K., 2023. Documentation of the Applied Dynamic Analysis of the
Global Economy (ADAGE) model. Technical Report. Available at lit!ps://githiib.cotn/RTIIntematiore IE.
113 Arrow, K.J., and G. Debreu. 1954. Existence of an equilibrium for a competitive economy. Econometrica 22:265-
290.
114 Armington, P. S. (1969). A Theory of Demand for Products Distinguished by Place of Production. Staff Papers -
International Monetary Fund, 16(1), 159-178.
29
-------
Figure 2.5-1: Representation of Economic Flows in the ADAGE model115
Region A Region B
Goods & Services
Region C
ADAGE includes additional detail for the energy, food, agriculture, and transportation
sectors. It runs in 5-year intervals from 2010 through 2050, and includes 8 global regions
(Africa, Brazil, China, EU 27, United States, Rest of Asia, Rest of South America, and Rest of
World; Figure 2.5-2). ADAGE is built off the GTAP v7.1 database which represents the global
economy in 2004,116 with additional data from other sources such as the International Energy
Agency, U.S. Energy Information Administration, and United Nations Food and Agriculture
Organization. These additional data help to extend the global economy from 2004 to 2010
through balanced growth and add more sectoral details and physical accounts. ADAGE tracks
inputs and outputs in monetary units, and also tracks commodities and resources in physical units
(such as energy units of fuel consumption, area of land, and mass of emissions).
115 Cai Y„ Beach R.. Woollacott J., Daenzer K., 2023. Documentation of the Applied Dynamic Analysis of the
Global Economy (ADAGE) model. Technical Report.
116 Narayanan, G. B.. and T. L. Walmsley (Eds.). 2008. Global Trade, Assistance, and Production: The GTAP 7
Data Base. West Lafayette, IN: Center for Global Trade Analysis, Purdue University.
http://www.gtap.agecon.purdue.edu/databases/v7/v7 doco.asp.
30
-------
Figure 2.5-2: ADAGE Regional Mapping
ADAGE models the markets for several agricultural commodities: wheat, corn, soybean,
sugarcane, sugar beet, rest of cereal grains, rest of oilseeds, and rest of crops, in addition to one
livestock category and one forestry category. The agricultural sector in the underlying GTAP
v7.1 database is more aggregated, so creating these commodities in ADAGE required
disaggregation using information on trade shares, consumption shares, cost shares, and own use
shares.117 This disaggregation was done with software called SplitCom118 and data from the
United Nations Food and Agricultural Organization FAOSTAT database and the United Nations
Cott)trade Database.1191'0 The "cereal grains" sector in GTAP v7.1 was split into corn and rest of
cereal grains, the oil seeds sector was split into soybean and rest of oilseeds, and the combined
sugarcane and sugar beet sector was split into sugarcane and sugar beet.
Agricultural sector details in ADAGE enable it to model several kinds of biofuels.
ADAGE includes 8 types of first-generation biofuels (corn ethanol, wheat ethanol, sugarcane
ethanol, sugar beet ethanol, soybean oil biodiesel, rape-mustard biodiesel, palm kernel biodiesel,
and corn oil biodiesel) and 5 types of advanced biofuels (ethanol from switchgrass, miscanthus,
agricultural residue, forest residue, and forest pulpwood). These biofuels are not included in the
GTAP 7.1 database and were split from GTAP v7.1 sectors using the SplitCom software and
secondary data from USDA's Economic Research Service, DOE's Energy Information
11 Beach, R.H.. D.K. Birur, L.M. Davis, andM.T. Ross. 2011. A dynamic general equilibrium analysis of U.S.
biofuels production. AAEA & NAREA Joint Annual Meeting, Pittsburgh, PA.
https://ageconsearch.iimn.edU/bitstream/103965/2/ADAGE-Biofuels AAEA Conference Paper.pdf.
118 Horridge, M.. J. Madden, and G. Wittwer. 2005. The impact of the 2002-2003 drought on Australia. Journal of
Policy Modeling 27(3):285-308.
119 Food and Agriculture Organization of the United Nations. 2012. FAOSTAT Database. Rome, Italy: FAO.
http://www.fao.Org/faostat/en/#data.
120 United Nations. 2012. UN Comtrade Database, http://eomtrade.un.org.
31
-------
Administration, and the United Nations Comtrade database.121'122'123 Corn ethanol and wheat
ethanol were split from the "food products sector" in GTAP v7.1, which receives inputs from
corn and wheat. Sugarcane ethanol and sugar beet ethanol were split from the chemicals sector.
Biodiesel from soybean, rapeseed, and palm oil were split from the vegetable oils and fats sector.
Distillers grains with solubles (DGS) and corn oil biodiesel are coproducts of corn ethanol
production. An oil meal coproduct was split from the vegetable oil sector in GTAP v7.1. Because
ADAGE does not explicitly represent rapeseed and palm oil production, the input shares of "rest
of oilseeds" is based on region-specific palm oil and rapeseed biodiesel yields (gallon of
biodiesel per ton of feedstock). Advanced biofuels were not included in the 2010 base year in
ADAGE but are allowed to enter the market in future years.
The energy sectors of the ADAGE model include coal, natural gas, crude oil, and refined
oil, and several categories of electricity generation technologies (conventional coal, conventional
natural gas, conventional oil, combined-cycle natural gas, nuclear, hydropower, geothermal,
wind, solar, and biomass). The supply of fossil fuels is limited by the availability of natural
resources, which is represented as a fixed factor in the model. Crude oil is used as an input for
refined oil and enters the production function in a fixed proportion. Electricity generation
technologies are combined into a single electricity output.
The transportation sector in ADAGE has been developed to include light duty vehicles,
freight trucks, buses, marine, aviation, freight rail and passenger rail. Biofuels can be consumed
in on-road transportation (light duty vehicles, buses, and trucks). Alternative fuel options
(hybrid, battery electric, fuel cell, and natural gas) are available for on-road vehicles. The GTAP
v7.1 database includes three types of transportation (air, water, and rest of transportation) and
was disaggregated using data from several sources.124
ADAGE includes six land types (cropland, pasture, managed forest, natural forest,
natural grassland, and other land125). Land use change is represented by the combination of a
given land type with materials, capital, and labor to produce a new land type. The amount of
conversion in a period is limited by a fixed factor that is substitutable with other inputs. Each
land type has its own endowment, land rent, and usage. The conversion cost between land types
is equal to the differences in land rents, involving input cost from the labor, capital, and materials
inputs for conversion activity. There are also constraints on the types of land that can be
converted to other types. For example, only pasture and managed forest can be converted directly
to cropland, but cropland can convert to any land type.126 A fixed factor elasticity is defined for
121 USD A, Economic Research Service (ERS). 2012. U.S. Bioenergy statistics. Washington, DC: U.S. Department
of Agriculture, https://www.ers.nsda.gov/data-prodncts/ns-bioenergv-statistics.
122 EIA. 2012. Petroleum & other liquids. Washington, DC: U.S. Department of Energy.
https://www.eia.gov/dnav/pet/pet move impctis a2 tins epooxe imO rnbbt a.htm.
123 United Nations. 2012. UN Comtrade Database, http://comtrade.nn.org.
124 Data sources include GCAM 4.2, the Bureau of Economic Analysis, the Bureau of Transportation Statistics, the
International Energy Agency, and the Energy Information Administration. For more details, see Cai Y., Beach R.,
Woollacott J., Daenzer K., 2023. Documentation of the Applied Dynamic Analysis of the Global Economy (ADAGE)
model. Technical Report.
125 "Other land" includes bare ground, wetlands, mangroves, salt marsh, glaciers, and lakes, and is assumed to be
constant over time.
126 Unmanaged forest can only be converted to managed forest, and grassland can only be converted to pasture.
Through these conversions, unmanaged forest and grassland could be converted to cropland over two time steps.
32
-------
each starting land type/ending land type pair. Elasticities are generally the same in every region.
However, the elasticities governing the conversion of natural forest to managed forest and
grassland to pasture vary by region. ADAGE models land in physical as well as monetary
quantities. Emissions from land use change are based on the differences in carbon stocks
(vegetative and soil carbon) between the land types, and emission factors (one for vegetative
carbon, and one for soil carbon) that represent the fraction of the change in carbon stock that
would occur over 20 years after land conversion. Land use change emissions and sequestration
are all reported in the model year in which the land use change occurs. Vegetative and soil
carbon stocks are based on data from GCAM 3.2, which were aggregated to ADAGE regions
using weighted land area.
ADAGE includes six types of greenhouse gases: carbon dioxide (CO2), methane (CH4),
nitrous oxide (N2O), perfluorocarbons (PFCs), hydrofluorocarbons (HFCs), and sulfur
hexafluoride (SF6). CO2 emissions from fossil fuel combustion are based on emissions factors
(kgC02/MMBTU) for coal, gas, and oil. The emission factors are differentiated by region and
based on data from EIA's International Energy Statistics. CO2 emission factors from sources
other than fossil fuel combustion and land use change are based on data from the Emissions
Database for Global Atmospheric Research (EDGAR) version 4.2.127 Non-CCh emission factors
are based on data from EPA.128
CGE models often represent individual economic sectors at a higher level of commodity
and technology aggregation than some PE models of those same economic sectors. However,
because CGE models capture the entire economy, they can be useful for determining impacts of
environmental policies across sectors and on GDP. In one study, the ADAGE model was used to
analyze projected impacts of the RFS on land use, crop production, crop prices, fossil energy
use, GHG emissions, and GDP.129 ADAGE has also been used to study the impact of oil prices
on biofuel expansion.130 In model comparison studies, ADAGE was used to analyze the GHG
abatement potential in Latin America,131 and the impacts of climate policy and agriculture,
forestry, and land use emissions.132
127 Joint Research Centre at European Commission. 2013. Emission Database for Global Atmospheric Research.
http://edgar.irc.ec.europa.eu/overview.php?v=42FT2010.
128 U.S. Environmental Protection Agency (EPA). 2012. Global Non-CCh GHG Emissions: 1990-2030. Washington,
DC: EPA. https://www.epa.gov/global-mitigation-non-co2-greetihonse-gases/global-non-co2-ghg-eniissions-1990-
2030.
129 Cai, Y., D.K. Birur, R.H. Beach, and L.M. Davis. (2013, August). Tradeoff of the U.S. Renewable Fuel Standard,
a General Equilibrium Analysis. Presented at 2013 AAEA & CAES Joint Annual Meeting, Washington, D.C.
130 Cai, Y., R.H. Beach, and Y. Zhang. (2014, March). Exploring the Implications of Oil Prices for Global Biofuels,
Food Security, and GHG Mitigation. Presented at 2014 AAEA Annual Meeting, Minneapolis, MN.
131 Clarke L., McFarland J., Octaviano C., vanRuijvenB., Beach R., DaenzerK., Herreras Martinez S., Lucena
A.F.P., Kitous A., Labriet M., Loboguerrero Rodriguez A.M., Mundra A., van der Zwaan B., 2016. Long-term
abatement potential and current policy trajectories in Latin American countries. Energy Econ. 56, 513-525.
http://dx.doi.Org/10.1016/i.eneco.2016.01.011.
132 Calvin K.V., Beach R., Gurgel A., Labriet M., Loboguerrero Rodriguez A.M., 2016. Agriculture, forestry, and
other land-use emissions in Latin America. Energy Econ. 56, 615-624.
http://dx.doi.Org/10.1016/i.eneco.2015.03.020.
33
-------
3 Comparison of Model Characteristics, Input Parameters, and Input Data
In this section we compare the characteristics of the five models described above in
Section 2. We compare the models across several characteristics that are important for biofuel
analysis. In later sections, we discuss how these model characteristics impact model results.
3.1 Model Characteristics
Table 3.1-1 summarizes some of the key characteristics of the five models featured in
Section 2. Although there are many ways to compare these models, we chose six key
characteristics based on their relevance to the definition of lifecycle greenhouse gas emissions in
Section 21 l(o)(l)(H) of the Clean Air Act.133 Specifically, we consider model sectoral coverage,
temporal resolution, regional coverage, GHG emissions coverage, land representation, and trade
dynamics. Differences among modeling frameworks along these coverage, resolution, and
dynamics characteristics may lead to significant differences in modeled perspectives on GHG
emissions outcomes. These six characteristics therefore provide a good starting point for
understanding the primary differences across these frameworks. We start our discussion based on
these six characteristics before touching on other key aspects of these models for biofuel GHG
analysis.
While we are not ruling out consideration or future use of other models, based on the
biofuel GHG modeling workshop and our review of the literature, we believe the models listed in
the table are the most likely to meet our needs for evaluating lifecycle GHG emissions. In
addition, the models selected provide a broad representation of the types of models that can be
used for lifecycle analysis.
133 Other important considerations are not included in this table, such as open access to the models.
34
-------
Table 3.1-1 Comparison of Key Characteristics Across Models
Characteristic
ADAGE
GCAM
GLOBIOM
GREET
GTAP
Type of Model
Computable
general
equilibrium
(CGE);
consequential
LCA
Integrated
assessment
model (IAM);
consequential
LCA
Partial
equilibrium
(PE);
consequential
LCA
Supply chain
LCA
Computable
general
equilibrium
(CGE);
consequential
LCA
Sectoral
Coverage
Economy-wide
with 36 sectors
Energy
(conventional
and renewable),
industry,
buildings,
transportation,
agriculture,
forestry, water
Agriculture,
forestry, and
bioenergy
Fuel supply
chains
including
energy
resource and
material inputs
Economy-wide
aggregated into
65 sectors
Temporal
Representation
Recursive
dynamic (5-
year time
steps)
Recursive
dynamic (5-year
time steps)
Recursive
dynamic (10-
year time steps)
Static (users
can select a
target year
from 1990-
2050)
Comparative
static
Regional
Coverage
8 economic
and spatial
regions
32 economic
regions; 384 land
regions (water
basins,
intersected with
economic
regions)
37 economic
regions; 10,000
spatial units
(grid cell)
Customizable
(typically U.S.
average)
19 economic
regions; 18
agro-ecological
zones
GHG Emissions
Coverage
Economy-wide
GHGs
including land
use change
Global GHGs
including land
use change
Crop
production,
livestock, and
land use change
Direct supply-
chain
emissions +
indirect land
use change
from CCLUB
module
Economy-wide
GHGs, with
land use change
GHGs
calculated with
the AEZ-EF
model
Land
Representation
(Arable land
categories
considered in
biofuel land use
change analysis)
Cropland,
pasture,
commercial
forest, non-
commercial
forest, natural
grassland,
other land
Cropland,
commercial
pasture and
forest, non-
commercial
pasture and
forest, shrubland,
grassland,
"protected" non-
commercial land
Cropland, other
agricultural
land, grassland,
commercial and
non-commercial
forest,
wetlands, other
natural land
Exogenous
(Land use
change
estimates from
GTAP-BIO
and CCLUB)
Cropland
(including
cropland-
pasture and
unused
cropland),
livestock
pasture,
"accessible"
forestry land
As observed above, modeling inherently involves trade-offs. For example, there may be
trade-offs between scope and detail, or between capabilities to understand individual supply
chains versus global impacts. Among the four model types considered in this exercise, the supply
chain LCA models, like GREET, have the most detailed technological representations but the
most limited scope. For example, the GREET model includes detailed representations of
numerous biofuel and energy production processes but does not include price-induced
interactions between supply chains or economic sectors or any other features which seek to
balance economic equilibria within or across sectors. PE models used for biofuel analysis tend to
35
-------
have a high level of detail in the agricultural sector, but limited interactions with other sectors.
For example, GLOBIOM has a detailed representation of crop production, livestock, and land
use, but does not include economic interactions between the agricultural and energy sectors (e.g.,
fuel prices are exogenous). CGE models are the broadest in economic scope, but they often
represent the world using a smaller number of physical regions and fewer specific technological
options within a given economic sector. IAMs focus on representing physical processes, but
often lack certain sectoral details relative to PE models, and treat more economic factors (e.g.,
global GDP) as exogenous relative to CGE models. When considering tradeoffs between these
methodological options, one must consider the goals of the analysis and whether cross-sectoral
impacts are potentially influential on the overall results. In instances where such impacts are
potentially influential, broader sectoral coverage is likely to be more critical. In instances where
such impacts are limited, or where the goal of the analysis is to understand GHG emissions from
a particular supply chain or sector, the narrower scope of a supply chain LCA or PE model may
be an acceptable tradeoff. Model comparison exercises can assist with these types of
assessments. We discuss below the extent to which cross-sectoral impacts appear relevant to
biofuel LCA modeling.
3.1.1 Sectoral Coverage
The modeling frameworks differ substantially in the scope of economic interactions that
they represent. Capturing a wide range of economic interactions is important for understanding
the overall GHG impacts, including indirect impacts, of crop-based biofuel production. Based on
economic theory, we expect increased consumption of crop-based biofuels to have complex
ripple effects through the entire world economy. For example, as the demand for feedstocks
increase, we expect the price of these commodities to increase, with consequences for
agricultural markets not only in the U.S., but around the world. These interactions are
complicated by the fact that the major crop-based biofuel feedstocks have coproducts (e.g.,
distiller grains, soybean meal) that are used as livestock feed. Given that producing biofuels
requires material (e.g., fertilizer) and energy (e.g., natural gas), increased biofuel production may
affect these input commodity markets as well. When biofuels displace gasoline or diesel in the
U.S., this change may affect consumer fuel prices and crude oil prices, which may in turn affect
other sectors of the economy.
Supply chain LCA models such as GREET do not include most of these economic
interactions. However, GREET includes agricultural sector interactions to a limited extent
through the exogenous addition of land use change GHG estimates. GLOBIOM models
economic interactions within and between the agricultural (including crops and livestock) and
forestry sectors. GLOBIOM also includes a bioenergy sector with limited economic interactions
other than through its consumption of feedstocks from the agricultural and forestry sectors.
GCAM models economic interactions within and among the energy, agriculture, forestry, and
water systems. The energy system in GCAM is highly developed, including energy production
from a broad range of technologies and resources, and energy consumption in the industrial,
commercial, residential, transportation, agriculture, and forestry sectors. As CGE models, GTAP
and ADAGE model interactions across the entire economy. Thus, CGE models include economic
interactions that the other modeling frameworks take as exogenous or do not include. As noted
above, however, this creates computational tradeoffs which often require CGE models to
36
-------
represent sectoral dynamics at a more highly aggregated level than other model types with
narrower scope.
The three models which represent energy market interactions (ADAGE, GCAM, and
GTAP) also differ in which energy commodities are represented and how demand for energy
commodities is linked to other model components. ADAGE represents production and bilateral
trade of crude oil, refined oil134, natural gas, coal, electricity, biodiesel (soy, palm kernel, rape-
mustard, corn oil), and ethanol (corn, wheat, sugarcane, sugar beet). ADAGE dynamically
represents the energy inputs required for extracting and refining petroleum and the inputs
required for production of biofuels. GCAM represents crude oil, refined oil, natural gas, coal,
electricity, biodiesel (soy, palm kernel, rapeseed, other oilseed-oil), and ethanol (corn, sugar
crops, energy grasses, crop residues). GCAM dynamically represents both the energy inputs
required for extracting and refining petroleum and the inputs required for growing and
transporting crops and producing biofuels.135 GTAP represents coal, crude oil, refined
petroleum, electricity, natural gas, corn ethanol, sugarcane ethanol, grain ethanol, soybean oil
biodiesel, rapeseed oil biodiesel, palm oil biodiesel, and other biodiesel. GTAP represents
production, consumption, and bilateral trade in these commodities.
3.1.2 Temporal Representation
Temporal representation, or the treatment of time dynamics, is another important
characteristic that differentiates the modeling frameworks. The ability to endogenously represent
temporal dynamics is an important model feature given that biofuel land use change emissions
occur over time (e.g., soil carbon levels change over multiple decades following land conversion)
and biofuel-induced effects are dependent on factors that change over time, such as crop yields
and overall demands of the population on land to produce food, feed, and fiber. GREET is
designed to simulate supply chains in a given year, and includes the flexibility for users to
choose background data (e.g., grid electricity mix) for future years extending out to 2050.136
GTAP is a comparative static model, meaning it simulates changes in the 2014 economy due to a
change in biofuel production or consumption.137 GLOBIOM, GCAM and ADAGE are recursive
dynamic models in which certain production, consumption, and investment decisions are made
on the basis of market conditions in each period with dependence on previous model periods
through capital and/or resource stocks. Conditions from previous periods are carried forward to
influence the next modeled period. This differentiates dynamic recursive frameworks
computationally from comparative static frameworks.
ADAGE and GCAM use 5-year time steps, whereas GLOBIOM uses 10-year time steps.
In ADAGE and GCAM, the time step represents a point in time (e.g., the 2020 time step
represents the estimated state of the world in the year 2020). In GLOBIOM, the time step
134 In these models, refined oil is an aggregation of all refined petroleum products, including gasoline and diesel.
135 Sampedro, J., Kyle, P., Ramig, C. W., Tanner, D., Huster, J. E., & Wise, M. A. (2021). Dynamic linking of
upstream energy and freight demands for bio and fossil energy pathways in the Global Change Analysis Model.
Applied Energy, 302, 117580. lit!ps ://doi.org/.1.0. .1.0.1.6/i.apenergy.202.1.. .1. .1.7580
136 However, as discussed above, if provided with sufficient data, GREET can estimate supply chain emissions for
different time periods
137 GTAP can model different time periods if the GTAP database is first manually projected forward (or backward)
based on assumptions. Due to time constraints, we were unable to perform such projections for this exercise.
37
-------
represents a long-term trend of changes over the applicable 10-year period (e.g., the 2020 time
step is a representative average of changes from 2011 to 2020).
3.1.3 Regional Coverage
Thorough understanding of the impacts of a change in biofuel consumption through LCA
requires consideration of significant indirect emissions. Many studies have shown that biofuel
consumption in the U.S. can have significant impacts in other regions of the world.138
Consequently, models need to represent all relevant regions to consider the full indirect impacts
of a change in biofuel consumption. Furthermore, regional representation is important due to
geographic variations related to terrestrial carbon stocks, agricultural yields, energy resources
and other factors. PE, CGE and IAM models often distinguish between economic regions and
biophysical regions. These models use solution algorithms to find market clearing conditions in,
and trade between, each of the economic regions. Biophysical regions are often defined based on
physical geography and geology to allocate economic activities and biophysical processes to
physical locations. GTAP models 19 economic regions and 18 non-contiguous AEZs (see
Figures 2.4-2 and 2.4-3). GLOBIOM models 37 economic regions and uses a spatially explicit
grid-cell approach to represent 10,000 spatial units worldwide. GCAM models 32 economic
regions and 235 global water basins—the intersection of the economic regions and water basins
produces 384 spatial subregions.139 ADAGE models 8 economic and geographic regions. In
contrast, GREET is not a geographic or regional model, but it can be customized to represent
biofuel production conditions for particular regions or supply chains. Data for GREET is
primarily representative of the USA. GREET also has modules that are designed to estimate soil
carbon and land use change emissions at a regional level. The FD-CIC module allows users to
estimate feedstock production emissions at county level, and the CCLUB module estimates
indirect land use change emissions based on the geographic regions represented by GTAP.
For this exercise, based on a template we provided to the modelers, ADAGE, GCAM,
and GLOBIOM reported results from eight mutually exclusive global regions: Africa, Brazil,
China, EU, USA, Rest of Asia, Rest of Latin America, and Rest of World. GTAP reported results
from 19 global regions. In this document, we generally present results from the USA region of
each model and an aggregation of the non-USA regions of each model.
3.1.4 GHG Emissions Coverage
There are notable differences in coverage of GHG emissions sources across the models.
These differences in which GHGs are included in each model lead to differences among biofuel
138 See for example, ICAO (2021). CORSIA Eligible Fuels ~ Lifecycle Assessment Methodology. CORSIA
Supporting Document. Version 3: 155; Plevin, R. J., J. Jones, P. Kyle, A. W. Levy, M. J. Shell and D. J. Tanner
(2022). "Choices in land representation materially affect modeled biofuel carbon intensity estimates." Journal of
Cleaner Production: 131477; Taheripour, F., X. Zhao and W. E. Tyner (2017). "The impact of considering land
intensification and updated data on biofuels land use change and emissions estimates." Biotechnology for Biofuels
10(1): 191.
139 Although we did not use it for this exercise, a spatial downscaling model called Demeter is able to present
GCAM land use results at higher spatial resolution (0.05° x 0.05°), but this tool is not used for this model
comparison. Chen, M., Vernon, C.R., Graham, N.T. et al. Global land use for 2015-2100 at 0.05° resolution under
diverse socioeconomic and climate scenarios. Sci Data 7, 320 (2020). https://doi.org/10.1038/s41597-020-00669-x.
38
-------
GHG emissions estimates produced from these models. As mentioned previously, GREET
estimates direct GHG emissions from a biofuel production supply chain and generally does not
include indirect market-mediated emissions from other sources and sectors. The exception is
indirect land use change emissions, which can be added exogenously to GREET results through
the CCLUB module. GLOBIOM endogenously calculates GHG emissions from agriculture,
including crop and livestock production, forestry, and land use change. GTAP reports three
overall categories of GHG emissions which collectively provide an estimate of global GHG
impacts: 1) fossil fuel combustion CO2 emissions, 2) non-CCh emissions including changes in
these emissions for energy and energy activities,140 and 3) land use change emissions.141
ADAGE endogenously calculates GHG emissions from the entire economy, including land use
change. GCAM endogenously calculates all global GHG emissions sources, including those
from the energy, agriculture, forestry and water systems, including from land use changes. Of the
five highlighted models, ADAGE, GCAM, and GTAP are the only models that capture GHG
emissions from market-mediated changes within the energy system.
It is important to note that although all five models seem to overlap in their coverage of
GHG emissions, they estimate GHG impacts using different methods. For example, GREET and
GLOBIOM both estimate GHG emissions from crop production, but they do so in fundamentally
different ways. GREET estimates the GHG emissions associated with producing the crops that
are directly used in the biofuel supply chain under evaluation. In contrast, GLOBIOM estimates
the GHG emissions associated with the market-mediated marginal changes in crop production
stemming from a biofuel shock (i.e., the difference in crop production emissions from a scenario
with a given amount of biofuel relative to a scenario absent that biofuel). ADAGE, GCAM and
GTAP represent a further departure from the GREET approach as they include market-mediated
GHG impacts from yet more economic sectors. A notable example is the inclusion of GHG
emissions from transportation fuel market effects in ADAGE, GCAM and GTAP. When these
models are shocked to consume more biofuels in a particular region, they estimate the effects of
the shock on transportation fuel prices and consumption, both in the region where the shock
occurs and all other global regions. Instead of assuming that biofuels displace gasoline or diesel
on an energy-equivalent basis, these models estimate the global market-mediated changes in
gasoline and diesel consumption associated with the biofuel shock and report the resulting GHG
emissions changes.
3.1.5 Land Representation
Categorization or binning of land into types is an important, but often overlooked,
consideration for land use change modeling. The ways in which land is categorized and the
assumptions regarding how much of it is available or unavailable for commercial use vary
widely across modeling frameworks. The GREET model does not explicitly represent land. But
it is able to add induced land use change emissions through the CCLUB module, which uses
GTAP. The other four models estimate interactions between cropland, pasture, forestry, and, in
some of these models, other land types as well. For example, GLOBIOM, ADAGE and GCAM
140 The non-C02 emissions category includes "other CO2", i.e., CO2 emissions from activities other than fossil fuel
combustion, see Chepeliev (2020). These include CH4, N20, and fluorinated gases (CF4, HFC134a, HFC23, SF6).
141 Land use change GHG emissions are calculated based on land category area changes from GTAP and emissions
factors from the AEZ-EF model.
39
-------
also model the expansion of commercial cropland, pasture and forestry activities into grassland
and forests that are not otherwise used for commercial production. By default, GLOBIOM and
GCAM both place various exogenous limits on conversion of certain lands, to broadly represent
land protection policies and regimes (e.g., protection of ecologically sensitive lands), though
these assumptions may be modified. In contrast, as discussed in Section 2.4, while the GTAP
databases represent managed and unmanaged lands, the GTAP-BIO model only allows managed
lands to be used for productive uses, excluding the possibility for "unmanaged" land, such as
rainforests or native grasslands, to be brought into agricultural or silvicultural production. As
shown in Figure 5.2-1, this assumption applies to a relatively large share of arable land and
means that GTAP employs a much different representation of commercially available land than
the other models. Additionally, the share of non-commercial land assumed to be protected or
unavailable for commercial use is also an important assumption across models. For example, to
the extent modeling assumes that policies will be implemented and enforced to protect natural
forests with high carbon stocks, this will likely reduce the land use change GHG estimates by a
significant amount compared to a scenario which assumes laxer enforcement of land
protections.142 Other differences in land representation, such as the representation of unused
cropland and the treatment of multicropping, could also impact model results, and are discussed
further in Sections 5.2 and 6.5, respectively. For land categories that are given the same name in
different models (e.g., cropland, pasture), the underlying definitions and data may be different -
investigating and potentially aligning these definitions and categorizations is a potential area for
further research.
3.1.6 Trade
A significant source of theoretical and practical variation across the models considered in
this comparison is their approach to representing commodity trade. ADAGE and GTAP
represent trade bilaterally using an Armington approach (i.e., assuming imperfect substitution
between the same product produced in different countries), however the degree of substitution
varies across traded items. GLOBIOM models trade bilaterally based on the spatial equilibrium
approach and assumes commodities to be homogenous and traded based on least expensive
production costs, though transportation costs and tariffs are also included. GCAM represents
trade in agricultural, livestock, forestry, and renewable fuel commodities through an Armington-
like approach and trade in all other commodities, including most energy commodities, through
homogenous global markets.143 These methods have areas of overlap and similarity but lead to
distinct structures of trade. These differences in structure have significance to the present model
comparison exercise for multiple reasons. The ability of these models to deviate from the
historical trade patterns to which they are calibrated varies. The willingness of simulated
economic actors to substitute imported goods for domestically produced goods, and vice versa,
also varies by model.
142 Mignone, B. K., Huster, J. E., Torkamani, S., O'Rourke, P., & Wise, M. (2022). Changes in Global Land Use and
CO2 Emissions from US Bioethanol Production: What Drives Differences in Estimates between Corn and Cellulosic
Ethanol?. Climate Change Economics, 13(04), 2250008.; Plevin, R. J., et al. (2022). "Choices in land representation
materially affect modeled biofuel carbon intensity estimates." Journal of Cleaner Production: 131477. Figure S9.
143 Note that the most recent public version of GCAM trades all energy goods through the Armington-like approach,
rather than through homogenous markets. This version of the model was not released in time for inclusion in this
exercise.
40
-------
3.2 Input Parameters and Data
In addition to the key model characteristics discussed above, it is also important to
consider differences in data and parameter inputs used within models for biofuel GHG analysis.
There have been very few published efforts to compare assumptions across these models or to
evaluate which parameters are highly influential on model results. However, the previous work
which has been done has suggested the parameter assumptions which are among the most
influential in biofuel GHG analysis are related to:
• Crop yields
• Crop intensification
• Land competition and land transitions
• Carbon stocks of different land types
• Trade
• Peatland emissions
• Substitutability in food and feed markets
In this section, we review this previously published literature related to data and
parameter inputs. We explore parameter sensitivity further through modeled scenarios in Section
9.
Assumptions related to crop yields and crop intensification are important for biofuel
GHG modeling. Global crop yield data is readily available from FAO; however, this data is
generally available at a country level and it is also crop-specific. Many models require data
inputs for subnational physical regions and must also aggregate many of the dozens of FAO-
reported crops into groups for computational tractability. Modelers must determine for
themselves how to downscale or aggregate data as needed. There may be differences in how the
models map this historical data to the crop categories and physical regions they represent.
Assumptions about how crop yields may change in the future are also influential and inherently
uncertain. Perhaps even more important for biofuel modeling are assumptions about how crop
yields may change in response to price changes. Plevin et al. (2015) performed a sensitivity
analysis of biophysical and economic inputs to the GTAP+AEZ-EF modeling framework, and
found the elasticity of crop yield with respect to price (YDEL) to be "by far" the most influential
parameter in terms of its effect on the estimated ILUC emissions associated with corn ethanol,
sugarcane ethanol and soybean oil biodiesel.144 In the GTAP model used in this model
comparison, the YDEL parameter may have less influence on the results, as it now accounts for
the ability of increased harvest frequency and use of "unused cropland" to increase crop
production without extensification..145 However, a sensitivity analysis with GCAM did not
identify crop yield assumptions to be among the most influential parameters determining corn
ethanol land use change GHG emissions.146 This suggests that input parameters that are highly
144 Plevin, R. J., et al. (2015). "Carbon Accounting and Economic Model Uncertainty of Emissions from Biofuels-
Induced Land Use Change." Environmental Science & Technology 49(5): 2656-2664.
145 Taheripour, F., et al. (2017). "The impact of considering land intensification and updated data onbiofuels land
use change and emissions estimates." Biotechnology forBiofuels 10(1): 191
146 Plevin, R. J., et al. (2022). "Choices in land representation materially affect modeled biofuel carbon intensity
estimates." Journal of Cleaner Production: 131477. Figure 7.
41
-------
influential in one model might not highly influential in another model due to structural
differences between frameworks.
The parameters which control land competition and land transitions within models are
also important. These parameters control the amount of substitution between land types that
occurs based on changes in commodity prices and land rental rates. A sensitivity analysis of
GCAM found the parameter controlling ease of transition between cropland, forest, and
grassland to be an influential parameter. A sensitivity analysis of GTAP also found that the
assumed elasticity of transformation between managed forest, cropland, and pasture is influential
for corn ethanol LUC GHG estimates.147
Sensitivity analysis using GCAM found other assumptions to be influential when
estimating corn ethanol land use change GHG emissions, including the soil carbon density of
cropland, ease of transition between crop types, the soil carbon density of grassland, and the soil
carbon density of other arable land.148 Other influential assumptions identified through
sensitivity analysis with GTAP include the relative productivity of newly converted cropland,
trade elasticities (i.e., ease of substitution among products imported from other countries) and
emissions from conversion of cropland pasture.149
Sensitivity analyses have shown that other influential assumptions within GTAP include,
but are not limited to, tropical peat soil oxidation and the share of palm oil expansion on peatland
for vegetable oil based biofuel modeling, and the share of vegetable oil biofuel feedstock that is
supplied through expanded vegetable oil production versus reduced demand and substitutions
with other products.150
Another influential assumption in biofuel GHG modeling is the choice of data sets for
soil carbon and biomass carbon stocks, and how these data are mapped to land categories and
regions to determine the GHG emissions from converting an acre of land from one use to
another. The soil and biomass carbon data sources used in each model are discussed in the model
descriptions above. Soil carbon data and analysis are active areas of research, and higher
resolution datasets have recently been produced using statistical methods and remote sensing
data.151 For example, the SoilGrids250m version 2.0 dataset provides soil carbon estimates for
the globe with quantified spatial uncertainty,152 and Spawn et al. (2020) developed global maps
147 Plevin, R. J., et al. (2015). "Carbon Accounting and Economic Model Uncertainty of Emissions from Biofuels-
Induced Land Use Change." Environmental Science & Technology 49(5): 2656-2664. Table S9 in the Supplemental
Information.
148 Plevin, R. J., et al. (2022). "Choices in land representation materially affect modeled biofuel carbon intensity
estimates." Journal of Cleaner Production: 131477. Figure 7.
149 Plevin, R. J., et al. (2015). "Carbon Accounting and Economic Model Uncertainty of Emissions from Biofuels-
Induced Land Use Change." Environmental Science & Technology 49(5): 2656-2664. Table S9 in the Supplemental
Information.
150ICAO (2021). CORSIA Eligible Fuels ~ Lifecycle Assessment Methodology. CORSIA Supporting Document.
Version 3: 155. Section 6.2
151 Spawn-Lee, Seth. (2022). "Carbon: Where is it and how can we know?" Presentation for EPA Biofuel GHG
Modeling Workshop. February 28, 2022. EPA-HQ-OAR-2021-0921-0022
152 Poggio, L., de Sousa, L. M., Batjes, N. H., Heuvelink, G. B. M., Kempen, B., Ribeiro, E., and Rossiter, D.:
SoilGrids 2.0: producing soil information for the globe with quantified spatial uncertainty, SOIL, 7, 217-240, 2021.
42
-------
of above and below ground biomass carbon density in the year 2010.153 With few exceptions,154
these newer data sets have not yet been incorporated into published estimates of biofuel land use
change.
Model Comparison Core Scenarios
4 Description of Core Modeled Scenarios
To compare the five models described above, we ran two scenarios through each
framework: 1) a reference case, 2) a corn ethanol scenario (also referred to as the "corn ethanol
shock"), and 3) a soybean oil biodiesel scenario (also referred to as the "soybean oil biodiesel
shock"). All of these scenarios are hypothetical and designed solely for the purpose of evaluating
and comparing the models. The modeled scenarios do not represent our forecast of what is likely
to occur in the future, nor should they be interpreted as reflecting EPA's expectations about
future biofuel policy decisions.
For the three dynamic models (ADAGE, GLOBIOM, and GCAM), we defined a
hypothetical reference case for modeling purposes with U.S. biofuel consumption volumes for
each modeled fuel set to constant values from 2020-2050, based on the 2016-2019 average from
EPA-Moderated Transaction System (EMTS) data (Table 4-1). We used the EMTS sum of
biodiesel and renewable diesel for the biodiesel baseline. For GTAP, the reference case is the
global economy as represented in the 2014 GTAP database.
The core GREET model, excluding the ILUC module, does not include an explicit
reference case for corn ethanol or soybean oil biodiesel. As discussed above, GREET does not
model GHG impacts resulting from a change in biofuel production relative to a reference case.
Instead, it estimates the GHG emissions associated with, or attributable to, each biofuel supply
chain. Although it does not include scenarios, GREET considers background and foreground
data. The foreground data represents the processes in the supply chain evaluated (e.g., corn
farming, ethanol production). The background data represents processes that are outside of the
supply chain, but that provide energy and material inputs to the supply chain (e.g., electricity
grid, natural gas supply chain, fertilizer supply chain). While GREET is a static time step model,
it provides default assumptions and estimates for individual years out to 2050. For the purposes
of this model comparison, we use GREET with the analysis year set to 2030.155
153 Spawn, S. A., et al. (2020). "Harmonized global maps of above and belowground biomass carbon density in the
year 2010." Scientific Data 7(1): 112.
154 Lark, T. J., et al. (2022). "Environmental outcomes of the US Renewable Fuel Standard." Proceedings of the
National Academy of Sciences 119(9): e2101084119.
155 Argonne National Lab updates GREET on an annual basis with modifications that impact results across many of
the pathways. Results in this section are from GREET-2022.
43
-------
Table 4-1: U.S. annual biofuel consumption volumes in the model reference case, for 2020-
2050156
Billion Gallons
Quad BTU
Ethanol from Corn
14.82
1.126
Biodiesel from Soybean Oil
1.19
0.14
Biodiesel from Canola/Rapeseed
Oil
0.26
0.03
Biodiesel from Palm Oil
0.09
0.01
Ethanol from Sugarcane
0.1
0.007
In addition to the reference case, we ran a corn ethanol scenario and a soybean oil
biodiesel scenario. The corn ethanol scenario is a consumption shock with an additional one
billion gallons (0.076 QBTU) of U.S. corn ethanol consumption in each year, with all other U.S.
biofuel consumption volumes set by assumption at the reference case levels. The soybean oil
biodiesel scenario is a consumption shock with an additional one billion gallons (0.118 QBTU)
of U.S. soybean oil biodiesel consumption in each year, with all other U.S. biofuel consumption
volumes set by assumption at the reference case levels. We selected the one billion gallon shock
size as a simple and reasonably sized shock that is large enough for the purposes of testing these
models. For the large economic models considered in our model comparison, it is necessary to
specify a change that is large enough to produce a tangible change in the model. We also did not
want to specify a shock that would be unreasonably large given current biofuel production levels.
As discussed above, these scenarios are hypothetical and designed solely for research purposes.
For the dynamic models (ADAGE, GCAM, GLOBIOM), the shocks increase linearly
from 2020 to 2030, such that that there is a 0.5 BG shock in 2025, and the full 1 BG shock is
reached in 2030. In these models, volumes are held at the 2030 value for 2030 to 2050 (Table 4-
2). The results from this exercise may be sensitive to the shape of the implemented shock of
time. We designed the scenarios with this ramp up to 2030 for a few reasons. First, these models
are primarily designed for evaluating future scenarios. While it is possible to set up these models
for retrospective analysis to simulate historical years ("hindcasting"), we did not have the time or
resources to complete such an analysis as part of this model comparison exercise. Second, we
designed the scenario with a linear ramp up to 2030 as that is the first future time period
represented in GLOBIOM.
For GTAP, these U.S. biofuel consumption volumes were added to the 2014 base year.
Because GTAP is a comparative static model, there is no ramp up period for the biofuel
consumption shocks in the modeled results for this framework.
156 To convert between gallons and Quad BTU, we used a lower heating value for ethanol of 0.076 Quad
BTU/Billion gallon, and a lower heating value for biodiesel of 0.118 Quad BTU/Billion gallon. For GTAP, the
reference case is 2014, which includes the following U.S. biofuel volumes: 14.29 billion gallons (1.09 Quad BTU)
of corn ethanol, 0.20 billion gallons (0.01 Quad BTU) of other ethanol, 0.68 billion gallons (0.08 Quad BTU) of
soybean oil biodiesel, and 0.61 billion gallons (0.07 Quad BTU) of other biodiesel.
44
-------
Table 4-2: U.S. corn ethanol and soybean oil biodiesel consumption volumes, in Quad BTU,
for ADAGE, GCAM, and GL
OBIOM
2020
2025
2030
2035
2040
2045
2050
Reference Case
Ethanol from Corn
1.126
1.126
1.126
1.126
1.126
1.126
1.126
Biodiesel from Soybean Oil
0.140
0.140
0.140
0.140
0.140
0.140
0.140
1 BG Soybean Oil Biodiesel <
2ase
Ethanol from Corn
1.126
1.126
1.126
1.126
1.126
1.126
1.126
Biodiesel from Soybean Oil
0.140
0.199
0.258
0.258
0.258
0.258
0.258
1 BG Corn Ethanol Case
Ethanol from Corn
1.126
1.164
1.202
1.202
1.202
1.202
1.202
Biodiesel from Soybean Oil
0.140
0.140
0.140
0.140
0.140
0.140
0.140
For these scenarios, we aligned the conversion factors for vegetable oil to biodiesel and
corn to ethanol across ADAGE, GCAM, and GLOBIOM (Table 4-3). These factors were aligned
to represent a standard dry mill process for production of corn ethanol, assuming natural gas use
to dry 100 percent of the DDG coproduct produced, and a transesterification process for
production of soybean oil biodiesel. The 2015 conversion factors are based on data received
from petitions under the RFS. For corn ethanol, the yield increase over time assumes that the
corn ethanol yield will approach the theoretical maximum efficiency of corn conversion to
ethanol by 2050, based on the assumed quantity of convertible material in a given quantity of
corn. Compared to our assumed 2020 yield, this is approximately a 10 percent increase in
ethanol yield per unit of corn feedstock. For soybean oil biodiesel, the yield increase over time
assumes that current state-of-the-art technology will become the nationwide industry average by
2050. Compared to our assumed 2020 yield, this is approximately a 5 percent increase in
biodiesel yield per unit of soybean oil feedstock. By default, the GTAP model uses conversion
assumptions based on historical data from 2014. While it is possible to adjust the conversion
yield in GTAP, we did not do so for his exercise in order to maintain the consistency of the 2014
database. In GTAP, the conversion factor for corn to ethanol is 2.8 gal/bushel, and the
conversion factor of soybean oil to biodiesel is 0.132 gal/lb oil. For the corn ethanol shock,
GTAP models a natural gas-fired dry mill corn ethanol process with dry DGS coproduct and no
corn oil coproduct. For the biodiesel shock, GTAP models a standard natural gas-fired
transesterification biodiesel production process. The GREET analysis relies on the assumptions
in GREET for 2030, which are a conversion factor for corn to ethanol of 2.92 gal/bushel, and a
conversion factor for soybean oil to biodiesel of 0.136 gal/lb oil. For 2030, GREET assumes by
default that 99.6 percent of the energy use in dry mill ethanol production will be from natural
gas, with the remainder from coal.
45
-------
Table 4-3: Conversion factors for vegetable oil to biodiesel and corn to ethanol, for
ADAGE,
GCAM, and GLOE
»IOM
Soybean oil
Corn conversion
conversion to
to ethanol
biodiesel
gal/bushel
gal/lb oil
2015
2.75
0.130
2020
2.78
0.132
2025
2.80
0.133
2030
2.85
0.134
2035
2.91
0.135
2040
2.96
0.135
2045
3.02
0.136
2050
3.06
0.136
Corn ethanol production creates DDG and corn oil coproducts. Table 4-4 shows the
assumptions in the models related to these coproducts. We did not align these assumptions across
the models. However, ADAGE, GCAM, and GLOBIOM already had similar DDG and corn oil
production assumptions. In GREET, less DDG and more corn oil is produced than in the other
models. In GTAP, more DDG is produced, and corn oil is not represented. ADAGE, GCAM, and
GLOBIOM all produce less DDG coproduct over time as corn ethanol production becomes more
efficient (i.e., more gallons per bushel) and a greater share of the initial feedstock mass is
converted to fuel. Soybean oil biodiesel production creates a glycerin coproduct. ADAGE,
GCAM, GLOBIOM and GTAP do not explicitly model this coproduct, while GREET does
explicitly model the glycerin coproduct.157
Table 4-4: Coproduct assumpt
ions for corn ethanol
DDG (lb/gal ethanol)
Corn oil (lb/gal ethanol)
ADAGE (2020)
5.9
0.2
ADAGE (2050)
5.1
0.2
GCAM (2020)
5.9
0.2
GCAM (2050)
5.1
0.2
GLOBIOM (2020)
5.9
0.2
GLOBIOM (2050)
5.1
0.2
GREET (2030)
4.2
0.4
GTAP (2014)
6.1
—
Note: Model year shown in parentheses.
A key assumption in soybean oil biodiesel production is the shares of soybean oil and
soybean meal produced per unit of soybeans crushed. Table 4-5 shows the soybean crush yield
share assumptions for each model. ADAGE, GCAM, and GLOBIOM all assume that 0.19 tons
of soybean oil are produced per ton of soybean crushed. These values are not assumed to change
over time in these models, and the assumptions are uniform across model regions. GREET and
157 In GREET, roughly 0.1 lb of glycerin is produced per pound of soy oil input.
46
-------
GTAP assume higher oil yields and lower meal yields relative to ADAGE, GCAM, and
GLOBIOM. In GTAP the amount of soybean oil produced from crushing varies by region.
Table 4-5: Production assumpi
tions for soybean oil biodiesel
Soybean oil (tons oil/tons
soybean)
Soybean meal (tons oil/tons
soybean)
ADAGE (2020)
0.19
0.8
ADAGE (2050)
0.19
0.8
GCAM (2020)
0.19
0.8
GCAM (2050)
0.19
0.8
GLOBIOM (2020)
0.19
0.8
GLOBIOM (2050)
0.19
0.8
GREET (2030)
0.22
0.78
GTAP (2014)158
0.2
0.8
Note: Model year shown in parentheses.
5 Comparison of Reference Case Estimates
In this section we compare the estimates and assumptions from the reference case. We
look, in turn, at the following elements from the reference case:
• Crop production
• Land use impacts
• Crop yields
• Energy consumption
• GHG emissions
The majority of these comparisons include ADAGE, GCAM, GLOBIOM, and GTAP.
The comparison of energy consumption does not include GLOBIOM as this model does not
endogenously consider energy markets. Only the comparisons of crop yield and GHG emissions
includes GREET. GREET is a supply chain LCA model that does not represent changes in
agricultural and economic markets between reference and modeled scenarios, as the other
models in this comparison exercise are designed to estimate.
5.1 Crop Production
ADAGE, GCAM, GLOBIOM, and GTAP each include different crops, which we
aggregated into common categories for reporting purposes to better enable comparison across the
models. Table 5.1-1 shows the crops included in each model, and how they are reported here. Of
the models, GLOBIOM includes the most disaggregated set of modeled crop categories. In
158 Values are approximate for the USA region. GTAP crushing rates are based on the mean data provided by the
World Oil data set. This data set shows the crushing rate for soybeans varies across countries, and is generally 18-
20 percent, with some rare cases of 17 percent (in Bangladesh and Thailand) and 21 percent (in Japan). The World
Oil data shows a crushing rate of 19.75 percent for the U.S. in 2014, which is implemented in the GTAP database
construction.
47
-------
ADAGE, palm fruit and rapeseed are not explicitly represented, but are included under "rest of
oilseeds."
Table 5.1-1: Crops represented in A
DAGE, GCAM, G1
LOBIOM, and GTAP
Model
Comparison
Category
ADAGE
GCAM
GLOBIOM
GTAP
Corn
Corn
Corn
Corn
Corn
Soybean
Soybean
Soybean
Soybean
Soybean
Wheat
Wheat
Wheat
Wheat, Durum
wheat*, Soft
wheat*
Wheat
Rice
Not explicitly
represented;
aggregated with
"other grains"
Rice
Rice
Paddy rice
Sugar crops
Sugarcane,
Sugar beet
Sugar crops
Sugar cane,
Sugar beet*
Sugar crops
Palm fruit
Not explicitly
represented;
aggregated with
"rest of oilseeds"
Oil palm and
coconuts
Palm fruit
Palm fruit
Rapeseed
Not explicitly
represented;
aggregated with
"rest of oilseeds"
Rapeseed
Rapeseed
Rapeseed
Other oil crops
Rest of oilseeds
Oil crops
Groundnut,
Sunflower
Other oil seeds
Other grains
Rest of cereal
grains
Other grain
Barley, Millet,
Sorghum
Other grain
Energy crops
None159
Herbaceous
biomass crop;
woody biomass
crop
Other crops
Rest of crops
Root/tuber;
Fiber crop;
Fodder herb,
Fodder grass,
Miscellaneous
crops
Cassava,
Chickpeas, Dry
beans, Potatoes,
Sweet potatoes,
Cotton, Peas*,
Rye*, Oat*,
Flax*
Other crops
*EU region only
159 ADAGE has the ability to model switchgrass and miscanthus, but production of those crops were not included in
these scenarios.
48
-------
Figure 5.1-1 shows the reference case crop production in 2014 (GTAP) and 2020 and
2050 (ADAGE, GCAM, and GLOBIOM). Total crop production in 2020 in the USA region is
highest in the ADAGE results and lowest in the GLOBIOM results. In the non-USA regions,
GCAM results have the highest 2020 crop production, and GLOBIOM results have the lowest
production. In 2050, the total production is again the highest in ADAGE results in the USA
region, and the highest in GCAM results in the non-USA region. The total crop production in the
USA region has a similar percent increase between 2020 and 2050 in the ADAGE and GCAM
results (30 percent and 27 percent, respectively). However, the ADAGE and GCAM results
differ in the growth rate of the production of individual crops. GLOBIOM results have a lower
percent increase in crop production (13 percent). In the non-USA regions, GCAM and
GLOBIOM results have a similar percent increase in total crop production (47 percent and 50
percent, respectively), whereas ADAGE results have a lower percent increase in total crop
production (21 percent).
Figure 5.1-1: Crop production (million metric tons) in the reference case160'161
Baseline Crop Production Commodity
'to o
"> L±J <> *>
H S 5 ° S 5 °
w « O § « O
LD (J
¦ Other Crops
¦ Corn
Other Grains
¦ Other Oil Crops
¦ Energy Crops
¦ Palm Fruit
¦ Rapeseed
Rice
¦ Soybean
¦ Sugar Crops
Wheat
Table 5.1-2 compares these modeled values with crop production data from FAOSTAT.
GTAP's crop production, which is calibrated to 2014 data, aligns closely with the FAOSTAT
2014 production data for corn and soybeans. 2020 crop production in ADAGE, GCAM and
GLOBIOM differs from the 2020 FAO values, for a few reasons. First, these models project
2020 production from a 2010, 2015, and 2000 model base year respectively. Long run economic
modeling projections do not, as a general methodological practice, attempt to build in exogenous
representation of short term historical economic shocks in modeled periods (i.e., times steps after
1611 Note that the USA and non-USA regions are shown on different scales to better show differences across the
models.
161 Reference case production values in the "Other Crops" category are mostly incomparable between models
because the models differ in which crops are represented in this category (see Table 5.1-1).
49
-------
the model base year), and these models should be expected to endogenously predict such shocks.
This alone leads to some variation in modeled estimates from the historical record for years like
2020, where a significant economic shock occurred in the form of the COVID-19 pandemic.
Second, as described in Section 3.1.2, the 2020 time step in ADAGE and GCAM represents a
slightly different time period than the 2020 time step in GLOBIOM. The ADAGE, GCAM, and
GLOBIOM crop production in 2020 generally falls within the range of production over the years
2015-2021, with a few exceptions. The ADAGE corn production results are higher than the FAO
range in the USA region, but lower than the FAO range in the non-USA regions. ADAGE and
GCAM soybean production results are both lower than the FAO range in the non-USA regions.
Table 5.1-2: Corn and soybean production (million metric tons) from reference case and
FAOSTAT data162
Data source
Corn, USA
Soybean, USA
Corn, Non-USA
Soybean, Non-
Region
Region
Region
USA Region
GTAP, 2014
361
107
678
199
FAOSTAT,
361
107
680
199
2014
ADAGE, 2020
462
114
622
199
GCAM, 2020
376
111
733
204
GLOBIOM,
368
99
742
219
2020
FAOSTAT,
358
115
805
240
2020
FAOSTAT,
345-412
97-121
708-826
216-251
2015-2021 range
5.2 Land Use
ADAGE, GCAM, GLOBIOM, and GTAP each include different land types, and different
assumptions about the reference area of each land type over time. For this exercise, for reporting
purposes we mapped land types to common categories across the models, as shown in Table 5.2-
1. Areas of land types in the "other non-arable land" category are held constant over time and
cannot convert to other land types.
162 FAOSTAT data from: https://www.fao.Org/faostat/en/#data. Non-USA values were calculated by subtracting the
United States production from the World production. FAOSTAT 2015-2021 range shows the highest and lowest
production from the years 2015 to 2021. These do not necessarily correspond to the 2015 and the 2021 values.
50
-------
Table 5.2-1: Land representation in ADAGE, GCAM
GLOBIOM, and <
HAP
Model
ADAGE
GCAM163
GLOBIOM
GTAP
Comparison
Category
Cropland
Cropland
Cropland
Cropland, short
rotation
plantation
Cropland*
Forest
Managed forest
Commercial
Managed forest
Forest164
(managed)
forest
Forest
Natural forest
Forest
Unmanaged
(unmanaged)
forest
Grassland
Natural grassland
Grassland
Grassland
Other arable
Not included
Other arable land
Other agricultural
Cropland
land
land, other
natural land
pasture*, "unused
land"*
Other non-
Other land:
Tundra,
Wetlands, "not
arable land
includes bare
Rock/ice/desert,
relevant" (e.g.
ground, wetlands,
Urban
ice, water bodies)
mangroves, salt
marsh, glaciers,
lakes
Pasture
Pasture
Intensively-
Pasture
Pasture165
(managed)
grazed pasture
Pasture
Not included
Other pasture
(unmanaged)
Shrubland
Not included
Shrubland
* GTAP results report an aggregated "Cropland" category which is meant to represent fallow cropland in addition to
actively cultivated cropland. For the scenario difference values, we are able to disaggregate those fallow land
categories - "cropland pasture" and "unused land" - and assign them to the "Other arable land" model comparison
category. For this model comparison exercise, GTAP assumes no change in U.S. Conservation Reserve Program
area due to the biofuel shocks.
Reference case land use for arable land is shown in Figure 5.2-1 for 2014 (GTAP) and
2020 and 2050 (ADAGE, GLOBIOM, and GCAM).166 The GTAP reference case land areas
differ most from the other models because GTAP does not include unmanaged land such as
unmanaged forest, grassland or shrubland.
163 In the version of GCAM used in this exercise, land types are further split by mineral soil and peat soil.
164 In the GTAP database the managed forest area is the sum of managed/commercial forest and "accessible" forest,
with accessibility determined based on an analysis of distance from roads.
165 In the GTAP database pasture area includes areas of grassland.
166 Land cover and land use changes in the model reference cases are based on the agricultural demand, differences
in land rent among land types, ease of substitution among land, and relative changes in land productivity.
51
-------
Figure 5.2-1: Arable land use (million metric hectares) in the reference case167'168
Reference Arable Land Cover
Model Year
ADAGE 2020
2050
< GCAM 2020
w 2050
GLOBIOM 2020
2050
GTAP 2014
0 100 200 300
Mha
0 100 200 300
Mha
0 100 200 300
Mha
0 100 200 300
Mha
0 100 200 300
Mha
Model
Year
OK IK 2K 3K 4K
Mha
OK IK 2K 3K 4K
Mha
OK IK 2K 3K 4K
Mha
OK IK 2K 3K 4K (
Mha
Land Cover Type
¦ Cropland
¦ Forest (managed)
I Forest (unmanaged)
¦ Grassland
¦ Other Arable Land
I Shrubland
I Pasture (managed)
¦ Pasture (unmanaged)
2K 3K 4K
Mha
For cropland, GLOBIOM shows lower area than other models in the non-USA regions.
For forest, ADAGE and GLOBIOM have similar area in the non-USA regions, and GCAM has
lower area. Because GTAP only represents managed forest, the total forest area is smaller than
the other models. But the managed forest area is larger than the other models. Grassland is
highest in ADAGE, followed by GCAM then GLOBIOM. For pasture, only GCAM
differentiates between managed and unmanaged pasture. GCAM has very little managed pasture
in the non-USA regions, but similar total pasture as GTAP. GTAP shows the largest area of
managed pasture, as it represents pasture and grassland jointly. ADAGE and GLOBIOM have
lower total pasture.
ADAGE, GCAM, and GLOBIOM all project an increase in cropland area and a decrease
in grassland area over time, both in the USA region and the non-USA regions. Each of these
models also shows a decrease in non-USA total forest area over time, with an increase in
managed forest and a decrease in unmanaged forest. In the USA region, GCAM and GLOBIOM
both show an increase in total forest area over time, with an increase in managed forest and a
decrease in unmanaged forest. In ADAGE, the USA region has a small decrease in managed
forest and increase in unmanaged forest, with an overall decrease in total forest area. For pasture,
ADAGE, GCAM, and GLOBIOM show different trends. In the non-USA regions, total pasture
decreases over time in ADAGE and GCAM, but increases in GLOBIOM. In the USA region,
total pasture increases over time in ADAGE, and decreases in GCAM and GLOBIOM. In
GCAM, managed pasture area increases over time, and unmanaged pasture area decreases over
time, in both the USA region and non-USA regions.
167 Note that the USA region and the non-USA region have different scales.
168 Cropland area in GTAP represents the sum of land cultivated for row crops, cropland pasture, and other unused
land that GTAP classifies as cropland. This differs from the "Cropland" category of land presented in Figure 6.6-2
and Figure 7.6-2 which illustrate changes in cropland compared to the reference case. In those figures, cropland
pasture and other unused cropland are assigned to the "Other Arable Land" category.
52
-------
The GLOBIOM and GCAM reference case results include reductions in "other arable"
land over time from 2020 to 2050. For GCAM, the other arable land category includes fallow,
unused, and unharvested cropland and also serves to represent differences in land area estimates
between FAO and other data sources. None of the models explicitly represent Conservation
Reserve Program (CRP) land in the USA as a unique land category. For agricultural land areas,
GLOBIOM and GCAM rely on FAO data, which does not explicitly list CRP. CRP may be
implicitly represented in the "other arable" category of GCAM and GLOBIOM, but without
explicitly accounting for the particular incentives offered to farmers by the program. ADAGE
does not include CRP and does not explicitly account for conservation management decisions.
The GTAP database includes data on CRP area, but the GTAP model included in our comparison
exercise assumes no change in CRP area due to the biofuel shocks, and this is the standard
assumption used in the GTAP model. Given that other studies focusing on the U.S. suggest that
biofuel consumption may have a significant effect on CRP area,169 this may be an area for future
research and model development.
5.3 Crop Yield
ADAGE, GCAM, and GLOBIOM use different exogenous assumptions about crop yield
growth over time. In GLOBIOM, exogenous yield improvements represent technological change
and multi-cropping. Crop yield growth is based on an extrapolation of historic yield trends from
FAO data. Exogenous assumptions on multi-cropping are based on a literature review and apply
to areas such as Brazil. In GCAM, exogenous yield growth is based on FAO data. In ADAGE,
land productivity by land type is from the linked EPPA-TEM model, and a 1 percent annual
growth in crop yield is assumed.
These models also have the ability to change crop yields endogenously, based on changes
in prices or other factors, as does the GTAP model. In ADAGE and GTAP, a nested CES
(constant elasticity of substitution) function governs the endogenous yield changes. Materials
(e.g., fertilizer) or energy (e.g., for farm equipment) can be substituted for land to increase the
yield. Additional capital or labor can also be invested to increase yields. GTAP imposes a
restriction on substitution among labor, land, and a mix of capital-energy in crop sectors to reach
a target for price-induced yield response. GCAM has four different technology options (rainfed
vs. irrigated; low-yield vs. high-yield), each with different yields. A logit function determines the
share of production in each of these technology options based on profit rates, and the prices of
fertilizer and irrigation water also affect the competition of these technologies. Yields within any
land use region, crop type, and irrigation level can increase or decrease by up to 20 percent based
on the profitability. GLOBIOM also has four management options with different intensity levels
(subsistence, low input, high input, irrigated high input). Crop production is represented at the
grid level, and GLOBIOM can reallocate production from one cell to another based on the
productivity and profitability.
Reference case corn and soy annual yields for these models are shown in Figure 5.3-1.
This figure also shows the 2014 yields in GTAP, and data and yield projections from USDA.
169 See for example, Chen, X., & Khanna, M. (2018). Effect of corn ethanol production on Conservation Reserve
Program acres in the US. Applied Energy, 225, 124-134.
53
-------
Models show a range in the crop yield and the yield growth rate. For corn, ADAGE and GCAM
have the highest yields in the USA region. For soybeans, GCAM has the highest yield and
ADAGE has the lowest yield in the USA region. USD A data and projections are generally within
the range of the modeled yields. In the USA region, the 2030 corn yield in GREET is 12.5 t/ha,
and the soybean yield is 3.7 t/ha. The non-USA region yield is weighted by crop production for
each individual region outside of the USA region. The corn and soybean yield in the non-USA
region is similar across models, although there is more variation in the soybean yields over time.
Figure 5.3-1: Corn and soybean yields (tons per hectare) in the reference case170
Corn Yields
Corn
15
Non-USA
11.0
10 L ""10 7
£ 9.6 9.6
11.0 10.9 10.9
Soybeans Yield
Model
¦
ADAGE
¦
GCAM
¦
GLOBIOM
¦
GTAP
15
¦
USDA - Data and Projections
10
5
0
2010 2020 2030 2040 2050 2010 2020 2030 2040 20501
Soybeans
USA
Non-USA
4
3.0 ^
3.6
4.0
4^.
.-—¦3.1
1.7 2'93£.
3.3
3.6
3.9
4
ID
3.2
3.3
3.6
3.9
3.4
4->
2
2.9 3 2
2.2-lT"2-6
2.6
2.8
3.0
2
0
0
2010
2020
2030
2040
2050
2010 2020
2030
2040
2050
5.4 Energy Consumption
Each model was given specifications for biofuel consumption in the USA region to stay
constant at specific levels in the reference case.171 However, constraints were not placed on
biofuel consumption in non-USA regions. Figure 5.4-1 shows the biofuel consumption in
ADAGE, GCAM, GLOBIOM, and GTAP. The models show very different reference case
amounts of biofuel consumption in the non-USA regions in 2020, and different projections over
1711 Yields reported from ADAGE, GLOBIOM, GTAP, and in the USDA data and projections are calculated as crop
production per harvested area (i.e., production per harvest). Yields reported from GCAM are calculated as crop
production per cultivated area (i.e., production from all harvests per cultivated area, where cultivated area is equal to
harvested area divided by harvest frequency).
171 ADAGE does not include rapeseed oil consumption in the USA region so that consumption volume is set at zero
instead of the specified amount.
54
-------
time through 2050. Since GLOBIOM does not endogenously represent energy markets, levels of
consumption of biofuels are set exogenously for all regions. For this exercise, consumption
levels of biofuels in the non-USA regions are held constant throughout the period of analysis.
GCAM shows similar total biofuel consumption in the non-USA region as GLOBIOM in 2020,
but the consumption more than doubles by 2050. ADAGE has much lower total biofuel
consumption in non-USA regions in 2020 than the other models, with almost no consumption of
soybean oil biodiesel.172 Biofuel consumption increases over time, with most of the increase in
ethanol from sugar crops. In GTAP, the 2014 non-USA biofuel consumption is higher than the
2020 consumption in ADAGE and lower than the 2020 consumption in GCAM and GLOBIOM.
There are also differences in the fuel categories, with most of the ethanol in GTAP coming from
an aggregated "other feedstocks" category rather than sugar crops, and most of the biodiesel
coming from "other oil crop oil."
Figure 5.4-1: Biofuel consumption (Quad BTU) in the reference case
4.00
33.00
I—
< CD
t/) "O
3 ^2.00
a
ADAGE
GCAM
GLOBIOM GTAP Commodity
Ethanol from Corn
¦ Ethanol from Sugar Crops
¦ Ethanol from Other Feedstocks
¦ Biodiesel from Soybean Oil
¦ Biodiesel from Rapeseed Oil
¦ Biodiesel from Palm Fruit Oil
Biodiesel from Other Oil Crop Oil
a)
i/>
as
CO
1.00
0.00
4.00
< i
U) 1
->3.00
. ~a
¦ a
o ^2.00
1.00
0.00
2020 2050
2020
2050
2020 2050
2014
ADAGE, GCAM, and GTAP show similar fossil fuel consumption in the reference case
(Figure 5.4-2).173 Consumption of natural gas, coal, and refined oil is slightly higher in the USA
region in 2020 in ADAGE than GCAM. In GTAP, the 2014 coal consumption in the USA is
higher than the 2020 consumption in ADAGE and GCAM, but the 2014 natural gas and refined
oil consumption is lower than the 2020 consumption in ADAGE and GCAM. In both ADAGE
172 ADAGE includes conventional vehicles and alternative fuel vehicles in its transportation sector. In this reference
run, ADAGE projects biofuel consumption in non-USA regions based on the relative competitiveness of
conventional and alternative fuel vehicles in the model over time. As electric vehicles become more competitive,
less biofuel is consumed. In the assumptions used by ADAGE in this run, soybean oil biodiesel is more costly to
produce than other biofuels in non-USA regions, so it is not consumed in these regions in the reference.
173 GLOBIOM does not model fossil energy consumption.
55
-------
and GCAM, natural gas consumption in the USA region increases over time, and coal
consumption decreases. In GCAM, refined oil consumption in the USA region decreases
between 2020 and 2050, whereas in ADAGE refined oil consumption increases. In the non-USA
regions in 2020, ADAGE has higher refined oil and natural gas consumption, but lower coal
consumption than GCAM. Both models show increases in consumption of these fossil fuels over
time in the non-USA regions, with ADAGE showing a larger increase. GTAP's 2014 non-USA
coal consumption is higher than the ADAGE and GCAM 2020 consumption, whereas the refined
oil consumption is lower. Natural gas consumption in 2014 in the non-USA region of GTAP is
slightly higher than GCAM's 2020 consumption. The differences between GTAP and other
models may reflect the difference in time periods represented. Differences across the models in
the reference case fossil fuel and biofuel consumption over time could impact the results of the
amount and type of fuel displaced in the biofuel volume shocks. Exploring the impact of these
differences could be an area for future research.
Figure 5.4-2: Fossil fuel consumption (Quad BTU) in the reference case
600
ADAGE
GCAM
GTAP
Commodity
¦ Coal
¦ Natural Gas
¦ Refined Oil
-------
Table 5.5-1: Greenhouse gases represented in each model
ADAGE
GCAM
GLOBIOM
GREET174
GTAP
CO2, CH4, HFC,
N2O, PFC, SFe
CO2, CH4,
HFC125,
HFC134a,
HFC 152a,
HFC227ea,
HFC23,
HFC236fa,
HFC32,
HFC365mfc,
N2O, PFC, SFe
CO2, CH4, N2O
CO2, CH4, N2O
CO2, CH4, N2O,
Fluorinated gases
(CF4, HFC134a,
HFC23, SFe)
Total GHG emissions in 2020 in the reference case are around 57 gigatons CO2
equivalents (GtCCheq) in ADAGE and 59 GtCCheq in GCAM. For comparison, the IPCC Sixth
Assessment Report estimates that global GHG emissions were 59±6.6 GtCCheq in 2019.175 In
both ADAGE and GCAM, CO2 is the largest contributor to the emissions, with methane the
second largest contributor. The GCAM reference case has higher non-CCh emissions in 2020
than ADAGE and GLOBIOM.
Figure 5.5-1 groups reference case emissions into a several broad categories. "Energy
from Fossil Fuels" includes all GHG emissions from fossil fuel combustion. Consequently, fossil
fuel emissions are not included in other categories. For example, emissions from diesel used to
drive tractors for crop production are included under "Energy from Fossil Fuels" rather than
"Crop Production." "Other (Industrial & Waste)" includes non-fossil fuel emissions from the
industrial and waste management sectors, such as CO2 from cement manufacturing and CH4
from landfills. "Livestock Production" includes emissions such as CH4 from enteric fermentation
and N2O and CH4 from manure. "LUC" includes emissions from biomass and soil carbon
associated with land use change. "Crop Production" includes emissions from crop inputs such as
N2O from fertilizer use and from crop production processes such as CH4 from rice production.
As shown in Figure 5.5-1, most emissions from ADAGE and GCAM come from CO2
from the energy from fossil fuels category. "Other (Industrial & Waste)" emissions are similar in
ADAGE and GCAM in 2020, but higher in GCAM than ADAGE by 2050. Emissions in this
category come from a mix of greenhouse gases. Emissions in this sector are not reported in
GLOBIOM. Emissions from livestock production are similar in ADAGE and GLOBIOM, and
higher in GCAM, and come primarily from methane. Land use change emissions are
significantly lower in ADAGE and GLOBIOM than GCAM. Crop production emissions are
similar in ADAGE and GCAM in 2020, but are 50 percent lower in GLOBIOM. Crop
production emissions increase over time in GCAM and GLOBIOM, but decrease over time in
ADAGE. GTAP reports land use change emissions by comparing land use areas between two
scenarios, but it does not track terrestrial carbon stocks or report total land use change emissions
174 GREET includes the ability to represent GWPs of short-lived climate forcers (volatile organic compounds,
carbon monoxide, NOx, and black carbon) but does not include them in results by default.
175 IPCC, 2023: Summary for Policymakers. In: Synthesis Report of the IPCC Sixth Assessment Report (AR6).
Available at: https://www.ipcc.cIi/report/ar6/svr/downloads/report/IPCC AR6 SYR SPM.pdf
57
-------
in each scenario. GTAP does also report several other categories of emissions, including
emissions from use of fossil fuels and total non-C02 emissions from sources other than land use
change. GREET is a supply chain LCA model that is designed to represent the emissions
emanating from the fuel supply chain rather than estimate the global economic impacts of a
change in biofuel consumption. GTAP and GREET are not included in Figure 5.5-1 because they
do not represent scenario-based emissions over time.
Figure 5.5-1: Global greenhouse gas emissions in ADAGE, GCAM, and GLOBIOM in the
reference case176
Model
ADAGE | GCAM | GLOBIOM |
C02
CH4
N20
Other GHGs
6K
0
Other (Industrial 0 4K
& Waste)
s 2K
0K_j
6K
0
Livestock 0
Production g ^
0K_j
4K
n 3K
LUC S 2K
5
IK
Emission Source
60 K
G>
Energy from o 40K
Fossil Fuels ^
5 20K
S 2K
o
Crop Production u
§ IK
OK
2030 2040 | 2030 2040 | 2030 2040
Year Year Year
176 Note that the rows of this figure use different scales. GTAP is not included in this figure because it does not
represent emissions over time, and due to time constraints, we do not have GTAP LUC emissions in the reference
case, or GHG emissions by gas for the source categories used in this figure. For comparison for GTAP, in the
reference case (2014), fossil fuel combustion and industrial CO2 emissions = 30,048 Mt, and other GHGs emissions
from all covered sources = 16,616 Mt CO:C. of which N;0 = 2,891 Mt CO2C. CH4 = 8742 Mt CO?c. fluorinated
gases = 986 Mt CO-c. and other CO2 = 3996 Mt CO?e. GREET is not included in this figure because it does not
include an explicit reference case, and therefore does not provide reference case emissions.
58
-------
5.6 Summary of Reference Case Estimates
The previous sections illustrate differences in the reference case in ADAGE, GCAM,
GLOBIOM, GTAP and GREET. Notable differences are observed across the models in crop
production, land use areas, biofuel and fossil fuel consumption in non-USA regions, and overall
emissions. These include differences in the reference case for 2020, as the models are initialized
with older data and define the 2020 time period in different ways.
Some of these differences could impact the results of the corn ethanol and soybean oil
biodiesel shocks from these models. For example, differences in the reference case crop yields
among models would cause differences in the amount of land needed to produce additional
crops. Differences in reference case biofuel and fossil fuel consumption among models could
affect energy sector responses the biofuel shocks. Potential future research could focus on how
the reference case influences the results of the biofuel shocks.
6 Comparison of Corn Ethanol Estimates
In this section, we present the results of the corn ethanol shock. The results in this section
show the difference between the corn ethanol shock and the reference case. We consider the
following elements in turn:
• Sources of corn ethanol to meet the shock
• Energy market impacts from the shock
• Crop production and consumption
• Trade impacts
• Yield changes
• Land use impacts
• Emissions: the modeled results of energy consumption, crop production, and land use
change described above come together in the modeled greenhouse gas emissions.
The majority of these comparisons include ADAGE, GCAM, GLOBIOM, and GTAP.
Only the comparison of GHG emissions includes GREET. GREET is a supply chain LCA model
that does not represent changes in agricultural and economic markets between reference and
modeled scenarios, as the other models in this comparison exercise are designed to estimate.
6.1 Sourcing Overview
The models included in this analysis have many options available for meeting the corn
ethanol consumption shock. For example, the USA region could produce additional corn ethanol,
import more corn ethanol, or export less corn ethanol. Additional imported corn ethanol supplies
could come from reduced consumption of corn ethanol in non-USA regions, or increased
production of corn ethanol. Increased domestic corn ethanol production could come from
diversion of corn from other uses, or increased production of corn, through yield increases or
increases in the area of corn cropland. This section will give an overview of the extent to which
the models rely on each of the available options for meeting the corn ethanol consumption shock.
59
-------
In the corn ethanol shock, most of the additional corn ethanol consumed in the USA
region comes from increased corn ethanol production in the USA region (Figure 6.1-1). In
ADAGE, GLOBIOM, and GTAP, the shock is met entirely by increased corn ethanol
production, with no change in gross imports or exports of corn ethanol in the USA region. In
GLOBIOM, because there is no energy sector, there cannot be a change in corn ethanol exports
or imports, so the shock must be met by corn ethanol production in the USA region.
In GCAM, up to 20 percent of the shock is met by changes in gross imports and exports
of corn ethanol, with the change in exports contributing to a larger percentage of the shock over
time. This change in exports is consistent with a reduction in the consumption of corn ethanol in
non-USA regions (blue bars, Figure 6.1-2).177 These GCAM results illustrate the potential
impact of dynamic energy sector modeling. Because some of the corn ethanol shock in GCAM is
met through changes in the energy sector in the non-USA regions, less new corn ethanol needs to
be produced, which reduces the impact on corn production and end uses.
Figure 6.1-1: Sources of additional corn ethanol consumed in the corn ethanol shock
relative to the reference case178
USA
ADAGE I GCAM I GLOBIOM IGTAP
o
CQ
u
O
-C
CO
c
o
0.08
0.07
0.06
£
o
u
30.05
i
o
i
CO
4—
~S0.04
~o
z>
£=
a
03
JZ
0.03
A->
LU
0.02
0.01
0.00
Source
¦ Gross Exports
¦ Gross Imports
¦ Production
ooooooooo
no <3- ld no ld no u">
o o o o o o o o o o
CM CM CM I CM CM CM I CM CM (M I CM
ADAGE, GCAM, GLOBIOM, and GTAP meet the corn ethanol shock through different
amounts of corn diversion from other uses, crop intensification, crop shifting to corn, and new
cropland (Figure 6.1-2). Based on the assumed conversion factor of corn to corn ethanol (Section
4), if all of the shock were met by new corn ethanol production, ADAGE, GCAM, and
GLOBIOM would need 8.9 million metric tons of additional corn for ethanol in 2030 and 8.3
million metric tons of additional corn for ethanol in 2050. GTAP would need 9.1 million metric
177 As shown in Figure 6.2-1, sugarcane ethanol is substituting for corn ethanol in non-USA regions of GCAM.
178 Red shows the contribution increased corn ethanol production in the USA region; orange shows the contribution
from increased corn ethanol gross imports to the USA region; blue shows the contribution from reduced corn
ethanol gross exports from the USA region.
60
-------
tons of additional corn for ethanol in 2014. The bottom panel of Figure 6.1-2 shows the sourcing
of corn for corn ethanol in units of million metric tons. In these results, GCAM needs less corn
feedstock than ADAGE, GLOBIOM, and GTAP because some of the shock is met by a decrease
in corn ethanol consumption in the non-USA region.
In these results, commodity diversion (reduced crop use for other purposes) accounts for
15-17 percent of the shock in ADAGE, 23-24 percent of the shock in GCAM, 26-40 percent of
the shock in GLOBIOM, and 57 percent of the shock in GTAP. These results are described more
in Section 6.3. Of the additional corn production, ADAGE, GCAM, GLOBIOM, and GTAP each
use a different mix of crop intensification (increased corn yields), shifting of cropland from other
crops to corn ("crop shifting" in Figure 6.1-2), and shifting land from other land types to
cropland ("new cropland" in Figure 6.1-2). In the GCAM results, most of the new corn comes
from new cropland. In the GLOBIOM and GTAP results, most of the new corn comes from
shifting of cropland from other crops to corn. In the ADAGE results, there is a transition over
time from more cropland shifting in 2030 to more new cropland in 2050. For GTAP, the primary
strategy for meeting the corn ethanol shock is commodity diversion, highlighted by a 1 percent
reduction in USA region feed consumption (DDG feed increases, corn feed decreases). However,
this reduction in total feed use has a much smaller impact (0.002 percent reduction) on USA
region meat and dairy production. Corn production and land use results are described in more
detail in Sections 6.3 and 6.6.
61
-------
Figure 6.1-2: Top panel: Percentage of the corn ethanol shock that is met by different
categories in 2030 and 2050. Bottom panel: Million metric tons of additional corn
production (red, orange, and yellow) and corn diverted to corn ethanol production from
other uses (green)179
ADAGE
GCAM
2030
2050
2030
2050
GLOBIOM 2030
GTAP
2014
0% 20% 40% 60% 80% 100%
Source Category
¦ Commodity Diversion
Crop Intensification
¦ Crop Shifting
¦ New Cropland
¦ Non-US Biofuel Consumption
ADAGE
Percent
GCAM GLOBIOM GTAP
£ 8
u
i £
+-> -M V)
U-l
-------
energy inputs required to grow, transport, and process additional feedstock. Correspondingly, a
reduction in the extraction and refining of petroleum would result in decreased demand for the
energy sources required in those processes. Finally, all of the above effects on demand for
energy sources will affect fuel prices, which, in turn, affect supply and demand for those fuels.
We refer to these adjustments in supply and demand to price as market-mediated effects.
Towards the end of this section, we present modeling results describing changes in liquid
fuel consumption relative to the size of the cumulative corn ethanol shock.180 These metrics
indicate whether one BTU of increased corn ethanol consumption in the USA region displaces
more or less than one BTU of refined oil181 or biofuel consumption, when averaged across all
years represented in the scenarios, and including the indirect effects discussed above. These
effects vary depending on whether they are considered within the USA region or non-USA
regions. As an illustration of the regional differentiation, we consider the expected effect of an
increase in corn ethanol consumption in the USA region on consumption of refined oil in the
non-USA regions. The primary theoretical mechanism for this effect is as follows: 1) biofuel
consumption increases in the USA region, displacing some quantity of refined oil consumption
in the USA region; 2) this reduces global demand for petroleum which puts downward pressure
on the price of crude and refined oil in non-USA regions; 3) the effect on crude and refined oil
prices leads to increasing demand for refined oil outside of the USA. The degree to which these
effects are reflected in the model results is presented in Figure 6.2-3 and the accompanying
discussion at the end of this section.
As discussed in Section 3, the models considered in this section differ in their
representations of energy markets. GREET is largely an attributional framework which includes
detailed accounting of the energy inputs for production of feedstocks, biofuels, and fossil fuels
but does not include a representation of markets for energy goods, the displacement effect of an
increase of biofuel use, nor of any other market mediated effects. GLOBIOM does not represent
energy commodities or markets, so it cannot be used to estimate the effects of a biofuel shock on
these markets. ADAGE, GCAM, and GTAP each represent a selection of energy commodities,
end use sectors, and market interactions.
180 I.e., the cumulative changes in energy consumption expressed as a percentage of the cumulative change in US
corn ethanol consumption over the duration of the modeled period.
181 In these models, refined oil is an aggregation of all refined petroleum products, including gasoline and diesel.
63
-------
Figure 6.2-1: Difference in consumption of energy commodities (quadrillion BTUs) in the
corn ethanol shock relative to the reference case in 2030 and 2050 (ADAGE, GCAM) and
2014 (GTAP)
ADAGE
Model / Region / Year
GCAM
<
fj)
z>
i—
CO
0.05
0.00
Commodity
GTAP | Ethanol (Corn)
1 Ethanol (Sugar Crops)
| Ethanol (Other)
| Biodiesel (All)
| Coal
Natural Gas
I Refined Oil
<
ZJ
-0.05
0.05
0.00
-0.05
2030 2050 2030 2050 2030 2050 2030 2050 2014 2014
ADAGE, GCAM, and GTAP results show differing estimated net impacts on biofuel
consumption and fossil fuel consumption under a one billion-gallon corn ethanol shock scenario
(Figure 6.2-1). As illustrated in Figure 6.1-1, a portion of the corn ethanol shock in GCAM is
met through decreased U.S. net exports of corn ethanol, the majority of which (95 percent in
2030) is a reduction in gross exports, as opposed to increased gross imports. This results in a
decrease in corn ethanol consumption in the non-USA regions (roughly ten percent when
compared to the total energy content of the corn ethanol shock in 2030) and an increase in
consumption of ethanol produced from sugar crops in non-USA regions (two percent of the
shock in 2030). While ADAGE and GTAP do represent trade in biofuel commodities (see Figure
6.2-2 below), the corn ethanol shock has little effect on trade of ethanol, and, consequently, little
effect on consumption of biofuels in non-USA regions, in the results from these models.
64
-------
Figure 6.2-2: Difference in U.S. net exports of energy commodities (quadrillion BTUs) in
the corn ethanol shock relative to the reference case in 2030 and 2050 (ADAGE, GCAM)
and 2014 (GTAP)
a
0.08
0.06
0.04
0.02
0.00
ADAGE
Model / Year
GCAM
GTAP
Commodity
¦
Crude Oil
¦
Refined Oil
¦
Coal
Natural Gas
¦
Biodiesel (Other)
¦
Biodiesel (Soy)
¦
Ethanol (Other)
¦
Ethanol(Corn)
-0.02
-0.04
2030 2050 | 2030 2050 | 2014
Results in all three models show increased consumption and decreased U.S. net exports
of natural gas, largely due to increased production of corn ethanol and drying of DDGs, though
the size of these impacts is notably smaller in GTAP results compared to in ADAGE and
GCAM. Impacts on natural gas use in the non-USA regions differ. GCAM results show
consistent and decreasing consumption of natural gas, corresponding with decreased demand for
natural gas used in ethanol production in non-USA regions and with other market mediated
effects. The lack of significant impacts on non-USA ethanol consumption in ADAGE and GTAP
results in a smaller effect on non-USA natural gas consumption in results from those models.
ADAGE, GCAM, and GTAP each model an aggregated refined oil commodity which
represents a range of petroleum products including gasoline, distillate fuel, and other industrial
chemicals and products. The primary displacement effect of increased corn ethanol consumption
is seen in the consumption of this modeled refined oil commodity. Within the USA region,
ADAGE, GCAM, and GTAP results show differing reductions in refined oil use; 0.068 and
0.079 quads in ADAGE and GCAM respectively in 2030, and 0.048 quads in GTAP in 2014.
The decrease in refined oil use in both ADAGE and GCAM is predominantly in the
transportation end use sector - this is the primary displacement effect - with some relatively
minor market mediated effects in other end use sectors. Results available from the GTAP model
did not disaggregate refined oil use by end use.
65
-------
The decrease in demand for crude and refined oil in the USA region observed in these
model results corresponds with a decrease in the price of these commodities. However, the
impact of the modeled shock on estimated prices of crude oil and refined oil is very small in
absolute terms because the one billion gallon shock represents only around one tenth of one
percent of global liquid fuel consumption. The result is a decrease in the estimated prices of
crude and refined oil by between one and three hundredths of one percent in the USA and non-
USA regions in ADAGE and GCAM results. Since crude and refined oil are globally traded, the
modeled price changes within and outside of the USA region are similar in direction and
magnitude. Outside of the USA region, all three model results show increased refined oil
consumption, largely driven by the downward price pressure on oil discussed above, though the
magnitude varies among models and model years.
Displacement and other net market impacts on refined oil consumption are often
presented in metrics normalized to the biofuel shock volume. This representation facilitates
comparisons of the effect across different studies and shock volumes. This indirect fuel use
effect is sometimes described in the literature as "oil rebound," though the scope of what is
included within the definition of "rebound" varies.
In the case of this model comparison exercise, we find it illustrative to consider the ratio
of cumulative net impacts on refined oil and other biofuels to the cumulative impacts on
consumption of corn ethanol in the USA region. These metrics indicate whether one BTU of
corn ethanol displaces more or less than one BTU of refined oil or other biofuel consumption,
when averaged across all years represented in the scenarios, and including the indirect effects
discussed above.
66
-------
Figure 6.2-3: Difference in liquid fuel consumption relative to the volume of the corn
ethanol shock182
Variable Group 1 / Model
Biofuel
100%
50%
0%
99%
99%
100%
92%
100%
100%
Refined Oil
Region
¦ USA
Non-USA
15%
23% 22%
-50%
-100%
-83%
-40%
¦61%
-107%
-68%
-83%
ADAGE GCAM GTAP ADAGE GCAM
GTAP
Figure 6.2-3 illustrates these cumulative relative effects within the USA region and non-
USA regions for both biofuels and refined oil. The left pane depicts the effect of the corn ethanol
shock on total biofuel consumption within the USA region (blue) and non-USA region (orange).
As discussed in Section 4, in the corn ethanol shock scenario, U.S. consumption of corn ethanol
is increased by one billion gallons, while U.S. consumption of all other biofuels is held constant
at reference case levels. Thus the cumulative difference in biofuel consumption in the USA
region between the corn ethanol scenario and the reference case is equivalent to the cumulative
size of the corn ethanol shock, which is the denominator of all of these relative metrics.
Therefore, by definition, the blue bar in the left pane is 100 percent, and represents the full
cumulative corn ethanol shock. Note that the scenarios in this model comparison exercise did not
place any additional constraints on consumption of biofuels in non-USA regions, so the
cumulative difference in consumption of biofuels in non-USA regions, depicted in orange on the
left pane of Figure 6.2-3, represents net impacts of the shock on consumption across all
represented biofuels. As discussed above, in the GCAM results for the corn ethanol scenario,
some of the required corn ethanol shock volume is met through adjustments in net trade of corn
ethanol. In the ADAGE and GTAP results for this scenario, the shock is met almost entirely
through increased corn ethanol production in the USA region. The cumulative effect of this
182 Values in the figure represent the difference between the shock and reference case of the given fuel category
(refined oil vs. liquid biofuels) and given region (USA region vs non-USA regions) divided by the difference in
consumption of liquid biofuels in the USA region (i.e., the shock volume). For ADAGE and GCAM, this is
calculated using cumulative volume differences between 2020 and 2050. For GTAP, which only estimates
differences in a single time step, the calculation uses only the volume differences in 2014.
67
-------
difference is seen in the orange bars; in GCAM, cumulative non-USA consumption of biofuels
decreases by eight percent of the cumulative USA corn ethanol shock volume, whereas ADAGE
and GTAP only show a one percent decrease in non-USA biofuel consumption. Thus, on net, the
shock scenario in GCAM increases global biofuel consumption by 92 percent of the total
specified cumulative shock, whereas the shock scenario in ADAGE and GTAP increases global
biofuel consumption by 99 percent of the total specified cumulative shock.
The righthand pane in Figure 6.2-3 illustrates the cumulative effects on refined oil
consumption within and outside the USA region. Under the corn ethanol shock scenario, that
additional volume is required to be consumed within the USA region, so the primary
displacement of refined oil used for transportation is within the USA region. If one BTU of
ethanol use displaced exactly one BTU of refined oil use in a given set of model results, and all
of the other indirect effects within the USA region discussed above were negligible, the blue bars
in this pane would show 100 percent. Thus, the size of the bar relative to 100 percent shows
whether the cumulative net impacts within the USA region are more or less than perfect energy
equivalent displacement.
As seen in the figure, there is greater than perfect displacement of refined oil in the USA
region in the GCAM results (107 percent). This displacement exceeds 100 percent primarily
because GCAM projects that the corn ethanol shock will increase the average price of fuel in the
USA region's gasoline pool. This causes a small decrease in USA region demand for gasoline in
addition to the energy equivalent displacement. In contrast, the ADAGE and GTAP results show
less than perfect displacement of refined oil in the USA region (83 percent and 61 percent,
respectively). In ADAGE, this difference is largely due to smaller reductions in refined oil
consumption in 2040 and 2050.
The effect on cumulative net non-USA oil consumption - a commonly used definition of
"oil rebound" in the literature - shows how global oil consumption changes as a result of the
shock. GCAM and GTAP results show larger increases in non-USA refined oil consumption (23
percent and 22 percent of the cumulative shock, respectively) than ADAGE (15 percent). The
global net effect of the shock on refined oil consumption is that, on average, 100 BTUs of corn
ethanol required to be consumed in the USA displaces 68 BTUs of global refined oil
consumption in ADAGE, 83 BTUs of global refined oil consumption in GCAM, and 40 BTUs of
global refined oil consumption in GTAP. That the estimated net effect of a U.S. biofuel shock on
global oil consumption amounts to less than one-for-one displacement makes intuitive sense; oil
and refined oil products are globally traded commodities. Any reduction in consumption of
refined oil in the USA makes available some additional supply to the rest of the world, which
would be expected to reduce the price of crude and refined oil globally and result in adjustments
to consumption patterns in all regions. We note, however, that the range of reductions in refined
oil use varies widely across the three models with energy sector representation, directly resulting
in the wide range of energy sector emissions savings estimated by these models. These emissions
results are presented in Section 6.7 below. Future research could better define and understand the
parameters and assumptions that lead to this range in reduction of refined oil consumption.
68
-------
6.3 Crop Production and Consumption
As shown in Section 6.1, ADAGE, GCAM, GLOBIOM, and GTAP results estimate
about 40-85 percent of the corn ethanol shock would be sourced from new corn production.
Estimated new corn production comes primarily from the USA region in these ADAGE, GCAM,
GLOBIOM, and GTAP results, with some new corn also produced in the non-USA regions in
the GCAM and GLOBIOM results (Figure 6.3-1). All four models estimate some reduction in
production of other crops in the USA region, though the magnitude varies.183 Soybean
production accounts for a large percentage of this decrease in all four models, but the
displacement of other crops is more variable across the results. GLOBIOM estimates the largest
decrease in non-corn USA crop production and GTAP the second largest, with GCAM and
ADAGE showing similar, more modest decreases.
Figure 6.3-1: Difference in commodity production (million metric tons) in the corn ethanol
shock relative to the reference case in 2014 (GTAP) and 2030 (ADAGE, GCAM,
GLOBIOM)
LD
CO
CO
c
o
u
ADAGE_
2030
GCAM
GLOBIOM
2014
GTAP
Commodity
¦ Corn
¦ Soybean
¦ Energy Crops
¦ Other Crops
¦ Other Grains
¦ Other Oil Crops
¦ Palm Fruit
¦ Rapeseed
¦ Rice
¦ Sugar Crops
¦ Wheat
-1
<
U1
ZD
<
LT)
ZD
<
LT)
ZD
<
lr
ZD
<
LT)
ZD
<
IT)
ZD
<
LT)
ZD
<
(/)
ZD
Results from three of the four models - GCAM, GLOBIOM, and GTAP - also estimate a
net increase in crop production in the non-USA region. These increases are multi-faceted, but
generally the crops with greater non-USA production are those for which U.S. net exports are
decreasing in the results for each respective model, i.e., some combination of corn, soybeans,
and/or wheat. One notable outlier to this general trend is the increase in sugar crop production in
GCAM. As shown in Section 6.2 and Figure 6.3-2, this additional sugar crop production is used
for fuel production in the non-USA regions of GCAM, which contributes to an increase in the
183 We also looked at forest product production for the models that are able to report it (ADAGE, GCAM,
GLOBIOM), and the change relative to the reference case is negligible.
69
-------
consumption of sugar crop ethanol. Conversely, in the ADAGE results, we observe a small net
decrease in crop production in the non-USA regions.
Globally, crop production increases in all four sets of model results. Most of the net
increase globally is from new corn production to produce additional corn ethanol. One exception
is the aforementioned increase in sugar crop production in GCAM; this is also occuring
indirectly to allow for greater consumption of corn ethanol in the USA region. We observe
substantial variation across the models regarding the magnitude of increased crop production,
and the share occurring within the USA region versus the non-USA regions. This is an area of
uncertainty across the models.
As explained in Section 6.1, in the ADAGE, GCAM, GLOBIOM, and GTAP results,
some of the corn ethanol shock is met by diversion of corn to fuel production from other end
uses. All four of these models show a reduction in the amount of corn used for feed, but there is
variation across the model results in how much the corn feed consumption is reduced (Figure
6.3-2). Part of the feed market impact may be attributable to the increase in corn prices which
follows from increased demand for corn in the shock case (changes in prices in the corn ethanol
shock case are discussed further below in Section 6.5). But it is also in part attributable to greater
production of corn DDG in the shock case.
DDG is a coproduct of corn ethanol production used almost exclusively for animal feed.
In these model results, the additional DDG produced from the additional corn ethanol production
is used for feed to replace the corn (that is, the DDG "backfills" for the corn diverted from feed
use to fuel use). Historically, USA-produced corn DDG is both consumed domestically and
exported. The degree to which future additional DDG production might be consumed
domestically versus exported is therefore a key uncertainty in forward-looking scenario analysis
for corn ethanol consumption. In the GLOBIOM results shown in Figure 6.3-2 below, the DDG
is consumed entirely within the USA region in 2030, displacing mostly corn in the feed market.
In ADAGE, GCAM, and GTAP, some of the additional DDG is consumed domestically and
some is exported for consumption in the non-USA regions (see also Figure 6.4-1). ADAGE
shows the largest share of exported DDG. Within the USA region, mostly corn is displaced in the
feed market. In non-USA regions, larger proportions of other crops are displaced, commensurate
with the dominant feed products in the affected regions. The results across all four models agree
however that, on a global basis, corn is the primary feed commodity displaced by additional
DDG. There is also good agreement across these four sets of results about the magnitude of
increased DDG production and consumption in response to the corn ethanol shock.
We observe from these results that there is more consistency among the models we
considered about the global magnitude of DDG consumption in response to a corn ethanol shock
than there is about where in the world that additional DDG consumption will occur. From this
we can conclude that exogenous assumptions about the location of DDG consumption carry
uncertainty. A possible area for further sensitivity analysis is to explore the potential impacts on
estimated GHG emissions should additional DDG be consumed primarily in the USA versus
primarily outside the USA.
70
-------
The ADAGE, GLOBIOM, and GTAP results estimate more additional corn for fuel
production than do the GCAM results. This is because, as discussed above, GCAM is meeting
some of the shock by reducing corn ethanol consumption in non-USA regions and reducing the
U.S. net exports of corn ethanol. To make up for the loss of corn ethanol in the GCAM results,
non-USA regions produce and consume some additional sugar crop-based ethanol. The question
of whether non-USA biofuel production and consumption would be measurably affected by
additional demand for corn ethanol in the USA therefore remains an uncertainty. However, it is
clear that such potential impacts on the energy sector may meaningfully affect the results; these
impacts cannot confidently be assumed to be zero.
The scenario results from ADAGE, GCAM, GLOBIOM, and GTAP consistently show
only minimal changes in the consumption of commodities for food, crushing, and other uses.
These results also consistently show only minimal changes in the consumption of commodities
and coproducts other than corn, DDG, and sugar crops.
Figure 6.3-2: Difference in consumption by end use (million metric tons) in the corn
ethanol shock relative to the reference case in 2014 (GTAP) and 2030 (ADAGE, GCAM,
GLOBIOM)184
Consumption by End Use
Fuel Production Feed Food Other Uses
Crushing
<
-------
impact of the corn ethanol shock on U.S. corn and other agricultural commodity exports to vary
by model. One would also expect the shares of domestic versus international consumption of the
DDG coproduct to vary by model, as imported DDG from the U.S. would be valued differently
based on how simulated economic actors are calibrated to value imported versus domestically
produced feed products.
Consistent with this expectation, we do observe ADAGE, GCAM, GLOBIOM, and
GTAP differ in their agricultural commodity trade responses to the corn ethanol shock. This is
illustrated by differences between the shock scenario and reference case in U.S. net exports of
crops and secondary agricultural commodities (see Figure 6.4-1). Results from all four models
show relatively minor changes in gross imports relative to gross exports, so the data displayed in
Figure 6.4-1 are roughly equivalent to differences in gross exports from the USA region. In
general, these reductions appear largely commensurate with the declines in crop production from
the USA region discussed in Section 6.3 above.
Figure 6.4-1: Difference in U.S. net exports of crops and secondary agricultural products
(million metric tons) in the corn ethanol shock relative to the reference case in 2030 and
2050 (ADAGE, GCAM, GLOBIOM) and 2014 (GTAP)
ADAGE
Model / Year
GCAM
GLOBIOM
GTAP
1.0
0.5
0.0
Commodity
| Other Crops
| Wheat
Vegetable Oil (Other)
| Meal (Other)
| Other Oil Crops
Vegetable Oil (Soy)
| Meal (Soy)
| Soybean
¦ DDG
¦ Corn
-2.0
2030
2050
2030
2050
2030
2050
2014
As discussed in Section 6.1, most of the corn ethanol shock in the ADAGE results is met
through additional corn production in the USA region, rather than imported corn. This results in
additional DDG production, roughly 41 percent of which is exported to the non-USA region.
There is very little change in trade of corn in the ADAGE results. In the GCAM results, the USA
72
-------
region reduces gross exports of corn to supply a portion of the additional demanded ethanol
feedstock. Of the additional DDG production in the USA region, roughly 18 percent is exported.
In these GCAM results, there are also decreases in U.S. net exports of other crops, most notably
soy and wheat. This is due to competition for land leading to some crop switching from other
crops to corn production in the USA region, resulting in less of these crops being available for
export. The GTAP results show a similar pattern as the GCAM results, i.e., net exports of DDG
increase while net exports of other commodities decrease relative to the reference case. Relative
to the GCAM results, the GTAP results include a smaller increase in DDG net exports, a smaller
decrease in corn net exports, but a larger decrease in net exports of other commodities such as
soybeans. As discussed in Section 6.1, in these GLOBIOM results most of the additional corn
used for ethanol feedstock in the corn ethanol shock scenario is produced in the USA region by
switching cropland from other crops to corn production. This results in greater reductions in the
production of other crops compared to what we observe in the ADAGE and GCAM results, most
notably in production of soy, wheat, and other crops. This results in larger decreases in exports
of those crops from the USA region in the GLOBIOM results. In these results, GLOBIOM
chooses to consume most of the additional DDG production domestically in 2030 and 2050,
which creates greater flexibility to divert corn used to meet the ethanol shock from the feed
market. In 2050, however, GLOBIOM estimates additional crop switching from soy to corn,
increasing the amount of corn which is used for animal feed and freeing up some DDG for
export in that model period.
6.5 Crop Yield
As discussed in Section 5.3 above, the four economic models included in this comparison
exercise all have the ability to increase crop yields in response to changes in crop price. The
theoretical basis for yields responding to price is similar across models; to the extent producers
see long-term revenue per ton of crop increasing, they may choose to invest in more expensive
but higher yielding agricultural technologies (i.e., invest more revenue in capital and material
inputs to production) and/or increase their personnel (i.e., invest more revenue in labor inputs to
production).
As discussed in Section 5.3 above, the endogenous mechanisms within each model which
simulate these decisions vary in structure. GCAM and GLOBIOM each represent four distinct
crop management options for each crop, though the characteristics of the four options in each
model are not fully aligned with one another. In ADAGE and GTAP, inputs of labor, capital, and
materials may be increased to generate higher yields through nested CES production functions.
The main similarity across these four models when it comes to changes in crop yield is that an
increase in crop price is the mechanism by which higher crop yields are induced. However, these
differences in endogenous yield response mechanisms indicate that each model would be
expected to simulate somewhat different patterns and magnitudes of crop yield response to a
given change in price.
Reference case yield trends are also an important factor in understanding differences
across models. As shown in Figure 5.3-1, reference case corn crop yield trends across the four
economic models are fairly similar in the historical periods of 2010 and 2015, though not
identical. However, for the three dynamic models, ADAGE, GCAM, and GLOBIOM, the trends
73
-------
in reference case corn yields diverge over time. Yields are calibrated to improve over time in all
three models however, reflecting a shared assumption that agricultural technologies will continue
to improve into the future. In reviewing the change in corn yields in our shock scenario relative
to the reference case shown by these dynamic models, the reader should keep in mind that yields
are improving over time in both the USA and non-USA regions in both scenarios, as they do in
the reference case.
As shown in Figure 6.1-2, crop intensification contributes to the sourcing of corn for the
ethanol shock to varying degrees across the models. In the biofuel volume shock scenarios
modeled for this exercise, we observe that the contributions from intensification are a minority of
the feedstock sourcing solution, accounting 15 percent or less of the additional feedstock
required. Intensification is a part of each model solution to at least some degree however, and we
can make some useful observations about how this effect is similar and different across the
models considered.
Before discussing the modeled crop yield results from this exercise, it is important first to
understand what is meant in this case by the term intensification. Increasing crop yield per
harvested unit of land is only one method of intensifying crop production. In regions of the world
where climatic conditions allow for it, multi-cropping (i.e., planting more than one crop per year)
is another option. GLOBIOM and GTAP consider this option explicitly to some extent by
distinguishing between the physical area on which crops are planted and the number of harvests
achieved annually on that area. In ADAGE and GCAM, no such distinction is made, and multi-
cropping is represented implicitly, embedded in the average yield for a given crop in a given
growing region. GTAP does not report total areas of multi-cropping in a given scenario, but it
does calculate and report changes between scenarios in harvested cropland area, unused cropland
and multi-cropping area. Thus, increasing the ratio of harvested to planted cropland area is a
distinct intensification strategy for GTAP.
Another intensification option is to shift production from less productive land or growing
regions to more productive land or regions. More productive land is assumed in these models to
garner a higher rental rate (i.e., the land is more expensive to purchase or use) because of the
higher revenues it can generate. As crop prices rise however, crop producers can potentially
afford more of this more expensive land. This intensification option is represented in all four
models to varying degrees, as the spatial detail of growing regions and land cover varies across
models.
When models report average yield for a given crop across a broad geopolitical region,
that output value mixes together some, but not necessarily all, of these effects. Depending on
how the reported yield value is calculated, different information about intensification may be
embedded. For the purposes of this section, yield output is calculated as regional production of a
crop divided by reported regional cropland use for that crop (these outputs are discussed in
greater detail in Section 6.6 below). Therefore, the reader should keep in mind that what is
discussed in this section as modeled crop yield output represents intensification more broadly
and is not only an improvement in the yield of a crop on specific acres of land through greater
investment in crop production inputs on that land.
74
-------
As shown in Figure 6.5-1 below, average USA region corn yields increase in all four
models in response to the corn ethanol shock. One can compare these results with the reference
case yields presented in Figure 5.3-1 and observe that these improvements are minor, less than a
1 percent improvement in USA region average yield in all cases. While improvements may be
larger in particular growing regions, the average yield across the USA region is instructive in
understanding why intensification plays only a minor role in the sourcing of corn for the ethanol
shock. As a collective, these four models estimate the corn ethanol shock modeled for this
comparison would induce relatively minor improvements in corn yield. This small observed
change in USA region corn yields is reasonable in light of the crop price changes. Figure 6.5-2
below shows that the change in corn price is also small, less than 0.5 percent in 2030. As
discussed above, crop price is the primary driver of increased crop yields and intensification in
general, and a small price change would be expected to induce a small yield response as well.
Looking at the non-USA results, there is even less effect on corn yield. This is not an
unexpected result. Figure 6.3-1 above shows the increase in corn production in response to the
shock is concentrated in the USA region. Figure 6.5-2 shows there is virtually zero change in
corn prices in the non-USA regions in response to the shock as well. This lack of perturbation of
the non-USA corn systems would not be expected to induce much change in corn yields.
Figure 6.5-1: Difference in corn yield in the corn ethanol shock relative to the reference
case in 2014 (GTAP) and 2030 (ADAGE, GCAM, GLOBIOM, GTAP)
Corn Shock (1BG)
Corn
2030
0.040
S ™ 0.020
0.000
0.038
2014
0.021
0.009
0.005
Model
¦ ADAGE
¦ GCAM
¦ GLOBIOM
¦ GTAP
0.040
: 0.020
0.000
0.002
0.000
0.000
0.000
ADAGE
GCAM
GLOBIOM
GTAP
75
-------
Figure 6.5-2: Percent difference in commodity prices in the corn ethanol shock relative to
the reference case185
0.5%
0.4%
0.3%
<
(/)
^0.2%
0.1%
0.0%
0.5%
0.4%
^0.3%
Z3
00.2%
0.1%
0.0%
ADAGE
GCAM
GLOBIOM
t
~
GTAP
I
Jj
Commodity
¦ Corn
¦ Soybean
¦ Energy Crops
¦ Other Crops
¦ Other Grains
¦ Other Oil Crops
¦ Palm Fruit
¦ Rapeseed
¦ Rice
¦ Sugar Crops
¦ Wheat
2030
2030
2030
2014
In the dynamic models, it is also instructive to consider the trend in yield change over
time, relative to the reference case. As shown in Figure 6.5-3 below, the pattern of this change
over time varies across the three dynamic models. Looking first at the results for the USA region,
in two of the three dynamic models, ADAGE and GCAM, the corn crop yield response to the
corn ethanol shock is strongest in 2030, the time step in which the shock reaches its peak. The
yield response diminishes thereafter over time, likely reflecting the fact that reference case yields
continue to improve in both of these models beyond 2030. The GLOBIOM results show a
different pattern. However, because all of these changes are fairly small compared to the
reference case corn yield, it is difficult to read much into the trends over time. Outside of the
USA region, none of the four models show a substantial change in corn yield. These responses
are consistent with the changes in corn area in each of the three models, described in Figure 6.6-
2 further below.
185 Average commodity prices for non-USA regions in GTAP results were not available for this exercise.
76
-------
Figure 6.5-3: Difference in corn yield in the corn ethanol shock relative to the reference
case in 2014 (GTAP) and over time from 2020 to 2050 (ADAGE, GCAM, GLOBIOM)
Model
0.035
0.030
0.025
J2 0.020
0.015
0.010
0.005
0.000
While the corn crop yield change results may appear to be somewhat different across
models based on Figure 6.5-3, when compared to reference case corn yields in each model they
are all relatively small. In ADAGE, GCAM, and GLOBIOM the percent differences in corn
yields in 2030 in the corn shock relative to the reference case are all less than one percent for the
USA and non-USA regions. We can observe from these results that the four economic models
generally agree that, in the specific scenarios modeled for this exercise, yields are not projected
to improve substantially in response to the corn ethanol shock. However, it is also notable that
even these small changes in corn yield are responsible for a small but notable percentage of the
additional corn produced to meet the shock.
From this exercise however, we cannot draw any firm conclusions from this yield
comparison regarding whether one method is superior to the others. All four of the models seem
to behave reasonably in these yield results. Sensitivity analysis may reveal the degree to which
GHG emissions results change when the underlying assumptions about crop yield responsiveness
to price are changed. This may indicate areas for further research.
6.6 Land Use
As described in Sections 6.1 and 6.3, in the ADAGE, GCAM, GLOBIOM, and GTAP
results, some of the corn ethanol shock is met by increased corn production, which comes from a
mix of cropland shifting from other crops to corn, land use change from other land types to
cropland, and changes in corn yield. As shown in Figure 6.6-1, corn cropland in the USA region
increases by 0.3 Mha in GTAP (2014) and 0.4-0.5 Mha in ADAGE, GCAM, and GLOBIOM
(2030). All of these model results show some amount of shifting of other crops to corn, but the
77
-------
amount of crop shifting varies. Model results also show differences in the impact on non-USA
regions.
In the GTAP and GLOBIOM results, most of the new corn cropland in the USA region
comes from shifting of other crops. In these model results, the area of soybean and wheat
increases in non-USA regions to make up for the loss of production of these crops in the USA
region. In both the GTAP and GLOBIOM results, the total cropland increases more in non-USA
regions than in the USA region, even though the corn for the corn ethanol shock is coming from
the USA region. In the ADAGE results there is some cropland shifting in the USA region, but a
larger net increase in cropland area in the USA region than seen in the GTAP or GLOBIOM
results. ADAGE has small amounts of cropland shifting in non-USA regions, with minimal
changes in total non-USA cropland. In the GCAM results, a much smaller fraction of the new
corn cropland is coming from crop shifting, and the net increase in cropland in the USA region is
higher than in the other models. The GCAM results also show an increase in corn cropland in
non-USA regions, reflecting the increased corn production in non-USA regions to meet the
shock.
Figure 6.6-1: Difference in cropland area by crop type (million hectares) in the corn
ethanol shock relative to the reference case in 2014 (GTAP) and 2030 (ADAGE, GCAM,
GLOBIOM)186
2030 2014 Land Use
ADAGE GCAM GLOBIOM GTAP _ _
¦ Corn
¦ Soybean
0.40 | | ¦ Sugar Crops
¦ Wheat
0 20 BOther Grains
¦ Other Oil Crops
¦ Other Crops
Lj.ll!
-0.20 — I
-0.40
<<<<<<<<
LOLOLOLOLOLOLOLO
ZD ZD ZD ZD Z> Z) | ZD ZD
a sz c c
o o o o
186 Horizontal lines show the net change in cropland. Cropland area shown represents land cultivated for row crops
in ADAGE and GCAM and harvested area in GLOBIOM and GTAP. When a single unit of land is harvested
multiple times in a single year, the area is counted multiple times as "harvested area" but only a single time as
"cultivated area."
78
-------
Each model considered here categorizes land in somewhat different ways (summarized in
Section 5.2), and each uses different methods for determining which land types, and how much
of each, are converted in response to economic stimuli in scenario runs (summarized in Section
2). In addition, the historical data sources on which the models rely to estimate reference case
land cover and land use differ in some ways, with data primarily coming either from FAO or
from the GTAP database.
The four economic models all choose to expand cropland to some degree to meet
growing crop demands in the corn ethanol shock, which subsequently causes changes in the area
of other land types in each model (Figure 6.6-2). In the ADAGE results for the corn ethanol
shock, most of the new cropland converted in the USA region comes from managed pasture. Due
to the land rent and net primary production (NPP)187 assumptions in ADAGE, that is the most
profitable conversion option. Very little land is converted outside the USA region in these
ADAGE results.
The GCAM results for the corn ethanol shock show decreasing cover for a mix of land
types in both USA and non-USA regions, with the largest shift in land use estimated to come
from unmanaged pasture. The change in USA land use is approximately three times greater than
the non-USA change in use. In the GLOBIOM results, very little new cropland is created in the
USA region; what change does occur comes largely from managed pasture. In the non-USA
region, the area of other arable land and grassland decreases relative to the reference case. As
explained in Section 2, in these model runs GLOBIOM does not allow forest conversion in the
USA and EU regions and restricts natural land conversion. The restriction on natural land
conversion may be a significant explanatory factor behind the observation in these GLOBIOM
results that the new corn cropland is mostly coming from crop shifting, rather than from a net
increase in cropland.
In the GTAP results, most of the new cropland comes from other arable land, which
includes the land types categorized in the GTAP results as "cropland pasture" and "unused
cropland." In the GTAP results, in the USA region, about 75 percent of the increase in harvested
area is explained by a reduction in cropland pasture area (land that fluctuates between cropland
and pasture and was unharvested in the reference case), 16 percent by a reduction in unused
cropland, 7 percent by a decrease in pasture, and 4 percent by an increase in multi-cropping. In
the GTAP results, in the non-USA regions, cropland pasture is once again the main source for
new harvested area (54 percent), followed by pasture (21 percent), unused cropland (12 percent),
forest (7 percent) and increased multi-cropping (6 percent). The GTAP results show no change in
unmanaged forest, grassland or pasture as these are not land categories in the GTAP model.
Each of the models has different assumptions about the carbon stock of different land
types in different regions. As shown in more detail in Section 6.7, the type and amount of land
converted and the carbon stock of the land types will factor in to the emissions from land use
change.
187 Net primary production is a measure of the rate of increase in plant biomass.
79
-------
Figure 6.6-2: Difference in land use (million hectares) in the corn ethanol shock relative to
the reference case in 2014 (GTAP) and 2030 (ADAGE, GCAM, GLOBIOM)188
Land Cover Type
e>
El
to
0.3
0.2
0.1
0.0
<-> -0.1
-0.2
ADAGE
2030
GCAM
GLOBIOM
2014
GTAP
I Cropland
I Grassland
I Managed Forest
I Un managed Forest
Managed Pasture
I Unmanaged Pasture
Shrubland
I Other Arable Land
I Other Not Arable Land
-0.3
USA
Non-USA
USA
Non-USA
USA
Non-USA
USA
Non-USA
Following the trends observed in the crop production results, the models show variation
in both the magnitude and location of land use change. As might be expected given their
differences in land competition structure and land categorization, these four models also present
diverse estimates regarding what types of land might be converted to cropland in response to
greater demand for corn ethanol. The models show some consistency in that they all convert a
significant share of the new cropland from pasture lands. Beyond this, some models convert
some generally smaller amount of forest land while others convert some amount of natural
grassland. Some of this uncertainty appears to be spatial in nature, that is, the models have
different estimates regarding where in the world cropland will expand. However, a significant
portion also appears attributable to differences in land conversion flexibility across the models.
Both factors are areas ripe for sensitivity and uncertainty analysis. As discussed in detail in
Sections 8 and 9, we have conducted some analyses of this sort for this exercise, but this remains
an area of potential for future research.
6.7 Emissions
The modeled results of energy consumption, crop production, and land use change
described above come together in the modeled greenhouse gas emissions. As shown in Figure
6.7-1, the modeled GHG emissions over time vary by model.
188 [n figure 6.6-2 and 7.6-2, "Cropland" area in GTAP represents land cultivated for row crops (calculated as the
change in harvested area minus the change in multicropping), while cropland pasture, and other unused cropland
have been reassigned to "Other Arable Land." This differs from Figure 5.2-1, in which cropland pasture and other
unused cropland are reported under the "Cropland" category.
80
-------
Figure 6.7-1: Difference in global greenhouse gas emissions in the corn ethanol shock
relative to the reference case189
GHG Emissions by Source
Model
Emission Source
ADAGE
GCAM
GLOBIOM
>-
CM
LUC O 0
KJ
£
-5_
7KT
^ 5
>
Energy from *
CM
Crop Production o o
+¦»
5
-5
- 5
„>.
Livestock n
0 0
Production u
2
-5
^ 5
>%
cT
Other (Industrial cm
& Waste) S 0
5
-5
2020 2030 2040 2050
Year
2030 2040
Year
2030 2040
Year
C02
CH4
N20
Other GHGs
Net GHG Emissions (All Represented Sources)
Model
ADAGE
GCAM
GLOBIOM
A
2020 2030 2040 2050
2020 2030 2040 2050 2020 2030 2040 2050
Year
Year
Year
189 GTAP is not included in this figure because it does not represent emissions over time, and due to time
constraints, we do not have GTAP GHG emissions by gas for the source categories used in this figure. For
comparison, for GTAP, in the corn ethanol scenario relative to the reference case (2014), LUC emissions = 0.46 Mt
CO?c. fossil fuel combustion and industrial CO2 emissions = -1.15 Mt. and other GHGs emissions from all covered
sources = 0.085 Mt COie, of which N20 = 0.41 Mt CO;C. CH4 = -0.28 Mt CO2C. fluorinated gases = 0.001 Mt
CO?c. and other CO;= -0.045 Mt CO:c: net total GHG emissions = -0.61 Mt CO;C. GREET is not included in this
figure because it does not represent scenario-based emissions over time. See Table 6.7-1 for carbon intensity values.
81
-------
Emissions from land use change show different patterns in the GCAM, ADAGE, and
GLOBIOM results due to the type of land use change occurring relative to the reference case and
to the carbon stock assumptions in each model. In the ADAGE results, most of the land use
change emissions that occur are attributable to the conversion of pasture to cropland. ADAGE
assumes that the soil carbon stock of cropland in the USA region is higher on a per-hectare basis
than the soil carbon stock of pasture.190 Therefore, the conversion of pasture to cropland causes
net carbon sequestration, and the emissions over time are less than in the reference case, but
close to zero. In GCAM, most of the cropland change is estimated to convert from land types
with relatively low carbon stocks, such as pasture and grassland. However, some of the land use
change is attributable to reduced future afforestation relative to what GCAM estimates would
occur in the future in the reference case. Even though the amount of change in future forest land
is small compared to the amount of change in other land types, the relatively higher carbon
stocks of forest compared to other land types lead to higher overall land use change emissions in
these GCAM results, relative to the other models. GLOBIOM shows conversion of cropland
from grassland and the other arable land aggregate category, which results in estimated LUC
emissions in between those of ADAGE and GCAM. The GCAM and GLOBIOM results show
land use change emissions peaking in 2030. This is because land conversion to cropland happens
primarily from 2020-2030 as more land is needed to increase corn production to meet the corn
ethanol shock.
"Energy from Fossil Fuels" (or "fossil fuel emissions") includes emissions associated
with producing biofuels (e.g., from consuming natural gas or electricity for process energy),
direct emissions associated with on-farm energy use to produce feedstock, and transporting both
biofuel feedstocks and finished fuels, as well as emissions from indirect impacts on the energy
sector, including displaced gasoline use for transportation that is replaced by corn ethanol. Of the
three models shown in Figure 6.7-1, these emissions are reported by ADAGE and GCAM. In the
corn ethanol results from these models, emissions from fossil fuels are lower than in the
reference case. Fossil fuel emissions reductions in the GCAM results become larger until 2030,
and then stay relatively constant through 2050. In the ADAGE results, emissions reductions
become larger until 2030 but then become smaller from 2030 to 2050 (while staying below the
reference case emissions). As shown in Section 6.2, fossil fuel consumption decreases in the corn
ethanol shock scenario relative to the reference case. GCAM results show the most reduction in
fossil fuel consumption, leading to a greater emissions reduction in the GCAM results than in the
ADAGE results. The drivers of these varying results in fossil fuel consumption are discussed in
Section 6.2 above.
Crop production emissions are higher than the reference case in the ADAGE, GCAM,
and GLOBIOM results. Changes in crop production emissions relative to the reference case are
due to changes in the types and quantities of crops grown in the models, and primarily come
from changes in N2O emissions, driven by both increased fertilizer use and direct nitrogen
fixation by soybeans. As shown in Section 6.3, ADAGE, GCAM, and GLOBIOM results all
show increases in corn production, with smaller changes in the production of other crops.
GLOBIOM results also show shifts in the location of soybean production. The increase in crop
production emissions is small in all of these model results. In the GLOBIOM results, the crop
190 These assumptions are based on an area-weighted average of carbon stocks from an earlier version of GCAM
(GCAM 3.2).
82
-------
production emissions increase over time. In the ADAGE and GCAM results, the crop production
emissions peak in 2030, and then decrease slightly until 2050. The change in emissions relative
to the reference case from the livestock sector and from industrial and waste management sectors
is very small.
The total change in GHG emissions across all sources over time varies across the models
(Figure 6.7-1). The ADAGE results show a net decrease in emissions from 2020-2040, primarily
driven by the decrease in CO2 emissions in the energy from fossil fuels category. From 2040-
2050, emissions are higher than in the reference case because the increase in N2O emissions from
crop production becomes larger than the decrease in CO2 emissions from fossil fuels. In the
GCAM results, net GHG emissions are greater than the reference case from 2020-2030 and less
than the reference case from 2035-2050, because the CO2 emissions from land use change
decline rapidly after 2030. In the GLOBIOM results, net emissions are greater than the reference
case from 2020-2050, because the largest contributors to emissions (CO2 from land use change
and N2O from crop production) are greater than the reference case over this time period.
There are a few commonalities across the ADAGE, GCAM, and GLOBIOM results of
emissions over time. All of these model results show small but positive emissions from crop
production relative to the reference case. The model results also all show very small emissions
from livestock production, waste management, and industry. There are also some key differences
in the emissions. Although GCAM and ADAGE both consider indirect impacts on the energy
sector, the emissions over time from the energy sector are very different. Future research could
explore the factors that determine the extent of refined oil displacement in each model through
sensitivity analysis. Additionally, there are large differences across the model results in the
amount of land use change emissions, due to differences in both the types of land converted and
the carbon stock assumptions. A sensitivity analysis of the carbon stock assumptions in GCAM
is shown in Section 9.2 below, and a sensitivity analysis of the land conversion elasticities in
ADAGE is shown in Section 9.3. Future research could focus on the impact of carbon stock
assumptions in other models, or on other model parameters that determine the types of land
converted.
As a next step in considering the lifecycle greenhouse gas emissions associated with the
corn ethanol shock in these model results, we calculated a carbon intensity (CI) for each category
of emissions. A CI is an estimate of the emissions per unit of fuel, which we express here in
kgC02eq/MMBTU. The CI calculated from a model run depends on the particular scenario and
model assumptions used. To calculate a CI for the ADAGE, GCAM, and GLOBIOM results, we
summed the emissions relative to the reference case from 2020 to 2050 to get the difference in
total cumulative emissions relative to the reference case. Then, we summed the difference in
corn ethanol consumption in the USA region (i.e., the corn ethanol shock) over 2020 to 2050 to
get the total cumulative biofuel consumption difference relative to the reference case. Finally, we
divided the cumulative emissions difference by the cumulative biofuel consumption difference to
estimate a CI. The calculated CI depends on the time horizon included in the calculation, because
the annual emissions vary over time. For example, emissions in the corn ethanol scenario relative
to the reference case may be higher from 2020-2030 than in later time steps, as is the case in
these GCAM and GLOBIOM results (Figure 6.7-1), or lower in 2020-2030 than in later time
steps, as is the case in these ADAGE results. Calculating a CI using only the results from 2020-
83
-------
2030 would result in a higher CI than considering emissions from 2020-2050 for GCAM and
GLOBIOM in this case. The opposite would be true for ADAGE in this case. For GTAP results,
we divided the emissions difference by the biofuel consumption difference in the USA region in
the single 2014 time step. GTAP emissions are given for a single year, but these results are
amortized over a 30 year time period. Results from GREET are already given as carbon
intensities, i.e., this is the metric GREET is designed to estimate.
When interpreting the ADAGE, GCAM, GLOBIOM, and GTAP CI results, a CI of zero
means that global GHG emissions are equal in the shock case and the reference case, a positive
CI means a greater quantity of GHGs are emitted globally relative to the reference case, and a
negative CI means a smaller quantity of GHGs are emitted globally relative to the reference case.
Importantly, a negative CI from one of these four models does not necessarily represent GHG
sequestration, but rather is best interpreted as a lower rate of emissions. Conversely, because
GREET is an attributional rather than consequential approach, a CI of zero means that the supply
chain for the fuel is estimated to not produce any emissions, a positive CI means that the supply
chain is estimated to release net GHG emissions, and a negative CI means that the supply chain
is estimated to achieve net GHG sequestration.191
Table 6.7-1 shows the CI of corn ethanol calculated using the emissions reported by each
model. Models are divided between those frameworks with energy markets (in the left side
columns) and models without energy markets (in the right side columns). This division is made
to reflect important differences in the sectors represented and the difficulty of direct
comparability between models on the left with models on the right. ADAGE, GCAM, and GTAP
include global emissions from every economic sector, including indirect, market-mediated
impacts. GREET includes detailed emissions estimates from fuel production, transport, and use,
but, as it is not a consequential model, it does not estimate the net change in GHG emissions
resulting from a change in biofuel consumption. Rather it estimates the emissions directly
attributable to the biofuel supply chain. GLOBIOM does not include any energy sector
emissions, but does include market impacts on crop production and the livestock sector.
Because of the differences outlined above, it would be inappropriate to compare all of the
emissions estimates across all of the models, but we can make several meaningful comparisons.
Results from the three models with energy markets (ADAGE, GCAM, GTAP) can be directly
compared, with the caveat that GTAP is representing 2014 while the other models are
representing a 2020-2050 scenario. Furthermore, we can compare the land use change emissions
estimates for all of the models, as GREET uses a consequential approach for this category of
emissions, again with proper caveats about temporal differences. We can also compare crop
production and livestock sector emissions estimates from ADAGE, GCAM and GLOBIOM.192
In the table below, we report emissions from "Agriculture, forestry and land use" for all five
191 This sentence about interpreting GREET CI estimates applies for biofuel pathways, such as corn ethanol and
soybean oil biodiesel, produced from "primary" feedstocks, but not for all pathways made with waste, byproduct or
residue feedstocks. For the waste, residue, and byproduct pathways, GREET sometimes considers emissions relative
to a baseline/counterfactual scenario, in which case a negative CI cannot always be interpreted as a net GHG
sequestration.
192 GTAP can also report emissions disaggregated into these source categories, but due to time constraints we did
not obtain such results from GTAP for this exercise.
84
-------
models as the sum of emissions from these stages; however, the GREET estimate for this
aggregate category is not directly comparable with the other models for reasons discussed below.
Energy sector emissions have a large impact on the CI in the ADAGE, GCAM, and
GTAP results. The energy sector CI is much lower (more negative) for the GCAM results than
for ADAGE and GTAP results, which is consistent with the greater cumulative global reduction
of refined oil use (shown in Figure 6.2-3) and lower emissions from fossil fuels over time
(shown in Figure 6.7-1). GREET reports the CI from fuel production and transportation but does
not consider indirect impacts on the energy sector, such as the energy rebound effects shown in
Section 6.2. The fuel production and transportation CI in the GREET results is based on the
amount of process energy needed for corn ethanol production as well as the amount of energy
needed to transport the feedstock and the fuel. This is why we use the label "Energy Sector" for
the first row in Table 6.7-1 for the three models with energy markets, but the label "Biofuel
Production" for this row for GREET.
85
-------
Table 6.7-1: Carbon intensity of corn ethanol (kgCCheq/MMBTU) calculated using
emissions reported by each model193
Models with Energy Markets
Models without Energy Markets
ADAGE
GCAM
GTAP
GLOBIOM
GREET
Sector/stage-
specific
emissions
Energy
from Fossil
Fuels
-15
-65
-15
Biofuel Production
X
29
Crop
Production
14
16
1
Crop Production
9
X
Feedstock
Production
X
16
Livestock
Sector
0.1
0.3
Livestock Sector
-1
X
Other
1
-1
Fuel Use
X
0.4
Land Use
Change
-1
31
6
Land Use Change
13
8
Totals
Agriculture,
forestry,
and land
use
14
47
7
Agriculture,
forestry, and land
use
21
24
Global
GHG
Impact
-1
-19
-8
Global GHG Impact
X
X
Supply
Chain GHG
Emissions
X
X
X
Supply Chain GHG
Emissions
X
53
The ADAGE and GCAM results show a similar CI from crop production. The crop
production CI from the GLOBIOM results is lower than these models, consistent with the lower
emissions over time in GLOBIOM relative to ADAGE and GCAM. GREET's feedstock
production CI is based on the energy and chemical inputs required to produce the amount of corn
needed for 1 MMBTU of ethanol. Unlike the other models, this value does not represent the
change in crop production emissions associated with an increase in ethanol production; in other
words, it does not include indirect impacts on the production of other types of crops. Livestock
and other sectors (including waste management and other industrial sectors) have only minor
impacts on the overall CI in ADAGE, GCAM, and GLOBIOM.
For the GTAP results, as discussed in Section 3.1.4, we have estimates of non-C02
emissions by greenhouse gas, but we do not have these emissions disaggregated by sector or
193 "X" means that the model does not report that category. For GTAP, emissions from crop production, the
livestock sector, and "other" are reported as an aggregated value of non-LUC, non-fossil fuel emissions. Negative
values for ADAGE, GCAM, GTAP, and GLOBIOM mean that emissions are lower than the reference case, whereas
positive values mean the emissions are higher than the reference case.
86
-------
lifecycle stage. GTAP can also report emissions disaggregated into these source categories, but
due to time constraints we did not obtain such results from GTAP for this exercise. The largest
changes, by gas, are an increase in N2O and a decrease in CH4. We believe the bulk of the
changes in these emissions are associated with changes in fertilizer N2O and livestock CH4, but
more work would be needed to confirm our intuition. For these reasons, in Table 6.7-1, we report
the aggregated non-CCh emissions estimate from GTAP across three rows combining Crop
Production, Livestock Sector and Other. This aggregated emissions estimate from GTAP is
lower than what the other models report for the sum of emissions from these three categories.
We would need to do more research to disaggregate these emissions and understand why they
are lower than estimates from the other models.
Land use change emissions are reported in all the models, and the CI results have wide
ranges across the models. As explained above, these differences are due to the type of land use
change and the carbon stocks of each land type in the models. GREET's LUC CI is based on
Argonne's CCLUB translation of a preestablished GTAP run using a different shock size (11.59
billion gallons of corn ethanol) from a 2004 baseline. This earlier GTAP run estimated a global
cropland area increase of 2.1 million hectares, with 47 percent of that additional land
requirement coming from the USA region, and forest land making up about 11 percent of the
land needed to convert to cropland.194
We can compare "Agriculture, forestry and land use change emissions" across four of the
models (ADAGE, GCAM, GLOBIOM, GTAP). For GTAP, we include the non-CCh emissions
in this category. For this category, the GCAM results include the highest emissions, driven by
the land use change emissions. Although the ADAGE results include lower land use change
emissions than the GTAP results, the aggregated agriculture and forest sector emissions are
higher for the ADAGE results, due to the difference in crop production emissions.
The total global CI can be compared across ADAGE, GCAM, and GTAP, because all of
these models represent the same sectors and include market impacts. The results from these
models show a range in corn ethanol CI, primarily due to differences in the energy sector CI and
land use change CI. For GLOBIOM and GREET, a total global CI cannot be calculated from the
model results because these models do not include all the relevant sectors and/or do not include
all the relevant market impacts. For GREET, we calculate the total supply chain CI. This is a
different metric than the other models' CIs, since GREET primarily uses an attributional
approach, coupled with consequential ILUC modeling from GTAP and CCLUB in lifecycle
analysis rather than a consequential approach. This value does not include any displacement of
fossil fuel consumption that would occur from the increased consumption of biofuels.195
194 Taheripour, Farzad, Wallace Tyner, and Michael Wang. 2011. "Global Land Use Changes Due to the U.S.
Cellulosic Biofuel Program Simulated with the GTAP Model." Argonne National Laboratory and Purdue
University. https://greet.es.an.Lgov/pnblication-lnc ethanol
195 GREET's ethanol CI estimates are often compared with GREET CI estimates for gasoline to derive a GHG
percent reduction relative to gasoline. In our 2010 RFS analysis, we similarly compared ethanol CI estimates from
models that do not include energy markets with a CI estimate for gasoline to calculate a percent reduction in
emissions.
87
-------
6.8 Summary of Corn Ethanol Estimates
Section 6 compares and contrasts the corn ethanol modeling estimates from ADAGE,
GCAM, GLOBIOM, GREET, and GTAP produced for this exercise. These models source the
corn ethanol required to meet the assumed shock in different ways in these results, but there are
some commonalities. Across frameworks, the two primary model strategies are to source corn
from new production and to divert corn from other uses. However, different models rely more on
one of these sourcing strategies or the other. Because of these differences in sourcing strategy,
the model results differ regarding the total additional corn production, crop trade, and land use
change impacts of the shock. The model results also have some other notable similarities and
differences. ADAGE, GCAM, GLOBIOM, and GTAP results all show a small amount of crop
yield intensification. The results also show a displacement of corn for feed use with DDG,
though there is disagreement regarding how much might be consumed in the USA region versus
exported and consumed elsewhere in the world. The models which explicitly include the energy
sector, ADAGE, GCAM, and GTAP, all show a decrease in refined oil consumption in the USA
region in their results, and an increase in non-USA regions. But there are notable differences
across these models in the total global displacement of refined oil. These factors all contribute to
differences in the estimated GHG emissions and CI of corn ethanol across the models, with
energy sector emissions and land use change emissions differing the most across the model
results.
The previous sections also highlight potential areas for future research. Sensitivity
analysis could better define the GHG emissions implications of model decisions regarding the
location of additional DDG consumption. Further research and sensitivity analysis could also
seek to better understand the parameters that influence land conversion to cropland. Furthermore,
research and sensitivity analysis could seek to better understand why model results show a range
in the reduction of refined oil consumption. These are only a few examples of the many research
topics that could help to explain what is driving differences in these model results.
7 Comparison of Soybean Oil Biodiesel Estimates
In this section, we present the results of the soybean oil biodiesel shock. The results in
this section show the difference between the soybean oil biodiesel shock and the reference case.
We consider the following elements in turn:
• Sources of soybean oil biodiesel to meet the shock
• Energy market impacts from the shock
• Crop production and consumption
• Trade impacts
• Yield changes
• Land use impacts
• Emissions: the modeled results of energy consumption, crop production, and land use
change described above come together in the modeled greenhouse gas emissions.
The majority of these comparisons include ADAGE, GCAM, GLOBIOM, and GTAP.
Only the comparison of GHG emissions includes GREET. GREET is a supply chain LCA model
88
-------
that does not represent changes in agricultural and economic markets between reference and
modeled scenarios, as the other models in this comparison exercise are designed to estimate.
7.1 Sourcing Overview
As in the corn ethanol runs, the models included in this analysis have many options
available for meeting the soybean oil biodiesel consumption shock, including increased
production of soybean oil biodiesel and changes in biodiesel imports and exports. Increased
soybean oil biodiesel production could come from diversion of soybeans or soybean oil from
other uses, increased crushing of existing soybean supplies, or increased production of soybeans.
This section will give an overview of the extent to which the models rely on each of these
options for meeting the soybean oil biodiesel consumption shock.
In the soybean oil biodiesel shock, the models show a range of solutions for meeting the
shock (Figure 7.1-1). In the ADAGE soybean oil biodiesel results, around half of the shock is
met by increased biodiesel production in the USA region, and half is met by increased gross
imports to the USA region. In the GCAM results, 77-79 percent of the shock is met by increased
soybean oil biodiesel production in the USA region, and 21-23 percent is met by a combination
of increased imports and reduced exports of soybean oil biodiesel. In GLOBIOM and GTAP, the
shock is met entirely by increased soybean oil biodiesel production in the USA region.
GLOBIOM does not have an energy market and therefore cannot trade biofuels, making
domestic biodiesel production the only option in this model.
Figure 7.1-1: Sources of additional soybean oil biodiesel consumed in the soybean oil
biodiesel shock relative to the reference case196
USA
GCAM
GLOBIOM GTAP
0.12
= 0.10
CD
eo
u
o
o
00
o
c
ru
CD
_Q
(
o
3
i
U1
i
CO
E
"O.
o
4—
=3
a
~cd
CO
CD
(
TD
O
ho
0.08
0.06
0.04
0.02
0.00
Source
¦ Gross Exports
Gross Imports
¦ Production
o o
lt>
o o o
CM CM CM
196 Red shows the contribution increased soybean oil biodiesel production in the USA region; orange shows the
contribution from increased soybean oil biodiesel gross imports to the USA region; blue shows the contribution
from reduced soybean oil biodiesel gross exports from the USA region.
89
-------
Although the ADAGE and GCAM results both meet a large percentage of the shock
through changes in soybean oil biodiesel imports, the impact on non-USA regions is very
different. In the GCAM results, 43-52 percent of the shock is met by reduced soybean oil
biodiesel consumption in non-USA regions (Figure 7.1-2). This latter share is larger than the
share of biofuel trade noted in Figure 7.1-1 above. The estimate in Figure 7.1-2 also includes
soybeans and soybean oil feedstock which are exported to the USA region rather than being
processed into biodiesel in their region of origin and consumed domestically. In contrast, the
ADAGE results do not show a reduction in soybean oil biodiesel consumption in other regions;
instead the increased imports are sourced from increased soybean oil biodiesel production in
non-USA regions. Energy market impacts are discussed further in Section 7.2.
ADAGE, GCAM, GLOBIOM, and GTAP meet the soybean oil biodiesel shock through
different amounts of soybean and soybean oil diversion from other uses, crop intensification,
crop shifting to soybean, and new cropland (Figure 7.1-2). Based on the assumed conversion
factor of soybean oil to soybean oil biodiesel (Section 4), if all of the shock were met by new
soybean oil biodiesel production, ADAGE, GCAM, and GLOBIOM would need 3.4 million
metric tons of additional soybean oil for biodiesel in 2030 and 3.3 million metric tons of
additional soybean oil for biodiesel in 2050 (bottom panel of Figure 7.1-2). GTAP would need
3.4 million metric tons of additional soybean oil for biodiesel in 2014. The GCAM results show
much less additional soybean oil is needed for the soybean oil biodiesel shock than in the
ADAGE, GLOBIOM, or GTAP results because soybean oil biodiesel consumption decreases in
the non-USA region in GCAM. Because soybean crushing yields about 19 percent extractable
soybean oil, if all of the additional soybean oil were coming from new soybean production,
ADAGE, GCAM, and GLOBIOM would require additional production of 17.8 million metric
tons of soybeans in 2030 and 17.6 million metric tons of soybeans in 2050. GTAP would require
an additional 18.1 million metric tons of soybeans in 2014.
In the ADAGE soybean oil biodiesel shock results, less than 5 percent of the shock is met
by commodity diversion, with the majority of the shock met by new soybean production. In the
GCAM results, because so much of the shock is met by reduction of soybean oil biodiesel
consumption in non-USA regions, much less additional soybean oil feedstock is needed than in
the other models. Of the additional soybean oil feedstock sourced in GCAM, around half comes
from commodity diversion, and half comes from new soybean production (primarily from new
cropland). In GLOBIOM and GTAP, the majority of the shock is met through commodity
diversion (85-88 percent and 83 percent, respectively). GTAP meets a small percentage of the
shock (2 percent) through a reduction of soybean oil biodiesel consumption in non-USA regions.
Commodity diversion and soybean production results are described more in Section 7.3, and land
use results are described in more detail in Section 7.6.
90
-------
Figure 7.1-2: Top panel: Percentage of the soybean oil biodiesel shock that is met by
different categories in 2030 and 2050. Bottom panel: Million metric tons of additional
soybean oil from new soybean production (red, orange, and yellow) and diversion from
other uses (green)197
ADAGE
2030
2050 1
GCAM
2030 |
2050 1
GLOBIOM
2030T1
2050 1
¦
GTAP
2014
¦ ¦
0%
50%
Percent
100%
Source Category
¦ Commodity Diversion
Crop Intensification
¦ Crop Shifting
¦ New Cropland
¦ Non-US Biofuel Consumption
ADAGE
GCAM
GLOBIOM GTAP
2030 2050 2030 2050 2030 2050 2014
7.2 Energy Market Impacts
The energy market mechanisms at play in the corn ethanol shock generally hold for
soybean oil biodiesel as well, though the magnitude and some of the detailed effects differ. We
refer to Section 6.2 above for a discussion of those principles. As noted in that section, of the
models considered under this model comparison exercise, ADAGE, GCAM, and GTAP include
explicit representations of energy commodities and energy commodity trade, end use sectors, and
energy market interactions.
The impacts of the soybean oil biodiesel shock on consumption of refined oil198 in the
USA region in ADAGE, GCAM, and GTAP broadly mirror the impacts seen under the corn
ethanol shock scenario; all three models show substantial displacement of refined oil use in the
USA region, with displacement in GCAM being the highest, displacement in ADAGE starting
somewhat less than in GCAM and declining over time, and GTAP having the smallest average
displacement of refined oil consumption in the USA region. Displacement of consumption of
197 A negative percent contribution means that there was decrease in soy production or an increase in non-fuel uses
of soybean. ADAGE has a negative percent contribution from commodity diversion in 2050 because some
additional soybeans were consumed for "other uses" - in this case, seed for additional soybean production.
GLOBIOM has a negative percent contribution from new cropland because soy cropland area decreased in non-USA
regions.
198 In these models, refined oil is an aggregation of all refined petroleum products, including gasoline and diesel.
91
-------
refined oil in the USA region results in reduced net imports of crude and refined oil, amounting
to 93 percent and 101 percent of the reduced USA consumption of refined oil in 2030 in
ADAGE results and GCAM results respectively.199
Figure 7.2-1: Difference in consumption of energy commodities (quadrillion BTUs) in the
soybean oil biodiesel shock relative to the reference case in 2030 and 2050 (ADAGE,
GCAM) and 2014 (GTAP)
ADAGE
Model / Region / Year
GCAM
GTAP
z>
I—
CO
bo (5
0.1
0.0
-0.1
0.1
I
Commodity
| Ethanol (All)
| Biodiesel (Other)
| Biodiesel (Soy)
| Coal
Natural Gas
I Refined Oil
0.0
-0.1
I
2030 2050
2030
2050 | 2030 2050 | 2030 2050 | 2014 | 2014
Trade in energy commodities plays a significant role in meeting the soybean oil biodiesel
shock in results from several of the models considered (see Figures 7.1-1 and 7.2-1). In ADAGE
and GCAM results, a substantial portion of the shock is met through greater net USA imports of
soybean oil biodiesel (48 percent and 23 percent of the shock in 2030 in ADAGE and GCAM
results respectively). In the ADAGE results, the increased net imports of soybean oil biodiesel in
the USA region are constituted almost exclusively of an increase in gross exports from the Rest
of Latin America region to the USA region. In the GCAM results, the increased net imports of
soybean oil biodiesel in the USA region are constituted of changes in exports of biodiesel across
multiple regions. It is notable that patterns of impacts of the soybean oil biodiesel shock on
biofuel trade in ADAGE and GCAM reflect the theoretical representations of trade in the two
models. In ADAGE, where trade is represented bilaterally and calibrated using historical trade
data, impacts occur almost exclusively in a region with large historical exports of biodiesel to the
USA. In GCAM, where commodities are exported to and imported from a global pool for each
commodity, impacts are distributed across multiple regions with historical exports (regardless of
destination) of biodiesel.
199 Data on trade of crude oil in GTAP results were not available for this exercise.
92
-------
We also note that GCAM's estimated reduction in consumption of soybean oil biodiesel
in the non-USA regions is greater in magnitude than the increased volume of biodiesel exported
to the USA region. This is because increased demand for soybeans and soybean oil puts upward
pressure on their prices and further reduces consumption for fuel, food, and other uses in the
non-USA regions.
Figure 7.2-2: Difference in U.S. net exports of energy commodities (quadrillion BTUs) in
the soybean oil biodiesel shock relative to the reference case in 2030 and 2050 (ADAGE,
GCAM) and 2014 (GTAP)
0.15
0.10
0.05
CO
Tj
a
0.00
-0.05
ADAGE
Model / Year
GCAM
GTAP
Commodity
| Crude Oil
| Refined Oil
| Coal
Natural Gas
| Biodiesel (Other)
| Biodiesel (Soy)
| Ethanol (Other)
Ethanol (Corn)
2030
2050
2030
2050
2014
Modeled changes in consumption of refined oil in non-USA regions are driven by two
main mechanisms in the results from ADAGE, GCAM, and GTAP. First, increased use of
soybean oil biodiesel in the USA region results in decreased consumption of refined oil in that
region (i.e., "the displacement effect"). This puts downward pressure on the global prices of
crude and refined oil, though the effect is small in absolute terms (between one and four
hundredths of a percent) due to the relatively small size of the one billion gallon shock compared
to global refined liquid fuel consumption. The result of this downward price pressure is some
increased demand for refined oil in non-USA regions. This effect is present in, and a contributing
factor to, the increased refined oil consumption seen in all three models in Figure 7.2-1. Second,
if a portion of the soybean oil biodiesel shock in the USA region is met through increased net
imports of soybean oil biodiesel, as is the case in ADAGE and GCAM, then the corresponding
non-USA regions with increased exports of biofuels have to make up that deficit in their liquid
93
-------
fuel markets by "backfilling" with either a) increased consumption of biofuels, likely coming
from increased production within those regions, or b) increased consumption of refined oil.
These two backfilling strategies are employed to different extents in ADAGE and GCAM
results. In the GCAM results, multiple regions increase exports of soybean oil biodiesel to meet
the increased demand in the USA region, but do not show commensurate increases in domestic
biodiesel production. This results in reduced consumption of biodiesel in those regions which is
backfilled with additional refined oil use. In contrast, in the ADAGE results, the increased
exports of soybean oil biodiesel from the Latin America region are met with increased
production, resulting in little impact on biofuel consumption in that region and obviating the
refined oil backfill effect shown in the GCAM results.
In summary, these dynamics explain the differences between the models in increasing
consumption of refined oil in non-USA regions. In GCAM results, deficits in liquid fuels
markets in non-USA regions are backfilled with refined oil, reducing the net global displacement
effect of the shock on refined oil consumption. In ADAGE results, deficits in liquid fuels
markets in non-USA regions are backfilled with increased biofuel production. In GTAP results,
there is little change in trade of biofuels, so there are no significant deficits in liquid fuel markets
in non-USA regions.
Finally, ADAGE and GCAM show increased natural gas consumption in the USA region,
albeit less than in the corn ethanol scenario, while GTAP shows little impact on natural gas
consumption in any region. The smaller impact on natural gas in the soybean oil biodiesel
scenario relative to the corn ethanol scenario is logical due to differences in the direct natural gas
demands of their respective fuel production technologies. The corn ethanol dry mill process
requires substantial natural gas for DDG drying, whereas the biodiesel transesterification
production process requires relatively little natural gas.
As discussed in Section 6.2, cumulative measures of the changes in refined oil and
biofuel consumption, relative to the size of the shock, are common and useful measures for
summarizing energy market impacts. These cumulative measures, illustrated in Figure 7.2-3
reflect the story presented above on the impacts of the soybean oil biodiesel shock on
consumption of other biofuels and refined oil globally.
94
-------
Figure 7.2-3: Difference in liquid fuel consumption relative to the volume of the soybean oil
biodiesel shock200
Variable Group 1 / Model
Biofuel
Refined Oil
Region
¦ USA
Non-USA
100%
99%
50%
0%
-50%
-100%
94%
100%
50%
100%
67%
24%
34%
-50%
-91%
-67%
-52%
-119%
-51%
¦86%
ADAGE GCAM GTAP ADAGE GCAM GTAP
In the lefthand pane of this figure, we see that the cumulative change in biofuel
consumption in the non-USA region amounts to one percent of the cumulative soybean oil
biodiesel shock in ADAGE, and 50 percent of the cumulative soybean oil biodiesel shock in
GCAM (largely attributable to reductions in soybean oil biodiesel consumption across a number
of non-USA regions), and six percent of the 2014 soybean oil biodiesel shock in GTAP.
In the righthand pane, we see similar directional effects on refined oil consumption in the
USA region as in the corn ethanol shock scenario discussed in Section 6.2; GCAM shows a
greater reduction in USA consumption of refined oil than the cumulative energy content of the
shocked biodiesel (119 percent), whereas ADAGE and GTAP show smaller reductions in USA
consumption of refined oil than the energy content of the shock (91 and 86 percent,
respectively). GCAM shows a much larger cumulative increase in non-USA refined oil
consumption outside of the USA region, which is driven by backfill of reduced biodiesel
consumption in the non-USA region.
The effect on cumulative net non-USA refined oil consumption - a commonly used
definition of "oil rebound" in the literature - shows how global oil consumption changes as a
21111 Values in the figure represent the difference between the shock and reference case of the given fuel category
(refined oil vs. liquid biofuels) and given region (USA region vs non-USA regions) divided by the difference in
consumption of liquid biofuels in the USA region (i.e., the shock volume). For ADAGE and GCAM, this is
calculated using cumulative volume differences between 2020 and 2050. For GTAP, which only estimates
differences in a single time step, the calculation uses only the volume differences in 2014.
95
-------
result of the shock. GCAM results show the largest increase in non-USA refined oil consumption
(67 percent of the cumulative shock) due to backfilling for traded biodiesel, as discussed above.
GTAP and ADAGE show more modest increases in non-USA refined oil consumption (34 and
24 percent respectively). The global net effect of the shock on refined oil consumption is that, on
average, for every 100 BTUs of soybean oil biodiesel required to be consumed in the USA, 67
BTUs of global refined oil consumption are displaced in ADAGE, 52 BTUs of global refined oil
consumption are displaced in GCAM, and 51 BTUs of global refined oil consumption are
displaced in GTAP. Future research could be done to better understand the parameters and
assumptions that lead to the range in reduction of refined oil consumption.
7.3 Crop Production and Consumption
As shown in Section 7.1, the ADAGE, GCAM, GLOBIOM, and GTAP results differ
notably in how much of the soybean oil biodiesel shock they each estimate would be sourced
from new soybean production. This is reflected in the estimated changes in soybean production
shown in Figure 7.3-1. The ADAGE results show the largest increase in global soybean
production, followed by GCAM, then GLOBIOM, and then GTAP. ADAGE and GCAM results
estimate the increase in soybean production would be split between the USA and non-USA
regions. In the GTAP results, the increase in production is estimated to occur almost entirely in
the USA region. In GLOBIOM, soybean production is estimated to increase in the USA region
but decrease in aggregate across the non-USA regions. ADAGE, GCAM, and GLOBIOM results
all show a decrease in corn production in the USA region as some of the new soybean area
displaces corn area.
In the non-USA region, the model results show an increase in the production of oil crops.
The ADAGE results show an increase in "other oil crop" production.201 In the GTAP, GCAM,
and GLOBIOM results, the increased oil crop production is primarily palm fruit. The GCAM
results show decreased corn production in non-USA regions, whereas the GLOBIOM results
show increased corn production in non-USA regions.
Globally, crop production increases in all four sets of model results.202 However, there is
much greater variation in the types and location of crop production across the models than there
was in the corn ethanol results. All four sets of the model results show an increase in soybean
production in the USA region, and a decrease in the production of other crops. There is
substantial variation in the crop production in the non-USA regions, particularly for soybean
production and palm fruit production. A comparison of Figures 6.1-2 and 7.1-2 lays plain one
important first order reason for this greater variability. The models show much greater diversity
in sourcing strategies for soybean oil biodiesel than they do for corn ethanol. This variation in
sourcing for soybean oil biodiesel results in more complex economic and environmental
outcomes than corn ethanol. Across the four economic models in this exercise, virtually all of the
corn for ethanol is produced in the USA region. This is largely attributable to the monolithic role
of the U.S. in historical global corn production and trade and to the fact that corn has no near-
201 As explained in Section 5.1, ADAGE does not explicitly represent oil crops other than soybeans. Therefore, for
ADAGE, "other oil crops" includes palm fruit.
202 We also looked at forest product production for the models that are able to report it (ADAGE, GCAM,
GLOBIOM), and the change relative to the reference case is negligible.
96
-------
perfect substitutes. By contrast, soybean oil does have near perfect substitutes for many end uses,
in the form of other vegetable oils. Additionally, soybean oil production and exports, and
vegetable oil production and exports more broadly, are historically distributed across more
regions. Marginal global demands for vegetable oil may reasonably be supplied from North
America, South America, or Asia. Thus, for soybean oil biodiesel, the models have a wider range
of options for the location of additional vegetable oil production. Also, soybean oil biodiesel
production has more complex impacts on the consumption and production of other crops than
corn ethanol production because of the wider range of end uses for soybean oil and meal, as
described below. The location of additional soybean production and the impact on the production
of other crops is a potential area for future research and sensitivity analysis.
Figure 7.3-1: Difference in commodity production (million metric tons) in the soybean oil
biodiesel shock relative to the reference case in 2014 (GTAP) and 2030 (ADAGE, GCAM,
GLOBIOM)
ADAGE
2030
GCAM
GLOBIOM
2014
GTAP
m
oo
10
X 0
Commodity
¦ Energy Crops
¦ Other Crops
Other Grains
¦ Other Oil Crops
¦ Palm Fruit
¦ Rapeseed
Rice
Sugar Crops
Wheat
¦ Soybean
¦ Corn
<
LO
ZD
<
U~)
ZD
c
o
<
ui
zd
<
V)
ZD
£Z
o
<
m
ZD
<
U~)
ZD
£Z
o
<
V)
ZD
<
LO
c
o
ADAGE, GCAM, GLOBIOM, and GTAP have slightly different pathways for producing
soybean oil biodiesel. In GCAM, GLOBIOM, and GTAP, soybean oil biodiesel is produced
from soybean oil. In ADAGE, soybean oil is not explicitly represented, and instead soybean oil
is part of an aggregated vegetable oil commodity. Soybean oil biodiesel in ADAGE can be
produced from vegetable oil or directly from soybeans.203 Soybean oil biodiesel produced from
soybeans produces oil crop meal (a generic vegetable meal commodity) as a coproduct.
2113 From a theoretical perspective, the latter strategy would represent a facility which co-locates crushing and
biodiesel production plants. Such a facility inputs whole soybeans and outputs biodiesel and soybean meal.
97
-------
The end use impacts of the soybean oil biodiesel shock are more complex than the
impacts in the corn ethanol shock because soybean oil biodiesel production can impact oilseed
markets, vegetable oil markets, and oil meal markets (Figure 7.3-2). The ADAGE, GCAM,
GLOBIOM, and GTAP results all show an increase in soybean crushing in the USA region. This
produces soybean oil and soybean meal in GCAM, GLOBIOM, and GTAP, and vegetable oil
and oil crop meal in ADAGE. In the GCAM, GLOBIOM, and GTAP results, additional soybean
oil is used for fuel production in the USA region. In the ADAGE results, some additional
vegetable oil is used for fuel production in the USA region, and additional soybean is also used
directly for fuel production. In the GCAM results, the additional soybean meal produced in the
USA region largely displaces corn for domestic feed use. We observe a similar trend in the
ADAGE results, where oil crop meal displaces corn for feed use in the USA region. In GTAP,
the additional soybean meal produced in the USA region displaces other oil crop meal for
domestic feed use. By contrast, all of the additional soybean meal produced in the USA region in
the GLOBIOM results is exported; this increase in USA soybean meal exports in turn depresses
non-USA production of feed crops, including soybeans. However, USA exports of DDG
decrease and more DDG is consumed in the USA region, displacing corn for feed use. In the
USA region, ADAGE, GCAM, and GLOBIOM results show only minimal impacts on food end
uses. In contrast, the GTAP results show a reduction in soybean oil for food use and no increases
in other types of crops for food use, implying a net reduction in food consumption. GTAP results
also show a reduction in soybean oil for "other uses," which includes soybean oil that is
industrially processed into other products.204 "Other uses" of soybeans increases in the ADAGE
results; this represents additional soybean seeds needed to grow more soybeans.
Non-USA regions show different impacts than the USA region. In the non-USA regions,
the ADAGE results show an increase in soybean consumption for crushing, an increase in
vegetable oil and soybean consumption for fuel production, an increase in soybean consumption
for other uses (seed), and feed displacement of other crops with oil crop meal. In the GCAM,
GLOBIOM, and GTAP results, there is an increase in oilseed crushing to make vegetable oil,
including palm fruit (GCAM, GLOBIOM, and GTAP), rapeseed (GCAM and GLOBIOM), and
other oil crops (GCAM and GTAP). ADAGE represents only two oil crop commodities,
soybeans and "other oil crop." The ADAGE results show an increase in the consumption of the
aggregated other oil crop for crushing. In the GLOBIOM results, the increased palm fruit
crushing helps backfill for reduced soybean crushing, which is due to decreased soybean
production in non-USA regions. In the ADAGE, GCAM, and GTAP results, the increased palm
fruit, rapeseed, and other oil crop crushing is in addition to increased soybean crushing.
These results also show impacts on the food and feed markets in the non-USA region. In
both the GCAM and GLOBIOM results, other vegetable oils replace soybean oil to at least some
extent in the food market in non-USA regions.205 GLOBIOM results show an overall reduction
in food consumption in the non-USA regions. GCAM results show a small reduction in food
consumption, but the overall change is close to zero. These food market impacts are smaller than
204 The "other uses" of soybean oil in GTAP can include processing for food products, such as margarine or salad
dressing, whereas the food end use includes soybean oil used directly for food, such as cooking oil.
205 In GLOBIOM results, palm fruit oil replaces soybean oil. In GCAM results, a mix of palm fruit oil, rapeseed oil,
and other oil crop oil replaces soybean oil.
98
-------
the feed market impacts. The GLOBIOM results also show displacement of soybean oil with
palm fruit oil for other uses (e.g., industrial uses such as cosmetics production) and an overall
increase in feed consumption, primarily from corn, soybean meal, and other crops. GCAM and
GTAP results show displacement of crops with soybean meal and other oil crop meal in the feed
market. The degree of substitution among feed commodities and food commodities, particularly
in the non-USA regions, is an area of difference across the model results.
Figure 7.3-2: Difference in consumption by end use (million metric tons) in the soybean oil
biodiesel shock relative to the reference case in 2014 (GTAP) and 2030 (ADAGE, GCAM,
GLOBIOM)206
Consumption by End Use
Fuel Production
Feed
Food
Other Uses
Crushing
10
<
i/)
I
-5
10
s. i, l|iiL.. -
I
Commodity
¦ Corn
¦ DDG
I Soybean
¦ Soybean Meal
¦ Soybean Oil
¦ Vegetable Oil (Total)
¦ Palm Fruit
¦ Palm Fruit Meal
¦ Palm Fruit Oil
¦ Other Oil Crops
¦ Other Oil Crops Oil
¦ Other Oil Crops Meal
¦ Other Crops
¦ Energy Crops
H Other Grains
¦ Rice
Sugar Crops
H Wheat
-5
7.4 Trade of Agricultural Commodities
As discussed in Section 3.1.6, ADAGE, GCAM, GLOBIOM, and GTAP all specify
commodity trade in somewhat different ways. From a theoretical perspective, we would expect
this to be relevant to a soybean oil biodiesel consumption shock scenario in several ways
analogous to those observed for corn ethanol in Section 6.4. Model results related to trade in
soybeans and other crops would be expected to vary by model. In addition, the assumed
elasticity of competition and degree of assumed fungibility between vegetable oils varies across
these modeling frameworks and would be expected to produce somewhat different results across
the models. Another consideration unique to soybean oil biodiesel scenarios is the treatment of
soybean meal trade.
2116 Results are shown in million metric tons of each feedstock. Because soybeans contain 19 percent oil, 10 million
metric tons of soybeans is equivalent to 1.9 million metric tons of soybean oil. ADAGE does not explicitly track
soybean oil or soybean meal, and those are included in "Other Oil Crops Oil" and "Other Oil Crops Meal,"
respectively.
99
-------
Figure 7.4-1: Difference in U.S. net exports of crops and secondary agricultural products
(million metric tons) in the soybean oil biodiesel shock relative to the reference case in 2030
and 2050 (ADAGE, GCAM, GLOBIOM) and 2014 (GTAP)
-4
-6
ADAGE
Model / Year
GCAM
GLOBIOM
GTAP
Commodity
| Other Crops
| Wheat
Vegetable Oil (Other)
| Meal (Other)
| Other Oil Crops
Vegetable Oil (Soy)
| Meal (Soy)
| Soybean
¦ DDG
Corn
-10
-12
2030 2050 2030 2050
2030 2050
2014
In ADAGE, of the additional soybean oil biodiesel produced in the USA region, a
sizeable portion is sourced from shifting cropland from corn production to soybean production.
Reduced corn production coincides with reduced use of corn for livestock feed in the USA
region, which is backfilled with the additional oilseed meal available in the soybean oil biodiesel
shock scenario. This results in relatively little change in U.S. net exports of agricultural goods in
ADAGE.
In GCAM, the USA region increases gross imports of soybean oil and decreases gross
exports of whole soybeans in order to meet the soybean oil biodiesel shock targets. There is a
smaller (relative to ADAGE) effect on crop production for non-soybean crops in the USA
region, so the additional soybean meal produced to meet the shock is not needed to backfill
deficits in livestock feed demand. A relatively small portion of the shock in GCAM (compared to
ADAGE) is met through crop shifting in the USA region, so livestock feed demand met by corn
and other crops is less affected by the soybean oil biodiesel shock. This results in increased gross
exports of soybean meal from the USA region in the soybean oil biodiesel shock in GCAM.
100
-------
GLOBIOM does not represent energy commodities nor their trade, so all of the biodiesel
needed to meet the soybean oil biodiesel shock must be produced in the USA region in
GLOBIOM. Additionally, GLOBIOM restricts the amount of natural land that can be converted
to crop production, so the majority of the additional feedstock needed to meet the soybean oil
biodiesel shock is sourced from either switching cropland from production of other crops to
soybean production, or from changes in net trade of soybeans and soybean oil in the USA region.
This results in reduced gross exports of soybeans and soybean oil and increased gross imports
soybean oil in the USA region. Crop switching reduces production of other crops in the USA
region, most notably corn, which results in decreased gross exports of corn and DDG, and wheat,
which results in increased gross imports of wheat to meet demands for food.
The GTAP results include a reduction in soybean exports, but a larger increase in exports
of soybean meal and other oilseed meals for livestock feed. Unlike the other models, the GTAP
results include an overall increase in the mass of USA region net crop and secondary crop
product exports. Relative to the other model results, the GTAP results include a smaller
reduction in soybean oil and soybean exports. Instead of reduced exports, the GTAP results
include reduced domestic consumption of soybeans and soybean oil for feed, food and other non-
biofuel purposes.
7.5 Crop Yield
As was observed in Section 6.5 above regarding corn crop yield modeling results, the
four economic models included in this comparison exercise all have the ability to increase crop
yields in response to changes in crop price. However, while these models share some similar
theoretical underpinnings regarding the economic logic of crop yield response to price, their
mechanisms for simulating this response vary in structure. Further, these models represent
additional methods of crop intensification beyond the ability to invest resources to increase yield
per acre on existing cropland.
Reference case yield trends are also an important factor in understanding differences
across models. As shown in Figure 5.3-1, reference case soybean crop yield trends across the
four economic models are fairly similar in the historical periods of 2010 and 2015, though not
identical. However, for the three dynamic models, ADAGE, GCAM, and GLOBIOM, the trends
in reference case soybean yields diverge over time. Yields are calibrated to improve over time in
all three models however, reflecting a shared assumption that agricultural technologies will
continue to improve into the future. In reviewing the change in soybean yields in our shock
scenario relative to the reference case shown by these dynamic models, the reader should keep in
mind that yields are improving over time in both the USA and non-USA regions in both
scenarios as they do in the reference case.
As shown in Figure 7.1-2 above, crop intensification contributes to the sourcing of
soybean oil for the biodiesel shock to varying degrees across the models. In both of the biofuel
volume shock scenarios modeled for this exercise, we observe that the contributions from
intensification are a minority of the feedstock sourcing solution, accounting for 17 percent or less
of the feedstock required. Intensification is a part of each model solution to at least some degree
101
-------
however, and we can make some useful observations about how this effect is similar and
different across the models considered.
As shown in Figure 7.5-1, average USA region soybean yields increase in all four models
in response to the soybean oil biodiesel shock. One can compare these results with the reference
case yields presented in Figure 5.3-1 and observe that these improvements are generally less than
a 1 percent increase relative to reference case yields, though in the case of ADAGE, USA region
average yield does increase by 1.3 percent in 2030. While improvements may be larger in
particular growing regions, the average yield across the USA region is instructive in
understanding why intensification plays only a minor role in the sourcing of soybean oil for the
biodiesel shock. As a collective, these four models estimate the soybean oil biodiesel shock
modeled for this comparison does not induce much improvement in soybean yield relative to
reference case yields. This small observed change in USA region soybean yields is reasonable in
light of the crop price changes observed in these results. Figure 7.5-2 shows that the change in
soybean price is also small, less than 2 percent in 2030. As discussed above, crop price is the
primary driver of increased crop yields and intensification in general, and a small price change
would be expected to induce a small yield response as well. These changes in soybean price are
largely a function of the changes in soybean oil and soybean meal prices, shown in Figure 7.5-3.
Figure 7.5-1: Difference in soybean yield in the soybean oil biodiesel shock relative to the
reference case in 2014 (GTAP) and 2030 (ADAGE, GCAM, GLOBIOM, GTAP)
Model
Soybeans | BADAGE
Soy Shock(1BG)
2030
2014
0.043
0.040
< ro
i/> .-c
Z)
0.020
0.000
GCAM
GLOBIOM
GTAP
0.017
0.005
0.012
Looking at the non-USA regions results, we see smaller average soybean yield responses
from all four models. We observe more yield response in the ADAGE and GLOBIOM results
than in the GCAM or GTAP results. ADAGE estimates the largest non-USA regional soybean
production response of the four models, so it is perhaps unsurprising from that perspective that it
also shows the strongest non-USA yield response. Soybean oil biodiesel produced in South
America provides a substantial share of the shock in the ADAGE results. The increased demand
of this new biodiesel production creates greater investment in soybean yields in this region. The
GLOBIOM results tell a different story. In these results, soybean production declines outside the
102
-------
USA region overall. As discussed in Section 7.3 above, the decline in non-USA soybean
production is primarily a response to the influx of USA-produced soybean meal into global feed
markets. However, it is notable that GLOBIOM appears to use intensification as a method for
mitigating the reduction in soybean production, rather than a means of further boosting increased
production, as is the case in the ADAGE results. Conversely, yields increase very little in GTAP
and GCAM as these models appear to focus on other strategies for supplying the needed soybean
oil. However, the responses from all four models are fairly small. These results, again, appear
reasonable in light of the very small soybean price changes in the non-USA regions observed in
Figure 7.5-2.
Figure 7.5-2: Percent difference in commodity prices in the soybean oil biodiesel shock
relative to the reference case207
2%
1%
0%
2%
fk 1%
ADAGE
GCAM
I
8
GLOBIOM
GTAP
Commodity
¦ Corn
¦ Soybean
¦ Energy Crops
¦ Other Crops
¦ Other Grains
¦ Other Oil Crops
¦ Palm Fruit
¦ Rapeseed
¦ Rice
¦ Sugar Crops
¦ Wheat
2030
2030
2030
2014
2117 Average commodity prices for non-USA regions in GTAP results were not available for this exercise.
103
-------
Figure 7.5-3: Percent difference in coproduct prices in the soybean oil biodiesel shock
relative to the reference case208
GLOBIO
GTAP
Commodity
Residue
st Residue
r Oil Crops Meal
r Oil Crops Oit
Fruit Meal
Fruit Oil
:seed Meal
seed Oil
ean Meal
ean OH
In the three dynamic models, ADAGE, GCAM, and GLOBIOM, we see somewhat
similar patterns of yield change over time. Figure 7.5-4 shows that all four of the models
estimate an increase in soybean yield in 2030 as the shock reaches its peak, both in the USA and
non-USA regions though the magnitudes of these increases vary by region and model. By 2050,
this increase tapers off in all models in both the USA and non-USA regions as well. The
magnitude of this tapering varies as well and that magnitude appears to positively correlate to
some degree with the magnitude of the 2030 increase in yield. In general, this tapering effect
appears attributable to improving reference case soybean yields over time.
208 Average commodity prices for non-USA regions in GTAP results were not available for this exercise.
104
-------
Figure 7.5-4: Difference in soybean yield in the soybean oil biodiesel shock relative to the
reference case in 2014 (GTAP) and over time from 2020 to 2050 (ADAGE, GCAM,
GLOBIOM)
USA
Non-USA
Model
¦ ADAGE
¦ GCAM
¦ GLOBIOM
¦ GTAP
0.00
2014 2020 2030 2040 2050 2014 2020 2030 2040 2050
While the soybean crop yield change results may appear to be somewhat different across
models based on the figures presented, they are all relatively small increases when compared to
reference case soybean yields in each model The largest increase in soybean yields in 2030 is
seen in the ADAGE results in the USA region - about 1.3 percent - while soybean yield changes
in the other models and regions are all less than one percent in 2030. We can observe from these
results that the four economic models generally agree that, in the specific scenarios modeled for
this exercise, yields are not projected to improve substantially in response to the soybean oil
biodiesel shock. However, it is also notable that even these small changes in soybean yield are
responsible for a small but notable percentage of the additional soybean oil produced to meet the
shock.
From this exercise however, we cannot draw any firm conclusions from this yield
comparison regarding whether one method is better than the others. All four of the models seem
to behave reasonably in these yield results. Sensitivity analysis may reveal the degree to which
GHG emissions results change when the underlying assumptions about crop yield responsiveness
to price are changed. This may indicate areas for further research.
7.6 Land Use
The increased soybean production comes from a mix of cropland shifting from other
crops to soybeans, land use change from other land types to cropland, and changes in soybean
yield. As shown in Figure 7.6-1, soybean cropland in the USA region increases by 0.3 Mha in
GTAP (2014), 2.7 Mha in ADAGE (2030), 0.7 Mha in GCAM (2030), and 1.1 Mha in
GLOBIOM (2030). In the non-USA regions, soybean cropland increases by 0.02 to 2.1 Mha in
105
-------
GTAP, ADAGE, and GCAM, and decreases by 1.2 Mha in GLOBIOM. All of these models
show some amount of shifting of other crops to soybeans, but the amount of crop shifting varies.
In the GTAP and GLOBIOM results, most new soybean cropland in the USA region
comes from shifting of other crops. In the GLOBIOM results, there is a shift in the non-USA
region from soybean cropland to corn, wheat, other grains, and other crops, to make up for the
lost production of these crops in the USA region. In both models, the total cropland increases
more in non-USA regions than in the USA region. In the ADAGE results, there is some cropland
shifting in the USA and non-USA regions, but a larger net increase in cropland area than in
GTAP or GLOBIOM. In the GCAM results, even though there is much less new soybean
cropland than in ADAGE, there is a similar net increase in total new cropland (horizontal line in
Figure 7.6-1) because there is less cropland shifting than in ADAGE.
Figure 7.6-1: Difference in cropland area by crop type (million hectares) in the soybean oil
biodiesel shock relative to the reference case in 2014 (GTAP) and 2030 (ADAGE, GCAM,
GLOBIOM)209
Commodity
2030 2014 _ r
ADAGE GCAM _| GLOBIOM GTAP " Lorn
¦ Soybean
»¦ Sugar Crops
¦ Other Grains
¦ Other Oil Crops
¦ Other Crops
all's-
< < i < < < < i < <
CO CO CO CO CO CO CO CO
ID D D Z3 Z) ZJ ZJ ZJ
c c c c
O O o o
z z z z
The net increase in cropland causes changes in the area of other land types in each model
(Figure 7.6-3). As described in Sections 2 and 6.6, the type of land use change in each model
depends on the model structure and constraints. In ADAGE, most of the increase in cropland in
the USA region is coming from managed pasture. In contrast, non-USA regions show large
2119 Horizontal lines show the net change in cropland. Cropland area shown represents land cultivated for row crops
in ADAGE and GCAM and harvested area in GLOBIOM and GTAP. When a single unit of land is harvested
multiple times in a single year, the area is counted multiple times as "harvested area" but only a single time as
"cultivated area."
15
m
^ ru
u J=
° *>
_c ^
LO
>>
o
if)
-1
106
-------
decreases in managed and unmanaged forest. In the non-USA region, the soybean production
and land use change are occurring the Rest of Latin America region. In the Rest of Latin
America region in ADAGE, the model assumes that forest productivity decreases over time,
which impacts land prices, and causes the reduction of forest area. GCAM results show a
decrease in a mix of land types in both the USA and non-USA regions, with the largest impact
on unmanaged pasture, similar to the corn shock. In the GLOBIOM results, the area of other
arable land and managed forest decreases relative to the reference in non-USA regions. The
restriction on natural land conversion in GLOBIOM could drive the result that the new soybean
cropland in the USA region comes from crop shifting, rather than land use change.
In the GTAP results, there is very little change in land use in the USA region, but in the
non-USA regions, cropland increases and other arable land decreases. In GTAP, in the non-USA
regions cropland pasture is the main source for new harvested area (53 percent), followed by
pasture (30 percent), unharvested cropland (11 percent), increased multi-cropping (5 percent),
and forest (1 percent). Because GTAP only represents managed land, the results show no
conversion of unmanaged forest, grassland, or unmanaged pasture.
Each of the models has different assumptions about the carbon stock of different land
types in different regions. As shown in more detail in Section 7.7, the type and amount of land
converted and the carbon stock of the land types will factor into the emissions from land use
change.
107
-------
Figure 7.6-2: Difference in land use (million hectares) in the soybean oil biodiesel shock
relative to the reference case in 2014 (GTAP) and 2030 (ADAGE, GCAM, GLOBIOM)210
Lard Type
o
U1
ADAGE
2030
GCAM
GLOBIOM
2014
GTAP
1.0
0.5
0.0
-0.5
-1.0
I Grassland
I Managed Forest
Managed Pasture
I Other Arable Land
I Other Not Arable Land
shrubland
I Unmanaged Forest
I Unmanaged Pasture
I Cropland
USA
Non-USA
USA
Non-USA
USA
Non-USA
USA
Non-USA
Following the trends observed in the crop production results, the models show variation
in both the magnitude and location of land use change. As might be expected given their
differences in land competition structure and land categorization, these four models also present
diverse estimates regarding what types of land might be converted to cropland in response to
greater demand for soybean oil biodiesel, in particular the extent of forest loss. Some of these
differences appear to be related to where in the world the results show that cropland will expand.
The differences also appear to be attributable to differences in land conversion flexibility across
the models. These are areas for potential future sensitivity and uncertainty analysis.
7.7 Emissions
The modeled results of energy consumption, crop production, and land use change
described above come together in the modeled greenhouse gas emissions. As shown in Figure
7.7-1, the modeled GHG emissions over time vary by model.
2111 In Figure 6.6-2 and 7.6-2, "Cropland" area in GTAP represents land cultivated for row crops (calculated as the
change in harvested area minus the change in multicropping), while cropland pasture, and other unused cropland
have been reassigned to "Other Arable Land." This differs from Figure 5.2-1, in which cropland pasture and other
unused cropland are reported under the "Cropland" category.
-------
Figure 7.7-1: Difference in global greenhouse gas emissions in the soybean oil biodiesel
shock relative to the reference case211
GHG Emissions by Source
Model
Emission Source
ADAGE
GCAM
GLOBIOM
w 40
>
LUC S 20
^ 0
r-\
v. 40
Energy from $ 20
Fossil Fuels <->
¦*->
S 0
_ 40
>N
~oT
Crop Production 0 t J
+-»
s 0
>_ 40
>.
Livestock oj 20
Production u
S 0
40
>*
Other (Industrial w 2o
& Waste) u
g 0
2030 2040
Year
2030 2040
Year
2030 2040
Year
C02
CH4
N20
Other GHGs
Net GHG Emissions (All Represented Sources)
Model
20
ADAGE
GCAM
GLOBIOM
r^\
A
—"
X——
2020 2030 2040 2050
Year
2020 2030 2040 2050
Year
2020 2030 2040 2050
Year
® GTAP is not included in this figure because it does not represent emissions over time, and due to time
constraints, we do not have GTAP GHG emissions by gas for the source categories used in (his figure. For
comparison, for GTAP, in the soybean oil biodiesel scenario relative to the reference case (2014), LUC emissions =
1.1 Mt CO;C. fossil fuel combustion and industrial CO2 emissions = -5.5 Mt, and other GHGs emissions from all
covered sources = -0.70 Mt COie, of w hich NjO = 0.13 Mt CO;C. CH4 = -0.72 Mt CO2C. fluorinated gases = 0.01 Mt
CO?c. and other CO2 = -0.13 Mt CO;c: net total GHG emissions = -5.1 Mt CO;e. GREET is not included in this
figure because it does not represent scenario-based emissions over time. See Table 7.7-1 for carbon intensity values.
109
-------
Emissions from land use change show different trends in ADAGE, GCAM, and
GLOBIOM results, due primarily to two factors: variation in the type(s) of land use change
occurring relative to the reference case, and variation in the underlying carbon stock data sets
and assumptions used in each model. In the ADAGE results, land use change emissions are the
highest of the models shown here. These emissions peak in 2030 in ADAGE and are higher than
the reference case throughout the entire model period. In the ADAGE results, the non-USA
region has a large amount of forest converted to cropland. Because forests have a higher carbon
stock than other land types, the ADAGE results show high land use change emissions. In
addition, emissions continue after 2030 because the assumptions and structure in ADAGE make
it cost effective to continue to convert land after 2030.
In the GCAM and GLOBIOM results, land use change emissions estimates are higher
than the reference case from 2020 to 2040, peaking in 2030. From 2040-2050, emissions are
slightly lower than the reference case. Emissions in the GCAM results are higher than in the
GLOBIOM results. In the GCAM results, most of the land use change is coming from lower
carbon land types, such as pasture and grassland. However, some of the land use change is
attributable to reduced amounts of estimated future afforestation relative to the reference case.
Even though the amount of change in forest land is small compared to the amount of change in
other land types, the high carbon stocks of forest land leads to higher land use change emissions.
The GLOBIOM results have less forest conversion than ADAGE and GCAM, and therefore
lower land use change emissions, especially earlier in the modeled period.
The "Energy from Fossil Fuels" (or "fossil fuel emissions") category includes emissions
associated with producing biofuels (e.g., from consuming natural gas or electricity for process
energy), direct emissions associated with on-farm energy use to produce feedstock, and
transporting both biofuel feedstocks and finished fuels, as well as emissions from indirect
impacts on the energy sector, including displaced diesel use for transportation that is replaced by
soybean biodiesel. In the soybean oil biodiesel results, ADAGE and GCAM show lower fossil
fuel emissions than in the reference case.212 In these results, the reduction in emissions from
fossil fuels becomes larger until 2030. From 2030-2050, fossil fuel emissions in the GCAM
results are relatively constant. In the ADAGE results, from 2030-2050 the reduction in emissions
becomes smaller, but emissions stay lower than in the reference case. As shown in Section 7.2,
refined oil consumption decreases in the soybean oil biodiesel shock scenario relative to the
reference case. Globally, the refined oil consumption decreases more in the ADAGE results than
the GCAM results. However, ADAGE results show a larger increase in global natural gas
consumption than the GCAM results, and an increase in coal consumption, rather than the
decrease seen in the GCAM results. The higher consumption of natural gas and coal in the
ADAGE results leads to a lower reduction in fossil fuel emissions in the ADAGE results than the
GCAM results.
Crop production emissions are higher than the reference case in the ADAGE, GCAM,
and GLOBIOM results, with GCAM results showing the largest increase. Changes in crop
production emissions relative to the reference case are due to changes in the types and quantities
of crops grown in the models, and primarily come from changes in N2O emissions, driven by
both increased fertilizer use and direct nitrogen fixation by soybeans. As shown in Section 7.3,
212 Emissions from "Energy from fossil fuels" are not reported by GLOBIOM.
110
-------
the ADAGE, GCAM, and GLOBIOM results all show increases in soybean production. These
results also show increased production of palm fruit and other oil crops. ADAGE and GCAM
results show a decrease in corn production, whereas GLOBIOM results show a shift in corn
production from the USA region to the non-USA regions. The crop production emissions are
small in all of these model results. Emissions peak in 2030 in the GCAM and GLOBIOM results,
and in 2040 in the ADAGE results, and then decrease until 2050. The change in emissions
relative to the reference case from the livestock sector and from industrial and waste
management sectors is very small.
The total change in GHG emissions across all sources over time varies across the models
(Figure 7.7-1). The ADAGE results show higher emissions than in the reference case from 2020-
2050, which is dominated by CO2 emissions from land use change. In the GCAM results, GHG
emissions are higher than in the reference case from 2020-2030 and lower than the reference
case from 2035-2050, because the CO2 emissions from land use change decline rapidly after
2030. In the GLOBIOM results, emissions are higher than in the reference case from 2020-2050,
and are dominated by CO2 emissions from land use change.
There are a few commonalities across the ADAGE, GCAM, and GLOBIOM results of
emissions over time. All of these model results show small but positive emissions from crop
production relative to the reference case. The model results also all show very small changes in
emissions from livestock production, waste management, and industry. The GCAM and ADAGE
results both show lower emissions from fossil fuel than the reference case, but there are
differences in the amount of fossil fuel emissions reduction. Future research could explore the
factors that determine the extent of refined oil displacement in each model through sensitivity
analysis. Additionally, there are large differences across the model results in the amount of land
use change emissions, due to differences in both the types of land converted and the carbon stock
assumptions. A sensitivity analysis of the carbon stock assumptions in GCAM is shown in
Section 9.2 below, and a sensitivity analysis of the land conversion elasticities in ADAGE is
shown in Section 9.3. Future research could focus on the impact of carbon stock assumptions in
other models, or on other model parameters that determine the types of land converted.
As explained in Section 6.7, we calculated a CI for each category of emissions, in
kgC02eq/MMBTU (Table 7.7-1). We also consider CI results from GREET. As explained in
Section 6.7, the models report emissions from different sectors. Models are divided between
those frameworks with energy markets (in the left side columns) and models without energy
markets (in the right side columns). This division is made to reflect important differences in the
sectors represented and the difficulty of direct comparability between models on the left with
models on the right. ADAGE, GCAM, and GTAP include global emissions from every economic
sector, including indirect, market-mediated impacts. GREET includes detailed emissions
assumptions from fuel production, transport, and use, but, as it is not a consequential model, it
does not estimate the net change in GHG emissions resulting from a change in biofuel
consumption. Rather it estimates the emissions directly attributable to the biofuel supply chain.
GLOBIOM does not include any energy sector emissions but does include market impacts on
crop production and the livestock sector.
Ill
-------
Because of the differences outlined above, it would be inappropriate to compare all of the
emissions estimates across all of the models, but we can make several meaningful comparisons.
Results from the three models with energy markets (ADAGE, GCAM, GTAP) can be directly
compared, with the caveat that GTAP is representing 2014 while the other models are
representing a 2020-2050 scenario. Furthermore, we can compare the land use change emissions
estimates for all of the models, as GREET uses a consequential approach for this category of
emissions, again with proper caveats about temporal differences. We can also compare crop
production and livestock sector emissions estimates from ADAGE, GCAM and GLOBIOM. In
the table below, we report emissions from "Agriculture, forestry and land use" for all five
models as the sum of emissions from these stages; however, the GREET estimate for this
aggregate category is not directly comparable with the other models for reasons discussed below.
Like in the corn ethanol shocks, energy sector emissions have a large impact on the CI of
soybean oil biodiesel in the ADAGE, GCAM, and GTAP results. The energy sector CI is higher
(less negative) for the ADAGE results than for the GCAM and GTAP results, which is consistent
with the smaller emissions reduction from fossil fuels over time shown in Figure 7.7-1,
particularly in the later model years. GREET reports the CI from fuel production and
transportation but does not consider indirect impacts on the energy sector, such as the energy
rebound effects shown in Section 7.2. The fuel production and transportation CI in the GREET
results is based on the amount of process energy needed for soybean oil biodiesel production as
well as the amount of energy needed to transport the feedstock and the fuel. This is why we use
the label "Energy Sector" for the first row in Table 7.7-1 for the three models with energy
markets, but the label "Biofuel Production" for this row for GREET.
112
-------
Table 7.7-1: Carbon intensity of soybean oil biodiesel (kgCCheq/MMBTU) calculated using
emissions reported by each model213
Models with Energy Markets
Models without Energy Markets
ADAGE
GCAM
GTAP
GLOBIOM
GREET
Sector/stage-
specific
emissions
Energy
from Fossil
Fuels
-28
-40
-46
Biofuel Production
X
13
Crop
Production
7
21
-6
Crop Production
11
X
Feedstock
Production
X
9
Livestock
Sector
0.7
-1.3
Livestock Sector
3
X
Other
1
0
Fuel Use
X
0.4
Land Use
Change
295
62
10
Land Use Change
23
10
Totals
Agriculture,
forestry,
and land
use
303
82
4
Agriculture,
forestry, and land
use
38
19
Global
GHG
Impact
276
42
-42
Global GHG Impact
X
X
Supply
Chain GHG
Emissions
X
X
X
Supply Chain GHG
Emissions
X
32
The ADAGE, GCAM, and GLOBIOM results show a range of CI from crop production.
The crop production CI from the GCAM results is higher than the other models, consistent with
the higher emissions over time in the GCAM results relative to the ADAGE and GLOBIOM
results. GREET's feedstock production CI is based on the energy and chemical inputs required to
produce the amount of soybean oil needed for 1 MMBTU of biodiesel. Unlike the other models,
this value does not consider indirect impacts on the production of other types of crops. Livestock
and other sectors (including waste management and other industrial sectors) have only minor
impacts on the overall CI in ADAGE, GCAM, and GLOBIOM.
For the GTAP results, we have estimates of non-C02 emissions by greenhouse gas, but
we do not have these emissions disaggregated by sector or lifecycle stage. The largest change, by
213 "X" means that the model does not report that category. For GTAP, emissions from crop production, the
livestock sector, and "other" are reported as an aggregated value of non-LUC, non-fossil fuel emissions. Negative
values for ADAGE, GCAM, GTAP, and GLOBIOM mean that emissions are lower than the reference case, whereas
positive values mean the emissions are higher than the reference case. For further discussion of how to interpret
positive and negative values, see Section 6.7.
113
-------
gas, is a decrease in CH4 emissions. We believe the bulk of the changes in these emissions are
associated with changes livestock CH4, but more work would be needed to confirm our intuition.
In Table 7.7-1, we report the aggregated non-CCh emissions estimate from GTAP across three
rows combining Crop Production, Livestock Sector and Other. GTAP shows a negative CI in this
aggregated category. We would need to do more research to understand why these emissions are
lower than estimates from the other models.
Land use change emissions are reported across all the models, and the CI results show
wide differences, consistent with the large differences in emissions shown in Figure 7.7-1. As
explained in Section 7.6, ADAGE results show conversion of forest land to cropland to grow
soybeans in non-USA regions, which results in a high estimated LUC CI. In contrast, GTAP
results show very little land use change, and therefore this model estimates a low LUC CI. Here
again, GREET's LUC CI is based on a GTAP run214 using a different shock size (0.812 billion
gallons of soybean oil biodiesel) using a 2004 baseline where around 13 percent of crop land
cover demand comes from forest land, and the remainder comes from land previously having
been pastureland.215
We can compare "Agriculture, forestry and land use change emissions" across four of the
models (ADAGE, GCAM, GLOBIOM, GTAP). For GTAP, we include the non-CCh emissions
in this category. For this category, the ADAGE results include the highest emissions, followed
by GCAM. These differences are driven by the land use change emissions.
The total global CI can be compared across ADAGE, GCAM, and GTAP, because all of
these models represent the same sectors and include market impacts. The results from these
models show a range in soybean oil biodiesel CI, primarily due to differences in the land use
change CI. For GLOBIOM and GREET, a total global CI cannot be calculated from the model
results because these models do not include all the relevant sectors and/or do not include all the
relevant market impacts. For GREET, we calculate the total supply chain CI. This is a
fundamentally different metric than the other models' CIs, since GREET primarily uses an
attributional approach to lifecycle analysis rather than a consequential approach. This value does
not include any displacement of fossil fuel consumption that would occur from the increased
consumption of biofuels.216
7.8 Summary of Soybean Oil Biodiesel Estimates
Section 7 compares and contrasts the soybean oil biodiesel modeling estimates from
ADAGE, GCAM, GLOBIOM, GREET, and GTAP produced for this exercise. These models
source the soybean oil biodiesel required to meet the assumed shock in different ways in these
214 We present the default soybean oil biodiesel run from GREET's LUC CCLUB tool here, referred to as "Soy
Biodiesel CARB Case 8"
215 Chen, Rui, Zhangcai Qin, Jeongwoo Han, Michael Wang, Farzad Taheripour, Wallace Tyner, Don O'Connor,
and James Duffield. 2018. "Life Cycle Energy and Greenhouse Gas Emission Effects of Biodiesel in the United
States with Induced Land Use Change Impacts." Bioresource Technology 251 (March): 249-58.
https://doi.Org/10.1016/i.biortech.2017.12.031.
216 GREET's biodiesel CI estimates are often compared with GREET CI estimates for diesel to derive a GHG
percent reduction relative to diesel. In our 2010 RFS analysis, we similarly compared biodiesel CI estimates from
models that do not include energy markets with a CI estimate for diesel to calculate a percent reduction in emissions.
114
-------
results. Some models rely primarily on crushing of new soybean production to produce
additional soybean oil feedstock. Other models rely primarily on diversion of soybean oil from
other uses. Some models also show a contribution from reduced soybean oil biodiesel
consumption in non-USA regions. In addition, the model results show differences in how much
of the new soybean oil biodiesel is produced in the USA region versus the non-USA regions.
Because of these differences in sourcing strategy, the model results differ regarding the amount
and location of soybean oil production, vegetable oil and biodiesel trade, and land use change
impacts of the shock. Notably, the amount and location of land use change, and the types of land
converted to cropland, differ substantially across the range of model results. The model results
also show differences in the impact on the food and feed markets, and different amounts of
displacement of palm oil or other oils. The model results also have some notable similarities.
ADAGE, GCAM, GLOBIOM, and GTAP results all show a small amount of crop yield
intensification. The models which explicitly include the energy sector, ADAGE, GCAM, and
GTAP, all show a decrease in refined oil consumption in the USA region in their results, and an
increase in non-USA regions. But there are differences across these models in the total global
displacement of refined oil. These factors all contribute to differences in the estimated GHG
emissions and CI of soybean oil biodiesel across the models, with the differences in land use
change emissions having the greatest impact on estimated CI.
The previous sections also highlight potential areas for future research. Sensitivity
analysis could test the impact of different degrees of substitution in feed and food markets.
Further research and sensitivity analysis could also seek to better understand the parameters that
influence land conversion to cropland. Furthermore, research and sensitivity analysis could seek
to better understand why model results show a range in the reduction of refined oil consumption.
These are only a few examples of the many research areas that could help us to understand what
is driving the variation in estimates across models.
Alternative Scenarios and Model Sensitivity Analysis
8 Alternative Volume Scenarios
To determine whether and how GHG emissions estimates from these models may vary
based on the volume of biofuels assumed, we ran alternative volume scenarios through the
models. The scenarios included half of the original soybean oil biodiesel shock (decreased to 500
million gallons) and a combined scenario in which both soybean oil biodiesel and corn ethanol
consumption are each increased by 1 billion gallons simultaneously. These new volume
scenarios were performed in ADAGE, GCAM, GLOBIOM, and GTAP using the same methods
for the core corn ethanol and soybean oil biodiesel scenarios. The alternative shock size was
chosen to compare how each model functions, and they are not necessarily meant to represent
realistic biofuel shock sizes.
8.1 Soybean Oil Biodiesel 500 Million Gallons (MG) Scenario
The 500 MG soybean oil biodiesel shock results generally indicate a linear relationship
between shock size and most output parameters. ADAGE, GCAM, and GTAP show a high
degree of linearity between volume shock assumptions and output values, with scenario changes
115
-------
from the reference case for the 500 MG soybean oil biodiesel shock generally being half the size
of those from the 1 BG shock. The GLOBIOM results show more nonlinear variability in output
values, but these nonlinearities tend to be quantitatively minor. To examine these questions of
model response linearity and for clarity of presentation, the 500 MG soybean oil biodiesel shock
has been normalized to show impacts per 1 billion gallons of soybean oil biodiesel in the results
presented in this section.
8.1.1 Energy Market Impacts
The models that include energy market impacts, ADAGE, GCAM, and GTAP, show a
linear relationship between shock size and global energy consumption. The size of the energy
sector impacts, expressed in quad BTUs per billion gallons (of shocked biodiesel), are generally
equal across the 500 MG and 1 BG soybean oil biodiesel scenarios, as illustrated in Figure 8.1.1-
1. GLOBIOM does not represent the energy sector and as such was not included in this section
of the analysis.
Figure 8.1.1-1: Difference in global energy consumption (Quad BTUs per BG of shocked
soybean oil biodiesel consumption) in the 500 MG and 1 BG soybean oil biodiesel shocks
relative to the reference case in 2030 (ADAGE and GCAM) and 2014 (GTAP)
(J
<
O
<
0.15
0.10
0.05
0.00
-0.05
Commodity
H Biodiesel from Other Oil Crop Oil
I Biodiesel from Soybean Oil
I Electricity
¦ Ethanol from Corn
H Ethanol from Sugar Crops
I Natural Gas
I Refined Oil
H Coal
<
u
0.10
0.05
0.00
-0.05
0.15
<3* CL
H <
O I—
00 (J
0.10
0.05
T3
TO
D
o
0.00
-0.05
Soy Shock (500MG)
Soy Shock (1BG)
116
-------
8.1.2 Crop production and consumption
Similar to energy consumption, ADAGE and GCAM show a generally linear relationship
between shock size and global commodity production impacts in the 500 MG soybean oil
biodiesel shock. GTAP also shows a generally linear relationship between commodity
production and shock size. GLOBIOM results have slight differences in production of corn and
soy between the 500 MG and 1 BG soybean oil biodiesel shocks, but these differences are minor.
Global commodity consumption by end use indicates a generally linear relationship with
respect to shock size across ADAGE, GCAM, and GLOBIOM in the year 2030, and there are
not any notable changes between the 500 MG and 1 BG soybean oil biodiesel scenarios. GTAP
also shows a generally linear relationship between global commodity consumption and shock
sizes in 2014.
However, in the 2050 time step, GLOBIOM results show nonlinearities in the global
crushing of palm fruit and the consumption of sugar crops and other crops for feed, with the 500
MG shock showing higher consumption per billion gallons.217 The nonlinearity for palm fruit is
attributable to the commodity substitution dynamics of GLOBIOM. As a commodity becomes
scarcer on the global market (soybean oil in this case), the price of that commodity increases and
there is increasing incentive to substitute less expensive alternatives (palm oil in this case).
However, that substitution becomes more expensive, i.e., the price of the substitute good
increases as greater quantities of the substituted product are demanded. In both the 500 MG and
1 BG soybean oil biodiesel shocks, increasing U.S. demand for soybean oil to produce biodiesel
leads to lower availability of soybean oil in other countries and higher prices for soybean oil and
soybeans. This shortfall is partly addressed with increased palm oil supply from Southeast Asia.
However, substitution of palm oil for soybean oil grows more costly per unit as demand rises.
For this reason, this substitution effect is less pronounced in the 1 BG case than in the 500 MG
case, where the total volume of additional palm oil demanded is smaller.
Regarding feed crops, the economic dynamics at play are somewhat similar. The 500 MG
soybean oil biodiesel shock generates less additional soybean meal than the 1 BG case, and U.S.
soybean meal prices are depressed by a smaller amount. This smaller price depression leads to a
less than proportional increase of the use of the meal as livestock feed abroad. The nonlinear
change in consumption of other feed products in the 500 MG case is related to the fact that,
unlike the other models considered in this exercise, GLOBIOM explicitly accounts for the need
for animal feed diets to be balanced nutritionally. Increasing consumption of one feed product, in
this case soybean meal, means that consumption of other complementary feed products must also
increase to maintain nutritional balance for livestock. In the 500 MG soybean oil biodiesel case
relative to the 1 BG case, the smaller increase in Non-USA consumption of soybean meal,
relative to the size of the shock, means that increased consumption of these other feed products is
also proportionally smaller. Figure 8.1.2-1 illustrates the differences in global commodity
217 In the 500 MG scenario results from GLOBIOM, consumption of palm fruit for crushing was 6.8 Mt per BG,
consumption of sugar crops for feed was 1.2 Mt per BG, and consumption of other crops for feed was 1.8 Mt per
BG. In the 1 BG scenario, consumption of palm fruit for crushing was 5.3 Mt per BG, consumption of sugar crops
for feed was 0.8 Mt per BG, and consumption of other crops for feed was 0.6 Mt per BG.
117
-------
consumption by end use in the 2050 time step for ADAGE, GCAM, and GLOBIOM, as well as
the 2014 time step for GTAP.
Figure 8.1.2-1: Difference in global commodity consumption by end use (Mt per BG of
shocked soybean oil biodiesel consumption) in the 500 MG and 1 BG soybean oil biodiesel
scenarios relative to the reference case in 2050 (ADAGE, GCAM, and GLOBIOM) and
2014 (GTAP)
e> 10
O Q.
Fuel Production
Feed
World
Food
Other Uses
Crushing
"II
u 10
m
£< CD
O LJ Q_
C\J^ ^
5 13 10
O ?
CD
O °-
o
Commodity
¦ Corn
¦ DDG
¦ Soybean
¦ Soybean Meal
¦ Soybean Oil
¦ Vegetable Oil (Total)
¦ Palm Fruit
¦ Palm Fruit Meal
¦ Palm Fruit Oil
¦ Other Oil Crops
¦ Other Oil Crops Oil
¦ Other Oil Crops Meal
¦ Other Crops
¦ Energy Crops
¦ Other Grains
¦ Rice
¦ Sugar Crops
¦ Wheat
Soy ShockSoy ShockSoy ShockSoy ShockSoy Shock Soy ShockSoy ShockSoy ShockSoy ShockSoy Shock
(500MG) (1BG) (500MG) (1BG) (S00MG) (1BG) (500MG) (1BG) (500MG) (1BG)
8.1.3 Land Use
The global land use change by land cover type in the 500 MG soybean oil biodiesel shock
has a relatively linear relationship in ADAGE, GCAM, and GTAP results, as seen in Figure
8.1.3-1. However, GLOBIOM results show an increase in global land converting to pasture per
billion gallons in the 500 MG shock (0.383 Mha per BG) relative to the 1 BG shock (0.233 Mha
per BG). Soybean meal and pasture are both livestock inputs and they are in competition with
each other to some extent to provide nutrition to livestock. When soybean meal prices fall as a
result of a supply influx, as occurs in the soybean oil biodiesel shocks, this reduces the
competitiveness of alternative forms of livestock nutrition, i.e., grazing on pasture land. In the
smaller 500 MG shock, soybean meal prices decrease less, which improves the competitiveness
of pasture relative to the larger 1 BG shock. As overall livestock demand rises in both of the
soybean oil biodiesel scenarios, pasture therefore captures a larger share of the nutrition supply
in the scenario where it is more competitive, i.e., the 500 MG shock. GLOBIOM results also
show a larger decrease in other arable land per billion gallons in the 500 MG shock (-0.964 Mha
per BG) compared to the 1 BG shock (-0.778 Mha per BG).
118
-------
Figure 8.1.3-1: Difference in land use (Mha per BG of shocked soybean oil biodiesel
consumption) for the 500 MG and 1 BG soybean oil biodiesel shocks relative to the
reference case in 2030 (ADAGE, GCAM, and GLOBIOM) and 2014 (GTAP)
2014
GTAP
Land Cover Type
I Cropland
I Forest
I Grassland
¦ Other Arable Land
I Other Non-arable Land
I Pasture
Soy Shock Soy Shock Soy Shock Soy Shock Soy Shock Soy Shock
(500MG) (1BG) (500MG) (1BG) (500MG) (1BG)
Soy Shock Soy Shock
(500MG) (1BG)
The GLOBIOM 500 MG results also show differences in where LUC occurs relative to
the 1 BG results (Figure 8.1.3-2). In the USA region, GLOBIOM results show a larger increase
in land conversion to pasture per billion gallon in the 500 MG scenario (0.325 Mha per BG) in
comparison to the 1 BG scenario (0.110 Mha per BG) and a larger decrease in other arable land
(-0.897 Mha per BG) compared to the 1 BG scenario (-0.666 Mha per BG). Forest has a smaller
decrease in land conversion in the 500 MG scenario (-0.145 Mha per BG) compared to the 1 BG
scenario (-0.21 Mha per BG) in GLOBIOM as well. In the non-USA regions, the 500 MG
GLOBIOM results show a greater increase in pasture and a greater decrease in other arable land
per billion gallons than the 1 BG results.
119
-------
Figure 8.1.3-2: Difference in land use by region (Mha per BG of shocked soybean oil
biodiesel consumption) for the 500 MG and 1 BG soybean oil biodiesel shocks relative to
the reference case in 2030 (ADAGE, GCAM, and GLOBIOM) and 2014 (GTAP)
Land Cover Type
ADAGE GCAM GLOBIOM GTAP ¦ Other Arable Land
I Other Non-arable Land
H Pasture
1.0 ¦ Grassland
I Forest
Soy Shock Soy Shock Soy Shock Soy Shock Soy Shock Soy Shock Soy Shock Soy Shock
(500MG) (1BG) (500MG) (1BG) (500MG) (1BG) (500MG) (1BG)
8.1.4 Emissions
In the 500 MG scenarios, ADAGE, GCAM, and GTAP results indicate a relatively linear
relationship between shock size and global GHG emissions. These models estimate a slight
percentage decrease in total cumulative GHG emissions in the 500 MG scenarios relative to the 1
BG scenarios, but these results are quantitatively minor (Table 8.1.4-1). In comparison to
ADAGE, GCAM, and GTAP, GLOBIOM results estimate a larger percentage decrease in global
cumulative emissions in the 500 MG soybean oil biodiesel scenario compared to the 1 BG
soybean oil biodiesel scenario.
Table 8.1.4-1: Percent difference in global accumulated GHG emissions per billion gallons
of soybean oil biodiesel shock in the 500 MG shock scenario relative to the 1 BG shock
scenario
ADAGE
GCAM
GLOBIOM
GTAP
Percent Difference (TOTAL GHG)
-2%
-2%
-24%
-6%
Percent Difference (LUC Only)
0%
-2%
-21%
-1%
120
-------
When examining global GHGs over time, in the 500 MG scenario, GLOBIOM results
estimate an increase in N2O emissions in 2050 compared to the 1 BG scenario (Figure 8.1.4-1).
While the accumulated GHGs in ADAGE remain relatively linear by the year 2050, when
examining emissions over time, ADAGE has more variability in each time step. This includes a
smaller increase in CO2 emissions in the year 2040 and conversely a larger increase in the year
2045 for the 500 MG shock in comparison to the 1 BG shock. GCAM indicates a generally linear
relationship between both the accumulated GHGs and the emissions over time.
Figure 8.1.4-1: Difference in global GHG emissions (lYItCCheq per BG of shocked soybean
oil biodiesel consumption) in the 500 MG and 1 BG soybean oil biodiesel shocks relative to
the reference case from 2020 through 2050m
g 40
LU 0
CD CL
< 0) 20
O CM
< O
u
ID
CO
i_
c\j
O
u
0
20
10
£ 10
a;
Q.
a
O ^
—1 O
CD CJ
o
CM
o
CNJ
Soy Shock (500MGL
World
Soy Shock (1BG)
GHG Type
¦ Other GHG
¦ HFC
¦ CH4
¦ N20
¦ C02
in
c\j
o
c\j
o in
ro m
o o
C\J CM
o in o
<3- «sj- m
000
CM CM CM
o in
CM CM
o o
CM CM
o
CO
o
CM
in o in
ro <3- <3-
000
CM CM CM
O
m
o
CM
Global GHG emissions by source also show a linear relationship over time. The patterns
between the 500 MG and 1 BG shocks tend to mirror each other in each model. However, in the
500 MG scenario, GLOBIOM shows a decrease in livestock production emissions in the year
2050 compared to the slight increase in livestock emissions in the 1 BG scenario.
8.1.5 Summary
Overall, the soybean oil biodiesel 500 MG shock results indicate a linear effect between
shock size and most output values for ADAGE, GCAM, and GTAP results. GLOBIOM results
show somewhat more nonlinearity with shock size for certain output parameters, which leads to
differences in the GHG emissions. But the nonlinearities observed in the GLOBIOM results tend
to be minor. GLOBIOM's global commodity consumption by end use estimates an increase in
palm fruit used for crushing per billion gallon, as well as an increase in sugar crops and other
218 GTAP is not included in this figure as it doesn't represent emissions over time. See Table for carbon intensity
values.
121
-------
crops used for feed in the 500 MG scenario relative to the 1 BG scenario. The most notable
difference in land use change is the increase in pasture and decrease in other arable land in the
non-USA region in the GLOBIOM 500 MG results relative to the 1 BG results. GLOBIOM also
estimated a decrease in global CO2 emissions in the 500 MG soybean oil biodiesel shock,
compared to the 1 BG shock. However, we can observe that, across ADAGE, GCAM, and
GTAP, the size of the biofuel shock does not appear to cause significant changes in the modeled
global GHG emissions results.
8.2 Combined Shock Volumes
In addition to the 500 MG soybean oil biodiesel scenario, a combined shock of 1 billion
gallons each of soybean oil biodiesel and corn ethanol was also performed. In the core scenarios
for corn ethanol and soybean oil biodiesel, presented in Section 6 and Section 7 respectively,
some models estimated an inverse relationship between corn and soybean production. For
instance, when we shocked the model with 1 BG of corn ethanol, soybean commodity production
would go down, as seen in Figure 6.3-1. However, historically volumes of corn ethanol and
soybean oil biodiesel consumption have grown alongside one another, though often at somewhat
different annual rates. This has resulted historically in simultaneous increases in demand for corn
starch and soybean oil from the biofuel sector. It is therefore worth considering whether modeled
LUC and emissions impacts in particular might differ from our core scenario results if the
models conduct a scenario where both corn ethanol and soybean oil biodiesel consumption in the
USA are assumed to increase simultaneously. The combined scenario was performed to examine
what would happen if both biofuels shocked the models.
There are a few general hypotheses regarding what impact such a combined volume
shock scenario might have relative to our core scenarios. One hypothesis is that the impacts will
be "additive", that is, the results will be approximately the sum of adding together impacts from
the corn ethanol and soybean oil biodiesel core scenarios. Another hypothesis is that increasing
demand for both fuels at the same time will create greater stress on the agricultural system than
either core scenario in isolation, since it will not be possible to simply decrease USA soybean
production in response to greater corn ethanol demand, or decrease USA corn production in
response to soybean oil biodiesel demand, as is estimated to occur in most of the core scenario
results. Such a result would be expected to create greater-than-additive modeled impacts on
LUC, crop production, and the resulting GHG emissions. The third hypothesis is that there could
be a counterbalance within variables with the combined shock, where the increase in one
variable could decrease another. We find the land and emissions estimates in the combined
scenario have a mostly additive effect in which modeling results in combined scenario are
generally equal in magnitude to the sum of the individual corn ethanol (1 BG) and soybean oil
biodiesel (1 BG) core scenarios.
8.2.1 Land Use
The combined scenario provides insight into how each of the models account for the
impact on other crop commodities when both corn ethanol and soybean oil biodiesel
consumption are increased simultaneously. Figures 8.2.1-1 and 8.2.1-2 illustrate the USA and
non-USA regional land use change by crop commodity in the years 2030 (ADAGE, GCAM, and
122
-------
GLOBIOM) and 2014 (GTAP). The 1 BG corn ethanol and 1 BG soybean oil biodiesel core
scenarios are stacked together in the left-hand columns of each commodity type with a line
indicating the sum of the two scenarios, and the combined scenario is on the right-hand side of
the columns with the line indicating the total from this scenario. To the extent the results of the
combined scenario are additive, we would expect the pair of lines for each crop commodity to be
similar in magnitude.
The figures below do in fact show each model estimates a generally additive relationship
between the corn and soy shocks, meaning that the sum of the impact magnitudes from the core
scenarios generally equals the total magnitude of the combined scenario. The most notable
difference is that GLOBIOM has a slightly larger increase in USA regional soybean land cover
as well as a slightly larger decrease in the non-USA regional soybean land cover in the combined
shock.219 Interestingly, we do not observe any notable changes in land cover for any other crop
commodities.
Figure 8.2.1-1: Difference in cropland area by crop in the corn ethanol shock, soybean oil
biodiesel shock, and combined shock relative to the reference case in the USA region in
2030 (ADAGE, GCAM, and GLOBIOM) and 2014 (GTAP)
USA
2.474
2.456
Other Crop Commodities
Scenario
I Corn & Soy Shock (1BG each)
I Corn Shock (1BG)
¦ Soy Shock (1BG)
2 «, 2
0.523 0.525
1.382 1.383
0
-0.214 -0.215
GLOBIOM
Mha
o ro
0.231 0.225
0.877 0.989
-1.037^^^1 -1.090®
GTAP
Mha
o ro
0.210 0.209
0.197 0.197
-0.296 -0.295
Corn Shock & Soy Combined Corn &
Shock (1BG) Soy Shock (1BG
each)
Corn Shock & Soy Combined Corn &
Shock (1BG) Soy Shock (1BG
each)
Corn Shock & Soy Combined Corn &
Shock (1BG) Soy Shock (1BG
each)
219 The detailed livestock feed market representation in GLOBIOM provides some explanation for this observation.
In the corn shock scenario, GLOBIOM estimates greater DDG production would displace some soybean meal used
for animal feed in the USA region, reducing the demand for soybeans and decreasing cropland used for soybeans. In
the combined shock scenario, demand for soybeans is driven by the soybean oil biodiesel target, and the
displacement effect of DDG in animal feed markets has less impact on cropland used for soybeans. This results in
surplus soybean meal in the USA region in the combined shock scenario, which is exported and displaces some
soybean production in non-USA regions.
123
-------
Figure 8.2.1-2: Difference in cropland area by crop in the corn ethanol shock, soybean oil
biodiesel shock, and combined shock relative to the reference case in non-USA regions in
2030 (ADAGE, GCAM, and GLOBIOM) and 2014 (GTAP)
I Soybean
2.258 2.260
Other Crop Commodities
Scenario
I Corn & Soy Shock (1BG each)
I Corn Shock (1BG)
I Soy Shock (1BG)
H < * 2
o £ 1 0.029 0.030 0.085 0.085 0.139 0.139
Corn Shock & Soy Combined Corn & Corn Shock & Soy Combined Corn & Corn Shock & Soy Combined Corn &
Shock (1BG) Soy Shock (1BG Shock (1BG) Soy Shock (1BG Shock (1BG) Soy Shock (1BG
each) each) each)
8.2.2 Emissions
To compare how the combined shock affects GHG emissions results in each model, we
analyzed the percent change from the combined shock relative to the sum of the core corn
ethanol and soybean oil biodiesel scenarios. ADAGE, GCAM, and GTAP estimate that the
combined scenario would results in relatively similar emissions to the sum of the individual 1
BG corn ethanol and soybean oil biodiesel core scenarios (Table 8.2.2-1). Similar to the soybean
oil biodiesel 500 MG scenario sensitivity, GLOBIOM estimates a larger percentage decrease
than the other models in cumulative LUC and total GHG emissions in the combined scenario.
Table 8.2.2-1: Percent difference in global accumulated emissions between the combined
ADAGE
GCAM
GLOBIOM
GTAP
Percent Difference (TOTAL GHG)
0%
3%
-27%
2%
Percent Difference (LUC Only)
0%
1%
-45%
5%
8.2.3 Summary
In this section we compared LUC and GHG emissions impacts from the combined
scenario to the sum of the core corn ethanol and soybean oil biodiesel scenarios. Overall, across
each of the models (ADAGE, GCAM, GLOBIOM, and GTAP), the results from the combined
scenario show an additive effect in which the combined scenario generally equals the sum of the
two core scenarios across many output values and parameters. GLOBIOM estimates slightly
more variability or nonlinearity in output values than the other models. The most notable
nonlinearity is the decrease in cumulative LUC emissions in the combined scenario. The results
124
-------
from these scenarios did not support the hypothesis that shocking the models with 1 BG corn
ethanol and 1 BG soybean oil biodiesel simultaneously creates greater stress on the agriculture
systems of these models.
9 Parameter Sensitivities
Sensitivity analysis assesses how uncertainty in the output of a model can be apportioned
to different sources of uncertainty in the model input.220 The NASEM (2022) study on LCA
Methods for transportation fuels recommends sensitivity analysis in several areas of the report.
For example, the report says, "LCA studies used to inform transportation fuel policy should be
explicit about the feedstock and regions to which the study applies and to the extent possible
should explicitly report sensitivity of results to variation in these assumptions."221 Following
these recommendations, we have conducted multiple sensitivity analyses as part of our model
comparison exercise.
When we model the environmental and economic impacts of biofuel production,
uncertainties arise in multiple forms. One type of uncertainty is model uncertainty, which is
related to the structure of the model employed. Two models with different structures and/or
solution techniques that otherwise are comparable in scope and use the same input data may
produce different results. One motivation for this model comparison exercise is to study model
uncertainty by comparing results of common scenarios from multiple models. The effect of
different models on GHG estimates is discussed above.
Another form of uncertainty is parameter or input uncertainty. Parameter uncertainty
naturally results as inputs to a model are not exactly known and/or the values of these inputs
cannot be exactly inferred.222 This section focuses on the effects of parameter uncertainty within
a given model. We performed multiple sensitivity analyses to study the influence of parameter
uncertainty on biofuel GHG emissions estimates. These sensitivity analyses are discussed in this
section. First, we performed stochastic sensitivity analysis, where input parameters are assigned
probability distributions, with GCAM, GLOBIOM and GREET. Second, we tested changes in
the soil organic carbon input data in GCAM. Third, we tested changes in land conversion
assumptions in ADAGE.
220 Saltelli, A. (2002), Sensitivity Analysis for Importance Assessment. Risk Analysis, 22: 579-590.
https://doi.org/Ht I 11 11)272-4332.00040
221 National Academies of Sciences, Engineering, and Medicine 2022. Current Methods for Life Cycle Analyses of
Low-Carbon Transportation Fuels in the United States. Washington, DC: The National Academies Press.
https://doi.org/10.17226/26402. Recommendation 4-6. Other relevant recommendations include but are not limited
to: 2-1, 2-2, 4-2, 4-4, 4-9, 4-10.
222 Related to parametric uncertainty is the concept of parametric variability which relates to the fact that even if
perfectly knowable, there is variability in values corresponding to parameter values in these systems. Models are
simplifications of reality and do not capture all the variability naturally occurring over time, space, and changing
conditions.
125
-------
9.1 Stochastic Parametric Sensitivities
9.1.1 GCAM
We ran a Monte Carlo simulation (MCS) with GCAM to explore the influence of a range
of parameters on the LCA estimates. The goals of the MCS are to test the behavior of the model,
evaluate the overall sensitivity of the CI estimates to variations in the input parameters, and to
test which parameters tend to have the largest influence on the results for this specific model.
We conducted this analysis using methods and software consistent with the MCS
described in Plevin et al. (2022).223 We ran the MCS by applying random values drawn from
distributions across 50 parameters. In this case, we use the term parameter to refer to a set of
related values in GCAM's input files. For example, for this analysis we call biomass carbon
density of grassland one parameter, even though GCAM uses independent grassland biomass
carbon input values for each water basin region. For each of the three MCE scenarios (i.e.,
reference, corn ethanol shock, soybean oil biodiesel shock), we ran 1,000 trials (3,000 total
model runs). The same set of randomly drawn parameter values were used for each of the three
scenarios. We consulted with the GCAM developers to determine the likely range of legitimate
values for each parameter and then set selected distributions for each parameter based on our
own subjective judgements. In some cases we were able to leverage previous research to
determine empirically based distribution shapes. Table 9.1.1-1 describes the parameters and
distributions used in our MCS.
Table 9.1.1-1: GCAM Monte Carlo Simulation Parameter Distributions224
Name
Distribution
Description
bd-biomassOil-
coef
Triangle(0.95, 1, 1.05)
The EJ of biomass oil required to produce anEJ of biodiesel.
Corn-etoh-corn-
coef
Triangle(0.98, 1, 1.02)
The Tg of corn required to produce an EJ of corn ethanol.
Crop-biomass-c
Triangle(0.7, 1, 1.3)
Biomass carbon density of cropland.
Grass-biomass-c
Triangle(0.7, 1, 1.3)
Biomass carbon density of unmanaged grass land.
Mgd-forest-
biomass-c
Triangle(0.7, 1, 1.3)
Biomass carbon density of managed forest land.
Mgd-pasture-
biomass-c
Triangle(0.7, 1, 1.3)
Biomass carbon density of managed pasture.
Other-arable-
biomass-c
Triangle(0.7, 1, 1.3)
Biomass carbon density of "other arable" land.
Shrub-biomass-c
Triangle(0.7, 1, 1.3)
Biomass carbon density of shrubland.
Unmgd-forest-
biomass-c
Triangle(0.7, 1, 1.3)
Biomass carbon density of unmanaged forest land.
Unmgd-pasture-
biomass-c-linked
Linked(grass-biomass-
c)
Biomass carbon density of unmanaged pasture (linked with
grass-biomass-c).
223 Plevin, R. J., Jones, J., Kyle, P., Levy, A. W., Shell, M. J., & Tanner, D. J. (2022). Choices in land representation
materially affect modeled biofuel carbon intensity estimates. Journal of cleaner production, 349, 131477. Section 2.5
describes the MCS.
224 Unless the parameter name includes an asterisk, the draws from the given distributions were multiplied by the
GCAM default values to produce values for each trial. For parameter names with an asterisk, values from the
distribution were used directly, replacing the default values.
126
-------
crop-soil-c
Triangle(0.7, 1, 1.3)
Soil carbon density of cropland.
Grass-soil-c
Triangle(0.7, 1, 1.3)
Soil carbon density of unmanaged grass land.
Mgd-forest-soil-c
Triangle(0.7, 1, 1.3)
Soil carbon density of managed forest land.
Mgd-pasture-soil-
c-linked
Linked(grass-soil-c)
Soil carbon density of managed pasture.
Other-arable-soil-c
Triangle(0.7, 1, 1.3)
Soil carbon density of "other arable" land.
Peat-C02-
emissions
Uniform(0.5, 2.0)
CO2 emissions from peatland conversion.
Peat-C02-
emissions-linked
Linked(peat-C02-
emissions)
CO2 emissions from peatland conversion on unmanaged land.
Shrub-soil-c
Triangle(0.7, 1, 1.3)
Soil carbon density of shrubland.
Unmgd-forest-soil-
c
Triangle(0.7, 1, 1.3)
Soil carbon density of unmanaged forest land.
Unmgd-pasture-
soil-c-linked
Linked(grass-soil-c)
Soil carbon density of unmanaged pasture (linked with grass-
soil-c).
N-fertilizer-rate
Triangle(0.7, 1, 1.3)
Quantity of N fertilizer required per mass of crop harvested.
Ag-energy-coef
Triangle(0.7, 1, 1.3)
Energy consumption coefficient for crop production.
Ag-energy-freight-
coef
Triangle(0.5, 1.0, 3.0)
Energy consumption coefficient for transport of ag and energy
commodities.
Crop-productivity
Triangle(0.7, 1, 1.3)
Annual change in agricultural productivity (yield).
Irrig-rainfed-logit-
exp
Triangle(0.333, 1, 3.0)
Logit exponent controlling competition between irrigated and
rainfed land.
Mgmt-level-logit-
exp
Triangle(0.333, 1, 3.0)
Logit exponent controlling competition between high and low
crop management levels.
N2o-emissions
Triangle(0.5, 1, 2.0)
N20 emissions intensity of agricultural production.
Veg-oil-demand-
logit-exp
Triangle(0.333, 1, 3.0)
Controls substitution among types of vegetable oil
water-wd-price
Triangle(0.333, 1, 3.0)
The price of withdrawn water.
Non-staples-
demand-share-
logit*
Uniform(-5.0, 0.0)
Logit exponent controlling shifting between non-staple foods.
Standard value is 0 in all regions.
Agro-forest-logit-
exp
Triangle(0.333, 1, 3.0)
Logit exponent controlling competition between forest-grass-
crop and pasture.
Cow-sheepgoat-
feed-logit
Triangle(0.5, 1, 2.0)
Logit exponent controlling competition between Beef, Dairy,
and SheepGoat, which determines the sharing between Mixed
and Pastoral subsectors.
Crop-logit-exp
Triangle(0.333, 1, 3.0)
Logit exponent controlling competition among crops.
Forest-grass-crop-
logit-exp
Triangle(0.1, 1.0, 3.0)
Logit exponent controlling competition among forest, grassland,
and cropland.
Forest-logit-exp
Triangle(0.333, 1, 3.0)
Logit exponent controlling competition between managed and
unmanaged forest.
Pasture-logit-exp
Triangle(0.333, 1, 3.0)
Logit exponent controlling competition between managed and
unmanaged pasture.
Regional-crop-
logit-exp
Triangle(0.333, 1, 3.0)
Logit exponent controlling competition between imports and
domestic ag products.
Traded-
commodity-logit-
exp
Triangle(0.333, 1, 3.0)
Logit exponent controlling competition in traded ag
commodities.
Traded-
commodity-
subsector-logit-exp
Triangle(0.333, 1, 3.0)
Logit exponent controlling competition among exports in each
traded commodity sector
127
-------
ng-upstream-ch4
Uniform(0.9, 1.3)
CH4 emissions upstream from natural gas production processes
and transport.
Population-factor*
Triangle(0.0, 0.5, 1.0)
Defines a path between the lower and higher bounds of the
UNDP 95 percent confidence interval around population
projections.
Resource-energy-
coef
Triangle(0.5, 1, 1.5)
Energy consumption coefficient for producing energy
commodities.
Biodiesel-
competition-logit-
exp
Triangle(0.5, 1, 2.0)
Controls substitution among types of biodiesel
pass-road-ldv-4 W -
logit-exp
Triangle(0.5, 1, 2.0)
Logit exponent controlling substitution among Compact Car,
Midsize Car, Large Car, Light Truck and SUV.
Pass-road-ldv-4W-
vehicle-logit-exp
Triangle(0.5, 1, 2.0)
Logit exponent controlling substitution among 4WD vehicle
fuel technology options include BEV, FCEV, Hybrid liquids,
Liquids, and NG.
pass-road-ldv-
logit-exp
Triangle(0.5, 1, 2.0)
Logit exponent controlling substitution between 2- and 4-wheel
light-duty vehicles.
Ref-fuel-enduse-
ex-US
Triangle(0.333, 1, 3.0)
Controls substitution in supplies of refined fuel for "end use"
outside the USA.
Staples-price-
elast*
empirical
Price elasticity of demand for staple foods
non-staples-price-
elast*
empirical
Own price elasticity of non-staple food demand.
Non-staples-
income-elast*
empirical
Income elasticity of non-staple food demand.
In some cases, combinations of parameters push the model beyond its ability to match
supply and demand in all markets simultaneously, in which case the model fails to solve. As
shown in the table above, we primarily used triangular distributions to reduce the likelihood,
relative to normal distributions, of outlier parameter draws, thus reducing the number of model
failures. Nonetheless, some of the trials failed to solve; the actual number of reference
case/shock pairs completed for each model version was 916 for corn ethanol (91.6 percent) and
918 for soybean oil biodiesel (91.8 percent). We investigated the source of failures and found the
parameter perturbations most likely causing the failures are some combination of: crop-logit-exp,
staples-price-elast, agro-forest-logit-exp, veg-oil-competition-logit-exp and forest-grass-crop-
logit-exp. The purpose of the MCS is to understand the model's response to parameter variation.
We could reduce the failure rate by narrowing the distributions for these parameters, but this
would come at the cost of gaining insights about how wider distributions influence the model.
Furthermore, evaluating which parameters tend to cause model failures provides valuable
information about the model. For these reasons, we did not to adjust our MCS setup to reduce the
failure rate.
The following figure presents the results of our MCS experiment with GCAM as
distributions of CI estimates for corn ethanol and soybean oil biodiesel. Although the figure
presents the MCS results in probabilistic terms, the actual probability of any given GHG
emissions impact cannot be determined from this analysis. Our sensitivity analysis only reveals
the likelihood of an outcome given all of the inputs into our analysis, such as the version of
GCAM, the reference parameter values, the solution technique, the definitions chosen for the
parameters evaluated, and the distributions for the parameters evaluated. Although the figure
128
-------
does not tell us the actual probability of a given outcome, it provides information about the
general tendency of the model and the variance of results due to parametric uncertainty.
Figure 9.1.1-1: Distribution of GCAM (a) land use change carbon intensity and (b) overall
carbon intensity estimates for corn ethanol and soybean oil biodiesel based on the MCS225
(a) LUC emissions only
soy
(b) Global GHGs
soy
-25 0 25 50 75 100 125
g C02 MJ"1
In the above figure, we present the distribution of land use change CI separately from the
distribution of overall CI. We extract the land use change CI to facilitate comparisons with other
studies or models that only report land use change emissions. While we do this separation to
facilitate comparison, we caution against considering the land use change estimates in isolation,
without considering the influence of scenario design and other sectors on the land use change
estimates. For example, in many of the soybean oil biodiesel trials, non-USA biodiesel
consumption decreases relative to the reference case, which tends to decrease land use change
emissions but tends to increase overall emissions because it is associated with greater use of
refined oil.
Based on the above figure, we observe that GCAM tends to estimate higher CI for
soybean oil biodiesel than corn ethanol, for both land use change and overall. The majority of
overall CI estimates for corn ethanol are less than zero, meaning that over the 2020-2050 period
considered, the modeled corn ethanol shock tends to result in a decrease in global GHG
225 Boxes indicate interquartile range; whiskers indicate 5th and 95th percentiles; vertical line indicates median
value. For corn ethanol, the median land use change carbon intensity is 22 gCO;c/MJ with 95 percent interval from
2 to 48 gCO;c/MJ. For corn ethanol, the median overall carbon intensity is -21 gC02e/MJ with 95 percent interval
from -48 to 8 gCO;c/MJ. For soybean oil biodiesel, the median land use change carbon intensity is 53 gCO;c/MJ
with 95 percent interval from 9 to 106 gCO;c/MJ. For soybean oil biodiesel, the median overall carbon intensity is
40 gCO;c/MJ with 95 percent interval from -5 to 93 gCO;c/MJ.
129
-------
emissions, inclusive of reductions in refined oil consumption. Conversely, a large majority of the
overall CI estimates for soybean oil biodiesel are greater than zero. The overall CI distributions
for the two fuels overlap, but in every trial (i.e., each set of runs with identical parameter values)
the overall CI of corn ethanol is at least 24 gCChe MJ"1 smaller than that of soybean oil biodiesel.
This is explained by the fact that that the most influential parameters have the same directional
effect on the CI estimates for both corn ethanol and soybean oil biodiesel. Finally, the figure
shows that the interval spanning the central 95 percent of CI estimates is about twice as wide for
soybean oil biodiesel relative to corn ethanol, indicating a higher level of parameter uncertainty
for soybean oil biodiesel.
As part of the MCS experiment, we identified the parameters most strongly influencing
the variance in GHG emissions results. We did this by computing the rank correlations between
the values for each random variable and the resulting GHG emissions across all MCS trials. The
rank correlations are squared and normalized to sum to one to produce an approximate
"contribution to variance." In the tornado charts below, the sign of the correlation is applied after
normalization. These figures show the strength of the influence of the 15 most influential input
parameters on the variance in the output (GHG emissions), in descending order, with the
magnitude and direction corresponding to the strength and direction of the correlation
respectively. A contribution to variance further from zero indicates that the parameter is more
influential. A positive contribution to variance indicates that as the parameter value increases or
decreases the CI estimates tend to move in the same direction. A negative contribution to
variance indicates the opposite. Following the figures, we discuss our interpretation of the
findings.
130
-------
Figure 9.1.1-2: Tornado chart of most the influential parameters on corn ethanol land
change carbon intensity estimates with GCAM
use
crop-soil-c
fbrest-grass-crop-logit-exp
crop-log it-exp
ag ro-forest-logit-exp
un mgd-forest-soi l-c
other-arable-soil-c
grass-soil-c
aop-productivity
ag-energy-freight-coef
mgmt-level-l og it-exp
water-wd-price
regional-crop-log it-exp
mgd-pasture-biomass-c
other-ara b le-biom ass-c
shrub-biomass-c
Sensitivity of ci-luc
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4
—I 1 1 1 1 1 1—
0.6
0.8
—I—
Contribution to variance
131
-------
Figure 9.1.1-3: Tornado chart of most the influential parameters on corn ethanol over;
carbon intensity estimates with GCAM
fbrest-grass-crop-logit-exp
crop-soil-c
n2o-emissions
un mgd-forest-soi l-c
crop-log it-exp
ag ro-forest-logit-exp
aop-productivity
ref-fu el-end use-ex-US
other-arable-soil-c
grass-soil-c
other-ara b le-biom ass-c
water-wd-price
ag-energy-coef
regional-crop-log it-exp
shrub-biomass-c
Sensitivity of ci-all
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
—i 1 1 1 1 1 1 1—
0.8
Contribution to variance
-------
Figure 9.1.1-4: Tornado chart of most the influential parameters on soybean oil biodiesel
land use change carbon intensity estimates with GCAM
fbrest-grass-crop-logit-exp
crop-soil-c
ref-fu el-end use-ex-US
crop-log it-exp
un mgd-forest-soi l-c
ag ro-forest-logit-exp
crop-productivity
veg-oii-demand-log it-exp
other-arabie-soii-c
grass-soil-c
biodiesel-competition-logit-exp
other-ara b le-biom ass-c
water-wd-price
shrub-biomass-c
mgmt-level-i og it-exp
Sensitivity of ci-luc
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
—I 1 1 1 1 1 1 1—
0.8
Contribution to variance
133
-------
Figure 9.1.1-5: Tornado chart of most the influential parameters on soybean oil biodiesel
overall carbon intensity estimates with GCAM
fbrest-grass-crop-logit-exp
crop-soil-c
n2o-emissions
crop-log it-exp
ag ro-forest-logit-exp
un mgd-forest-soi l-c
other-arabie-soil-c
grass-soil-c
other-ara b le-biom ass-c
crop-productivity
traded-commodity-iog it-exp
regional-crop-iogit-exp
water-wd-price
ag-energy-freight-coef
shrub-soil-c
Sensitivity of ci-all
-0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6
—i 1 1 1 1 1 1 1—
0.8
Contribution to variance
For overall CI, the tornado charts show that, for this MCS experiment, about 6
parameters have an outsized influence on the estimates. This does not mean the other parameters
have no effect, but rather that their influence is overwhelmed by the 6 most influential
parameters. The 6 most influential parameters for corn ethanol CI are also the 6 most influential
parameters for soybean oil biodiesel, with minor differences in their rank order. All of the 6 most
influential parameters for overall CI are directly related to emissions from land use and land use
change.
For both fuels, the most influential parameter is forest-grciss-crop-Iogit-exp, the
parameter controlling the flexibility of competition among forest, grassland, and cropland.
Higher values for this parameter mean more flexibility for price-driven land use changes among
these land categories. For example, given an increase in crop prices, higher values for this
parameter will translate to larger increases in crop area at the expense of grassland and forest
area. This finding helps to clarify that land conversion flexibility is not only a source of
uncertainty for GHG emissions impacts of biofuels between models, as we observe in Sections
6.6 and 7.6 above. It is also a source of uncertainty within models, at least for GCAM.
The other most influential parameters for both fuels are: 1) crop-soil-c, the soil carbon
density of cropland, 2) ti2o-emissiotis, the N2O emissions intensity of agriculture, 3) crop-logit-
134
-------
exp, the flexibility of competition among crops, 4) agro-forest-logit-exp, the flexibility of
competition between forest, grassland, cropland and pasture, and 5) unmgd-forest-soil-c, the soil
carbon density of unmanaged forest land.
When we look at the most influential parameters on the CI of land use change, we see
almost the same group of influential parameters, but with two exceptions. First, the n2o-
emissions parameter is absent from the tornado charts for land use change CI. N2O emissions are
an important component of crop production emissions in the GCAM results. This parameter is
only absent because we define land use change CI as the projected global change in CO2
emissions from LUC per unit of additional corn ethanol production, with both quantities summed
annually from 2021 through 2050 (i.e., it excludes N2O emission). The second exception is that
ref-fuel-enduse-ex-US parameter shows up as one of the most influential parameters for soybean
oil biodiesel land use change CI. This parameter controls substitution in supplies of refined fuel
outside the USA. For example, it controls substitution between biodiesel and petroleum diesel in
non-USA regions. As discussed above, in GCAM the soybean oil biodiesel shock tends to reduce
biodiesel consumption outside the USA, which increases petroleum diesel consumption and
requires less land for biodiesel feedstocks. Thus, higher values for ref-fuel-enduse-ex-US tends to
result in lower land use change emissions, but increases other emissions, resulting in a small net
effect on overall CI.
Overall, our MCS experiment with GCAM provides several insights. Parameter
uncertainty is an important factor for CI estimates of corn ethanol and soybean oil biodiesel with
GCAM. Based on this experiment, CI estimates for soybean oil biodiesel are more sensitive to
parameter uncertainty than such estimates for corn ethanol. Parameters related to land use change
have the most influence on CI estimates. In particular, parameters related to soil carbon densities
and ease of substitution between land categories are highly influential, and thus warrant special
attention.
9.1.2 GLOBIOM
We ran a Monte Carlo simulation (MCS) with GLOBIOM to explore the influence of a
range of parameters on land use change carbon intensity (LUC CI) for soybean oil biodiesel.226
The goals of the GLOBIOM MCS mirror those of the GCAM MCS discussed in Section 9.1.1; to
test the behavior of the model and to evaluate the overall sensitivity of the CI estimates to
variations in the input parameters.
The approach used in the GLOBIOM MCS was similar to that used in the GCAM MCS
described in Section 9.1.1. We ran the MCS by applying random values drawn from distributions
defined for 11 parameters. For each of two cases (i.e., a reference case and a soybean oil
226 The GLOBIOM MCS was conducted prior to the initiation of this MCE and, as such, differs somewhat in its
scenario design and assumptions. Differences between the version of GLOBIOM used in the MCE include some
minor updates of corn food consumption trends to better match historic development (2010, 2020) in a number of
different regions represented in GLOBIOM. The changes shift upward the food demand projections in both the
reference and shock scenarios. Additionally, the shock scenario in the MCS was specified as one billion gallons
gasoline equivalent of soybean oil biodiesel above reference case levels, whereas the shock in the MCE was
specified as one billion wet gallons of soybean oil biodiesel consumption above reference case levels.
135
-------
biodiesel shock), we ran 1,000 trials (2,000 scenario runs total). The same set of randomly drawn
parameter values were used for both of the two cases.
The eleven identified parameters were chosen by GLOBIOM developers based on expert
knowledge and previous research.227'228'229 These include seven economic parameters and four
biophysical parameters. The parameters and distributions used in the GLOBIOM MCS are
described below in Table 9.1.2-1. Each parameter distribution below represents a set of related
input values in GLOBIOM which are adjusted simultaneously based on the drawn value of the
parameter in a given trial. For example, a value drawn for the parameter labeled "Demand
elasticity (vegetable oils)" in Table 9.1.2-1 below is a multiplicative scalar which simultaneously
adjusts the demand elasticity for each vegetable oil and each region represented in GLOBIOM.
Three of the parameters in Table 9.1.2-1 represent collections of inputs which each have
independently drawn scalar values from the identical distribution. These parameter groups are
indicated with bold names and described in the Description column. When accounting for these
parameter groups, 72 separate values are drawn for each of 1,000 trials in the MCS.
227 Valin, H., D. Peters, M. van den Berg, S. Frank, P. Havlik, N. Forsell & C. Hamelinck (2015) The land use
change impact of biofuels consumed in the EU. Quantification of area and greenhouse gas impacts. Ecojys, Utrecht
(the Netherlands).
228 Nelson, G. C., H. Valin, R. D. Sands, P. Havlik, H. Ahammad, D. Deryng, J. Elliott, S. Fujimori, T. Hasegawa,
E. Heyhoe, P. Kyle, M. Von Lampe, H. Lotze-Campen, D. Mason d'Croz, H. van Meijl, D. van der Mensbrugghe, C.
Muller, A. Popp, R. Robertson, S. Robinson, E. Schmid, C. Schmitz, A. Tabeau & D. Willenbockel (2014) Climate
change effects on agriculture: economic responses to biophysical shocks. Proc Natl Acad Sci USA, 111, 3274-9.
https://doi.org/10.1073/pnas. 1222465110
229 Valin, H., R. D. Sands, D. van der Mensbrugghe, G. C. Nelson, H. Ahammad, E. Blanc, B. Bodirsky, S.
Fujimori, T. Hasegawa, P. Havlik, E. Heyhoe, P. Kyle, D. Mason-D'Croz, S. Paltsev, S. Rolinski, A. Tabeau, H. van
Meijl, M. von Lampe & D. Willenbockel (2014) The future of food demand: understanding differences in global
economic models. Agricultural Economics, 45, 51-67. https://doi.org/10.llll/agec.12089
136
-------
Table 9.1.2-1: GL<
DBIOM Monte Carlo simulation parameter distributions230'231
Name
Distribution
Description
Demand elasticity
(vegetable oils)
Log-uniform(0.5, 2)
Own-price and cross-price elasticities of demand for
vegetable oils. Determines adjustments in food uses of
vegetable oils.
Demand elasticity
(animal products)
Log-uniform(0.5, 2)
Own-price and cross-price elasticities of demand for animal
products (meat and dairy). Determines adjustments in food
uses of animal products.
Trade elasticity
(vegetable oils)
Log-uniform(0.75, 4)
Response of bilaterally traded quantities of vegetable oils to
changes in market prices.
Separate scalar values are drawn from identical distributions
for each of the four vegetable oils represented in
GLOBIOM.
Substitution
elasticity
(vegetable oils)
Log-uniform(0.75, 4)
Substitutability of vegetable oils for all uses, given a change
in their market price.
Separate scalar values are drawn from identical distributions
for each of 58 different global regions represented in
GLOBIOM.
Cropland and
pasture expansion
into natural
vegetation
Log-uniform(0.5, 2)
Extent to which cropland and grazing pasture can expand
into natural land uses, represented by land transition costs.
Separate scalar values are drawn from identical distributions
for cropland and grazing pasture.
Yield elasticity
(corn and soybean)
Log-uniform(0.9, 1.1)
Changes in corn and soybean yields in response to changes
in crop prices.
Yield projection
(corn and soy)
Log-uniform distribution
between SSP3 and SSP5
assumptions.
Exogenous yield change over time for corn in the USA
region and soybeans in the USA, Brazil, and Argentina
regions.
Expansion response
of palm into
peatland
Uniform(0.5, 1.5)
Degree of expansion of palm plantation into peatland in
Indonesia and Malaysia.232
Peatland emission
factor on
undisturbed forest*
Lognormal distribution on
range of 49 to 8549 tC02
ha-1 yr_1
Peatland emission intensity per unit of area converted in
Indonesia and Malaysia.
Emission factor for
carbon sequestration
in biomass on palm
plantations
Normal(0.59, 1, 1.41)
Carbon sequestration (as CO2) in palm plantations in
Indonesia and Malaysia per unit of area. Range based on
(IPCC 2019).233
Emission factors
from forest biomass
loss
Normal(0.5, 1, 1.5)
Emissions per unit of area due to forest clearing.
230 Bold parameter names indicate related groups of parameters. Unless the parameter name includes an asterisk,
the draws from the given distributions were multiplied by the GLOBIOM default values to produce values for each
trial. For parameter names with an asterisk, values from the distribution were used directly, replacing the default
values.
231 Note that some of the scalar distributions in this MCS are not balanced around the central value (scalar of 1). For
example, in the distribution for trade elasticity of vegetable oils (Log-uniform(0.75, 4)), roughly 17 percent of the
draws would be expected to be below one, and thus decrease the value of the given vegetable oil trade elasticity, and
roughly 83 percent of the draws would be expected to be above one, and thus increase that elasticity.
232 In GLOBIOM, expansion of palm plantations is assumed to occur in peatland and non-peatland at a fixed ratio,
which we adjust stochastically in this MCS analysis.
233 IPCC. 2019. 2019 Refinement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories. Volume 4:
Agriculture, Forestry and Other Land Use. Geneva (Switzerland): Intergovernmental Panel on Climate Change.
137
-------
Figure 9.1.2-1 below presents distributions of carbon intensity factors for a number of
different emissions categories, after excluding trials considered outliers.234 Although the figure
presents the MCS results in probabilistic terms, the actual probability of any given GHG
emissions impact cannot be determined from this analysis. Our sensitivity analysis only reveals
the likelihood of an outcome given all of the inputs into oar analysis, including the version of
GLOBIOM, the reference parameter values, and the distributions for the parameters evaluated.
Although the figure does not tell us the actual probability of a given outcome, it provides
information about the general tendency of the model and the variance of results due to
parametric uncertainty.
Figure 9.1.2-1: Distributions of carbon intensities from different categories of emissions for
soybean oil biodiesel based on the GLOBIOM MCS.235
LUC - Biomass 1
LUC -SOC
LUC - Peat
LUC - Total
Livestock production
Crop production
-50
50
100
gC02e MJ
-1
The MCS produced a range of LUC CI results (9.5, 40.6, and 73.5 gC02e/MJ for the 10th
percentile, mean, and 90th percentile respectively), with variation in emissions from biomass loss
accounting for a substantial portion of the variability in total LUC emissions. Note that the mean
value of total LUC CI for the GLOBIOM MCS is larger than the LUC CI estimate from the
234 Outliers are identified in these results based on the so-called "1.5 rule", assuming that the distribution of
emissions factors follows a normal distribution. According to this rule, a data point is considered an outlier if it is
less than (Q1 - 1.5*IQR) or greater than (Q3 + 1.5*IQR), where IQR is the interquartile range and Q1 and Q3 are
the first and third quartiles of the distribution, respectively. Outlier trials were identified using this rule for each of
three emissions categories - total land use change, crop production, and livestock production - after which all
identified outlier trials were excluded from the following results analysis. In total, 42 outlier trials were excluded
using this procedure.
235 Vertical lines within distributions represent mean values. "LUC - Biomass" includes emissions changes from
biomass loss from land use change, changes in agricultural biomass, natural reversion of land, and carbon
sequestered in harvested wood products. "LUC - SOC" emissions are land use change emissions from soil organic
carbon. "LUC - Peat" emissions are land use change emission from oxidation of peatlands. "LUC - Total" is the
sum of the above land use change emissions categories.
138
-------
soybean oil biodiesel shock scenario in the MCE. This difference arises for two reasons; 1) the
version of GLOBIOM used in the MCE was a more recent version of the model, with several
updated assumptions (see footnote above); and 2) some of the distributions of scalar values
applied to the parameters are weighted towards increasing the value of the parameter, which may
result in more trials showing CI values on one side of the central MCS scenario than the other.
This difference illustrates the limitation discussed above, but worth reiterating; distributions of
CI values produced through this MCS analysis are dependent on the inputs of the analysis and
should not be interpreted as representative of the probability of a given GHG emissions impact.
However, there are still meaningful observations we can make using these results.
GLOBIOM's estimates of GHG emissions from land use change, particularly emissions from
biomass loss but also from other subcategories of estimated LUC emissions, appear to be more
sensitive to parametric variations, at least for the parameters and distributions included in this
study, than estimates of emissions from livestock production and from crop production. This
observation reinforces the importance of continued study of model assumptions affecting LUC
and LUC CI and of considering uncertainty in LUC CI estimates.
In a process similar to that used in the GCAM MCS described in Section 9.1.1 above, we
identified the parameters most strongly influencing the variance in LUC CI. We did this by
computing the rank correlations between the values for each random variable and the resulting
LUC CI estimate across all MCS trials. The rank correlations are squared and normalized to sum
to one to produce an approximate "contribution to variance." In Figure 9.1.2-2 below, the sign of
the correlation is applied after normalization. This figure shows the strength of the influence of
each input parameter on the variance in the output (LUC CI), in descending order, with the
magnitude and direction corresponding to the strength and direction of the correlation
respectively. A contribution to variance further from zero indicates that the parameter is more
influential. A positive contribution to variance indicates that as the parameter value increases or
decreases the CI estimates tend to move in the same direction. A negative contribution to
variance indicates the opposite.
139
-------
Figure 9.1.2-2: Tornado chart of most the influential parameters in GLOBIOM MCS on
soybean oil biodiesel land use change carbon intensity.236
-0.1 0.0 0.1 0.2 0.3 0.4
Expansion response of palm into peatland
EF from forest biomass loss
Yield projection (corn and soy)
Substitution elasticity (vegetable oils)
Peatland EF on undisturbed forest
Yield elasticity (corn and soybean)
EF for carbon in biomass on palm plantations
Cropland and pasture expansion into natural vegetation
Demand elasticity (vegetable oils)
Trade elasticity (vegetable oils)
Demand elasticity (animal products)
Contribution to variance in LUC CI
The two parameters found to have the largest contribution to variance in LUC CI were
the expansion response of palm into peatland and the emissions factor from forest biomass loss.
The positive correlation of these parameters with LUC CI is logical; larger values of the first
result in greater expansion of palm plantations into peatland in response to the increased demand
for vegetable oils imposed under a soybean oil biodiesel shock. Larger values of the second
increase the emissions associated with forest loss in response to the shock. The sensitivity of
GHG emissions estimates to these parameters highlights the importance of further examination
of all of the models' parameterizations of land transitions, carbon fluxes, and representation of
peat lands.
The parameter with the third largest contribution to variance of LUC CI is the assumed
yield growth of corn and soy throughout the duration of the GLOBIOM run, which is negatively
correlated with LUC CI. Again, this relationship is logical; lower yield growth results in lower
yields in the future, which means that producing feedstock (soybeans) to meet the shock requires
additional cropland area and results in greater areas of land use change. The relative impact of
this parameter highlights the importance of considering the impact of assumptions about baseline
trends and how they continue into the future.
Finally, we note the relative importance (4th in Figure 9.1.2-2) of the substitution
elasticity of vegetable oils. Increasing the assumed substitutability of vegetable oils allows the
model to backfill more easily for deficits in soybean oil use with other oilseed oils, including
236 For parameters which represent groups of independently adjusted model inputs (indicated in bold), the
contributions to variance across all inputs within a given parameter group are summed. For all three of the grouped
parameters, this results in some cancellation because the signs of the calculated contributions to variance differ
among the inputs within a group. An alternative MCS design which instead used a single value applied to all model
inputs within these parameter groups may be expected to increase the relative contribution to variance of these
parameters.
140
-------
from palm and rapeseed. This results in increased diversion of soybean oil from food and other
uses. The impacts of this substitution on land use change and emissions are not straightforward,
vary by region and type of vegetable oil substitution, and interact with other parameters
perturbed in this MCS.237 This complicating layer of market interaction contributes to the wider
range of estimated GHG emissions impacts of soybean oil biodiesel relative to corn ethanol.
9.1.3 GREET
We worked with Argonne to develop the lifecycle GHG emissions analyses presented in
Section 6.7 and Section 7.7. These analyses rely on many input values from many sources
including government (e.g., USD A, EPA, DOE), academia, and industry. All these input values
are subject to some level of variation and uncertainty. We worked with Argonne to conduct
multiple sensitivity analyses with the GREET model238 to explore the influence of the inputs and
assumptions in the model framework on the results. This exercise allowed us to observe some of
the most influential and important factors to consider for further research to address uncertainty.
We conducted three sensitivity analyses, where we varied one parameter or assumption at a time,
and one stochastic sensitivity analysis (Section 9.1.3.4) where we varied all of the input
parameters simultaneously based on random draws from statistical distributions. Each of these
analyses are described in this section.
9.1.3.1 Parameter Input Data
To support our parametric sensitivity analyses we used data that Argonne has previously
collected from various sources. These data provide information about the variation in some of
the key input values to GREET. For farming input data, the main source of the variation is
geographic, and the source of variation for ethanol production data is differences among
individual corn ethanol facilities. The value and ranges for these parameters were used in both
the sensitivity and stochastic (Section 9.1.3.4) analyses discussed below. The tables below list
the parameter values and their ranges for corn ethanol and soybean oil biodiesel. The tables also
indicate the shape of the distribution used for each parameter for the stochastic analysis. For
parameters where Argonne had a relatively large data set on variation they used a normal
distribution, whereas they used a triangular distribution for parameters informed with less data
on variation.
Most of the data used in support of corn ethanol sensitivities is documented in Lee et al.
(2021).239 For corn farming, that includes data from USD A datasets (National Agricultural
Statistics Service [NASS], the Economic Research Service [ERS], and the Office of the Chief
237 For example, the effect on GHG emissions of greater substitution of palm oil for soybean oil used for food and
fuel production in Southeast Asia is amplified or muted by the parameters governing the expansion response of palm
plantations onto peatland, emissions factors associated with forest biomass loss, and the carbon in biomass on palm
plantations.
238 Sensitivity analyses presented in this section were run using GREET-2022 for the 2021 time step. This is the
default time step for the model. We decided to conduct sensitivity analyses for the 2021 time step as the data used to
inform the parameter ranges is more representative of 2021 than 2030.
239 Lee, Uisung, Hoyoung Kwon, May Wu, and Michael Wang (2021). "Retrospective Analysis of the US Corn
Ethanol Industry for 2005-2019: Implications for Greenhouse Gas Emission Reductions." Biofuels, Bioproducts and
Biorefining 15 (5): 1318-31.
141
-------
Economist [OCE] reports). Ethanol production data relies heavily on a corn ethanol
benchmarking and an agricultural consulting company that has conducted quarterly surveys of 65
dry mill ethanol facilities between 2005 - 2019 and includes ethanol yields (with corn inputs and
ethanol production), energy inputs by type (natural gas, coal, and electricity), chemical inputs,
and the yields of coproducts. Argonne used the 10th percentile (P10) and the 90th percentile
(P90) values as the high and low bounds of the ranges for ethanol production parameters in this
exercise. The full set of input parameters and their ranges for corn ethanol are shown below in
Table 9.1.3-1.
Table 9.1.3-1: GREET Corn Ethanol Sensitivity and Stochastic Simulation Input
Parameter Distributions for Model Year 2021
Name
Distribution240
Units
Farming: Corn yield
Normal (113, 178, 191)
bushels/acre
Farming: Corn yield (Nine states)241
Normal (153, 178, 191)
bushels/acre
Farming: N fertilizer
Normal (72, 158, 187)
lbs/acre
Farming: P fertilizer
Normal (33, 59, 89)
lbs/acre
Farming: K fertilizer
Normal (16, 60, 130)
lbs/acre
Farming: N20 rate
Normal (0.8, 1.26, 1.6)
percent
Farming: Herbicide
Normal (0.0, 2.3,3.2)
lbs/acre
Farming: Insecticide
Normal (0.0, 0.0, 0.2)
lbs/acre
Farming: Diesel
Normal (630,025; 927,625; 1,578,474)
BTU/acre
Farming: Gasoline
Normal (115,686; 143,155; 201,905)
BTU/acre
Farming: Natural gas
Normal (0; 85,504; 260,170)
BTU/acre
Farming: LPG
Normal (57,257; 183,004; 290,957)
BTU/acre
Farming: Electricity
Normal (72,741; 236,548; 950,459)
BTU/acre
Corn transportation distance
Normal (32, 40, 48)
miles
Ethanol: Yield
Triangular (2.7, 2.9, 3.0)
gal/bu
Ethanol: DGS yield
Triangular (3.7, 4.6, 5.5)
lbs/gal
Ethanol: Natural gas
Triangular (8,846; 22,386; 30,961)
BTU/gal
Ethanol: Electricity
Triangular (600; 2,098; 3,646)
BTU/gal
For soybean farming, the data informing the sensitivity analysis was mostly documented
in Xu et al. (2022)242 and primarily comes from USDA's National Agricultural Statistics Service
(NASS) Quick Stats database.243 Farm energy use data was obtained from USDA's ERS based
on the Agricultural Resource Management Survey. The farming data covers 19 major soybean-
240 In the parentheses, the first value is the P10 value, the middle value is the default assumption in GREET, and the
third value is the P90 value.
241 Corn is grown in many states in the United States but is primarily grown in the Midwest region across nine states.
For this sensitivity analysis, we present both the fuller range of corn yields across the U.S., and this subset of nine
primary corn growing states, which has a tighter range of corn yields.
242 Xu, Hui, Longwen Ou, Yuan Li, Troy R. Hawkins, and Michael Wang. 2022. "Life Cycle Greenhouse Gas
Emissions of Biodiesel and Renewable Diesel Production in the United States." Environmental Science &
Technology 56 (12): 7512-21. https://doi.org/10.1021/acs.est.2c00289.
243 USD A National Agricultural Statistics Service Quick Stats Database. Available at:
https://anickstats.nass.nsda.gov/
142
-------
producing U.S. states. Parameter data on biodiesel production (e.g., chemical inputs, energy
consumption, product yields) came from an Argonne-led industry survey conducted of biodiesel
producers in 2021 with support from what was then known as the National Biodiesel Board
(NBB) and is now known as Clean Fuels Alliance America as documented in Xu et al. The full
set of input parameter values and their ranges for soybean oil biodiesel are shown below in Table
9.1.3-2.
Table 9.1.3-2: GREET Soybean Oil Biodiesel Sensitivity and Stochastic Simulation Input
Parameter Distributions for Model year 2021
Name
Distribution244
Units
Farming: Soybean yield
Triangular (31.4, 50.6,61.7)
bushels/acre
Farming: N fertilizer
Triangular (1.3, 4.9, 15.6)
lbs/acre
Farming: P fertilizer
Triangular (12.4, 23.2, 54.8)
lbs/acre
Farming: K fertilizer
Triangular (2.9, 36.8, 92.6)
lbs/acre
Farming: Herbicide
Triangular (1.5, 2.2, 3.8)
lbs/acre
Farming: Insecticide
Triangular (0.002, 0.03, 0.40)
lbs/acre
Farming: Energy use
Triangular (338,791; 694,421; 1,373,805)
BTU/acre
Biodiesel production: Methanol use
Triangular (926, 945, 964)
BTU/lb BD
Biodiesel production: Energy use
Triangular (437, 514, 592)
BTU/lb BD
Biodiesel production: Biodiesel yield
Triangular (0.133, 0.136, 0.138)
gal BD/lb oil
Oil extraction: Oil yield
Triangular (4.4, 4.6, 4.9)
dry lbs
soybean/
lb soybean oil
Oil extraction: Energy use
Triangular (2,765; 3,073; 3,380)
BTU/lb oil
Biodiesel production: Glycerin yield
Triangular (0.09, 0.10,0.11)
lb/lb BD
9.1.3.2 Parameter Sensitivity Scenario Analysis
The first set of parametric sensitivities presented here was developed with Argonne and
assessed the modeling framework by considering variations and ranges of the key parameters
shown above and their individual impacts on the carbon intensities of corn ethanol and soybean
oil biodiesel produced in the United States. We conducted these sensitivity analyses by varying
each major input parameter shown in Table 9.1.3-1 for corn ethanol and Table 9.1.3-2 for
soybean oil biodiesel across their full range of values, each one at a time while keeping all the
other parameter values constant. By varying one parameter at a time, while holding others
constant, we can see the relative impact of each parameter on the final estimated LCA results.
This is also informative for identifying areas of uncertainty and necessary further research.
However, this "one at a time approach" provides less information than a stochastic analysis about
the potential range of results stemming from parameter uncertainty. This is because one at a time
analysis does not consider the effect of multiple parameters simultaneously varying from their
default input values. For example, if corn yield is higher than the default input value and
simultaneously the farming nitrogen fertilizer rate is actually lower than the default input value,
the actual carbon intensity may be lower than any of the results depicted in the Figure 9.1.3-1.
244 In the parentheses, the first value is the P10 value, the middle value is the default assumption in GREET, and the
third value is the P90 value.
143
-------
We used the parameter values in Table 9.1.3-1 for corn ethanol in GREET-2022
representing 2021 to conduct the sensitivity analysis of each individual parameter against a
baseline CI value of 45.9 gCCh/MJ derived using GREET's default assumptions (including
coproduct allocation assumptions). This value excludes LUC impacts from GREET's separate
CCLUB module that are discussed further below. Figure 9.1.3-1 shows the results of the
sensitivity analysis for corn ethanol minus GREET's CCLUB derived LUC impacts. Parameters
are ordered by their relative individual influence on the overall CI with the most impactful
parameters at the top of the figure.
Figure 9.1.3-1: Sensitivity analysis results of USA corn ethanol carbon intensity values
ranked by relative influence of each parameter's potential impact in GREET
Ethanol: Natural gas, Btu/gal (+38%/-60%)
Farming: Corn yield, bu/acre (+7%/-37%)
Farming: N fertilizer, lb/acre (+ 18%/-54%)
Ethanol: Electricity, Btu/gal (+ 74%/-71%)
Farming: Corn yield (9 states), bu/acre (+ 7%/-14%)
Ethanol: DGS yield, lb/gal (+ 20%/-20%)
Ethanol: Yield, gal/bu (+ 3%/-4%)
Farming: Electricity, Btu/acre (+ 302%/-69%)
Farming: Diesel, Btu/acre (+ 70%/-32%)
Farming: N20 rate, % (+ -37%/27%)
Farming: P fertilizer, lb/acre (+ 50%/-44%)
Farming: K fertilizer, lb/acre (+ 117%/-73%)
Farming: Herbicide, lb/acre (+39%/-100%)
Farming: Natural gas, Btu/acre (+ 204%/-100%)
Farming: LPG, Btu/acre (+ 59%/-69%)
Com transportation distance, miles (+ 20%/-20%)
Farming: Gasoline, Btu/acre (+ 41%/-19%)
Farming: Insecticide, lb/acre (+3354%/-100%)
Based on the data provided, overall CI for corn ethanol saw the largest variation and
influence in this exercise from the amount of natural gas used in processing and producing
ethanol in facilities with a wide range of efficiencies representing a difference of roughly 20
grams of CO2 per MJ of ethanol produced. Corn yields from farming corn was the next most
important factor when considering the variation in growing corn across the country. A subset of
these corn yields appears further down the list when considering only the nine states in the
Midwest. These states represent the majority of corn production volume and have higher corn
yields than most of the country. Corn farming and corn ethanol production do take place across
many states outside the Midwest,245 and we present both variations of this parameter for context.
Nitrogen fertilizer used to obtain higher crop yields was the third highest parameter of
importance in this sensitivity analysis.
We used the parameter values in Table 9.1-3 for soybean oil biodiesel in GREET-2022
representing 2021 to conduct the sensitivity analysis of each individual parameter against a
baseline CI value of 22.0 gCCh/MJ derived using GREET's default assumptions (including
coproduct allocation assumptions). This value also excludes LUC impacts from GREET's
separate CCLUB module that are discussed further below. Figure 9.1.3-2 shows the results of the
245 Geographic Representation of Corn Ethanol Production Ethanol Facilities in The United States. EIA (2023).
Available at: https://atlas.eia.gov/maps/3f984029aadc4647ac4025675799af90
144
-------
sensitivity analysis for soybean oil biodiesel minus GREET's CCLUB derived LUC impacts.
Parameters are ordered by their relative individual influence on the overall CI with the most
impactful parameters at the top of the figure.
Figure 9.1.3-2: Sensitivity analysis results of USA soybean oil biodiesel carbon intensity
values ranked by relative influence of each parameter's potential impact in GREET
21 21.5 22 22.5 23 23.5 24
Farming: Soybean yield, bushels/acre (-38%, +22%) 21.4 23.9
Farming: Energy use, Btu/acre (-51%,+98%) 21.2 23.5
Farming: N fertilizer, lb/acre (-74%, +220%) 21.6 23.1
Oil extraction: Oil yield, lb soybean/lb oil (-5%, +5%)
Farming: P fertilizer, lb/acre (-47%, +136%)
Biodiesel production: Energy use, Btu/lb BD (-15%, +15%)
Oil extraction: Energy use, Btu/lb oil (-10%, +10%)
Biodiesel production: Biodiesel yield, gal BD/lb oil (-2%, +2%)
Farming: K fertilizer, lb/acre (-92%, +152%)
Farming: Herbicide, lb/acre (-33%, +78%)
Biodiesel production: Methanol use, Btu/lb BD (-2%, +2%)
Farming: Insecticide, lb/acre (-92%, +1118%)
Biodiesel production: Glycerin yield, lb/lb BD (-10%, +10%) 22.0 II 22.0
Based on our input parameters and our GREET framework, the overall CI for soybean oil
biodiesel saw the most influence from the soybean crop yields. Energy used in growing soybean
on the field was the next most important factor. Nitrogen fertilizer used to obtain higher crop
yields was again the third highest parameter of importance in this sensitivity analysis. There was
not a wide variation of results in this exercise, and the greatest variation was in soybean farming
rather than soybean oil biodiesel production but that is due in part to a limited amount of
available data on variations in biodiesel production. The relatively small variation in estimates
suggests that variation in the parameters tested is not a large source of uncertainty for supply
chain LCA of soybean oil biodiesel. However, there are other assumptions that have a larger
influence on soybean oil biodiesel LCA estimates, as discussed in the sections that follow.
With some minor differences, we saw similarities between the most influential
parameters across corn ethanol and soybean oil biodiesel in this exercise. Crop yields and
nitrogen fertilizer as inputs were among the most influential factors in both scenarios and had
some of the largest impacts on these results based on the data provided. However, while both
sensitivities included farming practices, these did not include LUC parameters.
9.1.3.3 Allocation Sensitivity Analysis
Corn ethanol and soybean oil biodiesel production processes both yield biofuels as well
as economically significant coproducts. Dry mill corn ethanol production for example produces
distillers grains that are often used as livestock feed, and corn oil that is a vegetable oil that can
be used for cooking. Both have the potential to be further processed for producing biodiesel.
Similarly, soybean oil biodiesel transesterification results in coproducts such as soy meal which
is high in fiber and can be used as cattle feed, and glycerin that has a range of applications across
cosmetics and pharmaceuticals.
145
-------
For supply chain LCA models such as GREET, these coproducts are relevant because the
GHG impacts of the fuel of interest and its coproducts can be accounted for using various
methods and therefore yield different GHG results depending on the allocation methods used.
Allocation methods can use the economic values of the different product streams, the embedded
energy content (where applicable), or physical properties such as mass. This allocation
sensitivity analysis shows the variation in the CI values presented using the default input
parameters and how the resulting GHG emissions can vary quite significantly depending on the
LCA allocation methods selected.
For corn ethanol in GREET, Argonne uses a default displacement allocation method
whereby dried distillers grains are given a coproduct credit under the assumption they will be
used in place of conventional animal feeds such as corn and soybean meal. This results in the
estimated default CI value of 45.9 gCCh/MJ for corn ethanol shown in Figure 9.1.3-3, but this
result can vary significantly if the allocation method used is instead based on the energy content
of the ethanol and distillers grains or based on market value of the distillers grains versus the
ethanol fuel (which in turn relies on constantly varying and geographically diverse market
values). A hybrid method is also presented to allocate distillers grains, ethanol, and corn oil first
based on the market value first, and then energy allocation is used to calculate emissions for
ethanol and corn oil. The last results shown are a process-level allocation method that assigns
emission burdens of individual process steps to the product that is responsible for each specific
process. These last two allocation methods are further detailed in Wang et al. (2015).246 Based on
allocation method alone in this scenario, we derived a range between 32.2 - 48.4 gCCh/MJ for
corn ethanol (excluding LUC impacts).
Figure 9.1.3-3: Variations in the Carbon Intensity of Corn Ethanol Based on Various LCA
Allocation Methods
60
— 48.4
2 50 45 9 I - 45.3
'Mill
GREET default Energy allocation Market allocation Hybrid Process-level energy
allocation allocation
For soybean oil biodiesel, Argonne presents further delineations of LCA allocation
methods used either at the process level (assigning the GHG impacts based on the individual
steps that are involved, in this case soybean oil and soybean meal at the crushing facilities and
then between biodiesel and glycerin at the biodiesel plants) or the system level (in this instance
assigning the GHG burden across biodiesel, soy meal, and glycerin as products rather than
246 Wang, Zhichao, Jennifer B. Dunn, Jeongwoo Han, and Michael Q. Wang. 2015. "Influence of Corn Oil Recovery
on Life-Cycle Greenhouse Gas Emissions of Corn Ethanol and Corn Oil Biodiesel." Biotechnology for Biofuels 8
(1): 178. https://doi.org/10.1186/sl3068-015-035Q-8.
146
-------
individual steps). Within each of the process- and system-level allocation methods, there are the
same three methods of allocation shown for corn ethanol: mass, market value, and energy
allocation. Argonne by default uses a hybrid allocation method for soybean oil biodiesel in
GREET whereby mass-based allocation is used to account for the soybean meal coproduct from
soybean crushing and market-based allocation is used to account for the glycerine coproduct
from biodiesel production. This results in the estimated default CI value of 22.0 gCCh/MJ for
soybean oil biodiesel as shown in Figure 9.1.3-4. Based on different allocation methods alone in
this scenario, we derived a range between 18.4 - 33.7 gCCh/MJ for soybean oil biodiesel
(excluding LUC impacts), exemplifying how complicated it can be to perform LCA allocation
for various biofuels. This results in the estimated default CI value of 22.0 gCCh/MJ for soybean
oil biodiesel as shown in Figure 9.1.3-4. Based on different allocation methods alone in this
scenario, we derived a range between 18.4 - 33.7 gCCh/MJ for soybean oil biodiesel (excluding
LUC impacts).
Figure 9.1.3-4: Variations in the Carbon Intensity of Soybean Oil Biodiesel Based on
Various LCA Allocation Methods
33.7
22.0
32.2
29.6
18.4
Hybrid
(GREET Default)
Mass Market Value Energy
Process-Level Allocation
Mass Market Value Energy
System Level Allocation
As illustrated by the figures above in this allocation sensitivity analysis section,
coproduct allocation methods can have a significant impact on biofuel LCA estimates when
using a supply chain LCA model such as GREET. As with the above sections, these results did
not include GREET's reported LUC GHG emissions that come from CCLUB and rely on GTAP
data.
9.1.3.4 Stochastic Parameter Analysis
Relying on the same parameter inputs and distributions shown in Tables 9.1.3-1 and
9.1.3-2, we also conducted a sensitivity analysis using the stochastic tool built into the GREET
model. This tool allows for stochastic analyses of probable ranges of the different factors that
result in the likelihood of multiple outcomes, to conduct parameter uncertainty. This stochastic
tool also does not make changes to the land use change results that come from CCLUB
translating GTAP data but focuses on agricultural practices, fuel production, and transportation.
Therefore, the uncertainty present in LUC emissions estimates, discussed in other sections above
and below, is not considered here. Because GREET operates as a static attributional LCA
147
-------
framework, any uncertainties in market-mediated responses to biofuel consumption in the
agricultural or energy sectors is also not considered, nor are any uncertainties regarding dynamic
change over time.
A probability density function (PDF) was developed for the corn ethanol pathway
analyzed using the stochastic tool. GREET breaks down the corn ethanol pathway into the
following steps: farming energy, farming chemicals, ethanol production, coproducts, and tailpipe
fuel combustion (non-CCh emissions). The base values are presented along with what are known
as P10 and P90 values that make up the uncertainty bars. Ninety percent of the observations in
the stochastic analysis are above the P10 value, while ninety percent of observations fall below
the P90 value. Figure 9.1.3-5 below shows the stochastic analysis results for corn ethanol. This
stochastic analysis for corn ethanol relying on the input data provided would imply an 80 percent
probability that the GREET estimate for the fuel would be between 40.7 and 57.0 gCCh/MJ
(before accounting for LUC). The greatest variation identified based on data provided came from
farming chemicals used to support corn yields.
Figure 9.1.3-5: Stochastic analysis results of USA corn ethanol by lifecycle stage in GREET
(whiskers indicate P10 and P90 values)
70
i 60 —t-
IN
O 50
U
no 27.1
> 40
'
-------
results and an opportunity for further research. Soybean farming showed the greatest area of
uncertainty, which would be likely to be even greater if the scope of these data were expanded
beyond the United States. We also note that the estimates in Figure 9.1-3-6 are estimates of the
average supply chain GHG emissions associated with average soybean oil biodiesel. GREET
may estimate higher or lower LCA emissions for biodiesel produced from soybeans grown on a
particular farm or produced at a particular biodiesel facility.
Figure 9.1.3-6: Stochastic analysis results of USA soybean oil biodiesel by lifecycle stage in
GREET (whiskers indicate P10 and P90 values)
25
20
O
U
3B 15
c
0)
£ 10
c
o
_Q
ro
u
x
5.0
X
4.1
3.9
22.0
9.1
Soybean Farming
Soy Oil Extraction Soy Oil Transesterification
Fuel Distribution and
Combustion
Total
9.1.3.5 Land Use Change Sensitivity Analysis
As GREET is an attributional (or "supply chain") LCA model that does not endogenously
estimate indirect emissions such as those resulting from indirect land use change, GREET
incorporates a module called the Carbon Calculator for Land Use Change from Biofuels
Production (CCLUB) to account for indirect land use change emissions.247 CCLUB relies on a
selection of land use change estimates from GTAP studies conducted between 2011-2018, and
includes two corn ethanol and four soybean oil biodiesel scenarios that are described in Table 1-
1 of this document. We describe the CCLUB module in greater detail in Section 2.1 of this
document.
As a final parameter sensitivity analysis for GREET, we show a range of results
representing variations of soil organic carbon emission factors data sets and related assumptions
as options in the CCLUB module. By default, CCLUB relies on soil organic carbon emission
factors from the CENTURY model developed by Colorado State University for domestic land
use change calculations, and a separate dataset by Winrock International for international land
use change emission calculations.248 In our LUC sensitivity analysis, we present results using
both emission factors datasets where applicable, as well as varying the soil depth considered and
247 Kwon, Hoyoung, et al. (2021). Carbon calculator for land use change from biofuels production (CCLUB) users'
manual and technical documentation, Argonne National Lab, Argonne, IL. https://greet.es.anl.gov/publication-
cclub-manual-r7-2021
248 Ibid. See details about how these emission factor datasets are developed and used in the CCLUB manual.
149
-------
tillage practices. Similarly, we included results both based on assumptions about corn and
soybean crop yields increasing over time or remaining static.
CCLUB includes a forest prorating factor that is meant to adjust the forest land in GTAP
results to better align with the amount of accessible forest land as reported by the Cropland Data
Layer (CDL), a dataset developed by USDA's National Agricultural Statistics Service.249
Argonne accordingly applies this proration factor by region to the accessible forest land that
GTAP predicts will be converted in order to satisfy land needed to meet a given biofuel shock
based on a ratio of the differences between GTAP's assumed forest landcover versus what was in
USDA's CDL. This results in different amounts of assumed forest land to cropland conversions
and therefore LUC GHG emissions. We took the approach in this sensitivity analysis of
presenting results both with and without CCLUB making this forest proration factor adjustment.
GREET's default LUC scenario for corn ethanol is referred to as "Corn Ethanol 2011" in
CCLUB and is described in Taheripour et al. (2011).250 The scenario represents an increase in
USA corn ethanol production from 2004 levels (3.41 billion gallons) to 15 billion gallons (a
shock size of 11.59 billion gallons). Table 9.1.3-3 presents 20 different permutations and a range
of different emissions based on changing the assumptions for how CCLUB interprets this single
modeled GTAP scenario for land use change representing a corn shock. Argonne's pre-selected
options in CCLUB yield an estimate of 7.4 gCChe/MJ of corn ethanol for induced land use
change, while varying the assumptions in this sensitivity analysis yields a range between 6.5
gCChe/MJ to 9.7 gCChe/MJ when relying on CENTURY emission factors for domestic LUC
emissions, with the main differences coming from variations in the corn yield and tillage
practices. That estimated range expands to a high value of 16.2 gCChe/MJ if both the domestic
and international LUC emissions are based on the 2009 Winrock emissions factor data.
249 USDA National Agricultural Statistics Service (NASS) Cropland Data Layer (CDL) is available online at:
https://croplandcros.scinet.usda.gov/
250 Taheripour, F., et al. (2011). Global land use change due to the U.S. cellulosic biofuels program simulated with
the GTAP model, Argonne National Laboratory: 47.
150
-------
Table 9.1.3-3: CCLUB
Sensitivit
y Results
or "Corn Ethanol 2011" Scenario by Parameter
Select
Domestic
Emissions
Modeling
Scenario
Select
International
Emissions
Modeling
Scenario
Domestic
Emissions
Modeling
Scenario
Soil depth
considered in
modeling
Harvested
Wood
Product
(HWP)
Scenario
Tillage
Practice for
Corn and
Corn Stover
Production
Forest
Prorating
Factor
Domestic
(Data
Cell)
Foreign
(Data
Cell)
gCOze/MJ
Century
Winrock
yield
increase
30 cm
HEATH
No Till
Yes
109.6
432.7
6.7
Century
Winrock
yield
increase
100 cm
HEATH
No Till
Yes
91.5
432.7
6.5
Century
Winrock
yield
constant
30 cm
HEATH
No Till
Yes
235.6
432.7
8.3
Century
Winrock
yield
constant
100 cm
HEATH
No Till
Yes
245.7
432.7
8.4
Century
Winrock
yield
increase
30 cm
HEATH
No Till
No
146.3
432.7
7.2
Century
Winrock
yield
increase
100 cm
HEATH
No Till
No
130.9
432.7
7.0
Century
Winrock
yield
constant
30 cm
HEATH
No Till
No
274.2
432.7
8.8
Century
Winrock
yield
constant
100 cm
HEATH
No Till
No
287.4
432.7
8.9
Century
Winrock
yield
increase
30 cm
HEATH
US
Average
Yes
157.7
432.7
7.3
Century
Winrock
yield
increase
100 cm
HEATH
US
Average
Yes
162.4
432.7
7.4
Century
Winrock
yield
constant
30 cm
HEATH
US
Average
Yes
276.7
432.7
8.8
Century
Winrock
yield
constant
100 cm
HEATH
US
Average
Yes
307.9
432.7
9.2
Century
Winrock
yield
increase
30 cm
HEATH
US
Average
No
195.3
432.7
7.8
Century
Winrock
yield
increase
100 cm
HEATH
US
Average
No
203.5
432.7
7.9
Century
Winrock
yield
constant
30 cm
HEATH
US
Average
No
316.1
432.7
9.3
Century
Winrock
yield
constant
100 cm
HEATH
US
Average
No
351.2
432.7
9.7
Winrock
Winrock
871.1
432.7
16.2
GREET's default LUC scenario for soybean oil biodiesel is referred to as "Soy Biodiesel
CARB case 8" in CCLUB and is described in Chen et al. (2018)251 and Taheripour et al.
(20 1 7)252. The scenario represents an increase in U.S. soybean oil biodiesel production by 0.812
billion gallons. Table 9.1.3-4 presents eight different permutations and a range of different
emissions based on changing the assumptions for how CCLUB interprets this modeled GTAP
scenario for land use change representing a soybean shock. Argonne's pre-selected options in
CCLUB yield an estimate of 9.3 gCChe/MJ of soybean oil biodiesel for induced land use change,
251 Chen, R., Qin, Z., Han, J., Wang, M., Taheripour, F., Tyner, W., O'Connor, D., Duffield, J., 2018. Life cycle
energy and greenhouse gas emission effects of biodiesel in the United States with induced land use change impacts.
Bioresource Technology 251, 249-258. https://doi.Org/10.10.l.6/i.biortech.2017.12.031
252 Taheripour, F., Zhao, X., Tyner, W.E., 2017. The impact of considering land intensification and updated data on
biofuels land use change and emissions estimates. Biotechnol Biofuels 10, 191. https://doi.org/10.1186/sl3068-Q17-
0877-v
151
-------
while varying the assumptions in this sensitivity analysis yields a range between 9.0 gCChe/MJ
to 9.6 gCChe/MJ when relying on CENTURY emission factors alone for domestic LUC
emissions, with the variations primarily again coming from assumed soybean yield and tillage
practices. That estimated range expands significantly to a high value of 21.5 gCChe/MJ if both
the domestic and international LUC emissions are based on the 2009 Winrock emissions factor
data.
Table 9.1.3-4: CCLUB Sensitivity Results for "Soy Biodiesel CARB case 8" Scenario by
Parameter
Domestic
Emissions
Modeling
Scenario
International
Emissions
Modeling
Scenario
Harvested
Wood Product
(HWP)
Scenario
Tillage Practice
for Corn and
Corn Stover
Production
Forest
Prorating
Factor
Domestic
Emissions
Foreign
Emissions
gCOze/MJ
Century
Winrock
HEATH
No Till
Yes
24.4
1,105.7
9.0
Century
Winrock
HEATH
No Till
No
53.8
1,105.7
9.2
Century
Winrock
HEATH
US Average
Yes
68.2
1,105.7
9.3
Century
Winrock
HEATH
US Average
No
98.6
1,105.7
9.5
Winrock
Winrock
1,613.7
1,105.7
21.5
Both the corn ethanol and soybean oil biodiesel LUC sensitivity analysis results show
that even relying on the same LUC results from GTAP can yield significantly different emission
results based on assumption differences such as the emission factors used and other key data sets
or data interpretations.
We do not present results in this section with the intention of concluding what a range of
potential emissions the GREET model can be for corn ethanol and soybean oil biodiesel, as that
is outside the scope of this analysis. Instead, we mean to illustrate the variation in results that
come from key assumptions and where the model framework demonstrates the most variation in
its estimates based on those assumptions.
Across the various sensitivities we performed for GREET, corn ethanol and soybean oil
biodiesel each relied on a single LUC scenario provided by GTAP and interpreted by CCLUB.
While other models showed a significant variation in LUC impacts based on differing sensitivity
assumptions, the area of LUC was held constant for GREET. Instead, these sensitivities
highlighted variability associated with other assumptions. Our parameter and stochastic
sensitivities demonstrated the importance to emissions that corn and soybean yields have on
results and how they vary considerably across the country (they also vary over time). Data based
on industry surveys also suggested that there is still a significant range of efficiencies for energy
inputs both on fields and in biofuel facilities. On LCA allocation methods, we demonstrated how
impactful decisions are in emissions accounting for ethanol or biodiesel versus coproducts.
Similar to what is shown in the next section (Section 9.2), the soil carbon assumptions illustrated
in our GREET LUC sensitivity analysis had a relatively large impact based on the datasets used
to represent LUC emissions from static GTAP scenarios. Finally, some of these same areas seem
important for additional research. The uncertainty around farming chemical use for example was
also seen with our GCAM sensitivities.
152
-------
9.2 Soil Organic Carbon Sensitivities
Land use change emissions estimation is an important component of crop-based biofuel
lifecycle analysis, as demonstrated by the results we present in Sections 6.7 and 7.7. Estimates of
LUC emissions from the conversion of other land types to cropland vary to some extent based on
the type of land being converted. But beyond this another important area of variability is the
assumed carbon density of lands and the quantity of carbon emitted or sequestered when land
transitions from one state to another. The magnitude of this carbon exchange varies based on
climate, soil type, vegetation type, soil microbial activity, and numerous other factors. At the
time of the March 2010 RFS rule, most model soil carbon assumptions were based on field scale
sampling of soils and other estimation techniques, which were then extrapolated and applied to
much larger areas of land than their empirical samples covered. A small number of global
satellite-based data sets, such as the MODIS-based Winrock data we used to estimate LUC
emissions from the FAPRI model, also existed, but were relatively new. Over the last decade,
empirical satellite-based datasets have become more numerous and sophisticated, necessitating
revisitation of this area of science.253
We observed in Section 9.1.1 above that the GCAM results produced for this exercise are
sensitive to the assumed value of soil carbon density input parameters. For the analysis described
in Section 9.1.1, we stochastically varied the soil carbon and vegetation densities assumed in
GCAM, with independent distributions for each land category. The sensitivity analysis described
in this section is different, as it tests the influence of using different soil carbon data sources,
described below, to determine the baseline soil carbon densities.
The soil carbon assumptions of GCAM rely on a simple carbon cycle model that tracks
cohorts of soil and vegetation carbon over time, starting in 1750, the first spin-up year. In
previous versions of GCAM, average terminal carbon stocks (above and below ground
vegetative carbon and soil carbon) for each land use type were assumed exogenously based on
aggregate data, not differentiated by GCAM land use region. More recently, carbon stock data
acquisition and modeling capabilities have improved, and current vegetation and soil carbon
stock maps can be generated using sophisticated mathematical and statistical techniques. In an
additional set of runs, we tested the impacts of different soil carbon stocks on the land use
change emissions in GCAM.
The GCAM results presented in the core scenarios in Sections 5-7 use globally gridded soil
carbon stock data from SoilGrids 2017254 (30 cm depth) and vegetative carbon stock data from
Spawn et al. (2020).255 SoilGrids is based on soil profile observations from the WoSIS database
that have been interpolated via random forest machine algorithms to 250 m grid cells. Because
GCAM represents land at a water basin level, the model needs only one carbon stock input per
253 For more information on carbon stock datasets see: Spawn-Lee, S., "Carbon: Where is it and how can we know?"
EPA Workshop on Biofuel Greenhouse Gas Modeling, 2022. https://www.epa.gov/svstem/files/documents/2022-
03/bioftiel~gfag~model~worksfaop~measiire~map~soit~earbon~2022~02'~28.pdf
254 Hengl, T., Mendes de Jesus, J., Heuvelink, G. B., Ruiperez Gonzalez, M., Kilibarda, M., Blagotic, A.,. &
Guevara, M. A. (2017). SoilGrids250m: Global gridded soil information based on machine learning. PLoS one,
12(2), e0169748.
255 Spawn, S.A., Sullivan, C.C., Lark, T.J. et al. Harmonized global maps of above and belowground biomass carbon
density in the year 2010. Sci Data 7, 112 (2020).
153
-------
land type, per water basin.256 Summary statistics (the third quartile) were calculated for every
land use type in each basin to represent the steady state soil carbon stock at the beginning of
environmental simulation in 1700.257
To test the sensitivity of GCAM results to soil carbon stock assumptions, we tested
GCAM using 3 additional soil C datasets, as shown in Table 9.2-1. The Harmonized World Soils
Database (HWSD) uses a "paint by number" approach to categorize carbon stocks. The map was
built on several different global and regional expert-informed soil databases (SOTER, ESD, Soil
Map of China, WISE), built on a 30 arc-second resolution (approximately 1 km), and reprojected
with a grid scale size of 250 m. Each grid cell has estimates informed from these databases, with
areas lacking data filled in using machine learning estimates. In some countries, the soil
boundaries are defined polygons, with the center value assumed to be the value for the entire
polygon (hence the description as a "paint by number" approach). This type of map can result in
distinct boundaries at political or geological boundaries.
Table 9.2-1: Soil carbon stock datasets used for sensitivity analysis in GCAM
Dataset
Method
Depth
Resolution
Harmonized World Soils
Database (HWSD)258
Professionally derived
"Paint by Number"
30 cm
30 arc-second
Food and Agricultural
Organization Global Soil
Organic Carbon Map (FAO
GLOSIS)259
Combination raster of
country driven soil
maps
30 cm
30 arc-second
SoilGrids 2017260
Random forest
machine learning
30 cm
250 m
SoilGrids 2020261
Random forest
machine learning
30 cm
250 m
The FAO GLOSIS (Global Soil Information System) map is based on data collected and
reported by national institutions. The countries, under the guidance of the Intergovernmental
Technical Panel on Soils and the Global Soil Partnership Secretariat, used a uniform
methodology with modern soil digital mapping tools to create national maps, which were then
standardized to the global area. These maps were built on a 30 arc-second resolution
(approximately 1 km), and reprojected with a grid scale size of 250 m. Over 63 percent of the
256 Further description of the land allocation module in GCAM is available at: https://igcri.githiib.io/gcam~
doe/tand.fatnit
257 Since GCAM requires estimates of soil carbon from 1700, and the soil data we have represents modern day, the
moirai framework utilized the Q3 (third quartile) SoilGrids data, to represent a historic baseline.
258 Wieder, W.R., J. Boehnert, G.B. Bonan, and M. Langseth. 2014. Regridded Harmonized World Soil Database
vl.2. Data set. Available on-line [http://daac.ornl.gov] from Oak Ridge National Laboratory Distributed Active
Archive Center, Oak Ridge, Tennessee, USA. http://dx.doi.org/10.3334/ORNLDAAC/1247
259 FAO and ITPS. 2018. Global Soil Organic Carbon Map (GSOCmap) Technical Report. Rome. 162 pp.
https://www.fao.Org/3/I8891EN/i8891en.pdf
260 Hengl, T., Mendes de Jesus, J., Heuvelink, G. B., Ruiperez Gonzalez, M., Kilibarda, M., Blagotic, A.,. &
Guevara, M. A. (2017). SoilGrids250m: Global gridded soil information based on machine learning. PLoS one,
12(2), e0169748.
261 Poggio, L., de Sousa, L. M., Batjes, N. H., Heuvelink, G. B. M., Kempen, B., Ribeiro, E., and Rossiter, D.:
SoilGrids 2.0: producing soil information for the globe with quantified spatial uncertainty, SOIL, 7, 217-240, 2021.
154
-------
world map is based on country submissions. Countries that did not participate were filled in
using the SoilGrids 2017 map (1.9 percent of the world), and the remainder were calculated
using the Global Soil Partnership Secretariat partnerships and gap filling.
SoilGrids 2020 is an update of SoilGrids 2017. The SoilGrids 2020 estimate includes
more soil observations and a different set of environmental covariates than SoilGrids 2017. This
created a different interpolation of the data to a 250 m grid cell level. This method is more
computationally intensive than the method used for SoilGrids 2017, so the carbon stock is only
available for 0-30 cm depth. One benefit of SoilGrids 2020 over SoilGrids 2017 is that the
methods used to interpolate the SoilGrids 2017 map created some overestimates of SOC,
especially in the far northern latitudes (60-90°N).262 However, the soil carbon levels for the rest
of the world tended to be lower than most other soil carbon mapping estimates, so both 2017 and
2020 SoilGrids maps provide different information. We include SoilGrids 2017 in our analysis
because it is currently the default soil carbon dataset in GCAM v6.
In GCAM, land use change emissions are determined by the amount of land use change,
the location of land use change, and the difference in carbon stock between the starting and
ending land types. GCAM does not use soil carbon stock information to determine the types and
locations of land that change. Therefore, the quantity and location of land use change did not
vary across the runs, and differences in emissions are entirely based on differences in soil carbon
stock assumptions. Figure 9.2-1 shows the global emissions from land use change in the
reference case for each set of soil carbon stock assumptions. SoilGrids 2017 produces the highest
emissions and SoilGrids 2020 produces the lowest emissions.
Figure 9.2-1: Global emissions from land use change in the reference case using four soil
carbon datasets
o
4K
2K
OK
2020 2025 2030 2035 2040 2045 2050
262 Tifafi, M„ Guenet, B.. Hatte, C. (2018), Large differences in global and regional total soil carbon stock estimates
based on SoilGrids, HWSD, andNCSCD: Intercomparison and evaluation based on field data from USA, England,
Wales, and France. Global Biogeochemical Cycles, 32, (1), 42-56
155
-------
In Figure 9.2-2, we calculated the CI, as described in Sections 6.7 and 7.7. The CI is
based on the difference between the corn ethanol or soybean oil biodiesel scenario and the
reference case. The FAO GLOSIS dataset produces the lowest CI results, even though SoilGrids
2020 had the lowest LUC emissions in the reference case. This is because the corn ethanol and
soybean oil biodiesel scenarios had land use change in different locations than the reference case.
The CI of land use change varies greatly across the runs, from 9-31 kgC02e/MMBTU for corn
ethanol and 36-63 kgC02e/MMBTU for soybean oil biodiesel. For each of the soil carbon stock
assumptions, the CI from land use change is around twice as high for soybean oil biodiesel as for
corn ethanol.
Figure 9.2-2: Carbon intensity from land use change emissions for the corn ethanol shock
and the soybean oil biodiesel shock using a range of soil carbon datasets
60
50
=J
4—1
CD
E40
0)
o30
u
en
jx.
20
10
0
Soil C dataset
¦ HWSD
¦ FAO GLOSIS
¦ SoilGrids 2017
¦ SoilGrids 2020
Corn Shock(1BG) Soy Shock(1BG)
We draw no conclusions here about which soil carbon data set is most appropriate to use
for biofuel lifecycle analysis in GCAM or any other modeling framework. While this is a valid
scientific question, it was beyond the scope and resources of this exercise. Rather, our intention
is to show that the choice of soil carbon stock assumption, among commonly used datasets, can
have a large impact on the modeled CI of corn ethanol and soybean oil biodiesel within a given
modeling framework. Further work will be needed to explore how different soil carbon datasets
impact the results of other models, and to determine which soil carbon dataset is most
appropriate to use in this context.
9.3 Land Conversion Elasticity Sensitivities
In the soybean oil biodiesel results presented in Section 7, one of the major differences
between the ADAGE results and the results of the other models is the emissions from land use
change. We ran a set of sensitivity scenarios to determine whether changing the model
parameters changes the result that a large amount of forestland is converted to cropland.
156
-------
As explained in Section 2.5, the direction and magnitude of land use change in ADAGE
is determined by differences in prices between land types (which are in part driven by
differences in net primary production [NPP]) and fixed factor elasticities between the land types.
In the results presented above, the fixed factor elasticity from pasture to cropland is the same as
that from managed forest to cropland (Table 9.3-1). This means if prices of pasture and forest are
equal to each other, it is equally easy to convert forest to cropland and pasture to cropland. In
contrast, the fixed factor elasticity from cropland to pasture is higher than the fixed factor
elasticity from cropland to managed forest, meaning that given equal prices, more cropland
would convert to pasture than to managed forest. In these scenarios, because of assumptions of
NPP declining for forest and rising for pasture over time in key non-USA soybean-producing
regions, the price of managed forest declines while the price of pasture rises. Since the fixed
factor elasticity of converting these two land types to cropland is assumed to be equal, more of
the lower cost land, i.e., managed forest is converted in non-USA regions in these results.
Table 9.3-1: Fixed factor elasticity between land types in ADAGE core scenarios
Land Conversion
From
Cropland
Pastureland
Managed
Forestland
Natural
Forestland
Grassland
Cropland
0.26
0.26
Pastureland
0.3
0.02-0.509
To
Managed Forestland
0.15
0.02-0.509
Natural Forestland
0.15
0.15
Grassland
0.15
0.15
0.15
Note: Elasticity values for agricultural lands converting to other land types are assumed to be the same for all
regions. Elasticities for natural land conversion to agricultural land vary by region and range from 0.02 to 0.509.
We conducted a sensitivity analysis on the fixed factor elasticities between land types to
assess the impact of making it more difficult to convert forest to cropland than pasture to
cropland. The alternative elasticity values used in this sensitivity analysis are shown in Table
9.3-2. In this sensitivity, the fixed factor elasticities from pasture/managed forest to cropland
were swapped with the fixed factor elasticities from cropland to pasture/managed forest. In this
scenario, the fixed factor elasticity from pasture to cropland is twice as large as the fixed factor
elasticity from managed forest to cropland, making it easier to convert pasture than forest to
cropland.
157
-------
Table 9.3-2: Fixed factor elasticity between land types in ADAGE sensitivity runs
Land Conversion
From
Cropland
Pastureland
Managed
Forestland
Natural
Forestland
Grassland
Cropland
0.3
0.15
Pastureland
0.26
0.02-0.509
To
Managed Forestland
0.26
0.02-0.509
Natural Forestland
0.15
0.15
Grassland
0.15
0.15
0.15
Note: Elasticity values for agricultural lands converting to other land types are assumed to be the same for all
regions. Elasticities for natural land conversion to agricultural land vary by region and range from 0.02 to 0.509.
We focus on the results of the soybean oil biodiesel scenario. As shown in Figure 9.3-1,
the new runs ("Sensitivity") have more additional soybean cropland than the runs described in
Section 7 ("Core"). In the sensitivity runs, the soybean yield does not increase as much as in the
core runs, so more cropland is needed to produce soybeans for biodiesel. The sensitivity runs
also show a greater increase in total cropland. There is less shifting of land from other crop types
to soybean.
Figure 9.3-1: Difference in cropland area by crop type (million hectares) in the soybean oil
biodiesel shock relative to the reference case in 2030 for the original ADAGE runs ("Core")
and the fixed factor elasticity sensitivity runs ("Sensitivity")263
2030
USA
Non-USA
Land Use
¦ Corn
¦ Soybean
¦ Sugar Crops
¦ Wheat
¦ Other Grains
¦ Other Oil Crops
¦ Other Crops
Core
Sensitivity
Core
Sensitivity
In the sensitivity runs, there is a large change in the type of land converted to cropland,
relative to the core runs (Figure 9.3-2). In the USA region, managed pasture is still the primary
263 Horizontal lines show the net change in cropland.
158
-------
land type that is converted to cropland. However, in the non-USA regions, land is converted
from pasture and grassland rather than forest. Even though prices and production of the land
types did not change in this sensitivity, decreasing the land conversion elasticity of forest to
cropland resulted in a large reduction in the amount of forest conversion.
Figure 9.3-2: Difference in land use (million hectares) in the soybean oil biodiesel shock
relative to the reference case in 2030 for the original ADAGE runs ("Core") and the fixed
factor elasticity sensitivity runs ("Sensitivity")
2030
LD
CO
o
JZ
C/l
>
o
in
1.50
1.00
0.50
0.00
-0.50
-1.00
-1.50
USA
Non-USA
Land Type
¦ Managed Forest
¦ Unmanaged Forest
¦ Grassland
¦ Managed Pasture
¦ Unmanaged Pasture
Shrubland
¦ Other Arable Land
¦ Other Not Arable
¦ Cropland
Core Sensitivity | Core Sensitivity
As a result of the change to the land conversion elasticity, the estimated CI from land use
change decreased substantially, from 295 kgC02eq/MMBTU to 33 kgC02eq/MMBTU (Table
9.3-3). In the sensitivity runs, there is more total land use change, but much less emissions from
land use change. This emphasizes that the type of land converted and the carbon stock of the
converted land plays a major role in the emissions from land use change.
159
-------
Table 9.3-1: Carbon intensity of soybean oil biodiesel and corn ethanol
(kgCOieq/MMBTU) calculated using emissions reported by each ADAGE run
Soybean oil biodiesel
Corn ethanol
Core
Sensitivity
Core
Sensitivity
Energy Sector
-28
-30
-15
-17
Sector -
Crop Production
7
8
14
14
specific
Livestock Sector
0.7
0.7
0.1
0.1
emissions
Other
1
1
1
1
Land Use Change
295
33
-1
-1
Totals
Agriculture, forestry,
and land use
303
41
14
14
Global GHG Impact
276
12
-1
-3
The corn ethanol sensitivity scenario similarly shows less corn yield increase than the
core corn ethanol scenario, and more additional cropland. However, the core corn ethanol
scenario results in conversion of pasture to cropland, and this does not change in the sensitivity.
The estimated CI for the corn ethanol scenarios are shown in Table 9.3-3. The land use change
CI in the sensitivity is similar to the core run.
These results illustrate the importance of considering land parameter assumptions in the
models. We do not make conclusions here about which of these sets of results is more correct.
Rather, these results show that if there are assumptions in a model that allow more forest to be
converted in a biofuel scenario, then the emissions can be much higher. Future work could
explore whether there are other similarly important parameters in the models. For cases where
data are not available to set a parameter value (as is often the case for elasticity values), future
work could involve developing methods to use historical data to inform the choice of parameter
value.
9.4 Summary of Parameter Sensitivities
In this section we discussed the results of five sensitivity experiments testing the
influence of parameter input values on biofuel GHG impact estimates, including stochastic
analyses of GCAM, GLOBIOM, and the GREET model, a separate soil organic carbon
sensitivity analysis of GCAM, and a land conversion elasticity sensitivity of the ADAGE model.
Stochastic parameter experiments with GCAM indicate the assumptions relating to soil
carbon stocks, the ease of substitution between land and crop types, and the N2O emissions
intensity of agriculture are influential parameters for corn ethanol and soybean oil biodiesel
GHG impact estimates. The parameter controlling substitution between the non-USA regions
refined oil and biodiesel is also influential for the soybean oil biodiesel GHG estimates.
160
-------
A similar stochastic experiment with GLOBIOM considering only soybean oil biodiesel
GHG impact estimates finds that a different set of parameters are the most influential. For
example, the GLOBIOM experiment finds biomass carbon stock assumptions to be influential,
whereas these assumptions were not identified as influential by the stochastic GCAM
experiment. Other parameters that registered as influential in the GLOBIOM stochastic
experiment but not in the GCAM stochastic experiment include assumptions related to tropical
peat soil, substitution between vegetable oils, and yield elasticities for corn and soybeans.
The land conversion elasticity sensitivity experiment with the ADAGE model finds that
land use change GHG estimates for soybean oil biodiesel are highly sensitive to the assumed
fixed factor elasticities for forest and pasture to cropland. These results indicate that parameter
influence on biofuel GHG impact estimates is model dependent, i.e., a set of parameters that is
influential in one model may not be influential in another model.
The stochastic analyses conducted with the GREET model, using a specific set of
assumed parameter uncertainty distributions, suggest that supply chain LCA estimates for corn
ethanol are more sensitive to parameter input values than such estimates for soybean oil
biodiesel. Scenario sensitivity analyses with the GREET model indicate that corn ethanol and
soybean oil biodiesel estimates are more sensitive to coproduct allocation choices and
assumptions related to land conversion GHG emissions factors.
A parameter sensitivity analysis with different soil carbon datasets in GCAM indicates
that the initial steady state soil carbon conditions have a relatively large influence on land use
change GHG estimates. This suggests that estimates from the same model are likely to change
over time as science evolves and new data sets become available.
10 Summary of Findings and Future Research
Through this model comparison exercise, we aimed to move the science forward on
analyzing the lifecycle GHG impacts of the increased use of biofuel, understand model
differences, and examine how those differences impact model results. As described in Section 1,
this effort is consistent with recommendations from the NASEM report, "Current Methods for
Life Cycle Analyses of Low-Carbon Transportation Fuels in the United States," which
emphasizes the importance of comparing results across multiple economic models and
considering uncertainty.264 The detailed results and insights from this model comparison exercise
are explained in the sections above. This section summarizes our main findings, including areas
of similarity and difference across the models considered in this exercise, and potential areas for
future research.
264 NASEM recommendation 4-2: "Current and future LCFS [low carbon fuel standard] policies should strive to
reduce model uncertainties and compare results across multiple economic modeling approaches and transparently
communicate uncertainties." NASEM recommendation 4-3: "LCA studies used to inform policy should explicitly
consider parameter uncertainty, scenario uncertainty, and model uncertainty." National Academies of Sciences,
Engineering, and Medicine 2022. Current Methods for Life Cycle Analyses of Low-Carbon Transportation Fuels in
the United States. Washington, DC: The National Academies Press, https://doi.org/.1.0. .1.7226/26402.
161
-------
Some of these observations and findings are relevant only to certain models, based on
their characteristics and areas of coverage. As explained throughout this document, not every
model considered in this study includes all sectors of the economy or all types of interactions
discussed in this section. For example, we do not discuss GREET in any of our findings related
to economic interactions, nor do we discuss GREET and GLOBIOM in any of our findings
related to the energy sector. Models that are not listed in the findings of each subsection in this
summary do not model the features described in that subsection.
Framework Differences
Supply chain LCA models produce a fundamentally different analysis than economic
models. Supply chain LCA models generate detailed and transparent fuel production emissions
estimates. However, they do not evaluate all the indirect emissions associated with a change in
biofuel consumption. The economic models in our comparison are broad in scope, but they lack
certain supply chain details and are associated with greater variability. Their complexity makes it
difficult to identify the precise reasons that estimates vary across the models.
The emissions impacts observed in this exercise do not remain static over time in
frameworks with the ability to model dynamic change. The dynamic models considered in this
exercise, ADAGE, GCAM, and GLOBIOM, all agree that land use, crop production, livestock
markets, and energy markets would all be expected to adjust over time in response to a biofuel
shock, with cascading impacts on GHG emissions. Dynamically modeling the impacts of
biofuels over time results in different model solutions for GHG emissions than what would
be predicted by more simply extrapolating results in a single time step forward through
post hoc estimation. We make no conclusions about whether dynamic or static models are more
appropriate for different applications, but it is important to address the fact that they arrive at
different conclusions and to robustly consider the time period used for biofuel LCA modeling.265
Land Use Change and Emissions
Land use change and associated emissions magnitudes vary across the range of scenarios
presented in this exercise. Results between models show differences in the types of land which
transfer into cropland status between the reference and biofuel shock scenarios. Our Monte Carlo
and land conversion elasticity parameter sensitivity analyses show that these estimates can also
vary within individual models, depending on the parameter assumptions used. There are several
important factors in explaining these differences in LUC estimates among and within models.
Models use different economic equations, mathematical decision frameworks,266 and
assumptions to estimate which types of land to convert, in what quantities, and in which regions.
The quantities and location of LUC intersect with the global commodity market dynamics
discussed above. Differences in mathematical representations of LUC may lead to model results
which convert primarily one type of land or, conversely, results which spread the LUC impact
265 It is also important to consider the model reference case assumptions, including model projections into the future.
The parameter sensitivity analyses discussed in Section 9 suggest several concrete examples, such as the projection
of future crop yields, which critically influence model results.
266 For example, ADAGE and GTAP use a CES structure, GCAM uses logit nests, and GLOBIOM uses a global
gridded system.
162
-------
across multiple land types. Neither of these strategies necessarily leads to higher or lower LUC
emissions relative to the other. For example, the ADAGE modeling results demonstrate that
concentrating LUC to one type of conversion may lead to relatively larger LUC emissions
estimates (as shown in the soybean oil biodiesel results) or relatively smaller LUC emissions
estimates (as shown in the corn ethanol results). Within models, our sensitivity analyses
demonstrate that input parameter assumptions, such as those described in Sections 9.1.1 and 9.3,
may alter economic decisions and thus affect which land types are selected for conversion. This
model comparison and the associated sensitivity analyses have indicated that assumptions about
the ease of land substitution, especially from carbon-rich lands, remain a critical area of
uncertainty in biofuel LCA modeling. Future modeling efforts should robustly quantify this
uncertainty using either the types of methods described in this exercise or other rigorous
methods. This exercise highlights that inclusion of land use change emissions is critical for
biofuel lifecycle analysis and that frameworks must have the ability to robustly quantify
uncertainty in land use change and LUC emissions.
Further, spatial resolution in the land sector varies substantially across models and this
affects the scale at which economic land conversion decisions are made. This major area of
difference among models is critically tied to the scope of each model and the associated
computational burdens of land use modeling. It is unlikely that the CGE models, which must
necessarily resolve equations for more economic sectors, can achieve the spatial resolution
present in PE models and IAMs. However, the uncertainties created by coarser spatial resolution
may be quantifiable through targeted uncertainty analysis. Uncertainty also still exists at the
resolution represented by PE models and IAMs given that these LUC results are necessarily
estimates of the sum of economic decisions made by multiple actors. We conclude that there is
no one correct level of spatial resolution for biofuel LCA modeling. Sensitivity and
uncertainty analysis will be critical at all scales.
The economic models included in this exercise also restrict land conversion to varying
degrees, and the differences in assumptions across models are especially large for the most
carbon-rich arable lands (i.e., natural forests and grasslands). However, these assumptions are
also uniformly exogenous and previous literature has demonstrated that, to at least some extent,
they can be aligned across modeling frameworks. Future research could explore this space and
test whether LUC estimates across models become more similar when similar categories and
quantities of lands are available for conversion to cropland.
Additionally, the models use different assumptions about the carbon stocks of the
different land types, resulting in different emissions from land use change. A sensitivity analysis
using GCAM shows that when different soil carbon stock assumptions are used, there are large
differences in the resulting land use change emissions, even though the type and amount of land
converted is the same in each run. The stochastic parameter sensitivities conducted with GCAM,
GLOBIOM, and GREET also demonstrate that assumptions about soil carbon exchange from
LUC may substantially impact emissions results. Addressing variability and uncertainty in
soil carbon content globally and regionally will be critical to future biofuel LCA efforts. A
potential area for future research is to align carbon stock assumptions across multiple models to
better understand the relative impacts of land use change amount/type and carbon stocks on land
use change emissions.
163
-------
Energy Market Impacts
The models that include energy market impacts (ADAGE, GCAM, and GTAP) all
estimate significant indirect effects on fossil and/or bio-based energy consumption in the USA
and non-USA regions in both the corn ethanol and soybean oil biodiesel shocks. The results from
these models are in broad agreement that global displacement of refined oil267 consumption due
to the increase in biofuel consumption is estimated to generate net global energy emissions
savings. However, the amount of refined oil displaced globally was not equal to the increase in
biofuel consumption on an energy basis (i.e., a 1:1 displacement). This finding has broad
relevance to biofuel LCA because modeling efforts using frameworks which do not include an
energy sector generally assume 1:1 displacement by default. All three models in this study with
energy sectors show smaller global refined oil savings than would be expected from a 1:1
displacement. There are some directional differences regarding the impact in the USA region.
The ADAGE and GTAP results show less domestic refined oil displacement than would be
expected from a 1:1 displacement, while the GCAM results show more domestic refined oil
displacement than would be expected from a 1:1 displacement. However, the larger driver of the
global result is refined oil and biofuel consumption in the non-USA regions. Non-USA refined
oil consumption increases in the results from each of these models as a result of the shock. In
ADAGE and GCAM, there are significant changes in non-USA biofuel production and
consumption as well. In the ADAGE soybean oil biodiesel scenario, the non-USA regions
collectively produce more biodiesel and consume less of it, exporting that fuel to the USA region
instead. This reduced biodiesel consumption increases demand for fossil fuels. The increased
production is associated with agricultural sector emissions. The GCAM results show impacts on
non-USA biofuel production and consumption as well, particularly sugar crop ethanol in the corn
ethanol scenario, and soybean oil biodiesel in the soybean oil biodiesel scenario. These results
also show substantial changes in biofuel trade to and from the USA region in response to the
shocks. The results across all three models collectively indicate that the assumption of 1:1
displacement of refined oil for biofuel may be insufficient to capture the energy sector
impacts of biofuels; consequential modeling of the energy sector is an appropriate
methodology for capturing these impacts.
This insight illustrates the importance of including indirect energy market impacts in a
modeling framework. The ADAGE, GCAM, and GTAP results consistently indicate that the
assumption of a 1:1 refined oil displacement may be an overestimate of global fossil fuel
emissions savings. This becomes a crucial issue for biofuel lifecycle analysis, firstly, because
smaller fossil fuel emissions savings increase the estimated emissions intensity of the biofuel
being modeled and, secondly, because increased non-USA production of biofuels is associated
with emissions as well. However, further sensitivities would be needed to better understand the
driving factors behind the differences in the fossil fuel displacement across the models.
Global Trade
Global trade plays an important role in modeled emissions results from both the land and
energy sectors of these frameworks. Model results from the economic models considered in this
267 In these models, refined oil is an aggregation of all refined petroleum products, including gasoline and diesel.
164
-------
exercise consistently demonstrate that biofuel shocks can impact agricultural commodity trade
and energy trade in important ways. These include impacts on trade in refined oil and biofuels,
soybean meal and DDG feed products, and vegetable oils, among others. These changes in terms
of trade lead to differences in the energy emissions savings estimated by the models as well as
differences in the quantity of non-USA land use change estimated by the models. There is
general agreement among the economic models that these trade-driven impacts will occur to
some degree. However, despite the uniform agreement on the importance of trade-driven impacts
across the economic models included in this exercise, these models show different degrees of
trade responsiveness, which leads to results of differing magnitudes. Model trade structure and
assumed flexibility critically influence the modeled emissions results.
Commodity Substitutability
A second key factor, intertwined with trade, is commodity substitutability. Results in this
exercise from ADAGE, GCAM, GLOBIOM, and GTAP align in estimating commodity
substitution as a significant part of their scenario solution. As our sourcing analyses in Sections
6.1 and 7.1 above demonstrate, the degree to which this substitution occurs varies across models.
However, results from all of the models support two overarching findings: first, that estimates of
indirect GHG impacts are sensitive to whether and how substitution interactions are considered
and, second, that uncertainty in the ease of commodity substitution at different price points must
be considered. Key interactions include the substitutability of: biofuels for fossil fuels, one
biofuel for another, DDG and soybean meal for other feed products, and soybean oil for other
vegetable oils. Our modeling exercise has demonstrated that these commodity substitutability
relationships critically impact overall GHG emissions results from biofuel LCA modeling.
We summarize these critical impacts further below.
Crop and Coproduct Consumption by End Use
The results of the corn ethanol and soybean oil biodiesel scenarios also show significant
effects on end uses of biofuel feedstocks and coproducts across ADAGE, GCAM, GLOBIOM
and GTAP, most notably effects on corn, DDG, and soybean meal animal feed use and soybean
oil food use. In the corn ethanol scenario, the model results consistently show a decrease in corn
consumption for feed use and an increase in DDG consumption. However, the model results
differ crucially in their estimates regarding the location of DDG consumption (i.e., USA vs non-
USA regions) as well as the degree of displacement of other types of feed. Similarly, in the
soybean oil biodiesel scenario, the model results show an increase in soybean meal268 production
and use for feed. The models all estimate this influx of soybean meal will lead to a global
increase in feed use on a mass basis. However, the models differ regarding the location of
soybean meal production and the degree of displacement of other types of feed. Increased use of
DDG or soybean meal for feed can result in lower land use change emissions if these coproducts
displace crops for feed use. On the other hand, increased use of DDG or soybean meal for feed
can result in higher livestock sector emissions if their use causes an increase in total feed use,
rather than replacing other types of feed. Exploring the emissions impact of DDG and soybean
meal consumption location on overall GHG results is a potential area of future research, and one
which is closely related to further research into model commodity trade behavior more generally.
268 In ADAGE, the soybean meal is included in the aggregated "other oil seed meal" category.
165
-------
It is clear however that explicit modeling of the global livestock sector, including global feed
markets, is an important capability for estimating the emissions associated with an increase
in biofuel consumption. Modeling efforts which do not include these economic dynamics
exclude both critical drivers of overall GHG emissions and critical sources of uncertainty in
GHG modeling results.
In the soybean oil biodiesel scenario, the models differ in the amount of food
displacement. ADAGE results do not show any impact on food consumption. On the other hand,
GCAM, GLOBIOM, and GTAP results all show a decrease in the amount of soybean oil used for
food. In the GTAP results, a very small amount of the soybean oil is replaced by other oils; these
results also show an overall reduction in crops consumed for food. GTAP results also show a
decrease in soybean oil used for other uses (e.g., processing into other products) that is not
replaced by other oils. In the GCAM and GLOBIOM results, there is also a decrease in soybean
oil for food use. However, a major difference between these results and the GTAP results is that
the GCAM and GLOBIOM results show much greater replacement of soybean oil in the food
market with palm oil, rapeseed oil, and/or other crop oil, whereas the GTAP results show very
little replacement of soybean oil with other oils. The degree of substitution varies between
GCAM and GLOBIOM, with GLOBIOM results showing a net decrease in consumption of
crops for food, and GCAM results showing a nearly net zero change in consumption of crops for
food. Substitution of soybean oil with other oil types could result in a reduction of land use
change emissions from soybean production because less new soybean oil production is needed
for the biofuel shock. However, substitution of soybean oil with other vegetable oils could also
result in increased emissions from land use change.269 The effect of the number of vegetable oil
substitutes in a model on the lifecycle results, and the degree of substitution among feed
commodities and food commodities, particularly in the non-USA regions, is a potential area for
future study. Inclusion of explicit global vegetable oil competition is critical to biofuel
lifecycle analysis results because this competition affects the quantity and location of
estimated LUC emissions impacts.
Feedstock Production
Both intensification and extensification of corn and soybean feedstock production occur
across ADAGE, GCAM, GLOBIOM, and GTAP results in response to changing commodity
prices. In each of these models, extensification, including crop shifting, contributes to more of
the biofuel sourcing than intensification. All four models estimate yield increases of corn in the
corn ethanol scenario and soybeans in the soybean oil biodiesel scenario, but these increases are
small relative to the reference case yields. One factor could be that our volume shocks are not
large enough to induce much change in corn and soybean prices; indeed, the feedstock crop price
changes in these scenario results appear fairly small across models. In our soybean oil biofuel
volume sensitivity scenario, the models appear fairly stable in this area with respect to the size of
the shock, suggesting that shock size might not have significant influence on model yield
response. However, further research using a wider range of shock sizes and reference case
assumptions could test this hypothesis more rigorously than we have been able to in this
exercise.
269 For example, land use change to produce palm oil could result in increased emissions, particularly if the land
converted is peat land.
166
-------
We can observe generally that the models considered in this exercise do not see yield
improvements as a primary strategy for supplying additional biofuel feedstock, given our
scenario assumptions. Rather, feedstock crop extensification, including crop shifting, appears to
be relied upon more than intensification to increase the net supply of biofuel feedstock for
biofuel production across the economic modeling results presented in this exercise. This finding
appears to be robust across a wide range of uncertainty analyses. However, that is not to say crop
yield assumptions do not affect the results. Indeed, our parametric sensitivities do suggest that
crop productivity assumptions may be influential, though other parameters appear to be more
influential. Further research could better define this influence. The ability to endogenously
consider tradeoffs between intensification and extensification is an important capability for
estimating the emissions associated with an increase in biofuel consumption.
Soybean oil biodiesel and corn ethanol results vary
The models included in this study show greater diversity in feedstock sourcing strategies
for soybean oil biodiesel than they do for corn ethanol, and this wider range of options leads to
greater variability in the GHG results. There are several important reasons for this greater
diversity of strategies, which were explored throughout this document. For example, compared
to the corn ethanol results, there is less agreement among the models about where in the world
soybean oil biodiesel production would change in response to a change in USA region soybean
oil biodiesel consumption. Because of these differences in sourcing strategy, the model results
differ regarding the amount and location of soybean oil production, vegetable oil and biodiesel
trade, and land use change impacts of the shock.
Much of the new production of corn and corn ethanol in the corn ethanol shock results is
estimated to occur in the USA region. Conversely, in at least some of the modeling results, much
of the new production of soybeans, soybean oil and soybean oil biodiesel in the soybean oil
biodiesel shock results is estimated to occur outside the USA region. Partly for this reason, the
corn ethanol shock affects overall global trade, commodity production, and land use decisions to
a lesser extent than the soybean oil biodiesel shock. Across the suite of results from the MCE,
the USA imports more soybean oil biodiesel than corn ethanol. To the extent the increase in
USA consumption of soybean oil biodiesel increases non-USA soybean oil biodiesel exports,
some of the models choose to substitute this lost non-USA consumption of soybean oil biodiesel
with greater use of palm oil biodiesel or fossil fuels. To the extent that new biofuel feedstock
crops must be produced in these modeled scenarios to help satisfy demand for biofuels, each unit
of soybean oil biodiesel feedstock supplied in this way requires more land than does an
equivalent unit of corn ethanol feedstock supplied. This is because there is a lower yield per acre
of soybeans, and, implicitly, of soybean oil, compared to corn. Along with land use, soybean oil
biodiesel production also has much greater potential impacts on livestock production per unit of
fuel produced than does corn ethanol production. Soybean meal produced per gallon of soybean
oil biodiesel is greater than the amount of DDG produced per gallon of corn ethanol, which, all
else equal, can lead to a greater expansion of livestock production in the soybean oil biodiesel
scenario. These possibilities are realized to greater and lesser extents across the models and
across sensitivity analyses. Models included in the MCE produced a wider range of LCA
GHG estimates for soybean oil biodiesel than corn ethanol. This wider range of estimates is
167
-------
related to the greater diversity of feedstock sourcing strategies and the greater sensitivity of the
biodiesel estimates to the variability and uncertainty present in the parameter assumptions
discussed above.
Sensitivity Analysis
Alternative volume scenarios examine whether and how the assumed magnitude of the
volume shock of USA biofuel consumption impacts GHG emissions and other model output
values. In one scenario, where the soybean oil biodiesel volume is reduced to 500 MG, the
ADAGE, GCAM, and GTAP results do not differ substantially from the 1 BG scenario when
they are considered on a per billion gallon basis. GLOBIOM results do show some differences,
such as GHG emissions impacts per billion gallons, between the 1 BG and the 500 MG soybean
oil biodiesel shocks. In a combined scenario, in which corn ethanol and soybean oil biodiesel
were simultaneously increased by 1 BG each, the results generally equal the sum of impacts
observed in the individual 1 BG corn ethanol and soybean oil biodiesel core scenarios for
ADAGE, GCAM, and GTAP. GLOBIOM results for the combined scenarios show more
differences in the estimated output values, including GHG emissions, compared to the sum of the
individual scenarios. These results indicate that, within the range of volumes considered, shock
size does not lead to substantially different impacts on the modeled agriculture system and
estimated GHG emissions in most of the frameworks we have tested.
Finally, stochastic sensitivity analysis identifies which parameter assumptions are
particularly important for a particular model and scenario. Monte Carlo simulations with GCAM
indicate that assumptions relating to soil carbon stocks and the ease of substitution among land
types and crop types have a relatively large influence on the corn ethanol and soybean oil
biodiesel results. The parameter controlling substitution between non-USA regions refined oil
and biodiesel is also influential for the soybean oil biodiesel GHG estimates. A similar analysis
with GLOBIOM finds that biophysical parameters, including those governing the expansion
response of palm cultivation into peatland and governing the emissions associated with such
expansion, are influential on soybean oil biodiesel GHG estimates. Stochastic analysis with
GREET indicates that parameter assumptions have less influence on the supply chain LCA
estimates for corn ethanol and soybean oil biodiesel when using an attributional LCA model.
However, the sensitivity analysis with GREET shows more uncertainty associated with
coproduct allocation choices and for assumptions related to induced land use change GHG
emissions. Considered alongside the other results of this exercise, these parameter sensitivity
analyses indicate that substantial uncertainty in the emissions associated with corn ethanol
and soybean oil biodiesel remains, both within and across models, and that additional
research on economic model parameters remains a high priority. These sensitivity analyses
can help us allocate limited research resources by highlighting which types of parameters are
most influential. Additional parametric sensitivity analysis could help us further pinpoint specific
parameters for additional research and analysis.
Conclusions
In sum, we draw some important general conclusions from this model comparison
exercise. First, ADAGE, GCAM, GLOBIOM and GTAP estimate that substantial indirect effects
168
-------
would be induced by the corn ethanol and, especially, soybean oil biodiesel shocks that we ran
for this exercise. These indirect effects are important drivers in the modeled emissions associated
with these fuels, which highlights the importance of considering indirect effects in LCA.270
Second, we find substantial uncertainty regarding the overall greenhouse gas intensity of
the two biofuels examined in this exercise, corn ethanol and soybean oil biodiesel. Based on this
model comparison exercise, it is evident that variation in estimates remains high across models,
and within individual models when parameter uncertainty is considered. Although models have
advanced and new data has become available since EPA modeled the lifecycle GHG emissions
associated with corn ethanol and soybean oil biodiesel for the March 2010 RFS2 rule, there is
still a large degree of variation and uncertainty in lifecycle GHG estimates that consider
significant indirect emissions. The analyses we have conducted for this exercise highlight the
value of sensitivity analysis as a way of understanding which parameters and assumptions
influence the model results. Furthermore, given that uncertainty remains high for this type of
analysis, it is critical to perform robust uncertainty analysis and provide information about the
range of potential effects and risks of greater biofuel consumption. It is also important to
compare model results and parameters to historic observation.
To summarize, we find that the following model characteristics are critical for evaluating
the GHG impacts, including direct and indirect emissions, associated with a change in biofuel
consumption:
1. Supply chain LCA models produce a fundamentally different analysis than
economic models. Supply chain LCA models evaluate the GHG emissions emanating
from a particular supply chain, whereas economic models evaluate the GHG impacts of a
change in biofuel consumption. Supply chain LCA models generate detailed and
transparent fuel production emissions estimates. However, they do not evaluate all of the
indirect emissions associated with a change in biofuel consumption. The economic
models in our comparison are broad in scope, but they lack certain supply chain details.
2. Land use change emissions are a major contributor to the overall emissions.
ADAGE, GCAM, GLOBIOM, and GTAP all include land use change and land use
change emissions. GREET includes a static estimate of land use change emissions using
previous GTAP results with a different shock size and a 2004 baseline. Estimates of land
use change vary significantly. Drivers of variation in these estimates include differences
in assumptions related to trade, the substitutability of food and feed products, and land
conversion, as well as structural differences in how models represent land categories.
3. This exercise showed that when impacts of biofuel consumption on global energy
markets are considered, GHG emissions estimates are significantly altered. The
270 This finding also supports NASEM recommendation 2-2: "When a decision-maker wishes to understand the
consequences of a proposed decision or action on net GHG emissions, CLCA [consequential lifecycle analysis] is
appropriate. Modelers should provide transparency, justification, and sensitivity/robustness analysis for modeling
choices for the scenarios modeled with and without the proposed decision or action." National Academies of
Sciences, Engineering, and Medicine 2022. Current Methods for Life Cycle Analyses of Low-Carbon
Transportation Fuels in the United States. Washington, DC: The National Academies Press.
https://doi.org/10.17226/26402.
169
-------
models that include energy sector results (ADAGE, GCAM, and GTAP) all estimate that
displacement of refined oil for biofuel is less than 1:1, reducing the GHG emission
reductions associated with the biofuels modeled. This indicates that economic modeling
of the energy sector may be required to avoid overestimating the emissions reductions
from fossil fuel consumption.
4. Model trade structure and assumed flexibility influence the modeled emissions
results. There is general agreement among the economic models that these trade-driven
impacts will occur to some degree. However, these models show different degrees of
trade responsiveness, which impacts trade flows at differing magnitudes across model
results.
5. Certain commodity consumption dynamics appear to substantially influence GHG
emissions results. DDG and soybean meal's impact on the livestock and feed sectors can
affect the estimated GHG emissions associated with biofuels. Explicit modeling of the
global livestock sector, including global feed markets, is an important capability for
estimating the emissions associated with an increase in biofuel consumption.
6. The degree to which other vegetable oils replace soybean oil diverted to fuel
production from other markets can impact GHG emissions associated with soybean
oil biodiesel. Results in this exercise from economic models (ADAGE, GCAM,
GLOBIOM, and GTAP) align in estimating commodity substitution as a significant part
of their scenario solution. Inclusion of explicit global vegetable oil competition is critical
to biofuel lifecycle analysis results because this competition affects the quantity and
location of estimated LUC emissions impacts.
7. The ability to endogenously consider tradeoffs between intensification and
extensification is an important capability for estimating the emissions associated
with an increase in biofuel consumption. Both intensification and extensification of
corn and soybean feedstock production occur across ADAGE, GCAM, GLOBIOM, and
GTAP results in response to changing commodity prices. The degree of crop yield
intensification influences the amount of extensification needed to produce new feedstock
for biofuels. ADAGE, GCAM, GLOBIOM, and GTAP can all model increased crop
yields in response to crop prices. GLOBIOM and GTAP also explicitly consider multi-
cropping.
8. Models included in the MCE produced a wider range of LCA GHG estimates for
soybean oil biodiesel than corn ethanol. The models show much greater diversity in
feedstock sourcing strategies for soybean oil biodiesel than they do for corn ethanol, and
this wider range of options contributes to greater variability in the GHG results. There are
several important reasons for this greater diversity of strategies which were discussed
throughout this document.
9. This exercise demonstrated that a wide range of results can be obtained by varying
parameter values, highlighting the importance of sensitivity and uncertainty
analysis. Stochastic uncertainty analysis can currently be performed with GCAM,
GLOBIOM, and GREET, and Monte Carlo analysis can be performed with GCAM and
GLOBIOM. Other types of sensitivity analysis, such as varying individual parameters,
can be performed with ADAGE and GTAP as well. Sensitivity analysis, which considers
170
-------
uncertainty within a given model, can help identify which parameters influence model
results. However, pinpointing the direct causes of why one estimate differs from another
would require additional research.
Next Steps
A primary goal of this modeling exercise is to help advance the science related to
understanding how different modeling tools can be used to assess the GHG impacts of biofuels.
We understand that there is significant interest amongst stakeholders in a separate but related
topic: namely, how to determine which models, methods, and data are best suited for evaluating
the GHG impact of biofuels. Some stakeholders have suggested that EPA should include criteria
for such evaluative purposes as part of this MCE.
This MCE intentionally does not directly address that subject, nor does it include
proposed criteria. We have in this document instead focused on improving our understanding of
the current state of science for biofuel GHG modeling, including, but not limited to, how the
different models vary, how those variations affect results, and which parameters are critical to
model results. We have not developed a set of criteria against which different models can be
assessed, though we recognize that the development and use of such criteria could be critical in
helping to inform future policy decisions. EPA notes that the criteria used to assess different
models could vary greatly depending on the context in which lifecycle GHG modeling is being
used. For example, the criteria could differ if the context was a holistic program-wide regulatory
analysis as opposed to an assessment of individual fuel pathways. Criteria might also differ
based on the extent to which fuel volumes from a given individual biofuel pathway appear likely
to have impacts on the broader energy or agricultural sectors. To the extent EPA goes on to
develop criteria against which we evaluate different models, this model comparison exercise
provides critical information which will help EPA's work.
The preceding sections of this document note areas for further research, and we are
interested in hearing stakeholder input on those suggestions. EPA is also interested in feedback
and evaluation from outside researchers and organizations on this model comparison exercise.
We plan to directly engage with stakeholders to collect input, consider our outstanding research
needs in this area, and identify those lines of inquiry most critical to future decisions.
171
------- |