Dynamic Linking of Upstream Energy and Freight Demands for Bio and Fossil Energy Pathways in the Global Change Analysis Model

Abstract

Comprehensive study of the environmental impacts associated with demand for an energy resource or carrier in any one sector requires a full consideration of the direct and indirect impacts on the rest of the regional and global energy system. Biofuels are especially complex since they have feedbacks to both the energy system and to regional and global crop markets. In this study, we present a strategy for dynamically including the upstream energy and transportation links to the Global Change Analysis Model. We incorporate the following inter-sectoral linkages: energy inputs to crop production, energy inputs to fossil resource production, and freight transport requirements of energy and agricultural commodities. We assess the implications of explicitly including these links by measuring the global impacts of increased corn ethanol demand in the United States with and without these links included. Although the net global impact of the upstream links on energy and emissions are relatively modest in the scenarios analyzed, the inclusion of these links illustrates interesting trade-offs in energy and transportation demand among fossil fuel and agriculture sectors. We find that the increment in agricultural energy driven by the additional biofuel production associated with the corn ethanol shock is higher than the decrease of energy associated with the displaced fossil fuel consumption. However, this effect is compensated by the reduction in freight transportation requirements of energy. These sectoral interactions suggest that this level of modeling detail could be important in evaluating future analytical questions.

1. INTRODUCTION

A complete understanding of the environmental impacts of energy demands requires consideration of the direct and indirect impacts on the rest of the global energy system [1,2]. Energy is consumed in the production of energy and agricultural resources, whether it is energy for extracting fossil fuels or energy for planting and harvesting crops. Energy is also consumed in transporting resources from production sites to processing facilities and in transporting final goods to distribution and fueling facilities [3]. Many energy and agricultural resources are traded globally and travel internationally on ships, while others are generally produced and consumed within one region and are more often transported on trains, barges, and trucks. These differences in transportation methods may lead to differences in transport energy profiles among commodities. Moreover, the increase in global energy demand combined with the depletion of fossil resources in the last century has increased the energy needs associated with present-day fossil energy resource production and transformation [4,5]. These dynamics are nuanced, and the extent to which capturing them will change overall assessments of energy system impacts remains unclear. Explicitly including such dynamics in modeling of the energy system can help to shed light on this question. Without considering these dynamics in energy sector modeling and assessing their impact, we cannot conclude whether and how much omitting such dynamics affects the results of environmental impact assessments. Considering the full range of potential system impacts is critical to our understanding of the energy consumption needs and net greenhouse gas (GHG) emissions associated with any form of energy [6–8]. Biofuels are especially complex in this regard since they are linked to both energy markets and agricultural markets. The analysis of energy use and emissions impacts associated with bioenergy involves not just the direct energy for production and transport of bioenergy feedstock crops, but also indirect effects on energy resources and other agricultural crops that may be displaced, as well as all their associated energy consumption for production and transport. Some studies have found that the direct energy requirements (and the associated GHG emissions) for biofuel production are larger than for traditional fossil fuels on a per-unit basis [9]. Moreover, Pimentel and Patzek, (2005) concludes that energy outputs from biofuel production could be smaller than the associated energy requirements. Other studies have found that updated data and the incorporation of co-products into the assessment suggest energy and GHG emissions associated with biofuel production have been historically overestimated [11,12]¹. These divergences² show the value in considering the whole-system energy and emissions effects associated with bioenergy, including energy consumption and emissions associated with producing and transporting fossil energy and agricultural goods which are “upstream” of bioenergy production. This study describes an approach for estimating upstream energy requirements from production and transportation of both fossil energy and agricultural commodities.

The literature includes a wide variety of potential tools and frameworks for conducting energy and emissions impact assessments of bioenergy, including, but not limited to, traditional life cycle analysis (LCA) models, and global economic models such as computable general equilibrium (CGE) modeling frameworks, and partial equilibrium (PE) modeling frameworks. All three framework types can be complementarily or individually used to conduct biofuel LCA. The decision of which method to use for a given analysis must balance tradeoffs between analytical detail and economic and physical system sectoral coverage, and each of these types of frameworks may have advantages or disadvantages depending on the application.

Models in the traditional LCA group consist of accounting tools designed to provide detailed calculations for the energy, material, and other inputs involved in all stages of the production and use of an energy or other resource at a particular point in time. One example is the Greenhouse gases, Regulated Emissions, and Energy use in Transportation (GREET) model developed and maintained by the Argonne National Laboratory. The GREET platform has been used frequently as a standalone framework for biofuels life cycle assessment [18–20]. Another example is openLCA³, an open-source tool for sustainability and life cycle assessment maintained by the organization GreenDelta. While openLCA is a more generalized platform not specifically designed for analysis of biofuels, tailored versions of the tool have been applied frequently for this purpose [21–23]. Models like these generally rely on engineering principles of inputs and outputs and physics-based processes to provide static, detailed frameworks for quantifying direct inputs, outputs, and related quantities, such as water use or emissions, of a specific representation of technologies and energy pathways. Plevin et al., (2014) observe that most models of this type are “attributional” rather than “consequential”, meaning they have static representations of input requirements that do not account for the indirect impacts that occur as a consequence of the fuel or item of interest being present in the system. As a result, these models generally cannot endogenously account for indirect effects on other economic sectors or geographic regions on their own (e.g., impacts caused by shifts in the prices of and demand for agricultural or energy commodities). They can, however, be coupled with, or incorporate output from, CGE or PE economic models in an attempt to reflect indirect economic impacts. Several studies have used variations of these hybrid approaches, combining attributional accounting models with economic models to represent a broader range of impacts⁴.

Another approach to estimating the whole-system life cycle impacts of a technological shift towards bioenergy is CGE modeling⁵. In CGE models, all flows between every modeled sector of the economy are endogenous, both within and between model regions. International trade in each product or product aggregate and associated economic responses to changes in supply, demand, and price, are also directly modeled. CGE models dynamically represent the economic interactions among all the affected sectors, including direct and indirect upstream energy demands and feedbacks, such as displacements in other energy or agriculture sectors. CGE models also consider interactions of primary factors such as labor and capital. These qualities are beneficial to analyzing the energy impact of biofuels. However, as compared with other modeling approaches, CGE models often have relatively coarse sectoral detail, which may hinder analysis of specific technologies of interest. Moreover, CGE models represent interactions between sectors not as physical technologies but as abstract production functions that generally represent products, including energy and agriculture, in monetary units instead of physical quantities. Although methods have been constructed to translate results into physical quantities ex post, this representation remains a limitation for modeling detailed physical quantities such as land use and technologies in the energy system [30].

A third category of modeling framework that has been applied to bioenergy life cycle analysis is partial equilibrium (PE) modeling. Such models are “partial” in that they do not represent all interactions between modeled economic sectors (as CGE models do) but instead focus on sectors or markets of interest. This focus allows for technologically detailed representations with modeled commodity flows represented in physical units. PE models are closer to accounting models in the level of detail they can support, while still relying on economic principles and capturing intersectoral economic dynamics. Several PE models have been used for biofuel analysis, which may focus exclusively on the USA⁶, endogenously consider US and international systems⁷, or, additionally, serve as components of larger integrated assessment models⁸. However, linking decision variables in couplings of separable optimization models is cumbersome, time-consuming to execute, and is usually limited to sharing prices, supplies and demands of bioenergy. We are not aware of any analysis in the literature that has used a full interlinked representation of energy demand inputs for agriculture production and transportation with separate optimization models.

This study aims to address and assess the consequences of several known shortcomings of PE-based methods by incorporating four intersectoral linkages, normally absent from PE energy-system models, into the Global Change Analysis Model (GCAM): agricultural energy consumption (i.e., energy inputs to crop production), energy used for fossil resource production, freight transport requirements of energy and freight transport requirements of agricultural commodities. Energy system models typically represent both agricultural energy use and the energy used for oil and gas production in the industrial sector⁹. In such models, simulation of a shift from oil-derived liquid fuels to corn-derived ethanol would not consider the consequent variations of the agricultural energy use, nor the changes in energy used for oil extraction or production, and the impacts thereof. Similarly, to the extent that the freight transportation sector is represented individually in energy system models, it is usually a “final demand” sector, not one whose outputs are consumed by other modeled activities. In many frameworks, the quantity and type of freight transportation service is not dynamic and does not adjust as energy and agricultural transport quantities change. As a result, analysis of a shift from gasoline to biofuels for transportation would not capture the fact that these fuels, and their primary feedstocks (e.g., crude oil, crops such as corn), may entail different upstream transportation distances and modes, and therefore incur different upstream energy demands. Because these new dynamic linkages occur “upstream” of final energy demands, we refer to them hereafter as “upstream linkages”. Absent consideration of these upstream linkages, an analysis of changes in consumption of energy or agricultural goods would also not account for any change in total freight transportation service that might result from such a shift. The sectoral and system-wide impacts of these shortcomings in PE models remains unclear and warrants more explicit consideration.

In this study, we estimate how a corn ethanol shock in the USA affects the energy requirements for fossil energy resource production, agricultural energy use, freight transportation of energy goods, and freight transportation of agricultural goods. Additionally, we explore the whole-system implications of the corn ethanol shock with and without considering these upstream linkages. Our method demonstrates a path by which partial equilibrium frameworks can explicitly consider these dynamics. Our analysis provides an initial estimate of the extent to which the explicit modeling of these linkages modifies both the sectoral and economywide energy balances and flows. Both are innovative contributions for the integrated assessment community in general, and to the literature on fossil fuel and biofuel energy resource requirements and greenhouse gas impacts. In addition, the insights obtained from this study will be of interest for stakeholders associated with the design and implementation of long-term energy strategies.

2. MATERIALS AND METHODS

2.1. GLOBAL CHANGE ANALYSIS MODEL (GCAM)

GCAM is a partial equilibrium, integrated assessment model that explores both human and Earth dynamics and consistently resolves interactions between energy, economy, land, water, and physical Earth systems within a fully integrated computational system. For each model time period, the model iterates until it finds a vector of prices that clears all markets and satisfies all consistency conditions. The model 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 model has been in continuous development for over 30 years and has been applied in several studies and model inter-comparison activities, including the IPCC’s Representative Concentration Pathways [37] and Shared Socioeconomic Pathways [38]. GCAM is an open-source community model that can be downloaded from a public repository¹⁰.

The energy module in GCAM consists of depletable and renewable resources¹¹, energy transformation and distribution sectors (electricity, refining, gas processing, hydrogen production, and district services), and final energy demand sectors (buildings, industry, and transportation)¹². Each sector 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 logit [41] or modified logit model [42], depending on the object. This logit choice model, that avoids the “winner take all” result, is summarized in the following equation:

$$S_i=\frac{a_i*c_i^{\sigma }}{{\sum }_j{a}_j*c_j^{\sigma }}$$ Eq (1)

The market share of each discrete technology “i”, is determined by a share-weight parameter that reflects the specific preferences for a particular choice (α), the cost (c; which includes fuel and non-fuel costs), and an exogenous logit exponent that determines the price responsiveness of the competition (σ). 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. While all these parameters are documented and can be consulted in online repository¹³, the Supplementary Material (hereinafter SM) includes a dedicated excel file with the subset of the input information that are identified as the most relevant for the presented analysis.

The agriculture and land use module differentiates more than 350 land use regions globally, generated as the intersection of 32 geo-political regions with 235 global water basins. 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 [43,44]. This competition follows a similar profit-based logit structure (see Eq 1), so economic land use decisions are based on the relative profitability of using land for competing purposes. Profitability of lands that are not in commercial production in the historical calibration period are inferred from the profitability of proximate lands used for agriculture and forestry. The mentioned file in the SM also incorporates some parameters and assumptions that directly influence land competition, within the context of the developed analysis, while all the information read in by the model can be found in the online repository. 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. A complete description of the land use module can be found in Kyle et al., (2011).

Finally, we note that in this study we use a modified version of GCAM 5.1 [46,47], with some enhancements. This modified version of the model is labeled as GCAM-T-2020 (hereafter “GCAM-T”). GCAM-T can be downloaded from a public repository¹⁴, which incorporates detailed documentation of the developed modifications. In GCAM-T, the land protection representation has been upgraded; where in GCAM 5.1 a default fraction of non-commercial lands are excluded from future conversion, this study excludes lands understood to be protected and lands that are not suitable agricultural production, based on Zabel et al., (2014)¹⁵. Another relevant feature developed for this version is the disaggregation of oil crops from the standard GCAM configuration (Kyle et al., 2011, Table A2). Oil crops are disaggregated into soybeans, rapeseed, and other oil crops (e.g., sunflower, safflower, peanut), which provides additional value for the analysis of regional agricultural markets in the framework of this study¹⁶. In addition, GCAM-T also includes representation of regional agricultural markets with global trade, with region-specific crop prices. Therefore, the agricultural demand for each commodity within a region can be supplied by either domestic or imported crops, which has direct implications for this study.

2.2. UPSTREAM LINKAGES WITHIN GCAM

In this paper, we develop novel “upstream linkages” which dynamically account for direct and indirect energy and freight demands associated with resource and bioenergy production in GCAM. Specifically, agricultural energy use and the energy used for fossil energy production are represented within the industrial energy use sector, and the entirety of freight transportation is treated as a final demand. Figure 1 summarizes how these novel linkages create new dynamic relationships between economic sectors in GCAM, using the ethanol and petroleum refining sectors as examples.

Figure 1:

Subset of the sector/subsector interconnections in GCAM-T. Solid lines represent the interlinks in GCAM-core version, and the “dashed” lines the upstream linkages developed for GCAM-T. A more detailed description of these linkages can be found in the GCAM-T repository (https://github.com/gcamt/gcam-core).

Energy inputs for fossil fuel production are calibrated from the IEA energy balances (IEA, 2012), which estimate the energy used for coal mining and for oil and gas extraction. Due to a high degree of inter-regional and inter-annual heterogeneity of energy intensities, in future model time periods we assume that each unit of extracted natural gas requires approximately 0.05 units of gas, and each unit of extracted oil needs 0.02 and 0.04 units of gas and liquids¹⁷, respectively. The energy consumption by crop production in each model region is estimated from the quantity of liquid fuels assigned to the agricultural sector in the energy balances, downscaled to modeled crop types on the basis of relative cropland allocation. Fuel requirements per hectare of grass and tree bioenergy crops are obtained from Adler et al., (2007). For freight transport requirements of energy and agricultural commodities, we assign a distance coefficient to each resource imported or produced by the different regions based on the Commodity Flow Survey of the United States [51]. Note that for oil, the model assigns an international shipping distance coefficient for the imported oil, and also accounts for the regional road transport by trucks. However, pipeline transport is not part of the freight sector in GCAM, and as such the energy required to power oil and liquid fuel pipelines is not included in this study¹⁸. In addition, biomass liquids feedstocks (including corn for ethanol) are assumed to have shorter transport distances than crops used for other purposes as we assume that biorefineries are sited close to the croplands which supply them [52]. Therefore, the assumed transport distance between the average corn production site and USA corn ethanol refinery is shorter than the assumed transport distance between the average international crude oil production site and the average USA oil refinery, but the transport modes used are different. Table 1 summarizes the coefficients for these upstream linkages for the USA. In addition, a more detailed subset of these parameters can be found in the Supplementary Material, and all parameters are included in our GCAM-T repository¹⁹.

Table 1:

Summary of the input/output coefficients associated with the upstream linkages for the USA. Coefficients for the agricultural energy are averaged by basin, water technology (rainfed vs irrigated) and fertilizer use (Low or High use). The table includes a representative subset of freight transport requirements. Additional details on the coefficients can be found in the Supplementary Material.

Resource production and extraction; PJ/PJ
Sector	Input	Coefficient
Crude Oil	Refined liquids	0.04
	Wholesale gas	0.02
Natural Gas	Wholesale gas	0.05
Coal	Electricity	0.01
	Refined liquids	0.01
Agricultural energy; GJ/ton
Commodity		Coefficient
Corn		0.61
FiberCrop		2.47
FodderGrass		0.24
FodderHerb		0.31
MiscCrop		0.35
OilCrop		2.36
OtherGrain		1.66
PalmFruit		0.61
Rapeseed		3.36
Rice		0.74
RootTuber		0.13
Soybean		2.05
SugarCrop		0.09
Wheat		1.95
Freight inputs; million tonne-km per EJ
Sector	Subsector	Coefficient
Crops	Domestic	600
	Traded (internationally shipped)	1,000–4,000
Oil (internationally shipped)	Crude oil	50,000
	Unconventional oil	50,000
Refined Liquids	Petroleum liquids	2,800
	Ethanol	2,800
Biomass	Biomass	33,000
Biomass Oil	OilCrop	11,500
	Rapeseed	11,500
	Soybean	11,500
Coal	Coal	42,000

2.3. SCENARIOS

This study considers a two-by-two matrix of scenarios consisting of two counterfactual scenarios of the future, each conducted with and without explicitly considering several upstream linkages that have been added to GCAM for this study. The first counterfactual is a baseline scenario representing business-as-usual conditions (Baseline). We note that this baseline scenario assumes current volumes of corn ethanol and vegetable oil-based biodiesel consumption mandated under the U.S. Renewable Fuel Standard program are consumed and continue to be consumed into the future. The second is a “what-if” shock scenario wherein ethanol produced from corn is exogenously increased by 19 billion liters between 2020 and 2030 (compared to volumes produced in the baseline scenario) (CornShock). This study design, where a whole-system model is perturbed by a marginal amount of biofuels in order to analyze the net system responses, is frequently applied in the bioenergy literature for various purposes [25,33,53].

We then model two configurations of each scenario described above (Baseline and CornShock); one configuration which explicitly considers the upstream linkages described in Section 2.2 above (withLink) and another which does not (noLink). We note the difference in corn ethanol production between counterfactual scenarios provides a useful perturbation of the agricultural and energy systems examined in this study. These scenarios are however purely illustrative and are not intended to represent or mimic any current or potential bioenergy policies, though, as described further below, the Baseline scenario does take account of historically produced volumes of biofuels.

Other than the volume of USA corn ethanol consumption, all assumptions are held constant between the Baseline and CornShock scenarios within each approach (withLink or noLink). All differences between these scenarios are exclusively driven by the corn ethanol shock and the upstream linkages. In both the baseline and corn ethanol scenarios, the socioeconomic narrative follows the SSP2 (Shared Socioeconomic Pathway) approach, which is considered the “middle of the road” storyline [54]²⁰. For other assumptions, the scenarios assume a continuation of historical trends in the near term; in the long term, results are largely driven by the SSP2 socioeconomic storyline.

US corn ethanol consumption in the baseline follows historical production levels, reaching approximately 53 billion liters in 2015, as reported in the World Energy Balances of the International Energy Agency (IEA, 2019), and then remaining constant at that level through 2060. The CornShock implementation departs from the baseline in requiring a linear increase in the USA corn ethanol production from approximately 53 billion liters in 2020 to approximately 72 billion liters in 2030 and then remaining constant at that level through 2060. In energy terms, this shock represents about 14.2 EJ of corn ethanol cumulatively from 2020 through 2060. The increased corn ethanol consumption results in some amount of increased corn production, and an associated increase in natural gas consumption driven by increased fertilizer demand. Fertilizer production requires around 35 GJ of natural gas per tonne of nitrogen in produced fertilizer²¹. In addition, we assume that each energy unit of corn ethanol requires 0.3 units of natural gas for process heat and coproduct drying, in line with the requirements of a modern dry mill ethanol plant. Assumed natural gas requirements for oil refining are almost ten times lower (0.03), given that natural gas inputs are generally only used in onsite hydrogen production. A blend of other gaseous hydrocarbons generally provides process heat for oil refining.

We assume that ethanol yields improve by 0.4 percent per year to reflect technological learning, and that there are no further improvements in cost or efficiency after 2050. We also assume that the yields for corn increase over time based on the projections of the Food and Agriculture Organization (FAO). Therefore, the production of corn ethanol becomes relatively more efficient during the analyzed time horizon. Finally, Table 2 summarizes the main modeling assumptions related with corn ethanol refining. These parameters include the non-energy cost (fixed capital costs and all variable non-energy, non-feedstock costs), the amount of secondary output (Distiller’s dried grains with solubles; DDG) per unit of fuel, and the corn ethanol yield (input/output coefficient).

Table 2:

Summary of the key input parameters and modeling assumptions associated with corn ethanol refining.

Year	Non-energy cost of corn ethanol refining ($2010/GJ)	Distillers Grains Output (kgDDG/GJ)	Corn ethanol yield (kg Corn/ GJ)
2015	7.35	33.8	115.2
2020	7.23	33.3	114
2025	6.98	32.9	113.1
2030	6.93	32.1	111.1
2035	6.89	31.3	109
2040	6.84	30.5	107
2045	6.79	29.7	105
2050	6.75	29.1	103.5
2055	6.75	29.1	103.5
2060	6.75	29.1	103.5

3. RESULTS

In this section, we first demonstrate how the implementation of the corn ethanol shock has direct implications for fossil resource production, agricultural commodity production, and freight transportation. We compare the outcomes of the CornShock scenario with the baseline, in a context where upstream linkages are explicitly considered (withLink approach). We also test the robustness of these results to the implementation of alternative shocks. Following this we describe the effects of the shock on total energy balances and flows, distinguishing between impacts with (withLink) and without (noLink) including the novel upstream linkages to evaluate their implications for total modeled energy production and consumption in this context.

In our results, we find that exogenously increased consumption of corn ethanol in the USA impacts regional fossil resource consumption, causing impacts on production of crude oil and natural gas both regionally and globally. In cumulative terms (2020–2060), we find a reduction in the global production of crude oil of approximately 16,241 PJ as increasing USA corn ethanol consumption displaces some petroleum consumption. Cumulative global production of natural gas increases by approximately 3,729 PJ, driven by the additional needs for gas associated with nitrogenous fertilizer production, and with ethanol refining, as it needs more gas than crude oil in the refining process (see Section 2). This implies a net cumulative reduction of fossil fuel production of 12,512 PJ, at a global level, for the period 2020–2060. These results are shown in Figure 2²².

Figure 2:

Global differences in crude oil and natural gas production attributable to the implementation of the corn shock in the USA, per period and input including the upstream linkages (withLinks) (EJ).

We note that this difference in energy consumption decreases over time due to price equilibrium effects. The implementation of the corn ethanol shock increases the final price for refined liquids, which decreases total refined liquid production in the CornShock scenario compared to the baseline. As corn ethanol production is assumed to become more efficient over time (see Section 2), the absolute difference in total refined liquids production between scenarios decreases over time, as presented in Figure 2. The same dynamic can be observed for natural gas production. These differences over time will have subsequent implications for the energy needs for resource production (Figure 3) and agricultural energy (Figure 4). These figures show that differences across scenarios also decrease over time due to these dynamics. We also clarify that results in 2025 are smaller because the shock is not fully implemented and increases linearly until 2030.

Figure 3:

Global differences in energy needs for resource production attributable to the implementation of the corn shock in the USA, per period and input including the upstream linkages (withLinks) (EJ).

Figure 4:

Difference in agricultural energy associated with the implementation of the 19 billion liter per year corn shock in the USA, per period and commodity (withLinks) (EJ).

In the withLinks results, which account explicitly for the energy required to produce primary energy commodities, we can estimate the resulting changes in energy needs for fossil extraction associated with these net changes in crude oil and natural gas production. In cumulative terms, we find the consumption of refined petroleum liquids and natural gas associated with crude oil extraction decrease around 469 and 234 PJ, respectively. On the other hand, natural gas consumption for natural gas production increases by around 186 PJ in cumulative terms (2020–2060). The decrease of crude oil production is larger than the increase in natural gas production, resulting in a net decrease in energy consumption for fossil resource extraction. These effects are all larger in the short term and gradually decrease between 2030 and 2060, as shown in Figure 3. In aggregate, we find that the total decrease in net energy associated with global fossil energy consumed for fossil resource production is approximately 517 PJ cumulatively over the period modeled. This reduction in energy from fossil resource production is related to the implementation of the exogenous increment of corn ethanol, which displaces oil in the refining sector, with a subsequent decrease in oil extraction activities. This reduction represents approximately 4% of the cumulative decrease in fossil energy production globally.

Similarly, the implementation of the corn ethanol shock leads to an increase in corn demand. As explained in the methodology, the model includes the capability to represent regional agricultural markets, accounting for imports and exports for a set of crops. Therefore, the increase on corn demand could be satisfied either with domestic or imported corn. The modelled profit-based logit structure and the model calibration to match with historical data results in a strong preference towards domestic corn (see SM for more details). This leads to an increase in demand for corn cropland area within the USA and drives a corresponding decrease in area and production of other crops. Specifically, the increased demand for corn requires an additional 127–245 hectares each year the USA for domestically produced corn and 26–32 hectares in the rest of the world due to the equilibrium impacts of reduced exports of US corn. Within the USA, the largest increases in corn production occur in the Northern Mississippi River basin²³, which represents more than 36% of the production increase, followed by the Missouri (29%) and Ohio River basins (12%). Most of the increase in corn production is located on rainfed croplands (80–81%).

Increasing corn production implies additional use of agricultural machinery (e.g., tractors) and crop inputs (e.g., nitrogen fertilizer) with the corresponding energy requirements, as represented in the IEA energy balances. In cumulative terms (2020–2060), the agricultural energy demand of corn production increases by 669 PJ in the USA. The energy demands for soybeans and all other crops during this time period decrease by a more modest amount, 110 and 50 PJ, respectively. In addition, the implementation of the shock in the USA also has equilibrium effects on the agricultural sectors of the rest of the world, driven by changes in trade balances and crop prices. The global net response for agricultural energy demands largely reflects the energy intensities of production in the various regions of the world whose production volumes change. The variations in production of different commodities (mostly corn in terms of total crop tonnage) generates a net increase of 113 PJ in fossil energy consumption associated with agriculture outside the USA. On a global basis, we find a cumulative increase in fossil energy consumption for agricultural resource production of about 622 PJ. This estimated impact is of greater magnitude than the reduction in fossil energy for fossil resource production we estimate above and mitigates the global decrease in fossil energy consumption to some extent. The evolution of these energy requirements is summarized in Figure 4.

The implementation of the shock also has implications both for regional freight transport and international shipping, which are represented separately from one another in GCAM. The assumed transport distance between the average corn production site and USA corn ethanol refinery is shorter than the assumed transport distance between the average international crude oil production site and the average USA oil refinery (see Methods). Therefore, as the shock involves ethanol displacing oil, this reduces total tonne-km required to move energy goods; in cumulative terms (2020–2060), USA domestic freight transport decreases by 985 billion ton-kilometers (tonne-km), and international shipping by 5,967 billion tonne-km, differences of 0.7% and 1.1% respectively compared to baseline freight service demand.

Increased domestic demand for corn in the USA leads to reduced corn exports, which reduces demand for international shipping energy associated with these exports. Since we assume that biorefineries for ethanol production are closer to croplands [52], average transport distances for dried distiller’s grains with solubles (DDGS), a coproduct of corn ethanol production consumed by the feed sector, are also assumed to be shorter than those for competing feed sources. This results in a net decline in animal feed tonne-km. These effects on the freight system over the analyzed time horizon are presented in Figure 5. These estimated impacts demonstrate that the corn ethanol shock causes direct and indirect impacts on the energy requirements for motor fuel production, which is consistent with the aforementioned literature (see Introduction).

Figure 5:

Cumulative difference in domestic and international freight transport associated with the implementation of the 19 billion liter per year corn shock in the USA, per region and sector (withLinks) (B-ton-km).

We have conducted a sensitivity analysis to evaluate how robust the obtained insights are to alternative corn ethanol volume shocks. We have analyzed how the energy inputs to resource production, the energy associated with the agricultural sector, and the freight transport inputs of energy and agricultural commodities vary in a response to the implementation of ten different biofuel shocks. In five of these scenarios, the shock is sequentially reduced in 10-percentage-point increments (i.e., from −10% to −50%) compared to the central shock (see Section 2 for more details on the central shock). In the other five scenarios, the implemented corn ethanol shocks are sequentially increased in 10-percentage-point increments (i.e., from 10% to 50%) compared to the central shock.

This sensitivity analysis first shows that the insights obtained in this study are robust to the implementation of alternative shocks in the USA. Broadly, decreasing the exogenously incremented corn ethanol production shock volume decreases the energy inputs associated with oil production and freight input requirements, while increasing the energy associated with the agricultural sector and with natural gas production in all the cases explored. The inverse occurs when the exogenous shock is increased. However, the response to the alternative shocks vary across the upstream linkages analyzed. Energy inputs to resource extraction and production have most stable response in the analyzed sensitivity shocks. Energy associated with natural gas production increases by roughly 0.012 units for every unit of ethanol produced, in all the cases. Likewise, the reduction on crude oil production per unit of ethanol represents around −0.05 units in the explored sensitivity scenarios. On the other hand, our sensitivity results for energy inputs to crop production and freight requirements show a less uniformly proportional response to the implementation of alternative corn ethanol volumes. In the analyzed sensitivity scenarios, for every unit of corn ethanol produced, agricultural energy inputs increase from roughly 0.04 to 0.05 units. We observe that, as the size of the assumed ethanol volume shock grows, the per-unit change in agricultural energy consumption decreases at a relatively, though not completely, linear rate. The decrease of freight inputs requirements of energy and agricultural crops ranges from −0.106 to −0.121 million ton-km/PJ, depending on the alternative exogenous increments. As the volume shock grows, the per-unit decrease in total freight energy consumption associated with the shock becomes smaller. The rate of decrease in per-unit freight consumption follows a generally logarithmic trend. The results of our sensitivity analysis are summarized in Figure 6.

Figure 6:

Differences in upstream linkages per unit of produced corn ethanol associated with the implementation of alternative corn ethanol shocks. The panel in the left shows energy inputs to resource production, and energy inputs to crop production. The panel in the right shows freight transport requirements of energy and agricultural commodities.

We observe numerous direct impacts of the corn ethanol shock on fossil fuel production, crop production, and freight transport volumes. However, given that total final energy in the USA in the analyzed time horizon represents between 67–79 EJ/yr, the explicit consideration of the upstream energy linkages resulting from an 0.4 EJ/yr corn ethanol shock had little net effect on total energy balances and flows. Figure 7 analyzes these differences by region and fuel. In the USA, the cumulative difference in total final energy demand, driven by the implementation of the corn ethanol shock, is −566 PJ with these linkages and −268 PJ without considering the linkages. In the rest of the world, the divergences are even smaller; our results estimate impacts of 267 PJ with the linkages and 272 PJ without them. Even though there exist some differences, especially in the USA, they do not imply a significant difference for overall energy flows. According to our modeling projections, cumulative (2020–2060) total final energy demand in USA would represent around 3,000 EJ. Therefore, the difference in USA between the two approaches (withLink and noLink) would represent less than a 0.01% difference in the total final energy demand.

Figure 7:

Cumulative difference (2020–2060) in total final energy demand driven by the corn ethanol shock by region and fuel, considering and not the upstream linkages (PJ)

4. DISCUSSION

The implementation of the corn ethanol shock will have implications in the different systems represented within GCAM. The increase in corn ethanol production would have a direct impact on the energy system, due to associated natural gas demands, and displacement of oil. In addition, this change in the energy mix produces subsequent effects in further modules, such as the agricultural and land use system. Many of these effects respond somewhat differently depending on whether the upstream linkages are represented or not. Given that the shocks analyzed in this paper are small compared to system-wide energy demand, it is unsurprising that these differences associated with including the upstream linkages have a modest effect on the overall global energy balance. However, we note these linkages may have more significant impacts on overall energy flows in a scenario which implements a shock that more substantially modifies the structure of the global energy mix.

Our results suggest that the increase in agricultural energy consumption driven by the shock is larger than the reduction in direct energy consumption for fossil resource production. Conversely, our results also suggest that domestically produced bioenergy commodities and their coproducts may be associated with less freight transportation service on average than fossil energy commodities. Different studies have envisioned expanded deployment of bioenergy in different future scenarios [56–58], while recent studies focus on the role of biofuels in determined sectors [27]. The outcomes presented in this study suggest that the incorporation of upstream linkages may provide valuable information to these analyses, especially with regards to impacts within the fossil energy and agricultural sectors.

While we have tested the sensitivity of our model to the most important assumption, the size of the corn ethanol shock, we acknowledge that there is uncertainty in many other parameters and assumptions we have made to explicitly model these upstream linkages. For example, there is uncertainty regarding the input-output coefficients of agricultural energy, the energy requirements for fossil resource extraction and production, and the freight transport distances and modes for each commodity. While the base-year volumes of energy associated with agricultural energy and resource extraction and production are calibrated to the energy balances of the Integrated Energy Agency (IEA, 2012), the freight transport distances and associated energy consumption rely on bottom-up estimation. Hence, further research ought to focus on this parameter to understand and, where possible, reduce uncertainty, especially for regional or country-level analysis.

Finally, the level of flexibility assumed in GCAM for regional production of globally traded fossil fuels also impacts the region-level results in this study. In the model, fossil energy commodity markets are resolved at the global level, and the regional production volumes are more flexible from year to year than in the real world, where production volumes reflect long-lived investments and the character of reserves, among other factors. While recent versions of GCAM explicitly represent of capital stocks and reserves of fossil energy resources, this feature is not included in GCAM-T, which is in part why all regions in this study are assigned the same energy intensities of oil and gas production. Future research could improve the inter-regional allocation of fossil energy production as well as incorporating realistic region-specific estimates of the energy intensities of resource production over time.

5. CONCLUSION

A comprehensive assessment of the environmental impacts associated with energy commodity production requires consideration of direct energy for commodity production and for freight transport, and consideration of the indirect effects on the energy resources and the agricultural crops that it may be displacing. Integrated assessment modeling is a useful method to assess the whole-system energy implications of different policies and shocks, into the future as the systems evolve. This study considers the global and regional effects of a shock on corn ethanol consumption in the USA, with and without considering four inter-sectoral linkages: agricultural energy consumption, fossil resource extraction energy consumption, freight transport requirements of energy resources and freight transport requirements of agricultural commodities.

Our results demonstrate that a modeled increase in corn ethanol consumption in the USA has sector-specific implications for each of the analyzed upstream energy linkages. Overall, we find that, for the modeled corn ethanol shock, the increase in agricultural energy associated with the additional bioenergy production outweighs the reduction in energy inputs for fossil resource extraction and production in cumulative terms (2020–2060). In our results, this is offset by a reduction in freight transport needs. But data underlying the freight transport assumptions are limited and should be the focus of future research.

We conclude that enhancing an integrated assessment model to include these upstream energy linkages can provide a significant improvement upon existing methods of estimating the whole-system impacts of energy policies decades into the future. This approach retains the benefits of using long-term economic equilibrium models, while also incorporating some of the advantages of life cycle analysis, which have a heavy focus on the myriad upstream linkages that are involved with production of specific energy commodities. The approach used here may be particularly helpful for assessment of scenarios that project significant increases in bioenergy commodity demands, due to the different character of upstream energy use and transport between fossil and bioenergy commodities. Further, our finding of a small net effect from including these upstream linkages in overall global and regional energy balances and flows is a valuable outcome for the integrated modeling community. Our results suggest upstream energy may not have significant implications at a system-wide level, depending on the scale of the shock, but could be considered by these integrated assessment models for more detailed analysis. The finding that these dynamics lead to noticeably different emissions outcomes suggests that they could produce globally meaningful changes in energy consumption and emissions in bioenergy scenarios which substantially alter the global energy mix. Consideration of these upstream linkages can therefore inform any such future work.

DATA STATEMENT

Data used in this paper were from GCAM data system or simulation results. The GCAM-T version used in this study can be downloaded from a public repository (https://github.com/gcamt/gcam-core), and from Zenodo (https://zenodo.org/record/4705472#.YH9VZOhKiUk). The repository includes a detailed documentation of this version with all the developed modifications. Moreover, an additional data repository will be made available in Zenodo. This complementary repository will contain an excel file with a representative set of parameters with relevant model parameters and assumptions, all the GCAM results for the four scenarios analyzed (“scen_data_upstream_energy.dat” and additional comma-separated values (csv) files), the R code for generating main figures, and all the files required for running the “noLink” scenarios (note that the upstream linkages are incorporated by default in GCAM-T, so no additional files are required).