Anda di halaman 1dari 14

Bioresource Technology 101 (2010) 654–667

Contents lists available at ScienceDirect

Bioresource Technology
journal homepage: www.elsevier.com/locate/biortech

Life cycle evaluation of emerging lignocellulosic ethanol conversion technologies


Sabrina Spatari a,*, David M. Bagley b, Heather L. MacLean c
a
Drexel University, Civil, Architectural, and Environmental Engineering, 3141 Chestnut Street, Philadelphia, PA 19104, USA
b
University of Wyoming, Department of Civil and Architectural Engineering, 1000 E. University Avenue, Department 3295, Laramie, WY 82071, USA
c
University of Toronto, Department of Civil Engineering and School of Public Policy and Governance, 35 St. George Street, Toronto, Ontario, Canada M5S 1A4

a r t i c l e i n f o a b s t r a c t

Article history: Lignocellulosic ethanol holds promise for addressing climate change and energy security issues associ-
Received 28 January 2008 ated with personal transportation through lowering the fuel mixes’ carbon intensity and petroleum
Received in revised form 30 March 2009 demand. We compare the technological features and life cycle environmental impacts of near- and
Accepted 14 August 2009
mid-term ethanol bioconversion technologies in the United States. Key uncertainties in the major pro-
Available online 16 September 2009
cesses: pre-treatment, hydrolysis, and fermentation are evaluated. The potential to reduce fossil energy
use and greenhouse gas (GHG) emissions varies among bioconversion processes, although all options
Keywords:
studied are considerably more attractive than gasoline. Anticipated future performance is found to be
Biofuels
Lignocellulosic ethanol
considerably more attractive than that published in the literature as being achieved to date. Electricity
Life cycle assessment co-product credits are important in characterizing the GHG impacts of different ethanol production path-
Uncertainty analysis ways; however, in the absence of near-term liquid transportation fuel alternatives to gasoline, optimizing
Monte Carlo analysis ethanol facilities to produce ethanol (as opposed to co-products) is important for reducing the carbon
intensity of the road transportation sector and for energy security.
Ó 2009 Elsevier Ltd. All rights reserved.

1. Introduction To support these recent policy initiatives and, more broadly, to


make progress in improving the environmental performance of our
Ethanol from lignocellulosic feedstocks has received much re- light-duty vehicle transportation system, a suite of emerging tech-
cent attention for its potential role as a light-duty vehicle fuel nologies for converting lignocellulosic feedstocks to ethanol is un-
due to volatile crude oil prices, ethanol’s potential to reduce North der development at laboratory, pilot, and demonstration scales
America’s dependence on imported petroleum, and recent policies (Wyman et al., 2005a; Eggeman and Elander, 2005; Aden et al.,
to reduce the carbon intensity of transportation fuels and to in- 2002; US DOE, 2007). The technologies consist of different pre-
crease the production of biofuels (Solomon et al., 2007; Sperling treatment methods (Wyman et al., 2005a; Mosier et al., 2005)
and Yeh, 2009). The importance of biofuels in the United States and enzymatic hydrolysis and fermentation orientations (Lynd
was reflected in the Energy Policy Act (US Congress, 2005) and et al., 2002). As these developing technologies may differ in their
more recently in the Energy Independence and Security Act (EISA), feedstock requirements as well as process energy and chemical in-
which calls for further increasing the use of these fuels in transpor- puts, they exhibit a range of life cycle energy and environmental
tation, and requiring the use of lignocellulose-based fuels begin- performance, an aspect that has not been discussed in detail in
ning in 2010 (US Congress, 2007). Lignocellulosic ethanol the literature (Spatari, 2007).
possesses environmental and resource conservation advantages Life cycle assessment (LCA) studies have shown environmental
over both petroleum-based fuels and corn-derived ethanol, includ- and social benefits associated with lignocellulosic ethanol and its
ing potentially lower life cycle greenhouse gas (GHG) emissions use in light-duty vehicles compared to gasoline and corn ethanol
and reduced usage of fossil fuel resources, particularly imported fueled vehicles (e.g., Spatari et al., 2005; Sheehan et al., 2003). To
petroleum (Fu et al., 2003; Sheehan et al., 2003; Spatari et al., date, most studies have focused on modeling the upstream agricul-
2005; Farrell et al., 2006). Production of the fuel could stimulate tural processes in specific geographic locations as well as the end
rural economic development (Kim and Dale, 2004), and ethanol uses and have considered only a limited set of feedstocks in addi-
is compatible with existing light-duty vehicles in blends with gas- tion to a single pre-treatment-conversion technology. Further-
oline at low (10%) and high (85%) volumes. more, these published estimates are based on ASPEN simulations
of commercial scale nth plants.1 While it is important to examine

1
* Corresponding author. Tel.: +1 215 571 3557; fax: +1 215 895 1363. Industrial facilities that have overcome cost and performance penalties of the
E-mail address: spatari@drexel.edu (S. Spatari). first-of-a-kind (pioneer) facilities.

0960-8524/$ - see front matter Ó 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.biortech.2009.08.067
S. Spatari et al. / Bioresource Technology 101 (2010) 654–667 655

the predicted best case scenarios, it is also useful to understand per- (c. 2010) and two advanced mid-term (c. 2020) lignocellulosic eth-
formance to date and evaluate the transition from present to future anol conversion technologies in the United States.
performance. Normally LCA is conducted on the complete cycle of production
Another gap in most published LCA studies is that they have not and use of a product; however, because we are comparing the
accounted for the impacts associated with the production and use same product (ethanol), we limit the analyses to the production
of pre-treatment chemicals, enzymes and nutrients used in the stages, which include feedstock production and transport, and eth-
conversion process (see MacLean and Spatari (2009)). anol conversion. This is sometimes termed a well-to-(plant) gate
Lignocellulosic ethanol has been attracting much attention due (WTG) analysis in the LCA literature of alternative fuels.
to its promise of assisting governments to make progress on reduc- To allow comparison of the different technology scenarios, we
ing GHG emissions from the transportation sector (Farrell et al., constructed LCA models with two different sets of input data.
2006). In spite of this attention there are, as yet, no commercial lig- One set of input data was developed from laboratory or process
nocellulose-to-ethanol facilities, and there is considerable uncer- development unit (PDU) scale data presented in the literature.
tainty regarding preferred technologies and their operating For models using this data set, we apply Monte Carlo analysis to
parameters. Therefore, the objectives of this paper are (1) to con- consider uncertainty in the predicted performance. The second
duct LCAs of lignocellulosic ethanol production from a larger set set of input data is generated from chemical plant designs simu-
of bioconversion technologies than has been previously examined; lated by ethanol technology developers using Aspen Plus software.
(2) to investigate how uncertainties in technological performance The Aspen simulations generally represent conditions set to maxi-
affect life cycle results through combining LCA with Monte Carlo mize ethanol production under specific operating conditions as-
simulation; (3) to include in the LCA the upstream impacts associ- sumed for future ‘‘nth” plants. For both sets of models, data for
ated with chemicals utilized in the conversion processes; and (4) to feedstock production and transport as well as all energy and emis-
understand, especially, the predicted ethanol yields, co-product sions associated with fuels and chemicals used throughout the eth-
electricity, fossil energy input and environmental emissions factors anol production life cycle were obtained from a combination of
including GHG emissions for the technologies examined both in the public (Wang, 2007) and private (PE, 2004) sources.
near-term (c. 2010) and mid-term (c. 2020) using LCA. The research We develop LCA models based on material and energy balances
demonstrates a life cycle approach/framework that is expected to to estimate the technological indicators, ethanol yield and electric-
be useful to researchers and technology developers as well as gov- ity co-production; and the resource–environmental impacts fossil
ernment entities (e.g. US Department of Energy) in estimating the energy input, petroleum input, GHG emissions (carbon dioxide
life cycle performance of technologies under development. (CO2), methane (CH4), and nitrous oxide (N2O)), air pollutants (car-
The technologies selected for study in this research are not bon monoxide (CO), nitrogen oxides (NOx), non-methane organic
meant to represent the full set of technologies under development gases (NMOG), sulfur oxides (SOx), and particulate matter (PM10))
(or their performance), ‘nor does their selection for modeling imply and solid waste.
endorsement by the authors.
2.1. Feedstock production, collection and transport
2. Methods
We consider corn stover (CS) and switchgrass (SG) feedstocks
We use LCA (ISO, 2006) to compare the energy and material in- for near-term LCA models. Life cycle data for the production, col-
puts and environmental emissions associated with eight near-term lection, and transportation of the feedstock include the energy

Table 1
Feedstock and transportation to conversion facility data model parameters.

Description Values Notes and data source


Feedstocks
Corn stover (CS) – near-term
Yield (dry metric ton/ha/yr) 9.7 Wang (2007)
Collection (MJ/dry metric ton) 260 MJ/dry metric ton Assumes 2-pass system. 50% removal of CS for ethanol
conversion. All data as per Wang (2007)
Dry weight composition (%) Glucan 37.4, xylan 21.1 arabinan 2.9, galactan 2, mannan Sheehan et al. (2003)
1.6, lignin 18
Nutrient replacement (g/dry kg CS) N 5, P 1.8, K 9.2 Sheehan et al. (2003), Kim and Dale (2004)
N2O emissions Direct: 44.1 Direct and indirect emissions calculated using IPCC
(g N2O/dry metric ton CS) Indirect: 66.7 (1997); co-product allocation described by Spatari (2007)
Switchgrass (SG) – near-term (mid-term)
Yield (dry metric ton/ha/yr) 11.2 Wang (2007)
Fertilizers (g/dry kg SG) N 11.7, P 0.16, K 0.25 Wang (2007)
Herbicide (g/dry kg SG) Atrazine 0.031 Wang (2007)
Harvesting (MJ/dry metric ton SG) 260 All data as per Wang (2007)
Dry weight composition (%) Glucan 32.3 (37.4), xylan 21.4 (24.9) arabinan 2.9 (3.4), Near-term: normalized average for near-term reported by
galactan 0.9 (1.1), mannan 0.3 (0.3), lignin 17.5 (23.6) Laser et al. (2009)
Mid-term: Laser et al. (2009)
N2O emissions 1.5% of N-fertilizer applied Wang (2007), Wu et al. (2006)
C sequestration due to land use change (g 53,500 Wang (1999)
CO2/dry metric ton SG)
Transportation – near-term (mid-term)
Distance 130 km (170 km) roundtrip Based on facility capacity of 2000 dry metric ton/day
(5000 dry metric ton/day)
Truck Class 8b heavy-duty, 2.1 L/km
Fuels/electricity (applicable to feedstock Assume US national electricity grid (Wang 2007)
production and transport in each LCA
model)

Note: Mid-term switchgrass values are noted in parentheses.


656 S. Spatari et al. / Bioresource Technology 101 (2010) 654–667

and emissions associated with fertilizers, herbicides, and fuel to 70% based on literature estimates with a Monte Carlo analysis over
operate harvesting equipment (Wang, 2007). For assumptions, the production and use cycle of ethanol. Nutrients are added to re-
parameters and data sources on feedstock activities see Table 1. place those present in the CS removed from the field as noted by
Wang (2007) utilizes US data for the feedstock activities but does Sheehan et al. (2003) and Kim and Dale (2005). The CS feedstock
not site specific feedstock collection areas or biofuel production is collected in a two-pass system (Wang, 2007). Since estimates
facilities. We follow this example due to the prospective nature of emissions of N2O associated with N-fertilizer use vary widely
of this research and our focus on biofuel production, rather than (Sheehan et al., 2002; Wang, 2007), we use the IPCC (1997) method
agricultural activities. to estimate these emissions and a co-product allocation procedure
The model for CS collection assumes that 50% of the CS remains from Sheehan et al. (2003) to allocate the emissions to the CS col-
on the soil for maintaining soil organic carbon (SOC) and minimiz- lected and that remaining on the field.
ing erosion, and 50% is removed for ethanol conversion. The The SG production data are taken from Wang (2007) and are
amount of stover that can be sustainably removed in order to summarized in Table 1. Fertilizers and the herbicide atrazine are
maintain SOC, which in turn retains and recycles nutrients, im- added. Switchgrass is harvested using farming equipment specified
proves soil structure, enhances water exchange characteristics, by Wu et al. (2006); the equipment is assumed to be identical for
sustains microbial life and ultimately improves corn and subse- near and mid-term applications. We assume N application rates
quent stover yields, varies with agricultural practice and location. consistent with test plot studies (Vogel et al., 2002). Nitrous oxide
Some have argued that all stover should remain on the soil (Lal, emissions from N application are calculated as discussed by Wu
2004) while others indicate that between 5 and 8 Mg/ha should re- et al. (2006). The deep-rooting system of SG when planted on crop-
main on the land for till and no-till practices on continuous corn land and pastureland in the ratio 39:61 is assumed to result in
production, respectively (Wilhelm et al., 2007). These latter two re- additional sequestration of CO2 in the quantity 53,500 g/dry metric
moval rates correspond to approximately 48% and 18% stover re- ton SG as noted by Wang (1999), albeit, uptake of atmospheric CO2
moval for the yields we assume (Table 2). Although we do not can be offset by N2O release associated with N-fertilizer applied
examine variability in stover removal in this analysis, Spatari (Conant et al., 2005). The additional CO2 uptake is associated with
(2007) examines varying this model parameter between 35% and the cultivation of SG when grown on cropland assumed to formerly

Table 2
LCA variable input specifications (variable range and statistics) used in near-term Monte Carlo simulations and near- and mid-term input specifications for Aspen simulations.

Variable Variable Monte Carlo simulation Aspen simulation


notation
Range Mean Statistic Distribution Near- Mid-
terma termd
Low 2.5th High 97.5th Point Point
percentile percentile values values
Feedstocks
Corn stover (CS)
Glucan fraction (g glucan/g CS) fglucan 0.306 0.381 0.361 SD = 0.035 Student T 0.374
Xylan fraction (g xylan/g CS) fXylan 0.159 0.216 0.196 SD = 0.015 Logistic 0.211
Switchgrass (SG)
Glucan fraction (g glucan/g SG) fglucan 0.268 0.375 0.334 SD = 0.023 Logistic 0.323 0.375
Xylan fraction (g xylan/g SG) fXylan 0.176 0.249 0.219 SD = 0.047 Student T 0.211 0.249
Pre-treatment
Dilute acid (NREL)
Hemicellulose conversion (g hemicellulose sugar/ cj 0.6 0.75 0.675 SD = 0.038 Normal 0.75 0.95
g hemicellulose)
Hydrolysis conversion (g glucose/g glucan) cj 0.616 0.654 0.635 SD = 0.0097 Normal 0.8 0.95
AFEX
Hemicellulose conversion cj 0.65 0.85 0.75 SD = 0.051 Normal 0.776 0.95
Hydrolysis conversion cj 0.9 1 0.95 SD = 0.0097 Normal 0.95 0.95
Fermentation
Glucose-to-ethanol conversion (fraction of Yglu_eth 0.85 0.95 0.90 SD = 0.026 Normal 0.92b 0.95
theoretical yield) 0.95c
Xylose-to-ethanol conversion (fraction of Yxyl_eth 0.5 0.90 0.70 SD = 0.103 Normal 0.85b 0.95
theoretical yield) 0.85c
Arabinose-to-ethanol conversion (fraction of Ya_eth 0.01 0.72 0.283 Mode = 0 Triangular 0b 0.95
theoretical yield) 0.85c
Galactose-to-ethanol conversion (fraction of Ygal_eth 0.01 0.72 0.283 Mode = 0 Triangular 0b 0.95
theoretical yield) 0.85c
Mannose-to-ethanol conversion (fraction of Yman_eth 0.01 0.72 0.283 Mode = 0 Triangular 0b 0.95
theoretical yield) 0.85c

Notes: SD, standard deviation.


a
Near-term Aspen simulations with the SG feedstock compositions were modified from simulations run with the CS feedstock to adjust for differences in cellulose,
hemicellulose, and lignin concentrations as was done in previous LCA models (Spatari et al., 2005). Yields of ethanol from glucose, xylose, arabinose, galactose, and mannose
for the near-term models were estimated for two scenarios, one in which the fermentation yield in the first line of each row was assumed, and one in which the fermentation
yields in the second line shown were assumed.
b
The Aspen simulations whereby only glucose and xylose are fermented (first line) were set to correspond to Aspen model simulations investigated by the CAFI group, and
summarized in Wyman et al. (2005a,b), and in costing models by Eggeman and Elander (2005), and represent plausible near-term ethanol yield estimates.
c
The Aspen simulations whereby all five sugars are fermented (second line) were set to correspond to Aspen model simulations reported by Aden et al. (2002) and
assumed by Sheehan et al. (2003) in their LCA model of corn stover, and represent more optimistic estimates of ethanol yields relative to the estimates shown in the first line.
d
A Monte Carlo simulation was not undertaken for the mid-term models due to the difficulty in assigning plausible probability distributions to the advanced SG feedstock
and mature CBP technology assumed in those models. Only the advanced SG feedstock was examined for the mid-term Aspen simulations based on the model design reported
by Laser et al. (2009).
S. Spatari et al. / Bioresource Technology 101 (2010) 654–667 657

be used for corn production Wu et al. (2006). The CS and SG feed- We examine two possible pre-treatment technologies, dilute
stocks are assumed to be transported to a 2000 dry metric ton/day sulfuric acid developed at the National Renewable Energy Labora-
ethanol conversion facility. tory (NREL) and ammonia fibre explosion (AFEX) developed by
In the mid-term model, the properties of the ‘‘advanced” SG Bruce Dale and colleagues at Michigan State University. These
(adv SG) feedstock are taken from Laser et al. (2009). Laser et al. re- technologies were selected for near-term LCA model development,
ports a mature SG crop having higher fractions of cellulose, hemi- based on data availability, from a set of promising emerging tech-
cellulose and lignin, and a decline in ash, protein, and soluble nologies and in particular in the case of the NREL technology be-
carbohydrate relative to a normalized average SG measured from cause it has been the most widely assumed technology modeled
25 samples, owing to a late season single-cut harvest, which sup- in almost all published lignocellulose-to-ethanol LCAs. Further-
ports cell-wall growth (sources of cellulose and hemicellulose) more, NREL’s process described by Aden et al. (2002) is the tech-
and lignin and advances in breeding relative to current SG samples. nology under investigation for Renewable Fuel Standard (RFS)
Inputs of fertilizers and herbicide, as well as feedstock harvesting policy development. Both technologies employ solid–liquid sepa-
energy demand and N2O emissions are assumed to be the same ration techniques to separate the lignin in the feedstock and direct
as the near-term SG models as noted by Wu et al. (2006). Only it to an energy recovery unit for steam and electricity generation.
the feedstock transport distance is assumed to be greater for the Simultaneous saccharification and co-fermentation (SSCF) of the
mid-term model due to the larger collection distance required for two dominant sugars present in the feedstock, glucose and xylose,
a 5,000 dry metric ton/day advanced ethanol conversion facility. and SSCF of all six-carbon (C6) and five-carbon (C5) sugars present
in the feedstock, are assumed for two sets of near-term models.
The mid-term technology LCA models assume the use of an ad-
2.2. Ethanol conversion: pre-treatment, hydrolysis, and fermentation; vanced AFEX (adv AFEX) pre-treatment technology (relative to the
and electricity co-production near-term AFEX) in combination with consolidated bioprocessing
(CBP) enzyme production and hydrolysis–fermentation technol-
To produce ethanol from lignocellulose via biochemical conver- ogy. CBP is under development at Dartmouth College by Lee Lynd
sion the cellulose and hemicellulose fractions are digested into and colleagues.
sugars using a pre-treatment followed by a hydrolysis process. The dilute sulfuric acid (NREL) pre-treatment process is
The sugars are in turn fermented to alcohol, while the lignin frac- modeled from the process design by NREL (2001). The CS is treated
tion is separated and used as a source of energy to drive the pro- with dilute sulfuric acid at 190 °C and 177 psi, which releases
cess. The portions of cellulose and hemicellulose not converted to hemicellulose sugars into solution and prepares the cellulose for
sugars, and the sugars not fermented to ethanol are directed to the hydrolysis. Flash cooling then removes degradation products from
facility’s energy recovery equipment along with the lignin. Use of the pre-treatment reaction directing them to wastewater treat-
the lignin fraction and other biosolids produced during conversion, ment, and solid and liquids are separated. The ion exchange unit
including the non-digested cellulose/hemicellulose fibres and with ammonia eluant originally specified by Wooley et al. (1999)
unfermented sugars, for process energy improves the attractive- for treating yellow poplar (YP) and included in the NREL (2001)
ness of lignocellulosic technology, as it greatly reduces process fos- CS model along with overliming with the addition of lime, is re-
sil energy requirements (Aden et al., 2002). Fig. 1 summarizes the moved for our model based on a more recently published design
processing steps involved in the bioconversion of lignocellulose to by Aden et al. (2002) which includes overliming only and not ion
ethanol for the LCA models developed in this research. exchange due to lower fractions of acetates and consequently

CBP
Enzyme
production
Wastewater
Feed handling Pre-treatment Ethanol recovery
treatment &
system
lignin separation
Hydrolysis &
fermentation

-SSCF

Steam & Syrup & solids


Electricity Ethanol

Energy recovery:
steam & Finished
electricity
Electricity product

Lignin & biogas

Legend
System Boundary
Process
Flow

Fig. 1. Plant design model for the biochemical conversion of lignocellulose to ethanol. Notes: The model assumes energy recovery for steam and electricity production from
lignocellulose fractions not converted to ethanol. Adapted from model designs described by Aden et al. (2002) and Laser et al. (2009). CBP, consolidated bioprocessing; SSCF,
simultaneous saccharification and co-fermentation.
658 S. Spatari et al. / Bioresource Technology 101 (2010) 654–667

lower acetic acid by-product formation in CS versus YP, and a cost producing high ethanol yields (Ingram et al., 1987, 1991), however
savings for eliminating ammonia requirements. NREL has tested continued research effort is necessary to attain commercial-scale
the dilute acid technology at PDU scale at NREL. Additions of dilute fermentation of all cellulose and hemicelluloses sugars. At present,
sulfuric acid, lime and other input chemicals to NREL pre-treat- arabinose remains the most difficult sugar to ferment. Mannose
ment are noted in Table 3. and galactose can be fermented more easily to ethanol with the
The AFEX pre-treatment process treats biomass with liquid appropriate organism (e.g., S. cerevisiae) as suggested by Mabee
anhydrous ammonia (NH3) at moderate temperatures (60– (2006). Much research continues to improve the co-fermentation
100 °C) and high pressure (250–300 psi) for five minutes, after of hexose and pentose sugars (Ingram et al., 1987, 1991; Lawford
which time the pressure is rapidly reduced, causing an ‘‘explosion” and Rousseau, 1992, 2003). Our models assume use of cellulase en-
and allowing the cellulose and hemicellulose fibres to expand and zymes produced from Trichoderma reesei.
be more amenable to enzymatic hydrolysis. This makes the lignin During CBP enzyme production, hydrolysis and fermentation of
fraction soluble, and allows for nearly complete conversion of cel- all sugars take place in one reaction vessel using one organism for
lulose and hemicellulose fibres to fermentable sugars (Teymouri enzyme production and fermentation of all C5 and C6 sugars (Lynd,
et al., 2005). Most NH3 is recovered and recycled (97.3%). NH3 to 1996; Lynd et al., 2002; Laser et al., 2009). To date no single organ-
dry biomass loadings of 1:1 are needed to ensure high sugar yields ism can perform all functions required by CBP technology, for
and are assumed for the near-term models, however recent work example, the cellulase producing enzyme (T. reesei) we assume in
by Sendich et al. (2008) has shown that NH3 to dry biomass load- our near-term models does not ferment sugars to ethanol. Like-
ings as low as 0.3:1 do not compromise the sugar generation effi- wise, the fermentative organism we assume (Z. mobilis) does not
ciency as long as the NH3 is kept in the liquid phase. This latter produce cellulases, and other potential ethanol-producing organ-
ammonia to biomass ratio is assumed in the mid-term models. isms either produce low ethanol yields or grow under aerobic con-
The technology has been tested at laboratory-scale. The AFEX ditions (Lynd, 1996). However, researchers at Dartmouth College
parameters used as input for the pre-treatment module of the continue to conduct metabolic engineering research to develop
LCA models are summarized in Table 2 and chemical inputs in Ta- such an organism (Lynd et al., 2002). The models developed for
ble 3. Aspen simulation values of the AFEX-pre-treated CS feed- CBP assume such an organism exists and that it is used for both
stock were provided by Eggeman (2005), who employed mean enzyme production and fermentation. The mid-term models are
estimates from CS pre-treatment laboratory tests conducted by assumed to represent high attainment of technological perfor-
Teymouri et al. (2005). mance in the bioconversion of lignocellulose to ethanol.
Simultaneous saccharification and co-fermentation (SSCF) is The variables used as input for the pre-treatment, hydrolysis
accomplished by combining the enzymatic hydrolysis of cellulose and fermentation modules of the near-term LCA models are sum-
to glucose process and the co-fermentation of C5 sugars (xylose marized in Table 2, including the values employed in the Monte
and potentially arabinose) and C6 sugars (glucose and potentially Carlo analysis (developed from laboratory and PDU-scale data),
mannose and galactose) process in one reaction vessel (Lynd, and the values used in the nth plant Aspen simulations. Inputs into
1996; Lynd et al., 2002). To date, the co-fermentation of glucose the mid-term models, which only employ single-point values from
and xylose has been tested at bench and pilot scales successfully the Aspen simulation, are summarized in Table 2. The equations
with the organisms Saccharomyces cerevisiae (Hahn-Hägerdal developed for the LCA models are described in Section 2.4.
et al., 2001) and Zymomonas mobilis (Aden et al., 2002). While glu-
cose fermentation is a commercial-scale technology, xylose fer- 2.3. Chemical inputs, electricity credits, and model designations
mentation to date has been performed under controlled
laboratory-scale conditions using genetically modified organisms A series of cradle-to-gate modules that detail the chemical load-
(Sheehan et al., 2003) and at demonstration scale (Iogen, 2007); ings used in pre-treatment, hydrolysis, and fermentation were
the simultaneous fermentation of arabinose, mannose, and galact- developed for the two near-term (NREL–SSCF and AFEX–SSCF)
ose using one organism has been successful at laboratory-scale, and one mid-term (adv-AFEX–CBP) ethanol conversion processes

Table 3
Data for chemical input to ethanol conversion unit operations.

Conversion process Chemical Chemical input Assumptions and data sources


(kg/dry metric ton feedstock)
NREL–SSCFa H2SO4 26 Chemical loading specifications for H2SO4, LPG, cellulase and DAP taken from
Ca(OH)2 29 (NREL, 2001); for Ca(OH)2 taken from Sheehan et al. (2002). Chemical LCI data
LPG 0.2 sets taken from GaBi (PE, 2004). Chemical inputs are assumed for CS and SG
Cellulase 9.2 models
DAP 1.9
AFEX–SSCF NH3 20 Chemical loading specifications taken from MSU data (Teymouri et al., 2005;
LPG 0.2 Eggeman, 2005). Chemical LCI data sets taken from GaBi (PE, 2004). Chemical
Cellulase 9.6 inputs are assumed for CS and SG models
DAP 2.2
Advanced: AFEX–CBP NH3 8.1 Chemical loading specifications taken from Dartmouth College (Laser et al.,
DAP 2.2 2009) and Sendich (2006). LCI modules for all chemical and energy inputs taken
from GaBi (PE, 2004)
Cellulase production NH3 48 Input specifications (mass loadings) for each chemical and electricity input
(NH4)2 SO4 16 taken from NREL data on enzyme broth manufacture (Sheehan et al., 2002). The
CaO 58 broth has a concentration of 83.2 g cellulase/kg broth. LCI modules for all
Electricity (in MJ) 8.6 chemical and energy inputs, including electricity, taken from GaBi (PE, 2004).
LPG 0.2 The US national electricity grid mix is assumed
H2SO4 78

Notes: Diammonium phosphate ((NH4)2HPO4) is abbreviated as DAP; LPG, liquified petroleum gas; LCI, life cycle inventory; MSU, Michigan State University.
a
Sulfuric acid requirements specified in a design by Lloyd and Wyman (2005) call for significantly lower quantities (between 5 and 10 kg sulfuric acid/dry metric ton
biomass) than specified by NREL noted above.
S. Spatari et al. / Bioresource Technology 101 (2010) 654–667 659

examined. Data on chemicals used were taken from the commer- Yi is the yield (by weight) of sugar i, and Yi_E is the yield (by weight)
cial life cycle inventory database GaBi 4.2 (PE, 2004). Data on en- of ethanol from sugar i where i = 1 to 5 corresponds to glucose, xy-
zyme manufacture were not available in the database. Enzyme lose, arabinose, galactose and mannose, respectively. The yield of
manufacture can vary in its energy inputs and emission outputs each sugar following pre-treatment for the hemicellulose sugars,
depending on the enzyme family and the energy mix in the loca- and hydrolysis for the glucose in the cellulose fraction is estimated
tion of manufacture (Nielson et al., 2006). MacLean and Spatari as follows:
(2009) detail the chemical and energy inputs and GHG emissions
Y i ¼ fi  cj  bk ; ð2Þ
for near-term enzyme production. The mid-term CBP model does
not use this module because it is assumed to produce the enzymes where fi corresponds to the fraction of the sugar polymer in the
needed in-house. Chemical loadings and data sources are summa- feedstock while cj is the conversion efficiency for converting each
rized in Table 3. All models assume the co-production of electricity sugar polymer to sugar where, j = 1 for converting glucan to glucose
and steam from the combustible biomass portions not used for during hydrolysis, and j = 2–5 for converting xylan to xylose, arab-
ethanol production, namely the lignin, biogas from the anaerobic inan to arabinose, galactan to galactose and mannan to mannose,
digester used for treating process wastewater, undigested solids respectively, in the pre-treatment step. The term bk is a stoichiom-
(hemicellulose/cellulose fibres), and unfermented sugars. All of etric conversion parameter for the fermentation of each sugar to
the steam and most of the electricity is used to operate the plant; ethanol. b1 and b2 correspond to six-carbon (C6) and five-carbon
however, excess electricity not used by the plant may be exported (C5) sugars, respectively, and are determined by Eqs. (3) and (4):
to the electricity grid. All near-term and one mid-term model(s) as-
sume conversion of residual biomass, using a combuster-boiler and MWC6
b1 ¼ cC6 ð3Þ
MWC6 polymer
;
turbogenerator, known as Rankine (R) power generation, as speci-
fied in Aden et al. (2002); while one mid-term model assumes
steam and electricity production via the more efficient, relative MWC5
b2 ¼ cC5 ð4Þ
MWC5 polymer
:
to Rankine, integrated gasification combined-cycle (IGCC) technol-
ogy specified by Laser et al. (2009). We assume that all excess elec-
In Eqs. (3) and (4) cC6 is the theoretical conversion of C6 sugars
tricity is sold to regional electricity grids in the vicinity of the
to ethanol (0.511 g ethanol/g C6-sugar) and cC5 is the theoretical
ethanol facility, but assume this electricity displaces electricity
conversion of C5 sugars to ethanol (0.504 g ethanol/g C5-sugar).
from the average national US electricity grid since we do not define
MWC6/MWC6_polymer (180/162) and MWC5/MWC5_polymer (150/132)
specific plant locations. An electricity co-product credit is applied
represent the fractions of molecular weights of the C6 sugars to su-
following the system expansion method (ISO, 2006). We specify
gar polymers and C5 sugar to C5 sugar polymers, respectively.
the feedstock, pre-treatment, hydrolysis and fermentation pro-
The material balance around the ethanol conversion facility as-
cesses and energy recovery system abbreviations included in the
sumes that all combustible portions of the biomass that are not
10 LCA models in Table 4.
converted to ethanol could be converted to electricity and sold to
the grid since the process steam and electricity requirements are
2.4. LCA model equations
assumed to be fixed over different conversion efficiencies. Using
an overall material balance around the ethanol conversion facility,
We develop a set of equations to estimate the technological,
we estimate the excess electricity (EEl) to be sold to the grid as
environmental and energy metrics for the different technology
follows:
pathways shown in Table 4 based on the material balances of the
conversion processes shown in Fig. 1. All metrics are normalized EEl ¼ cb Eb ÿ Ep ; ð5Þ
to the volumetric flowrate Q_ E (L/h) of ethanol produced in the con-
where cb is a coefficient developed from the Aspen simulation mate-
version facility, which is also the functional unit:
rial balances, that determines the conversion to electrical energy
a X
5 from the biomass delivered. For near-term AFEX–SSCF pathways it
Q_ E ¼ M
_F  Y i  Y i E: ð1Þ is assumed to be 13%; for NREL–SSCF pathways, it is assumed to
qEtOH i¼1
be 16%; and for mid-term adv-AFEX–CBP it is assumed to be 29%
_ F represents the input mass flow rate of feedstock
In Eq. (1), M and 47% for the Rankine and IGCC electricity conversion processes,
(kg/h); a is the ethanol recovery efficiency following distillation, respectively. Eb is the energy delivered to the boiler, described be-
taken as 99.4% and 99.7%, for near and mid-term models, respec- low, and Ep is the fixed process energy and steam requirement used
tively; and qEtOH, is the density of ethanol (0.789 kg/L). The term by the ethanol conversion process; it is assumed to be 69,000 MJ/h

Table 4
Summary of lignocellulosic ethanol life cycle models and designations.

Model Designation Notes


T1 NREL-CS-2-SSCF-R NREL pre-treatment of CS with SSCF hydrolysis–fermentation of 2 sugars (glucose and xylose) and energy recovery via R
T2 AFEX-CS-2-SSCF-R AFEX pre-treatment of CS with SSCF hydrolysis–fermentation of 2 sugars and energy recovery via R
T3 NREL-SG-2-SSCF-R NREL pre-treatment of SG with SSCF hydrolysis–fermentation of 2 sugars and energy recovery via R
T4 AFEX-SG-2-SSCF-R AFEX pre-treatment of SG with SSCF hydrolysis–fermentation of 2 sugars and energy recovery via R
T5 NREL-CS-5-SSCF-R NREL pre-treatment of CS with SSCF hydrolysis–fermentation of 5 sugars (glucose, xylose, arabinose, mannose and galactose)
and energy recovery via R
T6 AFEX-CS-5-SSCF-R AFEX pre-treatment of CS with SSCF hydrolysis–fermentation of 5 sugars and energy recovery via R
T7 NREL-SG-5-SSCF-R NREL pre-treatment of SG with SSCF hydrolysis–fermentation of 5 sugars and energy recovery via R
T8 AFEX-SG-5-SSCF-R AFEX pre-treatment of SG with SSCF hydrolysis–fermentation of 5 sugars and energy recovery via R
T9 AFEX ad-SG-5-CBP-R Advanced AFEX pre-treatment of mid-term SG with CBP; fermentation of 5 sugars; and energy recovery via R
T10 AFEX ad-SG-5-CBP-I Advanced AFEX pre-treatment of mid-term SG with CBP; fermentation of 5 sugars; and energy recovery via IGCC

Notes: Pre-treatment systems: NREL = dilute acid, NREL technology; AFEX = ammonia fibre explosion, MSU technology; advanced AFEX ad = advanced AFEX technology, MSU
and Dartmouth College technology; Feedstocks: CS = corn stover, and SG = switchgrass, Hydrolysis–fermentation technologies; SSCF = simultaneous saccharification and co-
fermentation, and CBP = consolidated bioprocessing; Energy recovery: R = Rankine cycle, IGCC = integrated gasification combined-cycle.
660 S. Spatari et al. / Bioresource Technology 101 (2010) 654–667

for the AFEX–SSCF, 53,000 MJ/h for the NREL–SSCF, and 116,000 MJ/ ily have smaller ranges than those in Table 2 due to refinement in
h for the adv-AFEX–CBP production pathways, determined through design necessary for successful commercialization.
the Aspen simulations. Statistical distributions for the CS and SG feedstock fractions
The energy delivered to the boiler, Eb, is determined by the mass were generated with low/high ranges set to correspond to the
and energy content of the species delivered to the boiler. Eq. (6) 2.5th and 97.5th percentiles (95% confidence interval) in Table 2.
estimates Eb as the difference between the total energy in the input The compositions of 15 CS samples and 57 SG samples were ana-
biomass (MiLHVi) and the weighted average of the sugars con- lyzed to obtain the values (NREL, 2001). The hemicellulose and
verted to ethanol and their LHV. Therefore, at higher ethanol yields, hydrolysis conversion yields (values denoted as c) were assumed
less of the biomass is available for conversion to electricity. to be normally distributed. For NREL pre-treatment, the means
were set to those reported in laboratory studies (Aden et al.,
X
n X
5
Eb ¼ Mi LHV i ÿ M F  LHV E Yi  Yi E: ð6Þ 2002), as was the 2.5th percentile value for c2 while the 97.5th per-
i¼1 i¼1 centile value for c2 was set to the NREL (2001) Aspen simulation
value. The standard deviation for c1 was set equal to that of the
We model the technological performance of the different etha-
AFEX c1. For AFEX pre-treatment, the means, 2.5th and 97.5th per-
nol conversion facilities using the metrics ethanol yield YE (L/dry
centile values were taken from experimental data (Teymouri et al.,
metric ton, Eq. (7)) and electricity co-production EEl (MW, Eq.
2005).
(5)). The life cycle resource–environmental metrics are the fossil
The fermentation yield values for glucose and xylose were esti-
energy input, Efos, and the petroleum input, Epet (MJ/L ethanol,
_ GHG , and air pollutant emissions for dif- mated from a sensitivity analysis conducted by Aden et al. (2002)
Eq. (8); GHG emissions, M
_ j (g constituent/L ethanol, Eq. (9); and and normal distributions were assumed. The yield values for arab-
ferent pollutant species, M
_ SW (kg solid waste/L ethanol, Eq. (10)): inose, galactose and mannose were assumed to have triangular dis-
solid waste generation, M
tributions with the modes equal to 0 and the high values arising
Q_ E from the design case of Aden et al. (2002) and also used by Sheehan
YE ¼ ð7Þ et al. (2003). The triangular distributions may be overly conserva-
_F
;
M
tive for galactose and mannose but reflect that the fermentation
X
n X
n reactor will be less likely to be optimized for these low concentra-
ei;j
Ei ¼ ð8Þ tion sugars (Mabee, 2006). Both Aden et al. and Sheehan et al. re-
Q_ E
;
i¼1 j¼1 flect up-to-date experimental observations conducted by NREL
using the Z. mobilis organism. However, there are other organisms
where e is the energy input needed (MJ/h) for each step of the life
(e.g., recombinant Escherichia coli KO11 and K. oxytoca developed
cycle, i corresponds to fossil energy (i = 1) and petroleum (i = 2),
by Dr. Lonnie Ingram from the University of Florida) that have
and j corresponds to different inputs across the life cycle (e.g., farm-
achieved, at laboratory-scale, ethanol yields in the 90–95% range
ing energy, fertilizers, process chemicals).
for all sugars. We revisit this salient point in Section 3.4 when
X
n X
n
mi;j we discuss model sensitivity, and pre-treatment and fermentation
_i¼
M ð9Þ technology choices.
Q_ E
;
i¼1 j¼1
The Monte Carlo simulations employed 10,000 runs, for the
where m is the emission rate (g/h) for each step of the life cycle, near-term models only. We exclude a Monte Carlo analysis of
i = 1–6 corresponds to emissions of GHGs (weighted CO2, N2O, uncertainty for the mid-term pathways due to insufficient data
and CH4 in g CO2 equivalent), CO, NOX, NMOG, SOX, and PM10 to for assigning plausible variable ranges and probability distribu-
air; and j corresponds to emissions of each air emission species i tions for those technologies at this time.
associated with inputs across the life cycle:
X
n
mSW;i 3. Results and discussion
_ SW ¼
M ð10Þ
Q_ E
:
i¼1 We present results of technical and environmental metrics
Here, mSW,i corresponds to solid waste emissions from the ethanol developed for the ten near- and mid-term LCA models. Near-term
conversion process only. results are presented both as stochastic and point estimates based
The life cycle metrics calculated in Eqs. ()(8)–(10) include the on the Monte Carlo and Aspen simulation results, respectively. We
subtraction of an electricity co-product credit determined through discuss differences between the point and stochastic estimates in
Eq. (5). Section 3.3. Air pollutant emissions, solid waste generation and
the two mid-term technology model results are presented only
as point estimates based on the Aspen simulations. All model re-
2.5. Monte Carlo analysis
sults, unless otherwise specified, are presented as well-to-gate
(WTG), including all activities up through the ethanol production
Because lignocellulosic ethanol conversion technologies are not
step.
yet at commercial scale there is considerable uncertainty sur-
rounding the performance of these facilities. To examine this
uncertainty we obtained data on input variables reflecting perfor- 3.1. Ethanol yield and co-product electricity
mance that has been achieved at laboratory and/or PDU scale as re-
ported in the literature. We use Monte Carlo simulation to generate Ethanol and electricity co-product yields for the ten lignocellu-
stochastic distributions of performance metrics estimated through losic ethanol pathways are shown in Figs. 2 and 3, respectively. In
Eqs. (5), (7), and (8). Table 2 presents the values of the variables all near-term cases, ethanol yields are higher for the AFEX models
and the statistical distributions for each variable for the Monte Car- (T2, T4, T6, T8) than for the NREL models (T1, T3, T5, T7), whether
lo analysis. We selected the variable settings based on actual per- SG or CS is the feedstock. This is due to higher pre-treatment and
formance reported in the literature up to the time of our study. hydrolysis yields; and consequently, lower co-product electricity
Because the data used in the Monte Carlo analysis come from lab- for the AFEX cases since more of the feedstock is devoted to etha-
oratory-scale studies they are not meant to represent operating nol production. While the Aspen model results show that CS etha-
parameters for commercial-scale facilities, which would necessar- nol yields are consistently higher than those of SG for all models
S. Spatari et al. / Bioresource Technology 101 (2010) 654–667 661

450 to ethanol (i.e., in the form of a concentrated sugar syrup), can


Aspen model, Nth plant
be converted to electricity. While there may be limitations in con-
400 verting the syrup to electricity in practice, and likely a different
process design would be utilized if a facility were to be optimized
Yield (L/Mg)

350
for electricity rather than ethanol production, we assume the en-
300 ergy present in the syrup can be converted with the same effi-
ciency as other solids from the biomass in Eq. (6), in order to
250 estimate the potential for use of all residual biomass for excess
electricity.
200 When comparing near and mid-term Aspen simulation results,
ethanol yield and co-product electricity both increase due to the
150
T1 T2 T3 T4 T5 T6 T7 T8 T9/T10
compositionally advanced SG feedstock and mature technology.
Among the mid-term models, with its high efficiency, the IGCC en-
Fig. 2. Near- and mid-term model pathway ethanol yields. Notes: Near-term ergy recovery/electricity generation technology outperforms the
stochastic estimates based on Monte Carlo simulation are presented as box-plots. Rankine. Jin et al. (2009) studied a suite of IGCC technologies and
The top of the box represents the 75th percentile, the middle line represents the
found that IGCC, while requiring higher capital expenditure rela-
median (50th percentile) and the bottom of the box represents the 25th percentile.
The whiskers correspond to the 2.5th and 97.5th percentiles. Near and mid-term tive to Rankine, could sell electricity at a lower cost with the same
results shown as single-point estimates are based on Aspen simulations of nth internal rate of return due to its higher generating efficiency.
plants.

3.2. Life cycle fossil energy and petroleum input and greenhouse gas
140 emissions
Aspen simulation Nth plant
Co-product Electricity (MW)

120
We present WTG (ethanol production life cycle) fossil energy
100 input (Fig. 4), petroleum input (Fig. 5), and GHG emissions
80 (Fig. 6), respectively, for the ten lignocellulosic ethanol pathways.
All models assume relatively low fossil energy inputs throughout
60 the life cycle and include a credit for the electricity co-product,
40 which is why most results are negative. To provide as an example
a quantitative estimate of WTG GHG emissions (all presented in g
20
CO2 eq./L of ethanol), the near-term AFEX models range from ÿ170
0 to ÿ50 (Aspen Models) and ÿ350 to ÿ290 (median stochastic anal-
T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 ysis) while the near-term NREL models range from ÿ600 to ÿ250
(Aspen models) and ÿ1380 to ÿ1300 (median stochastic analysis).
Fig. 3. Near- and mid-term model pathway electricity co-product yields. Notes:
Near-term stochastic estimates based on Monte Carlo simulation are presented as
The fossil energy input and GHG emissions are both lower
box-plots. The top of the box represents the 75th percentile, the middle line (more negative) for NREL pre-treatment systems due to the larger
represents the median (50th percentile) and the bottom of the box represents the electricity credit. This observation would not be obvious without
25th percentile. The whiskers correspond to the 2.5th and 97.5th percentiles. Near examining how fossil energy inputs and GHG emissions vary with
and mid-term results shown as single-point estimates are based on Aspen
ethanol yields. The credits result from off-setting fossil energy and
simulations of nth plants.
GHG emissions from the US national electricity generation mix,
which uses mostly coal (approximately 56%) and other fossil en-
shown, when variations in glucan and xylan concentrations are ta- ergy inputs (natural gas and some crude oil), a technique following
ken into consideration, as shown in the stochastic estimates, etha- the ISO (2006) system expansion co-product allocation method,
nol yields do not vary significantly between the feedstocks with and one employed in most LCAs of lignocellulosic biofuels (e.g.,
the pre-treatment system held constant. Mid-term models, which Kim et al., 2009). Therefore, from a resource and climate change
use the adv-SG feedstock, assume comparatively higher concentra- perspective, in comparing the two pre-treatment processes, using
tions of cellulose, xylan, and lignin (the main source for thermal,
electrical, and co-product energy) relative to the near-term SG
5
feedstock (see Table 1), as the authors assume improved composi- Aspen simulation Nth plant
Fossil Energy Input (MJ/L)

tional yields due to maturation of the energy crop and a late season 0
single-cut harvest (Laser et al., 2009). This, along with the high
conversion efficiencies assumed in pre-treatment, hydrolysis, and -5
fermentation with the mature CBP technology (Table 2) is the T8 T9
T2 T4 T6
cause for the increase in both ethanol yield as well as electricity -10
production. The higher electricity production capacity is also due T10
to the larger plant capacity (5000 dry metric tons/day) assumed -15
in the mid-term relative to the near-term (2000 dry metric tons/
-20
day). T7
T1 T3 T5
Both Aspen simulation and stochastic results indicate that at
-25
higher ethanol yields, electricity production is lower due to con-
verting more of the biomass to ethanol rather than to electricity. Fig. 4. WTG fossil energy inputs for near- and mid-term model pathway. Notes:
The stochastic estimates show this pattern more pointedly over a Near-term stochastic estimates based on Monte Carlo simulation are presented as
range: ethanol yields at the upper percentile (97.5th) correspond box-plots. The top of the box represents the 75th percentile, the middle line
represents the median (50th percentile) and the bottom of the box represents the
to electricity yields at the lower percentile (2.5th), with the reverse 25th percentile. The whiskers correspond to the 2.5th and 97.5th percentiles. Near
true at the low ethanol/high electricity yield end, a pattern result- and mid-term results shown as single-point estimates are based on Aspen
ing from the assumption that any leftover biomass not converted simulations of nth plants.
662 S. Spatari et al. / Bioresource Technology 101 (2010) 654–667

1.6 the median and 2.5th percentile. Only petroleum input declines for
the mid-term technologies relative to all near-term technologies,
Petroleum Input (MJ/L)

and more so for the IGCC than for the Rankine pathway. This is pri-
1.2
marily due to the petroleum input not being significantly affected
by the credit, therefore it is not negative for any of the Aspen sim-
0.8 ulations shown. Additionally, petroleum input does not change
appreciably with improvements in ethanol yield, nor does it de-
pend on the pre-treatment system or the feedstock selected, as it
0.4
Aspen simulation Nth plant only varies from 1.2 to 1.4 MJ/L for all near-term models.

0 3.3. Simulation results and Monte Carlo analysis


T1 T2 T3 T4 T5 T6 T7 T8 T9 T10
With few exceptions, the Aspen model point-estimates fall out-
Fig. 5. WTG petroleum inputs for near- and mid-term model pathway. Notes: Near-
term stochastic estimates based on Monte Carlo simulation are presented as box-
side of the 95% confidence interval (CI) resulting from the Monte
plots. The top of the box represents the 75th percentile, the middle line represents Carlo analysis of fossil energy input and GHG emissions. This is be-
the median (50th percentile) and the bottom of the box represents the 25th cause many of the values of the model variables reported in the lit-
percentile. The whiskers correspond to the 2.5th and 97.5th percentiles. Near and erature based on recent experimental measurements indicate
mid-term results shown as single-point estimates are based on Aspen simulations
lower performance than that assumed in the Aspen simulations.
of nth plants.
In other words, the Aspen simulation values are not close to the
medians and in many cases lie outside of the 95% CI range specified
in the Monte Carlo models (see Table 2). The NREL Aspen model
200
designs assume hemicellulose and hydrolysis conversions higher
GHG emissions (g CO2 eq./L)

-200 than those that have been measured experimentally to date as re-
ported in the literature (the latter being selected as mean values in
-600 T9
the Monte Carlo analysis). The dilute acid pre-treatment system
T2 T6 T8
T4 tested at PDU-scale by NREL has not reported the high hemicellu-
-1000
T10 lose and hydrolysis conversion efficiencies that are noted in their
-1400 Aspen model (Aden et al., 2002; Sheehan et al., 2003). In contrast,
for certain AFEX model results, the Aspen point estimates lie with-
-1800 in the 95% CI due to researchers having demonstrated sugar yields
T1 T3 T5 T7 from hemicellulose and cellulose feedstock fractions in the labora-
-2200
Aspen model, Nth plant tory that are in line with those in the Aspen model. Of all near-term
models, only Aspen model estimates for the T4 model lie close to
Fig. 6. WTG GHG emissions for near- and mid-term model pathway. Notes: Near- the interquartile range. It is important to recognize, however, that
term stochastic estimates based on Monte Carlo simulation are presented as box- the NREL pre-treatment technology does not represent all dilute
plots. The top of the box represents the 75th percentile, the middle line represents
the median (50th percentile) and the bottom of the box represents the 25th
acid pre-treatment performance; Lloyd and Wyman (2005) and
percentile. The whiskers correspond to the 2.5th and 97.5th percentiles. Near and Wyman et al. (2005a,2005b) report higher hemicellulose sugar
mid-term results shown as single-point estimates are based on Aspen simulations and glucose yields (comparable to AFEX results) than those of NREL
of nth plants. (Sheehan et al., 2003; Aden et al., 2002) when testing CS substrates
with dilute sulfuric acid at laboratory-scale (comparable to AFEX
laboratory-scale results). We compare some technological differ-
more of the input biomass to produce electricity rather than etha- ences between NREL pre-treatment and results from Lloyd and
nol decreases total fossil energy consumption and GHG emissions Wyman (2005) and AFEX in Section 3.4. In addition, progress in
per liter of ethanol output, an observation that is consistent with the field that was not reported in peer-reviewed literature up to
results from literature (Wyman, 1994; Larson, 2006; Zhang et al., the time of this study is not reflected in our analysis. Comparing
2007). It does not affect total petroleum consumption appreciably technologies that are at an early stage of technological develop-
(Fig. 5), which is in accordance with results of Farrell et al. (2006), ment is a challenge, in particular because it is difficult to account
who show that petroleum consumption is not significant in the for how efficiency improvements in the technologies may evolve
ethanol life cycle, whether commercial-scale corn or future ligno- at the laboratory, pilot, or commercial scales.
cellulosic ethanol technologies are considered. For energy security Aspen simulation results in Fig. 2 indicate that ethanol yields
however, the higher ethanol yields achieved through the use of the increase by 9–12% when comparing one pre-treatment system
AFEX–SSCF production pathways, while not reducing total fossil (e.g., AFEX) with one feedstock (e.g., CS) in the two sugar versus
energy input and GHGs as much as the NREL–SSCF pathways, could five sugar fermentation cases. On the other hand, when examining
allow for greater displacement of gasoline if a large supply of bio- the stochastic results, there is hardly a difference between two and
mass is available to produce ethanol in the US in the range of bil- five sugar fermentation. This is due in part to the conservative
lions of gallons, important for energy security. probability distributions assigned to the fermentation of the three
In comparing mid-term with near-term Aspen simulation re- minor sugars, arabinose, galactose, and mannose in the Monte Car-
sults, when the Rankine electricity conversion system (T9) is em- lo simulation. We applied a conservative probability distribution
ployed, fossil energy inputs as well as GHG emissions decline for arabinose, the pentose sugar with the highest concentration fol-
slightly relative to those of the AFEX–SSCF models (T2, T4, T6, lowing xylose, because of the difficulty in fermenting it with the
and T8), but increase slightly relative to those of the NREL–SSCF organism Z. mobilis reported in literature (Sheehan et al., 2003).
models (T1, T3, T5, and T7). On the other hand, using IGCC electric- This would not be the case if a different organism were used, one
ity conversion (T10) reduces fossil energy and GHG emissions rel- shown, at laboratory-scale, to ferment at high yields all sugars
ative to all near-term Aspen simulation results (T1–T8) due to its (e.g., those developed by Ingram et al., 1987, 1991). What remains
large electricity credit. Stochastic fossil energy and GHG emissions uncertain at this time is how those organisms, whose success has
for NREL–SSCF models are lower than the mid-term IGCC model at been demonstrated at laboratory-scale, may perform in terms of
S. Spatari et al. / Bioresource Technology 101 (2010) 654–667 663

yield and cost at commercial scale. We use the same distribution holds additional uncertainties, including those related to electricity
for galactose and mannose fermentation since conditions in the markets, decision makers’ goals, policies, etc., not accounted for in
SSCF unit operation would more likely be set to optimize for glu- the results presented herein. Moreover, the assumption of includ-
cose and xylose fermentation (Mabee, 2006), which are in much ing a co-product credit itself carries uncertainty because these eth-
higher concentrations in the feedstock. anol plants do not exist and therefore secure markets for the co-
All near-term models were sensitive to the xylose fermentation products also do not exist. When evaluating future lignocellulosic
yield due to its high standard deviation (Table 2) in the stochastic ethanol facilities, the question of allocating a credit for the electric-
estimates, indicating that it is likely among the most important ity co-product is uncertain due to uncertainties in the regional
conversion variables requiring improved reliability in order to raise electricity markets (future generation mixes and transmission
ethanol yields. Following glucose, xylose is the most important su- capabilities) into which the co-product would be sold. We assumed
gar for fermentation because of its concentration in herbaceous a credit based on the US national average mix. However, if the co-
feedstocks. While there has been progress in developing organisms product were sold into regions with higher/lower fractions of coal-
to ferment xylose (Lawford and Rousseau, 2003; Sheehan et al., based electricity, the credit would be more/less pronounced.
2002), it remains a reaction process that has yet to be practiced To examine the effects of electricity credits on life cycle GHG
at commercial scale. For this reason, the probability distribution emissions, and to demonstrate the effect that improved dilute acid
for xylose fermentation (normal distribution with mean set at pre-treatment and fermentative organism choice has on WTG life
0.70 g ethanol/g xylose and larger standard deviation relative to cycle GHGs, we introduce Fig. 7, which shows GHG emission re-
that employed for glucose fermentation) assumed in this research sults for select near-term and mid-term pathways. In Fig. 7 we
was more conservative than if industrial-scale xylose fermentation introduce an additional case (L–W) that utilizes laboratory-scale
were practiced today, which would enable the selection of higher dilute acid pre-treatment measurements from Lloyd and Wyman
mean xylose-to-ethanol yields with tighter distributions as was as- (2005) along with fermentative organisms selected that are on
sumed for the glucose fermentation variable, which is based on par with those used in CBP. The L–W case is comparable to the
industrial-scale experience. A larger standard deviation for xylose near-term CS-AFEX (T6) case with respect to scale-up (from labo-
fermentation is also consistent with the sensitivity analysis that ratory measurement) and in its use of purchased enzymes, and
examined its effect on ethanol selling price by Aden et al. (2002). to the CBP models in its use of high-yielding fermentative organ-
Overall, the most robust results presented in Figs. 2–6 are the isms (95% conversion of all sugars to ethanol). The L–W case
near-term stochastic estimates. These ethanol production path- demonstrates high performance of dilute acid pre-treatment
ways make use of plausible ranges and probability distributions technology both in sugar and ethanol yields and net GHG
determined from laboratory results in combination with published emissions.
results from literature for setting the input variables that deter- When electricity credits are not considered, the uncertainty
mine ethanol yields and life cycle resource/environmental metrics. range tightens considerably over the 95% CI, and also the higher
The mid-term models incorporating mature lignocellulosic ethanol ethanol-yielding pathways have lower GHG emissions, the mid-
technology, while showing more promise in life cycle fossil energy term mature system (T9/T10) emitting the least of all pathways
and GHG emissions reductions than the near-term models, require shown. Among the near-term processes that do not assume elec-
additional research and development before we could assign cred- tricity credits, the AFEX technologies perform somewhat better
ible variable ranges and probability distributions with confidence. than the NREL ones under uncertainty, but probably more signifi-
cantly, the stochastic estimates lie close to the Aspen simulation
3.4. Co-product credit sensitivity estimates, therefore attesting to the promising performance of that
technology to date at laboratory-scale.
In view of energy and climate change impacts, the near-term,
lower ethanol-yielding technologies achieve the best performance
of all pathways shown in Figs. 4 and 6 due to exhibiting the lowest
1000
total fossil energy requirements and GHG emissions over the fuel Near-term Mid-term/
Advanced
production cycle. This apparent performance is due to the inclu-
GHG emissions (g CO2 eq/L)

500
sion of the electricity co-product credit. The credit assumes that
the electricity produced by the ethanol facility above that which
T9/T10
is required for the facility itself can be sold to the national US grid. 0
L-W
Under the system expansion method of co-product allocation, the CS
ethanol is credited with the energy use and emissions that would -500
T9
otherwise be associated with producing the same amount of elec- T6 T8
tricity using the average US electricity generation mix. Since the US -1000
mix is heavily reliant on fossil fuels, mostly coal, ethanol receives T10
significant fossil energy use and emissions credits, resulting in neg- -1500
ative fossil energy input and GHG emissions for ethanol production
as shown in Figs. 4 and 6. Furthermore, the wide 95% CI shown for T5
-2000
those same results is largely due to the electricity credit itself. The T7
electricity credit magnifies the negative fossil energy input and Stochastic w elec. Stochastic w/o elec. Aspen model elec. Aspen model w/o elec.
GHG emissions by a factor of 2–3 due to the large quantities of
Fig. 7. WTG GHG emissions including and excluding electricity co-product credits
both fossil energy and GHG emissions assumed to be displaced.
in near and mid-term time frames. Notes: Two cases are shown for each near-term
Since little electricity is generated in the US from petroleum technology: stochastic as box-plots over the 95% CI; and point estimates based on
(4.2%), the credit has little impact on that metric (Fig. 5). Inclusion Aspen models. In the mid-term/advanced technology case we include an Aspen
of the credit furthermore raises the question of whether, from a re- model pathway (L–W CS) that uses laboratory-scale dilute acid pre-treatment
source and climate change mitigation perspective, the biomass experimental data by Lloyd and Wyman (2005) and high fermentative yields of all
sugars, comparable to the CBP cases T9 and T10. The ‘L–W CS’ case is based on a
would be better utilized for displacing coal-derived electricity, an 2000 dry metric ton/day facility; 96% conversion of cellulose to glucose; 85%
observation that is consistent with results from literature (Wyman, conversion of hemicellulose to xylose, arabinose, galactose and mannose; pur-
1994; Larson, 2006; Zhang et al., 2007). This latter point however, chased enzymes; and ethanol yield 400 L/dry metric ton.
664 S. Spatari et al. / Bioresource Technology 101 (2010) 654–667

3.5. LCA model comparison with published results clude fuel transportation and distribution. We exclude electricity
credits from all pathways in Table 5 in order to assess input contri-
We compare our stochastic SG model results, as well-to-tank butions to each metric.
(WTT) results, to assess differences in the ethanol conversion mod- The GREET results are modeled after NREL processes and as-
ule, with those of GREET 1.7 (Wang, 2007), a publicly available sume relatively high ethanol yields and much lower electricity
peer-reviewed and widely cited LCA model that includes a herba- co-products (as they are based on the nth plant design discussed
ceous crop-based lignocellulosic ethanol pathway (which assumes earlier) compared with our models. Model T8, our process with
feedstock production identical to that for SG modeled in this work) the highest ethanol yield and lowest electricity capacity, shows re-
that utilizes dilute acid pre-treatment followed by biochemical sults closest to those of GREET; however, even with T8 the esti-
SSCF conversion (Table 5). The WTT results include those for the mated electricity co-product exceeds that of GREET by more than
WTG (presented up to this point in the paper) but additionally in- a factor of three. Our mean environmental results are noticeably

Table 5
Comparison of mean well-to-tank stochastic ethanol conversion models with GREET herbaceous crop model.

Yield (L/dry metric ton) Electricity Co-producta (MW) Electricity co-product (kWh/L) Fossil energy (MJ/L) Petroleum (MJ/L) GHG (g CO2 eq/L)
T3 210 ± 34 43 ± 5 2.5 ± 0.7 8.3 ± 1.2 3.1 ± 0.5 790 + 131/ÿ113
T4 290 ± 43 25 ± 6 1.1 ± 0.4 7.7 ± 1 2.2 ± 0.3 600 + 92/ÿ80
T7 220 ± 35 42 ± 5 2.3 ± 0.6 8.1 ± 1 3.0 ± 0.4 770 + 124/ÿ109
T8 290 ± 45 24 ± 6 1.0 ± 0.2 7.6 ± 1 2.2 ± 0.3 590 + 92/ÿ78
GREET 380 5 0.15 3.2 1.6 360

Notes: Model results do not include a co-product credit. GREET pathway is for 100% herbaceous (switchgrass) feedstock.
a
Electricity capacity for the GREET model was calculated from the electricity yield (kWh/gal), ethanol yield, and assuming a 2000 dry metric ton/day facility (GM/Argonne,
2005).

Table 6
Air pollutant emissions associated with feedstock (production, collection, and transport), process conversion, and electricity credit.

T1 T2 T3 T4 T5 T6 T7 T8 T9 T10
Air (g/L ethanol)
Carbon monoxide
Feedstock 0.5 0.5 0.7 0.6 0.5 0.4 0.6 0.5 0.4 0.4
Process 2.8 2.7 3.1 3.0 2.5 2.4 2.8 2.7 0.0 0.0
Electricity credit ÿ0.6 ÿ0.3 ÿ0.7 ÿ0.4 ÿ0.4 ÿ0.2 ÿ0.5 ÿ0.3 ÿ0.4 ÿ0.7
WTG 2.8 2.8 3.1 3.2 2.6 2.6 2.9 3.0 0.1 ÿ0.2
Nitrogen oxides
Feedstock 1 0.9 1 0.9 0.9 0.8 1 0.8 0.6 0.6
Process 3 3.0 3.3 3.3 2.7 2.7 3.0 3.0 0.1 0.1
Electricity credit ÿ3.3 ÿ1.8 ÿ3.9 ÿ2.3 ÿ2.4 ÿ1.2 ÿ3.2 ÿ1.8 ÿ2.1 ÿ4.0
WTG 0.7 2.1 0.4 1.9 1.2 2.3 0.8 2.1 ÿ1.4 ÿ3.3
Sulfur oxides
Feedstock 0.5 0.5 0.2 0.2 0.5 0.4 0.2 0.2 0.1 0.1
Process 4.0 1.6 4.3 1.8 3.5 1.5 4.0 1.6 0.07 0.07
Electricity credit ÿ6.2 ÿ3.5 ÿ7.4 ÿ4.3 ÿ4.5 ÿ2.3 ÿ6.0 ÿ3.3 ÿ3.9 ÿ7.5
WTG ÿ1.7 ÿ1.4 ÿ2.9 ÿ2.4 ÿ0.5 ÿ0.4 ÿ1.8 ÿ1.5 ÿ3.7 ÿ7.3
NMOG
Feedstock 0.2 0.2 0.4 0.3 0.2 0.2 0.3 0.3 0.2 0.2
Process 3.5 1.6 3.8 1.8 3.1 1.4 3.5 1.6 0.09 0.09
Electricity credit ÿ0.3 ÿ0.2 ÿ0.3 ÿ0.2 ÿ0.2 ÿ0.1 ÿ0.3 ÿ0.1 ÿ0.2 ÿ0.3
WTG 3.4 1.6 3.8 1.9 3.1 1.5 3.5 1.8 0.1 0.0
PM10 to air
Feedstock 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.08 0.08
Process 0.05 0.03 0.05 0.03 0.04 0.03 0.05 0.03 0.01 0.01
Electricity credit ÿ0.5 ÿ0.3 ÿ0.6 ÿ0.4 ÿ0.4 ÿ0.2 ÿ0.5 ÿ0.3 ÿ0.3 ÿ0.6
WTG ÿ0.3 ÿ0.1 ÿ0.4 ÿ0.2 ÿ0.2 ÿ0.06 ÿ0.3 ÿ0.1 ÿ0.2 ÿ0.5
NMOG disaggregated (process only)
Furfural 1.5 0 1.7 0 1.3 0 1.5 0 0 0
HMF 0.2 0 0.2 0 0.1 0 0.2 0 0 0
Ethanol 0.5 0.5 0.6 0.5 0.5 0.4 0.5 0.5 0.07 0.07
Acetic acid 1.1 0.9 1.2 1.0 0.9 0.8 1.1 0.9 0 0
Total 3.3 1.4 3.6 1.5 2.9 1.2 3.3 1.4 0.07 0.07

Solid (g/L ethanol)


Solid waste 340 210 370 240 300 190 340 220 88 88

For AFEX systems, including AFEX–SSCF and AFEX–CBP (T9 and T10), process-related emissions of PM are not monitored, rather the emissions shown result from process
inputs (cellulase production, and chemical and fuel inputs). Notes: HMF, hydroxymethylfurfural; NMOG, non-methane organic gases; WTG, well-to-gate.
S. Spatari et al. / Bioresource Technology 101 (2010) 654–667 665

higher than those of GREET: 85–155% (fossil energy), 35–100% NOx, SOx, and PM10, but approximately the same quantity of CO.
(petroleum), and 60–120% (GHGs). While all models we develop The NREL models generate higher quantities of solid waste for
assume the same feedstock production, collection and transport landfilling due to the by-product formation of gypsum, which is
module as in GREET, our ethanol yields are considerably lower, not produced with AFEX pre-treatment, and which is specified
thus requiring more fossil energy input per L of ethanol produced, for disposal by landfill in the Aspen simulation. Through changes
and we additionally account for primary energy and emissions in feedstock composition and ethanol yield improvements, the
associated with process chemicals used in ethanol conversion. mid-term technology, which employs the advanced AFEX technol-
The GREET model considers only the energy associated with the ogy, reduces solid waste generation per L of ethanol produced rel-
fuel used directly in the ethanol conversion facility, that used for ative to the near-term AFEX-pre-treated models.
operating forklifts and estimates an input of 0.06 MJ/L (213 BTU/
gal) of primary fossil energy input (Wang, 2007). Our models in-
clude the life cycle energy inputs and emission outputs of replace- 4. Conclusions
ment pre-treatment chemicals (quantities of chemicals not
recovered in the process), enzymes, and nutrients, in addition to Exploration of domestic, renewable energy alternatives for
the fuel for operating forklifts; and estimate a primary fossil energy light-duty vehicles has been motivated by stakeholders’ acknowl-
input into and GHG emission from ethanol conversion of between edging the need for decreasing the demand for imported crude
1.6 and 2.7 MJ/L and 180–205 g CO2 eq./L, respectively. Both lower oil and decarbonizing the fuel mix, as well as more generally, the
ethanol yields (and uncertainty therein) and the addition of pro- need for near and mid-term improvements to the sustainability
cess chemical input energy and emissions give rise to differences of personal transportation. As shown in this research, lignocellu-
in this study’s lignocellulosic ethanol pathway results in relation lose-based ethanol produced through a set of emerging conversion
to the GREET pathway. technologies offers promise for making progress toward both en-
ergy security and climate change mitigation goals. The research
3.6. Air pollutant and solid waste impacts quantifies the resource investments (e.g., fossil, petroleum energy)
required to convert two potential feedstocks to ethanol utilizing
Table 6 shows WTG model air pollutant emissions and solid specific near and mid-term pre-treatment and hydrolysis–fermen-
waste generation. In all near-term scenarios (T1–T8), the ethanol tation technologies. Greenhouse gas and air pollutant emissions
conversion process dominates emissions of CO, NOx, SOx, and associated with the technologies are also quantified. The technolo-
NMOG. Even when factoring in the electricity credit reduction, gies examined are found to exhibit differences in both technologi-
the ethanol conversion (process)-related emissions of the four pol- cal and life cycle environmental performance. Surprisingly and
lutants dominate. This result is consistent with earlier work by perhaps counter-intuitively, low ethanol-yielding conversion tech-
Spatari et al. (2005), which did not include an electricity credit. nologies outperform, in terms of fossil energy inputs and GHG
With the addition of air pollutant controls on the energy recovery emissions, even the highest yielding mature mid-term technolo-
equipment, all air pollutants noted above would be reduced over gies when credits, based on the US national electricity generation
the fuel production cycle, although they could still be significant mix, are assumed to be displaced by the ethanol facility’s electric-
at the feedstock collection and transport steps. Mid-term air pollu- ity co-product. However, outside of biomass-based fuels, there are
tant emissions are also lower than near-term models due to higher no near-term low carbon liquid fuel alternatives for light-duty
ethanol yields, lower chemical loadings, and in situ enzyme pro- vehicle transportation available at reasonable cost. Converting bio-
duction. The near-term boiler/combustor design specified by Aden mass to ethanol is an option for reducing the carbon intensity of
et al. (2002), which we use in all near-term models, releases high the sector while making progress toward energy security goals.
SOX emissions originating from sulfur in the feedstock, and lacks Further reduction of the carbon intensity of lignocellulosic ethanol
emission control for that species (Sheehan et al., 2002). This issue is possible by lowering carbon emissions along its supply chain, for
was addressed by Laser et al. (2009) in the CBP design, and would example, using biobased fuels in farm equipment and for transport
need to be included in future improvements of near-term process (MacLean et al., 2000) and expanding the cadre of biorefinery prod-
designs, as it would result in a better life cycle profile for ethanol. ucts, such as producing process chemicals such as ammonia from
Installation of emission controls at ethanol conversion facilities anaerobic digester gas.
would be expected to be required to limit emissions; they would When examining the uncertainty in the performance of an eth-
increase the cost of production and would increase the facility’s de- anol technology based on published data and comparing this to the
mand for energy, which is produced on-site. This would change the best estimate of performance of an nth plant developed through
electricity co-product output, and associated credits, but would Aspen simulation, the difference in environmental performance is
likely not have a large impact on the carbon profile of the fuel life considerable. Most publicly available LCA models (e.g., GREET) as-
cycle. sume technology performance and associated ethanol yields that
The emissions of NMOG in Table 6 show that the AFEX pre- reflect nth plant performance. This assumption necessarily affects
treatment system emits less NMOG overall, and in particular no resource and environmental metrics as they are allocated to the
aldehydes, because it is an alkali pre-treatment system (Sendich, unit of fuel produced.
2006); whereas NREL’s pre-treatment generates some cyclic alde- Monte Carlo analysis represents a formal mechanism for appro-
hydes (furfural and hydroxymethyl furfural (HMF)) when hemicel- priately making predictions from uncertain input, which is a rele-
lulose reacts with dilute acid. Of the aldehyde emissions produced vant analytical approach for evaluating lignocellulosic ethanol due
in the NREL Aspen design, more than half are sent to the wastewa- to the technologies’ state of development. However, the results of
ter treatment system, but some stay with the pre-treated slurry the Monte Carlo analysis are only as robust as the data and proba-
and migrate through to various water recycle streams, a fraction bility distributions that go into the analysis. When additional data
of which is used to wash the incoming feedstock, where they vol- on pre-treatment, hydrolysis and fermentation yields reported at
atilize, and where emission controls would need to be installed. All laboratory and pilot-scale become available, it will be possible to
other air pollutant emissions, aside from NMOG, are roughly equiv- improve predicted life cycle performance of lignocellulosic ethanol
alent between the NREL and AFEX pre-treatment systems for the technologies further. We recognize that the technologies examined
ethanol conversion process. It is when co-product credits are taken herein represent a small sample from a larger pool of technologies
into consideration, that the NREL system emits lower quantities of that are improving with time due to the dynamic nature of the
666 S. Spatari et al. / Bioresource Technology 101 (2010) 654–667

lignocellulosic ethanol field. The technologies we selected for Laser, M., Jin, H., Jayawardhana, K., Lynd, L.R., 2009. Co-production of ethanol
and power from switchgrass. Biofuels, Bioproducts and Biorefining 3, 195–
investigation may have improved through additional research
218.
and development and continued industry initiatives since the time Lawford, H.G., Rousseau, J., 1992. Fuel ethanol from corn residue prehydrolysate by
of data collection for this study. a patented ethanologenic Escherichia coli B. Biotechnology Letters 24 (5), 421–
Most importantly, in spite of the variability and uncertainty in 426.
Lawford, H.G., Rousseau, J.D., 2003. Cellulosic fuel ethanol–Alternative fermentation
technological and environmental performance we report for the process designs with wild-type and recombinant Zymomonas mobilis. Applied
lignocellulosic ethanol pathways, all of these technologies offer a Biochemistry and Biotechnology 105–8, 457–469.
positive (fossil) energy gain and a substantial opportunity to re- Lloyd, T.A., Wyman, C.E., 2005. Total sugar yields for pretreatment by hemicellulose
hydrolysis coupled with enzymatic hydrolysis of the remaining solids.
duce GHG emissions relative to gasoline and corn ethanol, fuels Bioresource Technology 96, 1967–1977.
that are predominantly used in light-duty vehicles on the roads Lynd, L.R., 1996. Overview and evaluation of fuel ethanol from cellulosic biomass:
today. technology, economics, the environment, and policy. Annual Review of Energy
and the Environment 21, 403–465.
Lynd, L.R., Wiemer, P.J., van Zyl, W.H., Pretorius, I.S., 2002. Microbial cellulose
utilization: fundamentals and biotechnology. Microbiology and Molecular
Acknowledgements Biology Reviews 66, 506–577.
Mabee, W., 2006. Personal communication. Research Associate in Forest Science,
We thank Dr. Bruce Dale of Michigan State University and Dr. University of British Columbia, Vancouver, British Columbia, November 29,
2006.
Lee R. Lynd of Dartmouth College for providing data used in this re- MacLean, H.L., Spatari, S., 2009. Contribution of enzymes and process chemicals to
search and the Natural Sciences and Engineering Research Council the life cycle of ethanol. Environmental Research Letters 4, 1–10.
(Canada), AUTO21 Network Centre of Excellence and Government MacLean, H.L., Lave, L.B., Lankey, R., Joshi, S., 2000. A life cycle comparison of
alternative automobile fuels. Journal of the Air & Waste Management
of Ontario Early Researcher Award for financial support.
Association 50, 1769–1779.
Mosier, N., Wyman, C., Dale, B., Elander, R., Lee, Y.Y., Holzapple, M., Ladish, M., 2005.
Features of promising technologies for pretreatment of lignocellulosic biomass.
References Bioresource Technology 96, 673–686.
Nielson, P., Oxenbøll, K., Wenzel, H., 2006. Cradle to gate environmental assessment
Aden, A., Ruth, M., Ibsen, K., Jechura, J., Neeves, K., Sheehan, J., Wallace, R., 2002. of enzyme products produced industrially in Denmark by Novozymes A/S. The
Lignocellulosic biomass to ethanol process design and economics utilizing co- International Journal of Life Cycle Assessment 12, 432–438.
current dilute acid prehydrolysis and enzymatic hydrolysis for corn stover. NREL, 2001. Aspen Plant Design Model, un published.
NREL/TP-510-32438. National Renewable Energy Laboratory, Golden, CO. PE, 2004. GaBi 4: The Software System for Life Cycle Engineering. PE-Europe GmbH
Brinkman, N., Wang, M., Weber, T., Darlington, T., 2005. Well-to-Wheels Analysis of and IKP University of Stuttgart, March 2003.
Advanced Fuel/Vehicle Systems – A North American Study of Energy Use, Sendich, E.N., 2006. Personal communication, Michigan State University, East
Greenhouse Gas Emissions, and Criteria Pollutant Emissions. General Motors, Lansing, Michigan, June 12, 2006.
Argonne National Laboratory. <http://www.transportation.anl.gov/pdfs/TA/ Sendich, E.N., Laser, M., Seungdo, K., Alizadeh, H., Laureano-Perez, L., Dale, B., Lynd,
339.pdf>. L., 2008. Recent process improvements for the ammonia fibre expansion (AFEX)
Conant, R.T., Paustian, K., Del Grosso, S.J., Parton, W.J., 2005. Nitrogen pools and process and resulting reductions in minimum ethanol selling price. Bioresource
fluxes in grassland soils sequestering carbon. Nutrient Cycling in Technology 99 (17), 8429–8435.
Agroecosystems 71, 239–248. Sheehan, J., Aden, A., Riley, C., Paustian, K., Killian, K., Brenner, J., Lightle, D., Walsh,
Congress, US, 2005. Energy Policy Act. 109th Congress ed, pp. 551. M., Cushman, J., Nelson, R., 2002. Is Ethanol from Corn Stover Sustainable?
Eggeman, T., 2005. Personal communication, FRII Consulting, Denver, Colorado, (Draft). National Renewable Energy Laboratory, Golden, CO.
September 14, 2005. Sheehan, J., Aden, A., Paustian, K., Killian, K., Walsh, M., Nelson, R., 2003. Energy and
Eggeman, T., Elander, R.T., 2005. Process and economic analysis of pretreatment environmental aspects of using corn stover for fuel ethanol. Journal of Industrial
technologies. Bioresource Technology 96, 2019–2025. Ecology 7, 117–146.
Farrell, A.E., Plevin, R.J., Turner, B.T., Jones, A.D., O’Hare, M., Kammen, D., 2006. Solomon, B.D., Barnes, J.R., Halvorsen, K.E., 2007. Grain and cellulosic
Ethanol can contribute to energy and environmental goals. Science 311, 506– ethanol: history, economics and energy policy. Biomass and Bioenergy
508. 31, 416–425.
Fu, G.Z., Chan, A.W., Minns, D.E., 2003. Life cycle assessment of bio-ethanol derived Spatari, S., 2007. Biomass to ethanol pathways: evaluation of lignocellulosic ethanol
from cellulose. The International Journal of Life Cycle Assessment 8 (3), 137– production technologies. Ph.D. Dissertation. University of Toronto, Toronto, ON.
141. Spatari, S., Zhang, Y., MacLean, H.L., 2005. Life cycle assessment of switchgrass and
Hahn-Hägerdal, B., Wahlbom, C., Gardonyi, M., van Zyl, W.H., Otero, C.R.R., Jönsson, corn stover-derived ethanol fueled automobiles. Environmental Science &
L., 2001. Metabolic engineering of Saccharomyces cerevisiae for xylose Technology 39 (24), 9750–9758.
utilization. Advances in Biochemical Engineering/Biotechnology 73, 53–84. Sperling, D., Yeh, S., 2009. Low carbon fuel standards. Science and Technology
Ingram, L.O., Conway, T., Clark, D.P., Sewell, G.W., Preston, J., 1987. Genetic (Winter), 57–66.
engineering of ethanol production in Escherichia coli. Applied and Teymouri, F., Laureano-Perez, L., Alizadeh, H., Dale, B.E., 2005. Optimization of the
Environmental Microbiology 53 (10), 2420–2425. ammonia fiber explosion (AFEX) treatment parameters for enzymatic
Ingram, L., Conway, T., Alterthuin, F., 1991. Ethanol production by E. coli strains co- hydrolysis of corn stover. Bioresource Technology 96, 2014–2018.
expressing Zymomonas PDC and ADH genes. US Patent Office, National US Congress, 2007. Energy Independence and Security Act. 110th Congress ed., H.R.
Renewable Energy Laboratory (NREL), Washington, DC. 6, January 4, 2007, pp. 310.
Iogen, 2007. Iogen Corporation. <http://www.iogen.ca/cellulose_ethanol/ US DOE, 2007. DOE Selects Six Cellulosic Ethanol Plants for up to 385 Million in
what_is_ethanol/process.html> (accessed 21.01.07). Federal Funding. <http://www.energy.gov/news/4827.htm> (28.02.07).
IPCC, 1997. International Panel on Climate Change (IPCC). Revised 1996 IPCC Vogel, K.P., Brejda, J.J., Walters, D.T., Buxton, D.R., 2002. Switchgrass biomass
Guidelines for National Greenhouse Gas Inventories. The Greenhouse Inventory production in the Midwest USA: harvest and nitrogen management. Agronomy
Reference Manual, vol. 3. In: International Panel on Climate Change, United Journal 94, 413–420.
Nations Environment Program, Organization for Economic Cooperation and Wang, M.Q., 1999. GREET 1.5, Transportation Fuel-Cycle Model. Methodology, Use,
Development, International Energy Agency, Paris, France. and Results, vol. 1. ANL/ESD-39. Center for Transportation Research, Argonne
ISO, 2006. ISO 14041: Environmental management: life cycle assessment – Goal and National Laboratory.
scope definition and inventory analysis. ISO, Geneva. Wang, M.Q., 2007. GREET 1.7, Transportation Fuel-Cycle Model. Argonne National
Jin, H., Larson, E.D., Celik, F.E., 2009. Performance and cost analysis of future, Laboratory, Illinois, USA (accessed 23.11. 06).
commercially mature gasifi cation-based electric power generation from Wilhelm, W.W., Johnson, J.M.F., Karlen, D.L., Lightle, D.T., 2007. Corn stover to
switchgrass. Biofuels, Bioproducts and Biorefining 3, 142–173. sustain soil organic carbon further constrains biomass supply. Agronomy
Kim, S., Dale, B., 2004. Cumulative energy and global warming impact from the Journal 99, 1665–1667.
production of biobased products. Journal of Industrial Ecology 7, 147–162. Wooley, R., Ruth, M., Sheehan, J., Ibsen, K., Majdeski, H., Galvez. A., 1999.
Kim, S., Dale, B., 2005. Life cycle assessment of various cropping systems utilized for Lignocellulosic Biomass to Ethanol Process Design and Economics Utilizing
producing biofuels: bioethanol and biodiesel. Biomass and Bioenergy 29, 426– Co-current Dilute Acid Prehydrolysis and Enzymatic Hydrolysis: Current and
439. Futuristic Scenarios. NREL/TP-580-2615. National Renewable Energy
Kim, S., Dale, B.E., Jenkins, R., 2009. Life cycle assessment of corn grain and corn Laboratory, Golden, CO.
stover in the United States. The International Journal of Life Cycle Assessment Wu, M., Wang, M., Huo, H., 2006. Fuel-Cycle Assessment of Selected Bioethanol
14, 160–174. Production Pathways in the United States. Argonne National Laboratory, Energy
Lal, R., 2004. Is crop residue a waste? Journal of Soil and Water Conservation 59, Systems Division.
136A–139A. Wyman, C.E., 1994. Alternative fuels from biomass and their impact on carbon
Larson, E.D., 2006. A review of life-cycle analysis studies on liquid biofuel systems dioxide accumulation. Applied Biochemistry and Biotechnology 45 (46), 897–
for the transport sector. Energy for Sustainable Development X (2), 109–126. 915.
S. Spatari et al. / Bioresource Technology 101 (2010) 654–667 667

Wyman, C.E., Dale, B.E., Elander, R.T., Holtzapple, M., Ladisch, M.R., Lee, Y.Y., 2005a. pretreatment technologies to corn stover. Bioresource Technology 96, 2026–
Coordinated development of leading biomass pretreatment technologies. 2032.
Bioresource Technology 96, 1959–1966. Zhang, Y., Habibi, S., MacLean, H.L., 2007. Environmental and economic evaluation
Wyman, C.E., Dale, B.E., Elander, R.T., Holtzapple, M., Ladisch, M.R., Lee, Y.Y., 2005b. of bioenergy in Ontario, Canada. The Journal of the Air & Waste Management
Comparative sugar recovery data from laboratory scale application of leading Association 57, 919–933.

Anda mungkin juga menyukai