Applied Energy 307 (2022) 118201

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Comparing biofuels through the lens of sustainability: A data envelopment

analysis approach
Richard Cabrera-Jiménez a, Josep M. Mateo-Sanz a, Jordi Gavaldà a, Laureano Jiménez a,
Carlos Pozo b, *
a
Departament d’Enginyeria Quimica, Universitat Rovira i Virgili, Av. Països Catalans 26, 43007 Tarragona, Spain
b
Department of Chemical Engineering, Universitat Politècnica de Catalunya, EEBE, Av. Eduard Maristany 16, 08019 Barcelona, Spain

H I G H L I G H T S G R A P H I C A L A B S T R A C T

• We compare the performance of 72

biofuel routes in terms of 12 sustain­
ability indicators.
• DEA assigns an efficiency score to each
biofuel, ranking them from best to
worst.
• Renewable diesel was found the best
fuel type, followed by biodiesel and
ethanol.
• Waste biomass is preferred over ligno­
cellulosic and 1st generation carbon
sources.
• Benefits from demand-side measures are
on pair with improving biofuel
manufacturing.

A R T I C L E I N F O A B S T R A C T

Keywords: Liquid biofuels can facilitate the transition towards a more sustainable transportation sector by curbing carbon
Data envelopment analysis emissions while maintaining most of the current vehicle fleet. Today, a myriad of alternatives are available to
Renewable energy produce biofuels, where different decisions for the fuel type, blend, conversion process and carbon source will
Sustainable development
affect the final cost and environmental impact of the product. In this contribution, we analyze the performance of
Transport
Biofuels
72 different biofuels routes based on 12 indicators that cover the three sustainability dimensions: economic,
LCA environmental and social. The proposed multi-criteria approach combines Data Envelopment Analysis with Life
Cycle Assessment to evaluate biofuels from a cradle-to-wheel perspective, that is, considering the production
chain spanning from biomass production to the combustion of the biofuel in the engine. Results reveal that there
are 35 biofuels routes performing better than the rest, with renewable diesel being a better option than ethanol-
based blends or biodiesel, and waste biomass preferred over cellulosic biomass or bio-oils. The selection of the

Abbreviations: B20, Diesel fuel with up to 20%v/v FAME content; CIDI, Compression Ignition Direct Injection; DEA, Data Envelopment Analysis; DMU, Decision
Making Unit; E10, Gasoline fuel with up to 10%v/v bioethanol content; E85, Gasoline fuel with up to 85%v/v bioethanol content; FAME, Fatty Acid Methyl Ester;
FWET, Freshwater ecotoxicity; FWEU, Freshwater eutrophication; GWP, Global Warming Potential; HT, Human ecotoxicity; HVO, Hydrotreated Vegetable Oil; LCA,
Life Cycle Assessment; LCI, Life Cycle Inventory; LCIA, Life Cycle Impact Assessment; LO, Land Occupation; MSW, Municipal Solid Waste; NOx, Nitrogen Oxides;
PMFP, Fine Particulate Matter Formation Potential; POFP, Photochemical Oxidant Formation Potential; RDI, Renewable Diesel Production Based on SuperCetane;
RDII, Renewable Diesel Production Based on fluid catalytic cracker technology; SI, Spark Ignition; SBM, Slack Based Measure; TA, Terrestrial Acidification; TE,
Terrestrial Ecotoxicity; VRS, Variable return to scale.
* Corresponding author.
E-mail address: [email protected] (C. Pozo).

https://doi.org/10.1016/j.apenergy.2021.118201
Received 12 July 2021; Received in revised form 26 October 2021; Accepted 8 November 2021
Available online 27 November 2021
0306-2619/© 2021 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

carbon source proofed to be the most important decision, highlighting the need to consider regional aspects
related to soil and climate before promoting a certain biofuel. Overall, our results can help to derive effective
policies for the adoption of biofuels attaining the best performance at minimum cost and environmental risks.

1. Introduction 2050 in the United States [24]. Despite their advantages, biofuels are not
exempt from negative side-effects, mainly related to the competition for
The continued growth of world population and the adoption of land and water use [25].
higher standards of living have risen energy demand to unprecedented As aforementioned, biofuels can be produced using different sources
levels. In the scenarios developed before the COVID-19 crisis, energy and processes, each generating different environmental impacts and
demand was projected to grow by 12% between 2019 and 2030 [1]. achieving distinct performance in engines. In this context, the identifi­
Among energy-consuming sectors, transport is the main player by the cation of the most convenient biofuels considering simultaneously the
use of oil, covering 92% of fuel demand [2]. The widespread use of fossil three sustainability pillars –economic, environmental and social– calls
fuels is the main anthropogenic source of greenhouse gases, responsible for multi-criteria decision-making tools (MCDM) [26]. The usefulness of
for climate change [3]. This evidences the fact that current practices for such tools in solving environmental, socio-economic and technical
energy production are still far from sustainable [4], which raises con­ barriers involved in energy planning has been widely acknowledged
cerns on the associated impacts in several environmental dimensions [27].
such as global warming, human health, land use or resource depletion Different MCDM methods such as analytical hierarchy process [28],
[3–5]. Multi-attribute value theory [29] and Data Envelopment Analysis (DEA)
In 2018, only 3.7% of fuel demand for transport was covered by have been applied to assess different energy systems [30]. Amongst
renewable energy; with most of this being shouldered by biofuels (93%) MCDM tools, we resort here to DEA, a non-parametric method for
and the rest provided by renewable electricity [6]. Biofuels such as benchmarking alternatives [31]. The main advantage of DEA over other
biodiesel and bioethanol have been considered promising alternatives to multi-criteria assessment methods is its capacity to combine multiple
fossil fuels for sustainable development due to their high potential to indicators into a single performance score, avoiding the need to define
mitigate climate change [7–9]. Environmental pollution policies such as subjective weights between the indicators. This is very convenient in
the Paris Agreement and the European Green Deal consider the wide­ sustainability assessment as it allows to integrate indicators covering the
spread use of biofuels could importantly contribute to reaching reduc­ three sustainability dimensions into a single metric, classifying alter­
tion targets of 80–95% for greenhouse gas emissions by 2050 [10–12]. natives as efficient or inefficient. In addition, DEA provides information
Many countries, e.g., the USA, Brazil, EU, China, have launched biofuel on how much room for improvement is possible in inefficient alterna­
programs to reduce the use of fossil fuels in transport, and it is expected tives compared to the best-performing processes.
that the global share of biofuels in this sector will reach 17% by 2050 During the last years, some authors have combined Life Cycle
[6]. Assessment (LCA) with DEA to assess the overall level of sustainability of
Biofuels refer to solid, liquid, and gaseous fuels that are produced alternatives, enabling the identification of efficient processes with a
from renewable biological sources. The most common biofuel is bio­ focus on their sustainable performance. Examples of this combined
ethanol, representing 82% of the total biofuel produced today [13]. Its application include liquid fuels production [32], electricity generation
main manufacturers are the United States and Brazil, with an annual [33], bioenergy systems, [34], milk production [35], mussel cultivation
production volume of 59.7 and 34.4 billion liters in 2020 and 2019, [36], and grape production for vinification [37]. In the case of biofuels,
respectively [14–15]. The second most widely produced biofuel –and previous works using DEA focused on particular features or echelons of
the most common in Europe [16]– is biodiesel, obtained by trans­ the biofuel supply chain, e.g., cultivation locations [38], the biofuel
esterification of oils or fats. Raw materials for biodiesel include vege­ production process [39], or the logistic network [40]. In other cases, the
table oils, animal fats, and algae (third-generation biofuel), among focus was put on a particular carbon source, would it be sugarcane [41]
others [17]. Bioethanol and biodiesel share the feature that can be used or algae [42], evaluating the complete supply chain of individual
in internal combustion engines due to their high-octane number and products such as bioethanol [43] and biodiesel [44]. While some of
high heat of vaporization [18], being both suitable either as an additive these works assessed the life cycle of biofuels, their scope covered, at
in gasoline blends or as pure fuels in modified engines. most, stages up to the production of the fuel (cradle-to-tank), thus
Another relevant biofuel is renewable diesel (RD), sometimes called neglecting the combustion of the fuel during vehicle use (tank-to-
“second-generation biodiesel,” “green diesel,” or ‘‘HVO’’ (hydrotreated wheel). Since this is the stage where most of the emissions take place and
vegetable oil) [19]. This biofuel is chemically similar to petroleum diesel acknowledging that not all fuels show the same performance (in terms of
(i.e., composed mainly of paraffins) but can be produced from a emissions and energy efficiency) in vehicle engines, the inclusion of this
renewable feedstock containing triglycerides and fatty acids through stage in the analysis is crucial to obtain a holistic assessment of biofuels
various processes such as hydrotreating, gasification and pyrolysis [17]. throughout their complete life cycle.
Similar to biodiesel, its properties allow its use in conventional engines In this contribution, we evaluate the performance of 72 different
either as an additive or as a pure fuel [20–22]. routes for the production of biofuels considering the three sustainability
In addition to curbing greenhouse gas emissions, the production of dimensions, which are quantified here based on 12 different indicators.
biofuels can offer other ancillary benefits to society [23]. On the one The 72 routes result from selected combinations of four biofuel blends,
hand, it can diversify the supply of fuel to the transportation sector, six possible fuels (i.e., ethanol, biodiesel, RD or HVO, diesel and gaso­
providing a sustainable alternative to the existing transportation struc­ line) and 19 types of biological feedstocks. The analysis considers the
ture. On the other hand, it can also allow diversification of farmland whole life cycle of the biofuels, including cultivation, production, dis­
while strengthening domestic agriculture by promoting biofuel feed­ tribution, and final use of the fuel in combustion vehicles (i.e., cradle-to-
stocks according to their geographical location and resource availabil­ wheel), everything quantified via LCA [45]. The resulting MCDM
ity. In many cases, biofuels are suitable for current combustion engines problem is solved with DEA [30] with the objective of evaluating and
and fuel stations, providing an interim solution before the required identifying the most suitable biofuel routes, which will be deemed
infrastructure for electric vehicles is in place. Note that, while electricity efficient. For non-suitable biofuel routes, labelled as inefficient, we
is the fastest-growing energy source in the transportation sector, it is provide quantitative improvement targets that, if attained, would make
projected to account for less than 2% of transport fuel consumption in them efficient. Finally, the presented contribution aims to provide a

2
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

powerful framework for holistic assessments that could help policy- On the other hand, biofuels originally deemed efficient are further
makers to develop better-informed regulations and achieve this way ranked in Step 3 by using a different DEA model based on a so-called
the emission reduction targets of current environmental policies for the super-efficiency score [46]. The combination of these results with the
transportation sector. efficiency scores from Step 2 allows to build a sorted list from the best to
The remaining of this manuscript is structured in three sections as the worst-performing biofuels that could aid policy-makers in devel­
follows. Section 2 describes the methodology developed to evaluate oping effective regulations.
biofuel production from a sustainability perspective and a cradle-to- Finally, in Step 4, results are analyzed and interpreted considering
wheel scope, paying special attention to DEA and its integration in the the performance that selected biofuels could attain in different sce­
proposed framework. In Section 3, results are presented and analyzed in narios. Potential roadmaps for improvement are also discussed.
detail. Finally, in the conclusions, the implications for the technological,
political and social spheres are discussed. 2.1. Data acquisition

2. Methodology The methodology described is used to compare the performance of

72 biofuel routes. This myriad of biofuel alternatives is obtained by
The methodology used to assess the performance of biofuels consists combining selected options for the carbon source, the production pro­
of four main steps articulated around DEA, which is the cornerstone of cess, the fuel type and the car engine where the biofuel will be used in
our approach (Fig. 1). These steps are briefly summarized next, while (Fig. 2). Specifically, 19 types of biological feedstocks are considered as
further details are provided in the ensuing subsections. carbon sources; these cover lipids (i.e., vegetable oils, animal fats, and
Step 1 aims to obtain the data required to compute the indicators that algae), cellulosic material (e.g., crop residues or woody biomass) and
will be used to assess the sustainability performance of the biofuels. This dedicated energy crops (e.g., sugarcane, maize) [47]. Regarding biofuel
requires the collection of different types of data: from mass and energy production processes, four types are studied: (i) fermentation of sugars
balances for biofuel production processes, to traditional LCA data and (i.e., glucids) and (ii) biomass gasification to produce ethanol, (iii)
complementary information such as costs. transesterification of lipids (i.e., triglycerides) to produce FAME (fatty
With this information at hand, efficiency scores are computed for acid methyl esters), and (iv) hydrotreating of lipids (i.e., triglycerides) to
each biofuel using DEA in Step 2. To this end, each biofuel is modelled as produce hydrotreated vegetable oil (HVO) or renewable diesel (RD). In
a decision-making unit (DMU) in DEA and each sustainability indicator turn, each of the resulting biofuels can be blended differently to produce
is classified as an input or an output to the DMU (further details in the final commercial fuel. In the case of bioethanol, two blends with
section 2.2). This analysis allows to classify biofuels as efficient (i.e., gasoline are considered: E10, using 10% ethanol; and E85, using 85%
showing the best performance among alternatives) or inefficient (i.e., ethanol. These blends are used in spark ignition (SI) engines. In the case
inferior to the best-observed practices). For the latter, DEA also provides of biodiesels, a blend consisting of 20% biodiesel-80% conventional
improvement targets that, if attained, would make inefficient biofuels diesel is assumed for use in compression ignition direct injection (CIDI)
efficient. engines. Renewable diesel is obtained from two main processes, Super
cetane (i.e., labelled here as RDI) and fluid catalytic cracker technology
(i.e., named RDII) [47] and both are also used in CIDI engines but, in this
case, as pure fuels as they are not blended. In all cases, engines are
assumed to belong to a light vehicle carrying one single passenger.
Throughout this work, we use the term first-generation for biofuels
derived from edible agricultural feedstock such as grain or sugars (e.g.,
corn, sorghum), the term cellulosic for biofuels produced from ligno­
cellulosic biomass (e.g., willow, poplar) and the term bio-oil for biofuels
obtained from oleaginous plants (e.g., soy, palm) [48–49].
Overall, 50 ethanol-based fuel routes are considered as follows. On
the one hand, we study a total of 19 routes to produce ethanol though
the fermentation of sugars from dedicated energy crops. Among these,
10 routes comprise the direct fermentation of biomass sugars. This is the
case of five different production processes for corn, two processes
combining usage of corn stover and corn, one process for sweet sor­
ghum, one process for grain sorghum and one process for sugar cane.
The remaining nine routes rely each on a different cellulosic carbon
source and differ from the previous 10 in that the latter require a pre­
vious step consisting of an acid hydrolysis of the lignocellulose to pro­
duce simple sugars before these can be fermented into alcohol. Finally,
six cellulosic carbon sources (i.e., six out of the nine cellulosic sources
used) are employed to produce ethanol by biomass gasification. Overall,
this yields a total of 25 routes to produce ethanol from biomass.
Considering that the final ethanol product can be blended with gasoline
in two different proportions (E10 and E85), this yields a total of 50
ethanol-based biofuel production routes.
Additional 22 routes based on bio-oil are also considered. Biodiesel
can be produced from 8 additional carbon sources (in blue in Fig. 2)
through transesterification of lipids. The resulting product is blended
with diesel to form BD20. On the other hand, seven carbon sources can
be used to produce RD using two main processes, Super cetane (i.e., RDI)
and fluid catalytic cracker technology (i.e., RDII) [47]. These results in a
total of 14 additional biofuels routes (note that RD is used as a stand­
Fig. 1. Flowchart for the methodology proposed. alone fuel, i.e., not blended with diesel).

3
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

Fig. 2. Block-diagram providing the different alternatives considered as carbon source, fuel production process, blend and type of combustion engine. Carbon
sources are depicted with a different color depending on whether they are first-generation (e.g., corn), cellulosic (e.g., poplar) or bio-oils (e.g., palm). SI: Spark
Ignition; CIDI: Compression Ignition Direct Injection.

Overall, a total of 72 different biofuel routes are obtained: 50 for the be travelled with the biofuel; the environmental dimension is evaluated
case of ethanol that will be used in SI engines and 22 biofuels that will be through eight life-cycle impacts; and the performance in the social
used in CIDI engines. dimension is based on water use and land occupation since shortage of
For each of these 72 biofuel routes, 12 performance metrics covering these resources can trigger social conflicts [50]. These performance
the three sustainability dimensions are considered as follows. The eco­ metrics are assessed from a cradle-to-wheel perspective, thus accounting
nomic dimension is assessed through the cost and the distance that can for all the resources and emissions occurring from cradle-to-tank (i.e.,

4
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

during the farming stage, biomass transportation and conversion to fuel) computed by adding the corresponding impacts from the different life-
and from tank-to-wheel (i.e., combustion of the fuel in the vehicle cycle stages p (ImpactStage
u,p ), as shown in Eq. (3). For cradle-to-tank
engine). stages, impacts are computed as the product between the amount of
The starting point for the calculation of the 12 indicators are the data input f required in the stage (Inputf ,p ) and the life-cycle impact in
collected from the GREET 2020 database [47], which provides infor­ midpoint category u of producing a unit of input f (Ecovectoru,f ) (see Eq.
mation on the material and energy flows f (e.g., chemical reagents or
(4)). Ecovectors are obtained from Ecoinvent v3.7.1 database [56],
electricity) required in each production stage p (i.e., cultivation, biomass
using the activities reported in Table S18. For the combustion stage,
transportation or biomass to fuel conversion) involved in the trans­
direct emissions for different pollutants e (Emissione,p ), also provided by
formation of any carbon source into the corresponding fuel. These input
GREET, are converted into the corresponding impacts u by applying
flows, denoted here by Input Rawf,p and reported in Tables S1-S9 in the ReCiPe impact factors (IFu,e , Eq. (5)) [47,57–58]. The results of this
Supplementary Material, are obtained for one liter of biofuel since this is calculation (i.e., impacts for the combustion stage) are reported in
the calculation basis selected in this contribution. Arguably, only a Table S19 in the Supplementary Material.
certain share of these inputs should be attributed to the requirements of ∑
biofuels themselves since other by-products are also obtained during the Impactu = Impact Stage
u,p ∀u (3)
biofuel production process (e.g., corn-oil, electricity or glycerin). Ac­ p

cording to ISO 14040:2006 [51], allocation is the tool for partitioning ∑

input and output flows of a system between the product under study (e. Impact Stage
u,p = Inputf ,p Ecovectoru,f ∀u,p = {farm,transport,conversion}
g., biodiesel) and other by-products (e.g., glycerin). Among the different
f

allocation methods available in the literature, here an economic allo­ (4)

cation is used, as this is the baseline method for most LCA allocation ∑
Impact Stage
= Emissione,p IF u,e ∀u, p = {combustion} (5)
situations [52]. The economic allocation generates an allocation factor u,p
e
(denoted here by AF , with 0 ≤ AF ≤ 1) based on the quantity and the
economic value of the biofuel itself and the corresponding by-products. In the case of the GWP indicator, one final adjustment is required to
This allows to compute the input f of stage p attributed to biofuel pro­ account for the fact that, in the life cycle, emissions from biogenic car­
duction (Inputf,p ) as a certain share (AF) of the total input for the whole bon do not increase the total amount of carbon in the bio­
sphere–atmosphere system. This is because biogenic carbon originates
process (Input Raw
f,p ):
precisely by fixation of carbon from the CO2 absorbed during photo­
Inputf ,p = Input Raw
f ,p AF ∀f , p = {farm, transport, conversion} (1) synthesis, thus resulting in a net-zero cycle. Therefore, the CO2 absorbed
during biomass growth needs to be deducted from the total GWP ob­
Note that allocation only affects cradle-to-tank stages (i.e., farming, tained at the end of the fuel life-cycle to obtain the net balance of GHGs.
biomass transportation and conversion to fuel), since emissions incurred To this end, the carbon content of the fuel is expressed in terms of carbon
during combustion (i.e., tank-to-wheel) are solely attributable to the dioxide and discounted from the GWP obtained with Eq. (3). Carbon
biofuel itself. The interested reader is referred to section 1.3 in the contents considered for the different fuels are 54.4 %w/w, 76.2 %w/w,
Supplementary Material for further details on the economic allocation. 84.9 %w/w for ethanol, biodiesel, and renewable diesel, respectively
With allocated input values available, these are next used to compute [58], while fuel densities are provided in Table S21 in the Supplemen­
the different sustainability indicators. Calculations described next are tary Material.
repeated for every biofuel j, although subscript j has been dropped from Finally, social indicators (i.e., land occupation and water use) are
equations and variables for simplicity. Since the GREET database builds obtained as follows. The land occupation indicator (Land , in [ha⋅yr /l
upon United States (US) data, any complementary data used (e.g., power fuel]) accounts for the annual land used for growing the necessary crops,
generation matrix) will also be based on the US to preserve data ho­ neglecting land requirements for chemicals and energy production as
mogeneity, as suggested by Dyson. [53] . these are expected to be significantly smaller than for harvesting
First, the cost indicator (Cost) is computed as the summation of the biomass [56]. This indicator is computed from the amount of carbon
product between the amount of input f required in every stage p (cradle- feedstock needed to produce 1 L of biobased fuel (Crop , e.g., tons of
to-wheel) to produce one liter of the biofuel (Input f,p ) and the corre­ poplar needed for 1 L of ethanol) and the annual yield of the corre­
sponding unitary costs (UCf ), as shown in Eq. (2). Unitary costs are sponding crop (YieldCrops , in tons of crop per hectare and year) (see Eq.
obtained from different sources, as reported in Table S17. (6)). Data for Crop are obtained from GREET while the data for YieldCrops
∑ are reported in Table S25-S26, together with the corresponding data
Cost = Inputf , p UCf (2)
f , p∕
=farming sources.

Note that the costs of farming inputs and extraction are neglected as Land =
Crop
(6)
they are assumed to be included in the cost of the vegetable oil Yield Crops
feedstock. In the case of the water use indicator (Water), two contributions are
The other economic indicator, i.e., the distance that can be travelled considered: the life-cycle water consumption for chemicals and energy
by burning the biofuel in the corresponding engine, is directly retrieved production from cradle-to-wheel (WaterInputs ) plus the amount of water
from GREET, as this information is readily available in the database. consumed for growing the corresponding crops (WaterCrops ) (Eq. (7)).
As aforementioned, the environmental performance of the biofuel The former contribution is obtained by multiplying the amount of inputs
alternatives is quantified based on eight life-cycle impacts. Precisely, we (Input f,p ) by the life-cycle water consumption of producing one unit of
use eight midpoint indicators of the ReCiPe approach following a hier­
such input (WCInputs , as retrieved from Ecoinvent for activities in
archical perspective and assuming allocation at the point of substitution. f

We choose midpoint over endpoint indicators as the former are Table S18)(Eq. (8)). On the other hand, the amount of water required to
considered less uncertain and, therefore, more reliable than the latter grow the corresponding crop can be calculated from Eq. (9), where land
[54]. The indicators selected cover impacts related to human health (i. requirements are multiplied by the annual water consumption per
e., GWP, fine particulate matter formation, human ecotoxicity, photo­ hectare for the corresponding crop (LWCCrops , in mm of water per square
chemical oxidant formation potential) and ecosystems (i.e., terrestrial meter and year, see Tables S23-S24).
acidification, terrestrial ecotoxicity, freshwater eutrophication and Water = Water Inputs
+ Water Crops
(7)
freshwater ecotoxicity) [55]. The total impact in midpoint category u is

5
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

∑ 2.2. DEA fundamentals

Water Inputs
= Inputf ,p WCf Inputs (8)
f ,p
DEA [31] is a data-oriented approach for evaluating the relative ef­
Water Crops
= Land⋅LWC Crops
(9) ficiency of a set of n similar entities called decision-making units (DMUs,
indexed by j), which convert multiple inputs (i = 1,…,m) into multiple
Note that values for LWCCrops and YieldCrops correspond to agricultural outputs (r = 1,…,k) [61]. Although DEA was originally devised to assess
land in conditions appropriate for the cultivation of each particular crop, the productivity efficiency of production units, where inputs and out­
with rainfall and artificial irrigation being both valid options to satisfy puts nomenclature was meaningful, later it has been widely used as a
water requirements. If the performance of biofuel routes were to be MCDM tool in any context. In the latter case, inputs and outputs can be
evaluated for specific geographical areas, where annual rainfall is any performance metric of interest, with the general agreement that
known, we suggest computing the water requirements based only on inputs are metrics one is willing to minimize while outputs are metrics
irrigation, as this can make a difference in the results obtained for the one seeks to maximize [61]. Some model variations also consider the
efficiency scores of routes based on certain crops (see Fig. S1 in Sup­ potential existence of the so-called undesirable outputs, which are out­
plementary Material for further details). However, some of the consid­ puts to the production process one might want to reduce, e.g., polluting
ered crops (e.g., Miscanthus, switchgrass, poplar, willow) might be emissions [47].
suitable for marginal land, thus avoiding competition with food at the In this contribution, each of the 72 biofuel route alternatives is
expense of probably lower yields and larger water and chemical modelled as a DMU whose relative performance is evaluated based on
requirements. the 12 sustainability indicators described in the previous section and
In the case of materials that are considered residues or by-products of classified here as either inputs or outputs (desirable or undesirable, see
other crops (e.g., corn stover and forest residue), economic allocation Fig. 3). Dyson suggested that an appropriate discriminatory power could
factors of 15% [59] and 38% [60] respectively were applied for water be achieved in DEA if the number of DMUs is at least 2⋅(m⋅k), where m⋅k
use. The land required is obtained by multiplying the yield of these is the product of the number of inputs times the number of outputs [53];
materials per hectare [t/ha] by the amount of feedstock needed to such a condition is satisfied in the present analysis (i.e., 72 > 2(3⋅9)).
produce one liter of fuel. For each DMU, DEA returns a performance score, also called effi­
The final values for the 12 indicators for the 72 biofuels will be ciency score, lying between 0 and 1. DMUs (i.e., biofuels) with a score of
referred to as the nominal values and are provided in Table S22 in the 1 are referred to as efficient and are linearly combined to form the
Supplementary Material and summarized here in Table 1, where bio­ efficient frontier. Meanwhile, DMUs with a score strictly lower than 1
fuels are grouped into five categories according to their production are considered inefficient and are projected onto the efficient frontier to
process. generate the so-called virtual DMUs. Virtual DMUs can be understood as
Note that we retrieved the data used to calculate indicator values efficient versions of the projected DMU and allow the identification of
from the same source for all biofuel routes to ensure a fair comparison the improvements that the inefficient DMUs should target to become
between them. The only exception is farming data (i.e. water re­ efficient.
quirements and land yield for the different crops), which were retrieved While these basic elements are common for all DEA approaches, a
from different sources, but always under the common assumption of plethora of model variations has been put forward to date with the aim
adopting the most suitable conditions for growing each particular crop. of better aligning model assumptions with the problem under study.
Similarly, different production routes use different material inputs Some of the modelling choices include the returns-to-scale (RTS), model
whose cost could not be retrieved from a single data source but were orientation or the way in which the efficiency score is evaluated. These
always determined by the corresponding commodity market. In addi­ choices are explained in more detail in the following paragraphs.
tion, an uncertainty assessment was carried out to ensure reliable results The RTS aims to reflect whether DMUs operate or not at the same
and conclusions despite any potential data variation stemming from the scale. The most common choices are the constant returns-to-scale (CRS),
occasional use of different sources or assumptions (see Section 2.5 for which assumes the ratio between inputs and outputs is constant
further details on this matter). regardless of the level of inputs, and the variable returns-to-scale (VRS),
We next describe how indicator values are used in DEA to benchmark assuming a change in the inputs will produce a different change in the
the sustainability performance of the different biofuels routes studied. output depending on the input level [62].
Model orientation defines the way inefficient DMUs are projected

Table 1
Statistics of the sustainability indicators considered for the 72 biofuel routes. Values are for 1 L of fuel. Acronyms are provided in the table footnote.
Fuel type BD20 E10 E85 RDI RDII

Parameter Median (min–max) Median (min–max) Median (min–max) Median (min–max) Median (min–max)
Cost [US$] 0.80 (0.74–1.12) 0.69 (0.68–0.74) 0.31 (0.21–0.75) 0.62 (0.41–1.17) 0.70 (0.46–1.30)
LO [m2] 1.14 (0.04–3.65) 0.15 (0.01–0.28) 1.30 (0.001–2.36) 5.69 (0.20–11.04) 6.07 (0.22–11.67)
Water required [m3] 0.54 (0.01–1.79) 0.16 (0.001–0.43) 1.32 (0.01–3.66) 2.09 (0.23–5.54) 3.46 (0.26–9.02)
GWP [kg CO2-Eq] 2.52 (2.50–2.63) 2.51 (2.35–2.70) 0.85 (0.71–2.51) 0.59 (0.51–1.10) 0.31 (0.23–0.92)
FWET [10− 2 kg 1,4-DCE] 0.96 (0.91–1.37) 0.56 (0.01–0.90) 1.23 (0.01–3.62) 0.80 (0.66–2.69) 0.86 (0.75–3.02)
FWEU [10− 4 kg P-Eq] 0.45 (0.40–2.03) 0.44 (0.01–0.67) 0.85 (0.01–2.49) 0.64 (0.44–7.29) 0.70 (0.45–8.66)
HT [kg 1,4-DCE] 0.14 (0.14–0.29) 0.11 (0.09–0.15) 0.23 (0.10–0.49) 0.11 (0.10–0.75) 0.12 (0.10–0.88)
PMFP [10− 3 kg PM10-Eq] 1.59 (1.53–2.33) 1.45 (0.002–1.78) 1.78 (0.001–3.21) 1.04 (0.74–4.10) 1.18 (0.86–4.84)
POFP [10− 3 kg NMVOC] 5.31 (5.09–6.32) 4.36 (0.005–5.09) 5.64 (0.004–8.67) 4.00 (3.02–8.53) 4.40 (3.27–9.39)
TA [10− 3 kg SO2-Eq] 4.40 (4.28–5.45) 4.30 (0.005–5.23) 4.34 (0.004–7.91) 1.82 (1.28–6.15) 2.10 (1.50–7.29)
TE [10− 3 kg 1,4-DCE] 2.58 (2.58–2.63) 0.67 (0.005–3.12) 4.58 (0.004–15.30) 2.45 (2.43–2.67) 2.47 (2.44–2.72)
Distance (km) 15.12 10.73 8.63 14.57 14.57

* BD20: Diesel fuel with up to 20 %v/v FAME content; E10: Gasoline fuel with up to 10 %v/v bioethanol content; E85: Gasoline fuel with up to 85 %v/v bioethanol
content; RDI: Renewable Diesel Production Based on SuperCetane; RDII: Renewable Diesel Production Based on fluid catalytic cracker technology; LO: land occu­
pation; Water: water used in farming plus water depletion produced during chemicals manufacturing; GWP: global warming potential; FWEU: freshwater eutrophi­
cation; FWET: freshwater ecotoxicity; HT: human ecotoxicity; PMFP: fine particulate matter formation; POFP: photochemical oxidant formation potential; TA:
Terrestrial acidification; TE: terrestrial ecotoxicity.

6
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

Fig. 3. Inputs and (desirable and undesirable) outputs considered for each biofuel (DMU). Units for each indicator are provided between brackets. LO: land
occupation; Water: water used in farming plus water depletion; GWP: global warming potential; FWEU: freshwater eutrophication; FWET: freshwater ecotoxicity; HT:
human ecotoxicity; PMFP: fine particulate matter formation; POFP: photochemical oxidant formation potential; TA: Terrestrial acidification; TE: terrestrial
ecotoxicity.

onto the efficient frontier. In this regard, the most conventional alter­ original proportionality [64].
natives are input-orientated, which attempts to minimize inputs while Some of these concepts are illustrated in the following example
securing a certain level of output; and output-oriented models, where (Fig. 4), where DEA is used to assess the efficiency of four DMUs (A, B, C,
the opposite holds (i.e., outputs are expanded while maintaining the and D) against each other in a case considering two inputs and one
inputs at original levels). Non-oriented models, in which inputs and output. If the output is dummy (e.g., all DMUs show the same perfor­
outputs are allowed to change simultaneously, are also widely used. mance in this output), DMUs can be represented in a two-dimensional
Finally, models are commonly grouped in two categories depending cartesian plot as in Fig. 4. In this example, DEA would identify DMUs
on whether the efficiency measure is radial or non-radial. Radial mea­ B, C, and D as efficient because there is no other DMU showing better
sures belong to the Debreu–Farrell measures and force changes in all the performance, i.e., attaining lower inputs and/or higher output simul­
inputs (or all the outputs in an output-oriented model) to be propor­ taneously. Efficient DMUs form the so-called efficient frontier, which
tional [63]. In contrast, non-radial measures belong to the Par­ corresponds to segment C-B-D when a VRS is considered, as in this
eto–Koopmans measures [63] and allow inputs and outputs to vary in example. Then, the model would project inefficient DMU A onto the
any possible way so that inefficient DMUs attain the efficient frontier. efficient frontier to obtain the efficiency score and improvement targets
Examples of non-radial measures are Range Adjusted Measure, Russell for this unit. If the efficiency measure is radial and the model is input-
Measure, Additive Model and Slack Based Measure (SBM) models. Note oriented, then input 1 and input 2 would be decreased proportionally,
that not all possible model orientations can be used with any efficiency yielding virtual DMU A’. In contrast, using a non-radial SBM model, the
measure as these two choices are not always independent from each two inputs would be allowed to change non-proportionally. Indeed,
other. For instance, applying a non-radial model in cases where there is a Fig. 4 demonstrates this idea of non-proportionally wherein any pro­
linear dependence between inputs and outputs causes a loss of the jection in the quadrant A-A1-A2, as defined by slacks S−1 and S−2 (distance
between the assessed and the virtual DMU) would be permitted in an
SBM model. In this latter case, the virtual DMU of A could lie anywhere
in the segment A1-B-A2, provided that inputs are not allowed to worsen.
DEA models based on non-radial measures are agreed to have a
greater capacity to discriminate the DMUs under evaluation and yield a
lower number of efficient units [65], therefore, being the preferred
choice in environmental assessment. Among non-radial approaches, the
most widely used one is the SBM model proposed by Tone [66], which,
in its original formulation, treats undesirable outputs as inputs [67]. In
our case, this translates into DMUs having 11 inputs (three original plus
the eight undesirable outputs) and one output (the original desirable
output). Previous studies have used this model to investigate issues
related to water use relation with total factor productivity [68–69], the
relation between energy use efficiency and either GDP [70] or economy
development [71], the potential emission reductions and marginal
abatement costs of energy-related CO2 emissions [72], the measurement
of environmental efficiency of transportation sector based on CO2
emissions [73], and the relation between social fixed assets investment
and GDP in the industry with SO2 emissions [74]. The mathematical
model is described in detail in the next section.

2.3. SBM non-oriented model

Fig. 4. Difference in projection onto efficient frontier by the radial and the non-
radial SBM models. The SBM efficiency model proposed by Tone [75] is non-radial and

7
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

computes the efficiency score based on the excess of inputs (s−i , which execute standard DEA models under the assumption that the DMU
henceforth includes the original undesirable outputs) and the shortage assessed is excluded from the efficient frontier. In other words, in super-
of outputs (s+ r ). There are three variations of this model, i.e., input- efficiency DEA models, the virtual DMU must be constructed using the
oriented, output-oriented, and non-oriented, with the latter model remaining DMUs only [78]. For the case of the SBM model m.1, one can
referring to both input- and output-oriented. Working with the latter resort to the super-SBM model proposed by Tone [66] for evaluating
model prevents the need to decide between considering strong or weak efficient DMUs (ρ* = 1, S− = 0, S+ = 0) . The model formulation is as
disposability of environmental impacts, an assumption often made to follows:
deal with undesirable outputs [76]. Hence, without loss of generality, 1
∑m xi
we use the non-oriented SBM model dealing with undesirable outputs as δ* = min 1∑k
m i=1 xi0
(m.3)
inputs for evaluating DMUs.
yr
k r=1 yr0
∑ s−
1 − m1 mi=1 xii0 ∑
n
ρ* = min ∑ + (m.1) s.t. x ≥ λj xj
1 + 1k kr=1 ysr0r
j=1,∕
=0

∑
n
∑
n
s.t. λj xij + s−i = xi0 i = 1, 2, ⋯, m y≤ λj yj
j=1 j=1,∕
=0

∑
n
x ≥ x0 , y ≤ y0 , λ ≥ 0
λj yrj − s+
r = yr0 r = 1, 2, ⋯, k
j=1
The previous SBM model (m.1) and the super SBM model (m.3)
− + selected for this work assume constant returns to scale (CRS), although
s ≥ 0, s ≥ 0
these models could be extended to variable returns to scale (VRS) by
∑n ∑n
λj ≥ 0 j = 1, 2, ⋯, n adding equations j=1 λj = 1 and =0 λj = 1 in models (m.1) and
j=1,∕
(m.3), respectively.
In this model, ρ is the SBM-efficiency score, xij is the value of input i
of DMU j, yrj is the value of output r of DMU j, and xio and yro are the
2.5. Dealing with data uncertainty in DEA
values of input i and output r of the DMU o under evaluation. In turn, s−i
and s+r are the input and output slacks, providing the distance from the
Regardless of the efforts invested in collecting data with the highest
DMU assessed to the efficient frontier. Slack variables in non-oriented
quality, DEA results might always be affected by data inaccuracies or
SBM models provide information regarding the degree of inefficiency
simplifications, which could lead to spurious efficiency scores and
attained by each input and output individually [77].
rankings. To overcome this, we consider uncertainty in our data in an
This fractional programming problem can be transformed into a
attempt to obtain more robust results and conclusions under different
linear programming problem using the Charnes–Cooper transformation
potential realizations of the uncertainty. Without loss of generality, we
as follows:
assume each indicator follows a uniform distribution spanning ±10% of
1 ∑m
Si− its nominal value. These distributions are then discretized into 100
τ* = mint − (m.2)
m i=1 xi0 different scenarios for each DMU using Monte Carlos sampling. Finally,
following the approach of Ewertowska et al. [79] 100 independent DEAs
1∑ k
Sr+ (i.e., one for each scenario) are solved, in addition to the nominal sce­
s.t. 1 = t + nario, yielding a distribution of efficiency scores for each DMU (rather
k r=1 yr0
than a single value).
∑
n
Λj Xij + Si− = xi0 t i = 1, 2, ⋯, m 3. Results
j=1

The non-oriented SBM efficiency model (m.2) and the non-oriented

∑
n
Λj Yrj − Sr+ = yr0 t r = 1, 2, ⋯, k SBM super-efficiency approach (m.3) were coded in GAMS v32.1.0
j=1 [80] and solved in an AMD Ryzen 5 4500U processor for each of the 72
DMUs in each scenario. Each instance took less than 1 s of CPU time to
S− ≥ 0, S+ ≥ 0, Λ ≥ 0, t > 0 be solved. The results obtained are described next, starting with the
efficiency and super-efficiency scores, then moving to the analysis of
Note that the optimal solution of model (m.2) (e.g., τ* , Λ* , t * , S− * , S+* )
inefficient alternatives and finally exploring different improvement
can be used to derive the optimal solution of model (m.1) using the
*
scenarios for the transportation sector. For the sake of simplicity, the
following relationships: ρ* = τ* , λ* = Λt* , S−
− *
*
= st* , S+* = s+*
t*
. discussion will focus on the results obtained for the nominal scenario,
except for cases where result distributions are explicitly mentioned.
2.4. Super-efficiency
3.1. Efficiency assessment
DEA evaluates the relative efficiency of DMUs but does not further
rank among the efficient units. This sometimes results in DEA providing Fig. 5 provides the combined results for the efficiency and super-
a long list of promising (efficient) alternatives, upon which decision- efficiency DEAs, with inefficient biofuels being represented based on
makers need to choose based on additional criteria. To provide more their efficiency score and efficient biofuels depicted based on their
accurate rankings without having to resort to additional considerations, super-efficiency score. Specifically, horizontal bars provide the effi­
the super-efficiency score has become an option to discriminate between ciency score in the nominal scenario, while overlapped boxplots provide
efficient DMUs. information on the distribution of the efficiency scores obtained in the
Super-efficiency models identify the best-performing DMUs by remaining 100 scenarios. Results reveal that 48% of the 72 biofuels
assigning an efficiency score greater than one, thus facilitating com­ routes analyzed are efficient in the nominal case (Fig. 5a), meaning that
parison with rankings based on parametric methods [46]. These models there is no other biofuel showing superior performance in all the

8
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

Fig. 5. Efficiency scores for biofuels. (Super)efficiency scores for the 72 biofuels routes are provided as horizontal bars in subplot (a), with biofuels sorted in
decreasing order of efficiency and efficient biofuels depicted with a green label. Histograms at the bottom of the figure group results per type of biofuel (subplot (b))
or type of feedstock (subplot (c)). ETOH corn A: Dry mill corn without oil extraction; ETOH corn B: Dry mill corn with oil extraction; ETOH corn C: Wet milling corn;
ETOH corn D: combined dry and wet milling corn; ETOH corn/stover A: integrated corn/stover ethanol (associated with corn); ETOH corn/stover B: integrated corn/
stover ethanol (associated with stover); ETOH corn E: Gen dry milling corn with oil extraction; ETOH sweet sorghum A: Conventional; ETOH sweet sorghum B:
Integrated.*G: Ethanol produced by gasification. BD20: Diesel fuel with up to 20 %v/v FAME content; E10: Gasoline fuel with up to 10 %v/v bioethanol content; E85:
Gasoline fuel with up to 85 %v/v bioethanol content; RDI: Renewable Diesel Production Based on SuperCetane; RDII: Renewable Diesel Production Based on fluid
catalytic cracker technology. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

9
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

sustainability indicators simultaneously. This implies there is a pool of [83–85]. Inspection of data (see Table 1) reveals that fuels used in SI
35 biofuels from which policy-makers can select the most suitable al­ engines have 15% lower median prices than the fuels used in CIDI en­
ternatives to promote according to the regional context (e.g., land gines, 0.67$/liter vs. 0.78$/liter. This difference is reversed when the
availability, farmer preferences or the most abundant type of vehicle -SI comparison considers fuel consumption per km, where CIDI-type fuels
vs. CIDI-). achieve a lower price (0.054$/km vs. 0.07$/km). Therefore, while
The highest efficiency score, standing at 1.61, is achieved by the renewable diesel generates greater environmental impacts per liter of
blend using 85% of ethanol from municipal waste, owning to different fuel burned (see Table S19 and S20), this is offset by the achievement of
factors. On the one hand, low cost, water and land occupation re­ longer distances travelled, which ultimately translate into lower impacts
quirements are allocated to MSW compared to other feedstocks (e.g., per km (i.e., lower inputs for the same level of output).
0.01 m3 of water/liter of E85 from MSW, compared to 0.76 m3 of water/ Overall, these results call for encouraging the use of renewable diesel
liter of E85 from dry mill corn without oil extraction). On the other over traditional biodiesel or bioethanol owing to their lower GWP in the
hand, this is also attributable to the production process itself, which life cycle, their lower fuel consumption rate, and their lower exhaust
takes advantage of low-cost fermentable sugar sources. In the case of particle emissions per km [86]. In cases where biodiesel is still to be
MSW, the energy demand of the process is self-satisfied by using either a used, cellulosic carbon sources are preferred over first-generation
fraction of the biomass feedstock or the residues from the fermented biomass; this might also avoid concerns about competition with food
biomass, also exporting any surplus of energy that might be produced. by growing crops in marginal land. In addition, the processes for con­
This makes the fossil carbon emissions, as well as the impacts associated verting cellulosic biomass into bioethanol typically devote part of the
with energy generation and transportation, lower for MSW than for any biomass feedstock to the cogeneration of heat and electricity for self-
first-generation biomass. Indeed, the production of 1 L of ethanol from consumption. This not only reduces the input requirements allocated
first-generation biomass (i.e., fermentation of simple sugars) emits on to the biofuel, but also lowers the dependence on the domestic elec­
average 0.47 kg CO2 eq, while the production of 1 L of ethanol from tricity mix by satisfying part of the energy demand of the process
cellulosic materials and an acid hydrolysis process emits only 0.21 kg through renewable sources (~74% on average) [47]. Cellulosic mate­
CO2 eq. rials are currently becoming more competitive, achieving better per­
Interestingly, the blend using 10% ethanol from MSW shows a formance and lower cost thanks to advances in the production of
modest efficiency in the nominal scenario (1.00) and even has an 11% enzymes for the degradation of lignocellulosic materials into simple
chance of being inefficient. This inferior result compared to the E85 fermentable sugars (e.g., pentoses, hexoses) [87]. However, replacing
blend stems from the increased amount of poor-performing gasoline the total diesel consumption in Europe (i.e., 287Mtoe [88]) with
present in the blend. Still E10 from MSW can achieve efficiencies as high renewable diesel from canola would require exploiting 90% of the total
as 1.11 in some scenarios; this would place it as the eighth fuel if sorted agricultural land available in the region (1.15 million km2 [89]), clearly
according to the maximum efficiency score displayed in any scenario. an unrealistic scenario.
Despite the promising results of biofuels based on MSW, the avail­ Inefficient units are mostly based on corn and sorghum grains, also
ability of waste suitable for biofuel production could limit the part of first-generation ethanol blends. Their low performance is due to
displacement of fossil fuels with these alternatives. As an example, 0.23 different factors. On the one hand, the feedstock costs are higher for
kg of dry MSW is generated per day and person in Europe. If all this these fuels than for lignocellulosic materials (e.g., 130$/t or 350$/t of
waste were used to produce E85, 0.08 L would be obtained, yet this corn and sorghum, respectively, compared to 58$/t for Miscanthus, as an
would only cover 1.8% of the daily per capita demand for fuel in the example of lignocellulosic material) [90]. Besides, environmental im­
region (i.e., 4.4 L/day person) [81–82]. pacts generated during the farming stage of corn and sorghum are more
We next turn our attention to the lowest efficiency score in the significant owing to the higher use of machinery, transportation, pesti­
nominal scenario (0.26), which corresponds to ethanol from corn (i.e., cides and fertilizers (e.g., 194 g of fertilizer per liter of corn-based
E85 from combined dry and wet milling corn). This can be explained by ethanol, compared to 10 g of fertilizer per liter of willow-based
its high resource requirements and low mileage achieved per liter of ethanol, see Tables S1 and S5).
biofuel (i.e., 8.63 km compared to 14.57 km in the case of any renewable One aspect that stands out is the low efficiency of fuels based on
diesel). Jatropha compared to soybeans, even though the former has a higher oil
Comparing the five different types of fuels studied (i.e., E10, E85, content, lower water requirements, and lower land occupation (see
BD20, RDI, RDII), it is observed that there is at least one efficient biofuel Table S28). This might be due to the three times higher energy
for each of them in the nominal scenario (Fig. 5b). This does not mean requirement for the farming stage per kg of feedstock compared to
that all fuel types performed equally well: whilst almost all the BD20, soybeans.
RDI and RDII fuels are found efficient, only 30% of ethanol-based fuels Inefficient biofuels based on bio-oil correspond to those coming from
(15 out of 50) achieve the efficient status (Fig. 5a). This indicates that corn and algae sources. This is not only due to the carbon source but
the fuel type alone is not enough to draw strong conclusions, and that rather to the need to mix these fuels with fossil diesel. Indeed, corn and
the carbon source should also be explored. algae are efficient when they are used to produce a fuel based on 100%
To this end, we classify biofuels into three groups depending on their renewable carbon (RDI and RDII), allowing for the reduction of carbon
carbon source: (i) bio-oils, consisting of animal fat and vegetable oil; (ii) emissions and other environmental impacts associated.
cellulosic material, i.e., those made of lignocellulosic biomass; and (iii) It is also observed that data uncertainty has a marginal role in
first-generation sources, including sugars and starch (Fig. 5c). Again, we shaping efficiency scores, at least to the extent of affecting the trends
find examples of efficient biofuels for any type of carbon source, yet observed. Most of the biofuels are efficient or inefficient in all the sce­
some patterns can still be observed. Bio-oil-based fuels (e.g., BD20, RDI, narios and the nominal case, with only five biofuel routes changing
RDII) show, on average, the highest efficiency scores, standing at 1.03, depending on the realization of the uncertainty. These are E85 Mis­
compared to 0.91 for biofuels based on lignocellulosic biomass and 0.56 canthus (with a 74% chance of being efficient), E10 sugar cane (90%
for those based on first-generation biomass. This also translates into a chance), E10 willow (90% chance), E10 poplar (60% chance) and E10
larger share of bio-oil-based fuels deemed efficient: 90%, compared to MSW (89% chance). Among them, only the aforementioned E10 from
46% in the case of biofuels from lignocellulosic feedstock and 59% for poplar and E85 from Miscanthus show their performance clearly affected
those based on first-generation biomass. These results are explained by (efficiency score between 0.76–1.00 for the former and 0.62–1.05 for the
the lower fuel consumption in CIDI engines compared to SI engines and, latter).
in the case of renewable diesel, by the possibility of using 100% bio- Given that biofuels are mainly considered a potential solution for the
based fuels (i.e., no blends) without affecting the engine performance climatic problem, and acknowledging that other environmental impacts

10
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

are also important, we next explore in detail the performance achieved (70%) and the second most important in GWP (25%) for algae-based
by some biofuels in terms of their GWP (Fig. 6a). It is observed that the renewable diesel, and therefore could be object of further research
combustion stage is the one that contributes the most to this impact aiming at improving the sustainability level of these biofuels.
category, being responsible for 80% of carbon emissions on average. The results obtained through the methodology applied in this
However, part of these emissions would come from biogenic carbon, contribution for the evaluation of biofuels are in agreement with those
which does not contribute towards the GWP because it does not increase found using other metrics such as RepSIM [92]. Despite differences
the total amount of carbon in the biosphere–atmosphere system in the between the two approaches exist, both methodologies combine eco­
life cycle. The share of emissions stemming from biogenic carbon de­ nomic, environmental and social indicators to perform a holistic sus­
pends on the fuel and the carbon source used, and can be as high as 93% tainability assessment, finding that the most sustainable alternatives
of combustion emissions for renewable diesel made from palm (i.e., 93% result from using low-value waste products as carbon sources (i. e.,
of total GWP without deducting biogenic CO2). In this particular MSW, tallow) and processes that involve cracking energy dense mole­
example, subtracting biogenic emissions would place the combustion cules and reforming them in the presence of hydrogen (e.g., HVO or
stage at 53% of the total GWP of the fuel. In contrast, biogenic emissions Fischer-Tropsh).
are low for biofuels based on blends with gasoline (e.g., E10 from wil­
low) or with conventional diesel (e.g., BD20 from palm), where even 3.2. Inefficiency assessment
after discounting biogenic emissions, the combustion stage still repre­
sents 82% and 81%, respectively, of the cradle-to-wheel GWP. Overall, Once identified, inefficient units are projected onto the efficient
this suggests that blends with low biofuel content (e.g., 20%) will have a frontier, and improvement targets are computed for their different sus­
limited benefit on climate change, which calls for policies promoting tainability indicators. These improvement targets are provided per DMU
pure biofuels or blends with higher biofuel shares. in Fig. 7 as the median percentual changes required with respect to the
A totally different picture emerges when other environmental im­ nominal values across the 100 different scenarios (i.e., decrease for in­
pacts are assessed. In the case of terrestrial ecotoxicity (Fig. 6b), the puts and undesirable outputs, and increase for outputs). In the interest of
production stage contributing the most towards the total impact de­ clarity, the information for the 37 inefficient DMUs is lumped, here, into
pends strongly on the fuel type and carbon source. For biofuels based on ten groups with similar carbon sources; the complete results are pro­
bio-oil, the combustion stage contributes 33% of total terrestrial eco­ vided in Table S32 in the Supplementary Material, while detailed results
toxicity, while for biofuels based on ethanol, combustion emissions for the nominal scenario are given in Table S31.
represent only 1% of the total terrestrial ecotoxicity. Indeed, the emis­ Most inefficient units need to achieve significant reductions in land
sions of polycyclic aromatic hydrocarbons (PAHs) released during bio­ occupation, water use, and terrestrial ecotoxicity to become efficient.
fuel combustion in a vehicle engine significantly affect the difference in This is especially evident in the case of E85 fuels due to the higher fuel
terrestrial ecotoxicity between the two fuel types. Note that, in the consumption of SI engines, causing, in turn, the increase of impacts from
absence of more specific data, we only differentiate PAH emissions be­ fuel production for the same mileage.
tween the two types of engines considered (CIDI vs SI), but not between Corn-based E10 requires improvements in all the inputs, with the
different blends used in the same engine. This assumption is based on the largest reductions observed in water use (96%), terrestrial ecotoxicity
observation that PAH emissions are mostly dictated by the engine (73%), land occupation (63%) and freshwater eutrophication (27%).
operating conditions [91]. This poor performance is mainly due to two factors. On the one hand,
In the case of third-generation biofuels, i.e., those using algae as corn farming is a very demanding process, requiring significant amounts
feedstock, the stage where algae is converted to renewable diesel is of land, water and energy compared to other crops (e.g., on average,
highly energy-intensive, mainly due to the oil extraction process. This 200% more energy than for cellulosic materials such as Miscanthus or
makes this stage the most important in terms of terrestrial ecotoxicity poplar). On the other hand, the conventional conversion process from

Fig. 6. Breakdown of the GWP (subplot (a)) and terrestrial ecotoxicity (subplot (b)) generated in the life cycle of selected biofuels. CMB: Combustion; FF: Fossil fuel
production; CM: Materials for the biofuel production; CE: Energy for biofuel production; FM: Farming materials; FE: Farming energy; BC: Biogenic Carbon.

11
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

Fig. 7. Median improvement targets for inefficient biofuels to become efficient considering 100 possible scenarios. E10 cellulosic: median improvement targets
across E10 switchgrass, E10 Miscanthus & E10 corn stover; E10 cellulosic sorghum: median improvement targets across E10 sweet sorghum B & E10 forage sorghum;
E10 corn: median improvement targets across E10 corn A-B-C-D-E & E10 corn/stover A-B; E85 cellulosic: median improvement targets across E85 switchgrass & E85
corn stover; E85 cellulosic sorghum: median improvement targets across E85 sweet sorghum A-B & E85 forage sorghum; E85 corn: median improvement targets
across E85 corn A-B-C-D-E & E85 corn/stover A-B; LO: land occupation; Water: water used in farming and water depletion; GWP: global warming potential; FWEU:
freshwater eutrophication; FWET: freshwater ecotoxicity; HT: human ecotoxicity; PMFP: fine particulate matter formation; POFP: photochemical oxidant formation
potential; TA: Terrestrial acidification; TE: terrestrial ecotoxicity.

first-generation biomass to E10 covers all its energy demand with the requirements and land occupation, respectively. The former impacts are
domestic energy matrix, being exposed to the cost and impacts of the mainly caused by the use and production of fertilizers, leaving rotational
country mix. This is a clear disadvantage compared to the conversion crops and organic fertilizers as the most promising option for their
process for cellulosic feedstocks, which satisfy part of their energy de­ abatement [94,95]. On the other hand, impacts on land occupation
mand by using a certain share of the biomass feedstock to generate heat might imply a yield increase that could be pursued by growing crops in
and electricity for self-consumption. best-endowed regions, i.e., on soils with adequate natural moisture
The higher ethanol content of corn-based E85 increases the resources available and non-winter climates [96]. Works by Castillo et al. [97] and
needed for crop farming, which in turn raises the improvements Zhang et al. [98]offer a suitability analysis of soils for different crops (e.
required for water use (94%), land occupation (95%) and terrestrial g., Miscanthus, switchgrass, poplar, Jatropha).
ecotoxicity (98%) to levels hardly achievable. Indeed, the first two in­ The production process for E10 and E85 based on cellulosic sorghum
dicators, clearly associated with farming of the grain, seem already (i.e., sweet and forage sorghum) is the same as for the other cellulosic
unattainable. Even if irrigation could be fully covered by rainfall in feedstocks, devoting part of the biomass to satisfy its own energy re­
certain regions, meeting the improvement target for land occupation quirements. Despite this, more demanding improvements are found
would entail almost doubling the yield: from the current 17.31 t/ha/yr when using sweet, and forage sorghum since growing these crops entails
(Table S25) up to a target yield of 34.27 t/ha/yr. On the other hand, higher costs and water requirements than other cellulosic feedstocks,
fertilizers and pesticides used during farming have a significant contri­ making them a poorer choice.
bution to terrestrial ecotoxicity (Fig. 6b), and again it is challenging to The inputs requiring the highest reductions for corn-based BD20 are
imagine that magnitude of reduction without affecting a crop yield that water use (72%) and land occupation (34%). Inefficiencies in these
should be further improved. In addition, E85 shows the lowest mileage categories are due to the low oil content of the corn grain (about 3–4%),
per liter of fuel used (i.e., higher fuel consumption per kilometer), which which results in larger feedstock requirements even after the economic
result in higher emissions from combustion and, therefore, higher re­ allocation (i.e., only 17% of the inputs for corn production are allocated
ductions in human ecotoxicity (74%), freshwater eutrophication (74%), to the biodiesel). An energy sector with a strong dependence on biomass
photochemical oxidant formation potential (57%), GWP (62%), fresh­ might alleviate global warming at the expense of imposing additional
water ecotoxicity (57%), fine particulate matter formation (57%) and burdens on land or freshwater use; however, the urgency to solve the
terrestrial acidification (46%). Furthermore, the mileage achieved climatic problem and the fact that land-system and freshwater use
should be improved by 2%; this could be pursued by using engines built planetary boundaries are not yet transgressed might fully justify the
to work with ethanol-blends (flex-fuels vehicles) or by incorporating transition [99,100].
turbochargers [93]. Algae-based BD20 shows high oil content (up to 35% in dry weight)
In the case of E10 from cellulosic feedstocks, reductions are required and low water requirements, which do not prevent it from needing
in most inputs, yet these are more modest compared to E10 based on important improvements in human toxicity (44%) and land occupation
first-generation biomass. The reason is that farming of cellulosic feed­ (35%). These two impacts are directly related to the energy needs for
stock requires less energy and materials than farming of first-generation algae drying and oil extraction, so advances in energy efficiency and the
biomass does, thus penalizing the contribution of biomass production oil extraction process, such as supercritical fluid extraction [101], could
for the latter. The most important improvements requested for E10 reduce the existing gap between the current and the target performance.
based on cellulosic biomass are reductions of 55% and 74% in water In addition, a reduction of 74% is required for terrestrial ecotoxicity.

12
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

Impacts in this category are generated mainly during diesel combustion 3.3. Enhancement scenarios
due to the generation of anthracene, fluoranthenes and pyrene [57].
This could be mitigated with the installation of Urea-based SCR systems, After identifying hotspots for inefficient biofuels in the previous
LNT Lean NOx Trap, or Exhaust Gas Recirculation that reduces the section, we next quantify the impact of adopting certain supply and
combustion temperature [102]. Finally, freshwater eutrophication demand-side measures to improve their sustainability performance. For
should be reduced by 61% for these biofuels. Although one could think the former, we focus on an improvement measure recurrently identified
this is the consequence of the water used for growing the algae, culti­ as promising in the previous section, considering only nominal values
vation is typically carried out in closed circuits where water is recircu­ for the indicators, namely the use of a renewable-based electricity ma­
lated. Consequently, 90% of the freshwater eutrophication stems from trix to supply energy for foreground processes. The mix proposed follows
the use of energy from the grid, which, in the case of the US-WECC the guidelines of the European Green Deal [12] for reduction of GWP
power mix, is dominated by coal (34%) and natural gas (18%). emissions and is based on 74% hydroelectric, 25% geothermal and 1%
Therefore, trying to meet this target implies either generating the wind. To complement the demand-side measures, we also analyze the
required energy internally using cleaner sources or relying on a more significance of adopting different demand-side measures, represented
sustainable energy mix. This latter option is explored in more detail in here by the use of different vehicles and passenger loads. Four scenarios
the next section. are considered in this regard on top of the reference case discussed so far
(i.e., labelled as scenario LVO1): light vehicles at minimum capacity (i.
e., one passenger, LV1); light vehicles at maximum capacity (i.e., five

Fig. 8. Changes achieved in inputs (i.e., costs, land occupation and water use) and selected undesirable outputs (i.e., global warming potential, freshwater eutro­
phication and fine particulate matter formation) by replacing the current mix (i.e., US-WECC) and vehicle type/usage for three fuels obtained from different types of
carbon sources: bio-oil, cellulose and first-generation feedstock. LVO1: base case without modifications (1 passenger); LV1: modified base case (1 passenger); LV5:
modified base case (5 passengers); PB10: public transport (10 passengers); PB30: public transport (30 passengers).

13
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

passengers, LV5) and the use of public transport (i.e., a bus) at 30% of its biofuel production processes and five biofuel blends.
maximum capacity (i.e., ten passengers, PB10) and at maximum ca­ The biofuel alternative with the highest efficiency score was based on
pacity (i.e., 30 passengers, scenario PB30). All these scenarios adopt the MSW, which suggests that these should be prioritized over other carbon
sustainable mix. sources. Fuels from natural oils also show a promising performance,
The consequences of adopting such measures and scenarios are with 20 of the 22 units analyzed deemed efficient. Among the remaining
calculated for a subset of six of the 12 performance indicators considered carbon sources, results agree with the recent trend of promoting the use
so far (i.e., cost, land occupation, water use, global warming potential, of cellulosic material for ethanol production. In terms of fuel type, our
freshwater eutrophication and fine particulate matter formation) for results suggest that policies should favor the widespread adoption of
three different biofuels (i.e., RDI based on algae, E85 based on corn and renewable diesel over traditional ethanol or biodiesel, since the former
E85 based on corn stover). The results obtained are shown on a per- achieved the best performance thanks to a higher fuel economy and a
capita basis in Fig. 8, where a comparison of scenarios LV1 and LVO1 higher biogenic carbon content in the fuel. The fuel type, however, was
allows us to assess the impact of the ceteris paribus change of the elec­ not found as impactful as the carbon source in achieving high efficiency
tricity source. scores.
Modifying the electricity matrix (i.e., comparison of scenario LV1 To complement policies for regulating biofuel supply, we also
with LVO1) allows algae-based RDI to achieve important reductions in explored the effectiveness of demand-side measures for the trans­
FWEU (82%), PMFP (52%) and GWP (44%). This change can be portation sector. We found that adopting responsible practices for
explained by the high electricity requirements of algae-based biofuel vehicle use could bring even more benefits than improving the biofuel
production and would suffice to attain the targets suggested by DEA for production processes or using cleaner energy sources.
some inefficient biofuels such as algae BD20 (e.g., 61% FWEU, 18% Finally, our analysis also provided targets for the improvement of
PMFP). In contrast, improvements in these three indicators for E85 inefficient biofuels that, if attained, would make them efficient. In this
biofuels are inexistent. This is because their production process requires regard, reductions in land occupation and water use, although highly
thermal energy rather than electricity. Although not explored here, the relevant and often identified as key in our results, might not be possible
implementation of heat pumps for waste heat recovery [103] could help to achieve depending on the type of crop and region. A case-by-case
these fuels to meet their improvement targets. Similarly, the remaining analysis is necessary for the farming stage to avoid the transportation
three indicators (i.e., cost, land occupation and water use) are barely of biomass over long distances and ensure no risks are imposed on food
affected by the change of the electricity mix, which suggests that other security. Indeed, the most appropriate feedstock might depend on the
measures would need to be pursued before improvement targets can be region of interest.
attained. For the case of cost reduction, governments can play a key role Promoting biological carbon sources today as an interim solution for
by providing economic incentives for biofuel production or discounting the transportation sector might proof useful even if the future is finally
certain taxes for the production or sale of biofuels. This would help dominated by electric vehicles since the infrastructure created for
alleviate the economic burden of some alternatives that can be key in the growing and transporting the biomass today could still be exploited by
achievement of environmental targets, that are becoming more bioenergy plants tomorrow. If combined with carbon capture and stor­
demanding. Meanwhile, we note that adopting the latest standards in age, these plants will remove carbon dioxide from the atmosphere, a
farming practices is not the only way to pursue improvements in land strategy deemed essential for meeting net-zero targets. In this context,
occupation and water use: increasing the efficiency of processes down­ multi-criteria approaches, such as the one presented in this contribution,
stream the supply chain (i.e., hydrolysis of lignocellulosic materials offer a powerful framework to perform holistic assessments with the
[104] or the fermentation process [105] will ultimately result in a lower capacity to minimize burden-shifting episodes and aid policy-makers in
demand for biomass feedstock and, therefore, lesser impacts from the development of better-informed policies.
farming.
More optimistic improvements are observed in all the cases when CRediT authorship contribution statement
supply and demand-side measures are combined (i.e., comparison of
scenario PB30 with LVO1). Revisiting the case of algae-based RDI, re­ Richard Cabrera-Jiménez: Conceptualization, Methodology, Soft­
ductions in FWEU, PMFP and GWP reach values as high as 97%, 93% ware, Writing – original draft, Visualization. Josep M. Mateo-Sanz:
and 92%; on average, 68% higher than in scenario LV1. Similar patterns Methodology, Writing – review & editing. Jordi Gavaldà: Validation,
are also observed for the rest of the biofuels and indicators, which, in Writing – review & editing. Laureano Jiménez: Resources, Writing –
this case, achieve improvements between 86% and 90%. These would review & editing, Supervision, Project administration. Carlos Pozo:
allow meeting the improvement targets requested by DEA for all the Conceptualization, Methodology, Validation, Writing – review & edit­
indicators in the case of E85 from corn stover and almost all the in­ ing, Supervision.
dicators except for land occupation, water use and terrestrial ecotoxicity
in the case of corn-based E85. Declaration of Competing Interest
These results highlight the importance of demand-side measures,
often overlooked in biofuel studies [106,107], since adopting a The authors declare that they have no known competing financial
responsible behavior can be, at least, as impactful as shifting to cleaner interests or personal relationships that could have appeared to influence
energy sources. Indeed, the greater the number of passengers in a certain the work reported in this paper.
vehicle, the greater the improvement in the performance of biofuels. The
only exception to this rule is the use of public transport at 30% of its Acknowledgements
capacity, which results in a worse alternative than a light vehicle with
five passengers. The authors would like to acknowledge financial support from the
Spanish Ministry of Economy and Competitiveness RTI2018-093849-
4. Conclusion BC33 and and Universitat Rovira i Virgili for the Marti-Franques
Research Grant (2019PMF-PIPF-82). This work was partially funded
In an effort to identify patterns that can aid in the development of by the Ministerio de Ciencia, Innovación y Universidades - Agencia
effective policies ensuring the sustainability transition in the trans­ Estatal de Investigación (AEI) (RED2018-102431-T). Carlos Pozo is a
portation sector, we combined LCA with DEA to assess the performance Serra Húnter fellow.
of 72 biofuel routes through the lens of sustainability. The different al­
ternatives result from the combination of 19 biological feedstocks, four

14
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

Appendix A. Supplementary material [29] Murrant D, Radcliffe J. Assessing energy storage technology options using a
multi-criteria decision analysis-based framework. Appl Energy 2018;231:
788–802. https://doi.org/10.1016/j.apenergy.2018.09.170.
Supplementary data to this article can be found online at https://doi. [30] Fernández D, Pozo C, Folgado R, Jiménez L, Guillén-gosálbez G. Productivity and
org/10.1016/j.apenergy.2021.118201. energy efficiency assessment of existing industrial gases facilities via data
envelopment analysis and the Malmquist index. Appl Energy 2018;212:1563–77.
https://doi.org/10.1016/j.apenergy.2017.12.008.
References [31] Charnes A, Cooper WW, Rhodes E. Measuring the efficiency of decision making
units. Eur J Oper Res 1978;2(6):429–44. https://doi.org/10.1016/0377-2217(78)
[1] World Energy Outlook 2020 – Analysis – IEA; n.d. https://www.iea.org/report 90138-8.
s/world-energy-outlook-2020 (accessed December 19, 2020). [32] Rodríguez-Vallejo Daniel F, Galán-Martín Ángel, Guillén-Gosálbez Gonzalo,
[2] World Energy Balances – Analysis – IEA; n.d. https://www.iea.org/reports Chachuat Benoît. Data envelopment analysis approach to targeting in sustainable
/world-energy-balances-overview (accessed December 19, 2020). chemical process design: Application to liquid fuels. AIChE J 2019;65(7):e16480.
[3] Masson-Delmotte V, Zhai P, Pörtner H-O, Roberts D, Skea J, Calvo E, et al. https://doi.org/10.1002/aic.v65.710.1002/aic.16480.
Climate Change and Land An IPCC Special Report on climate change, [33] Ewertowska A, Galán-Martín A, Guillén-Gosálbez G, Gavaldá J, Jiménez L.
desertification, land degradation, sustainable land management, food security, Assessment of the environmental efficiency of the electricity mix of the top
and greenhouse gas fluxes in terrestrial ecosystems Head of TSU (Operations) IT/ European economies via data envelopment analysis. J Cleaner Prod 2016;116:
Web Manager Senior Administrator; 2019. 13–22. https://doi.org/10.1016/j.jclepro.2015.11.100.
[4] Eurostat. Energy, transport and environment statistics 2019 edition. Belgium; [34] González-garcía S, Iribarren D, Susmozas A, Dufour J, Murphy RJ. Life cycle
2019. https://ec.europa.eu/eurostat/documents/3217494/10165279/KS-DK-1 assessment of two alternative bioenergy systems involving Salix spp. Biomass:
9-001-EN-N.pdf. Bioethanol Prod Power Gen 2012;95:111–22. https://doi.org/10.1016/j.
[5] Perera F. Pollution from fossil-fuel combustion is the leading environmental apenergy.2012.02.022.
threat to global pediatric health and equity: Solutions exist. Int J Environ Res [35] Iribarren Diego, Vázquez-Rowe Ian, Moreira María Teresa, Feijoo Gumersindo.
Public Health 2018;15(1):16. https://doi.org/10.3390/ijerph15010016. Further potentials in the joint implementation of life cycle assessment and data
[6] IEA. Renewables 2019 – Analysis and forecast to 2024; 2019. envelopment analysis. Sci Total Environ 2010;408(22):5265–72. https://doi.org/
[7] Demirbas A. Biofuels sources, biofuel policy, biofuel economy and global biofuel 10.1016/j.scitotenv.2010.07.078.
projections. Energy Convers Manage 2008;49(8):2106–16. https://doi.org/ [36] Lozano Sebastián, Iribarren Diego, Moreira M Teresa, Feijoo Gumersindo. The
10.1016/j.enconman.2008.02.020. link between operational efficiency and environmental impacts. A joint
[8] Transportation and Air Quality U.S. Environmnetal Protection Agency. application of Life Cycle Assessment and Data Envelopment Analysis. Sci Total
Renewable Fuel Standard Program (RFS2) Regulatory Impact Analysis (EPA-420- Environ 2009;407(5):1744–54. https://doi.org/10.1016/j.
R-10-006); 2010. scitotenv.2008.10.062.
[9] Food and Agriculture Organization of the United Nations. THE STATE OF FOOD [37] Vázquez-Rowe I, Villanueva-Rey P, Iribarren D, Teresa Moreira M, Feijoo G. Joint
AND AGRICULTURE; 2008. life cycle assessment and data envelopment analysis of grape production for
[10] Biofuels | Energy; n.d. https://ec.europa.eu/energy/topics/renewable-energy/b vinification in the Rías Baixas appellation (NW Spain). J Cleaner Prod 2012;27:
iofuels/overview_en (accessed January 7, 2021). 92–102. https://doi.org/10.1016/j.jclepro.2011.12.039.
[11] UNFCCC. ADOPTION OF THE PARIS AGREEMENT; 2015. [38] Babazadeh R, Razmi J, Pishvaee MS. Sustainable cultivation location
[12] European Commission. Sustainable transport | Mobility and Transport; n.d. htt optimization of the Jatropha curcas L. under uncertainty: A unified fuzzy data
ps://ec.europa.eu/transport/themes/sustainable_en (accessed February 11, envelopment analysis approach. Measur J Int Measur Confed 2016;89:252–60.
2021). https://doi.org/10.1016/j.measurement.2016.03.063.
[13] IEA-ETSAP, IRENA. Production of Liquid Biofules; 2013. [39] Costa AO, Oliveira LB, Lins MPE, Silva ACM, Araujo MSM, Pereira Jr. AO, et al.
[14] Barros S, Flake O. Brazil - Biofuels Annual; 2019. Sustainability analysis of biodiesel production: A review on different resources in
[15] U.S. Energy Information Administration (EIA). Biofuels explained, data and Brazil. Renew Sustain Energy Rev 2013;27:407–12. https://doi.org/10.1016/j.
statistics; 2021. https://www.eia.gov/energyexplained/biofuels/data-and-statisti rser.2013.06.005.
cs.php (accessed April 25, 2021). [40] Hong Jae-Dong, Mwakalonge Judith L. Biofuel logistics network scheme design
[16] Zhang Z, Lis M. Modeling Green Energy Development Based on Sustainable with combined data envelopment analysis approach. Energy 2020;209:118342.
Economic Growth in China. Sustainability 2020;12:1368. https://doi.org/ https://doi.org/10.1016/j.energy.2020.118342.
10.3390/su12041368. [41] Gilani H, Sahebi H, Oliveira Fabricio. Sustainable sugarcane-to-bioethanol supply
[17] Zhang Xiaolei. Essential scientific mapping of the value chain of chain network design: A robust possibilistic programming model. Appl Energy
thermochemically converted second-generation bio-fuels. Green Chem 2016;18 2020;278:115653. https://doi.org/10.1016/j.apenergy.2020.115653.
(19):5086–117. https://doi.org/10.1039/C6GC02335E. [42] Arabi Mahsa, Yaghoubi Saeed, Tajik Javad. Algal biofuel supply chain network
[18] Erdiwansyah, Mamat R, Sani MSM, Sudhakar K, Kadarohman Asep, Sardjono RE. design with variable demand under alternative fuel price uncertainty: A case
An overview of Higher alcohol and biodiesel as alternative fuels in engines. study. Comput Chem Eng 2019;130:106528. https://doi.org/10.1016/j.
Energy Rep 2019;5:467–79. https://doi.org/10.1016/j.egyr.2019.04.009. compchemeng.2019.106528.
[19] No SY. Application of hydrotreated vegetable oil from triglyceride based biomass [43] Ren J, Tan S, Dong L, Mazzi A, Scipioni A, Sovacool BK. Determining the life cycle
to CI engines - A review. Fuel 2014;115:88–96. https://doi.org/10.1016/j. energy efficiency of six biofuel systems in China: A Data Envelopment Analysis.
fuel.2013.07.001. Bioresour Technol 2014;162:1–7. https://doi.org/10.1016/j.
[20] Kalnes T, Marker T, Shonnard DR. Green diesel: A second generation biofuel. Int J biortech.2014.03.105.
Chem Reactor Eng 2007;5. https://doi.org/10.2202/1542-6580.1554. [44] Babazadeh R, Razmi J, Rabbani M, Pishvaee MS. An integrated data envelopment
[21] Knothe Gerhard. Biodiesel and renewable diesel: A comparison. Prog Energy analysis–mathematical programming approach to strategic biodiesel supply chain
Combust Sci 2010;36(3):364–73. https://doi.org/10.1016/j.pecs.2009.11.004. network design problem. J Cleaner Prod 2017;147:694–707. https://doi.org/
[22] Singh D, Subramanian KA, Singal SK. Emissions and fuel consumption 10.1016/j.jclepro.2015.09.038.
characteristics of a heavy duty diesel engine fueled with Hydroprocessed [45] The International Standards Organisation. INTERNATIONAL STANDARD
Renewable Diesel and Biodiesel. Appl Energy 2015;155:440–6. https://doi.org/ ASSESSMENT — Requirements and guilelines. Int J Life Cycle Assess 2006;2006:
10.1016/j.apenergy.2015.06.020. 652–68.
[23] European Commission. Biofuels for transport | Mobility and Transport; n.d. htt [46] Andersen Per, Petersen Niels Christian. A Procedure for Ranking Efficient Units in
ps://ec.europa.eu/transport/themes/urban/vehicles/road/biofuels_en (accessed Data Envelopment Analysis. Manage Sci 1993;39(10):1261–4. https://doi.org/
June 26, 2021). 10.1287/mnsc.39.10.1261.
[24] Energy Information Administration U. AEO2020 – Transportation; n.d. [47] Argonne GREET Model; n.d. https://greet.es.anl.gov/ (accessed January 11,
[25] Cobuloglu HI, Büyüktahtakin IE. Food vs. biofuel: An optimization approach to 2021).
the spatio-temporal analysis of land-use competition and environmental impacts. [48] Sims R, Taylor M, Saddler J, Mabee W. From 1st to 2dn generation biofuel
Appl Energy 2015;140:418–34. https://doi.org/10.1016/j. technologies. Paris; 2008.
apenergy.2014.11.080. [49] Onguglo Bonapas, Pacini Henrique, Kane Malick. Second generation biofuel
[26] Zurano-Cervelló Patricia, Pozo Carlos, Mateo-Sanz Josep María, markets. Geneva: FAPESP; 2016.
Jiménez Laureano, Guillén-Gosálbez Gonzalo. Sustainability efficiency [50] Pozo C, Galán-Martín A, Cortés-Borda D, Sales-Pardo M, Azapagic A, Guimerà R,
assessment of the electricity mix of the 28 EU member countries combining data et al. Reducing global environmental inequality: Determining regional quotas for
envelopment analysis and optimized projections. Energy Policy 2019;134: environmental burdens through systems optimisation. J Cleaner Prod 2020;270:
110921. https://doi.org/10.1016/j.enpol.2019.110921. 121828. https://doi.org/10.1016/j.jclepro.2020.121828.
[27] Kumar Abhishek, Sah Bikash, Singh Arvind R, Deng Yan, He Xiangning, [51] ISO (2006a). ISO 14040 Environmental management life cycle assessment
Kumar Praveen, et al. A review of multi criteria decision making (MCDM) principles and framework; n.d.
towards sustainable renewable energy development. Renew Sustain Energy Rev [52] Guinée JB, Heijungs R, Huppes G. Economic Allocation: Examples and Derived
2017;69:596–609. https://doi.org/10.1016/j.rser.2016.11.191. Decision Tree. Int J Life Cycle Assess 2004;9:23–33. https://doi.org/10.1007/
[28] Das Ridoy, Wang Yue, Putrus Ghanim, Kotter Richard, Marzband Mousa, BF02978533.
Herteleer Bert, et al. Multi-objective techno-economic-environmental [53] Dyson RG, Allen R, Camanho AS, Podinovski VV, Sarrico CS, Shale EA. Pitfalls
optimisation of electric vehicle for energy services. Appl Energy 2020;257: and protocols in DEA. Eur J Oper Res 2001;132(2):245–59. https://doi.org/
113965. https://doi.org/10.1016/j.apenergy.2019.113965. 10.1016/S0377-2217(00)00149-1.

15
R. Cabrera-Jiménez et al. Applied Energy 307 (2022) 118201

[54] Dong Ya Hong, Ng S Thomas. Comparing the midpoint and endpoint approaches [82] US EPA O. National Overview: Facts and Figures on Materials, Wastes and
based on ReCiPe - A study of commercial buildings in Hong Kong. Int J Life Cycle Recycling; n.d.
Assess 2014;19(7):1409–23. https://doi.org/10.1007/s11367-014-0743-0. [83] O’Driscoll R, Stettler MEJ, Molden N, Oxley T, ApSimon HM. Real world CO2 and
[55] Goedkoop M, Heijungs R, Huijbregts M, Schryver A de, Struijs J, Zelm R van. NOx emissions from 149 Euro 5 and 6 diesel, gasoline and hybrid passenger cars.
ReCiPe A life cycle impact assessment method; 2009, p. 126. Sci Total Environ 2018;621:282–90. https://doi.org/10.1016/j.
[56] Frischknecht R, Jungbluth N, Althaus HJ, Doka G, Dones R, Heck T, et al. The scitotenv.2017.11.271.
ecoinvent database: Overview and methodological framework. Int J Life Cycle [84] Tibaquirá J, Huertas J, Ospina S, Quirama L, Niño J. The Effect of Using Ethanol-
Assess 2005;10:3–9. https://doi.org/10.1065/lca2004.10.181.1. Gasoline Blends on the Mechanical, Energy and Environmental Performance of
[57] International Agency for Research on Cancer. Diesel and gasoline engine In-Use Vehicles. Energies 2018;11:221. https://doi.org/10.3390/en11010221.
exhausts. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans/ [85] Sullivan JL, Baker RE, Boyer BA, Hammerle RH, Kenney TE, Muniz L, et al. CO2
World Health Organization, International Agency for Research on Cancer 1989; emission benefit of diesel (versus Gasoline) powered vehicles. Environ Sci
46:41–185. Technol 2004;38:3217–23. https://doi.org/10.1021/es034928d.
[58] Hoekman S Kent, Broch Amber, Robbins Curtis, Ceniceros Eric, Natarajan Mani. [86] Pirjola Liisa, Kuuluvainen Heino, Timonen Hilkka, Saarikoski Sanna,
Review of biodiesel composition, properties, and specifications. Renew Sustain Teinilä Kimmo, Salo Laura, et al. Potential of renewable fuel to reduce diesel
Energy Rev 2012;16(1):143–69. https://doi.org/10.1016/j.rser.2011.07.143. exhaust particle emissions. Appl Energy 2019;254:113636. https://doi.org/
[59] Murphy CW, Kendall A. Life cycle inventory development for corn and stover 10.1016/j.apenergy.2019.113636.
production systems under different allocation methods. Biomass Bioenergy 2013; [87] Liu Chen-Guang, Xiao Yi, Xia Xiao-Xia, Zhao Xin-Qing, Peng Liangcai,
58:67–75. https://doi.org/10.1016/J.BIOMBIOE.2013.08.008. Srinophakun Penjit, et al. Cellulosic ethanol production: Progress, challenges and
[60] Moncada J, Cardona CA, Higuita JC, Vélez JJ, López-Suarez FE. Wood residue strategies for solutions. Biotechnol Adv 2019;37(3):491–504. https://doi.org/
(Pinus patula bark) as an alternative feedstock for producing ethanol and furfural 10.1016/j.biotechadv.2019.03.002.
in Colombia: experimental, techno-economic and environmental assessments. [88] Eurostat. Oil and petroleum products-a statistical overview Statistics Explained
Chem Eng Sci 2016;140:309–18. https://doi.org/10.1016/J.CES.2015.10.027. Production of crude oil; 2020.
[61] Cooper WW, Seiford LM, Zhu J. Handbook on Data Envelopment Analysis. 2nd [89] World Bank Group. Agricultural land (sq. km) | Data; n.d. https://data.wor
ed., 164. New York, NY 10013, USA: Springer US; 2011. https://doi.org/ ldbank.org/indicator/AG.LND.AGRI.K2?name_desc=false (accessed February 10,
10.1007/978-1-4419-6151-8. 2021).
[62] Seiford Lawrence M, Zhu Joe. An investigation of returns to scale in data [90] Khanna Madhu, Dhungana Basanta, Clifton-Brown John. Costs of producing
envelopment analysis. Omega 1999;27(1):1–11. https://doi.org/10.1016/S0305- miscanthus and switchgrass for bioenergy in Illinois. Biomass Bioenergy 2008;32
0483(98)00025-5. (6):482–93. https://doi.org/10.1016/j.biombioe.2007.11.003.
[63] Sueyoshi T, Goto M. Non-radial measurement. Environmental Assessment on [91] Dandajeh HA, Talibi M, Ladommatos N, Hellier P. Influence of Combustion
Energy and Sustainability by Data Envelopment Analysis. Chichester, UK: John Characteristics and Fuel Composition on Exhaust PAHs in a Compression Ignition
Wiley & Sons, Ltd; 2018. p. 95–113. https://doi.org/10.1002/9781118979259. Engine. Energies 2019;12:2575. https://doi.org/10.3390/EN12132575.
ch5. [92] Martín M. RePSIM metric for design of sustainable renewable based fuel and
[64] Avkiran NK, Tone K, Tsutsui M. Bridging radial and non-radial measures of power production processes. Energy 2016;114:833–45. https://doi.org/10.1016/
efficiency in DEA. Ann Oper Res 2008;164:127–38. https://doi.org/10.1007/ J.ENERGY.2016.08.031.
s10479-008-0356-8. [93] Wu X, Daniel R, Tian G, Xu H, Huang Z, Richardson D. Dual-injection: The
[65] Zhou P, Ang BW, Poh KL. Slacks-based efficiency measures for modeling flexible, bi-fuel concept for spark-ignition engines fuelled with various gasoline
environmental performance. Ecol Econ 2006;60(1):111–8. https://doi.org/ and biofuel blends. Appl Energy 2011;88:2305–14. https://doi.org/10.1016/j.
10.1016/j.ecolecon.2005.12.001. apenergy.2011.01.025.
[66] Tone Kaoru. A slacks-based measure of super-efficiency in data envelopment [94] Smith R, Slater FM. The effects of organic and inorganic fertilizer applications to
analysis. Eur J Oper Res 2002;143(1):32–41. https://doi.org/10.1016/S0377- Miscanthus×giganteus, Arundo donax and Phalaris arundinacea, when grown as
2217(01)00324-1. energy crops in Wales, UK. GCB Bioenergy 2010;2. https://doi.org/10.1111/
[67] Seiford Lawrence M, Zhu Joe. Modeling undesirable factors in efficiency j.1757-1707.2010.01051.x.
evaluation. Eur J Oper Res 2002;142(1):16–20. https://doi.org/10.1016/S0377- [95] Mudgal S, Lavelle P, Cachia F, Somogyi D, Majewski E, Fontaine L, et al.
2217(01)00293-4. Environmentl impacts of different crop rotations in European Union. Paris; 2010.
[68] Chen S. Environmental pollution emissions, regional productivity growth and [96] Yost MA, Kitchen NR, Sudduth KA, Allphin E. Miscanthus × Giganteus Growth
ecological economic development in China. China Economic Review 2015;35: and Nutrient Export on 22 Producer. Fields 2018. https://doi.org/10.1007/
171–82. https://doi.org/10.1016/j.chieco.2014.08.005. s12155-018-9907-2.
[69] Deng G, Li L, Song Y. Provincial water use efficiency measurement and factor [97] Perpiña Castillo Carolina, Lavalle Carlo, Baranzelli Claudia, Mubareka Sarah.
analysis in China: Based on SBM-DEA model. Ecol Ind 2016;69:12–8. https://doi. Modelling the spatial allocation of second-generation feedstock (lignocellulosic
org/10.1016/j.ecolind.2016.03.052. crops) in Europe. International Journal of Geographical Information Science
[70] Li LB, Hu JL. Ecological total-factor energy efficiency of regions in China. Energy 2015;29(10):1807–25. https://doi.org/10.1080/13658816.2015.1051486.
Policy 2012;46:216–24. https://doi.org/10.1016/j.enpol.2012.03.053. [98] Zhang B, Hastings A, Clifton-Brown JC, Jiang D, C Faaij AP. Modeled spatial
[71] Song ML, Zhang LL, Liu W, Fisher R. Bootstrap-DEA analysis of BRICS’ energy assessment of biomass productivity and technical potential of Miscanthus ×
efficiency based on small sample data. Appl Energy 2013;112:1049–55. https:// giganteus, Panicum virgatum L., and Jatropha on marginal land in China; 2020.
doi.org/10.1016/j.apenergy.2013.02.064. doi:10.1111/gcbb.12673.
[72] Choi Y, Zhang N, Zhou P. Efficiency and abatement costs of energy-related CO2 [99] Algunaibet IM, Pozo C, Angel Galán-Martín A, Huijbregts MAJ, Dowell N,
emissions in China: A slacks-based efficiency measure. Appl Energy 2012;98: Guillén-Gosálbez G. Powering sustainable development within planetary
198–208. https://doi.org/10.1016/j.apenergy.2012.03.024. boundaries Energy. Environ Sci 2019;12:1890. https://doi.org/10.1039/
[73] Chang YT, Zhang N, Danao D, Zhang N. Environmental efficiency analysis of c8ee03423k.
transportation system in China: A non-radial DEA approach. Energy Policy 2013; [100] Rockström Johan, Steffen Will, Noone Kevin, Persson Åsa, Chapin F Stuart,
58:277–83. https://doi.org/10.1016/j.enpol.2013.03.011. Lambin Eric F, et al. A safe operating space for humanity. Nature 2009;461
[74] Song Malin, Wang Shuhong, Liu Qingling. Environmental efficiency evaluation (7263):472–5. https://doi.org/10.1038/461472a.
considering the maximization of desirable outputs and its application. Math [101] Li P, Sakuragi K, Makino H. Extraction techniques in sustainable biofuel
Comput Modell 2013;58(5-6):1110–6. https://doi.org/10.1016/j. production: A concise review. Fuel Process Technol 2019;193:295–303. https://
mcm.2011.12.043. doi.org/10.1016/j.fuproc.2019.05.009.
[75] Tone K. Slacks-based measure of efficiency in data envelopment analysis. Eur J [102] Praveena V, Martin M Leenus Jesu. A review on various after treatment
Oper Res 2001;130:498–509. https://doi.org/10.1016/S0377-2217(99)00407-5. techniques to reduce NOx emissions in a CI engine. J Energy Inst 2018;91(5):
[76] Li H, Shi JF. Energy efficiency analysis on Chinese industrial sectors: An improved 704–20. https://doi.org/10.1016/j.joei.2017.05.010.
Super-SBM model with undesirable outputs. J Cleaner Prod 2014;65:97–107. [103] Venturini G, Pizarro-Alonso A, Münster M. How to maximise the value of residual
https://doi.org/10.1016/j.jclepro.2013.09.035. biomass resources: The case of straw in Denmark. Appl Energy 2019;250:369–88.
[77] Cecchini L, Venanzi S, Pierri A, Chiorri M. Environmental efficiency analysis and https://doi.org/10.1016/j.apenergy.2019.04.166.
estimation of CO2 abatement costs in dairy cattle farms in Umbria (Italy): A SBM- [104] Lin Shin-Ping, Kuo Tai-Ching, Wang Hsueh-Ting, Ting Yuwen, Hsieh Chang-Wei,
DEA model with undesirable output. J Cleaner Prod 2018;197:895–907. https:// Chen Yu-Kuo, et al. Enhanced bioethanol production using atmospheric cold
doi.org/10.1016/j.jclepro.2018.06.165. plasma-assisted detoxification of sugarcane bagasse hydrolysate. Bioresour
[78] Chiu Yung-Ho, Chen Yu-Chuan, Bai Xue-Jie. Efficiency and risk in Taiwan Technol 2020;313:123704. https://doi.org/10.1016/j.biortech.2020.123704.
banking: SBM super-DEA estimation. Appl Econ 2011;43(5):587–602. https:// [105] Farias D, Maugeri-Filho F. Sequential fed batch extractive fermentation for
doi.org/10.1080/00036840802599750. enhanced bioethanol production using recycled Spathaspora passalidarum and
[79] Ewertowska A, Pozo C, Gavaldà J, Jiménez L, Guillén-Gosálbez G. Combined use mixed sugar composition. Fuel 2021;288:119673. https://doi.org/10.1016/j.
of life cycle assessment, data envelopment analysis and Monte Carlo simulation fuel.2020.119673.
for quantifying environmental efficiencies under uncertainty. J Cleaner Prod [106] Rodionova MV, Poudyal RS, Tiwari I, Voloshin RA, Zharmukhamedov SK,
2017;166:771–83. https://doi.org/10.1016/j.jclepro.2017.07.215. Nam HG, et al. Biofuel production: Challenges and opportunities. Int J Hydrogen
[80] GAMS Development Corporation. General Algebraic Modeling System (GAMS) Energy 2017;42(12):8450–61. https://doi.org/10.1016/j.ijhydene.2016.11.125.
Release 32.1.0; 2020. [107] Mele FD, Guillén-Gosálbez G, Jiménez L. Optimal Planning of Supply Chains for
[81] Energy International Agency E. Today in Energy - U.S. Energy Information Bioethanol and Sugar Production with Economic and Environmental Concerns.
Administration (EIA); 2019. https://www.eia.gov/todayinenergy/detail.php? Comput Aided Chem Eng 2009;26:997–1002. https://doi.org/10.1016/S1570-
id=40893 (accessed February 11, 2021). 7946(09)70166-X.

16