Applied Energy 114 (2014) 366–376

Contents lists available at ScienceDirect

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

Optimizing biofuels production in an uncertain decision environment:

Conventional vs. advanced technologies
Jadwiga R. Ziolkowska ⇑
University of California at Berkeley, Department of Agricultural & Resource Economics, 207 Giannini Hall #3310, Berkeley, CA 94720-3310, USA

h i g h l i g h t s

 Algae biodiesel and switchgrass ethanol technologies are most sustainable.

 Corn ethanol needs to be limited by 0.52 billion l to optimize biofuels production.
 Biofuels production pattern changes at the 40% level of increased resources.
 The model is suitable to complement LCA and other deterministic approaches.

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

Article history: The question of increasing biofuels production and the development of different biofuels production
Received 7 June 2013 technologies has become controversial. On the one hand, production of corn-based biofuels creates a
Received in revised form 12 September ‘food/feed vs. fuel’ tradeoff condition, along with subsequent uncertainties for both consumers and pro-
2013
ducers (farmers). On the other, advanced biofuels (from, e.g., switchgrass, miscanthus, algae), although
Accepted 29 September 2013
acknowledged as environmentally friendly, are not available on a large commercial scale yet.
In addition, the limited resource availability for the production of biofuels feedstocks and the question
of a sustainable biofuels production are major issues impacting decision making. Most recently, climatic
Keywords:
Biofuels
conditions and the 2011–2012 drought in the US have imposed new uncertainties that need to be con-
Uncertainty sidered in policy- making processes.
Multi-criteria decision-making By using a multi-objective optimization model, the paper presents an approach of modeling sustainable
Linear programming biofuels production from conventional and advanced biofuels feedstocks, under the condition of limited
Fuzzy set theory resources and uncertainty resulting from incomplete information or missing knowledge about the con-
Policy evaluation sequences of possible policy actions.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction and research questions
In the face of the 2012 drought and increasing corn prices in the
US, the biofuels mandate policy supporting biofuels production be-
came more controversial. According to economic theory and mar-
ket relations in a situation of a limited market supply, corn
prices are predicted to spike and can cause a surge of other cereal
prices and later also increases in dairy and meat prices. Market
spillovers are possible, but uncertain, and if they occur, they could
result in further food crises in the short- and mid-term.
One of the theoretical solutions could be a waiver of the biofuels
mandates which would allow for shifting corn to food and feed
processing and away from fuel production. However, as of 2012,
the corn ethanol production in the US exceeded the mandated val-
ues, which means that the mandate is not binding and waiving it
⇑ Tel.: +1 510 642 3345; fax: +1 510 643 8911.
E-mail address: 11scientist@gmail.com
would most likely have little impact on corn ethanol production
[1].
Also, if theoretically the mandate policy is suspended, would it
be feasible to meet the environmental standards in the transport
sector? The advanced biofuels technologies (not competing with
food/feed) seem to offer a solution to the current problem; how-
ever, they are still limited and the advanced biofuels are not avail-
able in the desired quantities on the market.
Biofuels are classified into two groups: (1) conventional biofuels
produced from eatable crops (ethanol from corn, sugar beet, sugar
cane, cereals, cassava and other sugars; biodiesel from vegetable
oils, e.g., soybean, rapeseed, sunflower, palm oil) and (2) advanced
biofuels encompassing both (a) cellulosic ethanol produced from
agricultural waste, forest residue, municipal solid waste or energy
crops grown for fuel; (b) biodiesel from microalgae; and (c) drop-in
biofuels produced in biochemical processes and/or from geneti-
cally modified crops for fuel. Ethanol is blended with gasoline,
while biodiesel with diesel fuel to reach the environmental

0306-2619/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.apenergy.2013.09.060
 J.R. Ziolkowska / Applied Energy 114 (2014) 366–376
standards set by the Environmental Protection Agency (EPA). Cur-
rently, ethanol from corn and biodiesel from canola and soybean
are the main biofuels sources in the US.
In 2011/2012, the total ethanol production from corn amounted
to 13,559 million gallons and made 98.3% of the total ethanol pro-
duction in the country. Cellulosic ethanol was not produced on a
commercial scale in 2011/2012, but 9 million gallons of this fuel
are estimated to be supplied on the market in 2012/2013. The total
biodiesel production and consumption in 2011/2012 amounted to
1109 million gallons with no cellulosic diesel being produced in
that year. Both ethanol and biodiesel production are expected to
increase continuously in the following years, except for the decline
in corn ethanol in 2012/2013, which can be explained by the
drought and decreased feedstock yields. The prices for biofuels dif-
fer in different parts of the country with the average of $2.46 for a
gallon of corn-based ethanol in 2011/2012 in Omaha and $4.75 per
gallon biodiesel. The year 2012/2013 has been a breakthrough for
the cellulosic and other advanced biofuels technologies and thus
a limited production of cellulosic ethanol is expected to hit the
market by the end of the fiscal year [2].
In the US, the Renewable Fuel Standard (RFS) requires 13.2 bil-
lion gallons of conventional (mostly corn-based) biofuel to be pro-
duced in 2012. The goal of the RFS is to reduce greenhouse gas
emissions and the nation’s dependence on foreign oil. In 2010,
the RFS2 set environmental and volume requirements for different
types of advanced biofuels (cellulosic, advanced and renewable
biofuel). In the same year, the EPA downsized the mandatory stan-
dards for cellulosic ethanol due to the gap between the actual bio-
mass ethanol production (8 million gallons) and the targeted 100
million gallons [3].
An urgent question is about the production volume of biofuels
that could foster sustainable development and minimize market
distortions. Possible solutions to this question are not straightfor-
ward and need to be evaluated from different perspectives of the
applied technologies that bear several economic and environmen-
tal advantages as compared to conventional technologies.
In this paper, a fuzzy logic optimization model is presented that
analyzes objective-oriented and sustainable biofuels production
scenarios in the US, given limited and uncertain resource availabil-
ity. The model considers both traditional (corn, soybean, canola)
and advanced biofuels (switchgrass, algae) technologies, thus com-
paring their potentials in optimizing biofuels production in
general.
In the process of designing a sustainable biofuels policy, multi-
ple economic, environmental and social aspects and policy objec-
tives need to be considered. A comprehensive methodology for
such kind of an evaluation is, for instance, a multi-criteria analysis.
The research on multi-criteria methodology for biofuels production
and policies has evolved in recent years with four major studies of
Dinh et al. [4], Mohamadabadi et al. [5], Turcksin et al. [6] and Peri-
menis et al. [7] who evaluated multiple criteria in the biofuels pro-
duction chain. However, those studies do not address uncertainties
related to decision-making processes. De Vries et al. [8] used indi-
cators to evaluate resource efficiency of different crops that can be
used for biofuels production. The indicator analysis could be incor-
porated in the multi-criteria optimization framework presented in
this study.
Also, several authors addressed uncertainty of biofuels. Awudu
and Zhang [9,10]analyzed sustainability and profitability of the
biofuel supply chain under demand and price uncertainties and
found the biofuels supply chain to be most profitable if the ethanol
price is between $2.25 and $2.50. This study provides a valuable
contribution to market uncertainties; however it has a different fo-
cus than the analysis described in this paper which aims at pre-
senting a process of designing policies and decision-making
processes. Also Rajagopal and Plevin [11] addressed uncertainties
367
of biofuels production in the context of market related changes
based on two indicators: greenhouse gas emissions and indirect
land use changes. The research presented in this paper encom-
passes a large number of parameters in a bigger decision support
framework. Thus, it addresses a wider range of existing sustain-
ability aspects in biofuels policy making than in the above men-
tioned studies.
Moreover, the question of biofuels sustainability has been ad-
dressed broadly in the literature in recent years and the studies pro-
vide mainly analytical and descriptive analyses. Alam et al. [12]
analyzed long-term potentials of algae as a substitute for gasoline.
Sheehan [13], Diaz-Chavez [14] and Markevicius et al. [15] provided
an overview of sustainability aspects for biofuels. Amigun et al. [16]
went one step further and proposed policy incentives to improve
sustainable development in biofuels policies. Chalmers and Archer
[17] provided a case study of a sustainability reporting scheme for
biofuels and found an extensive stakeholder assessment a necessary
requirement of a successful evaluation process of biofuels schemes.
Also Scarlat and Dallemand [18] proposed a bioenergy certification
system as a possible way to boost sustainability of biofuels. In their
communication, Gasparatos et al. [19] argue about the necessity of a
unified appraisal approach for biofuels, while Gnansounou [20]
developed a logic-based model for assessing sustainability of biofu-
els and proved a knowledge-based system (involving stakeholders)
a suitable tool for sustainability assessment. Even though, sustain-
ability assessment is an important part of the biofuels feedstocks
analysis provided in this paper, the value added relates to measur-
ing uncertainty in biofuels decision-making processes rather than
sustainability of biofuels as such.
Among different studies in the literature, either on sustainabil-
ity or uncertainties of biofuels, none of them address the question
of uncertainty accompanying each decision-making process. Nei-
ther do they provide possible ways of designing policies in an
interactive decision-making process with policymakers. To fill this
gap, Ziolkowska et al. [21] proposed a multi-criteria framework for
several biofuels feedstocks, capturing uncertainties of decision
making, as well as for different policy instruments [22]. In this pa-
per, by including fuzzy logic in the PROMETHEE approach and the
LP model various aspects of missing or imprecise information,
complexity and vagueness of policy options and/or, among others,
subjectivity of expert assessments’ (as given with the expert elici-
tation approach) were evaluated. It is necessary to consider those
aspects to gain a broader picture of different biofuels production
technologies and decision-making processes. This would allow pol-
icymakers to comprehend potentials of different biofuels technol-
ogies first, before making drastic decisions about the future of
the biofuels industry, based on the knowledge about conventional
technologies only. Uncertainty aspects have been included in the
study in two steps:
(1) In the PROMETHEE approach that, by applying a concept of
the linguistic variable, addresses the question of missing or
imprecise information, vague concepts and subjective expert
assessments, and
(2) In the fuzzy LP model by determining the right-hand sides of
the constraints as fuzzy coefficients representing the avail-
able resources. Since the amount of resources available for
biofuels production purposes specifically, is not deter-
mined/known or enforced at this point of time, the paper
analyzes different scenarios and levels of biofuels produc-
tion in case of changing resource availability. It shows trade-
offs between the feedstocks in meeting the defined policy
objectives.
As most advanced biofuels are not produced commercially yet,
the model provides a tool that can be used to analyze sustainable
368 J.R. Ziolkowska / Applied Energy 114 (2014) 366–376

biofuels production scenarios, provided the advanced biofuels the analyzed feedstocks for meeting the defined biofuels policy
technologies to be in place. The paper addresses the following objectives, while the PROMETHEE decision function encompasses
questions: all biofuels policy objectives, as defined in this study.
The analyzed biofuels feedstocks were expressed as a set of
– How can biofuels production be designed in order to maximize alternatives A = {a1, a2, . . . , a5} representing corn (a1), switchgrass
the goals of a sustainable biofuels policy? (a2), soybean (a3), canola (a4) and algae (a5). One of the main steps
– How can uncertainties of resource availability and limited infor- in the PROMETHEE approach is the evaluation of the alternatives
mation on biofuels technologies be incorporated in decision- (here: biofuels feedstocks) in terms of the defined criteria (here:
making processes and what are potential subsequent implica- biofuels policy objectives).
tions for sustainable biofuels production and resource use? The policy objectives (cn) included in this study were divided
– What are production tradeoffs between the biofuels feedstocks into three groups: economic, environmental and social criteria
in the frame of a sustainable biofuels policy? with the respective sub-criteria, covering the issue of a sustainable
biofuels policy:
The paper is structured as follows. The next section presents
methodology and data sources for developing the optimization ap- Economic criteria: Environmental Social criteria:
proach. It briefly describes the way of deriving objective coeffi- Maximizing criteria: Protecting Improving social
cients by means of the PROMETHEE (Preference Ranking policy efficiency the environment environment
Organization METHod for Enrichment Evaluations) approach. It
? Reducing ? Reducing ? Increasing
also shows a way of applying fuzzy set theory to address uncer-
biofuels greenhouse gas consumer welfare
tainties in decision-making processes resulting from resource
production costs emissions (c5) (c9)
availability and missing information about the respective biofuels
(c1)
technologies and their contribution to the general biofuel produc-
? Increasing ? Reducing water ? Supporting
tion. Section three presents an optimization model for different
biofuels usage (c6) local
biofuels production scenarios, while section four shows an objec-
productivity/ha communities (c10)
tive-oriented biofuels production and tradeoffs between the feed-
(c2)
stocks, subject to the resource availability. Finally, conclusions
? Insuring ? Reducing land ? Improving
and possible extensions are formulated in section five.
national food use (c7) health and safety
security (c3) issues (c11)
2. Methodology and data ? Securing ? Protecting ? Creating new
farmers’ biodiversity and jobs (c12)
To analyze objective-oriented and sustainable biofuels produc- incomes (c4) landscapes (c8)
tion, a multi-objective linear programming optimization model
was developed, combining fuzzy logic and parametric program-
ming. Five biofuels feedstocks were included in the analysis: corn
The criteria were defined based on the most common state-
and switchgrass (for ethanol production) and soybean, canola
ments of biofuels policy objectives in policy reports and scientific
and algae (for biodiesel production). Only feedstocks were chosen
literature. The objectives were indicated as a set C = {c1, c2, . . . ,
that are used on a large scale for biofuels production in the US and
c12} and cover several aspects that can improve the effectiveness
represent conventional biofuels feedstocks (competing with food
of the biofuels policy.
production) (corn, soybean, canola) or feedstocks that are expected
The PROMETHEE methodology in this study covers both tangible
to be used on a commercial scale in the future, while being in an
(measurable) variables (e.g., biofuels production costs, GHG emis-
experimental and small scale production stage as of 2012 (ad-
sions, water use, land use) and intangible (non-measurable) vari-
vanced biofuels technologies from switchgrass, algae). Other feed-
ables (e.g., health and safety issues, national food security, etc.).
stocks that could potentially be a viable feedstock solution, but are
The measureable variables were derived from experiment-based
not produced in the US (e.g., jatropha), have not been included in
publications on different biofuels feedstocks describing the degree
the analysis. Also, feedstocks used on a very small scale (e.g.,
of the respective feedstocks of meeting the defined objectives
wheat, corn stover and forest residues) have not been included in
[23,4,24]. They were included in the LP (Linear Programming) model
the analysis, but could be assessed by the same means in the fu-
as constraints coefficients (see Table 3). The intangible variables
ture. Moreover, the selection of the feedstocks was determined
were assessed by twenty experts from the US Environmental Pro-
by the availability of statistical data on the feedstock sustainability
tection Agency (EPA), US Department of Agriculture (USDA), Food
potential as addressed in this paper. The procedures developed for
and Agricultural Policy Research Institute (FAPRI), International
this analysis can, in principle, be extended to any other feedstock
Food Policy Research Institute (IFPRI) and from several US universi-
whenever sufficient data becomes available.
ties. The experts assessed the relative importance of the defined pol-
For the analysis, data from the FAPRI Outlook Database on bio-
icy objectives and the relations between the feedstocks in terms of
fuels production from the enumerated feedstocks in the US in 2010
those objectives. For this study, only experts specialized in the field
was used. Also, the performance of the feedstocks in terms of the
of biofuels with an extensive knowledge about all analyzed biofuels
defined constraints related to the resource use (and derived from
technologies have been selected and their assessments were anon-
the PROMETHEE approach) was included in the model. The perfor-
ymous. A survey with farmers and citizens was omitted at that point
mance indicators were expressed with the constraint coefficients
of time, since neither of those groups possesses sufficient and de-
and will be described in detail in the following sections.
tailed knowledge on the discussed technologies and their economic,
environmental and social aspects, as defined. Thus, in this way, po-
2.1. PROMETHEE approach and derivation of objective coefficients tential evaluation biases were avoided in the first place.
The experts assessed the policy objectives with a numerical
The objective coefficients for the model have been generated by scale 1–10 (with 1 indicating the lowest importance and 10 repre-
the PROMETHEE (Preference Ranking Organization METHod for senting the highest level of importance) and assigned the relative
Enrichment Evaluations) approach. They denote the potential of weight vectors w = {w1, w2, . . . , w12} to the respective policy
 J.R. Ziolkowska / Applied Energy 114 (2014) 366–376 369

Table 1 µ(x)
Linguistic variables and triangular fuzzy numbers for biofuels feedstocks analysis.
Source: Author’s construction

Linguistic variable Triangular fuzzy number 1

Very low (VL) (0.00, 0.00, 0.25)

Low (L) (0.00, 0.25, 0.50)
Medium (M) (0.25, 0.50, 0.75) α
High (H) (0.50, 0.75, 1.00)
Very high (VH) (0.75, 1.00, 1.00)

objectives. In the next step, the experts assessed the relations be- 0
tween the biofuels feedstocks and the policy objectives by means a b c x
of a linguistic scale (a tool of fuzzy logic introduced by Zadeh
Fig. 1. Triangular fuzzy number and membership function. Source: Author’s
[25,26]). The linguistic scale included the following evaluation lev- construction according to Kahraman et al. (2006) [33].
els: ‘very low’ (VL), ‘low’ (L), ‘medium’ (M), ‘high’ (H), ‘very high’
(VH). Those assessments were further translated into fuzzy num-
bers. The process of assigning fuzzy sets to the enumerated linguis- e in X is represented as follows:
The fuzzy set A
tic terms will be explained in Section 2.2 (see also Table 1). The
choice of two different scales was advisable due to the fact that X
5
e¼
A  lA ðxi Þ
the preferences of decision makers with regard to the importance
i¼1
of the respective objectives can be described with precise numeri-
cal values, while the preferences of alternative solutions can be where lA(xi) denotes the value of the membership functions of the
determined only approximately. respective alternatives and  indicates the union operator in fuzzy
logic symbolic (compare: Hersh [32]).
2.2. Fuzzy relations and membership function In this paper, the triangular membership function (Fig. 1) was
applied, due to its simplicity in representing vagueness and uncer-
In this study, a fuzzy PROMETHEE approach was developed that tainty. According to the definition of the triangular membership
represents a methodological extension of a standard (crisp) PROM- function, an element x has the membership degree of 1 (and thus
ETHEE model as introduced initially by Brans [27]. The extension of a full membership), if and only if x = b, while it takes the member-
the model by fuzzy set theory allowed for addressing uncertainties ship value of 0 (and becomes a non-member) if and only if
resulting from imprecision of policy objectives, incomplete/vague x 6 a or x P c. Each membership value, such that:
information about policy options, imprecise information as well a < x < b and b < x < c, assigns an intermediate degree of mem-
as from potential subjectivity of experts’ assessments or inaccuracy bership (between 0 and 1) to the element x that is called a triangu-
of measurements (see also [28,29]). Other than a binary and prob- lar fuzzy number. Each triangular

fuzzy number is represented in
abilistic number, a linguistic variable assigns an interval range for this study as the triplet A ¼ ða; b; cÞ.
the membership function of an element in the fuzzy set. In this As the universe X can accommodate full membership, partial
way, a broader solution space can be defined for vague and thus membership or non-membership, fuzzy set theory allows for treat-
less specific variables. ing fuzziness in a quantitative way. In this study, the concept of
In this section, the concept of the membership function and fuzziness represented by Smithson and Verkuilen [30] and De Luca
fuzziness will be introduced briefly and the way of translating the and Termini [34] is used. According to the authors, a set is maxi-
linguistic scale to fuzzy numbers will be explained. The information mally fuzzy if all its elements have the membership of ½ in it,
on those specific steps of the analysis is limited to the extent rele- while it is completely crisp if all its elements have the membership
vant for understanding the optimization model in Section 3. of 0 or 1. Accordingly, the following criteria determine fuzziness
According to the fuzzy set theory, the membership of any ele- fuz(A) of the set A:
ment x belonging to the set A(sA(x)) can take only two values: 0
(non-member/‘false’) or 1 (member/‘true’), where A is a subset of fuzðAÞ ¼ 0; iff lA ðxÞ ¼ 0 or lA ðxÞ ¼ 1; 8x 2 A
the entire set X (universe of discourse). The mapping of the func-
tion is denoted by sA:X ? {0, 1}. Therefore, for each element 1
fuzðAÞ ¼ maximum; iff lA ðxÞ ¼ ; 8x 2 A
x e X, it applies: 2

1; iff x 2 A
sAðxÞ ¼ fuzðAÞ 6 fuzðBÞ; iff 8x 2 A; either lA ðxÞ 6 lB ðxÞ
0; iff x R A
1
In this study, the universe of discourse X is a finite set X = {x1, x2, whenever lB ðxÞ < ;
2
. . . , x5} with the set elements x1, x2, . . . , x5 denoting the analyzed
biofuels feedstocks. 1
Each element x in a fuzzy set à and in the universe X is de- or lA ðxÞ P lB ðxÞ whenever lB ðxÞ >
2
scribed by means of a membership function with the numerical
‘degree of membership’ on the real continuous and closed interval e
fuzðAÞ ¼ fuzð AÞ
between 0 (non-membership) and 1 (complete membership).
The fuzzy set A e in X is a set of ordered pairs In this paper, the concept of the membership function and fuzz-
e
A ¼ fx; le ðxÞ; 8x 2 Xg with the membership function: iness were used to define the fuzzy sets (fuzzy numbers) for eval-
A
uating the importance of the biofuels feedstocks in terms of the
lA : X ! H
biofuels policy objectives. As mentioned above, the linguistic scale
which maps all x e X into an ordered set H (called the ‘membership used by the experts was translated into fuzzy numbers, and thus
set’ and representing the

set [0, 1]), while l ðxÞ indicates the grade different importance levels were assigned to different interval sets
A
of membership of x in A for each x e X [30,31]. (Table 1).
370 J.R. Ziolkowska / Applied Energy 114 (2014) 366–376

The expert assessments for each biofuels feedstock in terms of
each policy objective (here: criterion) were averaged for all ex-
perts. Table 2 displays the input data used in the subsequent calcu-
lations and the optimization model, such as: the expert
assessments of criteria weights and the coefficients for the non-
measurable criteria as well as the coefficients for the measurable
criteria that have been expressed as fuzzy scalars (c1, c2, c5-c7).
By means of the PROMETHEE approach, fuzzy difference, fuzzy
preference index, fuzzy Aggregated Multi-criteria Preference Index,
and positive and negative flows have been calculated in multiple
analysis steps. At the end, the outcomes were defuzzified. The final
performance coefficients determined the complete ranking of the
biofuels feedstocks.
For the purpose of the optimization model and an objective-ori-
ented biofuels production analysis presented in this paper, only the
final coefficients of the PROMETHEE approach are relevant (row 3
7–9 in Table 3). Therefore, the exact process of deriving and calcu-
lating those coefficients is omitted here and is elaborated in Ziolk-
owska [35].
3. Optimization model for biofuels production subject to
resource availability
Most decisions are taken in an environment where the objec-
tives, constraints or parameters are incomplete and/or not pre-
cisely known [36]. According to Zeleny [37]: ‘‘Human decision
makers do not satisfy a well-defined criterion function subject to
well-structured constraints. They search for recognizable, harmonious,
optimal formulation and solution patterns (or structures). Traditional
notion of optimality, such that, maximize f(x) subject to x e X, is there-
fore inadequate. Fuzziness, ambiguity and imprecision facilitate and
integrate human (. . .) patterns of knowledge, decisions and
optimality’’.
The idea of incorporating fuzzy set theory into optimization ap-
proaches to analyze problems plagued by uncertainty and impreci-
sion was introduced by Bellman and Zadeh [38] and further
developed by, among others, Tanaka et al. [39], Negoita and Sularia
[40], Zimmerman [41], Orlovsky [42], Yager [43], and Farina and
Amato [44]. Fuzzy linear programming has been used to solve
real-world problems in many disciplines, e.g., agricultural econom-
ics [45,46], banking and finance, environmental management,
manufacturing and production, marketing, personal management,
transportation [47].
For the presented analysis, a fuzzy linear programming (LP)
model was developed to answer the question of an objective-ori-
ented biofuels production. The model is called an LP model with
fuzzy resources which indicates that the right hand of the con-
straints (representing the available resources) is expressed with
fuzzy numbers.
It needs to be underlined at the very outset that the model
should be understood as a tool that can be used in decision-making
processes in an interactive process with policymakers. Since the
model includes several assumptions (due to missing data or infor-
mation about the policy options and available resources), the re-
sults of the analysis should not be understood and interpreted as
definite solutions to the biofuels production problem. This is true
also due to some most recent developments in advanced biofuels
technologies that have not been included in this paper, and are still
in the experimental or pilot stage development process.
The paper points out benefits of integrating fuzzy set theory
into mathematical programming for multi-criteria decision-mak-
ing modeling of biofuels production under the condition of uncer-
tain and unknown resource availability. The uncertainty of
available resources is defined in this paper as a decision situation
where either the optimal resource use or the availability of the re-
sources for the cultivation of the respective biofuels feedstocks is
not known exactly.
3.1. Definition of the objective function
In order to compute an objective-oriented biofuels production
from the analyzed feedstocks, the objective function (reflecting

Table 2
Input data for the optimization model.

Criteria Weights Alternatives

Corn Switchgrass Soybean Canola Algae
c1 7.80 (0.42 0.42 0.42) (0.54 0.54 0.54) (0.57 0.57 0.57) (0.73 0.73 0.73) (0.62 0.62 0.62)
c2 6.80 (0.15 0.15 0.15) (0.47 0.47 0.47) (0.56 0.56 0.56) (0.86 0.86 0.86) (51.92 51.92 51.92)
c3 6.00 (0.18 0.36 0.60) (0.26 0.48 0.70) (0.15 0.36 0.60) (0.19 0.40 0.63) (0.40 0.63 0.81)
c4 6.30 (0.55 0.80 0.96) (0.33 0.58 0.83) (0.45 0.70 0.90) (0.36 0.61 0.85) (0.13 0.29 0.54)
c5 7.25 (0.76 0.76 0.76) (0.27 0.27 0.27) (3.44 3.44 3.44) (2.95 2.95 2.95) (0.11 0.11 0.11)
c6 6.15 (0.69 0.69 0.69) (0.48 0.48 0.48) (15.46 15.46 15.46) (15.46 15.46 15.46) (14.33 14.33 14.33)
c7 5.80 (66.00 66.00 66.00) (21.00 21.00 21.00) (18.00 18.00 18.00) (12.00 12.00 12.00) (0.20 0.20 0.20)
c8 5.60 (0.08 0.25 0.50) (0.31 0.56 0.80) (0.06 0.25 0.50) (0.11 0.33 0.56) (0.33 0.56 0.79)
c9 5.85 (0.16 0.39 0.64) (0.25 0.48 0.73) (0.16 0.39 0.64) (0.14 0.36 0.60) (0.30 0.54 0.78)
c10 5.70 (0.41 0.66 0.88) (0.35 0.60 0.81) (0.41 0.66 0.88) (0.34 0.59 0.81) (0.09 0.30 0.54)
c11 4.85 (0.10 0.34 0.59) (0.23 0.46 0.70) (0.15 0.39 0.64) (0.15 0.39 0.64) (0.19 0.43 0.66)
c12 5.65 (0.24 0.49 0.73) (0.28 0.53 0.76) (0.25 0.50 0.74) (0.19 0.44 0.68) (0.30 0.55 0.78)

Table 3
Model matrix for biofuels production (base-case scenario). Source: Author’s calculations.

1. Corn Switchgrass Soybean Canola Algae Sum/resource use

2. Current production 47.96 0.00 0.94 0.02 0.00 48.92 Current production (billion l biofuels)
3. Optimal production 47.44 0.50 0.75 0.02 0.22 48.92 Optimal production (billion l biofuels)
4. Objective coefficients 1.0 47.3 31.7 38.2 105.3 Objective coefficients
5. Upper bounds 71.94 0.50 1.41 0.03 0.50 74.38 Total upper bounds for the measures (billion l)
6. Lower bounds 38.37 0.00 0.75 0.02 0.00 39.14 Total lower bounds for the measures (billion l)
7. Constraint 1: GHG emissions 0.76 0.27 3.44 2.95 0.11 39.75 Constraint 1 coefficients (billion kg CO2 eq.)
8. Constraint 2: Water use 0.69 0.48 15.46 15.46 14.33 47.96 Constraint 2 coefficients (billion kl)
9. Constraint 3: Land use 66.00 21.00 18.00 12.00 0.20 3182.60 Constraint 3 coefficients (billion m2)
10. Constraint 4: Max. biofuels production 1.00 1.00 1.00 1.00 1.00 48.92 Coonstraint 4 coefficients (billion l)
 J.R. Ziolkowska / Applied Energy 114 (2014) 366–376 371

biofuels production) was maximized subject to several constraints teria (as applied in the fuzzy PROMETHEE approach) and the
related to available resources. The objective function has the fol- corresponding economic, environmental and social aspects of the
lowing form: biofuels policy (Section 2.1).
Currently, there is no legitimate limit for resource use in the
X
5
Max Z; s:t: Z ¼ zi x biofuels feedstock production. Therefore, for the modeling pur-
i¼1 poses, the resource use optimizing sustainability goals of the biofu-
els policy was estimated by maximizing the objective function in
where Z is the objective function, zi is the objective coefficients for the situation of the current biofuels production in 2010. The esti-
the respective alternatives (biofuels feedstocks) for i = 1, . . . , 5, x is mated resource values were then set as the restrictions for the
the current biofuels production from the respective feedstocks for inequality constraints. Three of the constraints need to be mini-
i = 1, 2, . . . , 5. mized: greenhouse gas (GHG) emissions, water use and the land
subject to: use constraints.
x P 0 and lC i ðxÞ; 8i: The following equation and Fig. 2 explain briefly the theory and
conditions

of minimizing fuzzy constraints. For the fuzzy con-
The objective function incorporates the economic, environmen- straint C i s:t:; i ¼ m1 þ 1 . . . ; m2 to be minimized, the membership
tal and social objectives of biofuels policy as presented in Sec- function takes a form:
tion 2.1. The function is defined as the sum product of the 8
> Xn
objective coefficients for the respective biofuels feedstocks (de- >
> 1; iff aij x < bi
>
>
rived from the PROMETHEE approach – see Section 2.2) and the >
> j¼1
>
> P 
current biofuels production from those feedstocks as of 2010. The >
>
< n
a xbi X n
objective function (representing the general biofuels production) is lC i ðxÞ ¼ 1  j¼1 ij
; iff b
0
aij x 6 bi þ bi
i 6
maximized subject to the following constraints: reduction of >
>
0
bi
>
> j¼1
greenhouse gas (GHG) emissions, reduction of water use, and >
>
>
> Xn
reduction of land use for biofuels feedstocks production, consider- >
> 0; iff aij x > bi þ bi
0
>
:
ing in addition the constraint of the possible maximum biofuels j¼1
production in this study.
The membership function indicates that:
An optimal production is defined as a production with possibly
low GHG emissions, possibly low inputs of water and land re- P
(1) If nj¼1 aij x < bi , then the ith constraint is absolutely satisfied
sources, and the highest biofuels production in general. The restric-
ðleCi
ðxÞ ¼ 1Þ.
tions for the constraints (right hand side of the constraint equation Pn 0 0
– see Table 3) were defined by using a fuzzy approach. In this way, (2) If j¼1 aij x > bi þ bi , where bi is the maximum tolerance

the question of uncertain resource availability (i.e., missing infor-
mation about available resources for cultivating the respective bio-
fuels feedstocks and/or about optimal resource use) was addressed.
It makes the analysis more comprehensive and realistic as com-
pared to deterministic approaches which define exact constraints
(with crisp numbers) for even uncertain events and situations.
For eliciting optimal biofuels production levels from different
feedstocks, a parametric programming approach was used. Differ-
ent than traditional sensitivity analysis, this approach allows for
analyzing optimal solutions by considering changes of several re-
sources simultaneously while the changes of the constraints can
be different for different resources.
3.2. Definition of the constraints and possible solution space
from bi, then the ith constraint is absolutely violated and
the degree of satisfaction is equal to zero ðle Ci
ðxÞ ¼ 0Þ.
Pn 0
(3) If j¼1 a ij x 2 ðbi ; bi þ bi Þ, then the membership function is
monotically decreasing, which means that the higher
achievement of the constraint, the lower the satisfaction of
a decision maker.
In the base-case scenario, the parameter h is equal to zero and
indicates that there is no uncertainty in resource availability. It also
displays the situation that would be given in any classical model
with crisp constraints. This paper provides an extension to stan-
dard deterministic optimization models, by applying fuzzy con-
straints. The model constraints were defined as follows:The
model constraints were defined as follows:

The objective function was maximized, subject to four con- (1) GHG emissions constraint: leCGHG ðxÞ
straints assuming uncertain/unknown resource availability repre-
sented with the parameter h. The parameter h is changing X
5
0
X
5

between 0 and 1 in different scenarios, thus indicating higher or

aiGHG x 6 biGHG þ h  biGHG ) aiGHG x 6 39:75 þ h  10
i¼1 i¼1
lower levels of available resources. The constraints were defined
for the base-case scenario with the parameter h = 0 denoting no with aiGHG denoting the GHG emission coefficients for the respective
uncertainty in terms of the available resources (low resource avail- biofuels feedstocks (performance of the biofuels feedstocks in terms
ability), and h = 1 denoting an extended and the highest level of
available resources, as defined in this model.
The constraints are defined as ‘available resources’ for biofuels
production. While this term is plausible for the water and land
use constraints, in case of the GHG emission constraint, the term
‘available resources’ is defined as the permitted level of the GHG
emissions related to biofuels production from the respective
feedstocks.
For the purpose of the analysis, only four constraints have been
included in the fuzzy LP model due to the abundance of conditions
restricting each other and limiting the possible solution space.
Although, only environment related constraints have been consid- Fig. 2. Membership function for the constraints to be minimized. Source: Author’s
ered, the objective function was calculated for all sustainability cri- construction.
372 J.R. Ziolkowska / Applied Energy 114 (2014) 366–376
of GHG emissions), biGHG denoting the current GHG emissions by the
0
respective biofuels feedstocks in the base-case scenario, biGHG repre-
senting the tolerance margin for GHG emissions set randomly and
deviating by 10 units from the resource use in the base-case
scenario.
The subscript i in the variable biGHG and other constraint vari-
ables, does not refer to the alternatives, but to the lower and upper
bounds of the constraints (see Fig. 2). The equation shows that the
right-hand side of the equation depends both on the h parameter
0
and the tolerance margin for the constraint biGHG . For this and the
0
other constraints to be minimized, bi was assumed to be equal
to 10 (setting a restriction upper bound on the resource use 10
times higher than the current optimal resource use). At the same
0
time, for the total biofuels production constraint, bi was assumed
to be 1. Those assumptions are random, as no exact policy require-
ments exist in terms of the resource use at this point of the biofuels
policy formulation. However, they logically allow for an exemplary
analysis of possible changes in biofuels production. Furthermore, it
should be underlined that the aim of the analysis is, among others,
to present an alternative modeling approach that can extend the
application of the standard crisp models. Thus, random assump-
tions allow for achieving this goal. They could differ and would
need to be accordingly adjusted, depending on data availability
for any other research problem or analysis region.
In the base-case scenario (h = 0), the constraint denotes that the
sum of GHG emissions from biofuels production from the analyzed
feedstocks should be less than or equal to 39.75 billion kg CO2-eq.
When introducing uncertainty in resource use, the parameter h
changes and further determines changes of the right-hand side of
the constraint. Thus, in this case, by permitting higher levels of
GHG emissions in the biofuels feedstock production process, the
possible solution space for this constraint is extended up to
0
49.75 billion kg CO2-eq. (if h = 1 and biGHG ¼ 10).
(2) Water usage constraint: leCWater ðxÞ
X
5
0
X
5 multi-objective LP problem with fuzzy resources for biofuels
aiWater x 6 biWater þ h  biWater ) aiWater x 6 47:96 þ h  10
i¼1 i¼1
with aiWater denoting the water use coefficients for the respective
biofuels feedstocks, biWater denoting the current water use for biofu-
0
els production from the respective feedstocks cumulatively, biWater
denoting the tolerance margin for water use deviating by 10 units
from the resource use in the base-case scenario.
The right-hand side of this constraint was set as an upper bound
(and the constraint is to be minimized). The constraint expresses
the goal of reaching maximum possible biofuels feedstocks yields
by a possible minimum input of water resources. In the base-case
scenario, the maximum water use for biofuels production should
be less than or equal to 47.96 billion kl (kiloliter) (which indicates
the current optimal water use for producing biofuels from the ana-
lyzed feedstocks). By introducing uncertainty in water use for bio-
fuels purposes, the right-hand side of the equation is accordingly
changing and reaches the maximum value of 57.96 billion kl if
h = 1.
(3) Land use constraint: leCLand ðxÞ
X
5
0
X
5 (indicating the largest distance to the maximum coefficient) was
aiLand x 6 biLand þ h  biLand ) aiLand x 6 3182:60 þ h  10
i¼1 i¼1
with aiLand denoting the land use coefficients for the respective bio-
fuels feedstocks, biLand denoting the current land use for biofuels
0
production from the respective feedstocks cumulatively, biLand
denoting the tolerance margin for land use deviating by 10 units
from the resource use in the base-case scenario.
Also the land use constraint is to be minimized, since biofuels
production is not supposed to compete for land resources with
other activities, e.g., agricultural production. In the base-case sce-
nario, the maximum land use for biofuels feedstock production
should be less than or equal to 3182.60 billion m2 (square meter)
which is the current optimal land use estimated with the model.
Considering a higher level of land resources available for biofuels
feedstock production, the right-hand side of the equation can reach
the maximum value of 3192.60 if h = 1 and is accordingly varying
between 3182.60 and 3192.60 billion m2 for all h from the interval
[0, 1].
(4) Lower bound for the total biofuels production: le
C Prod
ðxÞ
X
5
0
X
5
aiProd x P biProd  h  biProd ) aiProd x P 48:92  h  1
i¼1 i¼1
with aiProd denoting the production coefficients for the respective
biofuels feedstocks, biProd denoting the current biofuels production
0
from the respective feedstocks cumulatively, biProd denoting the tol-
erance margin for biofuels production deviating by 1 unit from the
current production in the base-case scenario.
The biofuels production in this model is to be maximized, thus,
the right-hand side of the constraint was set as a lower bound. In
the base-case scenario, the maximum level of biofuels production
should be equal or greater than 48.92 billion liter which denotes
the current biofuels production. By considering uncertainty (and
assuming a lower production level than given currently), the
right-hand side of the equation can vary between 47.92 billion liter
(for h = 1) and 48.92 billion liter (for h = 0).
In addition, the non-negativity constraint was considered, such
that:
x P 0; 8 i ¼ 1; 2; . . . ; 5
It excludes solutions with negative biofuels production levels.
Table 3 displays the model matrix and coefficients of the fuzzy
production.
Row 1 of the matrix displays the analyzed biofuels feedstocks,
while row 2 the current production of the respective biofuels feed-
stocks. As of 2010, no biofuels were produced from switchgrass
and algae on the commercial scale in the US. Therefore, current
production coefficients for biofuels from these feedstocks amount
to zero.
The calculation of an optimal (objective-oriented) production
(row 3) is based on the current biofuels production in 2010
(48.92 billion liters). Row 4 displays the objective coefficients from
the preference function in the PROMETHEE approach. Some coeffi-
cients in the PROMETHEE approach have a minus sign and would
be inadequately impacting the objective function in the optimiza-
tion model. To resolve this complication, the coefficients were
recalculated to absolute values by means of the semantic distance
approach which estimates distances between the alternatives in
terms of their performance. For this purpose, the highest PROM-
ETHEE coefficient was assigned 100% importance and the remain-
ing coefficients were compared to the maximum coefficient by
subtraction. In this way, a distance between the respective alterna-
tives was established by absolute values. The lowest coefficient
assigned the value of 1. Accordingly, the positive (absolute value)
coefficients were calculated based on the distance of the respective
alternatives to the maximum alternative. Thus, algae have the
highest importance coefficient followed by switchgrass, canola,
soybean and corn (which is congruent with the original results in
the PROMETHEE approach and represents the same absolute per-
formance distances between the alternatives).
 J.R. Ziolkowska / Applied Energy 114 (2014) 366–376
Furthermore, the upper and lower bounds were defined for the
biofuels production (row 5-6). The upper bound was calculated
based on an assumption of a 50% production increase, while the
lower bound was set at the level of at least 80% of the current pro-
duction in 2010. As the current biofuels production from switch-
grass and algae is equal to zero, an assumption was made for the
upper production bound to be 0.50 billion liter biofuels both from
the switchgrass and algae feedstocks respectively, while the cur-
rent production of 0 was set as the lower bound.
The coefficients for the constraints 1–3 (rows 7–9) represent
the performance of the biofuels feedstocks in terms of the respec-
tive objectives (as given in the PROMETHEE approach) and were
expressed in the respective constraint units per liter of biofuels.
The constraint 4 on the total biofuels feedstock production (row
10) attributes the value of 1 for each alternative to meet the condi-
tion of an optimal production to be equal to or greater than the
current production.
The right-hand side of the equations (resource use) was esti-
mated for the base-case scenario, as described above. By means
of the Visual Basic programming and the Simplex algorithm, the
objective function was maximized and optimal solutions for biofu-
els production were calculated. Further, with a stepwise parame-
terization of the h parameter between 0 and 1, biofuels
production was calculated for different scenarios of simulta-
neously changing resources.
4. Results and discussion
4.1. Optimal biofuels production in base-case scenario
The results show that changes in the production levels would be
necessary to maximize economic, environmental and social policy
criteria subject to the given constraints, as defined in the model. As
the second and third generation biofuels (here switchgrass and al-
gae, respectively) are only emerging technologies on the market,
and also due to several assumptions in the presented model, the
analysis should be understood as an attempt to demonstrate a tool
for interactive policy making under the condition of changing re-
source availability (uncertain resource availability), rather than a
definite optimal policy solution.
An objective-oriented biofuels production in the base-case sce-
nario shows that 47.44 billion l ethanol would be produced from
corn, 0.50 billion l ethanol from switchgrass, 0.75 billion l biodiesel
from soybean, 0.02 billion l biodiesel from canola, and 0.22 billion l
biodiesel from algae. The production levels are clearly determined,
among others, by the objective coefficients; and only biofuels pro-
duction from switchgrass (with the second high objective coeffi-
cient) reaches the upper production bound. The production of
0.6
0.50
0.4
0.22
0.2
Billion l
0.0
0.00
-0.2
-0.19
-0.4
-0.6 -0.52
Corn Switchgrass Soybean Canola Algae
Fig. 3. Difference between the optimal and current biofuels production in base-case
scenario. Source: Author’s calculations.
373
biofuels from soybean and canola is determined by the lower
bound that is binding in this case.
Fig. 3 shows the difference between the optimal (objective-ori-
ented) production in the base-case scenario and the current biofu-
els production from the respective feedstocks as of 2010. According
to the estimation, the base-case scenario production level maxi-
mizing the objective function would require to increase ethanol
production from switchgrass by 0.50 billion l and biodiesel produc-
tion from algae by 0.22 billion l. Also, it would require to decrease
ethanol production from corn by 0.52 billion l and biodiesel pro-
duction from soybean by 0.19 billion l. Biodiesel production from
canola would remain at the same level as currently given.
4.2. Biofuels production scenarios with changing resource availability
The base-case scenario in Section 4.1 presents a comparative-
static analysis of an objective-oriented biofuels production focused
on maximizing the defined economic, environmental and social
policy criteria. However, in reality, available resources or policy
standards are constantly changing, and thus, the final biofuels pro-
duction from the enumerated feedstocks will change in the course
of time also. In addition, in most decision situations and optimiza-
tion problems, available resources are not always precisely speci-
fied, and can be assessed only approximately. This applies
particularly to the biofuels production, due to changing economic,
environmental and policy conditions as well as technological
improvements. Since there are no standard levels or policy require-
ments/limitations determining resource use for biofuels feedstocks
production, resource allocation is arbitrary and unlimited as of
now.
The current draught in the US and bad corn yields raised doubts
about the effectiveness and sustainability of biofuels policies in
general. Despite those events, and also due to new technological
approaches in biofuels production, and fluctuating corn market
prices, biofuels still have good prospects to maintain a share in
the fuel market. Given the fact that, as of 2012 biofuels production
exceeded the mandated conventional biofuels levels and also that
the new emerging biofuels technologies offer more sustainable
solutions, an increase of biofuels production could be theoretically
expected in the long term, especially from second and third gener-
ation feedstocks.
In this situation, the question about the resource availability
arises and further about the impact of higher resource availability
for biofuels feedstock production on final biofuels production. An
alternative question would be: what would be the effect of a cap
on resource use for biofuels feedstocks production, due to either
policy regulations or scarcity of natural resources. Although this
question is highly interesting, it is not investigated specifically in
this paper. However, the cap on the resource use is introduced in
the model in the form of the upper bounds for the resource con-
straints and the current resource use.
Certainly, setting a limit (either a stringent or varying) for re-
source use can determine a new level of an optimal biofuels pro-
duction mix. In the following scenarios, a limit for resource use is
equal to the maximum resource use. Thus, it strictly impacts the
level of GHG emissions as well as water and land use for the biofu-
els feedstocks production. By means of parametric programming,
biofuels production scenarios were calculated in the situation of
simultaneously occurring changes of all resources. In this paper,
only the situation of increased resource availability is discussed.
For this purpose, the h parameter (included in the right-hand side
of the resource constraints – Table 3, rows 7–9) was stepwise
parameterized between 0 and 1. The parameterization of the h
parameter shows deviations from the resource values in the
base-case scenario (h = 0). In the parameterization process, the
374 J.R. Ziolkowska / Applied Energy 114 (2014) 366–376

Ethanol production from corn

48.00 Ethanol production from switchgrass and
biodiesel from soybean, canola and algae
47.90 1.20
47.80 1.00
Billion l

0.80

Billion l
47.70
0.60
47.60
0.40
47.50 0.20
47.40 0.00
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Switchgrass Soybean Canola Algae

Fig. 4. Optimal biofuels production scenarios subject to changing resources. Note: 0 represents the base-case scenario in terms of the available resources, while 1 indicates
the maximum resource availability in the model, based on the available resources in the base-case scenario. Source: Author’s calculations.

Ethanol from corn Ethanol from switchgrass

0.60
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.00 0.50
-0.10 0.40

Billion l
-0.20 0.30
Billion l

-0.30 0.20
-0.40
0.10
-0.50
0.00
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-0.60

Biodiesel from soybean Biodiesel from canola

0.012
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.20 0.010
0.15
0.008
Billion l

0.10
0.05 0.006
Billion l

0.00 0.004
-0.05
0.002
-0.10
-0.15 0.000
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
-0.20

Biodiesel from algae

0.60

0.50

0.40
Billion l

0.30

0.20

0.10

0.00
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Fig. 5. Absolute production changes as compared to the current biofuels production in the situation of changing available resources. Note: 0 represents the base-case scenario
in terms of the available resources, while 1 indicates the maximum resource availability in the model, based on the available resources in the base-case scenario. Source:
Author’s calculation.

available solution space for the right-hand of the linear constraints
(resource use) is accordingly extended.
Parametric programming and the LP model with fuzzy con-
straints allow for calculating changes that could not be detected
with a traditional sensitivity analysis only. While sensitivity anal-
ysis allows for a variation of a parameter or parameters (even
simultaneously), the rate of the parameter change could be only
the same (e.g., 10% change for all parameters at the same time).
With parametric programming, simultaneous changes of different
parameters (here: resources) by different proportional changes
for each parameter can be evaluated, which provides a more real-
istic analysis of changes occurring in the real world.
 J.R. Ziolkowska / Applied Energy 114 (2014) 366–376
As described in previous sections, the resource changes are ex-
0
pressed with the bi parameter for GHG emissions, water use and
land use. The parameter was set to 10, while the parameter
changes (indicating resource changes) occurred in 10% steps as
compared to the base-case scenario. Thus, the permitted GHG
emissions can vary between 39.75 kg CO2-eq. (if h = 0) and
49.75 kg CO2-eq. (if h = 1), the water use can vary between 47.96
billion kl and 57.96 billion kl, and the interval for the land use
changes is between 3182.60 billion m2 and 3192.60 billion m2.
The total biofuels production can vary between 48.92 billion l (if
h = 0) and 47.92 billion l (if h = 1).
The analysis shows that at the level of 40% resource increase
(h = 0.4) (as compared to the base-case scenario), corn ethanol pro-
duction increases from 47.56 to 47.92 billion l. The same increasing
tendency applies to biodiesel production from soybean and canola,
although the increase is less considerable than in the case of corn
ethanol (Fig. 4). In case of biodiesel from algae, the production of
biodiesel would increase only up to 40% resource availability in
the base-case scenario. If the resource increase is more than 40%,
biodiesel production from algae remains on an unchanged level
of 0.50 billion l. It is clearly limited by the upper bound of the de-
fined constraint in the fuzzy linear programming model. Only the
production of switchgrass ethanol is constant regardless of the
changes in available resources. Due to its high importance ex-
pressed with the high objective coefficient, it reaches the upper
bound of the constraint in the base-case scenario already.
A comparison of optimal biofuels production from the analyzed
feedstocks at different levels of available resources is displayed in
Fig. 5. With an increasing level of all resources (as compared to the
current biofuels production), corn ethanol production would de-
crease by 0.52 billion l in the base-case scenario (h = 0) and only
by 0.05 billion l in case of the maximum resource availability for
all resources defined in this model (when h = 1). The production
of switchgrass ethanol would increase by 0.50 billion l as com-
pared to the current production, regardless of any resource
changes. The soybean biodiesel production would decrease by
0.19 billion l if h = 0 and would subsequently increase by 0.16 bil-
lion l (if h = 1). Biodiesel production from canola would not change
in the base-case scenario and as long as h < 0.4 (and also as long as
algae is produced below the upper bound value), and would in-
crease by 0.01 billion l for each amount of the available resources
indicated by h > 0.4. A similar trend applies to algae biodiesel pro-
duction. At each level of available resources with h < 0.4, biodiesel
production from algae would increase stepwise and would reach a
maximum possible increase of 0.50 billion l for available resources
higher than 40% of the base-case scenario resource availability. The
analysis shows that biofuels production is highly sensitive to
changes of the available resources in the defined model as well
as when compared to current production. It proves practical bene-
fits of the presented optimization model with fuzzy resources and
including uncertainty factors related to resource use and resource
availability in biofuels production. The biofuels production levels
from the respective feedstocks are strongly dependent, among oth-
ers, on the defined upper and lower production bounds, as defined
in the model. As no scientific or political information exists on this
very specific issue, in this study, model relevant assumptions were
made. Thus, the analysis also shows the necessity of an interactive
analysis with policymakers that would allow for defining possible
and envisaged upper and lower production levels that could be
more realistic in terms of reaching the anticipated policy goals in
the long term.
5. Conclusions and outlook
The paper presents a linear programming model with fuzzy re-
sources for estimating an objective-oriented biofuels (ethanol and
375
biodiesel) production from traditional and advanced biofuels feed-
stocks. It shows practical benefits of applying this methodology in
interactive decision-making processes and designing sustainable
biofuels policies with policymakers.
The analysis shows algae and switchgrass to be most suitable
for biofuels production in terms of maximizing the defined eco-
nomic, environmental and social policy criteria simultaneously.
Corn, that is currently the most important feedstock for ethanol
production in the US, has the lowest sustainability potential among
the analyzed feedstocks. In order to maximize the economic, envi-
ronmental and social objectives of the biofuels policy, ethanol pro-
duction from switchgrass would need to be extended by
0.50 billion l, while biodiesel production from algae by 0.22 billion
l. At the same time, ethanol production from corn would need to be
limited by 0.52 billion l and biodiesel production from soybean by
0.19 billion l. Biodiesel production from canola could remain at the
same level as in the base-case scenario in 2010.
Also, biofuels production scenarios were analyzed for changing
resources. A breakthrough point in biofuels production is the level
of 40% increase in resource use as compared to the base-case sce-
nario. Biodiesel production from canola would not change in the
base-case scenario and would remain at the same level up to the
40% increase in available resources. At any level of available re-
sources higher than 40% of the base-case scenario level, biodiesel
production from canola would be higher by 0.01 billion l. A similar
trend applies to algae biodiesel. At each level of available resources
lower than 40% of the base-case resource availability, biodiesel
production from algae would increase stepwise and remain at
the maximum level of 0.50 billion l for each following resource
increase.
The linear programming methodology with fuzzy constraints
can be useful to policymakers dealing with decision situations
where information is unknown, limited, incomplete or imprecise.
As the current biofuels policy has not been evaluated by the EPA
or USDA in terms of multiple policy objectives and possible alterna-
tives (different production technologies), the presented approach
has a very high potential to be used as a practical tool for supporting
decision-making processes on the national and/or regional levels.
Currently, most policy decisions are based on cost-benefit analyses
and market-related indicators which are insufficient in addressing
complex multilevel problems of biofuels production, especially
environmental and social aspects. An interactive decision process
would allow the model to be accordingly adjusted and comple-
mented according to specific conditions and policy priorities in
the analyzed regions. It would help to more distinctively specify
the respective model variables, for instance, the upper and lower
bounds of the production level or limits for resource use.
As the model includes switchgrass and algae that are currently
not used on a commercial scale for biofuels production, the analy-
sis shows a decision situation reaching into the future. In the next
decades, advanced production technologies are expected to domi-
nate the biofuels market and provide feedstocks that will allow for
minimizing production costs and resource use, e.g., poplar for cel-
lulosic ethanol or pongamia for biodiesel production. The new
technologies are further expected to boost the experimental re-
search and provide solutions that have not been widely applied
yet.
In order to foster research and technological development,
interdisciplinary databases are necessary to provide data on con-
ventional and advanced feedstocks and biofuels. Due to many
uncertainties existing on different levels of biofuels assessment,
the development of a comprehensive database will be determined
by the current research studies and both experimental and com-
mercial biofuels production processes. This could also support
interdisciplinary analyses and integration of biofuels feedstocks
in other combined energy production systems [48–50].
376 J.R. Ziolkowska / Applied Energy 114 (2014) 366–376
The paper provides an example of a comparative-static analysis.
A possible extension could include a dynamic fuzzy linear pro-
gramming model for production changes occurring over time and
for predicting future biofuels production under changing condi-
tions of available resources. However, for this kind of analysis, a
larger data set would be required that is not available at this point
of time with the current limited knowledge, especially on
advanced biofuels feedstock costs and production trends. Also,
methodological extensions are recommended to complement mul-
ti-criteria approaches by new components, e.g., considering sub-
normality and convexity of fuzzy sets or incorporating neural
networks into the decision analysis. Those extensions, though
requiring large data sets, would allow for creating a more compre-
hensive picture of reality in biofuels production. The model can
also include indicators, currently used as separate methodologies,
e.g., net energy ratio commonly applied in Life-Cycle-Analysis. In
this way and by combining different methodologies, a more de-
tailed and real picture of biofuels production could be created. This,
again, would help to design policy solutions in a more effective and
sustainable way.
Acknowledgment
The author thanks Prof. William H. Meyers and Michael H.
Weiss, JD, for their valuable remarks and editing of this paper.
The author also acknowledges the valuable comments of the Editor
and anonymous Reviewers. The research was funded by the Euro-
pean Commission within the 7th EU Framework Program.
References
[1] Thompson W, Whistance J, Westhoff P, Binfield J, Renewable fuel standard
waiver options during the drought of 2012. FAPRI-MU Report #11–12. FAPRI:
University of Missouri – Columbia; 2012.
[2] FAPRI. FAPRI-MU 2013 baseline tables. Fapri: Columbia-Missouri; 2012.
[3] FAPRI (Food and Agricultural Policy Institute). World Biofuels: FAPRI 2010
Agricultural Outlook. FAPRI: Iowa State University; 2010.
[4] Dinh L, Guo Y, Mannan M. Sustainability evaluation of biodiesel production
using multicriteria decision-making. Environ Progr Sustain Energy
2009;28(1):38–46.
[5] Mohamadabadi H, Tichkowsky G, Kumar A. Development of a multi-criteria
assessment model for ranking renewable and non-renewable transportation
fuel vehicles. Energy 2009:112–25.
[6] Turcksin L, Macharis C, Lebeau K, Boureima F, VanMierlo J, Bram S, et al. A
multi-actor multi-criteria framework to assess the stakeholder support for
different biofuels options: the case of Belgium. Energy Policy 2011;39:200–14.
[7] Perimenis A, Walimwipi H, Zinoviev S, Muller-Langer F. Development of a
decision support tool for the assessment of biofuels. Energy Policy
2011;39:1782–93.
[8] De Vries S, Gerrie W, van de Ven J, van Ittersum M, Giller K. Resource use
efficiency and environmental performance of nine major biofuel crops,
processed by first-generation conversion techniques. Biomass Bioenergy
2010;34:588–601.
[9] Awudu I, Zhang J. Uncertainties and sustainability concepts in biofuel supply
chain management: a review. Renew Sustain Energy Rev 2012;16:1359–68.
[10] Awudu I, Zhang J. Stochastic production planning for a biofuel supply chain
under demand and price uncertainties. Appl Energy 2013;103:189–96.
[11] Rajagopal D, Plevin R. Implications of market-mediated emissions and
uncertainty for biofuel policies. Energy Policy 2013;56:75–82.
[12] Alam F, Date A, Rasjidin R, Mobin S, Moria H, Baqui A. Biofuel from algae – is it
a viable alternative? Procedia Eng 2012;49:221–7.
[13] Sheehan J. Biofuels and the conundrum of sustainability. Curr Opin Biotechnol
2009;20:318–24.
[14] Diaz-Chavez R. Assessing biofuels: aiming for sustainable development or
complying with the market? Energy Policy 2011;39:5763–9.
[15] Markevicius A, Katinas V, Perednis E, Tamasauskiene M. Trends and
sustainability criteria of the production and use of liquid biofuels. Renew
Sustain Energy Rev 2010;14:3226–31.
[16] Amigun B, Musango K, Stafford W. Biofuels and sustainability in Africa. Renew
Sustain Energy Rev 2011;15:1360–72.
[17] Chalmers J, Archer G. Development of a sustainability reporting scheme for
biofuels: a UK case study. Energy Policy 2011;39:5682–9.
[18] Scarlat N, Dallemand JF. Recent developments of biofuels/bioenergy
sustainability certification: a global overview. Energy Policy 2011;39:
1630–46.
[19] Gasparatos A, Lehtonen M, Stromberg P. Do we need a unified appraisal
framework to synthesize biofuel impacts? Biomass Bioenergy 2013;50:
75–80.
[20] Gnansounou E. Assessing the sustainability of biofuels: a logic-based model.
Energy 2011;36:2089–96.
[21] Ziolkowska J, Simon L, Zilberman D, Capturing uncertainties in evaluation of
biofuels feedstocks: a multi-criteria approach for the US. European Association
of Agricultural Economists 2011 ‘Change and Uncertainty Challenges for
Agriculture Food and Natural Resources’, August, 30–September, 2, 2011,
Zurich, Switzerland.
[22] Ziolkowska JR. Fuzzy multi-criteria framework for supporting biofuels policy
making. Int J Sci, Eng Technol 2011;59:341–5.
[23] Mata T, Martins A, Caetano N. Microalgae for biodiesel production and other
applications: a review. Renew Sustain Energy Rev 2010;14:217–32.
[24] Pimentel D, Patzek T. Ethanol production using corn, switchgrass, and wood;
biodiesel production using soybean and sunflower. Nat Resour Res
2005;14(1):65–76.
[25] Zadeh LA. Fuzzy sets. Inf Control 1965;8:338–53.
[26] Zadeh LA. Fuzzy algorithms. Inf Control 1968;12:99–102.
[27] Brans JP, Vincke P. A preference ranking organisation method: the
PROMETHEE method for multiple criteria decision-making. Manage Sci
1985;31(6):647–56.
[28] Sivanandam SN, Sumathi S, Deepa SN. Introduction to fuzzy logic using
MATLAB. Heidelberg: Springer; 2007.
[29] Novak V, Perfilieva I, Mockor J. Mathematical principles of fuzzy
logic. Boston: Kluwer Academic Publishers; 1999.
[30] Smithson M, Verkuilen J. Fuzzy set theory. Applications in the Social Sciences,
vol. 147. USA: SAGE Publications; 2006.
[31] Munda G. Social multi-criteria evaluation for a sustainable economy. Berlin-
Heidelberg: Springer; 2008.
[32] Hersh M. Mathematical modelling for sustainable development. Berlin-
Heidelberg: Springer; 2006.
[33] Kahraman C, Ertay T, Büyüközan G. A fuzzy optimization model for QFD
planning process using analytic network approach. Eur J Oper Res
2006;171:390–411.
[34] De Luca A, Termini S. A definition of an nonprobabilistic entropy in this setting
a fuzzy sets theory. Inf Control 1972;20:301–12.
[35] Ziolkowska JR. Fuzzy multi-criteria evaluations in complex decision systems.
How to deal with uncertainties in biofuels policies – sustainability study for
biofuels feedstocks in the United States. Aachen: Shaker Publisher; 2013.
[36] Liu B. Theory and practice of uncertain programming. Heidelberg: Physica
Verlag; 2002.
[37] Zeleny M. Fuzziness, knowledge and optimization: new optimality concepts.
In: Delgado M, Kacprzyk J, Verdegay JL, Vila M, editors. Fuzzy optimization.
Recent advances. Heidelberg: Physica Verlag; 1994. p. 3–20.
[38] Bellman R, Zadeh LA. Decision-making in a fuzzy environment. Manage Sci
1970;17:B141–164.
[39] Tanaka H, Okuda T, Asai K. On fuzzy mathematical programming. J Cybernet
1974;3:37–46.
[40] Negoita C, Sularia M. On fuzzy mathematical programming and tolerances in
planning. Econ Comput Econ Cybernet Stud Res 1976;1:3–15.
[41] Zimmerman HJ. Description and optimization of fuzzy system. Int J Gen Syst
1976;2:209–16.
[42] Orlovsky S. On programming with fuzzy constraint sets. Kybernetes
1977;6:197–202.
[43] Yager R. Mathematical programming with fuzzy constraints and a preference
on the objective. Kybernetes 1979;8:285–92.
[44] Farina M, Amato P. A fuzzy definition of optimality for many criteria
optimization problems. IEEE Trans Syst Man Cybern Part A Syst Hum
2004;34(3):315–26.
[45] Owsinski J, Zadrozny S, Kacprzyk J. Analysis of water use and needs in
agriculture through a fuzzy programming model. In: Kacprzyk J, Orlovski S,
editors. Optimization models using fuzzy sets and possibility
theory. Dordrecht: Reidel; 1987. p. 377–95.
[46] Leung Y. Interregional equilibrium and fuzzy linear programming. Environ
Plan 1988;A 20:25–40. 219–230.
[47] Jana B, Roy TK. Multi-objective fuzzy linear programming and its
application in transportation model. Tamsui Oxford J Math Sci 2005;
21(2):243–68.
[48] Daianova L, Dotzauer E, Thorin E, Yan J. Evaluation of a regional bioenergy
system with local production of biofuel for transportation, integrated with a
CHP plant. Appl Energy 2012;92:739–49.
[49] Song H, Starfelt F, Daianova L, Yan J. nfluence of drying process on the biomass-
based polygeneration system of bioethanol, power and heat. Appl Energy
2012;90:32–7.
[50] Wang X, Nordlander E, Thorin E, Yan J. Microalgal biomethane production
integrated with an existing biogas plant: a case study in Sweden. Appl Energy
2013;112:478–84.