You are on page 1of 15

Agricultural Systems 193 (2021) 103208

Contents lists available at ScienceDirect

Agricultural Systems
journal homepage: www.elsevier.com/locate/agsy

Review

Operations research for environmental assessment of crop-livestock


production systems
Mohammad Davoud Heidari a, *, Ian Turner a, Amir Ardestani-Jaafari b, Nathan Pelletier a, *
a
Food Systems PRISM Lab, University of British Columbia, 3247 University Way, Kelowna, BC V1V 1V7, Canada
b
Faculty of Management, University of British Columbia, 1137 Alumni Avenue, Kelowna, BC V1V 1V7, Canada

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

• Intrinsic variability in agricultural pro­


duction systems and the high number of
decision variables call for a rigorous
DSS.
• OR methods used for optimization of
agricultural and livestock systems for
environmental goals were reviewed.
• Three most commonly considered deci­
sion types were farm benchmarking,
output prediction and resource use
management.
• A framework (decision tree) considering
methodological choices with respect to
each method was proposed.
• It helps farmers to make informed de­
cisions based on their tresources, and
environmental impacts of their produc­
tion system.

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

Editor: Mark Van Wijk CONTEXT: Agricultural crop and livestock production systems are complex, essential to human well-being, and
fraught with sustainability challenges. In light of intrinsic variability in agricultural production systems and the
Keywords: high number of decision variables, decision support for optimization of sustainability outcomes should be sup­
Life cycle assessment ported by rigorous operations research.
Optimization
OBJECTIVE: Several operations research (OR) methods such as evolutionary algorithms, multi-objective opti­
Statistical analysis
mization, and data envelopment analysis (DEA) have been applied to optimization in agricultural contexts,
Decision-making
Variability taking into account different objective functions and decision variables, and life cycle-based evaluation of
Agricultural systems environmental outcomes in agriculture have become widespread. The current review evaluates the methods used
for optimization of agricultural and livestock systems for life cycle-based environmental sustainability goals.
METHODS: A Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) systematic review
method and three complementary search strategies were used to identify relevant articles. Strengths, weaknesses,
and performance issues for each method are considered and compared.
RESULTS AND CONCLUSIONS: Farm benchmarking, output prediction and resource use management are the
three most commonly considered decision types in crop-livestock production systems. To guide selection and
implementation of appropriate OR methods, a framework (decision tree) is proposed.

* Corresponding authors.
E-mail addresses: md.heidari@outlook.com (M.D. Heidari), nathan.pelletier@ubc.ca (N. Pelletier).

https://doi.org/10.1016/j.agsy.2021.103208
Received 12 January 2021; Received in revised form 16 June 2021; Accepted 21 June 2021
Available online 13 July 2021
0308-521X/© 2021 Elsevier Ltd. All rights reserved.
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

SIGNIFICANCE: The proposed decision tree provides an indication of necessary method-specific methodological
choices. Methodological choices with respect to each method are discussed.

1. Introduction environmental goals contribute to agricultural decision-making pro­


cesses is still lacking. More information is required to critically analyze
As the human population continues to grow (Gerland et al., 2014), so the necessary elements for taking environmental goals into account.
too must food production to satisfy increasing demand (Conforti, 2011). Moreover, despite increased attention to the potential utility and
At the same time, agri-food systems make significant contributions to a importance of applying optimization methods to improving sustain­
variety of environmental problems including biodiversity loss, ecotox­ ability outcomes for agricultural crop and livestock products, a sys­
icity, water pollution, climate change and land use (Clark et al., 2020). tematic basis for selecting appropriate optimization methods has not
Consequently, attention to measuring and managing the sustainability been advanced to date. With a focus on literature that integrates LCA
impacts of agri-food systems is growing. and optimization methods, the current review hence aims to:
Life cycle thinking is increasingly applied in support of
sustainability-oriented decision making. Within this context, life cycle 1- Identify and characterize trends in the application of OR methods for
assessment (LCA), which is a formalized methodology for quantifying reducing environmental impacts in agricultural crop and livestock
the cumulative resource use, energy demands, and environmental im­ production systems (Section 3);
pacts of a product or service throughout its entire life cycle (Hauschild 2- Identify the strengths and limitations of existing OR methods for
et al., 2017), is now widely applied to evaluate the sustainability im­ addressing the environmental impacts of agricultural crop and live­
pacts of food systems (Nemecek et al., 2015a, 2015b; Pelletier, 2015). stock production systems (Section 4);
Studies have variously addressed comparisons between alternative 3- Provide recommendations regarding best-fit OR methods for
production methods (Pelletier, 2017; Pelletier et al., 2008), assessment different applications/contexts in agricultural crop and livestock
of management decisions (MacWilliam et al., 2014), and evaluation of production systems (Section 5).
alternative technologies (Li and Mupondwa, 2014), among others.
Agricultural systems and their impacts are, however, highly het­ 2. Review methodology
erogeneous (Notarnicola et al., 2017). Differences in climatic conditions
(Nemecek et al., 2015b), soil type (Fedele et al., 2014), seed type This review used the Preferred Reporting Items for Systematic re­
(Nemecek et al., 2015a, 2015b), water availability (Fedele et al., 2014), views and Meta-Analyses (PRISMA) systematic review method (Moher
and farm practices (Fedele et al., 2014; Hokazono and Hayashi, 2015; et al., 2009), which includes “search strategy”, “screening criteria,” and
Knudsen et al., 2014; Nemecek et al., 2015a; Nemecek et al., 2012) may “extraction and synthesis of data” stages. Fig. 1 depicts how literature
lead to large differences in both the yields and environmental impacts were selected and screened in order to arrive at the final subset of (86)
characteristic of different farms and agricultural products (Boone et al., papers that were identified for detailed review. Relevant peer-reviewed
2016). In light of this heterogeneity, there is considerable potential to literature was first identified using key word searchers with the Web of
improve resource use efficiencies and reduce associated emissions Science search engine. Web of Science citation data was selected as it is
through identification and dissemination of sustainability best practices. accurate and its outcome can be easily reproduced by other researchers.
Operations research (in the US) / operational research (in the UK) A combination of logical operators “AND” or “OR” for search keywords
employs statistics, optimization techniques, and other analytical ap­ were applied (see Fig. 1). Relevant articles were identified by applying
proaches to model and solve complex decision-making problems (Car­ three complementary search strategies: Search I located articles that use
ravilla and Oliveira, 2013). Techniques from this field may hence help Environmental and Operations Research (OR) keywords in their title
realize efficiency gains and impact reductions in crop-livestock pro­ and crop-livestock keywords in their title and abstract, and Search II
duction systems when applied in concert with LCA. Possible analyses identified articles in the category of “Operations Research Management
include benchmarking the performance of farms relative to one another Science,” which is limited to OR Journals. Additional literature was
using data envelopment analysis (DEA) (Cooper and Seiford, 2007), and identified through searching the reference lists of selected literature
optimization using mathematical programming and machine learning. (Search III).
Due to considerable variability in farm level inputs, yields, and Two levels of screening were applied to select a subset of relevant
environmental impacts, optimization of agricultural systems using a literature. First, the Web of Science toolbox was used to remove pro­
single objective function will seldom be sufficient. Instead, multi- ceedings articles, editorial material and duplicates, as well as papers
objective optimization based on integration of an optimization method written in languages other than English. Second, irrelevant literature
and a mathematical algorithm that enables simultaneous consideration was identified and excluded by reviewing titles and abstracts. Only
of several objective functions will typically be required (Arora, 2015). literature that integrated LCA and optimization were selected for
Multi-objective optimization has been applied to a wide range of agri­ detailed consideration. For example, if a paper only considered opti­
cultural production systems covering both environmental and economic mization in the discussion section, it was discarded. Publications that
goals. Example applications include dairy (Breen et al., 2019), wheat were not open access or for which institutional access was unavailable
(Galán-Martín et al., 2017), and field crop (Klein et al., 2013; Cap­ were excluded as well. This review covers publications from 2000 on­
itanescu et al., 2017) production systems. Several studies have also re­ ward because beginning in the 2000s diverging approaches to LCA were
ported integration of different optimization methods into LCA of developed (e.g., spatially differentiated LCA, dynamic LCA, conse­
agricultural and livestock production systems (Ali et al., 2015; Moha­ quential LCA, and risk-based LCA). This was also the time that many
mad et al., 2014; Pishgar-Komleh et al., 2019; Vázquez-Rowe et al., international organizations invested in developing standardized
2012). methods and policy applications (i.e., UNEP/SETAC launched Life Cycle
Recently published literature reviews have variously addressed Initiative, European Commission communicates Integrated Product
optimization methods (Solano et al., 2020; Kaim et al., 2018), simula­ Policy). Specifically, literature published from 2000 to 2020 (July) were
tion modelling (Utomo et al., 2018), and machine learning methods considered. The initial database searches resulted in 246 hits.
(Liakos et al., 2018; Sharma et al., 2020) for application to agricultural Descriptive analysis of the consulted literature, including the rate of
supply chain research. Nonetheless, a general understanding of how publication of relevant literature over the last 20 years, is depicted in

2
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

Section 3 (Review Question 1). This section intends to describe the screened for information on objective functions, constraints, and deci­
trends in publication by providing quantitative descriptions of the dis­ sion variables to answer Review Question 2 (Section 4). Specific infor­
tribution of articles over time, and distribution of papers by crop or mation required for individual methods was identified by consulting
livestock type. Recently published literature reviews which analyzed the recently published literature reviews addressing individual methods.
application of different machine learning techniques and optimization More precisely, for Data Envelopment Analysis (DEA) papers (Section
methods to agricultural supply chain research (Kaim et al., 2018; Liakos 4.1), extracted data including case study type, spatial scope of the study,
et al., 2018; Sharma et al., 2020; Solano et al., 2020; Utomo et al., 2018) number of decision making units (DMUs), number of inputs and outputs,
were consulted to choose a set of application types that include all orientation of analysis, return to scale, and integration approach (Thies
reviewed methods. Based on these literature reviews and according to et al., 2019) were tabulated and discussed. Selected literature that
the purpose of the assessment, OR methods in each paper were cate­ applied Mathematical Programming (MP) (Section 4.2) and Evolu­
gorized into three groups: benchmarking (papers that employed Data tionary Algorithms (EI) (Section 4.3) were discussed on the basis of
Envelopment Analysis), resource use management (papers that tabulated information including scope of the study, objective functions,
employed mathematical programming and evolutionary algorithms), formulation type (linear programming, integer programming, and non-
and prediction (papers that employed neural networks). Depending on linear programming), decision type (operational, tactical, and strategic),
the OR method, the Methods sections of the selected literature were decision variables, and constraints (Liakos et al., 2018; Thies et al.,

Fig. 1. Selection of literature based on the PRISMA systematic review method.

3
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

2019). Consulted literature that applied Neural Networks (NN) (Section eco-efficiency of individual farms. Publications from the last five years
4.4) was investigated for their network architecture and performance mostly report using evolutionary algorithms in optimization of envi­
measures (Cabaneros et al., 2019). Selected literature were screened to ronmental impacts and cost of production. The overall trend of publi­
identify the strengths and limitations of each optimization modelling cation is parallel with increasing attention to LCA since the early 2000s,
method with respect to their integration approach with LCA based on when the Life Cycle Initiative was launched by UNEP/SETAC (Hauschild
analysis of the Results, Discussion and Conclusion sections. Limitations et al., 2017) and after 2009 when more attention has focused on specific
of the literature were classified based on the type of optimization applications of LCA such as dynamic LCA, carbon footprinting and
methods and were discussed (Review Question 2). The Results sections consequential LCA (Guinée et al., 2011).
of each study were screened to identify potential improvements, as The application areas of the studies can be divided into two cate­
indicated by the applied optimization method, with respect to envi­ gories: agricultural crops and livestock products (Fig. 3). Agricultural
ronmental impacts (the type of improvement, reduction percentage, crop production systems were the most frequently assessed. This may be
etc.). Section 5 of this review summarizes and discusses the consulted explained by the relatively wider range and volume of agricultural crops
articles based on their decision types and OR methods (i.e., DEA, MP, EA produced compared to livestock products. The agricultural crops
and NN). Based on the temporal scale and decision-makers, decisions considered include a diverse range of perishable and non-perishable
can be categorized under three types: operational, tactical and strategic crops such as cereals (Galán-Martín et al., 2017; Masuda, 2016;
(Le Gal et al., 2011). Operational decisions refer to short-term (daily or Nabavi-Pelesaraei et al., 2016), vegetables and melons (Khoshnevisan
seasonal) decisions that are usually made by farmers and agricultural et al., 2015), fruits and nuts (Fleskens and de Graaff, 2010), oilseed
agencies at local scales. Tactical decisions target medium-term goals and crops (Mohammadi et al., 2013; Mousavi-Avval et al., 2017), root crops
are taken by farmers and institutions at regional scales. Strategic de­ (Khoshnevisan et al., 2014; Khoshnevisan et al., 2013b), beverage and
cisions, which are also known as structural decisions, are long-term spice crops (Gatimbu et al., 2020; Khanali et al., 2017), sugar crops (De
decisions that are made by provincial, state and national agencies (Le Oliveira Florentino and Pato, 2014; Kaab et al., 2019a, 2019b), and fiber
Gal et al., 2011; Robert et al., 2018). Each decision type can be sup­ crops (Ullah et al., 2016; Ullah and Perret, 2014).
ported by different OR methods. Based on a synthesis of results from
Review Questions 1 and 2 we hence derive a generalized decision tree,
along with method-specific methodological choices, to support practi­
tioners in identifying best-fit OR methods for optimization of environ­
mental outcomes in crop-livestock systems (Review Question 3).

3. Descriptive analysis of the review database

In order to descriptively analyze trends in publication of relevant


literature (Review Question 1), article publication rate over time was
first determined (Fig. 2), and the foci of the publications were identified
via text mining. Case study types and spatial scale of the studies were
tallied as well.
As the distribution of articles published over time (Fig. 2) shows,
integration of OR methods into LCAs of agricultural and livestock pro­
duction systems has increased over the last five years (i.e., 52% of ar­
ticles). The high rate of publication in 2014 evinces a focus among
diverse researchers on using data envelopment analysis to improve the
Fig. 3. Distribution of papers by crop-livestock type.

Fig. 2. Search results by publication year.

4
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

The spatial scope of the reviewed articles is depicted in Fig. 4. In this


context, spatial scale refers to the area from which inventory data is
collected. Farm scale refers to articles that analyzed several variables
from a single farm, while region scale includes those studies analyzing
several farms within a specific region (e.g., a province) of a country.
Finally, country scale refers to studies that sampled data from several
regions of a country and are represented as a national average. The vast
majority of studies were performed at regional or national levels of
spatial resolution, suggesting the importance of using data from multiple
farms in most optimization modelling approaches.
Fig. 5 depicts the range and frequency of OR approaches that were
employed in the sustainability assessment of crop-livestock production
systems. The subsequent section discusses included articles on the basis
of each OR method. DEA has been most frequently applied, followed
closely be mathematical programming. Application of EIs and NNs have
been, comparatively, less frequently applied.

4. Application of operations research methods in environmental


assessment of crop-livestock systems
Fig. 5. Distribution of papers by operational research method; DEA: Data
Subsequent sections will discuss the articles that used OR methods
Envelopment Analysis; LP: Linear Programming; MIP: Mixed-Integer Program­
including DEA (Section 4.1), Evolutionary Algorithms (SECTION 4.2), ming; NLP: Non-Linear Programming; EA: Evolutionary Algorithm.
Mathematical Programming (Section 4.3) and other methods (Section
4.4).
levels (Cooper and Seiford, 2007). The majority of studies reviewed used
input-oriented models. This is likely the best choice for modelling crop-
4.1. Data envelopment analysis livestock production systems, as farmers have more control over their
inputs to, rather than the outputs from, their farm. Only seven of the
Data Envelopment Analysis (DEA) is a non-parametric method that studies considered employed output-oriented models (Berre et al., 2013;
has been frequently applied to measure eco-efficiency at the farm-level. Gerdessen and Pascucci, 2013; Gomes et al., 2015; Piot-Lepetit, 2014;
Table 1 summarizes the information extracted from literature reporting Rebolledo-Leiva et al., 2017; Reinhard et al., 2000; Rosano-Peña et al.,
application of DEA in environmental assessment of crop-livestock 2014).
systems. DEA models may also be classified based on the returns-to-scale
DEA is non-parametric statistical tool used to measure the relative model employed, which may be either Constant Returns to Scale
efficiency of similar Decision Making Units (DMU)s (Cooper and Seiford, (CRS), or Variable Returns to Scale (VRS). CRS models, also known as
2007) within a homogeneous economic sector - for example, farms that Charnes, Cooper, and Rhodes or CCR models, assume that increases in
use similar inputs and product similar products or services - and to input levels lead to proportional increases in outputs. These types of
determine a set of best performing DMUs as benchmarks from an models calculate technical efficiency, and operate under the assumption
operational perspective. DEA employs a frontier analysis that takes all that DMUs operate in perfect competition with one another and are
inputs and outputs into account and converts them to a single measure operating at optimum capacity. This, however, is often not the case in
called efficiency (Cooper and Seiford, 2007). reality due to several factors including economic (Berre et al., 2013;
The main objective of most DEA studies is to determine the most Hoang and Alauddin, 2012) and technical constraints (Piot-Lepetit,
efficient DMUs among those sampled based on those using the fewest 2014), climatic conditions, etc.
inputs or producing the most (or least) desirable (or undesirable) out­ In contrast, VRS models, also known as Banker, Charnes, and Cooper
puts. DEA models can be classified as input-oriented or output-oriented or BCC models, assume disproportionate increases in outputs following
models, where input-oriented models focus on improvement of effi­ increases in inputs (Cooper and Seiford, 2007). These models are used to
ciency by minimizing inputs assuming constant outputs, while output- calculate pure technical efficiency. Pure technical efficiency shows the
oriented models aim to maximize outputs assuming constant input way that production unit resources are managed. Since increases in
input level usually do not result in proportionate increases in output
level (i.e., constant return to scale) pure technical efficiency (i.e., vari­
able return to scale) is generally better suited for modelling the realities
of DMUs (Berre et al., 2013; Gomes et al., 2015; Reinhard et al., 2000).
In order to maximize the discriminatory power of DEA to differen­
tiate between efficient and inefficient DMUs, the ratio between number
of DMUs and inputs and outputs included should not be low (Sanjuan
et al., 2011). There are several rules of thumb for determining the
minimum number of DMUs given the number of inputs and outputs
included in the analysis. Dyson et al. (2001) suggest that the number of
DMUs must be at least two times the sum of the inputs and outputs (the
consulted literature generally followed this rule, although is was not
explicitly stated that the rule informed the choice of sample size). Others
suggest that the number of DMUs be at least three times the sum of the
inputs and outputs (Bowlin, 1998). While there are arguments that
larger populations enable discerning the efficiency frontier more accu­
rately, others counter that large data sets may actually decrease con­
sistency and add the effects of unrelated impacts to the results (Sarkis,
Fig. 4. Distribution of papers by spatial scope.

5
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

Table 1
Analysis of articles employing DEA.
Reference Scope Case study DMUs Inputs Outputs Input- Output- CRS1 VRS2 Integration approach
oriented oriented

(Beltrán-Esteve et al., 2017) Country Crop 200 9 2 n.a3 n.a n.a n.a Five step joint DEA-LCA
(Berre et al., 2013) Region Livestock 51 4 3 ✓ ✓ –
(Ebrahimi and Salehi, 2015) Region Crop 22 7 1 ✓ ✓ ✓ –
(Esfahani et al., 2017) Region Crop 42 4 1 ✓ ✓ ✓ –
(Gatimbu et al., 2020) Region Crop 54 5 3 ✓ ✓ –
(Gerdessen and Pascucci, 2013) Region Crop 252 8 4 ✓ ✓ ✓ –
(Gomes et al., 2015) Country Livestock 21 6 1 ✓ ✓ –
(Grados and Schrevens, 2019) Region Crop 58 5 1 ✓ ✓ ✓ –
(Hoang and Alauddin, 2012) Country Crop 29 3 3 ✓ ✓ –
(Hosseinzadeh-Bandbafha et al., Region Crop 120 6 1 ✓ ✓ ✓ –
2018)
(Houshyar et al., 2019) Region Crop 278 7 1 ✓ ✓ –
(Iribarren et al., 2010) Region Livestock 19 3 1 ✓ ✓ Five step joint DEA-LCA
(Iribarren et al., 2011) Region Livestock 72 8 3 ✓ ✓ Five step joint DEA-LCA
(Van Meensel et al., 2010) Country Livestock 117 11 1 ✓ ✓ ✓ –
(Kaab et al., 2019a) Region Crop 92 7–10 1 ✓ ✓ ✓ Five step joint DEA-LCA
(Khoshnevisan et al., 2013a) Region Crop 260 9 1 ✓ ✓ ✓ –
(Masuda, 2016) Region Crop 36 2 1 ✓ ✓ –
(Masuda, 2019) Country Crop 36 2 1 ✓ ✓ –
(Mohammadi et al., 2013) Region Crop 94 9 1 ✓ ✓ Five step joint DEA-LCA
(Mohseni et al., 2018) Region Crop 58 11 1 ✓ ✓ ✓ Five step joint DEA-LCA
(Nabavi-Pelesaraei et al., 2016) Region Crop 39 7 1 ✓ ✓ –
(Nabavi-Pelesaraei et al., 2017) Region Crop 240 7 1 ✓ ✓ ✓ Five step joint DEA-LCA
(Paramesh et al., 2018) Region Crop 70 10 1 ✓ ✓ ✓ Three step joint DEA-
LCA
(Payandeh et al., 2017) Region Livestock 90 8–11 2 ✓ ✓ ✓ Eight step joint DEA-LCA
(Piot-Lepetit, 2014) Country Livestock 90 12 3 ✓ ✓ ✓ –
(Rebolledo-Leiva et al., 2017) Region Crop 14 6 1 ✓ ✓ ✓ Three step joint DEA-
LCA
(Reinhard et al., 2000) Country Livestock 613 6 1 ✓ ✓ ✓ –
(Rosano-Peña et al., 2014) Country Crop 33 3 3 ✓ ✓ ✓ –
(Serra et al., 2014) Region Crop 190 6 3 ✓ ✓ –
(Skevas and Serra, 2016) Country Crop 132 6 2 ✓ ✓ –
(Skevas et al., 2012) Country Livestock 188 8 1 ✓ ✓ –
(Skevas et al., 2014) Country Crop 119 7 1 ✓ ✓ –
(Soteriades et al., 2016) Region Livestock 185 5 3 ✓ –
(Syp et al., 2015) Region Crop 55 7 1 ✓ ✓ Five step joint DEA-LCA
(Toma et al., 2017) Country Crop 21 5 1 ✓ ✓ ✓ –
(Ullah and Perret, 2014) Region Crop 200 7 3 ✓ ✓ –
(Ullah et al., 2016) Region Crop 169 8 1 ✓ ✓ Three step joint DEA-
LCA
(Vázquez-Rowe et al., 2012) Region Crop 40 10 1 ✓ ✓ Five step joint DEA-LCA
1
CRS: Constant Returns to Scale.
2
VRS: Variable Returns to Scale.

2002). Determining which position is more correct is, however, beyond 4) Analyzing the environmental impacts of the target virtual DMUs
the scope of the current review. using the associated, hypothetical LCIs. This stage results in esti­
Benchmarking is one of the main applications of joint LCA and DEA mation of the potential reductions in environmental impacts asso­
studies. It is used to improve the efficiency of inefficient DMUs by ciated with the target DMUs.
providing target efficiencies and identifying strategies to realize them. 5) This final step aims to verify calculated eco-efficiency by comparing
This can subsequently lead to reductions in environmental impacts, as the environmental impacts of the existing DMUs with the potential
quantified by LCA, as inefficient DMUs become more efficient. Addi­ environmental impacts of the target DMUs.
tionally, DEA provides users the ability to prioritize specific inputs and
outputs by assigning weights to them (Vázquez-Rowe et al., 2010). Joint There is an alternative approach for joint DEA and LCA application
DEA and LCA application can be conducted in a five-step approach that entails three steps only (Lozano et al., 2010; Paramesh et al., 2018).
(Iribarren et al., 2011; Iribarren et al., 2010; Mohammadi et al., 2013; This approach employs the same first two steps as the five-step
Vázquez-Rowe et al., 2010) that includes: approach); however, the third step differentiates it from the five-step
approach. The third step involves use of the LCI from the first step and
1) Data collection and creating life cycle inventories for individual the characterization factors from the second step to create a DEA matrix
DMUs; in order to calculate the eco-efficiency of individual DMUs. This, in turn,
2) Analyzing environmental impacts (LCIA stage) of individual DMUs; provides an estimation of potential environmental impacts and LCIs for
3) Applying DEA, considering the most relevant inputs and outputs target DMUs.
(choice of user). This stage results in measurements of the eco- DEA is the most widely used benchmarking approach (Galindro
efficiency of individual DMUs and the determination of the most et al., 2019). Joint DEA-LCA application at regional scale has been
efficient DMUs as benchmarks. Target DMUs, virtual DMUs that uses promoted in Europe (European Union, 2013), in particular when there is
fewer resources and produce more outputs, will be formulated at this a large number of farms with similar sets of inputs and outputs. There
stage. are, however, several drawbacks associated with DEA modelling. The
quality of results in DEA largely relies on the modelling choices that the

6
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

researcher makes (i.e., return to scale, and efficiency type) (Wojcik et al., and life cycle impact assessment phases. Depending on the goal of the
2019). In some studies, CCR models are frequently applied but result in study, objective functions are defined by an aggregated single environ­
high variances in the efficiency results, while the choice of return to mental impact or by multi-objective mathematical programming. Con­
scale has not been well explored (Wojcik et al., 2019). straints are defined on the basis of all activities from extraction of raw
materials to their end of life stage (Caldeira et al., 2019); (Tascione et al.,
4.1.1. Linear/non-linear/mixed integer programming 2014). Mathematical modelling using LP, MIP and NLP techniques often
Linear programming (LP), non-linear programming (NLP) and include several constraints (i.e., economic, environmental, and tech­
mixed-integer programming (MIP) are three powerful mathematical nical). Table 2 summarizes the information extracted from literature
programming (MP) techniques. They are also known as mathematical reporting application of linear/non-linear/mixed integer programming
optimization techniques. These three techniques can be combined with in environmental assessment of crop-livestock systems.
LCA to optimize objective function(s) with respect to a set of specific LP is the most prevalent technique used for optimization of envi­
constraints, and hence to identify a set of optimum solutions for ronmental goals in crop-livestock production systems (13 of 27 studies),
improving the environmental performance of a product system. MP followed by NLP (8 of 27 studies) and MIP (6 of 27 studies). While there
approaches can be integrated with LCA at both the life cycle inventory is a high level of uncertainty (arising from uncertain data points such as

Table 2
Analysis of articles employing mathematical programming.
Authors Scope Case Objectives Formulation Decision Constraint
study

(Annetts and Audsley, Country Crop Minimize the environmental outcome; maximize the profit LP1 Strategic Nutritional;
2002) economic
(Behera et al., 2015) Region Livestock Maximize net profit; minimize capital requirements minimize labor LP2 Operational; Economic
employment tactical
2
(Capitanescu et al., Country Crop Maximize farmer profit MIP Strategic Economic;
2017) environment
3
(Cobuloglu and Region Crop Maximize total benefit maximize environmental benefits NLP Tactical; Economic;
Büyüktahtakin, strategic environment;
2015) technical
(Cobuloglu and Region Crop Maximize total economic value MIP Operational; Economic; technical
Büyüktahtakin, tactical
2014)
(Cortignani and Region Crop Maximize land allocation maximize profit minimize water use LP Tactical; Economic
Severini, 2012) strategic
(Dowson et al., 2019) Region Livestock Maximize total profit NLP Tactical Nutritional
(Fasakhodi et al., Region Crop Maximize net return/water consumption maximize labor NLP Tactical Economic
2010) employment/water consumption
(Fleskens and de Region Crop Maximize gross margin maximize total labour input maximize LP Tactical Economic
Graaff, 2010) landscape value minimize water use minimize soil erosion minimize
burnt area maximize biodiversity value minimize pollution
(Galán-Martín et al., Region Crop Maximize crop production minimize environmental impacts LP Tactical Economic;
2017) environment
(Gebrezgabher et al., Country Livestock Maximize gross margin minimize GHG emissions, ammonia MIP Tactical Economic;
2014) emissions and land use change environment
(Giasson et al., 2002) Farm Livestock Minimize phosphorous index minimize cost NLP Operational Nutritional;
economic; technical
(Hassani et al., 2019) Region Livestock Maximize resilience maximize sustainability NLP Tactical Economic
(Huang et al., 2012) Region Crop Maximize long-term economic benefits minimize environmental NLP Tactical Economic
disbenefits
(Kang et al., 2020) Region Crop Maximize crop production LP Tactical; Economic
strategic
(Li et al., 2020) Region Crop Maximize net benefit minimize environmental impacts minimize MIP Tactical Economic;
social impacts environment;
technical
(Ma et al., 2018) Farm Crop Minimize quality degradation minimize cost MIP Tactical; Environment;
strategic technical
(Mansoori et al., 2009) Farm Crop Maximize farmer’s welfare minimize environmental burden LP Operational Economic
(Montazar, 2013) Region Crop Maximize net annual benefits NLP Tactical Economic
(von Ow et al., 2020) Country Crop Minimize aggregated environmental impacts LP Tactical Nutritional;
economic
(Rǎdulescu et al., Region Crop Minimize environmental risk maximize expected return minimize MIP Tactical Economic;
2014) financial risk environment
(Rohmer et al., 2019) Country Livestock Minimize environmental impacts minimize total cost LP Strategic Economic;
environment
(Viaggi et al., 2009) Region Crop Maximize profit LP Operational Economic; technical
(Xavier and Freitas, Region Crop Optimize 12 sustainability (social, economic and environmental) LP Operational –
2018) indicators
(Yuan et al., 2018) Region Crop Minimize environmental impacts LP Strategic Economic; technical
(Yuanyuan, 2020) Region Crop Maximize gross domestic product minimize environmental impacts NLP Tactical; Economic
strategic
(Zimmermann, 2008) Farm Livestock Maximize income maximize sustainability index LP Tactical; Economic
strategic
1
LP: Linear Programming.
2
MIP: Mixed-integer Linear Programming.
3
NLP: Non-Linear Programming.

7
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

revenue of farms and crop yields) with LP models when they are inte­ tactical decisions.
grated with LCA models, the simplicity of modelling and flexibility in Similar to the OR methods considered, mathematical programming
working with different decision variables make it the most popular has been mostly employed for crop production (20 of 27 studies). This
choice among MP models. LP models have been used to address diverse may be due to the high importance of cereal crops such as corn, wheat
decisions such as crop rotation, farm technology selection, economic and rice. Mathematical programming has been mostly employed for
decisions, optimization of cropping patterns and resource use manage­ regional-scale issues (18 of 27 studies), where objective functions target
ment between crop and livestock farms, resource allocation, and a sample of farms in a specific region.
simultaneously maximizing crop productivity while minimizing envi­ One of the limitations of LP models is high uncertainty in its pa­
ronmental impacts (Galán-Martín et al., 2017). rameters (Huang et al., 2012). In contrast, Stochastic Programming (SP)
The majority of the articles reviewed performed optimization across models can deal with uncertainty in objective functions - normally as
at least two objectives, however the optimization of several objectives probability distribution functions known as random (stochastic) pa­
was not applied simultaneously. Generally, they looked to maximize rameters (Huang et al., 2012). Application of mathematical program­
economic benefits for farms while minimizing environmental impacts. ming models is also sometimes criticized for limited stakeholder
Objective functions related to environmental impacts were mostly involvement in the modelling process, and for considering single aspects
emissions-based, focusing in particular on reducing GHG emissions in defining objective functions such as land use, irrigation, or farm
(Gebrezgabher et al., 2014). Economic objective functions usually seek technology (Udias et al., 2018). In order to overcome the drawbacks of
to maximize the difference between farm revenues and input costs. single objective LP models, multi-objective optimization models using
With respect to parameter uncertainty, deterministic, stochastic or LP, MIP and NLP models take several management aspects of a farm into
robust modelling approaches may be used (Jornada and Leon, 2016). account.
Deterministic modelling refers to the use of classical optimization al­
gorithms that are mainly dependent on linear algebra. When model 4.2. Evolutionary algorithms
parameters are deterministic, LP and MIP models are typically used
(Kong et al., 2019). When parameters are stochastic (Huang et al., Evolutionary algorithms (EAs) are potentially powerful tools for
2012), stochastic programming approaches are used. multi-objective optimization. Generally, the solution finding methods
Stochastic optimization including genetic algorithm, particle swarm used by EAs may be classified as elitist, or non-elitist. Both Pareto’s
optimization and other evolutionary algorithms refers to optimization frontier, and non-dominate sorting genetic algorithms (NSGA-II) are
modelling approaches where their objective functions (i.e., maximizing/ examples of elitist solution finding methods, in which multi-objective
minimizing) deal with random variables (i.e., probability space) (Huang problems are solved such that no solution for any single objective
et al., 2012). Robust optimizations are newer approaches that assume function dominates (i.e. is preferred) over the solution for any other
the parameters belong to deterministic variables (Maggioni et al., 2015). objective function. In contrast, non-elitist methods allow for potential
Depending on the type of products, several decision variables may be dominance of solutions over one another (Yusoff et al., 2011).
used to formulate the objective functions. These variables may be EAs can be easily customized depending on the studied problem. The
management-based (i.e. land allocation, time of seed sowing, types of two main approaches used for searching for solutions are genetic algo­
crop protection products and fertilizers, irrigation etc.), or resource- rithms (GAs), and particle swarm optimization (PSOs). GAs are a type of
based (i.e. seed, water, fertilizer, crop protection, and energy input search algorithm that work according to the principles of natural se­
amounts, yield, etc.). lection and evolution. These algorithms use probabilistic rules for pa­
The decisions supported by these analyses may address short-, me­ rameters, meaning it is a stochastic search algorithm. PSOs are another
dium-, or long-term issues. They may also be classified as operational, type of evolutionary algorithm that imitate swarm behavior of birds to
tactical, or strategic decisions. According to (Risbey et al., 1999), short- search among solution spaces. Table 3 summarizes the information
term decisions (i.e., operational) address seasonal decisions (< one extracted from literature reporting application of evolutionary algo­
year), medium-term decisions (i.e., tactical) cover multiple year (1–5 rithms in environmental assessment of crop-livestock systems.
years) decisions, and strategic decisions address long-term (> five years) Multi-objective EAs using GAs and PSOs have been applied to opti­
decisions. However, distinguishing between decisions on the basis of mization of many objectives in crop-livestock research, including
temporal scale is not always clear. maximizing farm profits, energy outputs, benefit/cost ratios, crop pro­
Strategic decisions typically refer to long-term issues such as de­ ductivity, and minimizing environmental impacts, farm costs, and en­
cisions on crop rotations, selection of optimum crop mix in multi-crop ergy surplus (Groot et al., 2012; Maiyar and Thakkar, 2019; Pastori,
systems, selection of farm equipment, selection of mechanization level 2017; Pishgar-Komleh et al., 2019). The majority of the consulted works,
of farms, monetary planning, reservoir management, and planning lo­ however, used elitist methods – in particular, genetic algorithms.
gistics for harvested products. Objective functions for these models are EA-based studies cover a wide range of decision contexts, including
traditionally presented as economic goals such as maximizing farm resource use management (Barak et al., 2016; López-Andrés et al., 2018;
utility, revenue, net present value, and profit, or minimizing input cost. Pishgar-Komleh et al., 2019; Shamshirband et al., 2015; Villalba et al.,
They may also be environmental goals such as minimizing land use, 2019), farm management practices (Breen et al., 2019), transportation
water use, GHG emissions or other environmental impacts. and logistics (Maiyar and Thakkar, 2019; Wang et al., 2018), and crop
Tactical decisions address issues in short to medium term planning selection (De Oliveira Florentino and Pato, 2014). Economic constraints
on farms such as water allocation, required labor, harvesting schedule are the most common constraint utilized; however, some studies
(He et al., 2018), or post-harvesting activities (Chandrasekaran and employed environmental (Chandrasekaran and Ranganathan, 2017;
Ranganathan, 2017). Objective functions associated with tactical Pastori, 2017; Shamshirband et al., 2015) and nutritional constraints
models are traditionally accompanied by an economic goal such as (De Oliveira Florentino and Pato, 2014) as well.
minimizing cost of inputs or maximizing farm profits, as well. Opera­ Integration of EAs with LCA usually proceeds obtaining the results of
tional decisions, which have been less frequently addressed than tactical an LCA study. The decision variables of joint LCA-EA models are typi­
decisions, deal with short term (day-to-day) issues and farm-scale cally the farm inputs (e.g., fuel consumption, seed, feed for livestock,
planning, such as scheduling of land preparation, mix and amount of energy consumption), while environmental impacts are entered as out­
farm inputs, harvest scheduling (Xie et al., 2018), and storage planning. puts. Objective functions of joint LCA-EA models are minimizing indi­
From the modelling point of view, some studies address both strategic vidual environmental impacts or aggregated impacts. Evolutionary
and tactical decisions, such as analyzing the performance of alternative algorithms (i.e., GA and PSO in this review) are then used to generate
farming practices (Yuanyuan, 2020), or combine operational and optimal alternatives (López-Andrés et al., 2018; Maiyar and Thakkar,

8
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

Table 3
Analysis of articles employing evolutionary algorithms.
Authors Scope Case Objectives Formulation name Decision variables Decision Constraint
study

(Pishgar-Komleh Region Crop Maximize economic output; MOGA NSGA-II Resource use management Operational; Economic
et al., 2019) minimize environmental using Pareto front; tactical
impacts PSO; ANN
(Breen et al., 2019) Farm Livestock Maximize annual net profit; MOGA with Eleven decision variables Operational; Economic
minimize annual farm CO2 weighting factors related to farm equipment; tactical
emissions management practices;
electricity tariffs
(Maiyar and Thakkar, Region Crop Minimize total cost PSO using Pareto Shipment quantity; route Tactical; Economic
2019) plots selection; hub location strategic
(Villalba et al., 2019) Farm Livestock Maximize financial margin, MOGA; NSGA-II Resource use management; Operational; Economic
minimize nitrogen; minimize feeding management; tactical
energy surplus reproduction management
(López-Andrés et al., Farm Livestock Maximize economic benefits; MOGA; Pareto front Resource use management Operational; Economic
2018) minimize environmental tactical
impacts
(Pastori, 2017) Country Crop Maximize farmer’s income; MOGA; Pareto front Fertilization and irrigation Tactical; Economic;
minimize nitrogen leaching practices strategic environment
(Mousavi-Avval et al., Region Crop Maximize output energy; MOGA; NSGA-II Resource use management Operational; Economic
2017) minimize environmental tactical
emissions; maximize benefit to
cost ratio
(Chandrasekaran and Country Crop Minimize total cost; maximize GA; constraint- Carbon dioxide loss at each Tactical Economic;
Ranganathan, total supply; minimize total dependent crossover stage of supply chain environment
2017) CO2 emission and mutation
(Barak et al., 2016) Region Crop Minimize GHG emissions; MOPSO; MOGA Energy use management Operational; Economic
maximize output energy; NSGA-II; NRGA; tactical
maximize benefit cost ratio Pareto front
(Shamshirband et al., Region Crop Minimize GHG emissions; MOGA; Pareto front Energy use management Operational; Environment
2015) maximize output energy; tactical
maximize benefit cost ratio
(Khoshnevisan et al., Region Crop Minimize environmental MOGA; Pareto front Resource use management Operational; Economic
2015) impacts tactical
(De Oliveira Region Crop Minimize collection and Bi-objective GA; Selection of crop varieties Tactical; Nutritional;
Florentino and transport costs; maximize Pareto front strategic economic
Pato, 2014) energy balance

MOGA: Multi-Objective Genetic Algorithm; NSGA-II: Non-dominated Sorting Genetic Algorithm II; PSO: Particle Swarm Optimization.

2019). Advantages of EAs include that they are conceptually easy to layer (input layer), one hidden layer with eight neurons and two outputs
understand and flexible in application, support multi-objective optimi­ (for example, two environmental impacts) in the output layer.
zation, and use stochastic optimization. The principle drawback of joint Feedforward NNs are the simplest types of NNs. They have no cyclic
EA-LCA is, however, the lack of a standardized integration method. How node connections, and they work in one direction. Inputs are supplied
to determine the stop point (i.e. where the global optimum is achieved) into the input layer and delivered from the output layer. The only in­
is also unclear in GAs (Sahoo et al., 2017; Sarker and Ray, 2009). ternal links in feedforward NNs are the links between weights.
Depending on the outputs of the analyzed data, the learning algo­
rithms of NNs can be supervised or unsupervised. If the desired output is
4.3. Neural network
known in the model, the learning algorithm is supervised. NN models
that are used for prediction and regression problems are supervised
An NN is a computing system consisting of a set of connected nodes
(Liakos et al., 2018). Unsupervised networks have no target outputs and
that are associated with a set of determinant weights. NNs are intended
their learning algorithm is different from supervised networks. The main
to simulate the action of human brain and train nodes. The nodes of an
application of unsupervised networks is clustering (i.e., looking for
NN are independent of each other, and thus act individually based on
natural groupings of data). Therefore, they are not further discussed in
their particular inputs and associated weights (Russell and Norvig,
this paper, but interested readers can refer to (Längkvist et al., 2014). In
2002). Weights determine the impact of connection among nodes. Initial
order to develop algorithms in supervised networks, it is necessary to
weights are determined by the NN, and these weights are subsequently
assign a set of samples (i.e., each individual farm is a sample) to train the
updated by learning algorithms.
NN.
There are at least three layers in each NN: the input layer, hidden
Table 4 summarizes the information extracted from literature
layer, and output layer. The input layer consists of the initial data fed
reporting application of NNs in environmental assessment of crop-
into the system, while the output layer consists of the prediction results.
livestock systems.
Hidden layers located between the input and output layer consist of a set
Modelling and prediction of farm inputs and outputs are the main
of neurons that have the responsibility for all computation tasks. Larger
application of NNs in crop-livestock production systems. In contrast with
numbers of hidden layers will result in more complex networks and
linear regression modelling, NNs can predict several outputs at a time.
longer modelling processes. Weights are parameters that are applied
They have been variously applied to modelling energy output (Elahi
within the nodes (i.e., neurons) of the hidden layers.
et al., 2019; Kaab et al., 2019b; Khoshnevisan et al., 2014; Khoshnevisan
The architecture of an NN is described by the number of hidden
et al., 2013c; Nabavi-Pelesaraei et al., 2018), farm yields (Khanali et al.,
layers and number of neurons in each layer. The first and last numbers
2017), GHG emissions (Khoshnevisan et al., 2014; Khoshnevisan et al.,
show input and output layers, respectively. Middle values show the
2013c), and other environmental impacts (Bonfiglio, 2011; Kaab et al.,
number of neurons in the hidden layers. An NN with these features:
2019b; Khanali et al., 2017; Khoshnevisan et al., 2013b; Nabavi-
8–8–2 (Bonfiglio, 2011), for example, comprises eight inputs in the first

9
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

Table 4
Analysis of articles employing NN.
Authors Scope Case Functionality Decision variables Decision Network architecture Performance
study measures

(Kaab et al., 2019b) Region Crop Prediction of energy Output energy; Tactical; Back-propagation feed-forward; R2; MAPE; RMSE
and environmental environmental strategic layers 9–10 - 5 - 11; 7–9 - 6 - 11; 70%
impacts impacts training 15% test 15% validation
(Elahi et al., 2019) Region Livestock Prediction of renewable Energy use on farms Tactical; Back-propagation feed-forward; R2; MSE
and non-renewable strategic layers 1–10 - 9
energy
(Nabavi-Pelesaraei Region Crop Prediction of energy Output energy; Tactical; Back-propagation feed-forward; R; RMSE; MAPE
et al., 2018) and environmental environmental strategic layers 12–6 - 8 - 1; 70% training 15%
impacts impacts test 15% validation
(Khanali et al., 2017) Region Crop Modelling of yield and Yield; environmental Tactical; Back-propagation feed-forward; R2; RMSE; MAPE
environmental impacts impacts strategic layers 8–10 - 11 -11; 8–8 - 10 -11;
8–12 - 7 - 11; 70% training 15% test
15% validation
(Khoshnevisan et al., Region Crop Prediction of output Output energy; GHG Tactical; Back-propagation feed-forward; R2; RMSE; MAE
2014) energy and GHG emissions strategic layers 12–8 - 2; 60% training 15% test
emissions 25% validation
(Khoshnevisan et al., Region Crop Prediction of energy use Energy use; GHG Tactical; Back-propagation feed-forward; R2; MSE; MRE
2013a, 2013b, and GHG emissions emissions strategic layers 11–3 - 2; 80% training 20% test
2013c)
(Khoshnevisan et al., Region Crop Prediction of Environmental Tactical; Back-propagation feed-forward; R2; RMSE; MAE
2013a, 2013b, environmental impacts impacts strategic layers 11–10 - 6; 60% training 25%
2013c) test 15% validation
(Bonfiglio, 2011) Region Crop Prediction of Environmental Tactical; Back-propagation feed-forward; R2; percentage of
environmental impacts impacts strategic layers 8–8 - 2; 70% training 30% predicted negative
validation values

MAPE: Mean Absolute Percentage Error; RMSE: Root Mean Square Error; MSE: Mean Squared Error; MAE: Mean Absolute Error; MRE: Mean Relative Error

Pelesaraei et al., 2018) in crop-livestock systems. primarily responsible for operational and tactical decisions (Carravilla
Modelling with an NN requires first defining the set of inputs and and Oliveira, 2013). Table 5 summarizes the three decision types (i.e.,
output(s). In the case of integrating an NN with LCA, the amounts of strategic, tactical, and operational) represented in the reviewed studies,
farm inputs are usually modelled as the input layer whereas life cycle and the OR methods applied (i.e., DEA, MP, EA, and NN).
environmental impacts are the output layer of the NN. Data collected Decisions using strategic models in the reviewed studies variously
data from farms must be broken down into three datasets including covered the whole supply chain of products, including production,
training, validation and test groups. Around 70–80% of data are used to processing and transportation aspects. Examples of such decisions
train the network, and the rest of the data are used for validation and include selection of crops, farm, infrastructure, and farm machinery
testing (Khoshnevisan et al., 2013c). size/level of mechanization (He et al., 2018). Benchmarking refers to
In order to train an NN, data from multiple farms should be utilized, identifying the most efficient farms, so that best practices may be
as was the case in all studies considered (i.e. regional scale). The per­ adapted and applied to other farms that are operating inefficiently
formance of an NN may be assessed using statistical measures such as the (Kahan, 2013). This is mainly implemented using DEA. Benchmarking
R and R2 values, and error indicators such as MAPE, RMSE, MSE, MAE using DEA is typically used to support mid-to-short term decisions (i.e.,
and MRE. These methods compare the outputs of the NN to the correct tactical and operational decisions) such as timing of farm operations like
outputs, as labelled in the test data. NNs have proven to be reliable in sowing, irrigation and harvesting. Prediction of farm outputs or envi­
dealing with large sets of variables. The capacity to analyze several ronmental impacts are the typical application of NNs. NNs support both
outputs at a time is another advantage of NNs, as is the relatively modest strategic and tactical decision making. Operational models (i.e.,
simulation time requirement (Sharif and Hammad, 2019). Similar to temporal-scale less than one year) are more important for decisions
EAs, the validity of the results from NNs is strongly conditioned by the impacting logistics for fresh and perishable products with a short shelf
choices that users make during modelling, but there is currently no life. Mathematical programming can be used with all decision types, but
consensus as to best practice for determining the appropriate structure was mainly applied for tactical and operational decisions in the studies
for NNs (Elahi et al., 2019; Kaab et al., 2019b). Rather, users rely on considered. Resource use management was the main application of MP
experimentation and checking the error indicators. and EA approaches and was used to support all three decision types
(operational, tactical and strategic) including type and amount of seed,
5. Best-fit OR methods for optimizing environmental outcomes fertilizers, crop protection products, water, fuel, and electricity
in crop-livestock systems consumption.
Considering the main applications of OR methods and the method­
This section summarizes the key features of OR methods with respect ological choices characteristic of each method a generalized decision
to their main applications for different decision types aimed at opti­ tree along with method-specific methodological choices is described in
mizing environmental outcomes in crop-livestock systems. Best-fit OR Fig. 6 to help users select best fit OR methods for optimization of envi­
approaches for specific decision types are identified. On this basis, a ronmental outcomes in crop-livestock production.
general framework in the format of a decision tree for selecting among Based on the nature of the methods and frequency of application as
OR methods is proposed. observed in the reviewed studies, the best-fit OR methods to support
Application of the reviewed OR methods to decision support in crop- optimization efforts in crop-livestock systems will generally be DEA for
livestock systems can be broadly categorized based on three decision benchmarking, NNs for prediction, and mathematical programming or
types: strategic, tactical and operational. Strategic decisions are usually evolutionary algorithms for resource management decision types.
made by governments and other national or global decision makers, Selecting inputs and outputs and choosing performance measures are
while other stakeholders including farmers and local farm managers are common methodological choices that are required for each OR method.

10
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

Table 5
Analysis of different OR methods in benchmarking, prediction, and resource use management decision support applications for crop-livestock
systems.

Fig. 6. A generalized decision tree for selecting best fit OR methods for optimizing environmental outcomes in crop-livestock production.

11
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

Error metrics such as MSE and MAE, and statistical measures such as the mathematical programming or evolutionary algorithms for resource
R and R2 values are typically used to measure the performance of NN management. The generalized decision tree proposed in Section 5 aims
and EI methods. These measures generally work under the concept of to help users to select the best-fit OR method for specific environmental
similarity and distance (Deza and Deza, 2016). In NNs, for example, optimization decision types in crop-livestock production, and to provide
these performance measures are used to compare the outputs of trained an indication of necessary method-specific methodological choices.
models with the actual outputs. For more on performance measures in Benchmarking problems are mostly addressed using DEA. Method­
OR methods, interested readers can refer to (Botchkarev, 2018). The ological choices for this approach include identifying the return to scale
compilation of decision variables is performed using selected inputs and (i.e., CRS, VRS) and modelling orientation (i.e., input-oriented, output-
outputs. Decision variables are then presented as objective functions for oriented). A challenging issue, however, is choosing the right sample
mathematical programming and evolutionary algorithm approaches and size (number of DMUs). Since there is no current consensus as to how to
as eco-efficiency objectives in the case of using benchmarking tech­ best determine the number of DMUs, further research in this area is
niques (e.g., DEA). recommended.
Prior to implementing selected OR methods, a number of key con­ Evolutionary algorithms (GA and PSO) and mathematical program­
siderations must be resolved including identifying the spatial scale and ming (LP, MIP, and NLP) are more suitable for resource use management
uncertainty aspects of the OR methods. The objectives of the study will (i.e., optimization) issues. The common methodological choices for
generally determine the spatial scale. As per Section 3 (Fig. 4), OR these two methods are defining decision variables and building objec­
methods may be implemented at three spatial scales: farm, region and tive functions. Farm inputs such as energy and seed consumptions are
country. Farm-level studies do not need to address the heterogeneity defined as decision variables, while environmental impacts are used as
among decision variables across different farms, as is required in the outputs of objective functions.
regional scale studies. The main advantages of farm-level models are the Supervised feedforward NNs are typically implemented to predict
ease of data collection, as data needs only to be collected from a single output (i.e., yield and environmental impacts) of farms. Determining the
farm. However, the outputs of a farm-level model cannot be applied at a appropriate structure for networks (i.e., number of hidden layers and
greater spatial scale due to potentially poor representativeness. number of neurons in each layer) is a key methodological consideration
Regional-scale studies are grounded in the assumption that the studied in this approach. Statistical measures such as the R and R2 values and
sample is representative of the studied population. At a larger scale, error indicators such as MSE and MAE are mainly applied to measure the
some studies randomly selected a sample of farms from several regions performance of NN and EI methods.
of a country to represent a national-average. These are known as
national-scale/country-level models. Macro-level models can reflect
national and global trends but are unable to show differences between Declaration of Competing Interest
farm-level decision variables.
As another key consideration, uncertainty and risk assessment are None.
essential elements in optimization modelling of agricultural systems, but
have been traditionally ignored in most model developments, in Acknowledgments
particular when LP models are employed (Dury et al., 2012). Agricul­
tural decisions are mainly affected by climate conditions and market This project is funded by the NSERC/Egg Farmers of Canada Indus­
prices (Han et al., 2019). Differences in these parameters result in trial Research Chair in Sustainability, with infrastructure funding from
additional model uncertainty (Nuthall, 2018). Stochastic modelling the Canada Foundation for Innovation John Evans Leaders Fund.
methods present promising approaches for addressing uncertainty in
agricultural decision-making contexts. Such approaches consider the References
possibility of measures over time, climate conditions, and market vari­
abilities, and sample values for decision variables accordingly. Some Aien, M., Rashidinejad, M., Fotuhi-Firuzabad, M., 2014. On possibilistic and probabilistic
uncertainty assessment of power flow problem: a review and a new approach.
optimization models, such as those developed by Dowson et al. (2019),
Renew. Sust. Energ. Rev. https://doi.org/10.1016/j.rser.2014.05.063.
do take into account uncertainty in input parameters and analyze its Ali, S.A., Tedone, L., De Mastro, G., 2015. Optimization of the environmental
impact on the model outputs. Objective functions of stochastic optimi­ performance of rainfed durum wheat by adjusting the management practices.
zation models employ scenario uncertainty (i.e., uncertainty due to J. Clean. Prod. 87, 105–118. https://doi.org/10.1016/j.jclepro.2014.09.029.
Annetts, J.E., Audsley, E., 2002. Multiple objective linear programming for
choices of probability distribution types) in their formulation. Integra­ environmental farm planning. J. Oper. Res. Soc. 53, 933–943. https://doi.org/
tion of uncertainty into these models provides solutions that are more 10.1057/palgrave.jors.2601404.
relevant to real world situations (Aien et al., 2014); however, doing so Arora, R.K., 2015. Optimization : Algorithms and Applications. Chapman and Hall/CRC.
Barak, S., Yousefi, M., Maghsoudlou, H., Jahangiri, S., 2016. Energy and GHG emissions
introduces a significant additional computational burden. Several management of agricultural systems using multi objective particle swarm
studies have addressed techniques to lessen computational pressure optimization algorithm: a case study. Stoch. Env. Res. Risk A. 30, 1167–1187.
rising from stochastic optimizations (Reddy et al., 2017). Compared to https://doi.org/10.1007/s00477-015-1098-1.
Behera, U.K., Kaechele, H., France, J., 2015. Integrated animal and cropping systems in
other OR methods, using stochastic optimization models in crop- single and multi-objective frameworks for enhancing the livelihood security of
livestock production systems is still relatively uncommon. farmers and agricultural sustainability in northern India. Anim. Prod. Sci. 55, 1338.
https://doi.org/10.1071/AN14526.
Beltrán-Esteve, M., Reig-Martínez, E., Estruch-Guitart, V., 2017. Assessing eco-efficiency:
6. Conclusions a metafrontier directional distance function approach using life cycle analysis.
Environ. Impact Assess. Rev. 63, 116–127. https://doi.org/10.1016/j.
To ensure net-positive outcomes from sustainability decision making eiar.2017.01.001.
Berre, D., Boussemart, J.P., Leleu, H., Tillard, E., 2013. Economic value of greenhouse
in crop-livestock production systems, it is important that stakeholders gases and nitrogen surpluses: society vs farmers’ valuation. Eur. J. Oper. Res. 226,
are able to make informed decisions based on the totality of the resource 325–331. https://doi.org/10.1016/j.ejor.2012.11.017.
burdens and environmental impacts of the systems they manage. The Bonfiglio, A., 2011. A neural network for evaluating environmental impact of decoupling
in rural systems. Comput. Environ. Urban. Syst. 35, 65–76. https://doi.org/10.1016/
aim of this study was to review and synthesize information regarding
j.compenvurbsys.2010.06.004.
operations research methods as applied to optimization or prediction of Boone, L., Van Linden, V., De Meester, S., Vandecasteele, B., Muylle, H., Roldán-Ruiz, I.,
environmental outcomes in crop-livestock systems. Nemecek, T., Dewulf, J., 2016. Environmental life cycle assessment of grain maize
A variety of models were found to support optimization efforts in production: an analysis of factors causing variability. Sci. Total Environ. 553,
551–564. https://doi.org/10.1016/j.scitotenv.2016.02.089.
crop-livestock systems, which were categorized on the basis of their Botchkarev, A., 2018. Performance Metrics (Error Measures) in Machine Learning
application: DEA for benchmarking; NNs for prediction; and Regression, Forecasting and Prognostics: Properties and Typology. Interdisciplinary

12
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

Journal of Information, Knowledge, and Management 14, 45–76. https://doi.org/ cross-compliance and Agri-environmental measures. Agric. Syst. 103, 521–534.
10.28945/4184. https://doi.org/10.1016/j.agsy.2010.05.005.
Bowlin, W.F., 1998. Measuring performance: an introduction to data envelopment Galán-Martín, Á., Vaskan, P., Antón, A., Esteller, L.J., Guillén-Gosálbez, G., 2017. Multi-
analysis (DEA). The Journal of Cost Analysis 15, 3–27. https://doi.org/10.1080/ objective optimization of rainfed and irrigated agricultural areas considering
08823871.1998.10462318. production and environmental criteria: a case study of wheat production in Spain.
Breen, M., Murphy, M.D., Upton, J., 2019. Development of a dairy multi-objective J. Clean. Prod. 140, 816–830. https://doi.org/10.1016/j.jclepro.2016.06.099.
optimization (DAIRYMOO) method for economic and environmental optimization of Galindro, B.M., Zanghelini, G.M., Soares, S.R., 2019. Use of benchmarking techniques to
dairy farms. Appl. Energy 242, 1697–1711. https://doi.org/10.1016/j. improve communication in life cycle assessment: a general review. J. Clean. Prod.
apenergy.2019.03.059. 213, 143–157. https://doi.org/10.1016/j.jclepro.2018.12.147.
Cabaneros, S.M., Calautit, J.K., Hughes, B.R., 2019. A review of artificial neural network Gatimbu, K.K., Ogada, M.J., Budambula, N.L.M., 2020. Environmental efficiency of
models for ambient air pollution prediction. Environ. Model. Softw. https://doi.org/ small-scale tea processors in Kenya: an inverse data envelopment analysis (DEA)
10.1016/j.envsoft.2019.06.014. approach. Environ. Dev. Sustain. 22, 3333–3345. https://doi.org/10.1007/s10668-
Caldeira, C., Freire, F., Olivetti, E.A., Kirchain, R., Dias, L.C., 2019. Analysis of cost- 019-00348-x.
environmental trade-offs in biodiesel production incorporating waste feedstocks: A Gebrezgabher, S.A., Meuwissen, M.P.M., Oude Lansink, A.G.J.M., 2014. A multiple
multi-objective programming approach. Journal of Cleaner Production 216, 64–73. criteria decision making approach to manure management systems in the
https://doi.org/10.1016/j.jclepro.2019.01.126. Netherlands. Eur. J. Oper. Res. 232, 643–653. https://doi.org/10.1016/j.
Capitanescu, F., Marvuglia, A., Navarrete Gutiérrez, T., Benetto, E., 2017. Multi-stage ejor.2013.08.006.
farm management optimization under environmental and crop rotation constraints. Gerdessen, J.C., Pascucci, S., 2013. Data envelopment analysis of sustainability
J. Clean. Prod. 147, 197–205. https://doi.org/10.1016/j.jclepro.2017.01.076. indicators of european agricultural systems at regional level. Agric. Syst. 118, 78–90.
Carravilla, M.A., Oliveira, J.F., 2013. Operations research in agriculture: better decisions https://doi.org/10.1016/j.agsy.2013.03.004.
for a scarce and uncertain world. Agris On-line Papers in Economics and Informatics Gerland, P., Raftery, A.E., Ševčíková, H., Li, N., Gu, D., Spoorenberg, T., Alkema, L.,
2, 37–46. Fosdick, B.K., Chunn, J., Lalic, N., Bay, G., Buettner, T., Heilig, G.K., Wilmoth, J.,
Chandrasekaran, M., Ranganathan, R., 2017. Modelling and optimisation of Indian 2014. World population stabilization unlikely this century. Science 346, 234–237.
traditional agriculture supply chain to reduce post-harvest loss and CO2 emission. https://doi.org/10.1126/SCIENCE.1257469.
Ind. Manag. Data Syst. 117, 1817–1841. https://doi.org/10.1108/IMDS-09-2016- Giasson, E., Bryant, R.B., Bills, N.L., 2002. Environmental and economic optimization of
0383. dairy manure management. Agron. J. 94, 757–766. https://doi.org/10.2134/
Clark, M.A., Domingo, N.G.G., Colgan, K., Thakrar, S.K., Tilman, D., Lynch, J., agronj2002.7570.
Azevedo, I.L., Hill, J.D., 2020. Global food system emissions could preclude Gomes, E.G., De Abreu, U.G.P., De Mello, J.C.C.B.S., De Carvalho, T.B., De Zen, S., 2015.
achieving the 1.5◦ and 2◦ C climate change targets. Science (New York, N.Y.) 370, Economic and socio-environmental performance assessment of beef cattle
705–708. https://doi.org/10.1126/science.aba7357. production systems: a data envelopment analysis (DEA) approach with weight
Cobuloglu, H.I., Büyüktahtakin, I.E., 2014. A mixed-integer optimization model for the restrictions. Rev. Bras. Zootec. 44, 219–225. https://doi.org/10.1590/S1806-
economic and environmental analysis of biomass production. Biomass Bioenergy 67, 92902015000600004.
8–23. https://doi.org/10.1016/j.biombioe.2014.03.025. Grados, D., Schrevens, E., 2019. Multidimensional analysis of environmental impacts
Cobuloglu, H.I., Büyüktahtakin, I.E., 2015. Food vs. biofuel: an optimization approach to from potato agricultural production in the Peruvian Central Andes. Sci. Total
the spatio-temporal analysis of land-use competition and environmental impacts. Environ. 663, 927–934. https://doi.org/10.1016/j.scitotenv.2019.01.414.
Appl. Energy 140, 418–434. https://doi.org/10.1016/j.apenergy.2014.11.080. Groot, J.C.J., Oomen, G.J.M., Rossing, W.A.H., 2012. Multi-objective optimization and
Conforti, P., 2011. Looking Ahead in worLd Food and agricuLture: Perspectives to 2050 design of farming systems. Agric. Syst. 110, 63–77. https://doi.org/10.1016/j.
Agricultural Development Economics Division Economic and Social Development agsy.2012.03.012.
Department. Guinée, J.B., Heijungs, R., Huppes, G., Zamagni, A., Masoni, P., Buonamici, R., Ekvall, T.,
Cooper, W.W., Seiford, L.M., 2007. Data Envelopment Analysis a Comprehensive Text Rydberg, T., 2011. Life cycle assessment: Past, present, and future. Environ. Sci.
with Models, Applications, References and DEA-Solver Software Second Edition. Technol. 45, 90–96. https://doi.org/10.1021/es101316v.
Springer. Han, E., Baethgen, W.E., Ines, A.V.M., Mer, F., Souza, J.S., Berterretche, M., Atunez, G.,
Cortignani, R., Severini, S., 2012. A constrained optimization model based on Barreira, C., 2019. SIMAGRI: An agro-climate decision support tool. Comput.
generalized maximum entropy to assess the impact of reforming agricultural policy Electron. Agric. 161, 241–251. https://doi.org/10.1016/j.compag.2018.06.034.
on the sustainability of irrigated areas. Agric. Econ. 43, 621–633. https://doi.org/ Hassani, L., Daneshvar Kakhki, M., Sabouhi Sabouni, M., Ghanbari, R., 2019. The
10.1111/j.1574-0862.2012.00608.x. optimization of resilience and sustainability using mathematical programming
De Oliveira Florentino, H., Pato, M.V., 2014. A bi-objective genetic approach for the models and metaheuristic algorithms. J. Clean. Prod. 228, 1062–1072. https://doi.
selection of sugarcane varieties to comply with environmental and economic org/10.1016/j.jclepro.2019.04.324.
requirements. J. Oper. Res. Soc. 65, 842–854. https://doi.org/10.1057/ Hauschild, M.Z., Rosenbaum, R.K., Olsen, S.I., 2017. Life Cycle Assessment: Theory and
jors.2013.21. Practice, Life Cycle Assessment: Theory and Practice. Springer International
Deza, M.M., Deza, E., 2016. Encyclopedia of Distances, Encyclopedia of Distances. Publishing. https://doi.org/10.1007/978-3-319-56475-3.
Springer, Berlin Heidelberg.. https://doi.org/10.1007/978-3-662-52844-0 He, P., Li, J., Wang, X., 2018. Wheat harvest schedule model for agricultural machinery
Dowson, O., Philpott, A., Mason, A., Downward, A., 2019. A multi-stage stochastic cooperatives considering fragmental farmlands. Comput. Electron. Agric. 145,
optimization model of a pastoral dairy farm. Eur. J. Oper. Res. 274, 1077–1089. 226–234. https://doi.org/10.1016/j.compag.2017.12.042.
https://doi.org/10.1016/j.ejor.2018.10.033. Hoang, V.N., Alauddin, M., 2012. Input-orientated data envelopment analysis framework
Dury, J., Schaller, N., Garcia, F., Reynaud, A., Bergez, J.E., 2012. Models to support for measuring and decomposing economic, environmental and ecological efficiency:
cropping plan and crop rotation decisions. A review. Agronomy for Sustainable an application to OECD agriculture. Environ. Resour. Econ. 51, 431–452. https://
Development. doi. https://doi.org/10.1007/s13593-011-0037-x. doi.org/10.1007/s10640-011-9506-6.
Dyson, R.G., Allen, R., Camanho, A.S., Podinovski, V.V., Sarrico, C.S., Shale, E.A., 2001. Hokazono, S., Hayashi, K., 2015. Life cycle assessment of organic paddy rotation systems
Pitfalls and protocols in DEA. Eur. J. Oper. Res. 132, 245–259. https://doi.org/ using land- and product-based indicators: a case study in Japan. Int. J. Life Cycle
10.1016/S0377-2217(00)00149-1. Assess. 20, 1061–1075. https://doi.org/10.1007/s11367-015-0906-7.
Ebrahimi, R., Salehi, M., 2015. Investigation of CO2 emission reduction and improving Hosseinzadeh-Bandbafha, H., Nabavi-Pelesaraei, A., Khanali, M., Ghahderijani, M.,
energy use efficiency of button mushroom production using data envelopment Chau, K. Wing, 2018. Application of data envelopment analysis approach for
analysis. J. Clean. Prod. 103, 112–119. https://doi.org/10.1016/j. optimization of energy use and reduction of greenhouse gas emission in peanut
jclepro.2014.02.032. production of Iran. J. Clean. Prod. 172, 1327–1335. https://doi.org/10.1016/j.
Elahi, E., Weijun, C., Jha, S.K., Zhang, H., 2019. Estimation of realistic renewable and jclepro.2017.10.282.
non-renewable energy use targets for livestock production systems utilising an Houshyar, E., Chen, B., Chen, G.Q., 2019. Environmental impacts of rice production
artificial neural network method: a step towards livestock sustainability. Energy 183, analyzed via social capital development: an Iranian case study with a life cycle
191–204. https://doi.org/10.1016/j.energy.2019.06.084. assessment/data envelopment analysis approach. Ecol. Indic. 105, 675–687. https://
Esfahani, S.M.J., Naderi, K.M., Saadi, H., Dourandish, A., 2017. Efficiency and doi.org/10.1016/j.ecolind.2018.07.040.
Sustainability of Silage Corn Production by Data Envelopment Analysis and Multi- Huang, Y., Li, Y.P., Chen, X., Ma, Y.G., 2012. Optimization of the irrigation water
Functional Ecological Footprint: Evidence from Sarayan County, Iran. J. Agr. Sci. resources for agricultural sustainability in Tarim River basin, China. Agric. Water
Tech. 19, 1453–1468. In this issue. Manag. 107, 74–85. https://doi.org/10.1016/j.agwat.2012.01.012.
European Union, 2013. European Commission recommendations n. 2013/19/EU of 9 Iribarren, D., Vázquez-Rowe, I., Moreira, M.T., Feijoo, G., 2010. Further potentials in the
april 2013 on the use of common methods to measure and communicate the life joint implementation of life cycle assessment and data envelopment analysis. Sci.
cycle environmental performance of products and organizations. Total Environ. 408, 5265–5272. https://doi.org/10.1016/j.scitotenv.2010.07.078.
Fasakhodi, A.A., Nouri, S.H., Amini, M., 2010. Water resources sustainability and Iribarren, D., Hospido, A., Moreira, M.T., Feijoo, G., 2011. Benchmarking environmental
optimal cropping pattern in farming systems; a multi-objective fractional goal and operational parameters through eco-efficiency criteria for dairy farms. Sci. Total
programming approach. Water Resour. Manag. 24, 4639–4657. https://doi.org/ Environ. 409, 1786–1798. https://doi.org/10.1016/j.scitotenv.2011.02.013.
10.1007/s11269-010-9683-z. Jornada, D., Leon, V.J., 2016. Biobjective robust optimization over the efficient set for
Fedele, A., Mazzi, A., Niero, M., Zuliani, F., Scipioni, A., 2014. Can the life cycle Pareto set reduction. Eur. J. Oper. Res. 252, 573–586. https://doi.org/10.1016/j.
assessment methodology be adopted to support a single farm on its environmental ejor.2016.01.017.
impacts forecast evaluation between conventional and. J. Clean. Prod. 69, 49–59. Kaab, A., Sharifi, M., Mobli, H., Nabavi-Pelesaraei, A., Chau, K. Wing, 2019a. Use of
Fleskens, L., de Graaff, J., 2010. Conserving natural resources in olive orchards on optimization techniques for energy use efficiency and environmental life cycle
sloping land: alternative goal programming approaches towards effective design of assessment modification in sugarcane production. Energy 181, 1298–1320. https://
doi.org/10.1016/j.energy.2019.06.002.

13
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

Kaab, A., Sharifi, M., Mobli, H., Nabavi-Pelesaraei, A., Chau, K. Wing, 2019b. Combined study in Iran. China Agricultural Economic Review 1, 478–484. https://doi.org/
life cycle assessment and artificial intelligence for prediction of output energy and 10.1108/17561370910989284.
environmental impacts of sugarcane production. Sci. Total Environ. 664, Masuda, K., 2016. Measuring eco-efficiency of wheat production in Japan: a combined
1005–1019. https://doi.org/10.1016/j.scitotenv.2019.02.004. application of life cycle assessment and data envelopment analysis. J. Clean. Prod.
Kahan, D., 2013. FARM Management Extension Guide FARM BUSINESS ANALYSIS Using 126, 373–381. https://doi.org/10.1016/j.jclepro.2016.03.090.
Benchmarking. FAO. Masuda, K., 2019. Eco-efficiency assessment of intensive Rice production in Japan: joint
Kaim, A., Cord, A.F., Volk, M., 2018. A review of multi-criteria optimization techniques application of life cycle assessment and data envelopment analysis. Sustainability 11,
for agricultural land use allocation. Environ. Model. Softw. 105, 79–93. https://doi. 5368. https://doi.org/10.3390/su11195368.
org/10.1016/j.envsoft.2018.03.031. Mohamad, R.S., Verrastro, V., Cardone, G., Bteich, M.R., Favia, M., Moretti, M.,
Kang, Y., Ozdogan, M., Zhu, X., Ye, Z., Hain, C., Anderson, M., 2020. Comparative Roma, R., 2014. Optimization of organic and conventional olive agricultural
assessment of environmental variables and machine learning algorithms for maize practices from a life cycle assessment and life cycle costing perspectives. J. Clean.
yield prediction in the US Midwest. Environ. Res. Lett. 15, 064005 https://doi.org/ Prod. 70, 78–89. https://doi.org/10.1016/j.jclepro.2014.02.033.
10.1088/1748-9326/ab7df9. Mohammadi, A., Rafiee, S., Jafari, A., Dalgaard, T., Knudsen, M.T., Keyhani, A., Mousavi-
Khanali, M., Mobli, H., Hosseinzadeh-Bandbafha, H., 2017. Modeling of yield and Avval, S.H., Hermansen, J.E., 2013. Potential greenhouse gas emission reductions in
environmental impact categories in tea processing units based on artificial neural soybean farming: a combined use of life cycle assessment and data envelopment
networks. Environ. Sci. Pollut. Res. 24, 26324–26340. https://doi.org/10.1007/ analysis. J. Clean. Prod. 54, 89–100. https://doi.org/10.1016/j.jclepro.2013.05.019.
s11356-017-0234-5. Moher, D., Liberati, A., Tetzlaff, J., Altman, D.G., 2009. Preferred reporting items for
Khoshnevisan, B., Rafiee, S., Omid, M., Mousazadeh, H., 2013a. Applying data systematic reviews and meta-analyses: the PRISMA statement. PLoS Med. https://
envelopment analysis approach to improve energy efficiency and reduce GHG doi.org/10.1371/journal.pmed.1000097.
(greenhouse gas) emission of wheat production. Energy 58, 588–593. https://doi. Mohseni, P., Borghei, A.M., Khanali, M., 2018. Coupled life cycle assessment and data
org/10.1016/j.energy.2013.06.030. envelopment analysis for mitigation of environmental impacts and enhancement of
Khoshnevisan, B., Rafiee, S., Omid, M., Mousazadeh, H., Sefeedpari, P., 2013b. energy efficiency in grape production. J. Clean. Prod. 197, 937–947. https://doi.
Prognostication of environmental indices in potato production using artificial neural org/10.1016/j.jclepro.2018.06.243.
networks. J. Clean. Prod. 52, 402–409. https://doi.org/10.1016/j. Montazar, A., 2013. A decision tool for optimal irrigated crop planning and water
jclepro.2013.03.028. resources sustainability. J. Glob. Optim. 55, 641–654. https://doi.org/10.1007/
Khoshnevisan, B., Rafiee, S., Omid, M., Yousefi, M., Movahedi, M., 2013c. Modeling of s10898-011-9803-1.
energy consumption and GHG (greenhouse gas) emissions in wheat production in Mousavi-Avval, S.H., Rafiee, S., Sharifi, M., Hosseinpour, S., Notarnicola, B., Tassielli, G.,
Esfahan province of Iran using artificial neural networks. Energy 52, 333–338. Renzulli, P.A., 2017. Application of multi-objective genetic algorithms for
https://doi.org/10.1016/j.energy.2013.01.028. optimization of energy, economics and environmental life cycle assessment in
Khoshnevisan, B., Rafiee, S., Omid, M., Mousazadeh, H., Rajaeifar, M.A., 2014. oilseed production. J. Clean. Prod. 140, 804–815. https://doi.org/10.1016/j.
Application of artificial neural networks for prediction of output energy and GHG jclepro.2016.03.075.
emissions in potato production in Iran. Agric. Syst. 123, 120–127. https://doi.org/ Nabavi-Pelesaraei, A., Hosseinzadeh-Bandbafha, H., Qasemi-Kordkheili, P., Kouchaki-
10.1016/j.agsy.2013.10.003. Penchah, H., Riahi-Dorcheh, F., 2016. Applying optimization techniques to improve
Khoshnevisan, B., Bolandnazar, E., Shamshirband, S., Shariati, H.M., Anuar, N.B., Mat of energy efficiency and GHG (greenhouse gas) emissions of wheat production.
Kiah, M.L., 2015. Decreasing environmental impacts of cropping systems using life Energy 103, 672–678. https://doi.org/10.1016/j.energy.2016.03.003.
cycle assessment (LCA) and multi-objective genetic algorithm. J. Clean. Prod. 86, Nabavi-Pelesaraei, A., Rafiee, S., Mohtasebi, S.S., Hosseinzadeh-Bandbafha, H., Chau, K.
67–77. https://doi.org/10.1016/j.jclepro.2014.08.062. Wing, 2017. Energy consumption enhancement and environmental life cycle
Klein, T., Holzkämper, A., Calanca, P., Seppelt, R., Fuhrer, J., 2013. Adapting assessment in paddy production using optimization techniques. J. Clean. Prod. 162,
agricultural land management to climate change: a regional multi-objective 571–586. https://doi.org/10.1016/j.jclepro.2017.06.071.
optimization approach. Landsc. Ecol. 28, 2029–2047. https://doi.org/10.1007/ Nabavi-Pelesaraei, A., Rafiee, S., Mohtasebi, S.S., Hosseinzadeh-Bandbafha, H., Chau, K.
s10980-013-9939-0. Wing, 2018. Integration of artificial intelligence methods and life cycle assessment to
Knudsen, M.T., Meyer-Aurich, A., Olesen, J.E., Chirinda, N., Hermansen, J.E., 2014. predict energy output and environmental impacts of paddy production. Sci. Total
Carbon footprints of crops from organic and conventional arable crop rotations - Environ. 631–632, 1279–1294. https://doi.org/10.1016/j.scitotenv.2018.03.088.
using a life cycle assessment approach. J. Clean. Prod. 64, 609–618. https://doi.org/ Nemecek, T., Weiler, K., Plassmann, K., Schnetzer, J., Gaillard, G., Jefferies, D., García-
10.1016/j.jclepro.2013.07.009. Suárez, T., King, H., Milà, I., Canals, L., 2012. Estimation of the variability in global
Kong, Q., Kuriyan, K., Shah, N., Guo, M., 2019. Development of a responsive warming potential of worldwide crop production using a modular extrapolation
optimisation framework for decision-making in precision agriculture. Comput. approach. J. Clean. Prod. 31, 106–117. https://doi.org/10.1016/j.
Chem. Eng. 131, 106585. https://doi.org/10.1016/j.compchemeng.2019.106585. jclepro.2012.03.005.
Längkvist, M., Karlsson, L., Loutfi, A., 2014. A review of unsupervised feature learning Nemecek, T., Bengoa, X., Lansche, J., Mouron, P., Rossi, V., 2015a. Methodological
and deep learning for time-series modeling. Pattern Recognition Letters 42, 11–24. Guidelines for the Life Cycle Inventory of Agricultural Products. World Food LCA
https://doi.org/10.1016/j.patrec.2014.01.008. Database.
Le Gal, P.Y., Dugué, P., Faure, G., Novak, S., 2011. How does research address the design Nemecek, T., Hayer, F., Bonnin, E., Carrouée, B., 2015b. Designing eco-efficient crop
of innovative agricultural production systems at the farm level? A review. rotations using life cycle assessment of crop combinations. Eur. J. Agron. 65, 40–51.
Agricultural Systems. doi. https://doi.org/10.1016/j.agsy.2011.07.007. Notarnicola, B., Sala, S., Anton, A., McLaren, S.J., Saouter, E., Sonesson, U., 2017. The
Li, M., Fu, Q., Singh, V.P., Liu, D., Li, T., Zhou, Y., 2020. Managing agricultural water and role of life cycle assessment in supporting sustainable Agri-food systems: a review of
land resources with tradeoff between economic, environmental, and social the challenges. J. Clean. Prod. 140, 399–409. https://doi.org/10.1016/j.
considerations: a multi-objective non-linear optimization model under uncertainty. jclepro.2016.06.071.
Agric. Syst. 178, 102685. https://doi.org/10.1016/j.agsy.2019.102685. Nuthall, P. (Ed.), 2018. Farm Business Management, the Human Factor, 2nd, Edition. ed.
Li, X., Mupondwa, E., 2014. Life cycle assessment of camelina oil derived biodiesel and Lincoln University, New Zealand.
jet fuel in the Canadian Prairies. Science of The Total Environment 481, 17–26. Paramesh, V., Arunachalam, V., Nikkhah, A., Das, B., Ghnimi, S., 2018. Optimization of
https://doi.org/10.1016/J.SCITOTENV.2014.02.003. energy consumption and environmental impacts of arecanut production through
Liakos, K., Busato, P., Moshou, D., Pearson, S., Bochtis, D., 2018. Machine learning in coupled data envelopment analysis and life cycle assessment. J. Clean. Prod. 203,
agriculture: a review. Sensors 18, 2674. https://doi.org/10.3390/s18082674. 674–684. https://doi.org/10.1016/j.jclepro.2018.08.263.
López-Andrés, J.J., Aguilar-Lasserre, A.A., Morales-Mendoza, L.F., Azzaro-Pantel, C., Pastori, M., 2017. A multi-objective approach to evaluate the economic and
Pérez-Gallardo, J.R., Rico-Contreras, J.O., 2018. Environmental impact assessment environmental impacts of alternative water and nutrient management strategies in
of chicken meat production via an integrated methodology based on LCA, simulation Africa. J. Environ. Inf. 29, 16–28. https://doi.org/10.3808/jei.201500313.
and genetic algorithms. J. Clean. Prod. 174, 477–491. https://doi.org/10.1016/j. Payandeh, Z., Kheiralipour, K., Karimi, M., Khoshnevisan, B., 2017. Joint data
jclepro.2017.10.307. envelopment analysis and life cycle assessment for environmental impact reduction
Lozano, S., Iribarren, D., Moreira, M.T., Feijoo, G., 2010. Environmental impact in broiler production systems. Energy 127, 768–774. https://doi.org/10.1016/j.
efficiency in mussel cultivation. Resour. Conserv. Recycl. 54, 1269–1277. https:// energy.2017.03.112.
doi.org/10.1016/j.resconrec.2010.04.004. Pelletier, N., 2015. Life cycle thinking, measurement and Management for Food System
Ma, Q., Wang, W., Peng, Y., Song, X., 2018. An optimization approach to the intermodal Sustainability. Environ. Sci. Technol. 49, 7515–7519. https://doi.org/10.1021/acs.
transportation network in fruit cold chain, considering cost, quality degradation and est.5b00441.
carbon dioxide footprint. Polish Maritime Research 25, 61–69. https://doi.org/ Pelletier, N., 2017. Life cycle assessment of Canadian egg products, with differentiation
10.2478/pomr-2018-0007. by hen housing system type. J. Clean. Prod. 152, 167–180. https://doi.org/10.1016/
MacWilliam, S., Wismer, M., Kulshreshtha, S., 2014. Life cycle and economic assessment j.jclepro.2017.03.050.
of Western Canadian pulse systems: the inclusion of pulses in crop rotations. Agric. Pelletier, N., Arsenault, N., Tyedmers, P., 2008. Scenario modeling potential eco-
Syst. 123, 43–53. https://doi.org/10.1016/j.agsy.2013.08.009. efficiency gains from a transition to organic agriculture: life cycle perspectives on
Maggioni, F., Potra, F.A., Bertocchi, M., 2015. Stochastic versus robust optimization for a Canadian canola, corn, soy, and wheat production. Environ. Manag. 42, 989–1001.
transportation problem. Odysseus 215–218. https://doi.org/10.1007/s00267-008-9155-x.
Maiyar, L.M., Thakkar, J.J., 2019. Modelling and analysis of intermodal food grain Piot-Lepetit, I., 2014. Technological externalities and environmental policy: how to
transportation under hub disruption towards sustainability. Int. J. Prod. Econ. 217, simulate manure management regulation within a DEA framework. Ann. Oper. Res.
281–297. https://doi.org/10.1016/j.ijpe.2018.07.021. 214, 31–48. https://doi.org/10.1007/s10479-010-0744-8.
Mansoori, H., Kohansal, M.R., Ghousi, M.F.K., 2009. Introducing a lexicographic goal Pishgar-Komleh, S.H., Akram, A., Keyhani, A., Sefeedpari, P., Shine, P., Brandao, M.,
programming for environmental conservation program in farm activities: a case 2019. Integration of life cycle assessment, artificial neural networks, and
metaheuristic optimization algorithms for optimization of tomato-based cropping

14
M.D. Heidari et al. Agricultural Systems 193 (2021) 103208

systems in Iran. Int. J. Life Cycle Assess. 1–13. https://doi.org/10.1007/s11367-019- Syp, A., Faber, A., Borzecka-Walker, M., Osuch, D., 2015. Assessment of greenhouse gas
01707-6. emissions in winter wheat farms using data envelopment analysis approach. Polish
Rǎdulescu, M., Rǎdulescu, C.Z., Zbǎganu, G., 2014. A portfolio theory approach to crop Journal of Environmental Studies 24, 2197–2203. https://doi.org/10.15244/
planning under environmental constraints. Ann. Oper. Res. 219, 243–264. https:// PJOES/39682.
doi.org/10.1007/s10479-011-0902-7. Tascione, V., Mosca, R., Raggi, A., 2014. LCA and linear programming for the
Rebolledo-Leiva, R., Angulo-Meza, L., Iriarte, A., González-Araya, M.C., 2017. Joint environmental optimization of waste management systems: A simulation, in:
carbon footprint assessment and data envelopment analysis for the reduction of Pathways to Environmental Sustainability: Methodologies and Experiences. Springer
greenhouse gas emissions in agriculture production. Sci. Total Environ. 593–594, International Publishing, pp. 13–22. https://doi.org/10.1007/978-3-319-03826-1_2.
36–46. https://doi.org/10.1016/j.scitotenv.2017.03.147. Thies, C., Kieckhäfer, K., Spengler, T.S., Sodhi, M.S., 2019. Operations research for
Reddy, S.S., Sandeep, V., Jung, C.-M., 2017. Review of stochastic optimization methods sustainability assessment of products: a review. Eur. J. Oper. Res. https://doi.org/
for smart grid. Frontiers in Energy 11, 197–209. https://doi.org/10.1007/s11708- 10.1016/j.ejor.2018.04.039.
017-0457-7. Toma, P., Miglietta, P.P., Zurlini, G., Valente, D., Petrosillo, I., 2017. A non-parametric
Reinhard, S., Knox Lovell, C.A., Thijssen, G.J., 2000. Environmental efficiency with bootstrap-data envelopment analysis approach for environmental policy planning
multiple environmentally detrimental variables; estimated with SFA and DEA. Eur. and management of agricultural efficiency in EU countries. Ecol. Indic. 83, 132–143.
J. Oper. Res. 121, 287–303. https://doi.org/10.1016/S0377-2217(99)00218-0. https://doi.org/10.1016/j.ecolind.2017.07.049.
Risbey, J., Kandlikar, M., Dowlatabadi, H., Graetz, D., 1999. Scale, context, and decision Udias, A., Pastori, M., Dondeynaz, C., Carmona Moreno, C., Ali, A., Cattaneo, L., Cano, J.,
making in agricultural adaptation to climate variability and change. Mitig. Adapt. 2018. A decision support tool to enhance agricultural growth in the Mékrou river
Strateg. Glob. Chang. 4, 137–165. https://doi.org/10.1023/A:1009636607038. basin (West Africa). Comput. Electron. Agric. 154, 467–481. https://doi.org/
Robert, M., Thomas, A., Sekhar, M., Raynal, H., Casellas, É., Casel, P., Chabrier, P., 10.1016/j.compag.2018.09.037.
Joannon, A., Bergez, J.É., 2018. A dynamic model for water management at the farm Ullah, A., Perret, S.R., 2014. Technical- and environmental-efficiency analysis of
level integrating strategic, tactical and operational decisions. Environ. Model. Softw. irrigated cotton-cropping systems in Punjab, Pakistan using data envelopment
100, 123–135. https://doi.org/10.1016/j.envsoft.2017.11.013. analysis. Environ. Manag. 54, 288–300. https://doi.org/10.1007/s00267-014-0300-
Rohmer, S.U.K., Gerdessen, J.C., Claassen, G.D.H., 2019. Sustainable supply chain design 4.
in the food system with dietary considerations: a multi-objective analysis. Eur. J. Ullah, A., Perret, S.R., Gheewala, S.H., Soni, P., 2016. Eco-efficiency of cotton-cropping
Oper. Res. 273, 1149–1164. https://doi.org/10.1016/j.ejor.2018.09.006. systems in Pakistan: an integrated approach of life cycle assessment and data
Rosano-Peña, C., Guarnieri, P., Sobreiro, V.A., Serrano, A.L.M., Kimura, H., 2014. envelopment analysis. J. Clean. Prod. 134, 623–632. https://doi.org/10.1016/j.
A measure of sustainability of Brazilian agribusiness using directional distance jclepro.2015.10.112.
functions and data envelopment analysis. Int J Sust Dev World 21, 210–222. https:// Utomo, D.S., Onggo, B.S., Eldridge, S., 2018. Applications of agent-based modelling and
doi.org/10.1080/13504509.2014.901992. simulation in the Agri-food supply chains. Eur. J. Oper. Res. https://doi.org/
Russell, S., Norvig, P., 2002. Artificial Intelligence: A Modern Approach, Third edition. 10.1016/j.ejor.2017.10.041.
Pearson. Van Meensel, J., Lauwers, L., Van Huylenbroeck, G., Van Passel, S., 2010. Comparing
Sahoo, S., Russo, T.A., Elliott, J., Foster, I., 2017. Machine learning algorithms for frontier methods for economic-environmental trade-off analysis. Eur. J. Oper. Res.
modeling groundwater level changes in agricultural regions of the U.S. Water 207, 1027–1040. https://doi.org/10.1016/j.ejor.2010.05.026.
Resour. Res. 53, 3878–3895. https://doi.org/10.1002/2016WR019933. Vázquez-Rowe, I., Iribarren, D., Moreira, M.T., Feijoo, G., 2010. Combined application of
Sanjuan, N., Ribal, J., Clemente, G., Fenollosa, M.L., 2011. Measuring and improving life cycle assessment and data envelopment analysis as a methodological approach
eco-efficiency using data envelopment analysis. J. Ind. Ecol. 15, 614–628. https:// for the assessment of fisheries. Int. J. Life Cycle Assess. 15, 272–283. https://doi.org/
doi.org/10.1111/j.1530-9290.2011.00347.x. 10.1007/s11367-010-0154-9.
Sarker, R., Ray, T., 2009. An improved evolutionary algorithm for solving multi- Vázquez-Rowe, I., Villanueva-Rey, P., Iribarren, D., Teresa Moreira, M., Feijoo, G., 2012.
objective crop planning models. Comput. Electron. Agric. 68, 191–199. https://doi. Joint life cycle assessment and data envelopment analysis of grape production for
org/10.1016/j.compag.2009.06.002. vinification in the Rías Baixas appellation (NW Spain). J. Clean. Prod. 27, 92–102.
Sarkis, J., 2002. Preparing your Data for DEA, in: Productivity Analysis in the Service https://doi.org/10.1016/j.jclepro.2011.12.039.
Sector with Data Envelopment Analysis. Viaggi, D., Bartolini, F., Raggi, M., 2009. Combining linear programming and principal-
Serra, T., Chambers, R.G., Oude Lansink, A., 2014. Measuring technical and agent models: an example from environmental regulation in agriculture. Environ.
environmental efficiency in a state-contingent technology. Eur. J. Oper. Res. 236, Model. Softw. 24, 703–710. https://doi.org/10.1016/j.envsoft.2008.10.014.
706–717. https://doi.org/10.1016/j.ejor.2013.12.037. Villalba, D., Díez-Unquera, B., Carrascal, A., Bernués, A., Ruiz, R., 2019. Multi-objective
Shamshirband, S., Khoshnevisan, B., Yousefi, M., Bolandnazar, E., Anuar, N.B., simulation and optimisation of dairy sheep farms: exploring trade-offs between
Wahab, A.W.A., Khan, S.U.R., 2015. A multi-objective evolutionary algorithm for economic and environmental outcomes. Agric. Syst. 173, 107–118. https://doi.org/
energy management of agricultural systems - a case study in Iran. Renew. Sust. 10.1016/j.agsy.2019.01.011.
Energ. Rev. https://doi.org/10.1016/j.rser.2014.12.038. von Ow, A., Waldvogel, T., Nemecek, T., 2020. Environmental optimization of the Swiss
Sharif, S.A., Hammad, A., 2019. Developing surrogate ANN for selecting near-optimal population’s diet using domestic production resources. J. Clean. Prod. 248, 119241.
building energy renovation methods considering energy consumption, LCC and LCA. https://doi.org/10.1016/j.jclepro.2019.119241.
Journal of Building Engineering 25, 100790. https://doi.org/10.1016/j. Wang, S., Tao, F., Shi, Y., 2018. Optimization of location–routing problem for cold chain
jobe.2019.100790. logistics considering carbon footprint. Int. J. Environ. Res. Public Health 15, 86.
Sharma, R., Kamble, S.S., Gunasekaran, A., Kumar, V., Kumar, A., 2020. A systematic https://doi.org/10.3390/ijerph15010086.
literature review on machine learning applications for sustainable agriculture supply Wojcik, V., Dyckhoff, H., Clermont, M., 2019. Is data envelopment analysis a suitable
chain performance. Comput. Oper. Res. 119, 104926. https://doi.org/10.1016/j. tool for performance measurement and benchmarking in non-production contexts?
cor.2020.104926. Bus. Res. 12, 559–595. https://doi.org/10.1007/s40685-018-0077-z.
Skevas, T., Serra, T., 2016. The role of pest pressure in technical and environmental Xavier, A., Freitas, Costa, Fragoso, R., Rosário, M., 2018. A regional composite indicator
inefficiency analysis of Dutch arable farms: an event-specific data envelopment for analysing agricultural sustainability in Portugal: a goal programming approach.
approach. J. Prod. Anal. 46, 139–153. https://doi.org/10.1007/s11123-016-0476-0. Ecol. Indic. 89, 84–100. https://doi.org/10.1016/j.ecolind.2018.01.048.
Skevas, T., Lansink, A.O., Stefanou, S.E., 2012. Measuring technical efficiency in the Xie, Y.L., Xia, D.X., Ji, L., Huang, G.H., 2018. An inexact stochastic-fuzzy optimization
presence of pesticide spillovers and production uncertainty: the case of Dutch arable model for agricultural water allocation and land resources utilization management
farms. Eur. J. Oper. Res. 223, 550–559. https://doi.org/10.1016/j.ejor.2012.06.034. under considering effective rainfall. Ecol. Indic. 92, 301–311. https://doi.org/
Skevas, T., Stefanou, S.E., Oude Lansink, A., 2014. Pesticide use, environmental 10.1016/j.ecolind.2017.09.026.
spillovers and efficiency: a DEA risk-adjusted efficiency approach applied to Dutch Yuan, K.Y., Lin, Y.C., Te Chiueh, P., Lo, S.L., 2018. Spatial optimization of the food,
arable farming. Eur. J. Oper. Res. 237, 658–664. https://doi.org/10.1016/j. energy, and water nexus: a life cycle assessment-based approach. Energy Policy 119,
ejor.2014.01.046. 502–514. https://doi.org/10.1016/j.enpol.2018.05.009.
Solano, N.E.C., García Llinás, G.A., Montoya-Torres, J.R., 2020. Towards the integration Yuanyuan, Z., 2020. Research on multi-objective planning model for agricultural
of lean principles and optimization for agricultural production systems: a conceptual pollution, environmental regulation and economic development. Arch. Latinoam.
review proposition. J. Sci. Food Agric. 100, 453–464. https://doi.org/10.1002/ Nutr. 70, 423–433.
jsfa.10018. Yusoff, Y., Ngadiman, M.S., Zain, A.M., 2011. Overview of NSGA-II for optimizing
Soteriades, A.D., Stott, A.W., Moreau, S., Charroin, T., Blanchard, M., Liu, J., machining process parameters. In: Procedia Engineering, 15. Elsevier,
Faverdin, P., 2016. The relationship of dairy farm eco-efficiency with intensification pp. 3978–3983. https://doi.org/10.1016/j.proeng.2011.08.74.
and self-sufficiency. Evidence from the French dairy sector using life cycle analysis, Zimmermann, A., 2008. Optimization of sustainable dairy-cow feeding systems with an
data envelopment analysis and partial least squares structural equation modelling. economic-ecological LP farm model using various optimization processes. J. Sustain.
PLoS One 11, 1–21. https://doi.org/10.1371/journal.pone.0166445. Agric. 32, 77–94. https://doi.org/10.1080/10440040802121411.

15

You might also like