Journal of Cleaner Production 165 (2017) 13e26

Contents lists available at ScienceDirect

Journal of Cleaner Production

journal homepage: www.elsevier.com/locate/jclepro

System dynamic modeling of energy savings in the US food industry

Yuan Xu a, *, Joseph Szmerekovsky b
a
Transportation and Logistics Program, North Dakota State University, 1320 Albrecht Blvd, Fargo, ND 58105, USA
b
Department of Management and Marketing, North Dakota State University, NDSU Department 2420, PO Box 6050, Fargo, ND 58108, USA

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

Article history: In the United States, the food industry alone accounts for approximately 19% of the total energy
Received 16 January 2017 consumed. The objective of this study is to forecast the food-related energy consumption and evaluate
Received in revised form policy alternatives for reducing food related energy consumption in the medium and long term. Influ-
7 July 2017
ential factors of energy consumption in the food industry include population growth rate, GDP, agri-
Accepted 12 July 2017
Available online 14 July 2017
culture's share of GDP, food waste, technologies, and agricultural investment. Based on the life cycle of
the food industry, five life stages were considered: agricultural production, industrial processing and
packaging, transportation, wholesale and retail, and household. In this study, based on the causal re-
Keywords:
Food industry
lationships and feedback loops between these factors and the energy consumed for each life stage, a
System dynamics system dynamics (SD) model was designed to simulate the situation for the US food industry. Policy
Energy consumption options of reducing waste percentage and improving industrial energy productivity are incorporated for
Policy analysis building the energy use mitigation scenarios. By implementing the integrated mitigation scenarios, a
9.43% reduction in total food system energy consumption can be attained in 2030.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction in oil equivalents to supply their food (Pimentel et al., 2008).

Owing to oil shortage, air pollution and climate change, energy
savings and emission reductions are important issues with global
attention. The food industry, which accounts for 30% of the global
energy consumption and 20% of greenhouse gas emissions has been
increasingly recognized for its potential for energy savings
(Monforti-ferrario et al., 2015). From the production of crops and
livestock, to the processing, packaging, distributing, storing, pre-
paring, serving and disposing of food products, energy plays an
important and necessary role for every stage of the food industry. In
the United States, the food industry takes a 10% share of the US GDP
(Egilmez et al., 2014) and around 17.3 million full- and part-time
jobs are related to agriculturedabout 9.3 percent of total US
employment (Glaser and Morrison, 2016). Although the food in-
dustry is neither energy nor emission intensive compared with
other industries such as iron or steel, it is still an important energy
consumer due to its massive scale. Overall, the food industry ac-
counts for approximately 19% of the total primary energy
consumed in the US, and each American requires nearly 2000 l/year
* Corresponding author.
E-mail addresses: yuan.xu@ndsu.edu (Y. Xu), joseph.szmerekovsky@ndsu.edu
(J. Szmerekovsky).
Additionally, in order to accommodate a growing population the
industry has consequentially grown in recent years. The average
annual growth rate of US food-related energy use was about 23
percent (Azzam, 2012). Therefore, it is necessary to seek energy
saving approaches for the food industry.
The food industry includes several important compo-
nentsdsuch as regulation, education, research and financial ser-
vices–which are beyond the scope of our interest here because they
represent very little energy consumption. We are actually modeling
the food system. The food system includes all the processes
involved in feeding a population: agricultural production, food
manufacturing (formulation, food processing, and packaging),
transportation and distribution, wholesale and retail, and food
preparation and disposal in the home. Based on the life cycle of the
food system, five life stages were considered: agricultural produc-
tion, industrial processing and packaging, transportation, whole-
sale and retail, and household (Minn and Seager, 2009). In addition
to the rapid population growth, several other factors also influence
energy consumption throughout the US food system, such as GDP
growth rate, agriculture's share of GDP, food waste, technologies,
and agricultural investment. These factors were chosen because
they were always used in previous literature for analyzing the food
system or building SD models for other industries (Ansari and Seifi,
2012; Canning et al., 2010; Minn and Seager, 2009; Wallgren and

http://dx.doi.org/10.1016/j.jclepro.2017.07.093
0959-6526/© 2017 Elsevier Ltd. All rights reserved.
14 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26
Mattias, 2009).
In this study, based on the causal relationships and feedback
loops between these factors and energy consumed for each life
stage of the food system, an SD model integrated with life cycle
analysis (LCA) was built to capture the constant shifts and changes
occurring in the US food system. The model was validated on data
from 1980 to 2014 and used to predict the trends of total US food
energy consumption through 2030. The objective here is to esti-
mate potential energy consumption in the future and to assess the
effects of two potential policy approaches for saving energy in the
US food industry: specifically, reducing the percentage of food
waste and improving industrial energy productivity.
Many papers have discussed energy conservation in the food
industry in different parts of the world, but none has comprehen-
sively examined the energy conservation status in all of the five life
stages in the food system. This paper addresses the following
important issues:

the impacts of population and GDP growth on food demand;

the inherent structures of energy consumption under diverse
scenarios for each life stage and the whole food system;

the impact of investments in public agricultural research and
development (R&D) on agricultural energy intensity;

the impact of agriculture's share of GDP on agriculture energy
use; and

analysis of potential energy savings attained by two policies:
reducing waste percentage and improving energy productivity
in industrial processing.
The rest of this paper is organized as follows: Section 2 presents
literature relevant to this study and how this research is derived;
Section 3 discusses the methodology, the modeling framework of
this study and the model validation; Section 4 develops the sensi-
tivity analysis, the policy analysis and the result discussion; and
Section 5 provides the conclusion and the limitations of this paper.
2. Literature review
Much research has analyzed food related topics, especially is-
sues like food security (Godfray et al., 2010; Maxwell, 1996;
Schmidhuber and Tubiello, 2007) and food quality (Cen and He,
2007; Grunert, 2005). Since the food industry is not an energy-
intensive industry and cost for energy inputs is only approxi-
mately 3% of the total cost of production (Muller et al., 2007), much
less literature deals with energy consumption in the food industry.
Energy conservation in the food industry has been studied by
several authors in different parts of the world, such as the US
(Azzam, 2012; Canning et al., 2010; Specht et al., 2014), China (Lin
and Xie, 2015, 2016; Ma et al., 2012), the UK (Hall and Howe,
2011), France (Seck et al., 2013, 2015), the Netherlands (Ramírez
et al., 2006), Switzerland (Muller et al., 2007), Portugal (Nunes
et al., 2014, 2016), and Sweden (Wallgren and Mattias, 2009).
Ramírez et al. (2006) developed feasible indicators to monitor
energy efficiency developments in the food industry based on
physical production data at the company level. To our knowledge,
this was the first time that energy usage per unit of physical output
was examined in a non-energy intensive industry at a high level of
aggregation. Hall and Howe (2011) analyzed the influence of
improving different kinds of energy intensive technologies on en-
ergy savings for the food processing industry. Azzam (2012)
answered the question of how much of US energy consumption
goes to the US food system by compiling diverse previous research
findings into a table and provided several energy efficient actions.
Specht et al. (2014) introduced an environmentally friendly build-
ing named Zfarming for future urban food production promising
resource savings and energy efficiency. In other efforts to increase
energy efficiency and save energy, Nunes et al. (2014) devised a
predictive tool for assessing the energy performance of cold storage
in agrifood industries, and Seck et al. (2013, 2015) developed a
bottom-up model to analyze the impact of heat recovery using heat
pumps in industrial processes. Lin and Xie (2015) studied the
substitution relationships between energy and other input factors
like capital and labor and the rebound effect of the food industry in
China. Most of this research provides tools for measuring energy
efficiency development in a single stage of the food industry.
However, we are unaware of any study that has comprehensively
examined the energy conservation status in all of the five life stages
in the food system.
Energy conservation or carbon emission reductions have always
been a hot topic for researchers focusing on diverse industries.
Methods employed in such studies can be divided into three main
categories: top-down, bottom-up, and hybrid approaches. The top-
down approach, such as Stochastic Frontier Analysis (SFA) and Data
Envelopment Analysis (DEA), refers to the decomposition of a
scenario into a set of equations. DEA has been applied in the
transport sector (Park et al., 2016), the power industry (Hampf and
Rødseth, 2015), and the iron and steel industry (Yang et al., 2016).
SFA has been used in the electricity sector (Chen et al., 2015), the
chemical industry (Lin and Long, 2015), and the thermal power
industry (Lin and Long, 2015). Bottom up approaches, such as
conservation supply curve (CSC) and regression, analyze the global
impacts associated with an industry by assembling the various local
disaggregated influences. Chen et al. (2017) applied CSC to analyze
the costs and the potential of energy conservation in China's coal-
fired power industry. CSC has also been used in India's cement,
iron and steel industries by Morrow et al. (2014). A hybrid approach
combines both bottom-up and top-down approaches.
For the food industry, Gowreesunker and Tassou (2015) sum-
marized many different approaches for modeling the food energy
flow, including life cycle analysis (LCA), MARKAL model, regression
analysis, input-output (IO) analysis and SD model. Among them,
LCA, MARKAL model and regression analysis belong to bottom-up
approaches, but IO analysis and SD model are top-down ap-
proaches. MARKAL is a demand-driven model used to compare
multi-period energy-related systems at the national level to
represent its evolution. It is often used to determine the trade-offs
between different objectives such as costs and environmental in-
dicators (Gowreesunker and Tassou, 2015). Regression is a simple
model to reflect the causal relationships between the dependent
and independent variables in the food system. Using statistical
methods or empirical observations to determine suitable influen-
tial variables is especially important for regression analysis
(Mendenhall and Sincich, 2012). Among these five methods, LCA
and IO analysis are the most popular quantitative approaches. IO is
a technique that represents the interdependencies between
different branches of economic sectors. Food-energy IO analysis is
usually used at a product specific level. This method always needs a
wide range of data presented in tabular form and can be attained
from national statistical offices on a yearly basis. Canning et al.
(2010) employed the IO model to analyze the energy flow in the
US food system between 1997 and 2007, and Zhang et al. (2012)
built a multi-regional IO model to track the use of energy for
various sectors in China in 2007. LCA, also known as cradle to grave
analysis, is a technique for evaluating environmental impacts
associated with all the processes or activities throughout the life
cycle of a product. LCA analysis was developed in the 1960s. Since
then, more and more researchers have used LCA to monitor energy
flows in the food industry, including Monforti-ferrario et al. (2015)
and Dutilh and Kramer (2000). Another powerful approach to
model a complex system is SD, a computer aided approach to
 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26
understanding the behavior of a system over time. SD models
provide an effective way to predict dynamic implications of
policies.
SD was first proposed by Forrester (1961). Since then, system
dynamics has been widely used in various fields of study such as
sociology, ecology, engineering, and medical sciences. For example,
Hsu (2012) developed an SD model to assess the effects of policies
like capital subsidies and feed-in-tariffs on solar PV installations.
Using this model to simulate the effect of applying different com-
binations of promotion policies under budget limitations and CO2
emission reduction goals in the long term, policymakers can carry
out cost/benefit analysis and decide the best policy to adopt in real
world. SD was also used by Saysel and Hekimog
lu (2013) to build a
quantity model to explore the options for carbon mitigation in the
Turkish electric power industry. After simulating the industry on an
annual basis for 30 years, the analysis revealed that adopting policy
options such as feed-in-tariffs, investment subsidies and carbon
taxes could reduce CO2 emissions by 50%. Ansari and Seifi (2012)
presented an SD model to study the effect of subsidy reform of
energy prices in Iran on energy consumption in the Iranian iron and
steel industry, and Ansari and Seifi (2013) presented another SD
model to explore the impact of this reform on production and en-
ergy consumption for the Iranian cement industry. In both models,
energy consumption in the industry was estimated under various
production and export scenarios while taking into account new
energy prices. Corrective policies were suggested to achieve the
most energy savings based on the simulated outlook of energy
demand for each scenario. Other SD models have been employed to
examine semiconductor industry development (Chen and Jan,
2005), natural gas industry in the UK (Chi et al., 2009) and in
China (Li et al., 2011), and highway sustainability (Egilmez and
Tatari, 2012). SD has often been used to explain the social and
corporate behavior of a complex system over time, thereby making
it useful for assessing the effects of energy conservation and carbon
emission reduction policies in diverse industries.
In this research, SD modeling is used to estimate the total energy
consumption for the entire food sector instead of considering each
food type or category in detail. In addition, to better illustrate this
complex system, SD is integrated with some concepts of LCA for a
better understanding of the process. IO and MARKAL are impossible
to apply without vast and accurate data which is not available for
our problem. Regression is useful to get a rough indication of policy
impacts for a simple case, but errors and uncertainties increase a lot
when applying it to a complex system. Dutilh and Kramer (2000)
employed LCA to compare energy requirements for different cate-
gories of food in each production and consumption stage and
concluded that there is a remarkable relationship between energy
requirements and market value throughout the production chain.
They demonstrate which life stages should be incorporated and
how to estimate energy demand when modeling energy con-
sumption. However, they only provide a rough estimate for the
energy requirements for some major groups of food products. In
contrast, all types of food products will be considered as a whole in
this research to provide existing and future information of energy
consumption for the entire food sector. This makes assessing which
energy saving procedures should be used in the food industry and
their effectiveness more clear. Roy et al. (2009) indicate that agri-
cultural production has been the hotspot for researchers' focus
during the lifetime of food products. Some other LCA studies on
agricultural products include both agricultural production and in-
dustrial processing as priorities (Audsley et al., 1997; Berlin et al.,
2007). Thus, in this research, the life stages of agricultural pro-
duction and industrial processing and packaging are important
focus points. Feng et al. (2013) used an SD model to analyze urban
15
energy consumption and CO2 emissions in Beijing, China. They
separated the urban system into six subsystems, including agri-
cultural consumption. The methodology of using agricultural en-
ergy intensity and agricultural GDP to measure agriculture energy
use in this research is derived from Feng et al. (2013). According to
Suh (2015), agricultural energy intensity is mainly influenced by
agricultural budget and output. Since public research and devel-
opment (R&D) investment is a major factor in the agricultural
budget (Heisey et al., 2011), public R&D investment and agricultural
output were chosen as two parameters affecting the agricultural
energy intensity in this paper.
3. Methodology
SD is a powerful method for analyzing and simulating complex
systems and is often used for policy analysis. Vensim, a simulation
software for improving the performance of real systems, was used.
This software has causal tracing, Monte Carlo sensitivity, optimi-
zation and sub-scripting capabilities (Zhang et al., 2012).
An SD model consists of feedback and causal loops, stock, flow
and auxiliary variables. A feedback loop is a closed sequence of
causes and effects. Stock is the state of an element that accumulates
the influences it receives over time. Flow is the increase or decrease
in the value of a stock during a period of time. The auxiliary vari-
ables are the rates of change for the stocks by determining the
values of flow for a period of time. After defining and testing these
components, the system was simulated for a reasonable time
period with respect to initial baseline values of past data. Then, by
comparing the simulation output with past data, the model was
validated. Finally, a series of scenario analysis for the variation of
key input parameters was conducted to examine the system's
responses.
3.1. Reference mode
In SD, the reference mode is simply another name for a plot of
the behavior of key variables over time. Reference mode, based on
historical information, gives clues to appropriate model structure
and can be used to check plausibility once the model is built. The
reference mode used in the SD model in this paper is the increasing
energy consumption trend in the US food industry. There is a sig-
nificant amount of data on energy use and economic activity in
different parts of the food sector that can be found from the
Department of Energy, Department of Agriculture and Department
of Commerce, but there isn't any official government figure on the
total amount of US energy consumption specifically for the food
industry. Indeed, because of the complexity of the modern food
industry, isolating food activity from other aspects of life is nearly
impossible, since food energy use is distributed across the resi-
dential, industrial, commercial, and transportation sectors (Minn
and Seager, 2009). In this study, we use the proportion of energy
used for food and data of total US energy consumption to estimate
the energy consumption in the US food sector. Annual Data for the
period 1980e2014 for total US energy consumption was attained
from the US Department of Energy, EIA (2016). According to the
food energy analysis models of Steinhart 1974 and Minn 2009, food
energy took 12.7% of total US energy consumption in 1970 and
17.8% in 2007 (Minn and Seager, 2009). Based on these data, we
estimate that the percentage of total US energy consumption that is
attributable to the food system grows at a rate of (17.8%e12.7%)/
(2007e1970) ¼ 0.14% per year, starting at 12.7% in 1970. This per-
centage is then multiplied by total US energy consumption to es-
timate US food system energy consumption from 1980 to 2014. The
reference mode for US food energy consumption is shown in Fig. 1.
16 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26
Fig. 1. Total energy consumption in the food industry (1980e2014).
3.2. Stock and flow diagram
A stock and flow diagram is the algebraic representation of a
model based on the causal relationships of parameters. In this
section, each parameter in the stock and flow diagram is explained.
Then, a sub-model to estimate food demand is introduced. Finally, a
comprehensive stock and flow diagram is constructed.
Data source of variables and comprehensive regression equa-
tions used in the model are shown in Appendix. All variables
measured in dollars are in real terms of 2009. Customer price index
(CPI), which was found from Bureau of Labor Statistics (2016), was
used to convert nominal dollar value into constant dollar value in
2009 to exclude the effects of inflation. For example, the CPIs for
2005 and 2009 are 195.3 and 214.5, respectively, then the value of
261.5 billion dollars for agricultural output in 2005 would be equal
to 287.2 billion dollars in 2009. This is calculated as 261.5*(214.5/
195.3).
3.2.1. Exogenous variables
GDP and population are two important exogenous variables
which directly affect the demand for food (Ansari and Seifi, 2012,
2013). The GDP directly reflects the economic development of a
country. Population growth is closely related to the increase of food
demand. Existing data for the GDP growth rate and the population
growth rate for 1980e2014 were averaged and then used to show
the pattern in the model. The average GDP growth rate is 0.272%
and average population growth rate is 0.998%.
3.2.2. Endogenous variables
(1) Food requirement: Food requirement measures the quantity
of food needed to be produced to satisfy total food demand
and post-consumer food waste. Market equilibrium, where
food supply is exactly equal to food demand, is a prerequisite
for this research. Food waste takes a large proportion of the
food supply and can't be ignored. According to Buzby et al.
(2014), the amount of food loss and waste is between 30
and 40 percent of total food supply from food industry. Thus,
food requirement was calculated by summing total food
demand and estimated food waste.
(2) Agricultural output: Agricultural output measures the value
of agricultural products produced during a certain period (a
year). Correlation and regression analysis were conducted
between food requirements and agricultural output. The
results shown in Table 1 indicate that there is a strong cor-
relation between food requirements and agricultural output
(0.957). Thus, the regression equation is given by
Agricultural output ¼ ð0:139food requirement þ 36830:8Þ=1000
R2 ¼ 0:92; p value ¼ 0:000
(1)
(3) Agriculture energy use: Direct energy consumption, farm
fuel, and indirect energy consumption, like farm chemicals,
machinery and irrigation, were used to calculate the total
energy input of agriculture production. The idea of using the
product of agricultural energy intensity and agricultural GDP
to measure agriculture energy use was derived from Feng
et al. (2013). Agricultural energy intensity was mainly
influenced by agricultural budget and output effects (Suh,
2015). Public R&D investment is a major factor in the agri-
cultural budget, and it has a significant impact on agricultural
productivity (Heisey et al., 2011). Thus, a regression model
was built using agricultural energy intensity, public R&D
investment, and agricultural output. The related index data
was obtained from the relationship among these variables.
Agriculture share of GDP measures the contribution of agri-
culture sector to the national economy. Relevant data from
1980 to 2014 were gathered from the Bureau of Economic
Analysis (BEA, 2016) and an average value of agriculture
share was thus calculated and used in the model.
Agriculture energy use ¼ agricultural energy intensity
(2)
agricultural GDP
Table 1
Correlation analysis between food requirement and agricultural output.
Food requirement Agricultural output
Food requirement 1 0.957
Agricultural output 0.957 1
 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26
Agricultural energy intensity ¼ ð582:509  0:00958398
“public R&D investment”  0:964287agricultural outputÞ=10000
R2 ¼ 0:60; p value ¼ 0:000
Agricultural GDP ¼ agriculture shareGDP=100
(4) Processing and packaging, transportation, retail and house-
hold: The output for processing and packaging, trans-
portation, retail and household is assumed to be related to
the output of the former stage. In this research, price spread
measures the ratio of the value of food products from each
life stage to its former stage. The ratio of farm value to retail
price from 1980 to 1997 can be found through Economic
Research Service (USDA, 2016). Extra years from 1998 to 2014
were extended based on the trend of the index in former
years. In the model, the price spreads were assumed to be the
same for any two neighboring life stages from agriculture to
retail for convenience, and the index for each year was
calculated by evenly dividing gathered ratio data into three
parts. For example, the ratio of farm value to retail price is
0.37 in 1980, thus, the price spread is 1.39 ((1/0.37)^(1/3)).
The value of food products for household equals that of retail
since no trading activity occurs during the customer phase.
(5) Industrial energy use: Industrial energy use measures the
energy consumption for food processing and packaging.
Energy productivity is defined as the ratio of industrial food
output divided by energy input. Thus, industrial energy use
equals food output of processing and packaging divided by
energy productivity.
Industrial energy use ¼“processing; packaging”=
energy productivity
(6) Total food system energy consumption: Total food system
energy consumption equals the sum of energy consumed by
agriculture, processing and packaging, transportation, retail
and household (Minn and Seager, 2009). Three regression
models were built to calculate transportation energy use,
commercial energy use and household energy use as shown
Fig. 2. Sub-model food demand stock and flow diagram.
17
in the appendix. R-squares for these three models are 0.926,
0.944 and 0.926, respectively.
Total food system energy consumption
(3) ¼ agriculture energy use þ industrial energy use
þ transportation energy use þ commercial energy use
(4) þ household energy use
(6)
3.2.3. Demand
Total food demand increases with the growth of population and
per capita food demand. Per capita food demand increases by per
capita gross domestic product growth. The structure of food de-
mand is derived from Ansari and Seifi (2012) and is presented in
Fig. 2. Population and GDP are stock variables, population growth
and GDP growth are flow variables, and the others are auxiliary
variables. The arrow shows the flow. Per capita food demand is
represented by per capita food expenditures in this paper. Relevant
data was gathered from Economic Research Service in USDA (2016).
The relationship between per capita food demand and GDP per
capita is given by
Per capita food demand ¼ 0:029*GDP per capita þ 2:578
R2 ¼ 0:60; p value ¼ 0:000
(7)
3.2.4. Stock and flow diagram
Based on the life cycle assessment of food products, energy
consumption of the food industry is the sum of energy used during
the following stages: agriculture, processing and packaging,
transportation and customer phase. The overall stock and flow di-
(5) agram is shown in Fig. 3.
3.3. Model validation
Model validation is the process of testing the accuracy of the
model behaviors. In this research, the reference mode is the energy
consumption in the food industry between 1980 and 2014. Calcu-
lations were performed using Vensim PLE 32 on Dell OPTIPLEX
9020 computer running with 3.40 GHz CPU and 16 GB of RAM.
18 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26

Fig. 3. Final stock and flow diagram.

Equations used in Vensim are shown in the Appendix.
The SD model was simulated for 35 years and the total food
system energy consumption obtained from the model at yearly
intervals was compared with the real data. The comparison of the
actual and the SD model's outputs is shown in Fig. 4. The R-square is
0.907.
One-way ANOVA statistical hypothesis test was used to compare
the actual energy consumption trend with the trend obtained with
the proposed SD model. Some assumptions should be met before
using the one-way ANOVA test: a. normal distribution; b. equal
variance; c. no outliers.
Minitab was used to test these assumptions. Normality test and
outlier test were applied to both actual and estimated energy
consumption data for the 1980e2014 period. Both data have no
outliers (p-value > 0.05). Estimated data is normally distributed
because p-value is bigger than 0.05. However, p-value for the actual
data is smaller than 0.05, indicating it's not normally distributed.
Fortunately, ANOVA is pretty tolerant of violations of normality.
According to the output of two-sample variance test, both data
have equal variances (p-value > 0.05).
Since all three assumptions are met, the one-way ANOVA test
was employed for the comparison. The result of the ANOVA test is
shown in Table 2. According to these results, there is no statistically
significant difference between two samples (p-
value ¼ 0.732 > 0.05). Thus, the reference mode provided by the
proposed SD model is valid.

20

18
Quadrillion Btus

16

14

12

10

8
1980 1985 1990 1995 2000 2005 2010 2015

Energy ConsumpƟon: Reference Mode Energy consumpƟon: SD model

Fig. 4. Validation of results.

 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26 19

Table 2 intensity, and agriculture's share of GDP directly influence the en-
Results of the one-way ANOVA test. ergy use by agriculture sector. Thus, population, GDP, public R&D
Source of Variation SS df MS F P-value F crit investment and agriculture share were chosen for sensitivity
Between Groups 0.79039 1 0.79039 0.11758 0.73272 3.98189
analysis in this study.
Within Groups 457.0771 68 6.72172

4.1.1. Impact of population on total food demand

Total 457.8674 69
4. Result analysis and discussion
4.1. Sensitivity analysis
Sensitivity tests were conducted to ascertain whether little
variations of input parameters can cause shifts in the behavior of
the system. Total food system energy consumption depends on
total food demand, food production, agricultural energy intensity
and industrial energy productivity. As discussed above, population
and GDP decide the total food demand, public R&D investment
plays an important role in controlling the agricultural energy
Population is a major factor in controlling food demand, which
in turn makes population the primary determinant for agricultural
production, and consequently for total food system energy con-
sumption. In addition to the 0.998% population growth rate, a
scenario of þ25% population growth rate (P1) starting in 2015,
corresponding to 1.248% annual growth, was designed. Fig. 5
compares the baseline (0.998% growth rate) with the P1 scenario
(1.248% growth rate).
4.1.2. Impact of GDP on total food demand
The impact of GDP growth on food demand was examined by
raising the GDP growth rate from 0.027 (Baseline) to 0.03 (G1) for
the years 2015 onwards. According to Fig. 6, a small increase in GDP
contributes significantly to the increased food demand in the food

Fig. 5. The estimated population trends for baseline and P1.

Fig. 6. The estimated total food demand trends for baseline and G1.
20 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26

Fig. 7. The estimated total food energy consumption trends for baseline and RD1.

system. Total food demand is positively related to total food system
energy consumption, thus, GDP has a positive impact on the total
food system energy consumption.
4.1.3. Impact of public R&D investment on total food system energy
consumption
As discussed above, agriculture energy intensity is mainly
influenced by agricultural budget and output effects. Public R&D
investment, as the major factor in the agricultural budget, is
consequently an important factor in agricultural energy use.
However, public R&D investment differs every year, so its sensi-
tivity should be analyzed. A 5% yearly growth rate of public R&D
investment starting from 2015 (RD1) was compared with the
baseline constant R&D investment. As shown in Fig. 7, the public
R&D investment has a negative impact on the total food system
energy consumption, but the impact is not very big.
4.1.4. Impact of agriculture share on total food system energy
consumption
Agriculture's share of GDP has a positive impact on agricultural
GDP, and agricultural GDP has a positive impact on the agriculture
energy use. And agriculture's share of GDP can be temporally
changed and is not connected to the year in this model. Thus, ag-
riculture's share of GDP, as an exogenous variable, should be tested
for its influence on the total food energy consumption. In addition
to the 0.852% share of the GDP given to agriculture (Baseline), an
additional scenario with a 1% share of the GDP for agriculture (AS1)
starting in 2015 was investigated. As shown in Fig. 8, agriculture's
share of GDP has a positive impact on the total food system energy
consumption.
4.2. Scenario generation
In this study, the main objective is to reduce total energy

Fig. 8. The estimated total food energy consumption trends for baseline and AS1.
 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26
consumption in the food supply chain. To realize this goal, two
potential strategies were considered and implemented in the SD
model: reducing waste percentage (WP) and improving energy
productivity (EP). Two action plans were implemented based on
each policy. Different scenarios were thus generated under the
broad categories of baseline scenario and alternative scenarios.
4.2.1. Baseline scenario
The baseline scenario was generated for the existing growth rate
of population and GDP without any policy interventions. Public
R&D investment remains constant as of the year 2014 and agri-
culture's share of GDP remains 0.852%. As mentioned earlier, the
population growth and GDP have positive effects on the total food
demand as well as total food system energy consumption. The
impacts of public R&D investment and agriculture's share of GDP
on total food energy consumption were tested through sensitivity
analysis. Thus, these four parameters should be controlled for the
baseline scenario and alternative scenarios. The projected trends of
total food energy consumption for the baseline were used as
feedback for applying policy options of reducing waste percentage
and improving energy productivity.
4.2.2. Alternative scenarios
Two alternative scenarios were considered: Policy 1 and Policy
2. Policy 1 (WP) concerns reducing waste percentage and Policy 2
(EP) is about improving energy productivity in industrial process-
ing. Each policy has two action plans.
(1) Policy 1; reducing waste percentage (WP): The USDA esti-
mates that the amount of food loss and waste is between 30
and 40 percent of the food system (Buzby et al., 2014). On
average, 35% was used for the waste percentage in the SD
model. In 2015, Agriculture Secretary Tom Vilsack and
Environmental Protection Agency Deputy Administrator Stan
Meiburg set the first US national food waste reduction goal,
calling for a 50% reduction by 2030 (USDA, 2015). Based on
this, there are two action plans considered in this study. Plan
1 (WP1) attains the 50% food waste reduction goal, and Plan
2 (WP2) achieves 25% food waste reduction by 2030. If no
action is taken, the trend is shown as the baseline. The ex-
pected waste percentage trends are shown in Fig. 9. Fig. 9
shows that reducing waste percentage is applied at yearly
Fig. 9. The expected waste percentage trends.
21
intervals, and annual reduction of waste percentage is 0.0109
(0.35*0.5/16) for WP1 and 0.0055 (0.35*0.25/16) for WP2.
(2) Policy 2; improving energy productivity (EP): Energy pro-
ductivity is measured by output divided by energy input. The
energy efficiency in food processing and packaging facilities
can be improved by upgrading existing equipment or intro-
ducing new energy-saving equipment such as refrigeration,
pumps or fan systems. Research indicates that 10e15% in-
dustrial energy savings could be realized by implementing
new technologies (MnTAP, 2009). In other words, imple-
menting new technologies could achieve an 11.1e17.6% in-
crease in energy productivity. According to equation (5),
these are calculated as (1/(1e10%) e 1) and (1/(1e15%) e 1).
Two plans were applied to improve energy productivity. Plan
1 (EP1) attains a 17.6% increase in energy productivity by
2016, and plan 2 (EP2) achieves an 11.1% increase in energy
productivity by 2016. The expected energy productivity
trends are shown in Fig. 10. Fig. 10 shows that energy pro-
ductivity improves significantly in 2015 for EP1 and EP2,
then, the growth rates are the same as the baseline starting
in 2016.
4.3. Scenario analysis
The SD model has been run under every possible combination of
policy scenarios for a life span of 50 years using Vensim software.
The results obtained were discussed to ascertain and compare the
impacts of various policy options on total food system energy
consumption.
4.3.1. Baseline scenario
For the baseline scenario, the growth rate of population and
GDP, the public R&D investment and agriculture's share of GDP
were kept constant. No waste reduction plan was implemented or
new technology employed to increase industrial energy produc-
tivity. Under these situations, the population in the US is projected
to reach 375.14 million by 2030, and total food demand is projected
to be 1675 billion dollars measured in 2009 dollars. The estimated
trend of total food energy consumption is shown in Fig. 11. The total
food system energy consumption shows an increasing trend and is
estimated to be 26.44 quadrillion Btus in the year 2030.
22 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26

Fig. 10. The expected energy productivity trends.

Fig. 11. The estimated total food energy consumption trends for all scenarios.

4.3.2. Alternative scenarios
Two policies were applied in the SD model and each policy has
two plans. WP1 attains 50% food waste reduction by 2030 and WP2
attains 25% food waste reduction by 2030; EP1 achieves a 17.6%
increase in energy productivity by 2016 and EP2 attains an 11.1%
increase in energy productivity by 2016. These four plans and the
baseline scenario were tested and results were compared with each
other. Results in terms of energy consumption in quadrillion Btus in
2030 are show in Fig. 12.
There are four possible combinations of these four plans: 1)
reducing waste percentage by 50% and at the same time attaining a
17.6% increase in energy productivity (WP1þEP1); 2) reducing
 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26 23

Fig. 12. The food energy consumption results of single plans in 2030.

waste percentage by 50% and attaining an 11.1% increase in energy
productivity (WP1þEP2); 3) reducing waste percentage by 25% and
attaining a 17.6% increase in energy productivity (WP2þEP1); and
4) reducing waste percentage by 25% and attaining an 11.1% in-
crease in energy productivity (WP2þEP2). These four combinations
were tested and results are shown in Fig. 13.
From the results in Fig. 11, projected total food energy con-
sumption under WP1 and WP2 show increasing trends for the
period 2015e2030, but the growth rates are lower than the base-
line scenario since reducing waste percentage is applied at yearly
intervals. However, under EP1 and EP2, the estimated total food
energy consumption decreases substantially for the year 2016, then
it increases with the same growth rate as the baseline scenario
starting from 2017. There is a sudden decrease in the estimated
total food energy consumption in 2016 because new technologies
for improving energy productivity would be introduced that same
year. WP1 has the lowest growth rate, and EP1 has the lowest value
for projected total food energy consumption in 2016. As shown in
Fig. 11, the lines for WP1 and EP1 intersect in the year 2020. Thus,
when only considering single plans, EP1 consumes the least
amount of energy before the year 2020, but WP1 consumes the
least amount of energy for the years 2020 onwards.
From the results in Fig. 12, in 2030, estimated total food energy
consumption is 26.4418 quadrillion Btus if no action is taken. WP1
saves the most energy compared with the baseline; 7.15%
((26.4418e24.5509)/26.4418) energy reduction will be attained by
reaching the 50% food waste reduction goal. This suggests a greater
opportunity for energy savings from reducing food waste than from
improving energy productivity in the processing and packaging
stage in the mid run.
From the results shown in Fig. 13, the lowest energy consump-
tion is reached by implementing both WP1 and EP1. In 2030, a
9.43% ((26.4418e23.9495)/26.4418) energy consumption reduction
is achieved by this combination compared with the baseline.
In the above scenario analysis, the population growth rate, the
GDP growth rate, the public R&D investment and agriculture's
share of GDP remain constant. Population and GDP are major fac-
tors in controlling total food demand, which in turn is the primary
determinant of agricultural production and total food system en-
ergy consumption. Population stabilization or a decrease in GDP

Fig. 13. The food energy consumption results of plan combinations in 2030.
24 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26

would decrease the food demand in the food system and thus
reduce total food energy consumption. Public R&D investment has
a negative impact on total food system energy consumption and
agriculture's share of GDP has a positive impact on total food en-
ergy consumption. An increase in public R&D investment or a
decrease in agriculture's share of GDP would reduce the total food
energy consumption. A 9.43% energy consumption reduction could
be attained in 2030 by a combined scenario by reducing waste
percentage by 50% and attaining a 17.6% increase in energy pro-
ductivity. Population growth rate, GDP growth rate, public R&D
investment and agriculture's share of GDP are important exogenous
variables that typically change over time. Therefore, targets for
energy consumption must allow for future changes.
5. Conclusion and future work
A system dynamics model for energy consumption in the food
industry was developed. The model was built to make projections
of food related energy consumption in the United States over a time
of 50 years. The objective was to evaluate policy alternatives for
reducing total food energy consumption. Two potential strategies
were considered and implemented in the SD model: reducing
waste percentage and improving energy productivity. By reducing
waste percentage at yearly intervals, the growth rate of the total
food energy consumption decreases for future years. By employing
new technologies to improve energy productivity in 2016, there is a
sudden decrease in total food energy consumption for the year
2016, then it increases with the same growth rate as the baseline
scenario starting from 2017. When only considering single plans,
EP1 consumes the least amount of energy before the year 2020, but
WP1 consumes the least amount of energy for the years 2020 on-
wards. The final results show that a 9.43% reduction in total food
energy consumption can be attained in 2030 by combining both
EP1 and WP1. Population growth rate, GDP growth rate and agri-
culture's share of GDP have positive impacts on total food system
energy consumption and public R&D investment has a negative
impact on total food energy consumption. Therefore, targets for
energy consumption must allow for future changes of these
variables.
The food industry is a very complex system which includes
many different food types. Different types of food experience
diverse life cycles and consume different degrees of energy for each
stage. Future work can specifically address the difference of energy
consumption among different food types into modeling. Also, other
influential factors such as household affluence and machinery
could be considered as parameters. Although household energy use
consumes a large percentage of total food energy, this part is a
small part of this model since the focus of this research is on the
agricultural sector. Further research can further address the
household sector.
Acknowledgments
I thank Megan Even and Yong shin Park for critically reading the
manuscript and helpful discussions. This research did not receive
any specific grant from funding agencies in the public, commercial,
or not-for-profit sectors.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.jclepro.2017.07.093.
Appendix

Table 1
Main variables and parameters of SD model.

Variable Abbreviation Unit Equations Data Source

Population growth rate POPR % 0.998 The world bank, 2016

Population growth POPG thousand POPG¼INTEGER(POP*POPR)
Population POP thousand POPt ¼ POP(t-1)þPOPG
GDP growth rate GDPR % 0.272 BEA, 2016
GDP growth GDPG billions of dollars GDPG ¼ INTEGER(GDP*GDPR)
GDP GDP billions of dollars GDPt ¼ GDP(t-1)þGDPG
GDP per capita CGDP thousands of dollars CGDP ¼ GDP*1000/POP
Per capita food demand CFD thousands of dollars CFD¼CGDP*0.029 þ 2.578 USDA, 2015
R2 ¼ 0.91, p-value ¼ 0.000
Total demand TD millions of dollars TD ¼ POP*CFD
Total waste percentage WP % 35 Buzby et al., 2014
Food requirement FR millions of dollars FR ¼ TD/(1-WP)
Agricultural output OA billions of dollars OA¼(0.139*FR þ 36830.8)/1000 USDA, 2016
R2 ¼ 0.92, p-value ¼ 0.000
Public R&D investment RD millions of dollars (1980,1629.72), (1992,2142.36), (1996,1927.79), USDA, 2016
(2001,2591.69), (2007,2329.28), (2010,2566.49),
(2013,1970.72), (2014,2183.37)
Agricultural energy intensity EIA e EIA¼(582.509e0.009*RD-0.964*OA)/10000
R2 ¼ 0.60, p-value ¼ 0.000
Agriculture share AS % 0.852 U.S.D.A. and Gross Domestic
Product, 2016
Agricultural GDP AGDP billions of dollars AGDP ¼ AS*GDP
Agriculture energy use EUA quadrillion Btus EUA ¼ EIA*AGDP Minn and Seager, 2009
Price spread PS e PS ¼ 0.014*(Time-1979)þ1.372 USDA, 2016
R2 ¼ 0.96, p-value ¼ 0.000
Processing, packaging PP billions of dollars PP¼OA*PS
Energy productivity EP e EP ¼ 0.664*(Time-1979) þ 132.402
R2 ¼ 0.55, p-value ¼ 0.000
Industrial energy use EUI quadrillion Btus EUI ¼ PP/EP Minn and Seager, 2009
Transportation TRT billions of dollars TRT ¼ PP*PS
Transportation energy use EUT quadrillion Btus EUT ¼ 0.00096*TRT þ 0.49433 Minn and Seager, 2009
R2 ¼ 0.94, p-value ¼ 0.000
 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26 25

Table 1 (continued )

Variable Abbreviation Unit Equations Data Source

Retail RT billions of dollars RT ¼ TRT*PS

Commercial energy use EUC quadrillion Btus EUC ¼ 0.000811*RT þ 1.13214 Minn and Seager, 2009
R2 ¼ 0.93, p-value ¼ 0.000
Household HH billions of dollars HH ¼ RT
Household energy use EUH quadrillion Btus EUH ¼ 0.00229*HH þ 3.20908 Minn and Seager, 2009
R2 ¼ 0.93, p-value ¼ 0.000
Total food system energy consumption TFSEC quadrillion Btus TFSEC ¼ EUA þ EUI þ EUT þ EUC þ EUH

References Energy Convers. Manag. 105, 20e29.

Lin, B., Xie, X., 2016. CO2 emissions of China's food industry : an input - output
approach. J. Clean. Prod. 112, 1410e1421.
Ansari, N., Seifi, A., 2012. A system dynamics analysis of energy consumption and
Ma, C., Chen, M., Hong, G., 2012. Energy conservation status in Taiwanese food
corrective policies in Iranian iron and steel industry. Energy 43 (1), 334e343.
industry. Energy Policy 50, 458e463.
Ansari, N., Seifi, A., 2013. A system dynamics model for analyzing energy con-
Maxwell, S., 1996. Food security: a post-modern perspective. Food Policy 21 (2),
sumption and CO2 emission in Iranian cement industry under various pro-
155e170.
duction and export scenarios. Energy Policy 58, 75e89.
Mendenhall, W., Sincich, T., 2012. A Second Course in Statistics: Regression Analysis,
Audsley, E., Alber, S., Clift, R., Cowell, S., Crettaz, P., Gaillard, G., Pearce, D., 1997.
seventh ed. Pearson.
Harmonisation of environmental life cycle assessment for agriculture. In: Final
Minn, M., Seager, G., 2009. Energy Use in American Food Production. Retrieved April
Report, Concerted Action AIR3-CT94-2028. European Commission, DG VI
10, 2016, from. http://www.pulsecanada.com/uploads/8c/96/
Agriculture, p. 139.
8c966601850c2c80ccdd00081c0560fb/2009-05-11-Minn-Energy-Use-in-
Azzam, A., 2012. Energy consumption in the U.S. food system. Cornhusker Econ.
American-Food-Production.pdf.
Berlin, J., Sonesson, U., Tillman, A.-M., 2007. A life cycle based method to minimise
MnTAP, 2009. Increasing Energy Efficiency in Food Processing. Retrieved May 5,
environmental impact of dairy production through product sequencing.
2016, from. http://www.mntap.umn.edu/food/energy.htm.
J. Clean. Prod. 15 (4), 347e356.
Monforti-ferrario, F., Dallemand, J., Pascua, I.P., Motola, V., Banja, M., Scarlat, N.,
Buzby, J.C., Farah-Wells, H., Hyman, J., 2014. The estimated amount, value, and
Renzulli, P., 2015. Energy Use in the EU Food Sector: State of Play and Oppor-
calories of postharvest food losses at the retail and consumer levels in the
tunities for Improvement.
United States. USDA-ERS Econ. Inf. Bull. Number 121.
Morrow, W.R., Hasanbeigi, A., Sathaye, J., Xu, T., 2014. Assessment of energy effi-
Canning, P., Charles, A., Huang, S., Polenske, K.R., Waters, A., 2010. Energy use in the
ciency improvement and CO2 emission reduction potentials in India's cement
U.S. food system. U. S. Dep. Agric. Econ. Res. Rep. 94.
and iron & steel industries. J. Clean. Prod. 65, 131e141.
Cen, H., He, Y., 2007. Theory and application of near infrared reflectance spectros-
Muller, D.C.A., Marechal, F.M.A., Wolewinski, T., Roux, P.J., 2007. An energy man-
copy in determination of food quality. Trends Food Sci. Technol. 18 (2), 72e83.
agement method for the food industry. Appl. Therm. Eng. 27 (16), 2677e2686.
Chen, H., Kang, J.-N., Liao, H., Tang, B.-J., Wei, Y.-M., 2017. Costs and potentials of
Nunes, J., Neves, D., Gaspar, P.D., Silva, P.D., Andrade, L.P., 2014. Predictive tool of
energy conservation in China's coal-fired power industry: a bottom-up
energy performance of cold storage in agrifood industries: the Portuguese case
approach considering price uncertainties. Energy Policy 104 (January), 23e32.
study. Energy Convers. Manag. 88, 758e767.
Chen, J.H., Jan, T.S., 2005. A system dynamics model of the semiconductor industry
Nunes, J., Silva, P.D., Andrade, L.P., Gaspar, P.D., 2016. Key points on the energy
development in Taiwan. J. Oper. Res. Soc. 56 (10), 1141e1150.
sustainable development of the food industry e case study of the Portuguese
Chen, Z., Barros, C.P., Borges, M.R., 2015. A Bayesian stochastic frontier analysis of
sausages industry. Renew. Sustain. Energy Rev. 57, 393e411.
Chinese fossil-fuel electricity generation companies. Energy Econ. 48, 136e144.
Park, Y.S., Lim, S.H., Egilmez, G., Szmerekovsky, J., 2016. Environmental efficiency
Chi, K.C., Nuttall, W.J., Reiner, D.M., 2009. Dynamics of the UK natural gas industry:
assessment of U.S. transport sector: a slack-based data envelopment analysis
system dynamics modelling and long-term energy policy analysis. Technol.
approach. Transp. Res. Part D Transp. Environ.
Forecast. Soc. Change 76 (3), 339e357.
Pimentel, D., Williamson, S., Alexander, C.E., Gonzalez-Pagan, O., Kontak, C.,
Dutilh, C.E., Kramer, K.J., 2000. Energy consumption in the food chain. AMBIO A J.
Mulkey, S.E., 2008. Reducing energy inputs in the US food system. Hum. Ecol. 36
Hum. Environ. 29 (2), 98e101.
(4), 459e471.
Egilmez, G., Kucukvar, M., Tatari, O., Bhutta, M.K.S., 2014. Resources, conservation
Ramírez, C.A., Blok, K., Neelis, M., Patel, M., 2006. Adding apples and oranges: the
and recycling supply chain sustainability assessment of the U. S. food
monitoring of energy efficiency in the Dutch food industry. Energy Policy 34
manufacturing sectors : a life cycle-based frontier approach. Resour. Conserv.
(14), 1720e1735.
Recycl. 82, 8e20.
Roy, P., Nei, D., Orikasa, T., Xu, Q., Okadome, H., Nakamura, N., Shiina, T., 2009.
Egilmez, G., Tatari, O., 2012. A dynamic modeling approach to highway sustain-
A review of life cycle assessment (LCA) on some food products. J. Food Eng. 90
ability: strategies to reduce overall impact. Transp. Res. Part A Policy Pract. 46
(1), 1e10.
(7), 1086e1096.
Saysel, A.K., Hekimog
lu, M., 2013. Exploring the options for carbon dioxide miti-
Feng, Y.Y., Chen, S.Q., Zhang, L.X., 2013. System dynamics modeling for urban energy
gation in Turkish electric power industry: system dynamics approach. Energy
consumption and CO2 emissions: a case study of Beijing, China. Ecol. Model.
Policy 60, 675e686.
252, 44e52.
Schmidhuber, J., Tubiello, F.N., 2007. Global food security under climate change.
Forrester, J.W., 1961. Industrial Dynamics. The M.I.T. Press, Cambridge, MA.
Proc. Natl. Acad. Sci. 104 (50), 19703e19708.
Glaser, L., Morrison, R.M., 2016. Ag and Food Sectors and the Economy. U.S.
Seck, G.S., Guerassimoff, G., Maïzi, N., 2013. Heat recovery with heat pumps in non-
Department of Agriculture, Economic Research Service. Retrieved May 5, 2016,
energy intensive industry: a detailed bottom-up model analysis in the French
from. http://www.ers.usda.gov/data-products/ag-and-food-statistics-charting-
food & drink industry. Appl. Energy 111, 489e504.
the-essentials/ag-and-food-sectors-and-the-economy.aspx.
Seck, G.S., Guerassimoff, G., Maïzi, N., 2015. Heat recovery using heat pumps in non-
Godfray, H.C.J., Beddington, J.R., Crute, I.R., Haddad, L., Lawrence, D., Muir, J.F.,
energy intensive industry: are energy saving certificates a solution for the food
Smaling, E.M.A., 2010. Food security: the challenge of feeding 9 billion people.
and drink industry in France? Appl. Energy 156, 374e389.
Science 327 (5967), 812e818.
Specht, K., Siebert, R., Hartmann, I., Freisinger, U.B., Sawicka, M., Werner, A.,
Gowreesunker, B.L., Tassou, S.A., 2015. Approaches for modelling the energy flow in
Dierich, A., 2014. Urban agriculture of the future: an overview of sustainability
food chains. Energy, Sustain. Soc. 5 (1), 7.
aspects of food production in and on buildings. Agric. Hum. Values 31 (1),
Grunert, K.G., 2005. Food quality and safety: consumer perception and demand.
33e51.
Eur. Rev. Agric. Econ. 32 (3), 369e391.
Suh, D., 2015. Declining energy intensity in the U.S. agricultural sector: implications
Hall, G.M., Howe, J., 2011. Energy from waste and the food processing industry.
for factor substitution and technological change. Sustainability 7 (10),
Process Saf. Environ. Prot. 90 (3), 203e212.
13192e13205.
Hampf, B., Rødseth, K.L., 2015. Carbon dioxide emission standards for U.S. power
The World Bank, 2016. World Development Indicators. Population. http://data.
plants: an efficiency analysis perspective. Energy Econ. 50, 140e153.
worldbank.org/indicator/SP.POP.TOTL?locations¼US. (Accessed 10 June 2010).
Heisey, P., Wang, S.L., Fuglie, K., 2011. Public agricultural research spending and
U.S. Department of Commerce, Bureau of Economic Analysis (BEA), 2016. Agricul-
future US agricultural productivity growth: scenarios for 2010-2050. U.S. Dep.
ture's Share of GDP. http://www.bea.gov/industry/gdpbyind_data.htm.
Agric. Econ. Res. Serv.
(Accessed 10 June 2010).
Hsu, C., 2012. Using a system dynamics model to assess the effects of capital sub-
U.S. Department of Commerce, Bureau of Economic Analysis (BEA), 2016. Gross
sidies and feed-in tariffs on solar PV installations. Appl. Energy 100, 205e217.
Domestic Product [GDP]. http://www.bea.gov/national/. (Accessed 10 June
Li, J., Dong, X., Shangguan, J., Hook, M., 2011. Forecasting the growth of China's
2010).
natural gas consumption. Energy 36 (3), 1380e1385.
U.S. Department of Agriculture, Economic Research Service, 2016. Agricultural
Lin, B., Long, H., 2015. A stochastic frontier analysis of energy efficiency of China's
Output. http://www.ers.usda.gov/data-products/agricultural-productivity-in-
chemical industry. J. Clean. Prod. 87 (C), 235e244.
the-us.aspx. (Accessed 10 June 2010).
Lin, B., Xie, X., 2015. Factor substitution and rebound effect in China's food industry.
U.S. Department of Agriculture, Economic Research Service, 2016. Agricultural
26 Y. Xu, J. Szmerekovsky / Journal of Cleaner Production 165 (2017) 13e26
Research Funding in the Public and Private Sectors. http://www.ers.usda.gov/
data-products/agricultural-research-funding-in-the-public-and-private-
sectors.aspx. (Accessed 10 June 2010).
U.S. Department of Agriculture, Economic Research Service, 2015. Per Capita Food
Expenditure. http://www.ers.usda.gov/data-products/food-expenditures.aspx.
(Accessed 10 June 2010).
U.S. Department of Agriculture, Economic Research Service, 2016. Price Spread.
http://www.ers.usda.gov/media/308011/aer780h_1_.pdf. (Accessed 10 June
2010).
U.S. Department of Agriculture, 2015. USDA and EPA Join with Private Sector,
Charitable Organizations to Set Nation's First Food Waste Reduction Goals.
Retrieved January 1, 2016, from. http://www.usda.gov/wps/portal/usda/
usdahome?contentid¼2015/09/0257.xml&navid¼NEWS_
RELEASE&navtype¼RT&parentnav¼LATEST_RELEASES&edeployment_
action¼retrievecontent.
U.S. Department of Commerce, Bureau of Economic Analysis (BEA), 2016.
Benchmark Input-Output Accounts (Various Years). http://www.bea.gov/
industry/index.htm#benchmark_io. (Accessed 10 June 2010).
U.S. Department of Energy, Energy Information Administration. Annual Energy
Review (various years), www.eia.doe.gov/aer/. Accessed 10 June 2016.
U.S. Department of Labor, Bureau of Labor Statistics (BLS), 2016. Customer Price
Index (CPI). http://data.bls.gov/pdq/SurveyOutputServlet. (Accessed 10 June
2010).
Wallgren, C., Mattias, H., 2009. Eating energy d identifying possibilities for reduced
energy use in the future food supply system. Energy Policy 37, 5803e5813.
Yang, W., Shi, J., Qiao, H., Shao, Y., Wang, S., 2016. Regional technical efficiency of
Chinese iron and steel industry based on bootstrap network data envelopment
analysis. Socio Econ. Plan. Sci. 57, 1e11.
Zhang, X.C., Ma, C., Zhan, S.F., Chen, W.P., 2012. Evaluation and simulation for
ecological risk based on emergy analysis and pressure-state-response model in
a coastal city, China. Procedia Environ. Sci. 13, 221e231.