Quality in Competitive Fresh Produce Supply Chains

with
Application to Farmers’ Markets

Deniz Besik and Anna Nagurney

Department of Operations and Information Management
Isenberg School of Management
University of Massachusetts, Amherst, Massachusetts 01003

September 2016; revised December 2016; revised March 2017

Socio-Economic Planning Sciences (2017), 60, pp 62-76.

Abstract: Fresh produce supply chains have special characteristics, notably, that the qual-
ity of the product (fruit or vegetable) deteriorates continuously over time, even under ideal
conditions. In this paper, we begin with explicit formulae for fresh produce quality dete-
rioration based on chemistry and temperature and provide a path-based framework. We
then focus on farmers’ markets, the popularity of which has been growing due to consumers’
greater awareness of and interest in product quality and emphasis on health. Farmers’ mar-
kets, as examples of direct to consumer channels and shorter supply chains, are studied in the
framework of game theory in both uncapacitated and capacitated versions. A case study of
apples in Massachusetts, under various scenarios, including production disruptions, provides
quantitative evidence of the applicability of our supply chain network approach.

Key words: quality, food supply chains, fresh produce, oligopolistic competition, food
deterioration, product differentiation

1
1. Introduction

Food is essential to life and to well-being. Fresh produce in the form of fruits and
vegetables, in particular, is noted for its nutrients in terms of vitamins and minerals and
is an important component of healthy diets of both children and adults alike (O’Connor
(2013)). Eating nutritious foods can reduce the incidence of illnesses and malnutrition and
can even prolong life. Children who are well-nourished are better able to concentrate and to
learn and grow. Adults who lack food security cannot prosper and neither can their families.
According to Neff et al. (2009) food hardship is closely correlated with obesity and related
diseases of the heart, diabetes, and cancers.

Knowledgeable modern consumers are increasingly demanding high quality in their food
products, including fresh produce, and, yet, they may be unaware of the great distances the
food has traveled through intricate supply chains and the length of time from the initial
production or “picking” of the fruits and vegetables to the ultimate delivery. Moreover,
consumers, faced with information asymmetry, may not know how long the food may have
been lying on the grocers’ and retailers’ shelves, even once delivered and unpacked. The
great distances traveled create issues in terms of quality since fresh produce is perishable
(see Nahmias (2011) and Nagurney et al. (2013)). As noted in Yu and Nagurney (2013),
food supply chains are distinct from other product supply chains in that the quality of food
products is decreasing with time, even with the utilization of the most advanced facilities and
conditions (Sloof, Tijskens, and Wilkinson (1996) and Zhang, Habenicht, and Spieß (2003)).

The transformation of global food supply chains since the early 1900s has been nothing
short of remarkable. As reported in Martinez et al. (2010), in the early 1900s, much of
the food bought and consumed in the United States was grown locally and about 40% of
Americans resided on farms, whereas in 2000 only 1% did (cf. Pirog (2009)). Consumers
over a century ago obtained information as to the quality of the foods through direct contact
with farmers. Except for various food preservation activities, few foods were processed or
packaged, and fresh produce, fish, and dairy products usually traveled less than 24 hours to
market (see Giovannucci, Barham, and Pirog (2010)). Foods were consumed based on their
availability and growing seasons.

According to a report by DeWeerdt (2016), in 1993, a Swedish researcher determined that

the ingredients of a typical Swedish breakfast consisting of an apple, bread, butter, cheese,
coffee, cream, orange juice, and sugar, had traveled a distance equal to the circumference
of our planet before reaching the consumer. In 2005, Pirog and Benjamin, through the
use of a mathematical formula, computed that the milk, sugar, and strawberries that make

2
up a container of strawberry yogurt jointly journeyed 2,216 miles only to the processing
plant. According to Pirog et al. (2010), calculations made by Hendrickson (1997) examining
transportation and fuel requirements estimated that fresh produce in the United States
traveled an estimated 1,500 miles with fresh produce arriving in Austin, Texas estimated to
have traveled an average of 1,129 miles.

Consumers tend to connect the terms ‘fresh,’ ‘good quality,’ and ‘tasty’ to the products
that are being locally produced (Holloway and Kneafsey (2000)). The definition of the
term ‘local’ varies between regions and communities (Martinez (2010)); however, the most
common definition is that local food is food grown within a county or within a state (Wilkins,
Bokaer-Smith, and Hilchey (1996)). According to a national survey conducted by Bond,
Thilmany, and Bond (2009), four out of five people living in the United States buy local
food. Thus, the demand for locally grown fresh produce is also increasing and demand and
supply management becomes more crucial for farmers. Successful harvest and post-harvest
decisions can help farms to meet growing demand in local fresh produce markets.

It has now been established that the quality of food, including that of fresh produce, along
with quality preservation and degradation, can be determined using chemical formulae, which
include both time and temperature (cf. Labuza (1982), Taoukis and Labuza (1989), Tijskens
and Polderdijk (1996), Rong, Akkerman, and Grunow (2011)). Hence, in order to capture
quality associated with fresh produce in a supply chain one must be aware of the various
supply chain network economic activities such as “production,” storage, transportation, etc.,
as well as the durations and temperatures associated with these activities. For example, in
terms of kinetics (cf. Labuza (1982)), the quality degradation of food such as meat and fish
follows first order reactions whereas that of many fresh fruits and vegetables follows zero-
order reactions with the order of the reaction corresponding to the power of the differential
equation for quality.

Consequently, the network topology of a fresh produce supply chain and, in particular, the
length of a path in terms of time from an origin node to a destination node can significantly
influence the quality of the fresh produce that consumers purchase and consume. This
gives impetus to the investigation of local food systems and associated shorter marketing
channels. For example, a U.S. Congressionally mandated report by the U.S. Department of
Agriculture (cf. Low et al. (2015)) noted that the sale of food through direct-to-consumer
channels, such as through farmers’ markets, and intermediated marketing channels, that is,
sales to institutions, including hospitals and schools, or regional distributors, appears to be
increasing. As reported therein, there were 8,268 farmers’ markets in the United States in
2014, with the number having increased by 180% since 2006. 163,675 farms, or 7.8% of the

3
U.S. ones, were marketing foods locally in 2012 and of these farms, 70% used only direct-to-
consumer channels. This growth implies increasing consumer interest in locally sourced fresh
produce due to quality and health perspectives. Interestingly, and perhaps paradoxically, as
noted by Babiak (2013), concurrent with the growth in farmers’ markets across the United
States, food insecurity has reached epidemic levels with nearly 1 in 5 Americans, or 47.8
million people, relying on food stamp benefits.

Various authors have emphasized the change in quality of food products in the supply
chain until the final points of demand (see Sloof, Tijskens, and Wilkinson (1996), Van der
Vorst (2000), Lowe and Preckel (2004), Ahumada and Villalobos (2009, 2011), Blackburn
and Scudder (2009), Akkerman, Farahani, and Grunow (2010), and Aiello, La Scalia, and
Micale (2012), Yu and Nagurney (2013)). Amorim, Costa, and Almada-Lobo (2014) utilize
demand functions that depend on product quality and also price and then construct demands
for different products based on age. They propose deterministic and stochastic production
planning models that capture consumers’ desire for fresher products. Liu, Zhang, and Tang
(2015) also utilize demand functions that depend on price and quality but they depend
continuously on time. The authors determine the dynamic pricing and investment strategies
to reduce the deterioration rate of the quality for perishable foods. However, there has been
only limited research done in terms of farmers’ markets, which are examples of short supply
chains (cf. Fabbrizzi, Menghini, and Marinelli (2014) for a review), and provide an excellent
setting in which quality can be captured over time since consumers can obtain information
directly from the producers, that is, the farmers.

Moreover, optimization has been the primary methodological framework, in the case of
food supply chain model settings, including longer ones, whereas it is clear that farmers
compete in food supply chains including in direct to consumer chains as in farmers’ markets.
Our supply chain model utilizes game theory. Farmers’ operational costs can depend on
their fresh produce flows as well as those of the others since they compete for resources.
Furthermore, demand prices are differentiated by brand since consumers reveal their pref-
erences through their demands for the quality of the individual farmer’s produce as well as
that of the others’. Hence, in order to capture competition, both operational costs as well
as demand price functions depend on vectors of variables. The work of Yu and Nagurney
(2013) is one of the few that proposes a game theory model for oligopolistic competition in
brand differentiated fresh produce supply chains with perishability. This paper differs from
that paper is multiple significant ways. It models fresh produce quality through kinetics,
and, hence, quality, which is time-dependent, is based on actual physical and chemical pa-
rameters. The quality of the fresh produce that arrived at the demand markets in the model

4
of Yu and Nagurney (2013) was identical. Also, both uncapacitated and capacitated versions
are presented here.

The majority of research on local food markets, especially farmers’ markets, pertains
mostly to consumer behavior and there are not many mathematical models that consider
demand or the supply side in local food supply chains. The literature on consumer behavior in
farmers’ markets provides valuable insights from the consumers’ and farms’ perspectives. For
instance, Holloway and Kneafsey (2000) argue that the quality criteria of the consumers can
be met by assigning food products to new consumption places, specifically, farmers’ markets.
According to the authors, consumers in the United Kingdom regard farmers’ markets as an
alternative space for supermarkets and even find them nostalgic. Asebo et al. (2007) have
also worked on farmer and consumer attitudes at farmers’ markets in Norway. They found
that, in Norway, consumers consider the origin of the food and the production processes
to be significant. Farmers’ markets fit perfectly into these considerations, since consumers
can meet the farmers face-to-face and learn about the production processes. Farmers also
benefit from farmers’ markets by being able to keep more of the retail price to themselves,
passing the middleman in the mainstream supply chain and meeting face to face with their
customers (Trobe (2001)). Besides, 19,000 farmers use farmers’ markets as their only sale
channel in the United States. Between 1998 and 2009, the number of farmers’ markets
increased by 92 percent, which reveals the positive tendency of consumers towards farmers’
markets. Therefore, it is important to reinforce the relationship between farms and the
consumers in local food markets, by proposing solutions to increase the profitability of farms
and consequently providing higher quality fresh food to consumers. However, despite the
significant interest in local food, a supply chain network model, which combines the quality
criteria of consumers and offers strategies for farms to increase their profits in farmers’
markets, does not exist in the literature.

Tong, Ren, and Mack (2012) present several of the few mathematical models for farmers’
markets, focusing on site selection using location theory in a spatio-temporal context and
provide a case study in Tucson, Arizona to illustrate the benefits of their approach. They
capture the distance that consumers may need to travel to the farmers’ markets and also
include the times of operation. They propose two types of linear mathematical programming
models to locate farmers’ markets, which are also extensions of the p-median problem in the
literature. In the first model, they propose an optimal location for farmers’ markets, while
minimizing the total overall additional travel distance of consumers. The additional travel
distance of an individual to a farmers’ market, in a time-frame, is represented by a parameter
in the objective function. The authors differentiate consumers as worker and non-workers in

5
the second model and suggest an optimal site for farmers’ markets. In addition, Bosona et
al. (2011) investigate the local food supply chain in Sweden, and propose alternative routes
to optimize transportation between farms and demand points. Furthermore, Aregu (2014),
from the Netherlands, creates a software called ‘Farmers Decision Enhancement Studio’ for
farmers to enhance their pricing decisions in farmers’ markets in Uganda.

In this paper, we develop a modeling and algorithmic framework for competitive farmers’
markets. The model is network-based and the farmers engage in Cournot competition over
space and time. The governing Nash equilibrium conditions are formulated as a variational
inequality problem. The novelty of the framework lies in that the quality of the fresh
produce product is captured as the produce propagates in the supply chain over space and
time with the consumers at the markets responding both to the price and the quality of
the fresh produce. Both uncapacitated and capacitated versions of the model are presented.
The latter can capture limitations in supply due to harvest problems or damage during the
growing season, limitations in transport and storage capabilities, and/or labor for harvesting
and processing. The game theory model can address questions of farmers as to which farmers’
markets they should serve; what the impact of a new competitor may be at one or more
markets in terms of profits, as well as the effects of capacity disruptions (or enhancements)
in their individual and others’ short supply chains. In addition, the model can ascertain the
impacts of changes in link parameters that capture quality. Policy makers, in turn, can also
obtain useful information as to the impacts of a greater number or fewer farms represented
at various markets and how reducing quality decay can affect farmers’ profits.

The paper is organized as follows. In Section 2, we provide some preliminaries, focusing on

the quality differential equations, which we then generalize to a path concept since the fresh
produce will proceed on multiple supply chain links from the harvesting point to ultimate
purchase at the farmers’ markets. Each link has associated with it both a time element as
well as a factor such as temperature, which also affects the fresh produce quality.

In Section 3, we first present the uncapacitated competitive supply chain network model
for farmers’ markets, give the governing Nash equilibrium conditions, and derive the varia-
tional inequality formulations in path flows and links flows. In addition, we construct the
capacitated path analogue and provide its variational inequality formulation under Nash
equilibrium. In Section 4, we present the algorithm, which resolves the variational inequal-
ity problems into subproblems in the path flows (and in the case of the capacitated model,
also in terms of Lagrange multipliers) that can be solved explicitly using formulae that we
present. In Section 5 we present a case study on apples in farmers’ markets. We summarize
and present our conclusions in Section 6.

6
2. Preliminaries on Food Quality Deterioration

Before we present the fresh food supply chain network model for farmers’ markets, it is
important to provide some preliminaries on the definition of quality, and the kinetics behind
the quality decay, since quality has major relevance for the fresh food industry. Afterwards,
we provide the generalization of the quality differential equations for fresh produce supply
chain networks by capturing them through the path concept.

Quality can be defined in multiple dimensions such as physical, emotional, and even
philosophical. Therefore, it is not easy to find a proper global definition of quality which
is valid for every type of industry and consumer. Reeves and Beednar (1994) present a
good discussion on the evolution of quality definitions. Nagurney and Li (2014) investigate
minimum quality standards in a spatial price equilibrium model and also consider information
asymmetry. Nagurney and Li (2016) present a plethora of competitive supply chain network
models with a focus on quality, where quality is defined as conformance to specifications as in
manufactured products. Nagurney et al. (2013), in turn, focus on perishable product supply
chains. Murdoch, Marsden, and Banks (2000) also present a valuable discussion on the
connection between quality and nature, especially in the food sector, from the perspective of
social sciences. However, there are more specific definitions of quality in the literature that
are construed specifically for the food sector. According to Kader (1997), the quality of fresh
fruits can be defined over the combination of their attributes, properties, or characteristics.
Furthermore, these attributes can be categorized by: color and appearance, flavor (taste and
aroma), texture, and nutritional value (Barrett, Beaulieu, and Shewfelt (2010)). Consumers
make the decision of purchase based on their sensory evaluations which consist of sight,
smell, taste, touch, and hearing (Abbott (1994)). Moreover, reports suggest that consumers
consider the quality attributes of appearance, firmness, freshness, and ripeness during the
initial phase of purchase (see Kader (1997) and Zind (1990)). Therefore, retaining the quality
of these fresh produce attributes is very important for both consumers and farmers.

Since fresh foods are biological products, they deteriorate and, consequently, lose quality
over time (Schouten et al. (2004), Singh and Anderson (2004)). Thus, an understanding
of the biochemical/physicochemical reactions, which cause the deterioration, is necessary in
order to be able to present an explanation of the quality loss (Singh and Cadwallader (2004)).
According to Taoukis and Labuza (1989), the rate of quality deterioration can be given as a
function of its microenvironment, gas composition, relative humidity, and temperature.

In this regard, Labuza (1984) captures the quality decay of a food attribute, Q, over time

7
t, through the differential equation:
∂Q
= kQn = Ae(−E/RT ) Qn . (1)
∂t

Here, k is the reaction rate and is defined by the Arrhenius formula, Ae(−E/RT ) , where A
is a pre-exponential constant, T is the temperature, E is the activation energy, and R is the
universal gas constant (Arrhenius (1889)). Moreover, n is the reaction order, which is a non-
negative integer and belongs to the set Z∗ = {0} ∪ Z+ . In general, the quality decay function
of the food attribute can be shown in terms of its reaction order. When the reaction order
∂Q
n is zero, that is, ∂t
= k, the quality decay rate of the food attribute Q can be expressed as
the function:
f0 (Q) = −kt. (2)
Furthermore, assuming that the initial quality is known and given as Q0 , we can define the
remaining quality Qt at time t, by using the zero order quality decay function as in Tijskens
and Polderdijk (1996):
Qt = Q0 + f0 (Q) = Q0 − kt. (3)

Recall that the reaction constant k can be written by the Arrhenius formula to show
the relationship between temperature and quality decay explicitly. Furthermore, having
the reaction order be 1 – also referred to as a first order reaction – leads to an exponential
function, which is observed commonly in food quality decay (Tijskens and Polderdijk (1996)).
This type of quality decay is given by the expression:

f1 (Q) = e−kt . (4)

Notice that, since this quality decay function has an exponential component, the quality
Qt at time t should be written as a multiplication of the initial quality Q0 and the quality
decay function, as shown in the following expression:

Qt = Q0 f1 (Q) = Q0 e−kt . (5)

Table 1 shows the decay kinetics, with the related quality attributes, of some fruits and
vegetables. For a detailed description of the quality decay kinetics for vegetables, see Aamir
et al. (2013). Notice that the quality attributes, in Table 1, are related to the appearance
and texture which are the two most important fresh produce characteristics for consumers.

We can now generalize each type of quality deterioration function in terms of the path
concept in a given fresh produce food supply chain network. We define a path p joining an

8
 Attribute Fresh Reaction Reference
Produce Order
Color Change Peaches First Toralles et al. (2005)
Color Change Raspberries First Ochoa et al. (2001)
Color Change Blueberries First Zhang, Guo, and Ma (2012)
Nutritional (Vitamin C) Strawberries First Castro et al. (2004)
Color Change Watermelons Zero Dermesonlouoglou, Giannakourou,
and Taoukis (2007)
Moisture Content Tomatoes First Krokida et al. (2003)
Color Change Cherries First Ochoa et al. (2001)
Texture Softening Apples First Tijskens (1979)
Nutritional (Vitamin C) Pears First Mrad et al. (2012)
Texture Softening Avocados First Maftoonazad and Ramaswamy (2008)
Nutritional (Vitamin C) Pineapples First Karim and Adebowale (2009)
Color Change Spinach Zero Aamir et al. (2013)
Color Change Asparagus First Aamir et al. (2013)
Color Change Peas First Aamir et al. (2013)
Texture Softening Beans First Aamir et al. (2013)
Texture Softening Brussel Sprouts First Aamir et al. (2013)
Texture Softening Carrots First Aamir et al. (2013)
Texture Softening Peas First Aamir et al. (2013)
Color Change Coriander Leaves First Aamir et al. (2013)

Table 1: Fresh Produce Attributes and Decay Kinetics

origin node i with a destination node j through directed links that it is comprised of in the
link set L. Furthermore, let βa denote the quality decay incurred on link a, which depends
on the reaction order n, reaction rate ka , and time ta on link a, according to:

 −ka ta , if n = 0, ∀a ∈ L,
βa ≡ (6)
 −ka ta
e , if n 6= 0, ∀a ∈ L.
Here, ka is the reaction constant related to the link a. Since each link on a path can have
different temperature conditions, the differentiation over the temperature of the links is
necessary. Thus, the reaction rate is described in the following equation for each link a by
the Arrhenius formula with the same parameters as in (1), except that the temperature is
now denoted for each link a as Ta , where:

ka = Ae(−EA /RTa ) . (7)

In addition, since (7) contains the exponential decay as a function and the other terms are
parameters, βa is defined generally for the reaction orders greater than zero. Before we show

9
the quality decay over a given path p, it is worthwhile to introduce the quality parameter
q0i , which represents the initial quality of the fresh produce, produced at the origin node i.
Also, Pji represents the set of all paths that have origin i and destination j. We now can
define the quality qp , over a path p, joining the origin farm node, i, with a destination node
farmer market, j, while incorporating the quality deterioration of the fresh produce as:
 X


 q 0i + βa , if n = 0, ∀a ∈ L, p ∈ Pji , ∀i, j,
a∈p


qp ≡ Y (8)
i

 q0i


 βa , if n = 1, ∀a ∈ L, p ∈ Pj , ∀i, j.
a∈p

By the above generalization, we differentiate our model from the existing works in the liter-
ature. Although we utilize Tijskens and Polderdijk (1996)’s quality decay formulations, the
authors have not provided a quality deterioration formulation for a path in a food supply
chain network. Rong, Akkerman, and Grunow (2011) build the food supply chain through a
mixed-integer programming model and incorporate the food quality decay into this model.
However, in this paper, we present a competitive supply chain game theory model where
the demand markets and the quality decay are product specific. Furthermore, Ketzenberg,
Bloemhof, and Gaukler (2015) present a model to assess the value of time and temperature
information of a perishable product. The authors provide a simulation study to evaluate the
remaining shelf life of fresh fish with randomly realized temperature and time information.
In their work, they provide a different food quality metric than ours which is not based on
quality kinetics and is not applied on a supply chain network.

We are now ready to present the competitive fresh produce supply chain network model
for farmers’ markets.

10
3. The Fresh Produce Farmers’ Market Supply Chain Network Models

In this section, we capture the economic behavior of farmers selling at farmers’ markets
and competing on quality and quantity of their fresh produce a la Cournot (1838) through
both an uncapacitated and a capacitated oligopoly model and we derive the variational
inequality formulations of the governing Nash (1950, 1951) equilibrium conditions. The
models consist of a finite number of I farms, that are run by farmers, typically, denoted by i,
and a finite number of demand points, M , which correspond to farmers’ markets, and with
a typical one denoted by j. Each farmers’ market takes place in a region and on a given day
of the week. We assume that the demand points correspond to farmers’ markets on different
days of the week since farmers may not have sufficient staff to be at multiple markets on the
same day. Farmers’ markets, on any given day, usually last no more than 6 hours and are
repeated in the same location and day over a season, which, in the Northeast of the United
States, for example, can last from May through October.

The uncapacitated model is presented in Section 3.1 and its capacitated analogue then
outlined in Section 3.2.

Before introducing the variables and the notation for the models, we provide a summary
of the assumptions that we have made:

1. The supply chain network topology illustrates a fixed time horizon in a given season
of the fresh fruit or vegetable, typically, a week.

2. The demand points represent the direct-to-consumer local farmers’ markets, in different
regions of a county or a state.

3. Farmers pick the harvest right before the beginning of the time horizon; therefore, a
farmers’ market that is available at the beginning of this time horizon sells the fresh
produce without storage. Farmers’ markets taking place on subsequent days store the
fresh produce before transporting it to the markets.

4. Consumers can buy products that are substitutes of one other within or across the
demand points. That means that the farmers’ markets provide multiple options of
fresh produce to consumers over space and time.

Since the local supply chains are short, the intermediate activities can be defined as:
harvesting, processing and packaging, storage, and transportation. Farmers compete at
the local farmers’ markets noncooperatively in an oligopolistic manner. The products are

11
differentiated at the markets according to their quality, which are presented by the quality
deterioration functions described in Section 2. In this regard, the corresponding fresh produce
food supply chain network topology is as in Figure 1. Based on the case study of concern
here, that is, apples, the time horizon of a week and the picking of apples once a week
is not unreasonable, according to experts that we have interviewed (Clements (2016a) and
Drew (2016)). If more choices were to be added in terms of multiple times to pick, for
example, then the network would be adapted/extended accordingly. Here our goal is to lay
the foundations for the modeling of both short and long food supply chain networks under
competition and with quality deterioration based on kinetics on paths consisting of distinct
economic activities such as transport, storage, etc.
Farms
1k · · · ik · · · Ik
Harvesting
k k k
? ? ?
··· ···
Processing/Packaging
k ··· k ··· k
? ? ?
aa  Q ! !


A aa 
Q !! A
QStorageA

A a
!! Q  A
 a a!a 
Transportation

AA
 k


!! k k k
U !
!

+

aa Qs k AU k
Q
 !

 ! Transportation
 a aaQ
A
 aQ A
!!

!  aQ
1k 2k Mk
!
+


AU


+ s
Q
a AU
···
Farmers’ Markets

Figure 1: The Supply Chain Network Topology of the Farmer’s Market Competitive Fresh
Produce Problem

Let G = [N, L] denote the graph with set of nodes N and links L in Figure 1. Each
farm i; i = 1, . . . , I, delivers fresh produce to the markets through its harvesting, processing,
storage, and transportation facilities, and seeks to determine its optimal strategies in terms
of how much of the produce to bring to each market in order to maximize profit. The link
set L consists of the sets of links L1 ∪ L2 . . . ∪ LI , where Li is the sequence of directed links of
farm i. Moreover, the links on the network correspond to the economic activities associated
with the farms.

The first set of directed links illustrates the harvesting/picking activity of each farmer
i; i = 1, . . . , I. The second set of directed links illustrates the processing activities such as
cleaning, sorting and labeling, and also packaging/packing of the fresh produce. The last
set of directed links corresponds to the storage and transport of the fresh produce. If the

12
demand point is a farmers’ market that is open on the first day of the week, the links only
capture the transportation cost. Otherwise, there is a storage link followed by a transport
link each subsequent demand point, that is, farmers’ market, as illustrated in Figure 1.

Recall that Pji denotes the set of paths joining farm node i with farmer market node j. In
addition, we let P i denote the set of paths emanating from farm node i to all farmer market
nodes j; j = 1, . . . , M . P then denotes the set of all paths in the supply chain network in
Figure 1. There are nP paths in the network and nL links.

3.1 The Uncapacitated Model

We now present the uncapacitated model.

The flow on each path, joining the farmer node i to the farmers’ market node j, is denoted
by xp , and all path flows must be nonnegative, that is,

xp ≥ 0, ∀p ∈ Pji ; i = 1, . . . , I; j = 1, . . . , M. (9)

Furthermore, the flow on a link a is equal to the sum of the path flows xp , on paths that
include the link a. This conservation of flow equation is expressed as:
X
fa = xp δap , ∀a ∈ L, (10)
p∈P

where δap is equal to 1 if the link a is included in the path p, and 0, otherwise. In the local
food supply chain network, there is usually a single path between farm i and the market j;
however, the path set Pji is introduced to present a more general definition.

Furthermore, the demand at the farmers’ market j for the fresh produce product of farmer
i is given by:
X
xp = dij , p ∈ Pji ; i = 1, . . . , I; j = 1, . . . , M. (11)
p∈Pji

We group the demands between all farms and farmer market pairs (i, j) into the vector
d and also the quality of their fresh produce on paths p ∈ Pji , ∀i, j, {qp }, into the vector q.
We assume that all vectors are column vectors.

The price of the product of farm i, in turn, at farmers’ market j, is denoted by ρij , and
depends not only on the demand for the farm’s fresh produce, but also on the quality of that
product, which we capture by the quality decay kinetics in (11). Also, the price depends,
in general, on the quantities of the competitors’ fresh produce at the markets as well as the
quality of their products.

13
 Hence, the demand price function ρij for farm i’s product at the farmers’ market j, is:

ρij = ρij (d, q), i = 1, . . . , I; j = 1, . . . , M. (12)

In view of (11) and (12), the demand price functions can be rewritten as:

ρ̂ij = ρ̂ij (x, q) ≡ ρij (d, q), i = 1, . . . , I; j = 1, . . . , M. (13)

The demand price functions are assumed to be continuous, continuously differentiable, and
monotone decreasing. Recall that the quality parameter vector q is defined with respect
to the decay function of the fresh produce which farms sell at the farmers’ markets and is
time-dependent. Thus, the demand price for a farm’s product is higher when the quality of
its fresh produce product is high and, if the demand is high, the price is lower. Note that
each farm is aware of the temperature conditions and the time associated with the links in
its supply chain network.

The competition among the farms for resources is reflected in the total operational costs
incurred in harvesting, processing, storage, and transportation. The quality-keeping cost
is related to the food deterioration and is also considered through the transportation and
the storage link costs. In general, the total operational cost of each link a, denoted by ĉa ,
depends on the flows on all the links in the fresh produce supply chain network, that is,

ĉa = ĉa (f ), ∀a ∈ L, (14)

where f is the vector of all the link flows. The total cost on each link is assumed to be
convex and continuously differentiable.

Let Xi denote the vector of path flows associated with farm i; i = 1, . . . , I, where Xi ≡
n
{{xp }|p ∈ P i }} ∈ R+P i , P i ≡ ∪j=1,...,M Pji , and nP i denotes the number of paths from farm
i to the farmers’ markets. Thus, X is the vector of all the farmers’ strategies, that is,
X ≡ {{Xi }|i = 1, . . . , I}.

The profit function of farm i is defined as the difference between its revenue and its total
costs, where the total costs are the total operational costs over Li . The profit function, which
we present is novel in terms of capturing the quality of the fresh produce in the demand price
functions, which is time-dependent.

The profit/utility function of farm i, denoted by Ui , is given by:

M
X X
Ui = ρij (d, q)dij − ĉa (f ). (15)
j=1 a∈Li

14
 In lieu of the conservation of flow expressions (10) and (11), and the functional expressions
(13) and (14), we can define Ûi (X) ≡ Ui for each farm i; i = 1, . . . , I, with the I-dimensional
vector Û being the vector of the profits of all the farms with respect to their earnings at the
farmers’ markets over a typical week:

Û = Û (X). (16)

We now state the fresh produce supply chain network Cournot-Nash Equilibrium condi-
tions for farmers’ markets in a region.

Definition 1: Fresh Produce Supply Chain Network Cournot-Nash Equilibrium

for Farmers’ Markets in the Uncapacitated Case

A path flow pattern X ∗ ∈ K = Ii=1 Ki constitutes a fresh produce supply chain network
Q

Cournot-Nash equilibrium if for each farm i; i = 1, . . . , I:

Ûi (Xi∗ , X̂i∗ ) ≥ Ûi (Xi , X̂i∗ ), ∀Xi ∈ Ki , (17)

n
where X̂i∗ ≡ (X1∗ , . . . , Xi−1
∗ ∗
, Xi+1 , . . . , XI∗ ) and Ki ≡ {Xi |Xi ∈ R+P i }.

A Cournot-Nash Equilibrium is established if no farm can unilaterally improve its profit

by changing its product flows throughout its supply chain network, given the product flow
decisions of the other farms.

Next, we derive the variational inequality formulations of the Cournot-Nash equilibrium

for the fresh produce supply chain network under oligopolistic competition satisfying Defi-
nition 1, in terms of both path flows and link flows (see, e.g., Cournot (1838), Nash (1950,
1951), Gabay and Moulin (1980), Nagurney (2006), and Nagurney et al. (2013)).

Theorem 1: Variational Inequality Formulations of the Uncapacitated Model

Assume that for each fresh produce farm i; i = 1, . . . , I, the profit function Ûi (X) is concave
with respect to the variables in Xi and is continuously differentiable. Then X ∗ ∈ K is a fresh
produce supply chain network Cournot-Nash equilibrium for farmers’ markets according to
Definition 1 if and only if it satisfies the variational inequality:
I
X
− h∇Xi Ûi (X ∗ ), Xi − Xi∗ i ≥ 0, ∀X ∈ K, (18)
i=1

where h·, ·i denotes the inner product in the corresponding Euclidean space and ∇Xi Ûi (X)
denotes the gradient of Ûi (X) with respect to Xi . Variational inequality (18), in turn, for

15
our uncapacitated model, is equivalent to the variational inequality that determines the vector
of equilibrium path flows x∗ ∈ K 1 such that:
 
I XM X ∗ M ∗
 ∂ Ĉp (x ) − ρ̂ij (x∗ , q) − ∂ ρ̂il (x , q)
X X X
x∗r  ×[xp −x∗p ] ≥ 0, ∀x ∈ K 1 , (19)
i=1 j=1 i
∂x p
l=1
∂x p i
p∈Pj r∈Pl

nP
where K 1 ≡ {x|x ∈ R+ }, and for each path p; p ∈ Pji ; i = 1, . . . , I; j = 1, . . . , M ,

∂ Ĉp (x) X X ∂ĉb (f ) ∂ ρ̂il (x, q) ∂ρil (d, q)

≡ δap and ≡ . (20)
∂xp i i
∂fa ∂xp ∂dij
a∈L b∈L

Variational inequality (19) can also be rewritten in terms of link flows as: determine the
vector of equilibrium link flows and the vector of equilibrium demands (f ∗ , d∗ ) ∈ K 2 , such
that: " #
I X X
X ∂ĉb (f ∗ )
× [fa − fa∗ ]
i=1 a∈Li b∈Li
∂f a

I X M
" M
#
∗
X X ∂ρ il (d , q)
+ −ρij (d∗ , q) − d∗il × [dij − d∗ij ] ≥ 0, ∀(f, d) ∈ K 2 , (21)
i=1 j=1 l=1
∂d ik

where K 2 ≡ {(f, d)|x ≥ 0, and (10) and (11) hold}.

Proof: (18) follows from Gabay and Moulin (1980); see, also, Masoumi, Yu, and Nagurney
(2012). (19) and (21) then follow using algebraic substitutions. 2

Variational inequalities (19) and (21) can be put into standard form (see Nagurney
(1999)): determine X ∗ ∈ K such that:

hF (X ∗ ), X − X ∗ i ≥ 0, ∀X ∈ K, (22)

where h·, ·i denotes the inner product in N -dimensional Euclidean space. Let X ≡ x and
 M
∂ Ĉp (x) X ∂ ρ̂il (x, q) X
F (X) ≡ − ρ̂ij (x, q) − xr ;
∂xp l=1
∂xp i r∈Pl

i
p ∈ Pj ; i = 1, . . . , I; j = 1, . . . , M , (23)

K ≡ K 1 , and N = nP , then (19) can be re-expressed as (22). Similarly, for the variational
inequality in terms of link flows, if we define the column vectors: X ≡ (f, d) and F (X) ≡
(F1 (X), F2 (X)), where
" #
X ∂ĉb (f )
F1 (X) = ; a ∈ Li ; i = 1, . . . , I ,
i
∂f a
b∈L

16
  
M
X ∂ ρ̂il (x, q) X
F2 (X) = −ρ̂ij (x, q) − xr ; p ∈ Pji ; i = 1, . . . , I; j = 1, . . . , M  , (24)
∂xp
l=1 r∈Pli

K ≡ K 2 , and N = nL + M , then (21) can be re-written as (22).

Since the feasible set K 1 is not compact, and the same holds for K 2 , we cannot obtain the
existence of a solution simply based on the assumption of the continuity of F . However, the
demand dij for each farm i’s product; i = 1, . . . , I, at every farmers’ market j; j = 1, . . . , M ,
may be assumed to be bounded, since the population requiring these products is finite
(although it may be large). Consequently, in light of (12), we have that:

Kc ≡ {x| 0 ≤ x ≤ c, }, (25)

where c > 0 and x ≤ c means that xp ≤ c for all p ∈ Pji ; i = 1, . . . , I, and j = 1, . . . , M .

Then Kc is a bounded, closed, and convex subset of K 1 . Thus, the following variational
inequality
hF (X c ), X − X c i ≥ 0, ∀X ∈ Kc , (26)
admits at least one solution X c ∈ Kc , since Kc is compact and F is continuous. Therefore,
following Kinderlehrer and Stampacchia (1980) (see also Theorem 1.5 in Nagurney (1999)),
we have the following theorem:

Theorem 2: Existence

There exists at least one solution to variational inequality (19) (equivalently, to (21)), since
there exists a c > 0, such that variational inequality (29) admits a solution in Kc with

xc ≤ c. (27)

In addition, we now provide a uniqueness result.

Theorem 3: Uniqueness

With Theorem 2, variational inequality (26) and, hence, variational inequalities (19) and
(21) admit at least one solution. Moreover, if the function F (X) of variational inequality
(21), as defined in (22), is strictly monotone on K ≡ K 2 , that is,

h(F (X 1 ) − F (X 2 )), X 1 − X 2 i > 0, ∀X 1 , X 2 ∈ K, X 1 6= X 2 , (28)

then the solution to variational inequality (21) is unique, that is, the equilibrium link flow
pattern and the equilibrium demand pattern are unique.

17
3.2 The Capacitated Model

As noted in the Introduction, challenges that farmers may be faced with include: labor
shortages, lower yield harvests due to weather conditions, as well as lower capacities in
both storage facilities and transport, which may arise due to disruptions. Hence, in this
subsection, we provide the capacitated version of the model in Section 3.1.

Specifically, we retain the objective function (15), for each farm i; i = 1, . . . , I, the
nonnegativity constraints (9), with conservation of flow equations (11), as well as the previous
notation, but now we have that
fa ≤ ua , ∀a ∈ L, (29a)

or, in view of (10):

X
xp δap ≤ ua , ∀a ∈ L, (29b)
p∈P

where ua is the positive upper bound on the flow on link a.

We define the feasible set Ki3 faced by farm i in the capacitated case as: Ki3 ≡ {Xi |Xi ∈
n
R+P i and (29b) holds for a ∈ Li }. Also, we define K 3 ≡ Ii=1 Ki3 .
Q

We, hence, have the following definition.

Definition 2: Fresh Produce Supply Chain Network Cournot-Nash Equilibrium

for Farmers’ Markets in the Capacitated Case

A path flow pattern X ∗ ∈ K 3 constitutes a fresh produce supply chain network Cournot-Nash
equilibrium in the capacitated case if for each farm i; i = 1, . . . , I:

Ûi (Xi∗ , X̂i∗ ) ≥ Ûi (Xi , X̂i∗ ), ∀Xi ∈ Ki3 , (30)

where, as before, X̂i∗ ≡ (X1∗ , . . . , Xi−1

∗ ∗
, Xi+1 , . . . , XI∗ )

Theorem 4: Variational Inequality Formulations of the Capacitated Model

Assume that for each fresh produce farm i; i = 1, . . . , I, the profit function Ûi (X) is concave
with respect to the variables in Xi , and is continuously differentiable. Then X ∗ ∈ K 3 is a
fresh produce supply chain network Cournot-Nash equilibrium for farmers’ markets in the
capacitated case according to Definition 2 if and only if it satisfies the variational inequality:
I
X
− h∇Xi Ûi (X ∗ ), Xi − Xi∗ i ≥ 0, ∀X ∈ K 3 . (31)
i=1

18
Variational inequality (31), in turn, for our capacitated model, is equivalent to the variational
inequality that determines the vector of equilibrium path flows x∗ ∈ K 3 such that:
 
I X M X ∗ M ∗
 ∂ Ĉp (x ) − ρ̂ij (x∗ , q) − ∂ ρ̂il (x , q)
X X X
x∗r  ×[xp −x∗p ] ≥ 0, ∀x ∈ K 3 . (32)
i=1 j=1 i
∂x p
l=1
∂x p i
p∈Pj r∈Pl

Moreover, variational inequality (32) is equivalent to the variational inequality problem:

nP nL
determine (x∗ , λ∗ ) ∈ K 4 , where K 4 ≡ {x ∈ R+ , λ ∈ R+ }, such that:
 
I X M X ∗ M ∗
 ∂ Ĉp (x ) − ρ̂ij (x∗ , q) − ∂ ρ̂il (x , q)
X X X X
x∗r + λ∗a δap  × [xp − x∗p ]
i=1 j=1 i
∂xp l=1
∂xp i a∈L
p∈Pj r∈Pl

" #
X X
+ ua − x∗p δap × [λa − λ∗a ] ≥ 0, ∀(x, λ) ∈ K 4 , (33)
a∈L p∈P

∂ Ĉp (x) ∂ ρ̂il (x,q)

where ∂xp
and ∂xp
are as defined in (20).

Proof: Variational inequality (31) again follows directly from Gabay and Moulin (1980).
Variational inequality (33) follows, in turn, from Bertsekas and Tsitsiklis (1989) (see also
Nagurney (2010)) with notice that λ∗ corresponds to the vector of optimal Lagrange multi-
pliers associated with constraints (29b) and λ is the vector of Lagrange multipliers associated
with the upper bound constraints on all links a ∈ L.

We will utilize variational inequality (33) for solution of the capacitated model and note
that variational inequality (33) can be put into the standard form (22) (see Nagurney (1999)),
if the vectors: X ≡ (x, λ) and F (X) ≡ (F3 (X), F4 (X)), where
 
M
∂ Ĉp (x) X ∂ ρ̂il (x, q) X X
F3 (X) =  − ρ̂ij (x, q) − xr + λa δap , p ∈ Pji ; i = 1, . . . , I; j = 1, . . . , M  ,
∂xp l=1
∂xp i a∈L r∈Pl

" # (34)
X
F4 (X) = ua − xp δap , a ∈ L . (35)
p∈P
4
With K ≡ K , (33) can be rewritten as (22).

The existence result below follows from the classical theory of variational inequalities (cf.
Kinderlehrer and Stampacchia (1980) and Nagurney (1999)).

19
Theorem 5: Existence of a Solution to the Capacitated Model

Existence of a solution X ∗ to variational inequality (31) is guaranteed since the feasible set
K 3 is compact.

4. The Algorithm

We now recall the Euler method, which is induced by the general iterative scheme of
Dupuis and Nagurney (1993). Specifically, at an iteration τ of the Euler method (see also
Nagurney and Zhang (1996)) one computes:

X τ +1 = PK (X τ − aτ F (X τ )), (36)

where PK is the projection on the feasible set K and F is the function that enters the
variational inequality problem (22).

As shown in Dupuis and Nagurney (1993) and Nagurney and Zhang (1996), for conver-
gence of the general iterative scheme, which induces the Euler method, the sequence {aτ }
must satisfy: ∞
P
τ =0 aτ = ∞, aτ > 0, aτ → 0, as τ → ∞. Specific conditions for convergence
of this scheme as well as various applications to the solutions of network oligopolies can
be found in Nagurney and Zhang (1996), Nagurney, Dupuis, and Zhang (1994), Nagurney
(2010), and Masoumi, Yu, and Nagurney (2012).

4.1 Explicit Formulae for the Euler Method Applied to the Uncapacitated Model

The elegance of this procedure for the computation of solutions to the uncapacitated fresh
produce farmers’ market supply chain network problem in Section 3.1 can be seen in the
following explicit formulae. In particular, we have the following closed form expressions for
the fresh produce path flows, for each path p ∈ Pji , ∀i, j:

M
X ∂ ρ̂il (xτ , q) X ∂ Ĉp (xτ )
xτp +1 = max{0, xτp τ
+ aτ (ρ̂ij (x , q) + xτr − )},
∂xp ∂xp
l=1 r∈Pli

∀p ∈ Pji ; i = 1, . . . , I; j = 1, . . . , M. (37)

Next, the closed form expression to compute the solution of the capacitated problem in
Section 3.2 is given.

20
4.2 Explicit Formulae for the Euler Method Applied to the Capacitated Model

The explicit formulae is shown to compute the solutions to the capacitated fresh produce
farmers’ market supply chain problem in Section 3.2. The closed form expressions for the
fresh produce path flows at iteration τ + 1 are as follows. For each path p ∈ Pji , ∀i, j,
compute:
M
X ∂ ρ̂il (xτ , q) X ∂ Ĉp (xτ ) X τ
xτp +1 = max{0, xτp τ
+ aτ (ρ̂ij (x , q) + xτr − − λa δap )},
∂xp ∂xp
l=1 r∈Pli a∈L

∀p ∈ Pji ; i = 1, . . . , I; j = 1, . . . , M. (38)
The Lagrange multipliers for each link a ∈ Li ; i = 1, . . . , I, can be computed as:
X
λτa+1 = max{0, λτa + aτ ( xτp δap − ua )}, ∀a ∈ L. (39)
p∈P

The number of strategic variables xp , as well as the number of the paths, in the supply chain
network, for both of the uncapacitated and capacitated supply chain networks, grow linearly
in terms of the number of nodes in the supply chain network, be it a farm, or a farmers’
market, etc. Therefore, even a fresh produce supply chain network with hundreds of demand
markets is still tractable within our proposed modeling and computational framework.

5. Case Study

We focus on apples for our case study. The United States holds second place in the world’s
apple production with 4 million metric tons of apples produced, with the leader being China,
which produced 33 million metric tons of apples in 2010 (USDA (2012)). Specifically, in our
case study we consider Golden Delicious apples. This type of apple is ranked in third place
among the other apple varieties in the United States with 25,000 metric tons of produc-
tion value. Our case study is based on farmers’ markets in western Massachusetts in the
United States. According to the USDA (2016) and the website of Community Involved in
Sustaining Agriculture (CISA), there are 8,558 farmers markets in the United States, 312
in Massachusetts, and 40 in western Massachusetts (CISA (2016)). Also, in Massachusetts,
369 in-state apple orchards produced 34 million pounds of apples in 2010. There are ap-
proximately 27 apple orchards in western Massachusetts. For this case study, we selected
three of the apple orchards located in western Massachusetts: Apex Orchards, the Park
Hill Orchard, and Sentinel Farm. The locations of these orchards/farms in Massachusetts
are, respectively, Shelburne, Easthampton, and Belchertown. These orchards/farms have
the possibility of selling their apples at the Amherst, Northampton, South Hadley, and/or
Belchertown farmers’ markets, which are open on different days in a given week during the

21
season, as shown in Figure 2. The Northampton Farmers’ Market is open on Tuesdays,
the South Hadley Farmers’ Market operates on Thursdays. On Saturdays the Amherst
Farmers’ Market is open, and the Belchertown Farmers’ Market is open on Sundays. Each
orchard/farm has its own harvesting, processing, storage, and transportation units. Notice
that the demand points in Figure 2 are sequenced depending on the day of the week. For
example, since the Northampton Farmers’ Market is open on Tuesdays, which is the first
farmers’ market of the week in this case study, it is numbered as the first demand point.
The South Hadley Farmers’ Market corresponds to demand point 2, and so on. Additionally,
apples are picked on Tuesdays at all the orchards/farms, so that there is no storage for the
Golden Delicious apples sold at the Northampton Farmers’ Market. However, each farm
should store its Golden Delicious apples for 2, 4, and 5 days for the South Hadley, Amherst,
and the Belchertown Farmers’ Markets, respectively.

Figure 2: The Supply Chain Network Topology for the Apple Farmers’ Markets Case Study
– Scenario 1

The Euler method for the uncapacitated (cf. (37)) and the capacitated problems (cf.
(38) and (39)) is implemented in MATLAB and the sequence aτ = {1, 12 , 21 , 31 , . . .}, with the
convergence tolerance being 10−6 , that is, the Euler method is deemed to have converged
if the absolute value of the difference of each successive variable iterate differs by no more
than this value. The code in Matlab is executed on a Macbook Pro laptop with a 2.8 GHz
Intel Core i5 processor and 8GB 1600 MHz DDR3 memory.

Scenario 1. In Scenario 1, it is assumed that the weather conditions in Massachusetts

are steady, which means that the temperature in the growing season of Golden Delicious

22
apples is at its seasonal normal and the harvesting season is for 2015. For this scenario,
we assume that orchard/farm i; i = 1, 2, 3, in the supply chain network has initial quality,
respectively, of: q01 = 1, q02 = 0.8, and q03 = 0.7. Apex Orchards are well-known for the
quality of their apples in western Massachusetts and, hence, the value for their apple initial
quality is q01 = 1. In addition, we assume that there is no capacity limit on the links in the
supply chain network. According to Table 1, we know that, for Golden Delicious apples, the
quality attribute, that of the texture softening, follows first order quality decay. The link
quality decay βa for every link a in the supply chain network and the parameters used in
the calculations are provided in Table 2. Also, we include the quality decay at the farmers’
markets by considering the selling/purchasing point as being the middle of the operation
period, since quality decay is a continuous process.

The orchards/farms in this case study are of different sizes and are located at different
altitudes. According to CISA, Apex, the Park Hill Orchards, and the Sentinel Farm have 170,
127, and 8 acres of land, respectively. Also, Apex Orchards are located at a higher altitude
than Park Hill and Sentinel Farm, but the altitude of Park Hill and the Sentinel are similar.
The harvesting season of apples is between September and October when temperatures are
between 19-22 ◦ C. In this case study, the harvesting operation is assumed to be realized at
the average temperature of the season depending on the location of the orchards. Notice that
Apex Orchards have the lowest harvesting temperature, since they are located at a higher
altitude where the temperature is lower. Furthermore, the duration of the link operations
are constructed according to the land size and the number of employees. For example, Apex
has a larger land size; therefore, the harvesting operation at this orchard takes a longer time.
Moreover, we assume that the larger the size of the orchard, the more employees it has. As a
result, the duration of harvesting and processing operations are the longest for Sentinel Farm
due to it having the lowest number of employees. The duration of transportation is calculated
as the summation of the actual transportation time between the orchard and the farmers’
market and half of the farmers’ market’s hours of operation, which the transportation link
is connected to in the supply chain network. The hours of operation of the Northampton,
South Hadley, Amherst, and Belchertown Farmers’ Markets are, respectively: 5, 6, 5.5, and
4 hours. The hours of storage are calculated according to the days between the harvesting
and when the farmers’ markets open in the week. For instance, Apple Orchards need to
store their apples for 2 days to sell at the South Hadley Farmers’ Market, which is open
on Thursdays. It is assumed that Apex Orchards have the controlled atmospheric storage
system to keep the temperature at 0◦ C which is the optimal storage temperature for apples
(Iowa State University Extension (2008)). However, it is assumed that Park Hill and Sentinel
Farm have regular storage which can lower the temperatures to 9◦ C and 12◦ C, respectively.

23
 Operations Link a Hours Temp (◦ C) βa
harvesting 1 4.00 19 0.992
processing 2 3.00 19 0.994
transportation 3 0.50 19 0.999
storage (2 days) 4 48.00 0 0.994
storage (4 days) 5 96.00 0 0.988
storage (5 days) 6 120.00 0 0.985
transportation 7 4.00 19 0.993
transportation 8 3.25 19 0.994
transportation 9 4.00 19 0.993
harvesting 10 3.00 22 0.992
processing 11 3.00 22 0.992
transportation 12 0.5 19 0.999
storage (2 days) 13 48.00 9 0.978
storage (4 days ) 14 96.00 9 0.957
storage (5 days) 15 120.00 9 0.947
transportation 16 3.75 19 0.993
transportation 17 5.16 19 0.990
transportation 18 3.00 19 0.992
harvesting 19 5.00 22 0.986
processing 20 5.00 22 0.986
transportation 21 0.50 22 0.998
storage (2 days) 22 48.00 12 0.967
storage (4 days ) 23 96.00 12 0.936
storage (5 days) 24 120.00 12 0.921
transportation 25 3.75 22 0.990
transportation 26 5.16 22 0.986
transportation 27 3.00 22 0.992

Table 2: Parameters for the Calculation of Quality Decay for the Case Study Scenario 1

The link quality decay βa is calculated by taking the universal gas constant and the activation
energy as 8.314 Jmol−1 K −1 and 88 kJmol−1 . The temperature and the time are converted,
respectively, to Kelvin and seconds for the quality decay calculations.

The total number of paths in the supply chain network is twelve and they are as follows:
path p1 = (1, 2, 3), p2 = (1, 2, 4, 7), p3 = (1, 2, 5, 8), p4 = (1, 2, 6, 9), p5 = (10, 11, 12),
p6 = (10, 11, 13, 16), p7 = (10, 11, 14, 17), p8 = (10, 11, 15, 18), p9 = (19, 20, 21), p10 =
(19, 20, 22, 25), p11 = (19, 20, 23, 26), and p11 = (19, 20, 24, 27). Also each path has its own
quality decay rate, which is calculated according to (11). Furthermore, the demand price
functions of the orchard/farms at the farmers’ markets are:

24
 Apex Orchards:

ρ11 (d, q) = −0.04d11 − 0.01d21 − 0.01d31 + 8qp1 − 4qp5 − 3qp9 + 30,

ρ12 (d, q) = −0.02d12 − 0.01d22 − 0.01d32 + 3qp2 − 2qp6 − 2qp10 + 25,

ρ13 (d, q) = −0.04d13 − 0.02d23 − 0.01d33 + 8qp3 − 4qp7 − 3qp11 + 30,

ρ14 (d, q) = −0.04d14 − 0.02d24 − 0.02d34 + 3qp4 − qp8 − 2qp12 + 25,

Park Hill Orchard:

ρ21 (d, q) = −0.04d21 − 0.02d11 − 0.02d31 + 3qp5 − 2qp1 − qp9 + 27,

ρ22 (d, q) = −0.04d22 − 0.01d12 − 0.02d32 + 3qp6 − 2qp2 − qp10 + 28,

ρ23 (d, q) = −0.04d23 − 0.02d13 − 0.02d33 + 4qp7 − 2qp3 − qp11 + 27,

ρ24 (d, q) = −0.02d24 − 0.01d14 − 0.01d34 + 2qp8 − qp4 − qp12 + 28,

Sentinel Farm:

ρ31 (d, q) = −0.04d31 − 0.02d11 − 0.02d21 + 4qp9 − qp1 − 2qp5 + 25,

ρ32 (d, q) = −0.04d32 − 0.01d12 − 0.02d22 + 4qp10 − 3qp2 − qp6 + 28,

ρ33 (d, q) = −0.02d23 − 0.01d13 − 0.01d33 + 4qp11 − 2qp3 − qp7 + 25,

ρ34 (d, q) = −0.04d34 − 0.02d14 − 0.02d24 + 3qp12 − 2qp4 − 2qp8 + 28.

The demand price functions are constructed according to the customer and orchard char-
acteristics. According to a former orchard owner, Colnes (2016), customers going to the
Amherst Farmers’ Market are more affluent and give importance to quality of the apples.
The Northampton Farmers’ Market is also similar to the Amherst Farmers’ Market in terms
of the consumers’ willingness to pay for higher quality. The demand price functions at the
South Hadley and Belchertown Farmers’ Markets are similar to one another, with the lowest
consumer willingness to pay in the supply chain network. Furthermore, some orchards are
assumed to have a more positive reputation at some farmers’ markets, which means that the
price of the Golden Delicious apples from an orchard may be less affected by the demand
for the apples of the other orchards at the same farmers’ market. For example, Apex is
assumed to have loyal customers at the Amherst Farmers’ Market whereas the Sentinel does
not; therefore, the price of Apex’s Golden Delicious apples is less affected by the demand for
the Sentinel’s apples.

25
 The total link cost functions and the computed equilibrium link flows are shown in Table
3. The cost functions are constructed based on the price data in Berkett (1994), King and
Gomez (2015), and the U.S. Energy Information Administration (2016) for the fuel price.
The flow unit is pecks and the total cost functions are constructed based on the dollar price
per peck.

Operations Link a ĉa (f ) fa∗

harvesting 1 0.02f12 + 3f1 165.8395
processing 2 0.015f22 + 3f2 165.8395
transportation 3 0.01f32 + 3f3 111.9827
storage (2 days) 4 0.01f42 + 3f4 0.0000
storage (4 days ) 5 0.015f52 + 4f5 53.8568
storage (5 days) 6 0.03f62 + 5f6 0.0000
transportation 7 0.02f72 + 6f7 0.0000
transportation 8 0.0125f82 + 4f8 53.8568
transportation 9 0.02f92 + 6.6f9 0.0000
2
harvesting 10 0.0125f10 + 6f10 94.7414
2
processing 11 0.0125f11 + 6f11 94.7414
2
transportation 12 0.0045f12 + f12 71.7812
2
storage (2 days) 13 0.01f13 + 1.67f13 22.9601
2
storage (4 days ) 14 0.015f14 + 6f14 0.0000
2
storage (5 days) 15 0.015f15 + 6.6f15 0.0000
2
transportation 16 0.0075f16 + 6f16 22.9601
2
transportation 17 0.01f17 + 6f17 0.0000
2
transportation 18 0.02f18 + 4f18 0.0000
2
harvesting 19 0.0125f19 + 6f19 98.5294
2
processing 20 0.015f20 + 4f20 98.5294
2
transportation 21 0.02f21 + 4f21 17.2084
2
storage (2 days) 22 0.007f22 + 1.67f22 32.4314
2
storage (4 days ) 23 0.009f23 + 6f23 0.0000
2
storage (5 days) 24 0.01f24 + 6f24 48.8896
2
transportation 25 0.005f25 + 6f25 32.4314
2
transportation 26 0.005f26 + 6f26 0.0000
2
transportation 27 0.0005f27 + 0.1f27 48.8896

Table 3: The Total Link Cost Functions and the Computed Equilibrium Link Flows for the
Case Study Scenario 1

26
 The computed equilibrium path flows and the quality decay of the paths are given in
Table 4. As mentioned earlier, the unit of the flows is pecks of apples. Notice that, in
equilibrium, there are paths and links with zero flows, which indicates the nonprofitable
farmers’ markets for specific orchards. For example, Apex serves the Northampton and
Amherst Farmers’ Markets; Park Hill serves the Northampton and South Hadley Farmers’
Markets, and Sentinel serves the Northampton, South Hadley, and Belchertown Farmers’
Markets. The game theory aspect of our model reveals the profitable markets and the
variational inequality solution returns positive path flows for these markets and zero for the
nonprofitable ones.

Path p qp x∗p
p1 0.9851 111.9827
p2 0.9733 0.0000
p3 0.9684 53.8568
p4 0.9645 0.0000
p5 0.7864 71.7812
p6 0.7645 22.9602
p7 0.7458 0.0000
p8 0.7395 0.0000
p9 0.6791 17.2084
p10 0.6514 32.4314
p11 0.6280 0.0000
p12 0.6217 48.8896

Table 4: The Path Quality Decay Rates and the Computed Equilibrium Path Flows for the
Case Study Scenario 1

By substituting the equilibrium path flow and quality decay values, the demand prices of
the orchard/farms at the profitable demand markets are found, in terms of dollars per peck,
as:

Apex Orchards:

ρ11 = 27.33, ρ12 = 24.53, ρ13 = 30.72, ρ14 = 25.42,

Park Hill Orchard:

ρ21 = 21.25, ρ22 = 26.13, ρ23 = 26.34, ρ24 = 27.40,

Sentinel Farm:

ρ31 = 20.79, ρ32 = 25.16, ρ33 = 24.29, ρ34 = 24.50.

27
 According to Clements (2016), who is an educator at the University of Massachusetts
Amherst Extension Fruit Program, the retail price of Golden Delicious apples is usually $2
per pound. A peck of apples is equal to 10-12 pounds which means the price of a peck can
be between $20-$24. Our results for the demand prices are close to this range and, hence,
are consistent with reality. Furthermore, the profits of the orchard/farms, in dollars, at the
equilibrium solution, are:

U1 (X ∗ ) = 1785.40, U2 (X ∗ ) = 484.03, U3 (X ∗ ) = 460.15.

Apex Orchards have the largest profit in Scenario 1, followed by Park Hill Orchard and
then the Sentinel Farm.

Scenario 2. In this scenario, it is assumed that a new orchard, which was solely selling to
retailers and wholesalers previously, is attracted by the demand for apples at the farmers’
markets. This hypothetical new orchard is called New Orchard, and is located in western
Massachusetts and enters the local food supply chain as depicted in Figure 3. As in Scenario
1, the uncapacitated variational inequality problem (22) is solved with the Euler method,
using the explicit formulae shown in (37).

Figure 3: The Supply Chain Network Topology for the Apple Farmers’ Markets Case Study
- Scenario 2

The quality decay parameters and the link quality decay rates of New Orchard are shown
in Table 5 and the total link cost functions are depicted in Table 6. The initial quality for
this orchard is, q04 = 1, and the other orchard/farms have the same initial quality values as

28
in Scenario 1. New Orchard is very similar to Apex Orchards in terms of its land size, the
number of employees, and the storage technology. Therefore, the New Orchards’ total link
cost functions for harvesting, processing, and storage are very similar to those of Apex. We
assume that New Orchard is located near Sentinel Farm in Belchertown at a lower altitude
than the other orchards/farms where the harvesting, processing, and transportation take
place at higher temperatures. The transportation duration from the New Orchard to the
farmers’ markets is similar to that of the Sentinel Farm.
Operations Link a Hours Temp (◦ C) βa
harvesting 28 4.00 19 0.988
processing 29 4.00 19 0.988
transportation 30 0.50 19 0.998
storage (2 days) 31 48.00 0 0.968
storage (4 days) 32 96.00 0 0.989
storage (5 days) 33 120.00 0 0.986
transportation 34 3.50 19 0.989
transportation 35 3.00 19 0.991
transportation 36 3.00 19 0.991

Table 5: New Orchard Parameters for the Calculation of Quality Decay for Scenario 2

Since New Orchard has the same initial quality of its Golden Delicious apples as that of
the Apex Orchards, it is expected that it may lose some of its loyal customers at the farmers’
markets which New Orchard is able to enter.
Operations Link a ĉa (f )
2
harvesting 28 0.02f28 + 3f28
2
processing 29 0.015f29 + 3f29
2
transportation 30 0.01f30 + 3f30
2
storage (2 days) 31 0.01f31 + 6f31
2
storage (4 days) 32 0.015f32 + 4f32
2
storage (5 days) 33 0.035f33 + 5f33
2
transportation 34 0.025f34 + 8f34
2
transportation 35 0.015f35 + 4f35
2
transportation 36 0.025f36 + 8f36

Table 6: The Total Link Cost Functions for New Orchard Under Scenario 2

With the entrance of New Orchard into the supply chain, the total number of paths in
the supply chain network for Scenario 2 is increased to sixteen with twelve of them being
the same as in Scenario 1. The additional paths and links are: path p13 = (28, 29, 30),
p14 = (28, 29, 31, 34), p15 = (28, 29, 32, 35), and p16 = (28, 29, 32, 36).

29
 Customers at the farmers’ markets have enough information about New Orchard’s Golden
Delicious apples. This means that the demand price functions of the other orchard/farms are
affected by the demand for the Golden Delicious apples of New Orchard. The new demand
price functions are:

Apex Orchards:

ρ11 (d, q) = −0.053d11 − 0.01d21 − 0.01d31 − 0.03d41 + 8qp1 − 2qp5 − 2qp9 − 4qp13 + 30,

ρ12 (d, q) = −0.03d12 − 0.01d22 − 0.01d32 − 0.004d42 + 3qp2 − 2qp6 − 2qp10 − qp14 + 25,

ρ13 (d, q) = −0.053d13 − 0.01d23 − 0.01d33 − 0.03d43 + 8qp3 − 2qp7 − 2qp11 − 4qp15 + 30,

ρ14 (d, q) = −0.03d14 − 0.01d24 − 0.014d34 − 0.004d44 + 3qp4 − qp8 − 2qp12 − qp15 + 25,

Park Hill Orchard:

ρ21 (d, q) = −0.05d21 − 0.01d11 − 0.01d31 − 0.01d41 + 3qp5 − qp1 − qp9 − qp13 + 27,

ρ22 (d, q) = −0.04d22 − 0.01d12 − 0.02d32 − 0.004d42 + 3qp6 − 2qp2 − qp10 − qp14 + 28,

ρ23 (d, q) = −0.05d23 − 0.02d13 − 0.01d33 − 0.02d43 + 4qp7 − 2qp3 − qp11 − 2qp15 + 27,

ρ24 (d, q) = −0.04d24 − 0.01d14 − 0.02d34 − 0.004d44 + 2qp8 − qp4 − qp12 − qp16 + 28,

Sentinel Farm:

ρ31 (d, q) = −0.05d31 − 0.01d11 − 0.01d21 − 0.01d41 + 2qp9 − qp1 − qp5 − qp13 + 25,

ρ32 (d, q) = −0.04d32 − 0.01d12 − 0.02d22 − 0.004d42 + 4qp10 − 3qp2 − qp6 − qp14 + 28,

ρ33 (d, q) = −0.05d33 − 0.02d13 − 0.01d23 − 0.02d43 + 4qp11 − 2qp3 − qp7 − 2qp15 + 25,

ρ34 (d, q) = −0.04d34 − 0.01d14 − 0.02d24 − 0.004d44 + 3qp12 − 2qp4 − 2qp8 − 2qp16 + 28.

According to the demand price functions, New Orchard is not very strong in the South
Hadley Farmers’ Market. However, it is very effective in the Northampton, Amherst, and
Belchertown Farmers’ Markets where Apex and Sentinel have been market leaders. In par-
ticular, New Orchard becomes a crucial competitor for Apex Orchards. Its demand price
functions are given below.

New Orchard:

ρ41 (d, q) = −0.053d41 − 0.03d11 − 0.01d21 − 0.01d31 + 5qp13 − 2qp1 − qp5 − qp9 + 30,

30
 ρ42 (d, q) = −0.0342 − 0.006d12 − 0.01d22 − 0.01d32 + 2qp14 − qp2 − qp6 − qp10 + 25,
ρ43 (d, q) = −0.053d43 − 0.03d13 − 0.01d23 − 0.01d33 + 5qp15 − 2qp3 − qp7 − qp11 + 30,
ρ44 (d, q) = −0.03d44 − 0.006d14 − 0.01d24 − 0.01d34 + 2qp16 − qp4 − qp8 − qp12 + 25.

For New Orchard, there is more competition in the Amherst and the Northampton Farm-
ers’ Markets where the customers are more health conscious and pay attention to the new-
comer. New Orchard’s demand price is mostly affected by Apex, which is a very important
competitor for New Orchard. For example, the demand price of New Orchard’s Golden
Delicious apples at the Northampton Farmers’ Market is affected mostly by the demand for
the Golden Delicious apples from Apex.

The computed equilibrium link flows, path flows, and the quality decay rates of the paths
are given in Table 7 and Table 8. New Orchard enters the Northampton and Amherst
Farmers’ Markets, since they are the only profitable farmers’ markets in the supply chain
network for it. Notice that some of the equilibrium path flows are lower than in Scenario 1.
For instance, the flow on the paths p1 and p3 are lower in Table 9 than they are in Table 4.
One explanation for this result is that New Orchard is a new market player whose quality of
Golden Delicious apples is very similar to that of Apex’s; therefore, some of Apex’s previous
customers now choose to buy apples from New Orchard and cause a drop in Golden Delicious
apple sales for Apex.

The demand prices of the orchard/farms, in dollars per peck, for the profitable demand
markets are:

Apex Orchards:

ρ11 = 23.49, ρ12 = 23.66, ρ13 = 27.49, ρ14 = 24.44,

Park Hill Orchard:

ρ21 = 21.46, ρ22 = 25.41, ρ23 = 25.49, ρ24 = 26.20,

Sentinel Farm:

ρ31 = 20.38, ρ32 = 24.38, ρ33 = 22.91, ρ34 = 23.08,

New Orchard:

ρ41 = 23.82, ρ42 = 23.99, ρ43 = 27.80, ρ44 = 24.21.

31
 Operations Link a fa∗
harvesting 1 124.0885
processing 2 124.0885
transportation 3 79.5849
storage (2 days) 4 0.0000
storage (4 days ) 5 44.5036
storage (5 days) 6 0.0000
transportation 7 0.0000
transportation 8 44.5036
transportation 9 0.0000
harvesting 10 87.4808
processing 11 87.4808
transportation 12 69.2348
storage (2 days) 13 18.2459
storage (4 days ) 14 0.0000
storage (5 days) 15 0.0000
transportation 16 18.2459
transportation 17 0.0000
transportation 18 0.0000
harvesting 19 86.0782
processing 20 86.0782
transportation 21 18.3520
storage (2 days) 22 30.9408
storage (4 days ) 23 0.0000
storage (5 days) 24 36.7854
transportation 25 30.9408
transportation 26 0.0000
transportation 27 36.7854
harvesting 28 126.1215
processing 29 126.1215
transportation 30 82.0895
storage (2 days) 31 0.0000
storage (4 days) 32 44.0319
storage (5 days) 33 0.0000
transportation 34 0.0000
transportation 35 44.0319
transportation 36 0.0000

Table 7: The Computed Equilibrium Link Flows for the Case Study Scenario 2

New Orchards’ entrance causes a price decrease for the other orchard/farms at the farmers’
markets due to competition in all prices except for ρ21 , which is almost the same as in scenario
1. For example, Apex sells its apples at $27.33 per peck at the Northampton Farmers’ Market
in Scenario 1, which decreases to $23.49 per peck.

32
 Path p qp x∗p
p1 0.9851 79.5849
p2 0.9733 0.0000
p3 0.9684 44.5036
p4 0.9645 0.0000
p5 0.7864 69.2348
p6 0.7645 18.2460
p7 0.7458 0.0000
p8 0.7395 0.0000
p9 0.6791 18.3520
p10 0.6514 30.9408
p11 0.6280 0.0000
p12 0.6217 36.7854
p13 0.9742 82.0895
p14 0.9345 0.0000
p15 0.9567 44.0319
p16 0.9538 0.0000

Table 8: The Path Quality Decay Rates and the Computed Equilibrium Path Flows for the
Case Study Scenario 2

The profits of the orchard/farms are:

U1 (X ∗ ) = 1097.39, U2 (X ∗ ) = 471.71, U3 (X ∗ ) = 345.45, U4 (X ∗ ) = 1142.19.

The profits of Apex, Park Hill Orchards, and the Sentinel Farm decrease from their values
in Scenario 1, as a result of the the entrance of New Orchard, which results in increased
competition. The largest profit is gained by New Orchard, followed by Apex, Park Hill, and
then Sentinel.

Scenario 3. This scenario is constructed to illustrate the apple shortage experienced in

western Massachusetts in 2016. We return to the original farmers as in Scenario 1. The
volatile weather conditions of an unexpected cold snap occurred in May. According to
various news articles, the cold snap damaged the green apple buds and an apple shortage
at the local markets, which includes the farmers’ markets, was expected. In this scenario,
we capture this apple shortage by imposing link capacities on the harvesting links of the
orchard/farms. The Euler method, with the explicit formulae given in (38) and (39) for the
capacitated problem, is used to compute the solution of the variational inequality formulation
(33). The supply chain network structure for this scenario is the same as that in Figure 2.
The total link costs functions, demand price functions of the orchard/farms, and the link
quality decay rates are also the same as the corresponding ones in Scenario 1.

33
 The link capacities (in pecks), the computed equilibrium link flows, and the computed
equilibrium Lagrange multipliers for this scenario are reported in Table 9.

Operations Link a Capacity fa∗ λ∗a

harvesting 1 20 20.0000 16.4077
processing 2 15000 20.0000 0.0000
transportation 3 15000 20.0000 0.0000
storage (2 days) 4 15000 0.0000 0.0000
storage (3 days ) 5 15000 0.0000 0.0000
storage (4 days) 6 15000 0.0000 0.0000
transportation 7 15000 0.0000 0.0000
transportation 8 15000 0.0000 0.0000
transportation 9 15000 0.0000 0.0000
harvesting 10 50 50.0000 6.4906
processing 11 15000 50.0000 0.0000
transportation 12 15000 50.0000 0.0000
storage (2 days) 13 15000 0.0000 0.0000
storage (3 days ) 14 15000 0.0000 0.0000
storage (4 days) 15 15000 0.0000 0.0000
transportation 16 15000 0.0000 0.0000
transportation 17 15000 0.0000 0.0000
transportation 18 15000 0.0000 0.0000
harvesting 19 60 60.0000 5.6685
processing 20 15000 60.0000 0.0000
transportation 21 15000 13.1918 0.0000
storage (2 days) 22 15000 18.7448 0.0000
storage (3 days ) 23 15000 0.0000 0.0000
storage (4 days) 24 15000 28.0624 0.0000
transportation 25 15000 18.7448 0.0000
transportation 26 15000 0.0000 0.0000
transportation 27 15000 28.0624 0.0000

Table 9: Link Capacities and Computed Equilibrium Link Flows for the Case Study in
Scenario 3

The capacities in Table 9 reflect the expected level of harvest damage at the orchard/farms.
For example, Apex is assumed to experience a larger damage of its Golden Delicious apples,
since it is located at a higher altitude, which causes the temperatures to drop lower than the
temperatures at the other orchard/farms and damages the green buds more. Therefore, the
capacity of Apex’s harvesting link is assumed to be the lowest. Park Hill Orchard and Sen-
tinel Farm are located at similar altitudes in western Massachusetts. We assume that they
experience similar temperatures which result in similar capacities imposed on the harvesting

34
links of these orchard/farms. The capacities on the other operational links remain relatively
large. Furthermore, the initial quality of Apex (q01 ), Park Hill Orchard (q02 ), and Sentinel
Farm (q03 ) apples are assumed to be 0.4, 0.5, and 0.6, respectively. Since the most damage
on the apple buds is assumed to happen at the Apex Orchards, the lowest initial quality is
assigned to this orchard.

From Table 9 and Table 10 one can see that Apex and Park Hill are now only at the
Northampton Farmers’ Market, whereas Sentinel serves the Northampton, South Hadley,
and Belchertown Farmers’ Markets. Moreover, links 1, 10, and 19 are at their capacities
and, hence, the associated Lagrange multipliers are positive.

The equilibrium path flows and the path quality decay values are reported in Table
10. Observe that the path flows of all the orchard/farms have decreased substantially with
respect to the path flows in Scenario 1. The experienced shortage is especially marked for
Apex since the path flows p1 , p2 , p3 , and p4 have decreased substantially from the results
reported in Scenario 1 in Table 4.

Path p qp x∗p
p1 0.3940 20.0000
p2 0.3893 0.0000
p3 0.3873 0.0000
p4 0.3858 0.0000
p5 0.4915 50.0000
p6 0.4778 0.0000
p7 0.4662 0.0000
p8 0.4622 0.0000
p9 0.5821 13.1918
p10 0.5584 18.7448
p11 0.5383 0.0000
p12 0.5329 28.0624

Table 10: Computed Equilibrium Path Flows for the Case Study Scenario 3

The demand prices of the orchard/farms, in dollars, are:

Apex Orchards:

ρ11 = 28.01, ρ12 = 23.91, ρ13 = 29.62, ρ14 = 24.35.

Park Hill Orchard:

ρ21 = 24.44, ρ22 = 27.72, ρ23 = 27.55, ρ24 = 27.72,

35
 Sentinel Farm:

ρ31 = 24.02, ρ32 = 27.84, ρ33 = 25.91, ρ34 = 26.78.

Since the supply is decreased, the prices of Golden Delicious apples increase in most of
the farmers’ markets. For example, Apex’s Golden Delicious apples are now $28.01 at the
Northampton Farmers’ Market in this scenario whereas the price was $27.33 in Scenario 1.
Additionally, since the quality of Apex’s apples is worse than in Scenario 1, this causes the
demand price to decrease at the Amherst Farmers’ Market. In Scenario 1, the price of its
apples at the Amherst Farmers’ Market was $30.72; however, now it is $29.62.

The profits of the orchard/farms in this scenario, in dollars, are:

U1 (X ∗ ) = 362.15, U2 (X ∗ ) = 498.28, U3 (X ∗ ) = 507.58.

The largest profit is achieved by Sentinel, followed by Park Hill, and Apex. In Scenario
1, Apex has the largest profit, which decreases substantially in this scenario due to it having
the lowest harvesting capacity and quality.

6. Summary and Conclusions

Fresh produce consists of both fruits and vegetables and such supply chains are especially
challenging since the quality of the product deteriorates continuously upon harvesting. At the
same time, consumers are demanding fresh products and are increasingly health conscious.
Farmers’ markets have increased in popularity internationally and, yet, the mathematical
modeling associated with such supply chains has been limited.

In this paper, we provide explicit formulae for a variety of fresh produce to capture
quality deterioration. We then identify the quality associated with different pathways in
supply chain networks. Subsequently, we focus on farmers’ markets, which are examples
of direct to consumer supply chains and are shorter supply chains since farmers bring the
product to markets at which consumers select their purchases. Specifically, we introduce a
game theory model for supply chain competition in a network framework for farmers’ markets
occurring within a period of time, such as a week. The farms are interested in maximizing
their profits and the consumers respond to the quality of the product. We provide both
qualitative properties of the equilibrium link flow pattern, propose a computational scheme,
and also illustrate our framework through numerical examples focused on peaches and then
in a case study for Golden Delicious apples and farmers’ markets in western Massachusetts.
Our framework considers both uncapacitated links in the supply chain network as well as

36
capacitated ones, which may occur due to crop failures, harvesting problems, labor shortages,
etc. This is the first game theory model for farmers’ markets and also the first competitive
fresh produce supply chain network model in which quality deterioration of fresh produce is
explicitly captured.

Acknowledgments This paper is dedicated to the memory of Robert Colnes, who passed
away at age 96 on May 30, 2016. He had been an apple farmer for decades and helpful
conversations with him inspired this research. The authors are grateful to the two anonymous
reviewers and to the Editor for constructive comments and suggestions on two earlier versions
of this paper.

References

Aamir M, Ovissipour M, Sablani SS, Rasco B. Predicting the quality of pasteurized vegeta-
bles using kinetic models: a review. International Journal of Food Science 2013; Article ID
271271.

Abbott JA. Quality measurement of fruits and vegetables. Postharvest Biology and Tech-
nology 1999; 15(3): 207-225.

Ahumada O, Villalobos JR. Application of planning models in the agri-food supply chain:
A review. European Journal of Operational Research 2007; 196(1): 1-20.

Ahumada O, Villalobos JR. A tactical model for planning the production and distribution
of fresh produce. Annals of Operations Research 2011; 190(1): 339-358.

Aiello G, La Scalia G, Micale R. Simulation analysis of cold chain performance based on

time-temperature data. Production Planning & Control 2012; 23(6): 468-476.

Akkerman R, Farahani P, Grunow M. Quality, safety and sustainability in food distribution:

A review of quantitative operations management approaches and challenges. OR Spectrum
2010; 32(4): 863-904.

Amorim P, Costa AM, Almada-Lobo B. Influence of consumer purchasing behaviour on the

production planning of perishable food. OR Spectrum 2014; 36(3): 669692.

Aregu R. Market and Price Decision Enhancement Services for Farmers in Uganda. Doctoral
Dissertation, University of Groningen, Groningen, The Netherlands; 2014.

Arrhenius S. Über die reaktionsgeschwindigkeit bei der inversion von rohrzucker durch suren.
Zeitschrift Für Physikalische Chemie 1889; 4: 226-248.

37
Asebo K, Moxnes Jervel A, Lieblein G, Svennerud M, Francis C. Farmer and consumer
attitudes at farmers markets in Norway. Journal of Sustainable Agriculture 2007; 30(4):
67-93.

Babiak L. Exploring Local Food System Practices and Perceptions: Insights from Florida’s
SNAP Authorized Farmers’ Markets. Master of Science Thesis. Department of Geogra-
phy, Environment, and Planning, College of Arts and Sciences, University of South Florida,
Tampa, Florida; 2013.

Barrett DM, Beaulieu JC, Shewfelt R. Color, flavor, texture, and nutritional quality of fresh-
cut fruits and vegetables: desirable levels, instrumental and sensory measurement, and the
effects of processing. Critical Reviews in Food Science and Nutrition 2010; 50(5): 369-389.

Berkett LP. Management Guide for Low-Input Sustainable Apple Production. DIANE Pub-
lishing; 1994 May 1.

Bertsekas DP, Tsitsiklis JN. Parallel and Distributed Computation. Prentice-Hall, Engle-
wood Cliffs, New Jersey; 1989.

Blackburn J, Scudder G. Supply chain strategies for perishable products: The case of fresh
produce. Production and Operations Management 2009; 18(2): 129-137.

Bond JK, Thilmany D, Bond C. What influences consumer choice of fresh produce purchase
location? Journal of Agricultural and Applied Economics 2009; 41(1): 61-74.

Bosona T, Gebresenbet G, Nordmark I, Ljungberg D. Integrated logistics network for the

supply chain of locally produced food, Part I: Location and route optimization analyses.
Journal of Service Science and Management 2011; 4(2): 174-183.

Castro I, Teixeira JA, Salengke S, Sastry SK, Vicente AA. Ohmic heating of strawberry
products: electrical conductivity measurements and ascorbic acid degradation kinetics. In-
novative Food Science & Emerging Technologies 2004; 5(1): 27-36.

Clements J. Private communication, University of Massachusetts Amherst. December 27,

2016a.

Clements J. Private communication, University of Massachusetts Amherst. June 25, 2016b.

Colnes R. Private communication, Salem, Massachusetts. February 6, 2016.

Community Involved in Sustaining Agriculture (CISA). Farmers markets in western Mas-

38
sachusetts; 2016.
http://www.farmfresh.org/food/farmersmarkets.php?zip=01002

Cournot AA. Researches into the Mathematical Principles of the Theory of Wealth 1838;
English translation: MacMillan, London, England 1897.

Dermesonlouoglou E, Giannakourou M, Taoukis, P. Kinetic modelling of the quality degra-

dation of frozen watermelon tissue: effect of the osmotic dehydration as a pre-treatment.
International Journal of Food Science & Technology 2007; 42(7): 790-798.

DeWeerdt S. Is local food better? World Watch Magazine 2016; 22(3). May/June.

Drew B. Private communication, Apex Orchards, Shelburne, Massachusetts. June 25, 2016.

Dupuis P, Nagurney A. Dynamical systems and variational inequalities. Annals of Opera-

tions Research 1993; 44(1): 9-42.

Fabbrizzi S, Menghini S, Marinelli N. The short food supply chain: A concrete example of
sustainability. A literature review. Rivista di Studi Sulla Sostenibilit 2014; 2: 189-206.

Gabay D, Moulin H. On the uniqueness and stability of Nash equilibria in noncooperative

games. In Applied Stochastic Control of Econometrics and Management Science. A Ben-
soussan, P Kleindorfer, CS Tapiero, Editors; North-Holland, Amsterdam, The Netherlands;
1980, 271-294.

Giovannucci D, Barham E, Pirog R. Defining and marketing “local” foods: Geographical

indications for US products. Journal of World Intellectual Property 2010; 15(2): 94-120.

Hendrickson J. Energy use in the US food system: A summary of existing research and
analysis. Center for Integrated Agricultural Systems, University of Wisconsin, Madison,
Wisconsin; 2004.

Holloway L, Kneafsey M. Reading the space of the farmers’ market: A preliminary investi-
gation from the UK. Sociologia Ruralis 2000; 40(3): 285-299.

Iowa State University Extension. Harvesting and storing apples; Ames, Iowa, 2008.

Kader AA. Fruit maturity, ripening, and quality relationships. In International Symposium
Effect of Pre-& Postharvest Factors in Fruit Storage 1997; 485: 203-208.

Karim OR, Adebowale AA. A dynamic method for kinetic model of ascorbic acid degradation

39
during air dehydration of pretreated pineapple slices. International Food Research Journal
2009; 16: 555-560.

Ketzenberg M, Bloemhof J, Gaukler G. Managing perishables with time and temperature

history. Production and Operations Management 2015; 24(1): 5470.

Kinderlehrer D, Stampacchia G. An Introduction to Variational Inequalities and Their Ap-

plications. Academic Press, New York; 1980.

King RP, Hand MS, Gomez MI, editors. Growing Local: Case Studies on Local Food Supply
Chains. University of Nebraska Press, Lincoln, Nebraska; 2015 Feb 1.

Krokida MK, Karathanos VT, Maroulis ZB, Marinos-Kouris D. Drying kinetics of some
vegetables. Journal of Food Engineering 2003; 59(4): 391-403.

Labuza TP. Shelf-life Dating of Foods. Food & Nutrition Press, Westport, CT; 1982.

Labuza TP. Application of chemical kinetics to deterioration of foods. Journal of Chemical

Education 1984; 61(4): 348-358.

Liu G, Zhang J, Tang W. Joint dynamic pricing and investment strategy for perishable foods
with price-quality dependent demand. Annals of Operations Research 2014; 226(1): 397416.

Low SA, Adalja A, Beaulieu E, Key N, Martinez S, Melton A, Perez A, Ralston K, Stewart
H, Suttles S, Vogel S, Jablonski BBR. Trends in U.S. local and regional food systems, AP-
068, U.S. Department of Agriculture, Economic Research Service, Washington DC; 2015.
January.

Lowe TJ, Preckel PV. Decision technologies for agribusiness problems: A brief review of
selected literature and a call for research. Manufacturing & Service Operations Management
2004; 6(3): 201-208.

Maftoonazad N, Ramaswamy HS. Effect of pectin-based coating on the kinetics of quality

change associated with stored avocados. Journal of Food Processing and Preservation 2008;
32(4); 621-643.

Martinez S, Hand M, Da Pra M, Pollack S, Ralston K, Smith T, Vogel S, Clark S, Lohr L,

Low S, Newman C. Local food systems concepts, impacts, and issues. Economic Research
Report Number 97. Economic Research Service, United States Department of Agriculture,
Washington DC; 2010.

40
Masoumi AH, Yu M, Nagurney A. A supply chain generalized network oligopoly model for
pharmaceuticals under brand differentiation and perishability. Transportation Research E
2012; 48(4), 762-780.

Mrad ND, Boudhrioua N, Kechaou N, Courtois F, Bonazzi C. Influence of air drying tem-
perature on kinetics, physicochemical properties, total phenolic content and ascorbic acid of
pears. Food and Bioproducts Processing 2012; 90(3): 433-441.

Murdoch J, Marsden T, Banks J. Quality, nature, and embeddedness: Some theoretical

considerations in the context of the food sector. Economic Geography 2000; 76(2): 107-125.

Nagurney A. Network Economics: A Variational Inequality Approach, second and revised

edition. Kluwer Academic Publishers, Boston, Massachusetts; 1999.

Nagurney A. Supply Chain Network Economics: Dynamics of Prices, Flows and Profits.
Edward Elgar Publishing Inc., Cheltenham, UK; 2006.

Nagurney A. Supply chain network design under profit maximization and oligopolistic com-
petition. Transportation Research E 2010; 46(3): 281-294.

Nagurney A, Dupuis P, Zhang D. A dynamical systems approach for network oligopolies and
variational inequalities. Annals of Regional Science 1994; 28: 263-283.

Nagurney A, Li D. Competing on Supply Chain Quality: A Network Economics Perspective.

Springer Business + Science Media, Springer International Publishing Switzerland; 2016.

Nagurney A, Yu M, Masoumi AH, Nagurney LS. Networks Against Time: Supply Chain
Analytics for Perishable Products. Springer Business and Science Media, New York; 2013.

Nagurney A, Zhang D. Projected Dynamical Systems and Variational Inequalities with Ap-
plications. Kluwer Academic Publishers, Norwell, Massachusetts; 1996.

Nahmias S. Perishable Inventory Systems. Springer Science & Business Media, New York;
2011.

Nash JF. Equilibrium points in n-person games. Proceedings of the National Academy of
Sciences, USA 1950; 36: 48-49.

Nash JF. Noncooperative games. Annals of Mathematics 1951; 54: 286-298.

Neff RA, Palmer AM, McKenzie SE, Lawrence RS. Food systems and public health dispar-

41
ities. Journal of Hunger and Environmental Nutrition 2009; 4: 282-314.

Ochoa MR, Kesseler AG, De Michelis A, Mugridge A, Chaves AR. Kinetics of colour change
of raspberry, sweet (Prunus avium) and sour (Prunus cerasus) cherries preserves packed in
glass containers: light and room temperature effects. Journal of Food Engineering 2001;
49(1): 55-62.

O’Connor A. Really? The claim: Fresh produce has more nutrients than canned. The New
York Times; 2013. May 27.

Pirog R. Local foods: farm fresh and environmentally friendly. Leopold Center for Sustain-
able Agriculture, Ames, Iowa; 2009.

Pirog R, Van Pelt T, Enshayan K, Cook E. Food, fuel, and freeways: An Iowa perspective
on how far food travels, fuel usage, and greenhouse gas emissions. Leopold Center for
Sustainable Agriculture, Iowa State University, Ames, Iowa; 2010.

Reeves CA, Bednar DA. Defining quality: alternatives and implications. Academy of Man-
agement Review 1994; 19(3): 419-445.

Rong A, Akkerman R, Grunow M. An optimization approach for managing fresh food quality
throughout the supply chain. International Journal of Production Economics 2011; 131(1):
421-429.

Schouten RE, Kooten OV, Jalink H, Kappers IF, Snel JFH, Jordi W, Steele R. Genetic
and physiological factors affecting colour and firmness. Understanding and Measuring the
Shelf-life of Food 2004; May 21: 69-90.

Singh RP, Anderson BA. The major types of food spoilage: an overview. Understanding
and Measuring the Shelf-life of Food 2004; 3-23.

Singh TK, Cadwallader KR. Ways of measuring shelf-life and spoilage. Understanding and
Measuring the Shelf-life of Food 2004; 165-183.

Sloof M, Tijskens LMM, Wilkinson EC. Concepts for modelling the quality of perishable
products. Trends in Food Science & Technology 1996; 7(5):165-171.

Taoukis PS, Labuza TP. Applicability of time-temperature indicators as shelf life monitors
of food products. Journal of Food Science 1989; 54(4): 783-787.

Tijskens LMM, Polderdijk JJ. A generic model for keeping quality of vegetable produce

42
during storage and distribution. Agricultural Systems 1996; 51(4): 431-452.

Tijskens LMM. Texture of Golden Delicious apples during storage. Lebensmittel-Wissenschaft

und Technologie 1979; 12: 138-142.

Tong D, Ren F, Mack J. Locating farmers’ markets with an incorporation of spatio-temporal

variation. Socio-Economic Planning Sciences 2012; 46(2): 149-156.

Toralles RP, Vendruscolo JL, Vendruscolo CT, del Pino FAB, Antunes PL. Properties of
polyphenoloxidase and peroxidase from Granada clingstone peaches. Brazilian Journal of
Food Technology 2005; 8(3): 233-242.

Trobe HL. Farmers’ markets: consuming local rural produce. International Journal of Con-
sumer Studies 2001; 25(3): 181-192.

United States Department of Agriculture (USDA). US apple statistics. Economic Research

Service, Washington DC; May 2012.
http://usda.mannlib.cornell.edu/MannUsda/viewDocumentInfo.do?documentID=1825

United States Department of Agriculture (USDA). Local food directories: National farmers
market directories. Agricultural Marketing Service, Washington DC; June 2016.
https://www.ams.usda.gov/local-food-directories/farmersmarkets

U.S. Energy Information Administration. Gasoline and fuel update. June 2016.
http://www.eia.gov/petroleum/gasdiesel/

Van der Vorst JGAJ. Effective food supply chains: Generating, modelling and evaluating
supply chain scenarios. Doctoral Dissertation, Wageningen University, The Netherlands;
2000.

Wilkins JL, Bokaer-Smith J, Hilchey D. Local foods and local agriculture: A survey of
attitudes among Northeastern consumers. Project Report. Division of Nutritional Sciences,
Cornell Cooperative Extension, Ithaca, New York; 1996.

Yu M, Nagurney A. Competitive food supply chain networks with application to fresh pro-
duce. European Journal of Operational Research 2013; 224(2): 273-282.

Zhang N, Guo Q, Ma Y. Stability of anthocyanin in wild blueberry fruit and its degradation
kinetics. In Information Technology and Agricultural Engineering 2012, Springer, Berlin,
Germany; 539-546.

43
Zhang G, Habenicht W, Spieß WEL. Improving the structure of deep frozen and chilled food
chain with tabu search procedure. Journal of Food Engineering 2003; 60: 67-79.

Zind T. Fresh trends 1990: A profile of fresh produce consumers. In The Packer Focus
1989-90, Vance Publishing Company, Overland Park, Kansas, 1990; 37-68.

44