See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/227520884

Supply Chain Strategies for Perishable Products: The Case of Fresh Produce

Article  in  Production and Operations Management · March 2009

DOI: 10.1111/j.1937-5956.2009.01016.x

CITATIONS READS
261 11,549

2 authors:

Joseph D. Blackburn Gary D. Scudder

Vanderbilt University Vanderbilt University
49 PUBLICATIONS   3,383 CITATIONS    52 PUBLICATIONS   2,774 CITATIONS   

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Joseph D. Blackburn on 18 November 2018.

The user has requested enhancement of the downloaded file.

Supply Chain Strategies for Perishable Products: The Case of Fresh Produce

Professor Joseph Blackburn

Owen Graduate School of Management
Vanderbilt University
Nashville, TN 37203
Joe.blackburn@owen.vanderbilt.edu
(615) 322-0645
(615) 343-7177 (fax)

Professor Gary Scudder

Owen Graduate School of Management
Vanderbilt University
Nashville, TN 37203
Gary.scudder@owen.vanderbilt.edu
(615) 322-2625
(615) 343-7177 (fax)

September 1, 2007
Revised September 4, 2007
OGSM Working Paper #03-14
Vanderbilt University
Nashville, TN
 Supply Chain Strategies for Perishable Products: The Case of Fresh Produce

Abstract

This paper examines supply chain design strategies for a specific type of

perishable product—fresh produce, using melons and sweet corn as examples. Similar to

other perishable products, melons and other types of produce reach their peak value at

time of harvest (or production); product value deteriorates exponentially post-harvest

until the product is cooled to dampen the deterioration. The supply chain design structure

that we propose is an extension of the efficient and responsive supply chain models

proposed by Fisher (1997). Using the product’s marginal value of time- the rate at which

the product loses value over time in the supply chain-- we show that the appropriate

model to minimize lost value in the supply chain is a hybrid of a responsive model from

post-harvest to cooling, followed by an efficient model in the remainder of the chain. We

also show that these two segments of the supply chain are only loosely-linked, implying

that little coordination is required across the chain to achieve value maximization. The

models we develop also provide insights into the use of a product’s marginal value of

time to develop supply chain strategies for other perishable products.

(Supply Chain Management; Perishable Products; Fresh Produce; Marginal Value

of Time)

2
 Supply Chain Strategies for Perishable Products: The Case of Fresh Produce

1. Introduction

This paper considers the problem of designing and managing effective supply

chains for a specific type of perishable product, fresh produce. Unlike conventional

products, the value of perishable products changes significantly over time in the supply

chain at rates that are often highly temperature and humidity dependent. We show that

these changes in product value make conventional supply chain strategies inappropriate.

For many products, a decision about supply chain strategy reduces to a choice

between responsiveness and efficiency. The appropriate choice depends on how the

product changes in value over the time interval between production and delivery to the

customer. To illustrate this, we define the term marginal value of time to be the change

in value of a unit of product per unit time at a given point in the supply chain; it measures

the cost of a unit time delay in the chain. When the marginal value of time remains

relatively constant over the supply chain, then a single design choice—responsiveness or

efficiency--applies consistently throughout the chain. This is the case for many

undifferentiated products However, for perishable products, we show that because of

dramatic changes in product value and in the cost of time delays, no single design choice

is appropriate for the entire chain, and we develop more complex supply chain strategies

that are combinations of speed and efficiency.

This study focuses on optimal supply chain design for perishable products, using

fresh melons and sweet corn as examples. These products are representative of other

perishable products in that they exhibit a rapid decline in value over time under certain

conditions and have limited shelf-life. With melons and sweet corn, we can also model

3
the effects on supply chain performance of actions to control the rate of perishability.

The results we obtain offer useful insights for using the time value of a product to

optimize supply chain performance.

The structure of the paper is as follows. The next section describes the problem

of maximizing value in the supply chain for fresh melons and sweet corn. We present a

review of relevant literature, followed by a summary of design strategies that have been

suggested for conventional supply chains. We then reinterpret those strategies in terms of

the marginal value of time in supply chains and show that the appropriate supply chain

for these products is a hybrid of conventional strategies. Finally, we analyze the specific

design problems posed in different segments of the supply chain and discuss

implementation issues.

2. Managing the Supply Chain for Melons

Figure 1 is a schematic of the sequence of activities in the supply chain for

melons. With melons and sweet corn, as with other fresh produce, the maximum quality

(and value) of the product is largely determined by actions taken in the early stages of the

process: seed production, growing conditions, planting practices, and harvesting methods.

As perishable products, quality begins to deteriorate once they are picked, and the supply

chain management problem is to control the loss in product quality over the remaining

stages in the chain—from the field to the customer. The focus of our study is

optimization of the supply chain post-harvest; we do not explore the agricultural issues

surrounding seed production and the growing operations.

In the large produce operations we have observed in California, melons and sweet

corn are hand picked and field packed, an extremely labor-intensive process. With

4
melons, workers pick the product and toss them onto an open flatbed truck driving

directly in front of them in the field. Workers on the truck sort the melons, discarding

damaged or immature fruit, and pack the salable items in cartons according to size.

Periodically, these cartons are transferred to a nearby truck. In the peak season, a truck is

filled with melons in about three to four hours, with multiple crews harvesting a given

field. When full, the truck is driven to the cooling shed, where the melons are hydro,

forced-air or vacuum cooled because proper cooling is necessary to preserve product

quality. The process for sweet corn is similar.

Figure 1

A Schematic of the Melon Supply Chain

Seed Growing Harvest/ Field Cooling/

Production Ops Pick Pack Storage

Retail/FS Retail/FS Retail

Transit Consumer
Warehouse Delivery Handling

At the time of harvest, a melon or an ear of corn is at its peak value in quality and

taste; sugar content largely determines quality, and the fresh-picked product is at its

maximum sweetness level. As an organic product, the freshly-picked product begins a

chemical process of respiration; respiration not only generates carbon dioxide (CO2) and

heat, but it also turns converts sugar to starch, causing the product to lose sweetness and

5
quality. Figure 2 displays laboratory measurements for the rate of respiration of melons

and sweet corn, showing that the respiration rate (and loss in sugar content) increases

significantly with temperature [Hardenburg et al. (1986)], and that sweet corn has higher

respiration rates than melons. Figure 3 displays the effect of respiration on quality (and

value) by showing that the loss of sweetness in corn over time follows an exponential

decay function whose decay rate increases dramatically with temperature [Appleman and

Arthur (1919)]; melons exhibit a similar functional relationship between sweetness, time

and temperature [Suslow, Cantwell and Michell (2002)].

Because freshly-picked produce can have an internal temperature reaching 90-95

degrees Fahrenheit, quickly removing field heat is critical to maintaining product quality;

therefore it is very important to move the product rapidly from the field to a cooling shed

to preserve product quality [Jobling (2002); Sargent et al. (2000) ] Hartz, Mayberry and

Valencia (1996) observe that rapid removal of field heat maximizes-post harvest life.

Once the melon or corn reaches the cooling shed and has been cooled to a temperature a

few degrees above freezing, product deterioration and value loss occur at a much lower

rate. The product value (its taste and appearance) can be maintained for several weeks,

provided that the “cold chain” is maintained throughout the remaining stages of the chain

as the product moves from the field to retailers across the U.S. [Perosio et al. (2001)].

6
 Figure 2

Respiration Rate for Melons and Sweet Corn

(Mg. CO2/kg. hr)

450

400

350

300
Respiration Rate

Sweet Corn
250

200

150
Melons
100

50

0
0 5 10 15 20 25 30
Temperature (deg. C.)

-------Figure 3 to be inserted here------

Our objective is to develop supply chain strategies to minimize lost value in the

post-harvest supply chain for fresh melons and sweet corn. That is, we seek to maximize

the value of the product delivered to the customer, net of the cost of managing the supply

chain process. Figure 4 shows schematically how the typical product loses value over the

supply chain: in the critical time period between picking and cooling (t0 to t1 in Figure 4),

product loses value at a rapid, exponential rate (as shown in Figure 3); in the interval

post-cooling (t1 to t2 in Figure 4) , the product’s value declines at a much slower rate.

We consider how to minimize the loss in product value over the entire post-harvest time

interval: before and after the “cold chain” is established.

7
 Figure 4
Declining Value of Product over Time

Value Field Temp.

0-50 C.
Purchase

t1
t0 Up to t2 Time
Time of “Cold-chain” 2 wks End of
Picking Established “Cold-Chain”

3. Literature Review

In developing supply chain strategies for perishable food products, we build upon

two distinct research streams: models for perishable inventory management and supply

chain design structures. We summarize the most relevant research in each of these

streams and integrate the concepts into a more general model for the supply chain for

perishable products.

Numerous models for managing the inventory of a perishable product models

have been developed (see Nahmias (1982) for a thorough review of the early literature).

Of particular relevance to the current study are models that deal with degradation of

product quality and value over time. In most of the early studies on perishable inventory,

perishability was defined as the number of units of product that become unusable or

obsolete over time—the decay was not in value, but in the number of units, and the decay

was modeled with a probability distribution. For example, Ghare and Schrader (1963)

8
developed an EOQ models for products in which the number of usable units were subject

to exponential decay. Covert and Philip (1973) and Philip (1974) extended this model,

but used the Weibull distribution to model item deterioration. Shah (1977) extended the

model to allow for shortages and backlogging, and Tadikamalla (1978) examined the

case of Gamma-distributed deterioration. Giri and Chaudhuri (1978) and Chakrabarty et

al. (1978) extended these models to include situations in which demand rate is dependent

upon either the inventory level or time.

One of the first papers to consider deterioration in product value over time was by

Fujiwara and Perera (1993), who developed EOQ models for inventory management

under the assumption that product value diminishes over time according to an exponential

distribution. However, they assume that the rate of deterioration of product value

increases with the age of the inventory. Research on the perishability of fresh produce

indicates that, unlike Fujiwara and Perera’s model, the loss in product value and quality is

at its highest rate immediately post-production (at harvest), and the rate of loss in value

declines until the produce finally “spoils” [Hardeman et al. (1986), Appleman and

Arthur (1919)].

To date, the perishability models that have been developed only consider

inventory management: determining appropriate levels of perishable stock to meet

demand. They do not consider broader supply chain design issues, which is the focus of

this paper. We build a model of perishability for fresh produce to examine how these

products should be managed throughout the supply chain.

A number of frameworks have been proposed for supply chain design. One of the

first was introduced by Fisher (1997), who devised a taxonomy for supply chains based

9
on the nature of the demand for the product. For functional products (stable, predictable

demand, long life cycle, slow “clockspeed”) Fisher argued that the supply chain should

be designed for cost efficiency; for innovative products (volatile demand, short life cycle,

fast “clockspeed) he maintained that the supply chain should be designed to be fast and

responsive. Lee (2002) expands upon Fisher’s taxonomy by suggesting that the supply

process should be either stable or evolving. A stable supply process has a well-

established supply base and mature manufacturing processes. In an evolving supply

process technologies are still early in their development with uncertain or limited

suppliers. In a recent article, Kopczak and Johnson (2003) discuss several changes in the

focus of business supply chain design to include such issues as coordination of activities

across companies, improving information flows, collaborative redesign of the supply

chain as well as its products and processes.

Lee and Fetzinger (1997) made a significant contribution to supply chain strategy

by introducing the concept of delayed product differentiation, or postponement. They

showed that delaying final product definition until further downstream in the chain

reduces variety in the early stages (in effect, making the product more functional). This

creates opportunities for supply chain designs that can be efficient in the early stages and

responsive in the final stages. In their studies of reverse supply chains, Blackburn et al.

[2006] use the concept of a returned product’s rapidly diminishing value over time to

show that an effective supply chain strategy with return streams can be early product

definition, or preponement, and that the early stages of the chain should be designed for

responsiveness. These studies suggest that supply chain strategies based on a simple

choice between efficiency and response can be inappropriate when the product undergoes

10
substantial differentiation or change in its value as it moves through the chain. In the

analysis that follows we show that this is the case for perishable produce: the value of the

product changes significantly and the appropriate supply chain structure is a combination

of responsiveness and cost efficiency.

In our analysis we show that the time-value profile for perishable produce

suggests a supply chain strategy based on product differentiation—a chain that is

responsive in the early stages and efficient in the later stages.

4. Maximizing Value in the Fresh Produce Supply Chain

Figure 4 shows schematically how the value of the product varies across the

supply chain for fresh produce, such as melons and sweet corn, Between t0 and t1 time

delays have a high cost because of rapid product deterioration until the product is cooled;

the supply chain must be responsive in this stage. Once cooled, product deterioration

slows and quality is stabilized: time delays have a much lower cost if the “cold chain t1

and t2. In the analysis that follows, we show that design decisions for the responsive and

efficient segments of the chain are only weakly linked, and we can separate the two

supply chain design problems into (1) the design of the responsive chain for the period

between picking and cooling; (2) devising a (cost) efficient chain for the maintenance of

the cold chain and distribution of product to retailers. We first examine the problem of

maximizing value in the responsive segment of the chain (from harvest to cooling), and

then discuss the joint problem to show that the efficient chain can be designed separately.

11
4.1 Maximizing Value in the Responsive Segment of the Chain

As previously noted, the time interval between picking of the product in the field

and transfer to the cooling shed is the most critical period for product quality in the entire

supply chain. Maximizing product value in this time interval between picking and

cooling can be modeled as an economic batch production model in which the key

decision variable is the size of the batch of product to be transferred for cooling.

Figure 5 indicates how the time interval between picking and cooling affects

product quality. From time of picking, t0, product sitting at field temperatures degrades

rapidly over the time interval between t0 until cooled (at time t1) to a temperature just

above freezing to arrest the decay in value and establish the “cold chain”. As previously

shown in Figure 3, research has established that the rate of decay in product sweetness

and quality (or value) due to heat production over the t1 – t0 interval can be adequately

modeled by an exponential decay function. In Figure 5 we model this loss in value,

generically, with an exponential function that equals value ,V, at time t0 and degrades

exponentially in time according to a function Vexp(-αt), where the parameter α depends

on both the product and the temperature (as was shown in Figure 3 for sweet corn).

[Suslow, Cantwell, and Mitchell (2002)].

12
 Figure 5
Deterioration of Product Value at Field Temperatures

Value

Vexp(-αt)

t0 t1 Time

Table 1 displays the observed α values at various temperatures for melons and

sweet corn.

Table 1
Rate of Exponential Decay in Product Value

Field Temp, °C. Melons Sweet Corn

0 0.001 0.005
10 0.003 0.015
20 0.006 0.027
30 0.03 0.13

13
 The loss in value during this critical time period depends on the transfer batch size

Q and the picking rate because they determine the length of time that the product remains

at high field heat, losing value. The appropriate value for Q is determined by a tradeoff

that pivots on the cost of time delays for the product.

Figure 6 is a schematic depicting the tradeoff facing the grower and distributor in

the selection of the optimal transfer batch size, Q. Assuming that the loss in product

quality equates to a cost of lost product value, then the loss in value is a concave

increasing function of Q. On the other hand, the cost of transferring batches of product to

the cooling facility involves a fixed transfer cost that is analogous to the setup cost in a

conventional lot-sizing problem.

Figure 6
Transfer Batch Cost Tradeoff for Perishable Products
Cost
Cost of Lost Product “Value”

Batch Transfer Cost

KD/Q

Transfer Batch
Q

To develop the model used to determine the optimal transfer batch, we assume

that the unit of analysis for product is a carton; individual units of product can vary

14
significantly in size and cartons are the standard transfer quantity.1We introduce the

following additional notation: let

D = total # of cartons picked over the harvest period;

Q = transfer batch size in cartons;
V= maximum value of a carton of product at time t0 (when picked);
p = picking rate (cartons per hour);
α = deterioration rate in value of product (dollars per hour);
K= batch transfer cost in $ (assumed to be independent of the lot size);
tr = transfer time (in hrs) from field to cooling shed.

4.2 Total Cost for the Harvest Period

The objective is to minimize the total cost of batch processing and loss in product

value over the harvest period. We assume continuous time by eliminating from

consideration time intervals when picking does not take place); then the harvest period is

of duration D/p, during which D/Q batches of product are picked. As previously noted,

cooling to remove field heat from the product arrests the respiration rate of the product,

stabilizing the rate of deterioration. Therefore, for a unit held at “field heat” for a time

interval t, the value at time t is Vexp(-αt), and the loss in value for a unit held for t time

units between picking and cooling is V[1- exp(-α t)]. Note that although the formulation

of our model has similarities to the traditional EOQ model, it differs in a significant way

from the modifications of the EOQ model that have been proposed for perishable

inventory management (which assume that the number of usable units, not their

individual value, are diminishing over time).

To construct the cost function for the optimal transfer batch size, consider the qth

unit picked in a batch Q. The qth unit is held in the field for a time t = (Q-q)/p, and the

1
For example, a carton of melon can contain 25-30 melons and weigh approximately 38 pounds.

15
loss in value for the qth unit is V[1- exp(-α t)], or V[1- exp(-α(Q-q)/p)], Therefore, the

loss in value of a batch of size Q can be expressed as

QV- ∫ V exp(−α (Q − q ) / p )dq = QV – [pV/ α][1-exp(-α Q/p)]. (1)

0

If we include the fixed transfer time tr , then expression (1) is modified as follows:

QV – [pV/ α] [exp(-αtr) ] [1-exp(-α Q/p)] (2)

The total cost per harvest period is simply the sum of the transfer costs (KD/Q) and the

cost in “loss in value” per batch incurred D/Q times. This gives the following expression:

TC(Q) = KD/Q + DV - [D/Q] [pV/ α][exp(-αtr )] [1-exp(-α Q/p)] (3)

Let τr represent the constant term exp(-αtr ), and we have

TC(Q) = KD/Q + DV - [D/Q][ τr pV/ α] [1-exp(-α Q/p)] (4)

Note that equation (4) is similar in structure to the traditional economic order

quantity (EOQ) problem: the sum of a setup (or transfer) cost term [KD/Q] and an

expression that represents the loss in value while product is held in the field (a cost of

holding stock). Yet there are some important differences: unlike the EOQ, it is the

“demand” that is instantaneous and the production (or picking) that occurs at a constant

rate. Thus, it is an economic batch production model in which the batch size is limited by

the economic loss in value prior to transfer to cooling. We do not include a traditional

inventory carrying cost in the total cost expression because, given the short time interval

(hours) and the rapid loss in product value at field temperatures, any differences in

inventory carrying cost as a function of Q are negligible. However, including an

inventory carrying cost in the model would be trivial.

16
4.3 Optimal Transfer Batch Size

Expression (4) implies that the batch size decision Q is independent of the value of D,

therefore the optimal transfer batch decision is to choose a value Q to minimize the

following modified total cost expression:

TCM(Q) = (1/Q)[K- (τr pV/α)(1-exp(-α Q/p)] (5)

To determine the optimal quantity Q, we take the derivative of (5), yielding

dTCM(Q)/dQ = (-1/Q2) )[K- ( τr pV/ α)(1-exp(-α Q/p)] –(τr V/Q) exp(-α Q/p) (6)

Set dTCM(Q)/dQ = 0, and we have

Q =exp(α Q/p)[(p/ α) -K/(τr V))] - p/ α (7)

We show in Appendix 1 that optimality conditions are satisfied; that is, at the solution to

(7), the second derivative is positive. The optimal value of Q can easily be found with a

spreadsheet using one of the available Solver routines.

We previously noted the similarity in form between equation (4) and the

traditional EOQ model. Although not obvious from (7), the optimal value of Q for our

model is bounded tightly from below by a EOQ-like square root expression. In Appendix

2 we show that this bound is

Q >= sqrt [2Kp/(τrαV)] (8)

We develop this bound, not because the optimal value of Q is difficult to compute, but to

show the close relationship between the perishable lot transfer problem and the traditional

EOQ for non-perishable items.

17
 If the “loss in value” is linearized—that is, V(exp(-αt)) is replaced by V(1-αt),

and the analysis repeated, then it is easily shown that (see Appendix 3), subject to the

constraint that αQ/p <1,

Q = sqrt [2Kp/( αV)]. (9)

That is, with a linear loss in product value over time, the optimal transfer quantity is

determined by a square root expression that is independent of the transfer time tr.

4.4 Example: Applying the Model in the Responsive Segment

To illustrate how the model can be used to maximize value in melon production,

we describe the problem faced by melon growers in California’s Central Valley. The

following description is based on our observations of melon harvesting, supplemented by

data from agribusiness research [Mayberry and Meister (2003)]. Melons are picked by

multiple teams of workers (10-20 workers) who move through the field behind a trailer

pulled by a tractor. As melons are picked, they are tossed to workers on the trailer who

pack them into cartons (of up to 30 melons). Picking rates by the team average about

50-60 cartons per hour. Cartons are stacked onto pallets, 42 cartons per pallet, and

trailers can hold about 12-14 pallets, or up to about 590 cartons of melons.

Full trailers are transported by truck to cooling sheds to remove field heat and to

preserve product quality. Cooling sheds are located throughout a growing region and

serve as both a cooling facility and a source for outbound truck shipments; they serve

several growers in a region. (These are typically owned and operated separately from the

growing operations. Thus, the location of these facilities is not considered here.)

Consequently, the time and cost to transfer a batch of cartons from the field to the cooling

shed depends on the location of the field. The time to transfer melons from the field to

18
cooling can vary from 15 minutes to up to an hour, and we estimated the cost of a truck

transfer to be between $50 and $100. Based upon yields, the average value of spring

harvested cantaloupes from California in 2003 was about $7.00 per carton.

For our example we assume the following: The value of a carton of melons at

time of picking is $7.002. For melons picked at a field temperature of 30 oC. (86 oF.),

Table 1 shows that the product loses value at a decay rate per hour of α = 0.03. The

picking rate p = 60 cartons per hour. We assume a transfer time, tr, of 0.5 hours and a

transfer cost, K, of $75.

Using equation (7), we find that under these conditions the optimal transfer

quantity is 216 cartons. With a picking rate of about 60 cartons/hour, the optimal Q value

indicates that melons should be transported to the cooling shed every 3 1/2 hours.

However, a full trailer load is 590 cartons, and a typical harvesting team would require

more than 9 hours to fill a trailer. Many growers use multiple picking teams during the

peak harvesting season, and, at a higher aggregate picking rate, these teams tend to fill a

trailer in about three hours, which is very close to the optimal transfer quantity.3

We note that the optimal transfer batch size is sensitive to the decay rate α, a

value that varies with harvest temperature (and with produce type). Figure 7 displays

optimal values of Q over a range of α values for our example problem. When the

deterioration rate is as low as α = 0.01 (corresponding to a field temperature of about 70

o
F), the optimal batch quantity for a single picking team is about 60% of a full trailer

2
In this example we have simply used the average value of cantaloupes based on University of California
data. Actual product value at time of picking would vary, and could be higher, but this would not
substantially alter the analysis or the conclusions.
3
The approximations also provide a reasonably tight lower bound on the optimal transfer batch size. The
lower bound, given by the square root expression (8), is 209 cartons,

19
quantity. The deterioration rate would need to be as low as .005 to justify transferring a

full trailer batch of 590 cartons.

Figure 7
Optimal Transfer Batch Size for Melon Example

400

350

300
Transfer Batch Size, Cartons

250

200

150

100

50

0
0.01 0.03 0.05 0.075 0.1 0.125
Deterioration Rate Alpha

5. Separability of the Responsive and Efficient Chain Designs

Once the “cold chain” has been established, the product can remain stable for two

to three weeks. The value of the product at the time the cold chain is established is a

function of Q, as was shown in (1): [pV/ α] [exp(-αtr )] [1-exp(-α Q/p)]

Product is then shipped by a refrigerated vehicle through a distribution center on

to the retailer. If the cold chain is maintained throughout, research on fresh produce

respiration rates indicates that the loss in value over time can be represented by a function

with a small negative slope (see Suslow, Cantwell and Mitchell (2002)). Since the cost

20
of time delays is small over this segment of the supply chain, cost efficiency—rather than

responsiveness—becomes the overriding objective and the design problem simplifies to

an economic choice among transportation modes (air, rail, or refrigerated truck). We

now construct a total cost model for the entire supply chain to show that the choice of

transportation mode has little effect on the optimal transfer batch, Q, in the responsive

segment of the chain.

The total cost analysis of the responsive and efficient supply chain is carried out

under the assumptions that t the loss in value over the efficient segment of the chain

follows an exponential decay rate: If V is the value of a unit at the beginning of the

efficient segment (the time at which the “cold chain” is established), then the value t

time units later is Vexp(-βt), where β << α. 4

The choices of transportation mode or carrier are typically made from a small,

finite set of size n; there is not a continuum of cost/time choices. In practice, the cost of

transport and the time to transport the product from cooling shed to retailer are essentially

determined by the mode of transportation. That is, shipment by refrigerated truck would

have a market-determined cost and time that are determined largely by distance and, to a

lesser extent, distribution center practices. Shipment by another mode of transport, such

as rail, would have a different cost and time. For this analysis, we simply assume that

there are n possible logistics choices. To construct a formal model of the supply chain

design decision, let

j denote the transportation mode;

tj = transportation time for mode j (=1,…n);

4
A linear decay function could also be assumed, since β and, therefore, βt are small, the two deterioration
functions are not substantially different, and both are good approximations to the real, observed loss in
product value. We show that, in either case, the efficient design decision has little effect on the responsive
segment. The analysis is similar and is omitted here.

21
 Cj = cost of transportation mode j (=1,…n).

Since refrigerated truck is the most common option used for the transport of fresh

produce, our model allows for the possibility of alternative truck carriers with different

time and cost profiles. In the analysis, each different truck alternative can be identified as

a different mode j.

5.1 Analysis under Exponential Decay

Assuming that product value deteriorates exponentially in this segment of the

chain, let τj = exp(-βtj ). Then, using expression (4) and including the loss in value over

the “efficient” segment of the chain and the transport cost, we have the following total

cost expression for the supply chain design:

TC(Q,j) = (D/Q)[K- (τr τj pV/α)(1-exp(-α Q/p)] + Cj (10)

Note that the only interaction between the choice of transportation mode and the optimal

transfer quantity Q is captured in the τj factor, and this factor is very close to 1 because

the deterioration rate β is small. The interaction is minimal.

Given a design choice j, the optimal value of Q is independent of the value of Cj

and by analysis similar to that carried out in the previous section we can develop an

analog of equation (8):

Q >= sqrt [2pK/( τβ τj αV)] (11)

This lower bound provides a very good approximation to the optimal value of Q for the

given transportation mode, and the optimal value can be determined by a non-linear

search procedure as described in the preceding section.

22
 Because there are only a small, finite set of design choices, j, for the efficient

segment of the chain, it is feasible to evaluate TC(Q,j) for each value of j to determine the

optimal design of both the efficient and the responsive segment of the chain.

5.2 Example: Applying the Model to the Efficient Segment of the Chain

Extending the example introduced in the preceding section, with β=0.02 per day

and tj=5 days, we calculate the optimal value Q = 227. The batch size Q is very similar to

the value calculated for the responsive segment alone, indicating that the efficient

segment of the chain has a minimal effect on the responsive chain. Doubling tj to 10

days for transportation mode j only increases the optimal Q value to 239. For field

managers, the appropriate decision is still to transport product to the cooling shed

approximately every 3-4 hours (at a 60 carton/hr picking rate). The solution is quite

robust with respect to the choice of transportation mode, and the decisions made in the

efficient and responsive chain can be decoupled without significant deviation from

optimality.

Because the product deteriorates at a much slower rate after the “cold chain” is

established, the choice of transportation mode j for the efficient segment of the chain is

insensitive to the shipment time. Therefore, the marginal value of time for the product

may be used to impute the cost of an additional day of shipping time in order to evaluate

alternative modes of transport. To compute the marginal cost in lost product value (over

the entire harvest season) of an additional day of shipping time, we take the partial

derivative with respect to tj of TC(Q,j), as given in (10). This yields

23
∂TC ⎛D⎞
= ( βτ j )⎜⎜ ⎟⎟(τ r ) (pV/α)[1-exp(-α Q/p)] (12)
∂t j ⎝Q⎠

To calculate the marginal cost of an additional day of transportation time, we assume a

six week harvest season, picking eight hours per day, to calculate D = 20,200 cartons of

melons. Using (12) and the data from our example, the incremental lost product value of

an additional day of transportation time is $90. On a per unit basis, the marginal cost is

$0.004, or less than a half cent per carton. The cost in lost product value of an additional

day in shipping is quite small, so the choice of shipping mode should be based primarily

on reliability (maintaining the cold chain) and cost of shipment. The marginal value of

time simplifies the evaluation of alternative transportation modes.

7. Summary and Conclusions

This paper examines a supply chain design problem for fresh produce, as an

example of a perishable product whose value declines exponentially post-production and

can then be stabilized. By using the marginal cost of time for a product to develop a

supply chain strategy, significant differences emerge between conventional supply chain

strategies and those needed for perishable products. The supply chain for melons

separates into two essentially independent segments: a “responsive” segment in which

product deterioration rates are high and an “efficient” segment with lower deterioration

rates. The “responsive” design is shown to be an optimal batch production problem that

is similar in structure to an EOQ model, and in the “efficient” segment, the design

decision reduces to a choice among competing transportation modes.

24
 An important result of this paper is that coordination across the chain is not a

requirement for supply chain optimization. The loose-linkage between the responsive

chain segment and the efficient segment means that each can be “optimized” without

affecting the other, or the overall quality of the product. By managing the process from

picking through cooling, growers can maximize product value in the responsive segment

of the chain by implementing optimal transfer batch sizes. Shipping decisions can be

made based on cost efficiency, subject to the constraint that the “cold chain” must be

maintained throughout.

Although we have specifically modeled the supply chain for melons and corn, we

note that our model also applies with minor modification to other fresh produce products

that mature in the field and reach their peak value at time of harvest. Other products,

notably flowers and seafood, have time-value patterns in the supply chain that are similar

to melons and so our general results about supply chain strategy also apply to these

classes of perishable products. We also note that for perishable products whose loss in

value cannot be stabilized, but continue to lose value at an exponential (or linear) rate, the

model we develop for the responsive segment of the chain can be used for supply chain

optimization.

This study introduces the concept of the marginal cost of time for a product as a

tool to analyze supply chain strategy. We have begun to explore extensions of the

concept to other product classes with different cost/time profiles. For example, there are

other types of fresh produce, such as tomatoes and bananas, which are often picked

before maturity and are allowed to ripen to their peak quality (and value) post-harvest.

Designing a supply chain for this type of product poses additional interesting questions

25
about the timing of production (or harvesting), managing the time interval while the

product ripens to its peak value, and preserving the product value throughout the rest of

the chain. The development of a supply chain strategy for such products with more

complex time-value profiles is an interesting future research topic.

Acknowledgements: This research was partially funded by the Dean’s Summer Research
Fund at the Owen Graduate School of Management.

26
 References

Appleman, C.O. and J.M. Arthur (1919). “Carbohydrate Metabolism in Green Sweet
Corn,” Journal of Agricultural Research, 17, 137-152.

Blackburn, J., Guide, V., Souza, G., and L. Van Wassenhove (2004). “Reverse Supply
Chains for Commercial Returns”, California Management Review, 46, 2, 6-22.

Chakrabarty, T., Giri, B.C., and K.S. Chaudhuri (1998). “An EOQ Model for Items with
Weibull Distribution Deterioration, Shortages, and Trended Demand: An
Extension of Philip’s Model”, Computers and Operations Research, 25, 7/8,
649-657.

Covert, R.P. and G.C. Philip (1973). “An EOQ Model for Items with Weibull
Distribution Deterioration”, AIIE Transactions, 5, 4, 323-326.

Fisher, M.L. (1997). “What is the Right Supply Chain for Your Product?” Harvard
Business Review, March-April, 105-116.

Fujiwara, O. and U.L.J.S.R. Perera (1993). “EOQ Models for Continuously Deteriorating
Products Using Linear and Exponential Penalty Costs”, European Journal of
Operational Research, 70, 1, 104-114.

Ghare, P.M. and G.F. Schrader (1963). “A Model for an Exponentially Decaying
Inventory”, The Journal of Industrial Engineering, XIV, 5, September-October,
238-243.

Giri, B.C. and K.S. Chaudhuri (1998). “Deterministic Models of Perishable Inventory
with Stock-Dependent Demand Rate and Nonlinear Holding Cost”, European
Journal of Operational Research, 105, 3, 467-474.

Hardenburg, R., Watada, A., and C. Wang (1986). The Commercial Storage of Fruits,
Vegetables, and Florist and Nursery Stocks, U.S. Department of Agriculture,
Agriculture Handbook No. 66.

Hardin, B. and J. Lee (1999). “Melons are on a Roll”, Agricultural Research, February ,
18-19.

Hartz, T.K., Mayberry, K.S., and Valencia, J. (1996). “Cantaloupe Production in

California”, Vegetable Research and Information Center, University of California.

Jobling, J. (2002). “Postharvest Management of Fruit and Vegetables”, Good Fruits and
Vegetables, January.

Kopczak, L., and E. Johnson (2003). "The Supply Chain Management Effect", Sloan
Management Review, 44, 3, 27-34.

27
Lee, H. (2002). "Aligning Supply Chain Strategies with Product Uncertainties",
California Management Review, 44, 3, 105-119.
Lee, H. and E. Feitzinger (1997). "Mass Customization at Hewlett-Packard: The Power
of Postponement", Harvard Business Review, Jan-Feb, 1997, 115-121.

Mayberry, K. and H. Meister (2003). "Sample Cost to Establish and Produce

Cantaloupes", University of Californai Cooperative Extension CN-IM-03-1.

Nahmias, S. (1982). “Perishable Inventory Theory: A Review”, Operations Research,

30, 4, July-August , 680-708.

Perosio, D.J., McLaughlin, E.W., Cuellar, S, and K.Park (2001). “FreshTrack 2001
Supply Chain Management in the Produce Industry”, Cornell University, Ithaca,
NY, September.

Philip, G.C. (1974) “A Generalized EOQ Model for Items with Weibull Distribution
Deterioration”, AIIE Transactions, 6, 2, 159-162.

Sargent, S.A., Ritenour, M.A. and J.K. Brecht (2000). “Handling, Cooling and Sanitation
Techniques for Maintaining Postharvest Quality”, Document HS719,
Horticultural Sciences Department, University of Florida, December.

Shah, Y.K. (1977). “An Order-level Lot Size Inventory Model for Deteriorating Items”,
AIIE Transactions, 9, 1, 108-112.

Suslow, T.V., Cantwell, M. and J. Mitchell (2002). “Cantaloupe: Recommendations for

Maintaining Postharvest Quality”, Department of Vegetable Crops, University of
California, Davis.

Tadikamalla, P.R. (1978). “An EOQ Model for Items with Gamma Distribution
Deterioration”, AIIE Transactions, 10, 1, 100-103.

Zuurbier, P.J.P.(1999). “Supply Chain Management in the Fresh Produce Industry: A

Mile to Go”, Journal of Food Distribution Research, 30, 1, 20-30.

28
Appendix 1: Verifying Optimality Conditions

From (8), we examine the second derivative of the TCM expression:

d 2TCM ⎛ 2 K ⎞ ⎡⎛ V ⎞ ⎤ ⎛ ' ⎞ ⎛ ⎞
=⎜ ⎟ − ⎢⎜ ⎟⎛⎜1 − ε −α Q ⎞⎟⎥ + ⎜ α V ⎟ε −α Q + ⎛⎜ V ⎞⎟ε −α Q + ⎜ V ⎟ε −α Q
' ' ' '

⎜ Q3 ⎟ ⎢⎜ Q 2α ' ⎟⎝ ⎠⎥ ⎜ Q ⎟ ⎜Q⎟ ⎜ ⎟
dQ 2 ⎝ ⎠ ⎣⎝ ⎠ ⎦ ⎝ ⎠ ⎝ ⎠ 2
⎝Q ⎠

Multiply by Q and rearrange—

Qd 2TCM ⎛ K ⎞ ⎡⎛ K ⎞ ⎛ V ⎞ ⎤
=⎜ ⎟ + ⎢⎜ ⎟−⎜ ⎟(1 − ε −α Q ) + ⎛⎜ V ⎞⎟ε −α Q ⎥ + V (1 + α ' )ε −α Q
' ' '

⎜ Q 2 ⎟ ⎢⎜ Q 2 ⎟ ⎜ Qα ' ⎟ ⎜ ⎟
dQ 2 ⎝ ⎠ ⎣⎝ ⎠ ⎝ ⎠ ⎝ ⎠
Q ⎥
⎦
⎛ K ⎞ ⎡ dTCM ⎤
=⎜ ⎟+
⎜ Q 2 ⎟ ⎢⎣ dQ ⎥⎦
+ V 1 + α ' ε −α ' Q ( )
⎝ ⎠

When Q is such that dTCM/dQ = 0, then we have

Qd 2TCM
+ V (1 + α ')ε −αQ ≥ 0 ,
K
=
dQ 2 Q2

verifying that the function is at its minimum.

Appendix 2: Derivation of Lower Bound on Optimal Transfer Batch Size Q

From (7), we have Q =exp(α Q/p)[(p/ α) -K/(τr V))] - p/ α

Let α/p = α’, and then the equation can be rewritten as:

Q = exp(α’Q)[(1/ α’) - K/(τr V) ] - 1/ α’

1−α’ K/(τr V) = (1 + α’Q) exp(-α’Q).

Replacing the exponential by its infinite series expansion yields

1−α’ K/(τr V) = (1 + α’Q) [1 - α’Q+(α’Q)2/2!- (α’Q)3/3!+…],

an expression that is equivalent to

Q2 = 2K/(τrα’V) + (2/3) (α’)Q3 – (1/4) (α’)2Q4 + (1/15) (α’)3Q5 …

For small α’ the terms in the expansion are diminishing, and

Q >= sqrt [2K/(τrα’V)]

29
 For the values of K, V, and α’ encountered in practice, this is a tight lower bound on the

optimal value of Q.

Appendix 3: Derivation of Optimal Transfer Quantity Under Linear Loss in Value

If the value of a unit held t time units is V[1-αt], then the loss in value over t time units is
Vαt. The loss in value of a batch of size Q can be expressed as

Q
VαQ 2
∫ Vα (Q − q) / p )dq =
2p
0
If we include the transfer time t, then the loss in value of a batch of size Q is
VαQ 2
+ Vαt r Q
2p

and the analog of expression (4) is

Vα Q 2
TC(Q) = KD/Q + [D/Q][ + Vα t r Q ]
2p
Expression (18) is independent of D and the transfer time tr. Taking the derivative and
solving for Q yields

2 Kp
Q= .
αV

30

View publication stats