Renewable Resource Management with Alternative Sources: the Case of Multiple Aquifers and a "Backstop" Resource

Abstract

While renewable resource economics is typically confined to one source and one aggregate demand, resource managers must often decide how to manage multiple sources of a resource simultaneously. In addition, studies of extraction sequencing are typically confined to non-renewable resources. We propose a dynamic optimization model to determine the efficient allocation of groundwater when two coastal aquifers are available for exploitation. We find that Herfindahl's least-cost-first result for nonrenewable resources does not necessarily apply to renewable resources, even when there is only one demand. Along the optimal trajectory extraction may switch from single to simultaneous use, depending on how the marginal opportunity cost of each resource evolves over time. A numerical simulation for the South Oahu aquifer system, which allows for differentiation of users by elevation and hence distribution costs, illustrates the switching behavior.

Full text

Renewable Resource Management with Alternative Sources: the Case of
Multiple Aquifers and a "Backstop" Resource
James Roumasset
a
Department of Economics
University of Hawai
‘
i at Mānoa
& University of Hawai
‘
i Economic Research Organization
Christopher Wada
b
Department of Economics
University of Hawai
‘
i at Mānoa
Working Paper No. 09-13R
c
October 14, 2009
Abstract
While renewable resource economics is typically confined to one source and one
aggregate demand, resource managers must often decide how to manage multiple sources
of a resource simultaneously. In addition, studies of extraction sequencing are typically
confined to non-renewable resources. We propose a dynamic optimization model to
determine the efficient allocation of groundwater when two coastal aquifers are available
for exploitation. We find that Herfindahl’s least-cost-first result for nonrenewable
resources does not necessarily apply to renewable resources, even when there is only one
demand. Along the optimal trajectory extraction may switch from single to simultaneous
use, depending on how the marginal opportunity cost of each resource evolves over time.
A numerical simulation for the South Oahu aquifer system, which allows for
differentiation of users by elevation and hence distribution costs, illustrates the switching
behavior.
Keywords: Renewable resources, dynamic optimization, multiple resources
JEL codes: Q25, Q28, C61
a
Professor, Department of Economics, University of Hawai‘i, Mānoa, and Environmental Director for the
University of Hawai‘i Economic Research Organization, Saunders 542, 2424 Maile Way, Honolulu, HI
96822 USA. E-mail: jimr@hawaii.edu
b
Ph.D. candidate, Department of Economics, University of Hawai‘i, Mānoa. E-mail: cawada@hawaii.edu
c
Revised version of WP 09-02, “Integrated Management of the South Oahu Aquifer System: A Spatial and
Temporal Approach.”

Renewable Resource Management with Alternative Sources: the Case of
Multiple Aquifers and a "Backstop" Resource
James Roumasset and Christopher Wada
1 Introduction
Efficient groundwater management is typically modeled as a single aquifer
serving a single group of consumers. In many cases, however, the water manager must
decide how fast to draw down or otherwise manage multiple aquifers. In Florida, for
example, five water management districts overly a large system of interconnected
aquifers (Florida DEP, 2009), while in California, over 500 distinct groundwater systems
underlie approximately 40% of the state’s surface area (California DWR, 2003). The
Oahu case is similar, albeit with not nearly as many aquifer systems to potentially draw
from. Previous studies of the Southern Oahu aquifer system, have estimated the optimal
intertemporal water allocation for the Pearl Harbor aquifer (Krulce, Roumasset, and
Wilson, 1997) and the optimal spatial and intertemporal allocation of water for the
Honolulu aquifer system (Pitafi and Roumasset, 2009), under the assumption that each
source serves separate water districts. In reality however, groundwater pumped from the
Pearl Harbor aquifer is currently being transferred
1
to the Honolulu consumption region,
suggesting that joint management is more appropriate.
The problem can be set up as one of a single demand supplied by multiple
sources. In the theory of nonrenewable resources, Herfindahl’s (Herfindahl, 1967) rule
1
Primary wells in both Honolulu and Pearl Harbor are connected directly to a common pipeline. Pressure
within the pipe is kept high enough that water flows at no additional pumping cost between regions toward
wherever groundwater is drawn out of the system.

3
states that when facing a single demand, a resource manager should extract deposits of a
resource in the order of unit extraction costs. When multiple demands exist, the least-
unit-extraction-cost-first rule is replaced by a least-price rule (Chakravorty and Krulce,
1994), according to which the optimal shadow price is given by the sum of extraction
cost, conversion cost, and the endogenous marginal user cost. Gaudet et al. (2001)
generalize the result to spatially differentiated resource sites and users. They find that in
the presence of setup costs, however, the least-price rule need not hold. If there is only
one demand, the extraction profile over multiple nonrenewable resources is again
determined exogenously, according to the sum of extraction and conversion costs for
each resource.
2
In what follows, we show that Herfindahl’s least-cost-first principle does not
extend to renewables, even in the presence of a single demand. A few recent studies
consider similar problems involving multiple renewable resources. Zeitouni and Dinar
(1997) construct a model with two adjacent sources of groundwater, but recharge is not
allowed to vary with the quantity of groundwater in stock, and they do not derive a
generalized rule to determine the optimal order of groundwater extraction. Horan and
Shortle (1999) develop a theoretical framework for the optimal management of multiple
Mink-Whale stocks but do not solve for the transitional dynamics. Costello and Polasky
(2008) construct a more general multiple renewable resource model, which incorporates
stochasticity in addition to space and time. They find that harvest closure is optimal
whenever the stock of a particular spatial patch falls below the patch-specific escapement
target for breeding. When that occurs, the expected biological returns from escapement
exceed the returns from current harvest.
2
See the discussions in Chakravorty et al. (2005) and Im et al. (2006).

4
However, their results are based on the assumption of state independent control,
i.e. they assume that marginal returns to harvest are independent of the amount harvested.
Although this may be a standard assumption in the fisheries literature, downward sloping
demand curves may be more common in other natural resource contexts. The current
study proposes an analytical model for the optimal joint management of two coastal
aquifers, where the natural net recharge of each aquifer varies with its respective head
level. Whereas most of the above-cited studies are analytical, we also provide a
numerical illustration of how to apply the model.
3
The rest of the paper is organized as follows. The next section extends the usual
single-aquifer economic-hydrologic optimization model to allow for multiple
groundwater resources. We derive a least-price rule for renewable resources analogous
to Chakravorty and Krulce’s rule for nonrenewables. We then contrast it with the
Herfindahl least-cost rule for the optimal extraction of nonrenewable resources. The
subsequent section utilizes data from the Honolulu and Pearl Harbor aquifers on the
island of Oahu, Hawaii to provide a numerical illustration of a two-aquifer problem. The
final section summarizes major analytical and empirical results and discusses general
conclusions and potential research extensions.
2 The model
Coastal groundwater aquifers are usually characterized by a “Ghyben-Herzberg”
lens (Mink, 1980) of freshwater sitting on an underlying layer of seawater. The upper
surface of the freshwater lens is buoyed up above sea level due to the difference in
density between the freshwater and the displaced saltwater (Figure 1). The head, or the
3
Zeitouni and Dinar (1997) numerically estimate a simplified version of their two-aquifer model.
However, they do not provide a description of the algorithm used for the estimation, and the aquifer net
recharge functions are state-independent.

5
distance between the top of the lens and mean sea level, is denoted h, and is one of
several measures of the freshwater stock in the aquifer. As the stock declines, i.e. the
lens contracts, the head level falls, and water extraction becomes more costly for several
reasons. Freshwater must be pumped longer distances, and that requires more energy. In
addition, when the lens contracts to the point where the lower surface reaches the bottom
of the deepest well, the remaining wells must take on larger shares of the pumping until
physical limitations on the rate of pumping or capacity restrictions necessitate the
construction of costly new wells. Thus, the extraction cost is a non-negative, decreasing,
convex
4
function of head:
0)(
≥
hc , 0)(
<
′
hc , and 0)(
≥
′
′
hc .
Figure 1: Coastal aquifer cross-section.
Adopted from Mink (1980)
4
This characteristic of the cost function ensures that the necessary conditions for the maximization problem
described below are also sufficient. In the application section, we assume a function that is linear in lift and
therefore linear in head, thus satisfying the convexity requirement.
Extraction
Well
Caprock
Seawater
Freshwater lens
Land
Head
Leakage

6
Leakage from a coastal aquifer is also a function of the head level. Low
permeability caprock bounds the freshwater lens along the coast,
5
but pressure from the
lens causes some freshwater to leak or discharge into the ocean as springflow and diffuse
seepage through the caprock. As the head level declines, leakage decreases both because
of the smaller surface area along the ocean boundary and because of the decrease in
pressure due to the shrinking of the lens. Thus leakage is a positive, increasing, convex
function of head: 0)(
≥
hl , 0)(
>
′
hl , and 0)(
≥
′
′
hl . Assuming that natural inflow from
precipitation and adjacent water bodies is fixed at some rate R, net recharge is defined as
)()( hlRhf
−
≡
. Since coastal groundwater is a renewable resource whose pre-extraction
growth rate depends on the stock of the resource, the following model may prove useful
for other renewables.
The cost of distributing water from the primary wells to consumers at spatially
heterogeneous locations varies, so the unit cost of transporting water to users in category
j is denoted
j
d
c. In particular, distribution costs are higher for users at higher elevations
because energy is required to boost the pumped water uphill. Consumption is also
differentiated across spatial categories, so
ij
t
q is the quantity extracted from aquifer i for
consumption in category j at time t, and
∑
j
ij
t
q
is the total quantity extracted for
category
j
in period
t
.
Because coastal aquifers are located near the ocean, access to seawater is
essentially unlimited. Desalination technology, although currently expensive, can
5
Caprock are coastal plain deposits (e.g. marine and terrestrial sediments, limestone, and reef deposits) that
impede discharge of groundwater to the sea. Although caprock borders the coastline of Southern Oahu and
is relevant to this study, it is not a general characteristic of coastal aquifers.

7
produce freshwater as an alternative to extraction. The unit cost of this backstop source
is assumed constant at
b
c
, and the quantity of desalinated water produced for
consumption in category
j
at time
t
is
j
t
b
.
The resource manager faces a non-autonomous optimal control problem with
bounded controls and state-space constraints. The problem is non-autonomous because
demand is allowed to grow over time. The control variables are restricted to be non-
negative, and the state-space is constrained by minimum allowable head levels for each
aquifer
i
h
min
.
6
Given a discount rate
0
>
r, the planner chooses the rates of extraction
and desalination over time to maximize the present value of net social benefits:
7
(1)
( )
∑ ∑ ∫∑
∞
=
+
−



















∑
+−+−
00
1
,
][])([),(
t j
bq
j
db
j
t
i
j
d
i
ti
ij
tj
t
bq
j
t
i
ij
t
j
t
ij
t
ccbchcqdxtxDMax
ρ
subject to
∑
−=−
+
j
ij
t
i
ti
i
t
i
ti
qhfhh )(][
1
γ
jihhbq
ii
t
j
t
ij
t
, ,0 ,0
min
∀≥≥≥
where
t
ρ
is the discount factor,
1
)1(
−
+≡
r
ρ
,
),(
1
txD
j
−
is the inverse demand function
for consumption category
j
, and
i
γ
is a height to volume conversion factor for aquifer
i
.
If the price for consumption category
j
is defined as ),(
1
tbqDp
j
t
i
ij
tj
j
t
+≡
∑
−
and
the marginal opportunity cost (MOC)
8
of a unit of water extracted from aquifer
i
as
6
Flow at the interface between the freshwater and underlying saltwater creates a thick transition zone
comprised of brackish water that varies in salinity. The EPA standard for salinity of potable water is 2% of
seawater salinity. As the aquifer is depleted, the transition zone eventually rises to the point where the well
bottoms reach the minimum allowable salinity. This quality consideration is incorporated into the model as
a head level constraint.
7
See Pitafi and Roumasset (2009) for a similar setup in the case of a single aquifer.
8
We use this terminology in honor of David Pearce (see e.g. Pearce et al., 1989).

8
i
ti
j
d
i
ti
ij
t
chc
1
1
)(
+
−
++≡
λργπ
, then the following can be derived from the necessary
conditions for the maximization problem (1):
9
(2)
ij
t
j
t
p
π
≤
, 0 then , if
=<
ij
t
q
j
db
j
t
ccp
+≤
, 0 then , if
=<
j
t
b
.
In other words, if the price for category
j
is less than the MOC of a unit of groundwater
extracted from aquifer
i
for consumption in
j
, then no water is extracted for that purpose.
Since it is never the case that zero water is consumed, it must be that
(
)
j
db
Ij
t
j
t
j
t
j
t
ccp
+=
,,...,,min
21
πππ
.
10
Consequently
mj
t
j
t
p
π
=
when aquifer
m
is being
used )0(
>
mj
t
q
and
njmj
t
j
t
p
ππ
==
,
nj
t
mj
t
j
t
p
ππ
∆=∆=∆
)0,0(
>>
nj
t
mj
t
qq
when aquifers
m
and
n
are being used simultaneously. This pricing and extraction rule accords with the
least-price rule for nonrenewables. Herfindahl’s least-cost rule for nonrenewables does
not extend to renewables, however, even in the case of a single demand.
Although the optimal order of resource extraction appears to be governed by a
fairly simple rule (2), solving for the MOC of each resource turns out to be a non-trivial
task. In the case of non-renewable resources with constant extraction costs, the entire
path of a given resource’s shadow price is determined once the initial value is specified.
When multiple renewable resources with stock-dependent extraction costs and growth are
considered, however, each feasible MOC path must be solved for in conjunction with the
associated feasible path of the aquifer head level. Forward-looking shooting algorithms
encounter difficulties of iterating on multiple starting values of MOC. In the presence of
9
The discrete-time Hamiltonian and its necessary conditions are in Appendix A.
10
Without loss of generality, suppose that
nj
t
mj
t
ππ
<
for
n
m
<
. Then let the price be determined by
kj
t
π
, where
1
>
k
, i.e.
ij
t
kj
t
j
t
p
ππ
>=
,
1,,1
−
=
ki
K
. But that violates one of the necessary
conditions, which states that
ij
t
j
t
p
π
≤
for all i.

9
multiple non-linearities, gradient ascent algorithms take many iterations to converge or
may not even converge at all. Because of these difficulties, it may be necessary to
condition the search algorithm on different possible orders of extraction.
Head levels change over time as a result of extraction and/or natural recharge.
Accordingly, the MOCs also change inasmuch as the extraction costs and shadow prices
are dependent on the head levels. Thus it may be optimal to extract exclusively from a
single aquifer for a finite period of time and then switch to pumping from more than one
source in the periods that follows. A planner with many aquifers at her disposal will face
multiple endogenously determined switching points as described by the optimal pricing
condition (2). In what follows, we explore the two-aquifer case, which is also the subject
of our subsequent numerical illustration. In the case of even two aquifers, many scenarios
are conceivable, but we focus on three of interest. Without loss of generality, aquifer A
is chosen as the aquifer to optimally extract from first in each scenario.
11
The actual
extraction pattern calculated in the numerical illustration is described by scenario 1.
Stage 1 Stage 2 Stage 3 Stage 4
Scenario 1 Extract from A
Extract only MSY from A
Extract from B
Steady State
Scenario 2 Extract from A
Extract from A
Extract from B
Extract only MSY from A
Extract from B
Steady
State
Scenario 3 Extract from A
Extract from A
Extract from B
Extract from A
Extract only MSY from B
Steady
State
Figure 2: Order of extraction in the case of two aquifers
12
11
We later establish a sufficient condition that determines which aquifer is optimally used first. Our
application illustrates the case wherein the condition is satisfied and indicates the larger aquifer for early
withdrawal.
12
We use “extract” to mean any removal of groundwater from an aquifer, even when the extraction rate
does not exceed net recharge, i.e. an aquifer’s head level may be rising while extraction is occurring.

10
In the initial stage of scenario 1, aquifer A is used exclusively, while aquifer B is
built up.
13
As the head level of aquifer B rises, the marginal extraction cost declines but
the shadow price rises, so the MOC may increase or decrease.
14
At the same time, the
MOC of aquifer A increases until the first switch-point occurs, whereupon aquifer A
reaches its minimum allowable head constraint.
In stage 2, extraction from aquifer A is limited to net natural recharge, and the
remaining optimal consumption is supplied by aquifer B. Price, now determined by the
MOC of aquifer B, continues to increase, and the shadow price of groundwater for
aquifer A rises apace. Although the extraction cost for aquifer A remains constant once
MSY extraction is imposed, the MUC still rises as the multiplier on the head constraint
becomes positive and rises (equation A.4).
The final stage of extraction is characterized by MSY extraction from both
aquifers. A steady state is maintained,
15
in which extraction from each aquifer is limited
to net recharge, and the price remains at the backstop cost. The stages of extraction for
scenario 1 are depicted graphically in Figure 3 below.
13
By sheer coincidence
)(
2
0
1
0
jj
ππ
=
it would be optimal to simultaneously draw down both aquifers
from the start.
14
Recall that
i
t
i
ti
ij
t
hc
1
)(
+
+=
ρλπ
. Since
0)(
<
′
i
ti
hc
by assumption, the first term on the left hand side
is decreasing as the head is building. Condition (A.4) in Appendix A reduces to
)(
1
1
1
i
ti
i
ti
i
t
i
t
hf
′
−=−
+
−
+
λργλρλ
when extraction is zero. Rearranging yields
0/)(/][
1
1
1
>
′
−=−
+
−
+
i
t
i
ti
i
ti
i
t
i
t
i
t
hfr
λλγλλλ
, i.e. the marginal user cost of aquifer i is always
increasing when extraction from aquifer i is zero. Hence, MOC may be increasing or decreasing.
15
Technically the system never reaches a steady state since demand is growing. However, it is meant that
the price and head levels remain constant in the “steady state.”

11
Figure 3: Hypothetical efficiency price path.
In the scenario depicted, aquifer A
reaches its head constraint at the first switch point.
16
The efficiency price is the lower
envelope of the MOC paths.
In scenario 2, stage 1 remains the same; aquifer A is used exclusively while
aquifer B is built up. At the first switch point, however, the MOCs of the two aquifers
become equal. Since neither aquifer is at its constraint, stage 2 is characterized by
simultaneous extraction of both aquifers, the rates of which are optimally chosen to keep
the MOCs equal. Eventually, aquifer A reaches its head constraint, and extraction is
limited to natural recharge. In the final stage, aquifer B is drawn down to its minimum
allowable head level, and the system reaches a steady state, in which both aquifers are
pumped at MSY. Scenario 3 is exactly the same as scenario 2 except that aquifer B
reaches its head constraint before aquifer A does.
16
The dotted portion of the MOC A curve represents what the MOC would be if the head constraint did not
bind at T
1
.
T
2
Time
T
1
MOC B Efficiency Price
MOC A
Backstop Cost

12
Generally for autonomous, single renewable resource problems, optimality calls
for monotonic state paths (Kamien and Schwartz, 1991). If the stock starts below its
steady state level, then it optimally builds monotonically, whereas if the stock starts
above its steady state level, then it is optimally drawn down monotonically. With
multiple resources, whether the problem is autonomous or not, optimality is not always
characterized by monotonic state paths. The result can be attributed to the resource-
specific extraction cost and growth functions, i.e. inter-resource comparisons must be
made.
It is not necessarily the case that the exclusively used aquifer is drawn all the way
down to its MSY level at the end of the first stage. We can verify that such an extraction
path is optimal, however, by checking that the following sufficient condition holds:
17
r
hfhc
r
hfhcp
hc
r
hfhc
r
hfhcp
hc
t
t
)()()()]([
)(
)()()()]([
)(
2
max2
2
max2
1
2
1
2
2
max2
2
max2
2
max2
1
min1
1
min1
1
1
1
1
1
min1
1
min1
1
min1
′
−
′
−
+
<
′
−
′
−
+
−−
−
−
γγ
γγ
for all ])},(),([min{
2
max2
1
min1 bt
chchcp ∈. In other words, if for all feasible head
combinations and prices, the extraction cost plus marginal user cost of aquifer 1 is less
than that of aquifer 2, then it is always optimal to draw aquifer 1 down to its MSY level
in the initial stage. In the illustration that follows, the sufficiency condition is satisfied,
thus ensuring that Pearl Harbor aquifer (PHA), the larger of the two, is optimally drawn
down first, and no water is extracted from the Honolulu aquifer (HNA) until PHA
reaches its minimum head level, where maximum sustainable yield is achieved.
17
See Appendix B for a proof of this result.

13
3 Application: Honolulu and Pearl Harbor aquifers
3.1 Functional forms and parameters
As previously discussed, the volume of water stored in a coastal aquifer is a
function of head, but it also depends on various hydrologic parameters such as the aquifer
boundaries, lens geometry, and rock porosity (Mink, 1980). Although the surfaces of the
lens are technically parabolic, the hydraulic gradient in southern Oahu is small enough
that storage and head are approximately linearly related. Following Krulce, Roumasset
and Wilson (1997), we suppose that 78.149 billion gallons of freshwater are stored per
foot of head in PHA. The net recharge function for PHA is constructed using Mink’s
recharge estimate of 220 million gallons per day (mgd) in combination with the leakage
function econometrically estimated by Krulce, Roumasset and Wilson (1997):
ttt
hhhl 022023.024972.0)(
2
+= . In a similar manner, we use Pitafi and Roumasset’s
(2009) volume-head conversion factor of 61 billion gallons of water per foot of head for
HNA and construct a net recharge function using Liu’s (2006) 64 mgd estimate of natural
inflow.
The extraction cost is specified as a convex (linear) function of lift:
)()(
i
tii
i
ti
hehc −=
ξ
, where lift is defined as the difference between the average ground
surface elevation of the wells,
i
e, and the head level. To simplify the discussion, we will
focus primarily on functional forms in the text and refer the reader to Table 1 for all of
the aquifer-specific parameter values. The energy-cost parameter,
i
ξ
, is calculated using
the initial unit extraction cost )(
0
i
i
hc , which is a volume-weighted average of unit
extraction costs for all primary wells in the initial period, and the initial head level,
i
h
0
.

14
Distribution costs
j
d
c are calculated for each elevation category j from booster station
pumping data (Table 2). The unit cost
b
c of desalinating water is estimated at $7.43/tg.
18
Parameter Honolulu
Pearl Harbor
i
e [ft] 50
272
i
ξ
[$/(tg*ft)] 0.00786
0.00121
)(
0
i
i
hc [$/tg] 0.22
0.31
i
h
0
[ft] 21.5
16
g 0.01
0.01
η
0.3
0.3
r 0.03
0.03
i
h
min
[ft]
19
21
15.125
Table 1: Parameter values for the Honolulu and Pearl Harbor aquifers
The demand for water is modeled as a constant elasticity function
η
α
−
=)(),(
j
t
gt
jtj
xetxD
. The coefficient
j
α
for each elevation category is calculated
using actual pumping data and the retail price for the year 2006 (Table 2). Following
Pitafi and Roumasset (2009), the exogenous rate of population growth g is assumed to be
1% in the baseline scenario. Demand elasticities vary considerably among studies, but
recent estimates for increasing block price structures (Olmstead, Hanemann, and Stavins,
2007) tend to be high relative to the elasticities used in previous studies of the southern
18
See Appendix C.
19
Taking into account average well depth below mean sea level, upconing, and the thickness of the
brackish transition zone, Liu (2006, 2007) estimates the minimum allowable head level required to avoid
seawater intrusion of the wells.

15
Oahu aquifer system. In the current study, the baseline value is taken as
η
=0.3. Finally
the interest rate r=3%.
20
Category Elevation (ft)
Distribution Cost ($/tg)
Qty (mgd)
Coefficient (
α
j
)
1 0
$1.81
70.34
89.48
2 500
$2.35
6.21
7.9
3 789
$3.21
1.17
1.49
4 1039
$4.37
0.65
0.83
5 1086
$5.62
0.17
0.21
6 1345
$6.90
0.12
0.15
7 0
$1.86
47.43
60.34
8 552
$2.37
3.7
4.7
9 887
$2.95
1.18
1.51
Table 2: Demand coefficients and distribution costs.
Categories 1-6 represent the
Honolulu consumption district and 7-9 the Pearl Harbor consumption district.
3.2 Computational strategy
The problem is solved using a forward-iterating algorithm. The initial and
terminal conditions for the head level are known. Growing demand ensures the
implementation of the backstop at some point since the size of the aquifer system is
finite, and steady state calculations reveal that the head level constraints are binding.
Thus the terminal price is the backstop cost, and the optimal path will be determined once
we know the correct initial shadow prices.
21
As described in section 2, the order of extraction will depend on which of the
resources is “cheaper.” In our study, the aquifers satisfy the sufficiency condition such
that Pearl Harbor aquifer is used first and is drawn down exclusively to its MSY head
20
The fact that the discount rate is greater than the population growth rate ensures convergence of the
objective functional.
21
Along the optimal trajectory, the price rises to the backstop cost and the head constraint becomes binding
at precisely the same time. See Appendix D for a proof of this result.

16
level, while Honolulu aquifer is allowed to build.
22
Once Pearl Harbor aquifer reaches its
MSY level, extraction is limited to recharge and Honolulu aquifer is drawn down to its
MSY level as the resources approach a steady state. Given the order and stages of
extraction, we proceed to solve for the endogenously determined switch-points and the
time at which the aquifers reach a steady state.
Following a "shooting" method,
23
trial values are assumed for the initial shadow
prices, and condition (A.5) allows one to solve for the shadow prices in the following
period. Once the period 2 shadow prices are determined, the price (A.2) and therefore the
rates of extraction can be ascertained for the current period. The rates of extraction
reveal the head levels in the next period via the equation of motion (A.6), and the whole
process can be repeated, using the period 2 head levels and shadow prices as the new
starting point. Eventually, one of the terminal conditions is reached. If at least one of the
other terminal conditions is inconsistent, then the initial guesses for the shadow prices are
revealed as incorrect. The guesses must be adjusted and the process repeated until all of
the initial and terminal conditions are satisfied for the head of each aquifer and the price,
so that the PV functional is maximized given the boundary conditions.
3.3 Results
We determine the optimal paths of price, extraction, and head for each of the
aquifers. The efficiency price for consumption category 1 starts at $2.15 and rises
relatively slowly until year 40, at which point PHA reaches its head level constraint.
After year 40, the price path is determined by the MOC of HNA, and the price begins to
22
If the sufficiency condition is not met, then the algorithm described will need to be adjusted, inasmuch as
there will likely be a stage of simultaneous extraction before either aquifer reaches its MSY level.
23
See e.g. Judd (1998).

17
rise more rapidly as HNA is depleted. During the second stage, any quantity demanded
at the optimal price in excess of maximum sustainable yield (MSY) for PHA is supplied
by extraction from HNA. Eventually, the price rises to the cost of the backstop as HNA
approaches its minimum allowable head level. The steady state is reached after 100 years,
and optimal consumption in excess of MSY for both aquifers from that point forward is
met by desalination. For a graphical example, see Figures 4 and 5 below.
Figure 5 depicts the MOC curves for each aquifer. Since the extraction cost
functions are linear and relatively flat in this application, the order of extraction is
determined primarily by differences in the net recharge functions. Intuitively, PHA is
drawn down first exclusively because the resulting decrease in leakage (increase in net
recharge) is greater than the increase in leakage (decrease in net recharge) that occurs as
HNA builds. It should also be noted that the PHA is building for an initial period even
when it is being used exclusively (Figure 4). Because the problem is non-autonomous,
non-monotonicity can result from the need to build the stock in anticipation of future
scarcity.

18
0 10 20 30 40 50 60 Years
5
10
15
20
25
30
Feet
Honolulu Head
10 20 30 40 50 60 Years
5
10
15
20
Feet
Pearl Harbor Head
Figure 4: Head paths (g=0.02, 90%R)
24
24
A scenario other than the baseline is chosen for illustrative purposes inasmuch as the MOCs are more
clearly differentiated.

19
Honolulu
Pearl Harbor
10 20 30 40 50 Years
2
4
6
8
tg
ê
day
Total Marginal Cost Paths
Highest Elevation
Lowest Elevation
10 20 30 40 50 60 Years
2
4
6
8
10
12
14
$
ê
tg
Efficiency Price Paths
Figure 5: MOC and efficiency price paths (g=0.02, 90%R)

20
3.4 Sensitivity analysis
Sensitivity analysis is performed to not only test the sensitivity of the results to
uncertain parameter values, but also to consider the possible impacts of climate change
and watershed degradation, caused by, for example, invasive species.
The three parameters with the most uncertainty in this study are the growth rate of
demand, the elasticity of demand for water, and the natural rate of infiltration to the
aquifer. More accurate measurement of the first two parameters is feasible but beyond
the scope of this study. The gains to such an endeavor are likely to be large, however.
Increasing the 1% demand growth per year assumed in the baseline scenario to 2% results
in the implementation of desalination 42 years closer to the present. Clearly, optimizing
with a wrongly assumed growth value can lead to large welfare losses; underestimating
growth would result in underpricing, and hence necessitate the implementation of a
backstop even sooner.
Scenario 1
τ
(yrs)
T
(yrs)
PV (million $)
PV lost recharge (million $)
Baseline 40
100
$12,047.80
-
%2
=
g
29
58
$15,912.90
-
90% R 24
83
$11,615.70
$432.10
80% R 12
63
$11,492.70
$555.10
70% R 8
46
$10,942.70
$1,105.10
%2
=
g
& 90% R 20
49
$16,713.70
$686.90
5.0
=
η
40
112
$10,457.80
-
Table 3: Sensitivity analysis
The baseline value for demand elasticity is based on existing studies of water
demand. Recent work however, (see e.g. Olmstead et al, 2007) suggests that elasticity
under increasing block pricing structures may be larger than previous estimates suggest.

21
Increasing the demand elasticity to -0.5 delays the need for desalination by 12 years.
This follows from the fact that the price need be increased by less to achieve any given
desired level of demand-side conservation. Small changes in elasticity have a large
impact on the optimal paths.
The third parameter of interest, natural infiltration to the aquifer or natural
recharge, can be broken down into several components: precipitation, evapotranspiration,
and runoff. In most economic studies of groundwater, natural recharge is taken as
constant for the entire planning horizon. This may have been a good approximation for
the past century, but with climate change looming on the horizon and threats of invasive
species on the rise, historic measurements of recharge are likely no longer accurate
approximations of future recharge. Ideally, a hydrologic-economic model would
incorporate factors that influence recharge such as climactic conditions and the state of
the watershed, inasmuch as global warming will certainly have an impact on recharge via
changes in quantities and patterns of precipitation, changes in evapotranspiration, and
changes in runoff; and increases in invasive species will affect runoff and
evapotranspiration as a result of their impact on landcover. However, regional climate
change and watershed models are still in a developmental stage, so incorporating the
science into an economic model is currently not feasible. Instead, we look at various
recharge scenarios that may result from the complicated physical interactions associated
with climate change or degradation of the watershed.
The numerical simulations are re-run assuming a 10%, 20%, and 30% reduction
in recharge. The reduction could be due to a decline in precipitation, increased runoff
due to more extreme rainfall events or damage to the watershed from invasive species, or

22
any combination of those factors. It is intuitive that lower recharge results in the need for
desalination sooner (17 years in the 10% reduction case); the growth of the aquifer is
slower so the resource is depleted more quickly. The $432.1 million present value of lost
recharge is calculated by taking the difference between the PV of the baseline scenario
and the 90%-recharge scenario. Although in reality the recharge would be declining
gradually over time, the 10% reduction is an approximation and the PV cost of doing
nothing is not trivial. The lost PV is even larger with higher reductions in recharge.
Preventing climate change on a local scale is not feasible, but investment in natural
capital within the watershed will help to mitigate the impact on recharge.
4 Conclusion
Although many real-world natural resource problems involve the use of multiple
resources, most renewable resource economic models focus on a single resource
supplying a single demand. When more than one source is available, optimal
management involves how much to extract from each source. We extend the usual
groundwater economics model to include multiple sources and find that at a given point
in time, optimality may call for extraction from a single aquifer exclusively, extraction
from both sources, or extraction from both aquifers and use of a backstop simultaneously.
Whereas Chakravorty and Krulce’s (1994)
least-price-first
rule for non-
renewables collapses to the Herfindahl least-cost-first rule for a single demand, optimal
extraction from multiple renewable resources such as aquifers is governed by the more
general least-price-first rule even when there is only one demand. Also contrary to the
nonrenewable case (e.g. Gaudet et al., 2001), a single demand can optimally be supplied
by more than one renewable resource in any given period. In the steady state, the

23
renewable characteristic of renewable resources ensures that all resources are used
simultaneously. Stages of simultaneous use prior to the steady state are also feasible.
While optimal extraction accords with
least-price-first,
those “prices” (which
refer to marginal opportunity costs along the optimal path) are endogenous. Which
resource has the lower initial marginal opportunity costs is not generally determinable
without actually solving the problem at hand. There are two countervailing forces;
extraction decreases leakage by lowering the head level, but at the same time it increases
extractions costs. Intuitively, if extraction costs are approximately constant for example,
optimality requires initially extracting exclusively from the aquifer for which the value of
net gained recharge (including movement along the leakage function) exceeds the value
of lost recharge from the unused aquifer. We find a sufficiency condition that not only
determines the order of extraction but also ensures complete drawdown of the first
resource to its MSY level before switching to the second resource.
In an application to the Southern Oahu aquifer system, we find that extraction
occurs in three stages. In the first, Pearl Harbor aquifer is used exclusively
25
while
Honolulu aquifer is built up. In this particular case, the order of extraction is determined
primarily by differences in the net recharge functions inasmuch as extraction costs are
linear and fairly flat. Intuitively, Pearl Harbor aquifer is drawn down first because the
resulting decrease in leakage (increase in net recharge) is greater than the increase in
leakage (decrease in net recharge) that occurs as Honolulu aquifer builds. Stage 2 begins
once Pearl Harbor reaches its minimum allowable head level. Extraction from Pearl
Harbor is limited to MSY and the remaining quantity demanded at the optimal price is
25
For an initial period, Pearl Harbor aquifer’s stock is actually increasing even when it is being used
exclusively. That both state variable approach paths are non-monotonic is not a general result. For
example, with a larger initial stock, Pearl Harbor would be optimally drawn down monotonically.

24
met by extraction from Honolulu. During stage 3, price rises to the backstop cost and
Honolulu reaches its head constraint. Beyond stage 3, extraction is limited to recharge
for each aquifer and the remaining consumption is met by desalination.
There are many possible research extensions to the current work, but a few in
particular stand out. First, the multiple-aquifer hydrologic-economic framework should
integrate watershed and regional climate models. As scientists gain a better
understanding of the physical models, this type of integration will become a more
feasible undertaking. Second while the simulation results suggest that there exist large
potential welfare gains resulting from optimal management of multiple resources, the
analysis does not indicate how such pricing could be achieved. Most water utilities are
required to keep a balanced budget, and optimal pricing generates excess revenue since
the price exceeds physical marginal costs. One possible solution is an increasing block
pricing structure with a free first block (see e.g. Pitafi and Roumasset, 2009). Third,
there may be cases in which multiple aquifers may be available to serve multiple
consumption districts with non-zero transportation costs between districts. In that case,
independent management may be optimal until scarcity increases enough to justify costly
transportation of water between districts. The endogenous boundary between locales
would adjust over time along the optimal trajectory. Finally, many natural resource
stocks are directly linked and adjacent aquifers are no exception. In the case of Southern
Oahu, subsurface inter-aquifer flow is relatively small so it is ignored in the analysis, but
that may be an exception rather than a generality. In the general case, the equations of
motion for the aquifer stock should be modified to include the state variable for each
interconnected aquifer.

25
Although the numerical simulation addresses a specific coastal aquifer problem,
the analytical framework is sufficiently general to have many other potential applications.
For example, a regional fish market likely draws from many fisheries. Thus a single
demand for fish channeled through that market is served by multiple fish stocks.
Similarly, an integrated lumber market would draw trees from multiple locales.
Appendix A
The corresponding discrete-time current value Hamiltonian
26
is:
(A.1)
( )
∑ ∑
∫










+−+−
∑
=
+
−
j
j
db
j
t
i
j
d
i
ti
ij
t
bq
j
ccbchcqdxtxDH
j
t
i
ij
t
][])([),(
0
1
( )
∑∑ ∑
−+







−+
+
−
i
ii
t
i
t
i j
ij
t
i
ti
i
ti
hhqhf ][])([
min1
1
µλγρ
and the Maximum Principle requires that :,
ji
∀
(A.2) ,0)(),(
1
11
≤−−−+=
∂
∂
+
−−
∑
i
ti
j
d
i
ti
i
j
t
ij
tj
ij
t
chctbqD
q
H
λργ
0≥
ij
t
q
, 0=
∂
∂
ij
t
ij
t
q
H
q
(A.3) ,0),(
1
≤−−+=
∂
∂
∑
−j
db
i
j
t
ij
tj
j
t
cctbqD
b
H
0≥
j
t
b
, 0=
∂
∂
j
t
j
t
b
H
b
(A.4)
i
t
i
ti
i
ti
i
ti
j
ij
t
i
t
i
t
i
t
hfhcq
h
H
µλργλρλ
−
′
−
′
=
∂
∂
−=−
+
−
+
∑
)()(
1
1
1
(A.5) 0
min
≥−=
∂
∂
ii
t
i
t
hh
H
µ
0≥
i
t
µ
, 0=
∂
∂
i
t
i
t
H
µ
µ
(A.6) ∑
−=
∂
∂
=−
+
+
j
ij
t
i
ti
i
t
i
t
i
ti
qhf
H
hh )(
)(
][
1
1
ρλ
γ
.
26
See, e.g. Conrad and Clark (1987).

26
Appendix B
Proposition:
When
r
hfhc
r
hfhcp
hc
r
hfhc
r
hfhcp
hc
t
t
)()()()]([
)(
)()()()]([
)(
2
max2
2
max2
1
2
1
2
2
max2
2
max2
2
max2
1
min1
1
min1
1
1
1
1
1
min1
1
min1
1
min1
′
−
′
−
+
<
′
−
′
−
+
−−
−
−
γγ
γγ
for all ])},(),([min{
2
max2
1
min1 bt
chchcp
∈, it is optimal to draw aquifer 1 down to its MSY
level before extracting from aquifer 2.
Proof:
Since we do not know the price and head paths before solving the entire
optimization problem, the proposition considers the most extreme head values, and hence
is a sufficient but not a necessary condition.
Since the terms on the left and right hand side of the inequality above are the
marginal opportunity costs for aquifer 1 and 2 respectively, we will show first that the left
hand side is the largest feasible value. Since
0
1
>c
and
0
1
<
′
c
,
1
min
h
results in the largest
possible value for the first term. By definition,
0
1
<
′
f
and
0
1
<
′
′
f
, so
1
min
h
ensures that
the second term is at its highest feasible value. Finally, since
1
c
′
is constant inasmuch as
0
1
=
′
′
c
,
1
min
h
yields the largest possible value for
1
f
and hence the largest possible value
for the third term. By an analogous argument, the RHS yields the lowest feasible value
for the marginal opportunity cost of aquifer 2. If the smallest feasible marginal
opportunity cost for aquifer 2 still exceeds the largest feasible marginal opportunity cost
for aquifer 1, then by the
least-price-first
rule, aquifer 1 should be drawn down
exclusively to its MSY level in the first stage of extraction.
□

27
Appendix C
The desalination cost estimate is based on reverse osmosis (RO) membrane technology.
The following simple amortization formula is used:
1)1(
)1(
−+
+⋅⋅
=
t
i
t
iiP
K
C
where
C
K
is the amortized cost,
P
is the original start up investment,
i
is the bond rate,
and
t
is the expected plant life. Adding the amortized start up cost to the operation and
maintenance costs yields the total annual cost of desalination. Dividing that cost by total
output per year then gives the unit cost of desalination. Values for plant capacity (5
mgd), set up cost ($40 million) and operation and maintenance costs ($5.75 million) are
obtained from a study by Oceanit (2003) and adjusted for inflation and increases in
energy costs. Amortization rates and estimated average plant life varies among
desalination studies, but for the current study
i
=7.5% and
t
=25 years. Finally, it is
assumed that buildings and equipment have no salvage value and capital replacement
costs are already incorporated in operation and maintenance costs. Thus the unit cost of
desalination is $7.43/tg.
The unit cost of desalination is an important factor in the optimal allocation of
groundwater. The current values of parameters that determine the desalination cost
estimate may change over time or may need adjustment. Sensitivity analysis can be
performed to address this issue.

28
Appendix D
Proposition:
If
bT
cp
=
1
and
min
2
hh
T
= in the optimal steady state, it must be
21
TT =
.
Proof:
Suppose
bT
cp =
1
,
min
2
hh
T
=, and
21
TT <
. Then for
),(
21
TTt ∈
,
bt
cp
=,
min
hh
t
>, and the head level must be drawn down further to satisfy its terminal condition.
However, drawing the head down would raise the marginal opportunity cost of
groundwater above
b
c
. If this were true, then the backstop would optimally be used to
satisfy all of the quantity demanded at the optimal price over the interval. But if
extraction is zero, then the head level rises and
min
hh
T
=
is never achieved, which is a
contradiction.
Suppose instead that
min
1
hh
T=, bT
cp
=
2
, and
21
TT <
. Then for
),(
21
TTt ∈
,
bt
cp
< and
min
hh
t=. Once the head constraint is binding at
1
T
, extraction is limited to
recharge and the price must be equal to b
c
, since desalination must be used to satisfy any
consumption in excess of natural recharge. But bT
cp
<
1
, which is a contradiction. Thus,
if bT
cp
=
1
and
min
2
hh
T=, it must be that
21
TT =
.
□
References
California Department of Water Resources (2003): “California’s Groundwater - Bulletin
118, Update 2003.” Available at:
http://www.groundwater.water.ca.gov/bulletin118/update2003/index.cfm.
Accessed April 30, 2009.
Chakravorty, Ujjayant, and Darrell L. Krulce (1994): “Heterogeneous Demand and Order

29
of Resource Extraction,”
Econometrica
, 62, 1445-1452.
Chakravorty, Ujjayant, Roumasset, James, and Kinping Tse (1997): “Endogenous
Substitution among Energy Resources and Global Warming,”
The Journal of
Political Economy
, 105, 1201-1234.
Chakravorty, U., Krulce, D. and J. Roumasset (2005): “Specialization and non-renewable
resources: Ricardo meets Ricardo,”
Journal of Economic Dynamics and Control
,
29: 1517-1545.
Chiang, Alpha C. (1992):
Elements of Dynamic Optimization
. Long Grove, Illinois:
Waveland Press.
Conrad, Jon M. and Colin W. Clark (1987):
Natural resource economics: Notes and
problems
. New York: Cambridge University Press.
Costello, Christopher and Stephen Polasky (2008): “Optimal harvesting of stochastic
spatial resources,”
Journal of Environmental Economics and Management
, 58, 1-
18.
Dalhuisen, Jasper M., Florax, Raymond J.G.M., de Groot, Henri L.F., and Peter Nijkamp
(2003): “Price and Income Elasticities of Residential Water Demand: A Meta-
Analysis,”
Land Economics
, 79, 292-308.
Florida Department of Environmental Protection. Water Management Districts
Information. http://www.dep.state.fl.us/water/waterpolicy/districts.htm.
Accessed April 30, 2009.
Herfindahl, O.C. (1967): “Depletion and economic theory.” In: Gaffney, M. (ed.),
Extractive Resources and Taxation. University of Wisconsin Press, pp. 63-90.
Horan, Richard D. and James S. Shortle (1999): “Optimal Management of Multiple

30
Renewable Resource Stocks: An Application to Mink Whales,”
Environmental
and Resource Economics
, 13, 435-458.
Hotelling, Harold (1931): “The Economics of Exhaustible Resources,”
The Journal of
Political Economy
, 39, 137-175.
Im, Eric I., Chakravorty, Ujjayant, and James Roumasset (2006): “Discontinuous
extraction of a nonrenewable resource,”
Economic Letters
, 90: 6-11.
Judd, Kenneth (1998):
Numerical Methods in Economics.
1
st
ed. Cambridge,
Massachusetts: MIT Press.
Kamien, Morton I., and Nancy L. Schwartz (1991):
Dynamic optimization: the calculus
of variations and optimal control in economics and management
. 2
nd
ed. New
York: North-Holland.
Krulce, Darrell L., Roumasset, James A., and Tom Wilson (1997): “Optimal management
of a Renewable and Replaceable Resource: The Case of Coastal Groundwater,”
American Journal of Agricultural Economics
, 79, 1218-1228.
Lau, L. Stephen, and John F. Mink (2006):
Hydrology of the Hawaiian Islands
.
Honolulu: University of Hawai‘i Press.
Liu, Clark C.K. (2006):
Analytical Groundwater Flow and Transport Modeling for the
Estimation of the Sustainable Yield of Pearl Harbor Aquifer
. Project Report PR-
2006-06. Honolulu: Water Resources Research Center.
Liu, Clark C.K. (2007):
RAM2 Modeling and the Determination of Sustainable Yields of
Hawaii Basal Aquifers
. Project Report PR-2008-06. Honolulu: Water Resources
Research Center.
Mink, John F. (1980):
State of the Groundwater Resources of Southern Oahu
. Honolulu:

31
Honolulu Board of Water Supply.
Oceanit (2003): Draft Environmental Impact Statement for the Proposed Kalaeloa
Desalination Facility. Prepared for: Honolulu Board of Water Supply. Available
at: http://www.e-monitoring.com/bws/publications/RelevantPublication.htm.
Accessed March 18, 2008.
Oki, Delwyn S. (2005): Numerical Simulation of the Effects of Low-Permeability
Valley-Fill barriers and the Redistribution of Ground-Water Withdrawals in the
Pearl Harbor Area, Oahu, Hawaii. Scientific Investigations Report 2005-5253,
U.S. Geological Survey.
Olmstead, Sheila M., Hanemann, W. Michael, and Robert N. Stavins (2007): “Water
Demand Under Alternative Price Structures,”
Journal of Environmental
Economics and Management
, 54, 181-198.
Pearce, David, Markandya, Anil and Edward Barbier (1989):
Blueprint for a Green
Economy
. London: Earthscan Publications Ltd.
Pitafi, Basharat A. and James A. Roumasset. 2009. “Pareto-Improving Water
Management Over Space and Time: The Honolulu Case,”
American Journal of
Agricultural Economics
, 91: 138-153.
US Department of Labor. 2008. Bureau of Labor Statistics Data. Available at:
http://www.bls.gov/data. Accessed March 17, 2008.
US Energy Information Administration. 2008. Petroleum Summary Data and Analysis.
Available at: http://tonto.eia.doe.gov/dnav/pet/hist/rwtcA.htm. Accessed March
14, 2008.
Younos, T. 2005. The Economics of Desalination.
Journal of Contemporary Water

32
Research & Education
132: 39-45.
Zeitouni, Naomi and Ariel Dinar (1997): “Mitigating Negative Water Quality and Quality
Externalities by Joint Management of Adjacent Aquifers,”
Environmental and
Resource Economics
, 9, 1-20.
Zhou, Y. and Tol, R.S.J. 2005. Evaluating the Costs of Desalination and Water Transport.
Water Resources Research
41(3): W03003.