Forgot your password?
Document Actions
Valuing Ecosystem Services as Productive Inputs
Discusses latest advances in valuing ecosystem services that support or protect economic activity, using the example of coastal wetlands. The two services discused in detail are the role of coastal wetlands as breeding grounds and fry nurseries for marine fisheries and protecting coastlines and economic activity from storm damages. The paper illustrates the methodology for mangroves in Thailand from 1979-2004, estimating the effects of mangrove deforestation in terms of loss of fishery values and increased expected storm damages.
tem
s
sy
o
ec
ing
alu es
V
vic
SUMMARY
xxx Valuing ecosystem services asCES,B.0266-4658
VALUING ECOSYSTEMMSH,ECOP
© CEPR, productive Article
BlackwellOriginal Policy
EDWARD SERVICES
Economic inputs
Publishing UK
Oxford, Ltd
BARBIER
2007.
r
se
This paper explores two methods for valuing ecosystems by valuing the services
that they yield to various categories of user and that are not directly valued in the
market, and illustrates the usefulness of these methods with an application to the
valuation of mangrove ecosystems in Thailand. The first method is known as the
production function approach and relies on the fact that ecosystems may be inputs
into the production of other goods or services that are themselves marketed, such as
fisheries. I discuss issues that arise in measuring the input into fisheries, particularly
those due to the fact that the fishery stock is changing over time, and the shadow
value of the ecosystem consists in its contribution to the maintenance of the stock
as well as its contribution to current output. The second method is known as the
expected damage approach and is used to value the services of storm protection in
terms of the reduction in expected future storm damage that the ecosystem can
provide. These two methods are shown to yield very different valuations of ecosystems
from those that would be derived by the methods typically used in cost-benefit
analyses. I argue that they represent a significant improvement on current practice.
— Edward B. Barbier
Economic Policy January 2007 Printed in Great Britain
© CEPR, CES, MSH, 2007.
VALUING ECOSYSTEM SERVICES 179
Valuing ecosystem services as
productive inputs
Edward B. Barbier
University of Wyoming
1. INTRODUCTION
Global concern over the disappearance of natural ecosystems and habitats has
prompted policymakers to consider the ‘value of ecosystem services’ in environmental
management decisions. These ‘services’ are broadly defined as ‘the benefits people
obtain from ecosystems’ (Millennium Ecosystem Assessment, 2003, p. 53).
However, our current understanding of key ecological and economic relationships
is sufficient to value only a handful of ecological services. An important objective of
this paper is to explain and illustrate through numerical examples the difficulties
faced in valuing natural ecosystems and their services, compared to ordinary economic
or financial assets. Specifically, the paper addresses the following three questions:
1. What progress has been made in valuing ecological services for policy analysis?
2. What are the unique measurement issues that need to be overcome?
3. How can future progress improve upon the shortcomings in existing methods?
I am grateful to David Aadland, Carlo Favero, Geoff Heal, Omer Moav and three anonymous referees for helpful comments.
The Managing Editor in charge of this paper was Paul Seabright.
Economic Policy January 2007 pp. 177–229 Printed in Great Britain
© CEPR, CES, MSH, 2007.
180 EDWARD B. BARBIER
1.1. Key challenges and policy context
As a report from the US National Academy of Science has emphasized, ‘the fundamental
challenge of valuing ecosystem services lies in providing an explicit description and
adequate assessment of the links between the structure and functions of natural systems,
the benefits (i.e., goods and services) derived by humanity, and their subsequent values’
(Heal et al., 2005, p. 2). Moreover, it has been increasingly recognized by economists
and ecologists that the greatest ‘challenge’ they face is in valuing the ecosystem
services provided by a certain class of key ecosystem functions – regulatory and habitat
functions. The diverse benefits of these functions include climate stability, maintenance
of biodiversity and beneficial species, erosion control, flood mitigation, storm protection,
groundwater recharge and pollution control (see Table 1 below).
One of the natural ecosystems that has seen extensive development and application
of methods to value ecosystem services has been coastal wetlands. This paper focuses
mainly on valuation approaches applied to these systems, and in particular their role
as a nursery and breeding habitat for near-shore fisheries and in providing storm
protection for coastal communities.
The paper employs a case study of mangrove ecosystems in Thailand to compare
and contrast approaches to valuing habitat and storm protection services. Global
mangrove area has been declining rapidly, with around 35% of the total area lost in
the past two decades (Valiela et al., 2001). Mangrove deforestation has been particularly
prevalent in Thailand and other Asian countries. The main cause of global mangrove
loss has been coastal economic development, especially aquaculture expansion
(Barbier and Cox, 2003). Yet ecologists maintain that global mangrove loss is contributing
to the decline of marine fisheries and leaving many coastal areas vulnerable to
natural disasters. Concern about the deteriorating ‘storm protection’ service of
mangroves reached new significance with the 26 December 2004 Asian tsunami that
caused widespread devastation and loss of life in Thailand and other Indian Ocean
countries.
The Thailand case study also illustrates the importance of valuing ecosystem
services to policy choices. Because these services are ‘non-marketed’, their benefits
are not considered in commercial development decisions. For example, the excessive
mangrove deforestation occurring in Thailand and other countries is clearly related
to the failure to measure explicitly the values of habitat and storm protection services
of mangroves. Consequently, these benefits have been largely ignored in national land
use policy decisions, and calls to improve protection of remaining mangrove forests
and to enlist the support of local coastal communities through legal recognition of
their de facto property rights over mangroves are unlikely to succeed in the face of
coastal development pressures on these resources (Barbier and Sathirathai, 2004).
Unless the value to local coastal communities of the ecosystem services provided by
protected mangroves is estimated, it is difficult to convince policymakers in Thailand
and other countries to consider alternative land use policies.
VALUING ECOSYSTEM SERVICES 181
Thus, as the Thailand case study reveals, the challenge of valuing ecosystem
services is also a policy challenge. Because the benefits of these services are important
and should be taken into account in any future policy to manage coastal wetlands in
Thailand and other countries, it is equally essential that economics continues to
develop and improve existing methodologies to value ecological services.
1.2. Outline and main results
The paper makes three contributions. The first is to demonstrate that valuing eco-
logical services as productive inputs is a viable methodology for policy analysis, and
to illustrate the key steps through a detailed case study of mangroves in Thailand.
The second contribution is to identify the measurement issues that make valuation of
non-marketed ecosystem services a unique challenge, yet one that is important for
many important policy decisions concerning the management of natural ecosystems.
The third contribution of the paper is to show, using the examples of habitat and
storm protection services, that improvements in methods for valuing these services
can correct for some shortcomings and measurement errors, thus yielding more accu-
rate valuation estimates. But even the preferred approaches display measurement
weaknesses that need to be addressed in future developments of ecosystem valuation
methodologies.
Section 2 discusses in more detail the importance of valuing ecosystem services,
especially those arising from the regulatory and habitat functions to environmental
decision-making. Section 3 reviews various methods for valuing these services.
Because the benefits arising from ecological regulatory and habitat functions mainly
support or protect valuable economic activities, the production function (PF)
approach of valuing these benefits as environmental inputs is a promising methodology.
However, the latter approach faces its own unique measurement issues. To illus-
trate the PF approach as well as its shortcomings, the section discusses recent
advances using the examples of the habitat and storm protection services of coastal
wetland ecosystems. Section 4 compares the application of the different methods to
valuing mangroves in Thailand. The case study indicates the importance of consid-
ering the key ecological-economic linkages underlying each service in choosing the
appropriate valuation approach, and how each approach influences the final valua-
tion estimates. In the case of valuing the mangroves’ habitat-fishery linkage, model-
ling the contribution of this linkage to growth in fish stocks over time appears to be
a key consideration. The case study also demonstrates the advantages of the expected
damage function approach as an alternative to the replacement cost method of
valuing the storm protection service of coastal wetlands. Section 5 concludes the
paper by discussing the key areas for further development in ecosystem valuation
methodologies, such as incorporating the effects of irreversibilities, uncertainties and
thresholds, and the application of integrated ecological-economic modelling to reflect
multiple ecological services and their benefits. Although substantial progress has been
182 EDWARD B. BARBIER
made in valuing some ecosystem services, many difficulties still remain. Future
progress in ecosystem valuation for policy analysis requires understanding the key
flaws in existing methods that need correcting.
2. BACKGROUND: VALUATION OF ECOSYSTEM SERVICES
The rapid disappearance of many ecosystems has raised concerns about the loss
of beneficial ‘services’. This raises two important questions. What are ecosystem
services, and why is it important to value these environmental flows?
2.1. Ecosystem services
Although in the current literature the term ‘ecosystem services’ lumps together a
variety of ‘benefits’, economics normally classifies these benefits into three different
categories: (i) ‘goods’ (e.g. products obtained from ecosystems, such as resource
harvests, water and genetic material); (ii) ‘services’ (e.g. recreational and tourism
benefits or certain ecological regulatory functions, such as water purification, climate
regulation, erosion control, etc.); and (iii) cultural benefits (e.g., spiritual and religious,
heritage, etc.).1 This paper focuses on methods to value a sub-set of the second category
of ecosystem ‘benefits’ – the services arising from regulatory and habitat functions.
Table 1 provides some examples of the links between regulatory and habitat functions
and the resulting ecosystem benefits.
2.2. Valuing environmental assets
The literature on ecological services implies that natural ecosystems are assets that
produce a flow of beneficial goods and services over time. In this regard, they are no
different from any other asset in an economy, and in principle, ecosystem services
should be valued in a similar manner. That is, regardless of whether or not there
exists a market for the goods and services produced by ecosystems, their social value
must equal the discounted net present value (NPV) of these flows.
However, what makes environmental assets special is that they give rise to particular
measurement problems that are different for conventional economic or financial
assets. This is especially the case for the benefits derived from the regulatory and
habitat functions of natural ecosystems.
For one, these assets and services fall in the special category of ‘nonrenewable
resources with renewable service flows’ (Just et al., 2004, p. 603). Although a natural
ecosystem providing such beneficial services is unlikely to increase, it can be depleted,
for example through habitat destruction, land conversion, pollution impacts and so
1
See Daily (1997), De Groot et al. (2002) and Millennium Ecosystem Assessment (2003) for the various definitions of ecosystem
services that are prevalent in the ecological literature.
VALUING ECOSYSTEM SERVICES 183
Table 1. Some services provided by ecosystem regulatory and habitat functions
Ecosystem functions Ecosystem processes Ecosystem services (benefits)
and components
Regulatory functions
Gas regulation Role of ecosystems in Ultraviolet-B protection
biogeochemical processes Maintenance of air quality
Influence of climate
Climate regulation Influence of land cover and Maintenance of
biologically mediated processes temperature, precipitation
Disturbance prevention Influence of system Storm protection
structure on dampening Flood mitigation
environmental disturbance
Water regulation Role of land cover in Drainage and natural irrigation
regulating run-off, river Flood mitigation
discharge and infiltration Groundwater recharge
Soil retention Role of vegetation root matrix Maintenance of arable land
and soil biota in soil structure Prevention of damage from
erosion and siltation
Soil formation Weathering of rock and Maintenance of productivity
organic matter accumulation on arable land
Nutrient regulation Role of biota in storage Maintenance of productive
and recycling of nutrients ecosystems
Waste treatment Removal or breakdown of Pollution control and
nutrients and compounds detoxification
Habitat functions
Niche and refuge Suitable living space for Maintenance of biodiversity
wild plants and animals Maintenance of beneficial
species
Nursery and breeding Suitable reproductive Maintenance of biodiversity
habitat andnursery grounds Maintenance of beneficial
species
Sources: Adapted from Heal et al. (2005, Table 3-3) and De Groot et al. (2002).
forth. Nevertheless, if the ecosystem is left intact, then the flow services from the
ecosystem’s regulatory and habitat functions are available in quantities that are not
affected by the rate at which they are used.
In addition, whereas the services from most assets in an economy are marketed,
the benefits arising from the regulatory and habitat functions of natural ecosystems
generally are not. If the aggregate willingness to pay for these benefits is not revealed
through market outcomes, then efficient management of such ecosystem services
requires explicit methods to measure this social value (e.g., see Freeman, 2003; Just
et al., 2004). A further concern over ecosystem services is that their beneficial flows
are threatened by the widespread disappearance of natural ecosystems and habitats
across the globe. The major cause of this disappearance is conversion of the land to
other uses, degradation of the functioning and integrity of natural ecosystems through
resource exploitation, pollution, and biodiversity loss, and habitat fragmentation
(Millennium Ecosystem Assessment, 2003). The failure to measure explicitly the
aggregate willingness to pay for otherwise non-marketed ecological services exacerbates
184 EDWARD B. BARBIER
these problems, as the benefits of these services are ‘underpriced’ in development
decisions as a consequence. Population and development pressures in many areas of
the world result in increased land demand by economic activities, which mean that
the opportunity cost of maintaining the land for natural ecosystems is rarely zero.
Unless the benefits arising from ecosystem services are explicitly measured, or ‘valued’,
then these non-marketed flows are likely to be ignored in land use decisions. Only
the benefits of the ‘marketed’ outputs from economic activities, such as agricultural
crops, urban housing and other commercial uses of land, will be taken into account,
and as a consequence, excessive conversion of natural ecosystem areas for development
will occur.
A further problem is the uncertainty over their future values of environmental
assets. It is possible, for example, that the benefits of natural ecosystem services may
increase in the future as more scientific information becomes available over time. In
addition, if environmental assets are depleted irreversibly through economic develop-
ment, their value will rise relative to the value of other economic assets (Krutilla and
Fisher, 1985). Because ecosystems are in fixed supply, lack close substitutes and are
difficult to restore, their beneficial services will decline as they are converted or
degraded. As a result, the value of ecosystem services is likely to rise relative to other
goods and services in the economy. This rising, but unknown, future scarcity value of
ecosystem benefits implies an additional ‘user cost’ to any decision that leads to
irreversible conversion today.
Valuation of environmental assets under conditions of uncertainty and irreversibility
clearly poses additional measurement problems. There is now a considerable literature
advocating various methods for estimating environmental values by measuring the
additional ‘premium’ that individuals are willing to pay to avoid the uncertainty
surrounding such values (see Ready, 1995 for a review). Similar methods are also
advocated for estimating the user costs associated with irreversible development, as
this also amounts to valuing the ‘option’ of avoiding reduced future choices for indi-
viduals (Just et al., 2004). However, it is difficult to implement such methods empiri-
cally, given the uncertainty over the future state of environmental assets and about
the future preferences and income of individuals. The general conclusion from studies
that attempt to allow for such uncertainties in valuing environmental assets is that
‘more empirical research is needed to determine under what conditions we can ignore
uncertainty in benefit estimation ...where uncertainty is over economic parameters
such as prices or preferences, the issues surrounding uncertainty may be empirically
unimportant’ (Ready, 1995, p. 590).
3. VALUING THE ENVIRONMENT AS INPUT
Uncertainty and irreversible loss are important issues to consider in valuing ecosys-
tem services. However, as emphasized by Heal et al. (2005), a more ‘fundamental
challenge’ in valuing these flows is that ecosystem services are largely not marketed,
VALUING ECOSYSTEM SERVICES 185
and unless some attempt is made to value the aggregate willingness to pay for these
services, then management of natural ecosystems and their services will not be
efficient. The following section describes advances in developing the ‘production
function’ approach, compared to other valuation methods, as a means to measuring
the aggregate willingness to pay for the largely non-marketed benefits of ecosystem
services.
3.1. Methods of valuing ecosystem services
Table 2 indicates various methods that can be used for valuing ecological services.2
However, some approaches are limited to specific benefits. For example, the travel
cost method is used principally for environmental values that enhance individuals’
enjoyment of recreation and tourism, averting behaviour models are best applied to
the health effects arising from pollution, and hedonic wage and property models are
used primarily for assessing work-related hazards and environmental impacts on
property values, respectively.
In contrast, stated preference methods, which include contingent valuation methods,
conjoint analysis and choice experiments, have the potential to be used widely in
valuing ecosystem goods and services. These valuation methods involve surveying
individuals who benefit from an ecological service or range of services, and analysing
the responses to measure individuals’ willingness to pay for the service or services.
For example, choice experiments of wetland restoration in southern Sweden
revealed that individuals’ willingness to pay for the restoration increased if the result
enhanced overall biodiversity but decreased if the restored wetlands were used mainly
for the introduction of Swedish crayfish for recreational fishing (Carlsson et al., 2003).
In some cases, stated preference methods are used to elicit ‘non-use values’, that is,
the additional ‘existence’ and ‘bequest’ values that individuals attach to ensuring that
a well-functioning system will be preserved for future generations to enjoy. A contin-
gent valuation study of mangrove-dependent coastal communities in Micronesia
demonstrated that the communities ‘place some value on the existence and ecosystem
functions of mangroves over and above the value of mangroves’ marketable products’
(Naylor and Drew, 1998, p. 488).
However, to implement a stated-preference study two key conditions are necessary:
(1) the information must be available to describe the change in a natural ecosystem
in terms of service that people care about, in order to place a value on those services;
and (2) the change in the natural ecosystem must be explained in the survey instrument
in a manner that people will understand and not reject the valuation scenario (Heal
et al., 2005). For many of the services arising from ecological regulatory and habitat
2
It is beyond the scope of this paper to discuss all the valuation methods listed in Table 2. See Freeman (2003), Heal et al. (2005)
and Pagiola et al. (2004) for more discussion of these various valuation methods and their application to valuing ecosystem goods
and services.
186 EDWARD B. BARBIER
Table 2. Various valuation methods applied to ecosystem services
Valuation Types of value Common types Ecosystem
methoda estimatedb of applications services valued
Travel cost Direct use Recreation Maintenance of
beneficial species,
productive ecosystems
and biodiversity
Averting Direct use Environmental impacts Pollution control
behaviour on human health and detoxification
Hedonic Direct and Environmental impacts Storm protection; flood
price indirect use on residential property mitigation; maintenance
and human morbidity of air quality
and mortality
Production Indirect use Commercial and recreational Maintenance of beneficial
function fishing; agricultural systems; species; maintenance of
control of invasive species; arable land and agricultural
watershed protection; productivity; prevention of
damage costs avoided damage from erosion and
siltation; groundwater
recharge; drainage and
natural irrigation; storm
protection; flood mitigation
Replacement Indirect use Damage costs avoided; Drainage and natural
cost freshwater supply irrigation; storm protection;
flood mitigation
Stated Use and non-use Recreation; environmental All of the above
preference impacts on human health
and residential property;
damage costs avoided;
existence and bequest
values of preserving
ecosystems
a
See Freeman (2003), Heal et al. (2005) and Pagiola et al. (2004) for more discussion of these various valuation
methods and their application to valuing ecosystem goods and services.
b
Typically, use values involve some human ‘interaction’ with the environment whereas non-use values do not,
as they represent an individual valuing the pure ‘existence’ of a natural habitat or ecosystem or wanting to
‘bequest’ it to future generations. Direct use values refer to both consumptive and non-consumptive uses that
involve some form of direct physical interaction with environmental goods and services, such as recreational
activities, resource harvesting, drinking clean water, breathing unpolluted air and so forth. Indirect use values
refer to those ecosystem services whose values can only be measured indirectly, since they are derived from
supporting and protecting activities that have directly measurable values, such as many of the services listed in
Table 1.
Source: Adapted from Heal et al. (2005, Table 4-2) and Table 1.
functions, one or both of these conditions may not hold. For instance, it has proven
very difficult to describe accurately through the hypothetical scenarios required by
stated-preference surveys how changes in ecosystem processes and components affect
ecosystem regulatory and habitat functions and thus the specific benefits arising from
these functions that individuals value. If there is considerable scientific uncertainty
surrounding these linkages, then not only is it difficult to construct such hypothetical
scenarios but also any responses elicited from individuals from stated-preference surveys
are likely to yield inaccurate measures of their willingness to pay for ecological services.
VALUING ECOSYSTEM SERVICES 187
In contrast to stated-preference methods, the advantage of PF approaches is that
they depend on only the first condition, and not both conditions, holding. That is,
for those regulatory and habitat functions where there is sufficient scientific knowl-
edge of how these functions link to specific ecological services that support or protect
economic activities, then it may be possible to employ the PF approach to value these
services. However, PF methods have their own measurement issues and limitations.
These are also discussed further in the rest of this section, and illustrated using
examples of key ecological services from coastal and estuarine wetlands.
3.2. The production function approach
Many of the beneficial services derived from regulatory and habitat functions are
commonly classified by economists as indirect use values (Barbier, 1994). The benefits
attributed to these services arise through their support or protection of activities that
have directly measurable values (see Table 2). For example, coastal and estuarine
wetlands, such as tropical mangroves and temperate marshlands, act as ‘natural
barriers’ by preventing or mitigating storms and floods that could affect property and
land values, agriculture, fishing and drinking supplies, as well as cause sickness and
death. Similarly, coastal and estuarine wetlands may also provide a nursery and
breeding habitat that supports the productivity of near-shore fisheries, which in turn
may be valued for their commercial or recreational catch.
Because the benefits of these ecosystem services appear to enhance the productivity
of economic activities, or protect them from possible damages, one possible method
of measuring the aggregate willingness to pay for such services is to estimate their
value as if they were a factor input in these productive activities. This is the essence
of the PF valuation approaches, also called ‘valuing the environment as input’
(Barbier, 1994 and 2000; Freeman, 2003, ch. 9).3
The basic modelling approach underlying PF methods is similar to determining
the additional value of a change in the supply of any factor input. If changes in the
regulatory and habitat functions of ecosystems affect the marketed production
activities of an economy, then the effects of these changes will be transmitted to
individuals through the price system via changes in the costs and prices of final goods
and services. This means that any resulting ‘improvements in the resource base or
environmental quality’ as a result of enhanced ecosystem services, ‘lower costs and
prices and increase the quantities of marketed goods, leading to increases in consumers’
and perhaps producers’ surpluses’ (Freeman, 2003, p. 259). The sum of consumer
and producer surpluses in turn provides a measure of the willingness to pay for the
improved ecosystem services.
3
The concept of ‘valuing’ the environment as input is not new. Dose-response and change-in-productivity models, which have
been used for some time, can be considered special cases of the PF approach in which the production responses to environmen-
tal quality changes are greatly simplified (Freeman, 1982).
188 EDWARD B. BARBIER
An adaptation of the PF methodology is required in the case where ecological
regulatory and habitat functions have a protective value, such as the storm protection
and flood mitigation services provided by coastal wetlands. In such cases, the envi-
ronment may be thought of producing a non-marketed service, such as ‘protection’
of economic activity, property and even human lives, which benefits individuals
through limiting damages. Applying PF approaches requires modelling the ‘produc-
tion’ of this protection service and estimating its value as an environmental input in
terms of the expected damages avoided.
Although this paper focuses mainly on applications of the PF approach to coastal
wetland ecosystems, as Table 2 indicates PF approaches are being increasingly
employed for a diverse range of environmental quality impacts and ecosystem ser-
vices. Some examples include maintenance of biodiversity and carbon sequestration in
tropical forests (Boscolo and Vincent, 2003); nutrient reduction in the Baltic Sea
(Gren et al., 1997); pollination service of tropical forests for coffee production in Costa
Rica (Ricketts et al., 2004); tropical watershed protection services (Kaiser and
Roumasset, 2002); groundwater recharge supporting irrigation farming in Nigeria
(Acharya and Barbier, 2000); coral reef habitat support of marine fisheries in Kenya
(Rodwell et al., 2002); marine reserves acting to enhance the ‘insurance value’ of
protecting commercial fish species in Sicily (Mardle et al., 2004) and in the northeast
cod fishery (Sumaila, 2002); and nutrient enrichment in the Black Sea affecting the
balance between invasive and beneficial species (Knowler et al., 2001).
3.3. Measurement issues for modelling habitat-fishery linkages
Applying PF methods to valuing ecosystem services has its own demands in terms of
ecological and economic data. To highlight these additional measurement issues, this
section draws on the example of valuing coastal wetlands as a nursery and breeding
habitat for commercial near-shore fisheries.
First, application of the PF approach requires properly specifying the habitat-
fishery PF model that links the physical effects of the change in this service to changes
in market prices and quantities and ultimately to consumer and producer surpluses.
As with many ecological services, it is difficult to measure directly changes in the
habitat and nursery function of coastal wetlands. Instead, the standard approach
adopted in coastal habitat-fishery PF models is to allow the wetland area to serve as
a proxy for the productivity contribution of the nursery and habitat function (see
Barbier, 2000 for further discussion). It is then relatively straightforward to estimate
the impacts of the change in the coastal wetland area input on fishery catch, in terms
of the marginal costs of fishery harvests and thus changes in consumer and producer
surpluses.
Second, market conditions and regulatory policies for the marketed output will
influence the values imputed to the environmental input (Freeman, 1991). For
instance, the offshore fishery supported by coastal wetlands may be subject to open
VALUING ECOSYSTEM SERVICES 189
access. Under these conditions, profits in the fishery would be dissipated, and
equilibrium prices would be equated to average and not marginal costs. As a
consequence, there is no producer surplus, and the welfare impact of a change in
wetland habitat is measured by the resulting change in consumer surplus only.
Third, if the ecological service supports a harvested natural resource system, such
as a fishery, forestry or a wildlife population, then it may be necessary to model how
changes in the stock or biological population may affect the future flow of benefits.
If the natural resource stock effects are not considered significant, then the environ-
mental changes can be modelled as impacting only current harvest, prices and
consumer and producer surpluses. If the stock effects are significant, then a change
in an ecological service will impact not only current but also future harvest and
market outcomes. In the PF valuation literature, the first approach is referred to as a
‘static model’ of environmental change on a natural resource production system,
whereas the second approach is referred to as a ‘dynamic model’ because it takes into
account the intertemporal stock effects of the environmental change (Barbier, 2000;
Freeman, 2003, ch. 9).
Finally, most natural ecosystems provide more than one beneficial service, and it
may be important to model any trade-offs among these services as an ecosystem is
altered or disturbed. Integrated economic-ecological modelling could capture more
fully the ecosystem functioning and dynamics underlying the provision of key services,
and can be used to value multiple services arising from natural ecosystems. For
instance, integrated modelling of an entire wetland-coral reef-sea grass system could
measure simultaneously the benefits of both the habitat-fishery linkage and the storm
protection service provided by the system. Examples of such multi-service ecosystem
modelling include analysis of salmon habitat restoration (Wu et al., 2003); eutrophi-
cation of small shallow lakes (Carpenter et al., 1999); changes in species diversity in
a marine ecosystem (Finnoff and Tschirhart, 2003); and introduction of exotic trout
species (Settle and Shogren, 2002).
To illustrate the first three of the above issues, I next explore two ways of measur-
ing the welfare effects of an environmental change on a productive natural resource
system with the example of the coastal habitat-fishery linkage. I will return to the
issue of integrated ecological-economic modelling of multiple ecological services in
Section 5.
3.3.1. Habitat-fishery linkages: static approaches. This section illustrates the
use of a static model to value how a change in coastal wetland habitat area affects
the market for commercially harvested fish. Many initial PF methods to value habitat-
fishery linkages have relied on this static approach. For example, using data from the
Lynne et al. (1981), Ellis and Fisher (1987) constructed such a model to value the
support by Florida marshlands for Gulf Coast crab fisheries in terms of the resulting
changes in consumer and producer surpluses from the marketed catch. Freeman
(1991) then extended Ellis and Fisher’s approach to show how the values imputed to
190 EDWARD B. BARBIER
the wetlands in the static model is influenced by whether or not the fishery is
open access or optimally managed. Sathirathai and Barbier (2001) also used a static
model of habitat-fishery linkages to value the role of mangroves in Thailand in
supporting near-shore fisheries under both open access and optimally managed
conditions.
As most near-shore fisheries are not optimally managed but open access, the following
illustration of the static model of habitat-fishery linkages assumes that the fishery is
open access. Any profits in the fishery will attract new entrants until all the profits
disappear, and in equilibrium, the welfare change in coastal wetland is in terms of its
impact on consumer surplus only.
As noted above, the general PF approach treats an ecological service, such as
coastal wetland habitat, as an ‘input’ into the economic activity, and like any other
input, its value can be equated with its impact on the productivity of any marketed
output. More formally, if h is the marketed harvest of the fishery, then its production
function can be denoted as:
h = h( E i . . . E k , S ) (1)
The area of coastal wetlands, S, may therefore have a direct influence on the marketed
fish catch, h, which is independent from the standard inputs of a commercial fishery,
Ei . . . E k .
A standard assumption in most static habitat-fishery models is that the production
function (1) takes the Cobb–Douglas form, h = AE aS b, where E is some aggregate
measure of total effort in the off-shore fishery and S is coastal wetland habitat area.
It follows that the optimal cost function of a cost-minimizing fishery is:
C * = C(h, w, S ) = wA−1/ah1/aS −b/a (2)
where w is the unit cost of effort. Assuming an iso-elastic market demand function,
P = p(h) = khη, η = 1/ε < 0 , then the market equilibrium for catch of the open access
fishery occurs where the total revenues of the fishery just equals cost, or price equals
average cost, i.e. P = C */h, which in this model becomes:
khη = wA−1/ah1−a/aS −b/a (3)
which can be rearranged to yield the equilibrium level of fish harvest:
a/β
w
A −1/β S −b/β , β = (1 + η ) a − 1
h= (4)
k
It follows from (4) that the marginal impact of a change in wetland habitat is:
a/β
b w
dh
A −1/β S −(b+β )/β
=− (5)
βk
dS
The change in consumer surplus, CS, resulting from a change in equilibrium
harvest levels (from h0 to h1) is:
VALUING ECOSYSTEM SERVICES 191
h1
k[( h1 )η+1 – ( h 0 )η+1 ]
∆CS = − k[( h1 )η+1 – ( h 0 )η+1 ]
p( h )dh − [ p1h1 − p 0h 0 ] =
η +1 (6)
h0
η[ p h − p h ]
11 00
=− .
η +1
By utilizing (5) and (6) it is possible to estimate the new equilibrium harvest and
price levels and thus the corresponding changes in consumer surplus associated with
a change in coastal wetland area, for a given demand elasticity, γ.
Figure 1 is the diagrammatic representation of the welfare measure of a change in
wetland area on an open access fishery corresponding to Equation (6). As shown in
the figure, a change in wetland area that serves as a breeding ground and nursery for
an open access fishery results in a shift in the average cost curve, AC, of the fishery.
The welfare impact is the change in consumer surplus (area P*ABC).
3.3.2. Habitat-fishery linkages: dynamic approaches. If the stock effects of a
change in coastal wetlands are significant, then valuing such changes in terms of the
impacts on current harvest and market outcomes is a flawed approach. To overcome
this shortcoming, a dynamic model of coastal habitat-fishery linkage incorporates
the change in wetland area within a multi-period harvesting model of the fishery. The
standard approach is to model the change in coastal wetland habitat as affecting the
biological growth function of the fishery (Barbier, 2003). As a result, any value
impacts of a change in this habitat-support function can be determined in terms of
changes in the long-run equilibrium conditions of the fishery. Alternatively, the
welfare analysis could be conducted in terms of the harvesting path that approaches
this equilibrium or the path that is moving away from initial conditions in the fishery.
Figure 1. The economic value effects of increased wetland area on an open
access fishery
Notes: AC: average cost; D: demand curve; P*: price per tonne; h*: fish catch in tonnes after change; P*ABC:
change in consumer and producer surplus.
Source: Adapted from Freeman (1991).
192 EDWARD B. BARBIER
Most attempts to value habitat-fishery linkages via a dynamic model that incorporates
stock effects have assumed that the fishery affected by the habitat change is in a long-run
equilibrium. Such a model has been applied, for example, in case studies of valuing
habitat fishery linkages in Mexico (Barbier and Strand, 1998), Thailand (Barbier et
al., 2002; Barbier, 2003) and the United States (Swallow, 1994). Similar ‘equilibrium’
dynamic approaches have been used to model other coastal environmental changes,
including the impacts of water quality on fisheries in the Chesapeake Bay (Kahn and
Kemp, 1985; McConnell and Strand, 1989) and the effects of mangrove deforestation
and shrimp larvae availability on aquaculture in Ecuador (Parks and Bonifaz, 1997).
However, valuing the change in coastal wetland habitat in terms of its impact on
the long-run equilibrium of the fishery raises additional methodological issues. First,
the assumption of prevailing steady state conditions is strong, and may not be a
realistic representation of harvesting and biological growth conditions in the near-
shore fisheries. Second, such an approach ignores both the convergence of stock and
harvest to the steady state and the short-run dynamics associated with the impacts of
the change in coastal habitat on the long-run equilibrium. The usual assumption is
that this change will lead to an instantaneous adjustment of the system to a new
steady state, but this in turn requires local stability conditions that may not be supported
by the parameters of the model.
There are examples of pure fisheries models that assume that the dynamic system
is not in equilibrium but is either on the approach to a steady state or is moving away
from initial fixed conditions. The latter approach has proven particularly useful in the
case of open access or regulated access fisheries (Bjørndal and Conrad, 1987; Homans and
Wilen, 1997). The following model shows how this approach can be adopted here to
the case of valuing a change in wetland habitat in terms of the dynamic path of an
open access fishery.
Defining Xt as the stock of fish measured in biomass units, any net change in
growth of this stock over time can be represented as:
∂ 2F ∂F
X t − X t −1 = F ( X t −1, St −1 ) − h( X t −1, Et −1 ), > 0, > 0. (7)
∂X t −1 ∂St −1
2
Thus, net expansion in the fish stock occurs as a result of biological growth in the
current period, F (Xt−1, St−1), net of any harvesting, h(Xt−1, Et−1), which is a function of
the stock as well as fishing effort, Et −1. The influence of the wetland habitat area, St −1,
as a breeding ground and nursery habitat on growth of the fish stock is assumed to
be positive, ∂F/∂St−1 > 0, as an increase in wetland area will mean more carrying
capacity for the fishery and thus greater biological growth.
As before, it is assumed that the near-shore fishery is open access. The standard
assumption for an open access fishery is that effort next period will adjust in response
to the real profits made in the past period (Clark, 1976; Bjørndal and Conrad, 1987).
Letting p(h) represent landed fish price per unit harvested, w the unit cost of effort
and φ > 0 the adjustment coefficient, then the fishing effort adjustment equation is:
VALUING ECOSYSTEM SERVICES 193
∂p( ht −1 )
Et − Et −1 = φ[ p( ht −1 ) h( X t −1, Et −1 ) − wEt −1 ],< 0. (8)
∂ht −1
Assume a conventional bioeconomic fishery model with biological growth character-
ized by a logistic function, F (Xt−1, St−1) = rXt−1[1 − Xt−1/K (St−1)], and harvesting by a
Schaefer production process, ht = qXtEt, where q is a ‘catchability’ coefficient, r is the
intrinsic growth rate and K (St) = α ln St, is the impact of coastal wetland area on
carrying capacity, K, of the fishery. The market demand function for harvested fish is
again assumed to be iso-elastic, i.e. p(h) = khη, η = 1/ε < 0. Substituting these
expressions into (7) and (8) yields:
X t −1
X t = rX t −1 1 − − ht −1 + X t −1 (9)
α ln St −1
Et = φRt −1 + (1 − φw )Et −1, Rt −1 = kht1−1η .
+
(10)
Both Xt and Et are predetermined, and so (9) and (10) can be estimated independently
(see Homans and Wilen, 1997). Following Schnute (1977), define the catch per unit
effort as ct = ht/Et = qXt. If Xt is predetermined so is ct. Substituting the expression
for catch per unit effort in (9) produces:
ct − ct −1 r ct −1
=r− − qEt −1. (11)
qα ln St −1
ct −1
Thus Equations (10) and (11) can also be estimated independently to determine the
biological and economic parameters of the model. For given initial effort, harvest and
wetland data, both the effort and stock paths of the fishery can be determined for
subsequent periods, and the consumer plus producer surplus can be estimated for
each period. Alternative effort and stock paths can then be determined as wetland
area changes in each period, and thus the resulting changes in consumer plus pro-
ducer surplus in each period are the corresponding estimates of the welfare impacts
of the coastal habitat change.4
3.4. Replacement cost and cost of treatment
In circumstances where an ecological service is unique to a specific ecosystem and is
difficult to value, then economists have sometimes resorted to using the cost of replac-
ing the service as a valuation approach.5 This method is usually invoked because of
the lack of data for many services arising from natural ecosystems.
For example, the presence of a wetland may reduce the cost of municipal water
treatment because the wetland system filters and removes pollutants. It is therefore
4
As along its dynamic path the open access fishery is not in equilibrium, producer surpluses, or losses, are relevant for the
welfare estimate of a change in coastal wetland habitat.
5
Such an approach to approximating the benefits of a service by the cost of providing an alternative is not used exclusively in
environmental valuation. For example, in the health economics literature this approach is referred to as ‘cost of illness’ (Dickie, 2003).
This involves adding up the costs of treating a patient for an illness as the measure of the benefit to the patient of staying disease-free.
194 EDWARD B. BARBIER
tempting to use the cost of an alternative treatment method, such as the building and
operation of an industrial water treatment plant, to represent the value of the wetland’s
natural water treatment service. Such an approach does not measure directly the
benefit derived from the wetland’s waste treatment service; instead, the approach is
estimating this benefit with the cost of providing the ecosystem service that people
value. Herein lies the main problem with the replacement cost method: it is using
‘costs’ as a measure of economic ‘benefit’. In economic terms, the implication is that
the ratio of costs to benefit of an ecological service is always equal to one.
The problems posed by the replacement cost method are illustrated in Figure 2,
in the case of waste water treatment service provided by an existing wetland ecosys-
tem. The cost of the waste water treatment service provided by the wetlands is ‘free’
and thus corresponds to the horizontal axis, MCS. Given the demand curve for water,
Q 1 amount of water is consumed. However, if the wetland is destroyed the marginal
cost of an alternative, human-built waste treatment facility is MCH. Thus, the ‘replacement
cost’ of using the treatment facility to provide Q 1 amount of water in the absence of
the wetlands is the difference between the two supply curves, or area 0BDQ 1.
However, this overestimates the benefit of having the wetlands provide the waste
treatment service. The true benefit of this ecosystem service is the demand curve, or total
willingness to pay, for Q 1 amount of water less the costs of providing it, or area 0ACQ 1.
For these reasons, economists consider that the replacement cost approach should
be used with caution. Shabman and Batie (1978) suggested that this method can
provide a reliable valuation estimation for an ecological service if the following con-
ditions are met: (1) the alternative considered provides the same services; (2) the
alternative compared for cost comparison should be the least-cost alternative; and (3)
there should be substantial evidence that the service would be demanded by society
if it were provided by that least-cost alternative. In the absence of any information
on benefits, and a decision has to be made to take some action, then treatment costs
become a way of looking for a cost-effective action.
Figure 2. Replacement cost estimation of an ecosystem service
Source: Adapted from Ellis and Fisher (1987).
VALUING ECOSYSTEM SERVICES 195
One of the best-known examples of a policy decision based on using the ‘replace-
ment cost’ method to assess the value of an ecosystem service is the provision of clean
drinking water by the Catskills Mountains for New York City (Heal et al., 2005). In
1996, New York City faced a choice: either it could build water filtration systems to
clean its water supply or the city could restore and protect the Catskill watersheds to
ensure high-quality drinking water. Because estimates indicated that building and
operating the filtration system would cost $6–8 billion whereas protecting and restoring
the watersheds would cost $1–1.5 billion, New York chose to protect the Catskills. In
this case, it was sufficient for the policy decision simply to demonstrate the cost-
effectiveness of restoring and protecting the ecological integrity of the Catskills watersheds
compared to the alternative of the human-constructed water filtration system. Thus,
clearly this is an example where the criteria established by Shabman and Batie (1978)
apply.
The main reason why economists have resorted to replacement cost approaches to
valuing an ecosystem service, however, is that there is often a lack of data on the
linkage between the initial ecological function, the processes and components of
ecosystems that facilitate this function, and the eventual ecological service that
benefits humans. The lack of such data makes it extremely difficult to construct
reliable hypothetical scenarios through stated preference surveys and similar methods
to elicit accurate responses from individuals about their willingness to pay for
ecological services. As an illustration, in the Catskills case study, a stated preference
survey may have elicited an estimate of the total willingness-to-pay by New York City
residents for the amount of freshwater provided – for example, the total demand for
freshwater Q 1 in Figure 2 – but it would have been very difficult to obtain a measure
of the willingness-to-pay to avoid losses in the water treatment service that occur
through changes in the land use in Catskills watershed that affect the free provision of
this ecological service.
Similarly, as pointed out by Chong (2005), it is very difficult to use stated preference
methods in tropical developing areas to assess the benefits to local communities
of the storm protection service of mangrove systems. Although there is sufficient
scientific evidence suggesting that such a service occurs, there is a lack of ecological
data on how loss of mangroves in specific locations will affect their ability to
provide storm protection to neighbouring communities. To date, the few studies
that have attempted to value the storm prevention and flood mitigation services
of the ‘natural’ storm barrier function of mangrove systems have employed the
replacement cost method by simply estimating the costs of replacing mangroves
with constructed barriers that perform the same services (Chong, 2005). Unfortu-
nately, such estimates not only make the classic error of estimating a ‘benefit’ by
a ‘cost’ but also may yield unrealistically high estimates, given that removing all the
mangroves and replacing them with constructed barriers is unlikely to be the least-
cost alternative to providing storm prevention and flood mitigation services in coastal
areas.
196 EDWARD B. BARBIER
3.5. Expected damage function approach
For some ecological services, an alternative to employing replacement cost methods
might be the expected damage function (EDF) approach.6
The EDF approach, which is a special category of ‘valuing’ the environment as
‘input’, is nominally straightforward; it assumes that the value of an asset that yields
a benefit in terms of reducing the probability and severity of some economic damage
is measured by the reduction in the expected damage. The essential step to imple-
menting this approach, which is to estimate how changes in the asset affect the
probability of the damaging event occurring, has been used routinely in risk analysis
and health economics, for example, as in the case of airline safety performance (Rose,
1990); highway fatalities (Michener and Tighe, 1992); drug safety (Olson, 2004); and
studies of the incidence of diseases and accident rates (Cameron and Trivedi, 1998;
Winkelmann, 2003). Here we show that the EDF approach can also be applied,
under certain circumstances, to value ecological services that also reduce the prob-
ability and severity of economic damages.
Recall that one of the special features of many regulatory and habitat services of
ecosystems is that they may protect nearby economic activities, property and even
human lives from possible damages. As indicated in Table 1, such services include
storm protection, flood mitigation, prevention of erosion and siltation, pollution con-
trol and maintenance of beneficial species. The EDF approach essentially ‘values’
these services through estimating how they mitigate damage costs.
The following example illustrates how the expected damage function (EDF ) methodology
can be applied to value the storm protection service provided by a coastal wetland,
such as a marshland or mangrove ecosystem. The starting point is the standard
‘compensating surplus’ approach to valuing a quantity or quality change in a non-
market environmental good or service (Freeman, 2003).
Assume that in a coastal region the local community owns all economic activity
and property, which may be threatened by damage from periodic natural storm
events. Assume also that the preferences of all households in the community are
sufficiently identical so that it can be represented by a single household. Let
m(px, z, u0) be the expenditure function of the representative household, that is, the
minimum expenditure required by the household to reach utility level, u0, given the
vector of prices, px, for all market-purchased commodities consumed by the household,
the expected number or incidence of storm events, z0.
Suppose the expected incidence of storms rises from z0 to z1. The resulting
expected damages to the property and economic livelihood of the household, E[D(z)],
translates into an exact measure of welfare loss through changes in the minimum
expenditure function:
6
The expected damage function approach predates many of the PF methods discussed so far, and has been used extensively to
estimate the risk of health impacts from pollution (Freeman, 1982, chs. 5 and 9).
VALUING ECOSYSTEM SERVICES 197
E[D(z)] = m(px, z1, u0) − m(px, z0, u0) = c(z) (12)
where c(z) is the compensating surplus. It is the minimum income compensation that
the household requires to maintain it at the utility level u0, despite the expected
increase in damaging storm events. Alternatively, c(z) can be viewed as the minimum
income that the household needs to avoid the increase in expected storm damages.
However, the presence of coastal wetlands could mitigate the expected incidence
of damaging storm events. Because of this storm protection service, the area of coastal
wetlands, S, may have a direct effect on reducing the ‘production’ of natural disasters,
in terms of their ability to inflict damages locally. Thus the ‘production function’ for
the incidence of potentially damaging natural disasters can be represented as:
z = z(S ), z′ < 0, z ″ > 0. (13)
It follows from (12) and (13) that ∂c(z)/∂S = ∂E[D(z)]/∂S < 0. An increase in wetland
area reduces expected storm damages and therefore also reduces the minimum
income compensation needed to maintain the household at its original utility level.
Alternatively, a loss in wetland area would increase expected storm damages and
raises the minimum compensation required by the household to maintain its welfare.
Thus, we can define the marginal willingness to pay, W(S ), for the protection services
of the wetland in terms of the marginal impact of a change in wetland area on
expected storm damages:
∂D
∂E[ D (z (S ))]
W (S ) = − = − E z ′, W ′ < 0. (14)
∂S ∂z
The ‘marginal valuation function’, W(S ), is analogous to the Hicksian compensated
demand function for marketed goods. The minus sign on the right-hand sign of (14)
allows this ‘demand’ function to be represented in the usual quadrant, and it has the
normal downward-sloping property (see Figure 3). Although an increase in S reduces
z and thus enables the household to avoid expected damages from storms, the addi-
tional value of this storm protection service to the household will fall as wetland area
increases in size. This relationship should hold across all households in the coastal
community. Consequently, as indicated in Figure 3, the marginal willingness to pay
by the community for more storm protection declines with S.
The value of a non-marginal change in wetland area, from S0 to S1, can be
measured as:
S1
− W (S )dS = E[ D (z (S ))] = c(S ). (15)
S0
If there is an increase in wetland area, then the value of this change is the total
amount of expected damage costs avoided. If there is a reduction in wetland area, as
shown in Figure 3, then the welfare loss is the total expected damages resulting from
the increased incidence of storm events. As indicated in (15), in both instances the
198 EDWARD B. BARBIER
Figure 3. Expected damage costs from a loss of wetland area
valuation would be a compensation surplus measure of a change in the area of
wetlands and the storm protection service that they provide.
As indicated in (14), an estimate of the marginal impact of a change in wetland
area on expected storm damages has two components: the influence of wetland area
on the expected incidence of economically damaging natural disaster events, z′, and
some measure of the additional economic damage incurred per event. Thus the right-
hand expression in (14) can be estimated, provided that there are sufficient data on
past storm events, and preferably across different coastal areas, and some estimate of
the economic damages inflicted by each event. The most important step in the
analysis is the first one, using the data on the incidence of past natural disasters and
changes in wetland area in coastal areas to estimate z(S ). One way this analysis can
be done is through employing count data models.
Count data models explain the number of times a particular event occurs over a
given period. In economics, count data models have been used to explain a variety
of phenomenon, such as explaining successful patents derived from firm R&D expen-
ditures, accident rates, disease incidence, crime rates and recreational visits (Cameron
and Trivedi, 1998; Greene, 2003, ch. 21; Winkelmann, 2003). Count data models
could be used to estimate whether a change in the area of coastal wetlands, S, reduces
the expected incidence of economically damaging storm events. The basic methodology
for such an application of count data models is described further in the appendix.
However, applying the EDF method to estimating the storm protection value of
coastal wetlands raises two additional measurement issues.
First, as the 2004 Asian tsunami and recent hurricanes in the United States have
demonstrated, the risks to vulnerable populations living in coastal areas from the
economic damages of storm events can be very large. This suggests that coastal
populations will display a degree of risk aversion to such events, in the sense that they
would like to see the least possible variance in expected storm damages. Applying
standard techniques, such as the capital-asset pricing model, this implies in turn that
VALUING ECOSYSTEM SERVICES 199
there should be a ‘risk premium’ attached to the storm protection value of coastal
wetlands that reduces the variance in expected economic damages from storm events
(Hirshleifer and Riley, 1992).
Second, estimating how coastal wetlands affect the expected number of economic
damaging events from the count data model and then multiplying the effect by the
average economic damages across events could be misleading under some extreme
circumstances. For instance, suppose a loss in wetland area is associated with a situ-
ation in which there is a change in the incidence of storms from one devastating
storm to two relatively minor storms per year. The count data model would then be
interpreted as not providing evidence against the null that the change in the wetland
area increases expected storm damages. Clearly, there needs to be a robustness check
on the count data model to ensure that such situations do not dominate the applica-
tion of the EDF approach.
4. CASE STUDY OF MANGROVE ECOSYSTEMS IN THAILAND
This section illustrates the application of the PF approach and the EDF approach to
valuation of ecological services with a case study of mangrove ecosystems in Thailand.
The two services of interest are the provision of a breeding and nursery habitat for
fisheries and the storm protection service of mangroves.
Both the dynamic and static PF approaches are used to estimate the value of the
mangrove-fishery habitat service. The EDF approach to estimating the storm protection
service of mangroves is contrasted with the replacement cost method.
4.1. Case study background
Many mangrove ecosystems, especially those in Asia, are threatened by rapid
deforestation. At least 35% of global mangrove area has been lost in the past two
decades; in Asia, 36% of mangrove area has been deforested, at the rate of 1.52%
per year (Valiela et al., 2001). Although many factors are behind global mangrove
deforestation, a major cause is aquaculture expansion in coastal areas, especially the
establishment of shrimp farms (Barbier and Cox, 2003). Aquaculture accounts for
52% of mangrove loss globally, with shrimp farming alone accounting for 38% of
mangrove deforestation; in Asia, aquaculture contributes 58% to mangrove loss with
shrimp farming accounting for 41% of total deforestation (Valiela et al., 2001).
Mangrove deforestation has been particularly prevalent in Thailand. Some estimates
suggest that over 1961–96 Thailand lost around 2050 km2 of mangrove forests, or
about 56% of the original area, mainly due to shrimp aquaculture and other coastal
developments (Charuppat and Charuppat, 1997). Since 1975, 50– 65% of Thailand’s
mangroves have been converted to shrimp farms (Aksornkoae and Tokrisna, 2004).
Figure 4 shows two long-run trend estimates of mangrove area in Thailand. In
1961, there were approximately 3700 km2 of mangroves, which declined steadily to
200 EDWARD B. BARBIER
Figure 4. Mangrove area (km2) in Thailand, 1961–2004
Notes: FAO estimates from FAO (2003). 2000 and 2004 data are estimated from 1990 –2000 annual average
mangrove loss of 18.0 km2. Thailand estimates from various Royal Thailand Forestry Department sources
reported in Aksornkoae and Tokrisna (2004). 2000 and 2004 data are estimated from 1993 –96 annual average
mangrove loss of 3.44 km2.
Sources: Based on FAO (2003), Aksornkoae and Tokrisna (2004) and author’s estimates.
around 2700 to 2900 km2 by 1980. Since then, mangrove deforestation has continued,
although there are disagreements over the rate of deforestation. For example, FAO
estimates based on long-run trend rates suggest a slower rate of decline, and indicate
that there may be almost 2400 km2 of mangroves still remaining. However, estimates
based on Thailand’s Royal Forestry Department studies suggest that rapid shrimp
farm expansion during the 1980s and early 1990s accelerated mangrove deforestation,
and as a consequence, the area of mangroves in 2004 may be much lower, closer to
1,645 km2.
Mangrove deforestation in Thailand has focused attention on the two principal
services provided by mangrove ecosystems, their role as nursery and breeding habitats
for offshore fisheries and as natural ‘storm barriers’ to periodic coastal storm events,
such as wind storms, tsunamis, storm surges and typhoons. In addition, many coastal
communities exploit mangroves directly for a variety of products, such as fuelwood,
timber, raw materials, honey and resins, and crabs and shellfish. One study estimated
that the annual value to local villagers of collecting these products was $88 per
hectare (ha), or approximately $823/ha in net present value terms over a 20-year
period and with a 10% discount rate (Sathirathai and Barbier, 2001).
4.1.1. Breeding and nursery habitat for fisheries. An extensive literature in
ecology has emphasized the role of coastal wetland habitats in supporting neighbouring
marine fisheries (for a review, see Mitsch and Gosselink, 1993; World Conservation
Monitoring Center; World Resources Institute 1996). Mangroves in Thailand also
provide this important habitat service (Aksornkoae et al., 2004).
VALUING ECOSYSTEM SERVICES 201
Thailand’s coastline is vast, stretching for 2815 km, of which 1878 km is on the
Gulf of Thailand and 937 km on the Andaman Sea (Indian Ocean) (Kaosa-ard and
Pednekar, 1998). Since 1972, the 3 km offshore coastal zone in southern Thailand
has been reserved for small-scale, artisanal marine fisheries. The Gulf of Thailand is
divided into four such major zones, and the Andaman Sea comprises a fifth zone.7
The mangroves along these coastal zones are thought to provide breeding grounds
and nurseries in support of several species of demersal fish and shellfish (mainly crab
and shrimp) in Thailand’s coastal waters.8 The artisanal marine fisheries of the five
major coastal zones of Thailand depend largely on shellfish but also some demersal
fish. For example, in 1994 shrimp, crab, squid and cuttlefish alone accounted for 67%
of all catch in the artisanal marine fisheries, and demersal fish accounted for 5.3%
(Kaosa-ard and Pednekar, 1998).
The coastal artisanal fisheries of Thailand are characterized by classic open access
conditions (Kaosa-ard and Pedneker, 1998; Wattana, 1998). Since the 1970s, there
have been approximately 36 000 –38 000 households engaged in small-scale fishing
activities. Although there are 2500 fishing communities scattered over the 24 coastal
provinces of Thailand, 90% of the artisanal fishing households are concentrated in
communities spread along the Southern Gulf of Thailand and Andaman Sea coasts.
While the number of households engaged in small-scale fishing has remained fairly
stable since 1985, the use of motorized boats has increased by more than 30%
(Wattana, 1998). Gill nets still remain the most common form of fishing gear used
by artisanal fishers. Although a licence fee and permit are required for fishing in
coastal waters, officials do not strictly enforce the law and users do not pay. Currently,
there is no legislation for supporting community-based fishery management (Kaosa-
ard and Pednekar, 1998).
4.1.2. Storm protection. The 26 December 2004 Indian Ocean tsunami disaster
has focused attention on the role of natural barriers, such as mangroves, in protecting
vulnerable coastlines and populations in the region from such storm events (UNEP,
2005; Wetlands International, 2005). Mangrove wetlands, which are found along
sheltered tropical and subtropical shores and estuaries, are particularly valuable in
minimizing damage to property and loss of human life by acting as a barrier against
tropical storms, such as typhoons, cyclones, hurricanes and tsunamis (Chong, 2005;
Massel et al., 1999; Mazda et al., 1997). Evidence from the 12 Indian Ocean countries
affected by the tsunami disaster, including Thailand, suggests that those coastal areas
7
The four Gulf of Thailand zones consist of the following coastal provinces: Trat, Chantaburi and Rayong (Zone 1); Chon
Buri, Chachoengsao, Samut Parkakan, Samut Sakhon, Samut Songkhram, Phetchaburi, Prachaup Khiri Khan (Zone 2);
Chumphon, Surat Thani, Nakhon Si Thammarat (Zone 3); and Songkhla, Patthani, Narathiwart (Zone 4). The fifth zone on
the Indian Ocean (Andaman Sea) consists of the following coastal provinces: Ranong, Phangnga, Phuket, Krabi, Trang and
Satun (Zone 5).
8
Mangrove-dependent demersal fish include those belonging to the Clupeidae, Chanidae, Ariidae, Pltosidae, Mugilidae, Lujanidae and
Latidae families. The shellfish include those belonging to the families of Panaeidae for shrimp and Grapsidae, Ocypodidae and Portnidae
for crab.
202 EDWARD B. BARBIER
that had dense and healthy mangrove forests suffered fewer losses and less damage
to property than those areas in which mangroves had been degraded or converted to
other land uses (Dahdouh-Guebas et al., 2005; Harakunarak and Aksornkoae, 2005;
Kathiresan and Rajendran, 2005; UNEP, 2005; Wetlands International, 2005).
In Thailand, the Asian tsunami affected all six coastal provinces along the Indian
Ocean (Andaman Sea) coast: Krabi, Phang Nga, Phuket, Ranong, Satun and Trang.
In Phang Nga, the most affected province, post-tsunami assessments suggest that
large mangrove forests in the north and south of the province significantly mitigated
the impact of the Tsunami. They suffered damage on their seaside fringe, but
reduced the tidal wave energy, providing protection to the inland population (UNEP,
2005; Harakunarak and Aksornkoae, 2005). Similar results were reported for those
shorelines in Ranong Province protected by dense and thriving mangrove forests. In
contrast, damages were relatively extensive along the Indian Ocean coast where
mangroves and other natural coastal barriers were removed or severely degraded
(Harakunarak and Aksornkoae, 2005).
With the overwhelming evidence of the storm protection service provided by intact
and healthy mangrove systems, since the tsunami disaster increased emphasis has
been placed on replanting degraded and deforested mangrove areas in Asia as a
means to bolstering coastal protection. For example, the Indonesian Minister for
Forestry has announced plans to reforest 600 000 hectares of depleted mangrove
forest throughout the nation over the next 5 years. The governments of Sri Lanka
and Thailand have also stated publicly intentions to rehabilitate and replant man-
grove areas (UNEP, 2005; Harakunarak and Aksornkoae, 2005).
Although the Asian tsunami has called attention to the storm protection service
provided by mangroves, the benefits of this service extends to protection against many
types of periodic coastal natural disaster events. As one post-tsunami assessment
noted: ‘It is important to recognize that any compromising of mangrove “protection
function” is relevant to a wide variety of storm events, and not just tsunamis. Whereas
the Indian Ocean area counted “only” 63 tsunamis between 1750 and 2004, there
were more than three tropical cyclones per year in roughly the same area’ (Dahdouh-
Guebas et al., 2005, pp. 445– 6).
The EM-DAT International Disaster Database shows that the number of coastal
natural disasters in Thailand has increased in both the frequency of occurrence and
in the number of events per year (see Figure 5). Over 1975–87, Thailand experienced
on average 0.54 coastal natural disasters per year, whereas between 1987–2004 the
incidence increased to 1.83 disasters per year. Thus, a recent World Bank report
identified the coastal and delta areas of Thailand as potentially high fatality (more
than 1000 deaths per event) and other damage ‘hotspots’ at risk from storm surge
events (Dilley et al., 2005, pp. 101–3).
The EM-DAT database also calculates the economic damage incurred per event.
Figure 6 plots the damages per coastal natural disaster in Thailand for 1975–2004.
The 2004 Asian tsunami with estimated damages of US$240 million (1996 prices)
VALUING ECOSYSTEM SERVICES 203
Figure 5. Coastal natural disasters in Thailand, 1975–2004
Notes: Over 1975–2004, coastal natural disasters included wave/surge (tsunami and tidal wave), wind storm
(cyclone/typhoon and tropical storm) and flood (significant rise of water level in coastal region. In order for
EM-DAT (2005) to record an event as a disaster, at least one or more of the following criteria must be fulfilled:
10 or more people reported killed; 100 people reported affected; declaration of a state of emergency; call for
international assistance.
Source: EM-DAT (2005). EM-DAT: The OFDA/CRED International Disaster Database. www.em-dat.net –
Université Catholique de Louvain, Brussels, Belgium.
Figure 6. Real damages per coastal natural disaster in Thailand, 1975–2004
Notes: The EM-DAT (2005) estimate of the economic impact of a disaster usually consists of direct (e.g. damage
to infrastructure, crops, housing) and indirect (e.g. loss of revenues, unemployment, market destabilization)
consequences on the local economy. However, the estimate of ‘zero’ economic damages may indicate that no
economic damages were recorded for an event. The estimates of economic damages are in thousands of US$
and converted to 1996 prices using Thailand’s GDP deflator.
Source: EM-DAT (2005). EM-DAT: The OFDA/CRED International Disaster Database. www.em-dat.net –
Université Catholique de Louvain, Brussels, Belgium.
was not the most damaging event to occur in Thailand. In fact, although the inci-
dence of coastal damages has increased since 1987, in recent years the real damages
per event has actually declined. For example, from 1979 to 1996, the economic
damages per event were around US$190 million whereas from 1996 to 2004, real
damages per event averaged US$61 million.
In sum, over the past two decades the rise in the number and frequency of coastal
natural disasters in Thailand (Figure 5) and the simultaneous rapid decline in coastal
mangrove systems over the same period (Figure 4) is likely to be more than a
204 EDWARD B. BARBIER
coincidence. Natural disasters occur when large numbers of economic assets are
damaged or destroyed during a natural hazard event. Thus an increase in the inci-
dence of coastal disasters is likely to have two sets of causes: the first is the natural
hazards themselves – tsunamis and other storm surges, tidal waves, typhoons or
cyclones, tropical storms and floods – but the second set is the increasing vulnerability
of coastal populations, infrastructure and economic activities to being harmed or
damaged by a hazard event.9 The widespread loss of mangroves in coastal areas
of Thailand may therefore have increased the vulnerability of these areas to more
incidences of natural disasters.
4.2. Valuation of habitat-fishery linkage service
This subsection compares and contrasts the static and dynamic approaches outlined
in Section 3.3 to valuing the habitat-fishery support service of mangroves in Thai-
land. As discussed above, in Thailand the near-shore artisanal fisheries supported by
this ecological service are not optimally managed but largely open access.
To conduct the static production function analysis of the mangrove-fishery linkage,
the methodology of Section 3.3.1 is applied to the same shellfish, demersal fishery
and mangrove data over 1983–96 as in Barbier (2003). These comprise pooled time-
series and cross-sectional data over the 1983–96 period for Thailand’s artisanal and
shellfish fisheries, as well as the extent of mangrove area, corresponding to the five
coastal zones along the Gulf and Thailand and Indian Ocean (Andaman Sea). Evi-
dence from domestic fish markets in Thailand suggest that the demand for fish is fairly
inelastic, and an elasticity of −0.5 was assumed for the iso-elastic market demand function.
Thus the static analysis calculation of the marginal impact of a change in wetland
area in Equation (5) requires specifying the unknown parameters of the Cobb –Douglas
production function for the fishery, h = AE aS b. Section A1 in the appendix explains
the approach used to estimate the unknown parameters (A, a, b) of the log-linear version
of the Cobb–Douglas production function and reports the resulting preferred estimations
(see Table A1). Using these results in Equations (5) and (6) allows calculation of the
welfare impacts of mangrove deforestation on Thailand’s two artisanal fisheries. The
results are depicted in Table 3, which displays both the point estimates and the
95% confidence bounds on these estimates through use of the standard errors.
All price and cost data for the fisheries used in the welfare analysis are in 1996 real
terms.
9
This view that natural disasters should not be viewed solely as ‘acts of God’ but clearly have an important anthropogenic
component to their cause is reflected in much of the current expert opinion on natural disaster management. This is summa-
rized succinctly by Dilley et al. (2005, p. 115): ‘Hazards are not the cause of disasters. By definition, disasters involve large human
or economic losses. Hazard events that occur in unpopulated areas and are not associated with losses do not constitute disasters.
Losses are created not only by hazards, therefore, but also by the intrinsic characteristics of the exposed infrastructure, land
uses, and economic activities that cause them to be damaged or destroyed when a hazard strikes. The socioeconomic contribu-
tion to disaster causality is potentially a source of disaster reduction. Disaster losses can be reduced by reducing exposure or
vulnerability to the hazards present in a given area.’
VALUING ECOSYSTEM SERVICES 205
Table 3. Valuation of mangrove-fishery linkage service, Thailand, 1996–2004 (US$)
Production function approach Average annual mangrove loss
FAO (18.0 km2)a Thailand (3.44 km2)b
Static analysis:
Annual welfare loss 99 004 (12 704–814 504) 18 884 (2425–154 307)
Net present value 570 167 108 756
(10% discount rate) (55 331– 4 690 750) (10 563–888 657)
Net present value 527 519 100 621
(12% discount rate) (52 233 – 4 339 883) (9972–822 186)
Net present value 472 407 90 108
(15% discount rate) (48 080 –3 886 476) (9179–736 288)
Dynamic analysis:
Net present value 1 980 128 373 404
(10% discount rate) (403 899–2 390,728) (164 506–691 573)
Net present value 1 760 374 331 995
(12% discount rate) (357 462–2 104 176) (147 571–614 058)
Net present value 1 484 461 279 999
(15% discount rate) (299 411–1 747 117) (126 178–516 691)
Notes: All valuations are based on mangrove-fishery linkage impacts on artisanal shellfish and demersal fisheries
in Thailand at 1996 prices. The demand elasticity for fish is assumed to be −0.5. Figures in parentheses
represent upper and lower bound welfare estimates based on the standard errors of the estimated parameters
in each model (see Section A1 in the appendix).
a
FAO estimates from FAO (2003). 2000 and 2004 data are estimated from 1990–2000 annul average mangrove
loss of 18.0 km2.
b
Thailand estimates from various Royal Thailand Forestry Department sources reported in Aksornkoae and
Tokrisna (2004). 2000 and 2004 data are estimated from 1993–96 annual average mangrove loss of 3.44 km 2.
Sources: Author’s calculations.
As Figure 4 shows, there are two different estimates of the 1996 –2004 annual
mangrove deforestation rates in Thailand, namely the FAO estimate of 18.0 km2 and
the Royal Thai Forestry Department estimate of 3.44 km2. For the welfare impacts
arising from the FAO estimates of annual average mangrove deforestation rates in
Thailand over 1996 –2004, the static analysis suggests that the annual loss in the
habitat-fishery support service is around US$99 000 ($13 000 to 815 000 with 95%
confidence). The net present value of these losses over the entire period is between
US$0.47 and 0.57 million ($48 000 to 4.7 million with 95% confidence). For the
much lower Thailand deforestation estimates, the annual welfare loss is just under
$19 000 ($2400 to 154 000 with 95% confidence) and the net present value of these
losses over the 1996 –2004 period is US$90 000 to 108 000 ($9000 to 0.9 million with
95% confidence).
Following the methodology of Section 3.3.2, we can also apply a dynamic produc-
tion function model to mangrove-fishery linkages in Thailand. As explained in the
section, this approach involves estimating the parameters of the dynamic mangrove-
fishery model, and then using these parameters to simulate the dynamic path of the
fishery and the corresponding consumer and surplus changes resulting from
mangrove deforestation. Because there are no data on the biomass stock, Xt, for
206 EDWARD B. BARBIER
Thailand’s near-shore fisheries, the appropriate dynamic model is the version indicating
the change over time in fishing effort, Et, and catch per unit effort, ct, i.e. Equations
(10) and (11). To compare with the static analysis, we use the same shellfish,
demersal fisheries and mangrove data, as well as assume the same iso-elastic demand,
from Barbier (2003) to estimate Equations (10) and (11) (see Section A2 in the
appendix). For example, the estimated parameters in the appendix correspond to
the following parameters of the dynamic production function model: b0 = r, b1 = −r/
qα, b2 = −q, b0/(b1*b2) = α, a1 = φ, a2 = 1−φw and −(a2 − 1)/a1 = w. These estimated
parameters are then employed to simulate the dynamic effort and stock paths (9) and
(10) of each fishery, starting from an initial level of effort, catch per unit effort and
mangrove area, and assuming a constant elasticity of demand of −0.5.10 By using
1996 data as the initial starting point in the simulation, i.e. for X0, E0, S0 and h0,
the dynamic paths yield effort, stock and harvest for each subsequent year from
1996–2004.
In the base case dynamic simulation, mangrove area is held constant at 1996 levels.
Two alternative paths for stock, effort and thus harvest are then also simulated,
corresponding to the two different estimates of the 1996–2004 annual mangrove
deforestation rates in Thailand, namely the FAO estimate of 18.0 km2 and the Royal
Thai Forestry Department estimate of 3.44 km2 respectively. The resulting changes
in consumer plus producer surpluses in each year over 1996 –2004, between each
deforestation simulation and the base case, provide the estimates of the welfare
impacts of the decline in the mangrove-fishery support service. That is, the changes
in consumer and producer surplus resulting from mangrove deforestation in
each subsequent year of the simulation are discounted to obtain a net present value
estimate of the resulting welfare loss. As in the static analysis, the discount rate
is varied from 10% to 15% (see Table 3). The standard errors for the parameters
of the model estimated from Equations (10) and (11) were also used to construct
both lower and upper confidence bounds on the simulation paths, and thus also
on the welfare estimates of the impacts of deforestation on the mangrove-fishery
linkage.
The results for the dynamic mangrove-fishery linkage analysis are also depicted in
Table 3, which indicates the welfare calculations associated with both the FAO and
Thailand deforestation estimates over 1996–2004. The table reports calculations
arising from the simulations based on the point estimates of the parameters of the
dynamic mangrove-fishery model. The ranges of values indicated in parentheses for
the dynamic analysis represent the lower and upper bound confidence intervals
10
Although there are no reliable stock data for Thailand’s near-shore fisheries, the Schaefer harvesting function, ht = qEtXt,
assumed in the model allows stock to be determined from catch per unit effort for a given estimated parameter q. That is,
Xt = ct/q, where ct = ht/Et. See Schnute (1977) for further details. The procedure employed here is to use the known harvest
and effort levels, as well as the estimated parameter q, for each fishery in the initial year 1996 to estimate the initial unknown
stock level, X0. Equations (9) and (10) were then used to simulate the dynamic path for Xt and Et in the subsequent years (1997–2004),
as well as the subsequent harvest, ht = qEt Xt. The dynamic simulation approach employed here is standard for an open access
fishery model (see Bjørndal and Conrad, 1987; Clark, 1976; Homans and Wilen, 1997).
VALUING ECOSYSTEM SERVICES 207
derived from the standard errors of the estimated model parameters (see Section A2
in the appendix). If the FAO estimate of mangrove deforestation over 1996–2004 is
used, then the net present value of the welfare loss ranges from around US$1.5 to
2.0 million ($0.3 to 2.4 million in the upper and lower bound simulation estimates).
In contrast, the lower Thailand deforestation estimation for 1996 –2004 suggests that
the net present value welfare loss from reduced mangrove support for fisheries is
around US$0.28 to 0.37 million ($0.13 to 0.69 million in the upper and lower bound
simulation estimates).
The welfare estimates in Table 3 indicate that the losses in the habitat-fishery
support service caused by mangrove deforestation in Thailand over 1996–2004 are
around three times greater for the dynamic production function approach compared
to the static analysis. In addition, the confidence bounds on the welfare estimates
produced with the static analysis are significantly larger, suggesting that the static
approach yields much more variable estimates of the welfare losses. Given the disparity
in estimates between the two approaches, a legitimate question to ask is whether
or not one approach should be preferred to the other in valuing habitat-fishery linkages.
It has been argued in the literature that, on the methodological grounds, the
‘dynamic’ PF approach is more appropriate for valuing how coastal wetland habitats
support offshore fisheries because this service implies that fish populations are more
likely to be affected over time (Barbier, 2000). If this is the case, then the environ-
mental ‘input’ of mangroves serving as breeding and nursery habitat for near-shore
fisheries should be modelled as part of the growth function of the fish stock. In
contrast, the static analysis, by definition, ignores stock effects and focuses exclusively
on the impact of changes in mangrove area on fishing effort and costs in the same
period in which the habitat service changes. The comparison of the dynamic and
static analysis in the Thailand case study of mangrove-fishery linkages confirms that,
by incorporating explicitly the multi-period stock effects resulting from mangrove loss,
the dynamic model produces much larger estimates for the value of changes in the
habitat-fishery support service. Since in this case study at least these stock effects
appear to be considerable, then they are clearly an important component of the
impacts of mangrove deforestation on the habitat-fishery service in Thailand.
In sum, the Thailand case study suggest caution in using the static analysis in
preference to the dynamic production function approach in valuing the ecological
service of coastal wetlands as breeding and nursery habitat for offshore fisheries. As
Table 3 indicates, the static approach could underestimate the value of this service as
well as yield more variable estimates. This may prove misleading for policy analysis,
particularly when considering options to preserve as opposed to convert coastal
wetlands. Certainly, the perception among coastal fishing communities throughout
Thailand is that the habitat-fishery service of mangroves is vital, and local fishers in
these communities have reported substantial losses in coastal fish stocks and yields,
which they attribute to recent deforestation (Aksornkoae et al., 2004; Sathirathai
and Barbier, 2001).
208 EDWARD B. BARBIER
4.3. Valuation of storm protection service
To date, the most prevalent method of valuing the storm protection service provided
by coastal wetlands is the replacement cost approach (Chong, 2005). This paper has
suggested the use of an alternative methodology, the EDF approach. The purpose of
the following subsection is to compare and contrast both approaches, using the
Thailand case study.
Sathirathai and Barbier (2001) employed the replacement cost method to estimate
the value of coastal protection and stabilization provided by mangroves in southern
Thailand. The same approach and data will be employed here. According to the
Harbor Department of the Royal Thai Ministry of Communications and Transport,
the unit cost of constructing artificial breakwaters to prevent coastal erosion and
damages from storm surges is estimated to be US$1011 (in 1996 prices) per metre
of coastline. Based on this estimate, the authors calculate the equivalent cost of
protecting the shoreline with a 75-metre width stand of mangrove is approximately
US$13.48 per m2, or US$134 801 per ha (1996 prices). Over a 20-year period and
assuming a 10% discount rate, the annualized value of this cost amounts to $14 169
per ha. This is the ‘replacement cost’ value of the storm protection function per ha
of mangrove.
The analysis for this paper uses this replacement cost value to calculate the annual
and net present value welfare losses associated with the two mangrove deforestation
estimates for Thailand over 1996–2004. The results are depicted in Table 4.
For the FAO mangrove deforestation estimate of 18.0 km2 per year over 1996–
2004, the annual welfare loss in storm protection service is around US$25.5 million,
and the net present value of this loss over the entire period ranges from US$121.7 to
146.9 million. For the Thailand deforestation estimation of 3.4 km2 per year, the
annual welfare loss in storm protection is about US$4.9 million, and the net present
value of this loss over the entire period ranges from US$23.2 to 28 million.
Section 3.5 describes the methodology for the EDF approach to estimating the
value of the storm protection service of coastal wetlands such as mangroves. As
emphasized in the appendix, the key step to this approach is to estimate the influence
of changes in coastal wetland area on the expected incidence of economically
damaging natural disaster events. The application of the EDF approach here employs
a count data model for this purpose. The details of the estimation are contained in
Section A3 of the appendix.
The analysis for Thailand over 1979–96 shows that loss of mangrove area in
Thailand increases the expected number of economically damaging natural disasters
affecting coastal provinces. Using this estimated ‘marginal effect’ (−0.00308), it is
possible to estimate the resulting impact on expected damages of natural coastal
disasters. For example, EM-DAT (2005) data show that over 1979–96 the estimated
real economic damages per coastal event per year in Thailand averaged around
US$189.9 million (1996 prices). This suggests that the marginal effect of a one-km2
VALUING ECOSYSTEM SERVICES 209
Table 4. Valuation of storm protection service, Thailand, 1996–2004 (US$)
Valuation approach Average annual mangrove loss
FAO (18.0 km2)a Thailand (3.44 km2)b
Replacement cost method:c
Annual welfare loss 25 504 821 4 869 720
Net present value (10% discount rate) 146 882 870 28 044 836
Net present value (12% discount rate) 135 896 056 25 947 087
Net present value (15% discount rate) 121 698 392 23 236 280
Expected damage function approach:
Annual welfare loss 3 382 169 645 769
( 2 341 686–5 797 339) (447 106–1 106 905)
Net present value 19 477 994 3 718 998
(10% discount rate) (13 485 827–33 387 014) (2 574 894–6 374 694)
Net present value 18 021 043 3 440 818
(12% discount rate) (12 477 089–30 889 671) (2 382 292–5 897 868)
Net present value 16 138 305 3 081 340
(15% discount rate) (11 173 553–27 662 490) (2 133 404–5 281 692)
Notes: Figures in parentheses represent upper and lower bound welfare estimates based on the 95% confidence
interval for the estimated coefficients in the model (see Section A3 in the appendix).
a
FAO estimates from FAO (2003). 2000 and 2004 data are estimated from 1990–2000 annual average mangrove
loss of 18.0 km2.
b
Thailand estimates from various Royal Thailand Forestry Department sources reported in Aksornkoae and
Tokrisna (2004). 2000 and 2004 data are estimated from 1993–96 annual average mangrove loss of 3.44 km 2
c
Re-calculated based on Sathirathai and Barbier (2001).
Sources: Author’s calculations.
loss of mangrove area is an increase in expected storm damages of about US$585 000
per km2. In Table 4, this latter calculation is combined with the FAO and Thailand
estimates of the average annual rates of deforestation to compute the welfare losses
in storm protection service for Thailand over 1996–2004. The table shows the wel-
fare calculations based both on the point estimates of the count data regression and
on using the standard errors to construct 95% confidence bounds on these estimates.
Table 4 shows that, for the FAO mangrove deforestation estimate of 18.0 km2 per
year over 1996–2004, the EDF approach estimates the annual welfare loss in storm
protection service to be around US$3.4 million ($2.3 to 5.8 million with 95% confi-
dence), and the net present value of this loss over the entire period ranges from
US$16.1 to 19.5 million ($11.2 to 33.4 million with 95% confidence). For the Thai-
land deforestation estimation of 3.4 km2 per year, the annual welfare loss in storm
protection is over US$0.65 million ($0.45 to 1.1 million with 95% confidence), and
the net present value of this loss over the entire period ranges from US$3.1 to 3.7
million ($2.1 to 6.4 million with 95% confidence).
Comparing the EDF approach and the replacement cost method of estimating the
welfare impacts of a loss of the storm protection service due to mangrove deforestation
confirms that the replacement cost method tends to produce extremely high
estimates – almost 4 times greater than even the largest upper-bound estimate
210 EDWARD B. BARBIER
calculated using the EDF approach. This suggests that the replacement cost method
should be used with caution, and when data are available, the EDF approach may
provide more reliable values of the storm protection service of coastal wetlands.
4.4. Land use policy implications
Valuation of the ecosystem services provided by mangroves are important for two
land use policy decisions in Thailand. First, although declining in recent years, con-
version of remaining mangroves to shrimp farm ponds and other commercial coastal
developments continues to be a major threat to Thailand’s remaining mangrove
areas. Second, since the December 2004 tsunami disaster, there is now considerable
interest in rehabilitating and restoring mangrove ecosystems as ‘natural barriers’ to
future coastal storm events.
To illustrate how improved and more accurate valuation of ecosystems can help
inform these two policy decisions, Table 5 compares the per hectare net returns to
shrimp farming, the costs of mangrove rehabilitation and the value of mangrove
services. All land uses are assumed to be instigated over 1996–2004 and are valued
in 1996 US dollars. The net economic returns to shrimp farming are based on non-
declining yields over a 5-year period of investment, with the pond abandoned in
subsequent years (Sathirathai and Barbier, 2001). These returns to shrimp aquaculture
are estimated to be $1078 to $1220 per ha. In comparison, the costs rehabilitating
mangrove ecosystems on land that has been converted to shrimp farms and then
abandoned are $8812 to $9318 per ha. Thus valuing the goods and services of
mangrove ecosystems can help to address two important policy questions: Do the net
economic returns to shrimp farming justify further mangrove conversion to this economic
activity, and is it worth investing in mangrove replanting and ecosystem rehabilitation
in abandoned shrimp farm areas?
As indicated in Table 5, if the older methods of valuing habitat-fishery linkages
with the static approach and storm protection with the replacement cost method are
employed, then mangrove ecosystem benefits are considerably higher than the net
economic returns to shrimp farming and the costs of replanting and rehabilitating
mangroves in abandoned farm areas. However, the static analysis undervalues the
habitat-fishery linkage of mangroves whereas the replacement cost method over-
inflates storm protection. The replacement cost method estimates storm protection
at $67 610 to 81 602 per ha, which is 99% of the value of all mangrove ecosystem
benefits. In contrast, the net income to local coastal communities from collected forest
products and the value of habitat-fishery linkages total to only $730 to $881 per ha,
which suggests that these two benefits of mangroves are insufficient on their own to
justify either halting conversion to shrimp farms or replanting and rehabilitating these
ecosystems on abandoned pond land.
If improved methods of valuing habitat-fishery linkages by the dynamic approach
and storm protection by the expected damage function method are employed, then
VALUING ECOSYSTEM SERVICES 211
Table 5. Comparison of land use values per hectare, Thailand, 1996–2004 (US$)
Land use Net present value per hectare
(10–15% discount rate)
Shrimp farming:a
Net economic returns 1078–1220
Mangrove ecosystem rehabilitation:b
Total cost 8812–9318
Ecosystem goods and services Older methods Improved methods
Net income from collected forest productsc 484–584 484–584
246–297d 708–987e
Habitat-fishery linkage
67 610–81 602f 8966–10 821g
Storm protection service
Total 68 341–82 484 10 158–12 392
a
Based on Sathirathai and Barbier (2001), updated to 1996 US$.
b
Based on costs of rehabilitating abandoned shrimp farms, replanting mangrove forests and maintaining and
protecting mangrove seedlings. From Sathirathai and Barbier (2001), updated to 1996 US$.
c
Based on Sathirathai and Barbier (2001), updated to 1996 US$.
d
Based on marginal value per ha based on the static analysis of this study (see Section 4.2).
e
Based on average value per hectare over 1996–2004 based on the dynamic analysis of this study and assuming
the estimated Thailand deforestation rate of 3.44 sq km per year (see Section 4.2).
f
Based on average value per hectare of replacement cost method of this study (see Section 4.3).
g
Based on marginal value per hectare of expected damage function approach of this study (see Section 4.3).
Sources: Author’s calculations.
the outcome is somewhat different. Although the total value of mangrove ecosystem
services is lowered to $10 158 to $12 392 per ha, it still exceeds the net economic
returns to shrimp farming. Storm protection service is still the largest benefit of
mangroves, but it no longer dominates the land use value comparison. The net
income to local coastal communities from collected forest products and the value of
habitat-fishery linkages total to $1192 to $1571 per ha, which now are greater than
the net economic returns to shrimp farming. The value of the storm protection,
however, is critical to the decision as to whether or not to replant and rehabilitate
mangrove ecosystems in abandoned pond areas. As shown in Table 5, storm protec-
tion benefits make mangrove rehabilitation an economically feasible land use option.
5. CONCLUSIONS
The case study of valuing mangroves in Thailand illustrates the potential use of the
PF approach to modelling key ecological regulatory and habitat services. The study
also indicates the importance of choosing the appropriate PF method for modelling
the key ecological-economic linkages underlying each service.
For example, the case study confirms that, if coastal wetlands such as mangroves
serve as a breeding and nursery habitat for a variety of near-shore fisheries, then it
seems more appropriate to model this environmental input as part of the growth
function of the fish stock. In comparison, not accounting for the stock effects of a
change in coastal nursery and breeding grounds may lead to an underestimation of
the value of this habitat-fishery linkage. The case study also illustrates how the EDF
212 EDWARD B. BARBIER
approach can be applied to valuing the storm protection service provided by mangroves,
and demonstrates why this method should be preferred to the less-reliable replacement
cost method, which has been used extensively in the literature to date (Chong, 2005).
The case study also points to some important policy implications for Thailand. In
recent decades, considerable mangrove deforestation has taken place in Thailand, mainly
as a result of shrimp farm expansion and other coastal economic developments (see Figure 4).
Over this period, mangrove conversion for these development activities was system-
atically encouraged by government land use policies (Aksornkoae and Tokrisna, 2004;
Barbier, 2003; Sathirathai and Barbier, 2001). Such policies were designed without
consideration of the value of the ecological services provided by mangroves, such as
their habitat support for coastal fisheries and storm protection. The case study of valuing
these ecological services for Thailand illustrates that their benefits are significant, and
should certainly not be ignored in future mangrove land management decisions.
The case study applications in this paper of valuing coastal storm protection and
habitat services have policy implications beyond Thailand as well. Even before
Hurricanes Katrina and Rita devastated the central Gulf Coast of the United States
in 2005, the US Army Corps of Engineers had proposed a $1.1 billion multi-year
programme to slow the rate of wetland loss and restore some wetlands in coastal
Louisiana. In the aftermath of these hurricanes, the US Congress is now considering
expanding the programme substantially to a $14 billion restoration effort (Zinn,
2005). As noted in Section 4, in the wake of the 2004 Asian tsunami, mangrove
restoration projects for enhanced coastal protection are underway in many countries
throughout the region. International donor groups are also supporting mangrove
restoration projects in Asia, especially in countries and regions devastated by the
tsunami (Check, 2005). In addition, there is mounting scientific evidence that near-
shore fisheries throughout the world are undergoing rapid decline, with loss of coastal
habitat and nursery grounds for these fisheries a contributory cause (Jackson et al.,
2001; Myers and Worm, 2003). Valuing the storm protection and habitat services of
coastal wetlands, as illustrated by the Thailand case study in this paper, can therefore
play a vital role in current and future debates about the state of coastal ecosystems
worldwide and the assessment of the costs and benefits of restoring these vital ecosystems.
Thus, valuing the non-market benefits of ecological regulatory and habitat services
is becoming increasingly important in assisting policymakers to manage critical
environmental assets. However, further progress applying production function
approaches and other methods to value ecological services faces two challenges.
First, for these methods to be applied effectively to valuing ecosystem services, it is important
that the key ecological and economic relationships are well understood. Unfortunately,
our knowledge of the ecological functions, let alone the ecosystem processes and
components, underlying many of the services listed in Table 1 is still incomplete.
Second, natural ecosystems are subject to stresses, rapid change and irreversible
losses, they tend to display threshold effects and other non-linearities that are difficult
to predict, let alone model in terms of their economic impacts. These uncertainties
VALUING ECOSYSTEM SERVICES 213
can affect the estimation of values from an ex ante (‘beforehand’) perspective. The
economic valuation literature recognizes that such uncertainties create the conditions
for option values, which arise from the difference between valuation under conditions
of certainty and uncertainty (e.g., see Freeman, 2003 and Just et al., 2004). The
standard approach recommended in the literature is to estimate this additional value
separately, through various techniques to measure an option price, that is, the amount
of money that an individual will pay or must be compensated to be indifferent from
the status quo condition of the ecosystem and the new, proposed condition. However,
in practice, estimating separate option prices for unknown ecological effects is very
difficult. Determining the appropriate risk premium for vulnerable populations
exposed to the irreversible ecological losses is also proving elusive. These are problems
currently affecting all economic valuation methods of ecosystem services, and not just
the production function approach. As one review of these studies concludes: ‘Given
the imperfect knowledge of the way people value natural ecosystems and their
goods and services, and our limited understanding of the underlying ecology and
biogeochemistry of aquatic ecosystems, calculations of the value of the changes
resulting from a policy intervention will always be approximate’ (Heal et al., 2005, p. 218).
Finally, Section 3 noted recent attempts to extend the production function
approach to the ecosystem level through integrated ecological-economic modelling.
This allows the ecosystem functioning and dynamics underlying the provision of
ecological services to be modelled and can be used to value multiple rather than
single services. For example, returning to the Thailand case study, it is well known
that both coral reefs and sea grasses complement the role of mangroves in providing
both the habitat-fishery and storm protection services. Thus full modelling of the
integrated mangrove–coral reef–sea grass system could improve measurement of the
benefits of both services. As we learn more about the important ecological and
economic role played by such services, it may be relevant to develop multi-service
ecosystem modelling to understand more fully what values are lost when such
integrated coastal and marine systems are disturbed or destroyed.
Discussion
Carlo A. Favero
IGIER, Bocconi University and CEPR
The objective of the paper is to apply a production function (and expected damage)
approach to ‘valuing the environment as input’, with an application to a mangrove
ecosystem in Thailand. I shall concentrate my discussion only on the production
function approach, but the main methodological points raised are naturally extended
to the expected damage function approach.
214 EDWARD B. BARBIER
An environmental good or service essentially serves as a factor input into production
that yields utility. The fundamental problem of the empirical application is the
evaluation of the following value function:
T
Bt + j
Vt = Et ∑
(1 + r ) j
j =0
where B is the social benefits in any time period of the mangrove ecosystem, and r is
the discount rate.
There are two fundamental questions:
1. What is the appropriate discount rate?
2. How to evaluate B?
The first question is not explicitly addressed and different scenarios on the discount
rate are adopted; the production function approach is the answer to the second question.
In theory the production function approach can be described as follows:
• Specification of a dynamic intertemporal optimization problem, where one of the
constraints is the production function relating the input of interest to the measurable
output.
• Solution of the model.
• Identification and estimation (or, whenever estimation is not possible, calibration) of
the technology and preference parameters of the model and of the auxiliary parameters.
• Dynamic stochastic simulation of the model to derive Et Bt+j and the associated
confidence intervals.
In practice two alternative approaches are considered: a static one and a dynamic
one. I shall not comment on the static approach because I find this inappropriate to
the very nature of the problem at hand, which is, by definition, dynamic.
In the dynamic approach a model is postulated to determine the dynamics of the
stock of fish measured in biomass units, Xt, the fishing effort, Et, the landed fish price
per unit harvested, p, and the harvest, h. The adopted model is described as follows:
X t +1 − X t = F ( X t , St ) − h( X t , Et )
Et +1 − Et = φ[ pt ht − wt Et ]
Xt
F ( X t , St ) = rX t 1 −
α ln St
ht = qX t Et
pt = khtη
where F(Xt, St) is biological growth in the current period, which is a function of St, the
mangrove area, h(Xt, Et), harvesting is a function of the stock as well as fishing effort,
Et. Fishing effort is modelled as a partial adjustment model in which the equilibrium
value is determined by fish price per unit harvested and the unit cost of effort, w.
VALUING ECOSYSTEM SERVICES 215
The model is estimated and then simulated keeping w exogenous and taking alter-
native scenarios for S, that by consequence is taken as exogenous. Using some
assumption for the discount rate the present value of a reduction in the mangrove
area is then computed.
The results are interesting but there are a number of important questions that the
modelling strategy leaves open:
• In the dynamic model S is exogenous and no law of motion for S is specified. The
model is not capable of explaining the reduction in S that we observe in the data.
In fact, if agents were acting following this model we would have never observed
a reduction in S, because a reduction in S has only costs and no benefits.
Macroeconometricians might see the applicability of the Lucas critique to this
model as an immediate consequence of the assumption of exogeneity of S.
• Expectations do not explicitly enter the model.
• What are the costs incurred in omitting from the model the dynamics of w?
• What is the performance of this model when evaluated in sample by dynamic
simulation?
• How is uncertainty added for estimation and more importantly for dynamic
simulation? The result reported in Table 4 seems to take account of only coefficient
uncertainty while, given the modelling choices, the fluctuations in the relevant variables
not explained by the adopted model are likely to be the main source of uncertainty.
I think that the answer to this set of open questions could further enhance the
potential of the interesting methods for valuing ecosystem services very well discussed
in this paper.
Omer Moav
Hebrew University, University of London Royal Holloway, Shalem Center, and CEPR
Edward Barbier demonstrates how basic micro theory can be implemented to
estimate the value of ecological services for human welfare. In particular, two methods
are developed: the production function approach and the expected damage function
approach. Both methods utilize exogenous variation in the size of the ecosystem
service, such as the size of a mangrove forest, on a beneficial outcome. According to
the former the outcome is welfare gained from the decline in the price of a consumption
good, such as fish, that utilizes the ecosystem as an input in its production process.
In the latter it is the economic value of the reduction in damage, arising, for instance,
from storms, that is reduced by a larger ecosystem.
The theory is rather straightforward. It is the availability of the data that the
estimation depends on, and it is not clear that for most practical problems there exists
sufficient exogenous variation in the ecosystem, allowing for a reliable assessment.
Nevertheless, Barbier convincingly illustrates that despite the difficulties, these methods
have the potential to provide important information about the value of the ecosystem
216 EDWARD B. BARBIER
and thereby the value of preventing its disappearance. This information can become
critical for policymakers, and might, even if only in marginal cases, generate the
crucial political force to reverse processes of natural habitat destruction.
The value of the functions of the ecosystem include, as stated by Barbier, ‘climate
stability, maintenance of biodiversity and beneficial species, erosion control, flood
mitigation, storm protection, groundwater recharge and pollution control’. This
statement reveals another limitation of the estimation methods. It focuses on a limited
set of benefits, implying a potentially huge underestimation of the value of the
ecosystem. First, due to information problems regarding most functions, it is difficult
to identify the size of the impact and/or its welfare value. For instance, most likely
exogenous variety in the ecosystem is not sufficient to estimate its effect on climate
stability, and the welfare value of biodiversity is a question hard to answer.
Second, the cost of preserving the ecosystem – giving up the benefits of its
alternative use – is paid by the local population, while many of its services extend
beyond that. As is well known, preserving natural ecosystems is a problem with large
externalities that go beyond borders. In other words, who cares? Do we expect the
poor fisherman in Thailand, or their government, to allocate a significant weight in
its welfare function to biodiversity? In fact, it is the population of the developed world
that cares, and this population’s willingness to pay a compensating price, could be
above and beyond the benefit of the ecosystem to the local population.
A more technical comment on the estimation process regards the open access
assumption and, in particular, the implicit assumption that changes in the habitat are
sufficiently small, relative to the economy, such that the producer’s surplus is
unchanged. Welfare gains from a larger ecosystem emerge only from the reduction
in consumer goods prices. This assumption adds to the bias in the estimation,
reducing the value of the ecosystem. To see this point, suppose that prices of the
consumption good are also given (traded good in an open economy). In this case
there is zero welfare gain from preserving the environment.
A final comment about the estimation method regards the implicit assumption of
stability of the steady-state equilibrium. However, non-monotonic convergence to the
steady state might characterize the dynamics of the ecosystem. For instance, the
population of a species might converge to its steady state in oscillations, implying that
a negative shock to the ecosystem might, once it is sufficiently fragile, result in extinction
of a species rather than a proportional reduction in the size of the natural population.
Beyond the problems of estimating the direct value of the ecological services for
human welfare, lies a somewhat deeper question regarding the long-run effects of the
utilization of natural resources for the benefit of mankind in the production process.
Maintaining natural habitats and benefiting their production services, or destroying
them and benefiting from their alternative land use for agriculture, might have different
long-run consequences on demographic variables, institutional development, and, in
particular, human capital promoting institutions (e.g., public schools, loans, and child
labour regulation), and the resulting accumulation of human capital.
VALUING ECOSYSTEM SERVICES 217
Natural resources, according to many studies, are a hurdle for the process of
development, in particular the accumulation of human capital. (e.g. Gylfason, 2001).
But to the best of my knowledge, we do not know yet how to make a distinction in
that regard between an open-access preserved ecosystem and agricultural land.
Therefore, depleting resources or increasing the size of agricultural land on the
account of the ecosystem, could have a significant impact on the economy.
Moreover, the transition from an open access ecosystem into private owned farmland
might have an impact on wealth inequality, in particular inequality in the ownership
of such land. Deninger and Squire (1998) show that inequality in land ownership has
a negative impact on economic growth. Engerman and Sokoloff (2000) provide evidence
that wealth inequality, brought about oppressive institutions (e.g., restricted access to
the democratic process and to education). They argue that these institutions were
designed to maintain the political power of the elite and to preserve the existing
inequality. Galor et al. (2005) provide evidence that inequality in the ownership of
agricultural land has a negative effect on public expenditure on education, and argue
that the elite of landowners might prevent public schooling, despite the support of the
owners of capital and the working class.
On the other hand, if the destruction of the ecosystem increases farmland and
thereby possibly promoting industrial development, and if the process does not
generate large wealth inequality, the return to human capital will most likely rise.
This could trigger a process of development stemming from reduced fertility and
increased investment in education. This brings us back to the main problem of
preventing the distractions of an ecosystem: the externality. Each small economy
might be better-off destroying the ecosystem, giving rise to an inefficient equilibrium.
The analysis suggested by Barbier, could, at least, highlight the benefits of preserving
natural habitats for the local economy.
APPENDIX: APPLICATION TO THAILAND CASE STUDY
This appendix outlines the econometric estimations for valuing habitat-fishery linkages
and the storm protection service of mangroves in the Thailand case study of Section 4.
A1. Static valuation of habitat-fishery linkage
To apply the static analysis of habitat-fishery linkages of Section 3.3.1 to the Thailand
case study, it is necessary to estimate the unknown parameters (A, a, b) of the log-
linear version of the Cobb–Douglas production function:
ln hit = A0 + a ln Eit + b ln Mit + µit (A1)
where i = 1, . . . , 5 zones, t = 1, . . . , 14 years (1983–96) and A0 = ln A.
Equation (A1) was estimated using the pooled data on demersal fisheries, shellfish
and mangrove area from Barbier (2003). These were the data on harvest, hit, and
218 EDWARD B. BARBIER
effort, Eit, for Thailand’s shellfish and demersal fisheries, as well as mangrove area,
Mit across the five coastal zones of Thailand and over the years1983–96. Various
regression procedures for a pooled data set were utilized and compared, including:
(i) ordinary least squares (OLS); (ii) one- and two-way panel analysis of fixed and
random effects; and (iii) a maximum likelihood estimation by an iterated generalized
least squares (GLS) procedure for a pooled time series and cross-sectional regression,
which allows for correction of any groupwise heteroscedasticity, cross-group correlation
and common or within-group autocorrelation. Table A1 indicates the best regression
model for the shellfish and demersal fisheries respectively, and the relevant test statistics.
For demersal fisheries, the preferred model shown in Table A1 is the GLS estimation
allowing for groupwise heteroscedasticity and correcting for both cross-group and
common autocorrelation. For the panel analysis of the demersal fisheries, the likelihood
ratio tests of the null hypothesis of zero individual and time effects across all five zones
and fourteen time periods were significant, thus rejecting the null hypothesis. In
addition, the Breusch–Pagan Lagrange multiplier (LM) statistic was also significant at
the 95% confidence level for both the one-way and two-models, which suggests
rejection of the null hypothesis of zero random disturbances. The Hausman test
statistic was also significant at the 99% confidence level, suggesting that the fixed
effects specification is preferred to the random effects. However, in both the one- and
two-way fixed effects model the t-test on the estimated parameter for a in Equation
(A1) was insignificant, suggesting the null hypothesis that a = 0 cannot be rejected.
As indicated in Table A1, from the pooled time series cross-sectional GLS regression
for demersal fisheries, the likelihood ratio (LR) test statistic of the null hypothesis for
homoscedasticity based on the least squares regression was computed to be 24.64,
which is statistically significant. Although not shown in the table, the alternative Wald
test for homoscedasticity is also statistically significant and confirms rejection of the
null hypothesis. Thus the GLS model with correction of groupwise heteroscedasticity
is preferred to the OLS regression. The LM statistic of 14.43 also reported in Table
A1 for demersal fisheries is a test of the null hypothesis of zero cross-sectional correlation,
which proves to be statistically significant. Although not indicated in the table, the
LR test statistic for groupwise heteroscedasticity as a restriction on cross-group
correlation was estimated to be 23.26, which is also statistically significant. Thus the
null hypothesis of zero cross-group correlation in the demersal fisheries regression can
be rejected. The common autocorrelation coefficient across all five zones was estimated
to be 0.484, and as shown in Table A1, once the GLS model for demersal fish was
corrected for this common autocorrelation, the null hypothesis that the coefficient a
= 0 is now rejected.
For shellfish, as indicated in Table A1 the preferred estimation of Equation (A1) is
the GLS estimation allowing for groupwise heteroscedasticity and correcting for
cross-group correlation, with A0 restricted to zero. For the panel analysis of shellfish,
the likelihood ratio tests and Breusch–Pagan LM tests of the null hypothesis of no
individual and time effects were significant, thus rejecting the null hypothesis. The
VALUING ECOSYSTEM SERVICES 219
Table A1. Estimates of Equation (A1) for Thailand’s shellfish and demersal
fisheries
Demersal fisherya Shellfish fisheryb
Coefficient
A0 11.213 (24.568)** –
A 0.341 (4.992)** 1.688 (38.254)**
B 0.100 (2.763)** 0.196 (3.693)**
Log-likelihoodc 5.401 –71.517
Likelihood ratio statisticd 24.643** 35.076**
Lagrange multiplier statistice 14.426* 21.304**
Notes: t-statistics are shown in parentheses.
a
Preferred model is groupwise heteroscedastic and correlated GLS, corrected for common autocorrelation.
b
Preferred model is groupwise heteroscedastic and correlated GLS, with A0 restricted to zero.
In the demersal fishery regression, correction of cross-group correlation Cov[eit,ejt] = σij leads to a positive
c
log-likelihood.
d
Tests the null hypothesis of homoscedasticity based on OLS.
e
Tests the null hypothesis of zero cross-group correlation based on OLS.
* Significant at 95% confidence level.
** Significant at 99% confidence level.
Sources: Author’s estimations.
Hausman test statistic was significant, suggesting that the fixed effects specification is
preferred to the random effects. However, in both the one- and two-way fixed effects
model the t-test on the estimated parameters for a and b in Equation (A1) was
insignificant, suggesting the null hypothesis a = b = 0 cannot be rejected. As indicated
in Table A1, from the pooled time series cross-sectional GLS regression of shellfish,
the LR test statistic of the null hypothesis for homoscedasticity based on the least
squares regression is 35.08, which is statistically significant. Although not shown in
the table, the alternative Wald test for homoscedasticity is also statistically significant
and confirms rejection of the null hypothesis. Thus the GLS model with correction
of groupwise heteroscedasticity is preferred to the OLS regression. The LM statistic
of 21.30 also reported in Table A1 for the shellfish regression is a test of the null
hypothesis of zero cross-sectional correlation, which proves to be statistically signifi-
cant. Although not indicated in the table, the LR test statistic for groupwise hetero-
scedasticity as a restriction on cross-group correlation was estimated to be 43.90,
which is also statistically significant. Thus the null hypothesis of zero cross-group
correlation in the shellfish regression can be rejected. As shown in Table A1, once
the GLS model of shellfish was corrected for groupwise heteroscedasticity and corre-
lation, the null hypotheses that a = 0 and b = 0 are now rejected.
The estimations of Equation (A1) for Thailand’s shellfish and demersal fisheries
were used in conditions (5) and (6) to calculate the welfare impacts of mangrove
deforestation over 1996–2004 on Thailand’s artisanal fisheries. The analysis uses the
same iso-elastic demand function as in Barbier (2003), with a demand elasticity, ε,
of −0.5. The results are reported in Table 3, which shows welfare calculations for
both the point estimates and upper and lower bounds on these estimates based on
the standard errors of the regression coefficients reported in Table A1.
220 EDWARD B. BARBIER
A2. Dynamic valuation of habitat-fishery linkage
The dynamic habitat-fishery modelling approach to valuing the habitat-fishery link-
age is outlined in Section 3.3.2. The main difficulty in applying this approach to
valuing mangrove-fishery linkages in Thailand is that data do not exist for the bio-
mass stock, Xt, of near-shore fisheries. Thus the appropriate system of equations to
estimate comprises (10) and (11). Because Et and ct are predetermined, both of these
equations can be estimated independently (Homans and Wilen, 1997). For both the
shellfish and demersal fisheries, the estimated equations are:
Eit = a0 + a1Rit −1 + a2 Eit −1 + µit −1 (A2)
cit − cit −1 c
= b0 + b1 it −1 + b2 Eit −1 + µit −1, (A3)
cit −1 ln M it −1
where i = 1, . . . , 5 zones, t − 1 = 1, . . . , 13 years (1983–95), Rit −1 = khit+η , a1 = φ, a2 =
1
−1
(1 − φw), b0 = r, b1 = −r/αq and b2 = −q. Both equations were estimated using the
pooled data on demersal fisheries, shellfish and mangrove area from Barbier (2003).
These were the data on harvest, hit−1, and effort, Eit−1, for Thailand’s shellfish and
demersal fisheries, as well as mangrove area, Mit−1 across the five coastal zones of
Thailand and over the years 1983–96. In addition, to calculate Rit−1 from hit−1 the
elasticity η = 1/ε = −2 was assumed as in the static analysis. Various regression
procedures for a pooled data set were utilized and compared, including: (i) OLS; (ii)
one- and two-way panel analysis of fixed and random effects; and (iii) a maximum
likelihood estimation by an iterated GLS procedure for a pooled time series and
cross-sectional regression, which allows for correction of any groupwise heteroscedas-
ticity, cross-group correlation and common or within-group autocorrelation. Tables A2
and A3 indicate the best regression models of Equations (A2) and (A3) for the shellfish
and demersal fisheries respectively, and the relevant test statistics.
For demersal fisheries, the preferred model for the effort equation (A2) is the GLS
estimation allowing for groupwise heteroscedasticity and corrected for common
autocorrelation. For the panel analysis, the likelihood ratio and Breusch–Pagan LM
tests of the null hypothesis of no individual and time effects were not significant; thus,
the null hypothesis cannot be rejected. However, as indicated in Table A2, from the
pooled time series cross-sectional regression of Equation (A2) for demersal fisheries,
the LR test statistic of the null hypothesis of homoscedasticity based on the OLS
regression is computed to be 93.22, which is statistically significant. Although not
shown in the table, the alternative Wald test for homoscedasticity is also statistically
significant and confirms rejection of the null hypothesis. Thus the GLS model with
correction of groupwise heteroscedasticity is preferred to the OLS regression. The
test statistics for the null hypothesis of zero cross-group correlation are mixed. The
LM statistic of 12.24 indicated in Table A2 is significant, whereas the LR test statistic
of 12.56 is not. When the GLS regression is corrected for groupwise correlation,
VALUING ECOSYSTEM SERVICES 221
however, the constant term a0 is no longer significant. The common autocorrelation
coefficient across all five zones is estimated to be 0.242, and although slight, correc-
tion of this autocorrelation improved the overall robustness of the GLS estimation.
As shown in Table A2, the preferred model for the effort equation (A2) for shellfish
is the one-way random effects estimation corrected for heteroscedasticity. The LR
and Wald tests of the pooled time series cross-sectional regressions of Equation (A2)
for shellfish indicated that the null hypothesis of homoscedasticity can be rejected.
Thus the GLS model with correction of groupwise heteroscedasticity is preferred to
the OLS regression. However, in all versions of the GLS regression the coefficient a1
was negative and statistically insignificant. The LR test for the presence of individual
effects is statistically significant, thus rejecting the null hypothesis of no such effects,
and although not shown, the equivalent F-test of the null hypothesis is also statistically
significant. Neither the Breusch–Pagan LM test of the null hypothesis of random
provincial-level disturbances nor the Hausman test of the random versus the fixed
effects specification is statistically significant. Although these results are somewhat
contradictory, they suggest that, if individual effects are present, they are likely to be
random. The LR test and F-test of the presence of time effects is not significant,
suggesting that the one-way is preferred to the two-way specification. Correction of
heteroscedasticity improves the robustness of the one-way random effects estimation
Table A2. Estimates of Equation (A2) for Thailand’s shellfish and demersal
fisheries
Demersal fisherya Shellfish fisheryb
Coefficient
a0 22.365 (2.254)* 808.720 (2.661)**
a1 0.00004 (4.375)** 0.000003 (0.233)
a2 0.84855 (21.703)** 0.70470 (8.183)**
Log-likelihood –380.903 –520.513
Likelihood ratio statistic for 93.223** –
homoscedasticityc
Likelihood ratio statistic for correlationd 12.552 –
Lagrange multiplier statistice 12.241* –
Likelihood ratio statistic for individual effectsf – 16.285**
Breusch–Pagan Lagrange multiplier statisticg – 0.04
Hausman test statistich – 1.88
Notes: t-statistics are shown in parentheses.
a
Preferred model is groupwise heteroscedastic GLS, corrected for common autocorrelation.
b
Preferred model is one-way random effects corrected for heteroscedasticity.
c
Tests the null hypothesis of homoscedasticity based on OLS.
d
Tests the null hypothesis of zero cross-group correlation based on OLS.
e
Tests the null hypothesis of zero cross-group correlation based on OLS.
f
Tests the null hypothesis of zero individual effects.
g
Tests the null hypothesis of zero random disturbances based on OLS.
h
Tests the null hypothesis of correlation between the individual effects and the error (i.e. random effects is
preferred to fixed effects estimation).
* Significant at 95% confidence level.
** Significant at 99% confidence level.
Sources: Author’s estimations.
222 EDWARD B. BARBIER
without affecting the parameter estimates. Although not shown in the table, the
preferred model displayed a very low estimated autocorrelation of 0.022.
As indicated in Table A3, the preferred model for the growth in catch per unit
effort equation (A3) for demersal fisheries is the GLS estimation allowing for group-
wise heteroscedasticity. For the panel analysis, the LR and Breusch–Pagan LM tests
of the null hypothesis of no individual and time effects were not significant; thus, the
null hypothesis cannot be rejected. However, from the pooled time series cross-
sectional regression, both the LR and Wald test statistics of the null hypothesis of
homoscedasticity are also statistically significant. Thus the GLS model with correction of
groupwise heteroscedasticity is preferred to the OLS regression. The test statistics for
the null hypothesis of zero cross-group correlation are mixed. The LM statistic of
11.03 indicated in Table A3 is significant, whereas the LR test statistic of 18.56 is
not. However, correcting the GLS regression for groupwise correlation does not affect
the estimation significantly. Although not shown in the table, the preferred model
displayed a very low estimated autocorrelation of −0.006.
Table A3 displays the preferred model for the growth in CPE equation (A3) for
shellfish, which is the GLS estimation allowing for groupwise and correlated hetero-
scedasticity and corrected for common autocorrelation. For the panel analysis, the
LR and Breusch–Pagan LM tests of the null hypothesis of no individual and time
effects were not significant; thus, the null hypothesis cannot be rejected. However,
from the pooled time series cross-sectional regression, both the LR and Wald test
statistics of the null hypothesis of homoscedasticity are also statistically significant.
Thus the GLS model with correction of groupwise heteroscedasticity is preferred to
the OLS regression. Although the LR and LM test statistics for the null hypothesis
Table A3. Estimates of Equation (A3) for Thailand’s shellfish and demersal
fisheries
Demersal fisherya Shellfish fisheryb
Coefficient
b0 0.4896 (2.908)** 0.2997 (2.371)*
b1 –0.000187 (–2.368)* –0.000201 (–2.354)*
b2 –0.000204 (–2.637)** –0.000060 (–2.007)*
Log-likelihood –22.337 –30.350
Likelihood ratio statistic for 24.627** 109.342**
homoscedasticityc
Likelihood ratio statistic for correlationd 18.235 11.434
Lagrange multiplier statistice 11.026* 8.491
Notes: t-statistics are shown in parentheses.
a
Preferred model is groupwise heteroscedastic GLS.
b
Preferred model is groupwise heteroscedastic and correlated GLS, corrected for common autocorrelation.
c
Tests the null hypothesis of homoscedasticity based on OLS.
d
Tests the null hypothesis of zero cross-group correlation based on OLS.
e
Tests the null hypothesis of zero cross-group correlation based on OLS.
* Significant at 95% confidence level.
** Significant at 99% confidence level.
Sources: Author’s estimations.
VALUING ECOSYSTEM SERVICES 223
of zero cross-group correlation are not significant, correcting the GLS regression for
groupwise correlation improves the significance confidence level of the estimated
parameter b2 from 90 to 95%. The common autocorrelation coefficient across all five
zones is estimated to be 0.147, and although slight, correction of this autocorrelation
improved the overall robustness of the GLS estimation.
Using the estimated parameters for Equations (A2) and (A3) for Thailand’s shellfish
and demersal fisheries allows simulation of the welfare impacts of mangrove deforestation
over 1996–2004 on Thailand’s artisanal fisheries. Again, the same demand function
with elasticity of −0.5 as in Barbier (2003) is employed. The results are reported in
Table 3, which shows welfare calculations for both the point estimates and upper and
lower bounds on these estimates based on the standard errors of the regression
coefficients reported in Tables A2 and A3.
A3. Expected damage function valuation of storm protection service
As discussed in Section 3.5, the key step in applying the expected damage function
approach to valuing the storm protection service of a coastal wetland such as mangroves
is to estimate how a change in mangrove area influences the expected incidence of
economically damaging natural disaster events.
Suppose that for a number of coastal regions, i = 1, . . . , N, and over a given period
of time, t = 1, . . . , T the ith coastal region could experience in any period t any number
of zit = 0, 1, 2, 3 . . . economically damaging storm event incidents. A common assumption
in count data models is that the count variable zit has a Poisson distribution, in which
case the expected number of storm events in each region per period is given by:
∂E[zit| it , xit ]
S
E[zit|sit , xit ] = λit = e α +β S +β ′x , = λit βS (A4)
i S it it
∂Sit
where as before, Sit is the area of wetlands, xit, are other factors, and αi accounts for
other possible ‘unobserved’ effects on the incidence of disasters specific to each
coastal region. Estimation of βS, along with an estimate of the conditional mean λit,
allows ∂Ζ/∂S in Equation (13) to be determined. One drawback of the Poisson dis-
tribution (Equation (A4)) is that it automatically implies ‘equidispersion’, that is, the
conditional variance of zit is also equal to λit. To test whether this is the case, the
Poisson method of estimating (A4) should be compared to other techniques, such as
the Negative Binomial model, which do not assume equidispersion in the variance of zit.
For the Thailand case study, the estimation of (A4) is:
ln E[zit|Mit, xit ] = ln λit = αi + βS Mit + β′xit + µit, (A5)
where i = 1, . . . , 21 coastal provinces, t = 1, . . . , 18 years (1979 –96). The EM-DAT
(2005) International Disaster Database contains data on the number of coastal disas-
ters occurring in Thailand since 1975 and the approximate location and date of its
impacts. From these data it is possible to determine zit, the number of economically
224 EDWARD B. BARBIER
damaging coastal natural disasters that occurred per province per year over 1979–
96. Mangrove area, Mit, is measured in terms of the annual mangrove area in square
kilometres for each of the 21 coastal provinces of Thailand over 1979–96. Two
control variables were included as the additional factors, xit, which may explain the
incidence of economically damaging coastal disasters, the population density of a
province and a yearly time trend variable. The inclusion of the population density
variable reflects the prevailing view in the natural disaster management literature that
‘hazard events that occur in unpopulated areas and are not associated with losses do
not constitute disasters’ (Dilley et al., 2005, p. 115).11 The yearly time trend was
included as a control because the number of coastal natural disasters seems to have
increased over time in Thailand (see Figure 5).12
Various regression procedures for a panel data set for count data models were
utilized and compared, including: (1) Poisson models assuming equidispersion, i.e.
equality of the conditional mean and the variance; (2) maximum likelihood estimation
of Negative Binomial models allowing for unequal dispersion; and (3) comparing
provincial to zonal fixed effects. Table A4 reports the best count data model for
estimating Equation (A5) and the relevant test statistics.
As shown in Table A4, the preferred specification of the count data model is the
Negative Binomial model with zonal fixed effects. In both the Poisson and Negative
Binomial panel models, the zonal fixed effects specification (with coastal zone 5 as
the default) is preferred to individual province effects, which is verified by LR tests of
the two specifications. Although the parameter estimates for the zonal fixed effects
are not shown, these estimated effects were significant at the 95% or 99% confidence
levels. As indicated in Table A4, two standard tests were employed for the null
hypothesis of equidispersion of the conditional mean and variance of the Poisson
specification of the count data model (Cameron and Trivedi, 1998; Greene, 2003).
Both the LM statistic and the t-test for equidispersion based on the residuals of the
Poisson regression are significant, indicating that the null hypothesis can be rejected,
and the Negative Binomial model that does not assume equidispersion is preferred
to the Poisson specification. The LR statistic reported in the table tests the null
hypothesis that the coefficients of the regressors are zero; as the statistic is significant,
the hypothesis is rejected.
The results displayed in Table A4 for the preferred model show that a change in
mangrove area has a significant influence on the incidence of coastal natural disasters
in Thailand, and with the predicted sign. The point estimate for βS indicates that a
1 km2 decline in mangrove area increases the expected number of disasters by 0.36%.
11
This view is also reflected in the criteria used in the International Disaster Database to decide which hazard events should
be recorded as ‘natural disasters’. In order for EM-DAT (2005) to record an event as a disaster, at least one or more of the
following criteria must be fulfilled: 10 or more people reported killed; 100 people reported affected; declaration of a state of
emergency; call for international assistance. The simple correlation between population density and mangrove area for the
sample is relatively low (−0.389).
12
This is a procedure recommended by Rose (1990), when such a trend effect is suspected.
VALUING ECOSYSTEM SERVICES 225
Table A4. Negative binomial estimation of Equation (A5) with zonal fixed effects
Parameter estimatea Marginal effectb
Variable
Mangrove area (Mit) –0.0036 (–4.448)** –0.0031 (–2.745)**
Population density (POPDENit) –0.0005 (–1.079) –0.0004 (–0.894)
Annual time trend (YRTRNit) 0.0781 (5.558)** 0.0669 (2.615)**
Dispersion parameter (αit) 0.0001
Estimated conditional mean (λ) 0.8559
Log-likelihood – 373.66
Lagrange multiplier statisticc 39.967**
Regression t-testd – 5.385**
Likelihood ratio statistice 74.919**
Notes: t-statistics shown in parentheses.
a
Parameter estimates for the zonal fixed effects are not shown. Zone 5 is the default and the fixed effects for
zones 1 to 4 were negative and significant at the 95% or 99% confidence levels.
Estimate of λitβ3 (see Equation (A4)).
b
c
Tests the null hypothesis of equidispersion in the Poisson model.
d
A regression-based test of the null hypothesis of equidispersion in the Poisson model.
e
Tests the null hypothesis that the restricted regression without the explanatory variables Mit, POPDENit and
YRTRNit is the preferred Negative Binomial model with zonal fixed effects.
* Significant at 95% confidence level.
** Significant at 99% confidence level.
Sources: Author’s estimations.
It is likely that the mangrove loss in Thailand, especially since the mid-1970s (see
Figure 4), has increased the expected number of economically damaging coastal
natural disasters per year. The estimated marginal effect corresponding to βS of a
change in mangrove area on coastal natural disasters (−0.0031) can be employed to
estimate the resulting impact of mangrove deforestation over 1979–96 in Thailand
on expected damages of natural coastal disasters. This is described further in Section
4.3 and shown in Table 4.
As discussed in Section 3.5, an underlying hypothesis of the expected damage
function methodology is that, if coastal wetland loss increases the incidence of natural
disaster per year, then wetland loss is also associated with increasing storm damages.
However, under certain circumstances the results of a count data model could provide a
misleading test of this null hypothesis. For instance, suppose a loss in wetland area is
associated with a change in the incidence of storms from one devastating storm to
two relatively minor storms per year. The count data model would then be inter-
preted as not providing evidence against the null that the change in the wetland area
increases expected storm damages, when what has actually happened is that total
storm damages have declined over time with wetland loss. This suggests the need for
a robustness check on the count data model, such as Equation (A5) in the Thailand
case study, to ensure that such situations do not dominate the application of the EDF
approach.
One possible robustness check is to test the null hypothesis directly; that is, are total
damages from storm events increasing with coastal wetland loss? In the Thailand case
study, the relevant estimation is
226 EDWARD B. BARBIER
Dit = αi + βS Mit + β′xit + µit (A6)
where the dependent variable, Dit, is total real damages from all storm events per
province per year over 1979–96. The EM-DAT (2005) database provides data on the
total economic damages per province per year in Thailand, and these data were
deflated using the 1996 GDP deflator. The standard regression procedures for the
panel analysis of Equation (A6) were performed, including comparing OLS with
fixed and random effects. Table A5 reports the OLS and random effects specifica-
tions for the preferred version of Equation (A6).
The preferred model in Table A5 is the pooled weighted least squares estimation
with correction for heteroscedasticity. The LR test of the null hypothesis of zero
individual effects across all 21 provinces is not statistically significant. Although not
shown in the table, an alternative F-test of the null hypothesis is also not significant.
Neither the Breusch–Pagan LM test of the null hypothesis of random provincial-level
disturbances nor the Hausman test of the random versus the fixed effects specification
is statistically significant. These tests confirm that in the panel analysis of Equation
(A5) of the weighted OLS regression is more efficient than either the random or fixed
effects models.
The weighted least squares regression in Table A5 indicates that, over 1979–96
and across the 21 coastal provinces of Thailand, total real storm damages increased
with mangrove loss. The point estimate suggests that a 1 km2 decline in mangrove
area increases real storm damages by around $52 per province per year. The regres-
sion also confirms that, for the Thailand case study, the null hypothesis that storm
damages increase with mangrove loss cannot be rejected.
Table A5. Panel estimation of Equation (A6) for total storm damages, Thailand
Variablea Pooled OLSb Random effectsb
Mangrove area (Mit) – 51.527 (–1.976)* –52.378 (–1.563)
Population density (POPDENit) – 12.896 (–0.343) –18.723 (–0.395)
Annual time trend (YRTRNit) 965.325 (2.058)* 983.653 (2.100)*
Constant 1 3748.820 (1.275) 1 4728.707 (1.153)
Log-likelihood –4598.425
Likelihood ratio statisticc 14.173
Lagrange multiplier statisticd 2.24
Hausman teste 1.04
Notes: t-statistics shown in parentheses.
a
Parameter estimates for the zonal fixed effects for Zone 1 and Zone 4 are not shown. Although neither
parameter was statistically significant, their inclusion improved the robustness of the overall regression.
b
Weighted least squares with robust covariance matrix to correct for heteroscedasticity.
c
Tests the null hypothesis of no fixed provincial effects.
d
Tests the null hypothesis of no random provincial effects.
e
Tests the null hypothesis that the random effects specification is preferred to the fixed effects. Test was
performed excluding the zonal fixed effect for Zone 4.
* Significant at 95% confidence level.
Sources: Author’s estimations.
VALUING ECOSYSTEM SERVICES 227
REFERENCES
Acharya, G. and E.B. Barbier (2000). ‘Valuing groundwater recharge through agricultural
production in the Hadejia-Jama’are wetlands in northern Nigeria’, Agricultural Economics, 22,
247–59.
Aksornkoae, S. and R. Tokrisna (2004). ‘Overview of shrimp farming and mangrove loss in
Thailand’, in E.B. Barbier and S. Sathirathai (eds.), Shrimp Farming and Mangrove Loss in
Thailand, Edward Elgar, London.
Aksornkoae, S., R. Tokrisna, W. Sugunnasil and S. Sathirathai (2004). ‘The importance of
mangroves: Ecological perspectives and socio-economic values’, in E.B. Barbier and S.
Sathirathai (eds.), Shrimp Farming and Mangrove Loss in Thailand, Edward Elgar, London.
Barbier, E.B. (1994). ‘Valuing environmental functions: Tropical wetlands’, Land Economics,
70(2), 155–73.
— (2000). ‘Valuing the environment as input: Applications to mangrove-fishery linkages’,
Ecological Economics, 35, 47–61.
— (2003). ‘Habitat-fishery linkages and mangrove loss in Thailand’, Contemporary Economic
Policy, 21(1), 59–77.
Barbier, E.B. and M. Cox (2003). ‘Does economic development lead to mangrove loss? A
cross-country analysis’, Contemporary Economic Policy, 21(4), 418–32.
Barbier, E.B. and S. Sathirathai (eds.) (2004). Shrimp Farming and Mangrove Loss in Thailand.
Edward Elgar, London.
Barbier, E.B. and I. Strand (1998). ‘Valuing mangrove-fishery linkages: A case study of
Campeche, Mexico’, Environmental and Resource Economics, 12, 151–66.
Barbier, E.B., I. Strand and S. Sathirathai (2002). ‘Do open access conditions affect the
valuation of an externality? Estimating the welfare effects of mangrove-fishery linkages’,
Environmental and Resource Economics, 21(4), 343–67.
Bjørndal, T. and J.M. Conrad (1987). ‘The dynamics of an open access fishery’, Canadian
Journal of Economics, 20, 74–85.
Boscolo, M. and J.R. Vincent (2003). ‘Nonconvexities in the production of timber, biodiversity,
and carbon sequestration’, Journal of Environmental Economics and Management, 46, 251–68.
Cameron, C.A. and P. Trivedi (1998). Regression Analysis of Count Data, Cambridge University
Press, Cambridge.
Carlsson, F., P. Frykblom and C. Lijenstolpe (2003). ‘Valuing wetland attributes: An applica-
tion of choice experiments’, Ecological Economics, 47, 95–103.
Carpenter, S.R., D. Ludwig and W.A. Brock (1999). ‘Management of eutrophication for lakes
subject to potentially irreversible change’, Ecological Applications, 9(3), 751–71.
Charuppat, T. and J. Charuppat (1997). The Use of Landsat-5 (TM) Satellite Images for Tracing the
Changes of Mangrove Forest Areas of Thailand, Royal Forestry Department, Bangkok.
Check, E. (2005). ‘Roots of recovery’, Nature, 438, 910–11.
Chong, J. (2005). Protective Values of Mangrove and Coral Ecosystems: A Review of Methods and Evidence,
IUCN, Gland, Switzerland.
Clark, C. (1976). Mathematical Bioeconomics, John Wiley and Sons, New York.
Dahdouh-Guebas, F., L.P. Jayatissa, D. Di Nitto, J.O. Bosire, D. Lo Seen and N. Koedam (2005). ‘How
effective were mangroves as a defence against the recent tsunami?’ Current Biology, 15(12), 443–47.
Daily, G. (ed.) (1997). Nature’s Services: Societal Dependence on Natural Ecosystems, Island Press,
Washington DC.
De Groot, R.S., M.A. Wilson and R.M.J. Boumans (2002). ‘A typology for the classification,
description and valuation of ecosystem functions, goods and services’, Ecological Economics,
41, 393–408.
Deninger, K. and L. Squire (1998). ‘New ways of looking at old issues: Inequality and growth’,
Journal of Development Economics, 57, 259–87.
Dickie, M. (2003). ‘Defensive behavior and damage cost methods’, in P. Champ, K.J. Boyle
and T.C. Brown (eds.), A Primer on Nonmarket Valuation, Kluwer, Boston, MA.
Dilley, M., R.S. Chen, U. Deichmann, A.L. Lerner-Lam and M. Arnold (2005). Natural Disaster
Hotspots: A Global Risk Analysis. Disaster Risk Management Series, Hazard Management
Unit, The World Bank, Washington, DC.
Ellis, G.M. and A.C. Fisher (1987). ‘Valuing the environment as input’, Journal of Environmental
Management, 25, 149–56.
228 EDWARD B. BARBIER
EM-DAT (2005). EM-DAT: The OFDA/CRED International Disaster Database. www.em-dat.net,
Université Catholique de Louvain, Brussels, Belgium.
Engerman, S. and Sokoloff, K.L. (2000). ‘Factor endowment, inequality, and paths of development
among new world economies’, UCLA.
FAO (2003). ‘Status and trends in mangrove area extent worldwide’ (by M.L. Wilkie and S.
Fortuna) Forest Resources Assessment Working Paper No. 63, Forest Resources Division,
Food and Agricultural Organization of the United Nations, Rome.
Finnoff, D. and J. Tschirhart (2003). ‘Harvesting in an eight-species ecosystem’, Journal of
Environmental Economics and Management, 45, 589–611.
Freeman, A.M. III. (1982). Air and Water Pollution Control: A Benefit-Cost Assessment, John Wiley,
New York.
— (1991). ‘Valuing environmental resources under alternative management regimes’, Ecological
Economics, 3, 247–56.
— (2003). The Measurement of Environmental and Resource Values: Theory and Methods, 2nd edn,
Resources for the Future, Washington DC.
Galor, O., O. Moav and D. Vollrath (2005). ‘Inequality in land ownership, the emergence of
human capital promoting institutions, and the great divergence’, Brown University.
Greene, W.H. (2003). Econometric Analysis, 5th edn, Prentice-Hall, Englewood Cliffs, NJ.
Gren, I-M., K. Elofsson and P. Jannke (1997). ‘Cost-effective nutrient reductions to the Baltic
Sea’, Environmental and Resource Economics, 10, 341–62.
Gylfason, T. (2001). ‘Natural resources, education, and economic development’, European
Economic Review, 45, 847–59.
Harakunarak, A. and S. Aksornkoae (2005). ‘Life-saving belts: Post-tsunami reassessment of
mangrove ecosystem values and management in Thailand’, Tropical Coasts, July, 48–55.
Heal, G.M., E.B. Barbier, K.J. Boyle, A.P. Covich, S.P. Gloss, C.H. Hershner, J.P. Hoehn,
C.M. Pringle, S. Polasky, K. Segerson and K. Shrader-Frechette (2005). Valuing Ecosystem Services:
Toward Better Environmental Decision Making, The National Academies Press, Washington DC.
Hirshleifer, J. and J.G. Riley (1992). The Analytics of Uncertainty and Information, Cambridge
University Press, Cambridge.
Homans, F.R. and J.E. Wilen (1997). ‘A model of regulated open access resource use’, Journal
of Environmental Economics and Management, 32, 1–21.
Jackson, J.B.C., M.X. Kirby, W.H. Berger, et al. (2001). ‘Historical overfishing and the recent
collapse of coastal ecosystems’, Science, 293, 629–38.
Just, R.E., D.L. Hueth and A. Schmitz (2004). The Welfare Economics of Public Policy: A Practical
Approach to Project and Policy Evaluation, Edward Elgar, Cheltenham, UK.
Kahn, J.R., and W.M. Kemp (1985). ‘Economic losses associated with the degradation of an
ecosystem: The case of submerged aquatic vegetation in Chesapeake Bay’, Journal of Envi-
ronmental Economics and Management, 12, 246–63.
Kaiser, B. and J. Roumasset (2002). ‘Valuing indirect ecosystem services: The case of tropical
watersheds’, Environment and Development Economics, 7, 701–14.
Kaosa-ard, M. and S. Pednekar (1998). ‘Background Report for the Thai Marine Rehabilitation
Plan 1997–2001’, Report submitted to the Joint Research Centre of the Commission of the
European Communities and the Department of Fisheries, Ministry of Agriculture and
Cooperatives, Thailand, Thailand Development Research Institute, Bangkok.
Kathiresan, K. and N. Rajendran (2005). ‘Coastal mangrove forests mitigated tsunami’,
Estuarine Coastal and Shelf Science, 65, 601–606.
Knowler, D., E.B. Barbier, and I. Strand (2001). ‘An open-access model of fisheries and
nutrient enrichment in the Black Sea’, Marine Resource Economics, 16, 195–217.
Krutilla, J.V. and A.C. Fisher (1985). The Economics of Natural Environments: Studies in the Valuation
of Commodity and Amenity Resources, Resources for the Future, Washington, DC.
Lynne, G.D., P. Conroy and F.J. Prochaska (1981). ‘Economic value of marsh areas for marine
production processes’, Journal of Environmental Economics and Management, 8, 175–86.
Mardle, S., C. James, C. Pipitone and M. Kienzle (2004). ‘Bioeconomic interactions in an
established fishing exclusion zone: The Gulf of Castellammare, NW Sicily’, Natural Resource
Modeling, 17(4), 393–447.
Massel, S.R., K. Furukawa and R.M. Brinkman (1999). ‘Surface wave propagation data in
mangrove forests’, Fluid Dynamics Research, 24, 219–49.
Mazda, Y., E. Wolanski, B. King, A. Sase, D. Ohtsuka and M. Magi (1997). ‘Drag force due
to vegetation in mangrove swamps’, Mangroves and Salt Marshes, 1, 193–99.
VALUING ECOSYSTEM SERVICES 229
McConnell, K.E. and I.E. Strand (1989). ‘Benefits from commercial fisheries when demand and
supply depend on water quality’, Journal of Environmental Economics and Management, 17, 284–92.
Michener, R. and C. Tighe (1992). ‘A Poisson regression model of highway fatalities’, American
Economic Review, 82(2), 452–56.
Millennium Ecosystem Assessment (2003). Ecosystems and Human Well-being: A Framework for
Assessment, Island Press, Washington DC.
Mitsch, W.J. and J.G. Gosselink (1993). Wetlands, 2nd edn, Van Norstrand Reinhold, New York.
Myers, R.A. and B. Worm (2003). ‘Rapid worldwide depletion of predatory fish communities’,
Nature, 423, 280–83.
Naylor, R. and M. Drew (1998). ‘Valuing mangrove resources in Kosrae, Micronesia’, Environ-
ment and Development Economics, 3, 471–90.
Olson, M.K. (2004). ‘Are novel drugs more risky for patients than less novel drugs?’ Journal of
Health Economics, 23, 1135–58.
Pagiola, S., K. von Ritter and J. Bishop (2004). How Much is an Ecosystem Worth? Assessing the
Economic Value of Conservation, The World Bank, Washington DC.
Parks, P. and M. Bonfaz (1997). ‘Nonsustainable use of renewable resources: Mangrove
deforestation and mariculture in Ecuador’, Marine Resource Economics, 9, 1–18.
Ready, R.C. (1995). ‘Environmental valuation under uncertainty’, in D.W. Bromley (ed.), The
Handbook of Environmental Economics, Blackwell Publishers, Cambridge, MA, pp. 568–93.
Ricketts, T.H., G.C. Daily, P.R. Ehrlich and C.D. Michener (2004). ‘Economic value of tropical
forest to coffee production’, Proceedings of the National Academy of Science, 101(304), 12579–82.
Rodwell, L.D., E.B. Barbier, C.M. Roberts and T.R. McClanahan (2002). ‘A model of tropical
marine reserve-fishery linkages’, Natural Resource Modeling, 15(4), 453–86.
Rose, N.L. (1990). ‘Profitability and product quality: Economic determinants of airline safety
performance’, Journal of Political Economy, 98(5), 944–64.
Sathirathai, S. and E.B. Barbier (2001). ‘Valuing mangrove conservation in Southern
Thailand’, Contemporary Economic Policy, 19, 109–22.
Schnute, J. (1977). ‘Improved estimates of the Schaefer Production Model: Theoretical
considerations’, Journal of the Fisheries Research Board of Canada, 34, 583–603.
Settle, C. and J.F. Shogren (2002). ‘Modeling native-exotic species within Yellowstone Lake’,
American Journal of Agricultural Economics, 84(5), 1323–28.
Shabman, L.A. and S.S. Batie (1978). ‘Economic value of natural coastal wetlands: A critique’,
Coastal Zone Management Journal, 4(3), 231–47.
Sugunnasil, W. (1998). ‘Fishing communities and coastal resource management in Southern
Thailand’, mimeo, Department of Social Sciences, Prince of Songkla University, Pattani, Thailand.
Sumaila, U.R. (2002). ‘Marine protected area performance in a model of a fishery’, Natural
Resource Modeling, 15(4), 439–51.
Swallow, S.K. (1994). ‘Renewable and nonrenewable resource theory applied to coastal
agriculture, forest, wetland and fishery linkages’, Marine Resource Economics, 9, 291–310.
UNEP (United Nations Environment Program) (2005). After the Tsunami: Rapid Environmental
Assessment Report, UNEP, Nairobi, 22 February; www.unep.org/tsunami/reports.
Valiela, I., J.L. Bowen and J.K. York (2001). ‘Mangrove forests: One of the world’s threatened
major tropical environments’, BioScience, 51(10), 807–15.
Wattana, S. (1998). ‘Fishing Communities and Coastal Resource Management in Southern Thai-
land’. Mimeo. Department of Social Sciences, Prince of Songkla University, Pattani, Thailand.
Wetlands International (2005). Natural Mitigation of Natural Disasters. Assessment Report to
Ramsar STRP12, Wetlands International, Jakarta, 2 February.
Winkelmann, R. (2003). Econometric Analysis of Count Data, 4th edn, Springer-Verlag, Berlin.
World Conservation Monitoring Centre (WCMC) (1992). ‘Wetlands’, in WCMC (ed.), Global
Biodiversity: Status of the Earth’s Living Resources, IUCN, Gland, Switzerland.
World Resources Institute (WRI) (1996). World Resources 1996–7, World Resources Institute,
Washington DC.
— (2001). World Resources 2000–2001. People and Ecosystems: The Fraying Web of Life, World
Resources Institute, Washington DC.
Wu, J., K. Skelton-Groth, W.G. Boggess and R.M. Adams (2003). ‘Pacific salmon restoration: Trade-offs
between economic efficiency and political acceptance’, Contemporary Economic Policy, 21(1), 78–89.
Zinn, J. (2005). Hurricanes Katrina and Rita and the Coastal Louisiana Ecosystem Restoration. CRS
Report for Congress, The Congressional Research Service, The Library of Congress,
Washington DC, 26 September.
s
sy
o
ec
ing
alu es
V
vic
SUMMARY
xxx Valuing ecosystem services asCES,B.0266-4658
VALUING ECOSYSTEMMSH,ECOP
© CEPR, productive Article
BlackwellOriginal Policy
EDWARD SERVICES
Economic inputs
Publishing UK
Oxford, Ltd
BARBIER
2007.
r
se
This paper explores two methods for valuing ecosystems by valuing the services
that they yield to various categories of user and that are not directly valued in the
market, and illustrates the usefulness of these methods with an application to the
valuation of mangrove ecosystems in Thailand. The first method is known as the
production function approach and relies on the fact that ecosystems may be inputs
into the production of other goods or services that are themselves marketed, such as
fisheries. I discuss issues that arise in measuring the input into fisheries, particularly
those due to the fact that the fishery stock is changing over time, and the shadow
value of the ecosystem consists in its contribution to the maintenance of the stock
as well as its contribution to current output. The second method is known as the
expected damage approach and is used to value the services of storm protection in
terms of the reduction in expected future storm damage that the ecosystem can
provide. These two methods are shown to yield very different valuations of ecosystems
from those that would be derived by the methods typically used in cost-benefit
analyses. I argue that they represent a significant improvement on current practice.
— Edward B. Barbier
Economic Policy January 2007 Printed in Great Britain
© CEPR, CES, MSH, 2007.
VALUING ECOSYSTEM SERVICES 179
Valuing ecosystem services as
productive inputs
Edward B. Barbier
University of Wyoming
1. INTRODUCTION
Global concern over the disappearance of natural ecosystems and habitats has
prompted policymakers to consider the ‘value of ecosystem services’ in environmental
management decisions. These ‘services’ are broadly defined as ‘the benefits people
obtain from ecosystems’ (Millennium Ecosystem Assessment, 2003, p. 53).
However, our current understanding of key ecological and economic relationships
is sufficient to value only a handful of ecological services. An important objective of
this paper is to explain and illustrate through numerical examples the difficulties
faced in valuing natural ecosystems and their services, compared to ordinary economic
or financial assets. Specifically, the paper addresses the following three questions:
1. What progress has been made in valuing ecological services for policy analysis?
2. What are the unique measurement issues that need to be overcome?
3. How can future progress improve upon the shortcomings in existing methods?
I am grateful to David Aadland, Carlo Favero, Geoff Heal, Omer Moav and three anonymous referees for helpful comments.
The Managing Editor in charge of this paper was Paul Seabright.
Economic Policy January 2007 pp. 177–229 Printed in Great Britain
© CEPR, CES, MSH, 2007.
180 EDWARD B. BARBIER
1.1. Key challenges and policy context
As a report from the US National Academy of Science has emphasized, ‘the fundamental
challenge of valuing ecosystem services lies in providing an explicit description and
adequate assessment of the links between the structure and functions of natural systems,
the benefits (i.e., goods and services) derived by humanity, and their subsequent values’
(Heal et al., 2005, p. 2). Moreover, it has been increasingly recognized by economists
and ecologists that the greatest ‘challenge’ they face is in valuing the ecosystem
services provided by a certain class of key ecosystem functions – regulatory and habitat
functions. The diverse benefits of these functions include climate stability, maintenance
of biodiversity and beneficial species, erosion control, flood mitigation, storm protection,
groundwater recharge and pollution control (see Table 1 below).
One of the natural ecosystems that has seen extensive development and application
of methods to value ecosystem services has been coastal wetlands. This paper focuses
mainly on valuation approaches applied to these systems, and in particular their role
as a nursery and breeding habitat for near-shore fisheries and in providing storm
protection for coastal communities.
The paper employs a case study of mangrove ecosystems in Thailand to compare
and contrast approaches to valuing habitat and storm protection services. Global
mangrove area has been declining rapidly, with around 35% of the total area lost in
the past two decades (Valiela et al., 2001). Mangrove deforestation has been particularly
prevalent in Thailand and other Asian countries. The main cause of global mangrove
loss has been coastal economic development, especially aquaculture expansion
(Barbier and Cox, 2003). Yet ecologists maintain that global mangrove loss is contributing
to the decline of marine fisheries and leaving many coastal areas vulnerable to
natural disasters. Concern about the deteriorating ‘storm protection’ service of
mangroves reached new significance with the 26 December 2004 Asian tsunami that
caused widespread devastation and loss of life in Thailand and other Indian Ocean
countries.
The Thailand case study also illustrates the importance of valuing ecosystem
services to policy choices. Because these services are ‘non-marketed’, their benefits
are not considered in commercial development decisions. For example, the excessive
mangrove deforestation occurring in Thailand and other countries is clearly related
to the failure to measure explicitly the values of habitat and storm protection services
of mangroves. Consequently, these benefits have been largely ignored in national land
use policy decisions, and calls to improve protection of remaining mangrove forests
and to enlist the support of local coastal communities through legal recognition of
their de facto property rights over mangroves are unlikely to succeed in the face of
coastal development pressures on these resources (Barbier and Sathirathai, 2004).
Unless the value to local coastal communities of the ecosystem services provided by
protected mangroves is estimated, it is difficult to convince policymakers in Thailand
and other countries to consider alternative land use policies.
VALUING ECOSYSTEM SERVICES 181
Thus, as the Thailand case study reveals, the challenge of valuing ecosystem
services is also a policy challenge. Because the benefits of these services are important
and should be taken into account in any future policy to manage coastal wetlands in
Thailand and other countries, it is equally essential that economics continues to
develop and improve existing methodologies to value ecological services.
1.2. Outline and main results
The paper makes three contributions. The first is to demonstrate that valuing eco-
logical services as productive inputs is a viable methodology for policy analysis, and
to illustrate the key steps through a detailed case study of mangroves in Thailand.
The second contribution is to identify the measurement issues that make valuation of
non-marketed ecosystem services a unique challenge, yet one that is important for
many important policy decisions concerning the management of natural ecosystems.
The third contribution of the paper is to show, using the examples of habitat and
storm protection services, that improvements in methods for valuing these services
can correct for some shortcomings and measurement errors, thus yielding more accu-
rate valuation estimates. But even the preferred approaches display measurement
weaknesses that need to be addressed in future developments of ecosystem valuation
methodologies.
Section 2 discusses in more detail the importance of valuing ecosystem services,
especially those arising from the regulatory and habitat functions to environmental
decision-making. Section 3 reviews various methods for valuing these services.
Because the benefits arising from ecological regulatory and habitat functions mainly
support or protect valuable economic activities, the production function (PF)
approach of valuing these benefits as environmental inputs is a promising methodology.
However, the latter approach faces its own unique measurement issues. To illus-
trate the PF approach as well as its shortcomings, the section discusses recent
advances using the examples of the habitat and storm protection services of coastal
wetland ecosystems. Section 4 compares the application of the different methods to
valuing mangroves in Thailand. The case study indicates the importance of consid-
ering the key ecological-economic linkages underlying each service in choosing the
appropriate valuation approach, and how each approach influences the final valua-
tion estimates. In the case of valuing the mangroves’ habitat-fishery linkage, model-
ling the contribution of this linkage to growth in fish stocks over time appears to be
a key consideration. The case study also demonstrates the advantages of the expected
damage function approach as an alternative to the replacement cost method of
valuing the storm protection service of coastal wetlands. Section 5 concludes the
paper by discussing the key areas for further development in ecosystem valuation
methodologies, such as incorporating the effects of irreversibilities, uncertainties and
thresholds, and the application of integrated ecological-economic modelling to reflect
multiple ecological services and their benefits. Although substantial progress has been
182 EDWARD B. BARBIER
made in valuing some ecosystem services, many difficulties still remain. Future
progress in ecosystem valuation for policy analysis requires understanding the key
flaws in existing methods that need correcting.
2. BACKGROUND: VALUATION OF ECOSYSTEM SERVICES
The rapid disappearance of many ecosystems has raised concerns about the loss
of beneficial ‘services’. This raises two important questions. What are ecosystem
services, and why is it important to value these environmental flows?
2.1. Ecosystem services
Although in the current literature the term ‘ecosystem services’ lumps together a
variety of ‘benefits’, economics normally classifies these benefits into three different
categories: (i) ‘goods’ (e.g. products obtained from ecosystems, such as resource
harvests, water and genetic material); (ii) ‘services’ (e.g. recreational and tourism
benefits or certain ecological regulatory functions, such as water purification, climate
regulation, erosion control, etc.); and (iii) cultural benefits (e.g., spiritual and religious,
heritage, etc.).1 This paper focuses on methods to value a sub-set of the second category
of ecosystem ‘benefits’ – the services arising from regulatory and habitat functions.
Table 1 provides some examples of the links between regulatory and habitat functions
and the resulting ecosystem benefits.
2.2. Valuing environmental assets
The literature on ecological services implies that natural ecosystems are assets that
produce a flow of beneficial goods and services over time. In this regard, they are no
different from any other asset in an economy, and in principle, ecosystem services
should be valued in a similar manner. That is, regardless of whether or not there
exists a market for the goods and services produced by ecosystems, their social value
must equal the discounted net present value (NPV) of these flows.
However, what makes environmental assets special is that they give rise to particular
measurement problems that are different for conventional economic or financial
assets. This is especially the case for the benefits derived from the regulatory and
habitat functions of natural ecosystems.
For one, these assets and services fall in the special category of ‘nonrenewable
resources with renewable service flows’ (Just et al., 2004, p. 603). Although a natural
ecosystem providing such beneficial services is unlikely to increase, it can be depleted,
for example through habitat destruction, land conversion, pollution impacts and so
1
See Daily (1997), De Groot et al. (2002) and Millennium Ecosystem Assessment (2003) for the various definitions of ecosystem
services that are prevalent in the ecological literature.
VALUING ECOSYSTEM SERVICES 183
Table 1. Some services provided by ecosystem regulatory and habitat functions
Ecosystem functions Ecosystem processes Ecosystem services (benefits)
and components
Regulatory functions
Gas regulation Role of ecosystems in Ultraviolet-B protection
biogeochemical processes Maintenance of air quality
Influence of climate
Climate regulation Influence of land cover and Maintenance of
biologically mediated processes temperature, precipitation
Disturbance prevention Influence of system Storm protection
structure on dampening Flood mitigation
environmental disturbance
Water regulation Role of land cover in Drainage and natural irrigation
regulating run-off, river Flood mitigation
discharge and infiltration Groundwater recharge
Soil retention Role of vegetation root matrix Maintenance of arable land
and soil biota in soil structure Prevention of damage from
erosion and siltation
Soil formation Weathering of rock and Maintenance of productivity
organic matter accumulation on arable land
Nutrient regulation Role of biota in storage Maintenance of productive
and recycling of nutrients ecosystems
Waste treatment Removal or breakdown of Pollution control and
nutrients and compounds detoxification
Habitat functions
Niche and refuge Suitable living space for Maintenance of biodiversity
wild plants and animals Maintenance of beneficial
species
Nursery and breeding Suitable reproductive Maintenance of biodiversity
habitat andnursery grounds Maintenance of beneficial
species
Sources: Adapted from Heal et al. (2005, Table 3-3) and De Groot et al. (2002).
forth. Nevertheless, if the ecosystem is left intact, then the flow services from the
ecosystem’s regulatory and habitat functions are available in quantities that are not
affected by the rate at which they are used.
In addition, whereas the services from most assets in an economy are marketed,
the benefits arising from the regulatory and habitat functions of natural ecosystems
generally are not. If the aggregate willingness to pay for these benefits is not revealed
through market outcomes, then efficient management of such ecosystem services
requires explicit methods to measure this social value (e.g., see Freeman, 2003; Just
et al., 2004). A further concern over ecosystem services is that their beneficial flows
are threatened by the widespread disappearance of natural ecosystems and habitats
across the globe. The major cause of this disappearance is conversion of the land to
other uses, degradation of the functioning and integrity of natural ecosystems through
resource exploitation, pollution, and biodiversity loss, and habitat fragmentation
(Millennium Ecosystem Assessment, 2003). The failure to measure explicitly the
aggregate willingness to pay for otherwise non-marketed ecological services exacerbates
184 EDWARD B. BARBIER
these problems, as the benefits of these services are ‘underpriced’ in development
decisions as a consequence. Population and development pressures in many areas of
the world result in increased land demand by economic activities, which mean that
the opportunity cost of maintaining the land for natural ecosystems is rarely zero.
Unless the benefits arising from ecosystem services are explicitly measured, or ‘valued’,
then these non-marketed flows are likely to be ignored in land use decisions. Only
the benefits of the ‘marketed’ outputs from economic activities, such as agricultural
crops, urban housing and other commercial uses of land, will be taken into account,
and as a consequence, excessive conversion of natural ecosystem areas for development
will occur.
A further problem is the uncertainty over their future values of environmental
assets. It is possible, for example, that the benefits of natural ecosystem services may
increase in the future as more scientific information becomes available over time. In
addition, if environmental assets are depleted irreversibly through economic develop-
ment, their value will rise relative to the value of other economic assets (Krutilla and
Fisher, 1985). Because ecosystems are in fixed supply, lack close substitutes and are
difficult to restore, their beneficial services will decline as they are converted or
degraded. As a result, the value of ecosystem services is likely to rise relative to other
goods and services in the economy. This rising, but unknown, future scarcity value of
ecosystem benefits implies an additional ‘user cost’ to any decision that leads to
irreversible conversion today.
Valuation of environmental assets under conditions of uncertainty and irreversibility
clearly poses additional measurement problems. There is now a considerable literature
advocating various methods for estimating environmental values by measuring the
additional ‘premium’ that individuals are willing to pay to avoid the uncertainty
surrounding such values (see Ready, 1995 for a review). Similar methods are also
advocated for estimating the user costs associated with irreversible development, as
this also amounts to valuing the ‘option’ of avoiding reduced future choices for indi-
viduals (Just et al., 2004). However, it is difficult to implement such methods empiri-
cally, given the uncertainty over the future state of environmental assets and about
the future preferences and income of individuals. The general conclusion from studies
that attempt to allow for such uncertainties in valuing environmental assets is that
‘more empirical research is needed to determine under what conditions we can ignore
uncertainty in benefit estimation ...where uncertainty is over economic parameters
such as prices or preferences, the issues surrounding uncertainty may be empirically
unimportant’ (Ready, 1995, p. 590).
3. VALUING THE ENVIRONMENT AS INPUT
Uncertainty and irreversible loss are important issues to consider in valuing ecosys-
tem services. However, as emphasized by Heal et al. (2005), a more ‘fundamental
challenge’ in valuing these flows is that ecosystem services are largely not marketed,
VALUING ECOSYSTEM SERVICES 185
and unless some attempt is made to value the aggregate willingness to pay for these
services, then management of natural ecosystems and their services will not be
efficient. The following section describes advances in developing the ‘production
function’ approach, compared to other valuation methods, as a means to measuring
the aggregate willingness to pay for the largely non-marketed benefits of ecosystem
services.
3.1. Methods of valuing ecosystem services
Table 2 indicates various methods that can be used for valuing ecological services.2
However, some approaches are limited to specific benefits. For example, the travel
cost method is used principally for environmental values that enhance individuals’
enjoyment of recreation and tourism, averting behaviour models are best applied to
the health effects arising from pollution, and hedonic wage and property models are
used primarily for assessing work-related hazards and environmental impacts on
property values, respectively.
In contrast, stated preference methods, which include contingent valuation methods,
conjoint analysis and choice experiments, have the potential to be used widely in
valuing ecosystem goods and services. These valuation methods involve surveying
individuals who benefit from an ecological service or range of services, and analysing
the responses to measure individuals’ willingness to pay for the service or services.
For example, choice experiments of wetland restoration in southern Sweden
revealed that individuals’ willingness to pay for the restoration increased if the result
enhanced overall biodiversity but decreased if the restored wetlands were used mainly
for the introduction of Swedish crayfish for recreational fishing (Carlsson et al., 2003).
In some cases, stated preference methods are used to elicit ‘non-use values’, that is,
the additional ‘existence’ and ‘bequest’ values that individuals attach to ensuring that
a well-functioning system will be preserved for future generations to enjoy. A contin-
gent valuation study of mangrove-dependent coastal communities in Micronesia
demonstrated that the communities ‘place some value on the existence and ecosystem
functions of mangroves over and above the value of mangroves’ marketable products’
(Naylor and Drew, 1998, p. 488).
However, to implement a stated-preference study two key conditions are necessary:
(1) the information must be available to describe the change in a natural ecosystem
in terms of service that people care about, in order to place a value on those services;
and (2) the change in the natural ecosystem must be explained in the survey instrument
in a manner that people will understand and not reject the valuation scenario (Heal
et al., 2005). For many of the services arising from ecological regulatory and habitat
2
It is beyond the scope of this paper to discuss all the valuation methods listed in Table 2. See Freeman (2003), Heal et al. (2005)
and Pagiola et al. (2004) for more discussion of these various valuation methods and their application to valuing ecosystem goods
and services.
186 EDWARD B. BARBIER
Table 2. Various valuation methods applied to ecosystem services
Valuation Types of value Common types Ecosystem
methoda estimatedb of applications services valued
Travel cost Direct use Recreation Maintenance of
beneficial species,
productive ecosystems
and biodiversity
Averting Direct use Environmental impacts Pollution control
behaviour on human health and detoxification
Hedonic Direct and Environmental impacts Storm protection; flood
price indirect use on residential property mitigation; maintenance
and human morbidity of air quality
and mortality
Production Indirect use Commercial and recreational Maintenance of beneficial
function fishing; agricultural systems; species; maintenance of
control of invasive species; arable land and agricultural
watershed protection; productivity; prevention of
damage costs avoided damage from erosion and
siltation; groundwater
recharge; drainage and
natural irrigation; storm
protection; flood mitigation
Replacement Indirect use Damage costs avoided; Drainage and natural
cost freshwater supply irrigation; storm protection;
flood mitigation
Stated Use and non-use Recreation; environmental All of the above
preference impacts on human health
and residential property;
damage costs avoided;
existence and bequest
values of preserving
ecosystems
a
See Freeman (2003), Heal et al. (2005) and Pagiola et al. (2004) for more discussion of these various valuation
methods and their application to valuing ecosystem goods and services.
b
Typically, use values involve some human ‘interaction’ with the environment whereas non-use values do not,
as they represent an individual valuing the pure ‘existence’ of a natural habitat or ecosystem or wanting to
‘bequest’ it to future generations. Direct use values refer to both consumptive and non-consumptive uses that
involve some form of direct physical interaction with environmental goods and services, such as recreational
activities, resource harvesting, drinking clean water, breathing unpolluted air and so forth. Indirect use values
refer to those ecosystem services whose values can only be measured indirectly, since they are derived from
supporting and protecting activities that have directly measurable values, such as many of the services listed in
Table 1.
Source: Adapted from Heal et al. (2005, Table 4-2) and Table 1.
functions, one or both of these conditions may not hold. For instance, it has proven
very difficult to describe accurately through the hypothetical scenarios required by
stated-preference surveys how changes in ecosystem processes and components affect
ecosystem regulatory and habitat functions and thus the specific benefits arising from
these functions that individuals value. If there is considerable scientific uncertainty
surrounding these linkages, then not only is it difficult to construct such hypothetical
scenarios but also any responses elicited from individuals from stated-preference surveys
are likely to yield inaccurate measures of their willingness to pay for ecological services.
VALUING ECOSYSTEM SERVICES 187
In contrast to stated-preference methods, the advantage of PF approaches is that
they depend on only the first condition, and not both conditions, holding. That is,
for those regulatory and habitat functions where there is sufficient scientific knowl-
edge of how these functions link to specific ecological services that support or protect
economic activities, then it may be possible to employ the PF approach to value these
services. However, PF methods have their own measurement issues and limitations.
These are also discussed further in the rest of this section, and illustrated using
examples of key ecological services from coastal and estuarine wetlands.
3.2. The production function approach
Many of the beneficial services derived from regulatory and habitat functions are
commonly classified by economists as indirect use values (Barbier, 1994). The benefits
attributed to these services arise through their support or protection of activities that
have directly measurable values (see Table 2). For example, coastal and estuarine
wetlands, such as tropical mangroves and temperate marshlands, act as ‘natural
barriers’ by preventing or mitigating storms and floods that could affect property and
land values, agriculture, fishing and drinking supplies, as well as cause sickness and
death. Similarly, coastal and estuarine wetlands may also provide a nursery and
breeding habitat that supports the productivity of near-shore fisheries, which in turn
may be valued for their commercial or recreational catch.
Because the benefits of these ecosystem services appear to enhance the productivity
of economic activities, or protect them from possible damages, one possible method
of measuring the aggregate willingness to pay for such services is to estimate their
value as if they were a factor input in these productive activities. This is the essence
of the PF valuation approaches, also called ‘valuing the environment as input’
(Barbier, 1994 and 2000; Freeman, 2003, ch. 9).3
The basic modelling approach underlying PF methods is similar to determining
the additional value of a change in the supply of any factor input. If changes in the
regulatory and habitat functions of ecosystems affect the marketed production
activities of an economy, then the effects of these changes will be transmitted to
individuals through the price system via changes in the costs and prices of final goods
and services. This means that any resulting ‘improvements in the resource base or
environmental quality’ as a result of enhanced ecosystem services, ‘lower costs and
prices and increase the quantities of marketed goods, leading to increases in consumers’
and perhaps producers’ surpluses’ (Freeman, 2003, p. 259). The sum of consumer
and producer surpluses in turn provides a measure of the willingness to pay for the
improved ecosystem services.
3
The concept of ‘valuing’ the environment as input is not new. Dose-response and change-in-productivity models, which have
been used for some time, can be considered special cases of the PF approach in which the production responses to environmen-
tal quality changes are greatly simplified (Freeman, 1982).
188 EDWARD B. BARBIER
An adaptation of the PF methodology is required in the case where ecological
regulatory and habitat functions have a protective value, such as the storm protection
and flood mitigation services provided by coastal wetlands. In such cases, the envi-
ronment may be thought of producing a non-marketed service, such as ‘protection’
of economic activity, property and even human lives, which benefits individuals
through limiting damages. Applying PF approaches requires modelling the ‘produc-
tion’ of this protection service and estimating its value as an environmental input in
terms of the expected damages avoided.
Although this paper focuses mainly on applications of the PF approach to coastal
wetland ecosystems, as Table 2 indicates PF approaches are being increasingly
employed for a diverse range of environmental quality impacts and ecosystem ser-
vices. Some examples include maintenance of biodiversity and carbon sequestration in
tropical forests (Boscolo and Vincent, 2003); nutrient reduction in the Baltic Sea
(Gren et al., 1997); pollination service of tropical forests for coffee production in Costa
Rica (Ricketts et al., 2004); tropical watershed protection services (Kaiser and
Roumasset, 2002); groundwater recharge supporting irrigation farming in Nigeria
(Acharya and Barbier, 2000); coral reef habitat support of marine fisheries in Kenya
(Rodwell et al., 2002); marine reserves acting to enhance the ‘insurance value’ of
protecting commercial fish species in Sicily (Mardle et al., 2004) and in the northeast
cod fishery (Sumaila, 2002); and nutrient enrichment in the Black Sea affecting the
balance between invasive and beneficial species (Knowler et al., 2001).
3.3. Measurement issues for modelling habitat-fishery linkages
Applying PF methods to valuing ecosystem services has its own demands in terms of
ecological and economic data. To highlight these additional measurement issues, this
section draws on the example of valuing coastal wetlands as a nursery and breeding
habitat for commercial near-shore fisheries.
First, application of the PF approach requires properly specifying the habitat-
fishery PF model that links the physical effects of the change in this service to changes
in market prices and quantities and ultimately to consumer and producer surpluses.
As with many ecological services, it is difficult to measure directly changes in the
habitat and nursery function of coastal wetlands. Instead, the standard approach
adopted in coastal habitat-fishery PF models is to allow the wetland area to serve as
a proxy for the productivity contribution of the nursery and habitat function (see
Barbier, 2000 for further discussion). It is then relatively straightforward to estimate
the impacts of the change in the coastal wetland area input on fishery catch, in terms
of the marginal costs of fishery harvests and thus changes in consumer and producer
surpluses.
Second, market conditions and regulatory policies for the marketed output will
influence the values imputed to the environmental input (Freeman, 1991). For
instance, the offshore fishery supported by coastal wetlands may be subject to open
VALUING ECOSYSTEM SERVICES 189
access. Under these conditions, profits in the fishery would be dissipated, and
equilibrium prices would be equated to average and not marginal costs. As a
consequence, there is no producer surplus, and the welfare impact of a change in
wetland habitat is measured by the resulting change in consumer surplus only.
Third, if the ecological service supports a harvested natural resource system, such
as a fishery, forestry or a wildlife population, then it may be necessary to model how
changes in the stock or biological population may affect the future flow of benefits.
If the natural resource stock effects are not considered significant, then the environ-
mental changes can be modelled as impacting only current harvest, prices and
consumer and producer surpluses. If the stock effects are significant, then a change
in an ecological service will impact not only current but also future harvest and
market outcomes. In the PF valuation literature, the first approach is referred to as a
‘static model’ of environmental change on a natural resource production system,
whereas the second approach is referred to as a ‘dynamic model’ because it takes into
account the intertemporal stock effects of the environmental change (Barbier, 2000;
Freeman, 2003, ch. 9).
Finally, most natural ecosystems provide more than one beneficial service, and it
may be important to model any trade-offs among these services as an ecosystem is
altered or disturbed. Integrated economic-ecological modelling could capture more
fully the ecosystem functioning and dynamics underlying the provision of key services,
and can be used to value multiple services arising from natural ecosystems. For
instance, integrated modelling of an entire wetland-coral reef-sea grass system could
measure simultaneously the benefits of both the habitat-fishery linkage and the storm
protection service provided by the system. Examples of such multi-service ecosystem
modelling include analysis of salmon habitat restoration (Wu et al., 2003); eutrophi-
cation of small shallow lakes (Carpenter et al., 1999); changes in species diversity in
a marine ecosystem (Finnoff and Tschirhart, 2003); and introduction of exotic trout
species (Settle and Shogren, 2002).
To illustrate the first three of the above issues, I next explore two ways of measur-
ing the welfare effects of an environmental change on a productive natural resource
system with the example of the coastal habitat-fishery linkage. I will return to the
issue of integrated ecological-economic modelling of multiple ecological services in
Section 5.
3.3.1. Habitat-fishery linkages: static approaches. This section illustrates the
use of a static model to value how a change in coastal wetland habitat area affects
the market for commercially harvested fish. Many initial PF methods to value habitat-
fishery linkages have relied on this static approach. For example, using data from the
Lynne et al. (1981), Ellis and Fisher (1987) constructed such a model to value the
support by Florida marshlands for Gulf Coast crab fisheries in terms of the resulting
changes in consumer and producer surpluses from the marketed catch. Freeman
(1991) then extended Ellis and Fisher’s approach to show how the values imputed to
190 EDWARD B. BARBIER
the wetlands in the static model is influenced by whether or not the fishery is
open access or optimally managed. Sathirathai and Barbier (2001) also used a static
model of habitat-fishery linkages to value the role of mangroves in Thailand in
supporting near-shore fisheries under both open access and optimally managed
conditions.
As most near-shore fisheries are not optimally managed but open access, the following
illustration of the static model of habitat-fishery linkages assumes that the fishery is
open access. Any profits in the fishery will attract new entrants until all the profits
disappear, and in equilibrium, the welfare change in coastal wetland is in terms of its
impact on consumer surplus only.
As noted above, the general PF approach treats an ecological service, such as
coastal wetland habitat, as an ‘input’ into the economic activity, and like any other
input, its value can be equated with its impact on the productivity of any marketed
output. More formally, if h is the marketed harvest of the fishery, then its production
function can be denoted as:
h = h( E i . . . E k , S ) (1)
The area of coastal wetlands, S, may therefore have a direct influence on the marketed
fish catch, h, which is independent from the standard inputs of a commercial fishery,
Ei . . . E k .
A standard assumption in most static habitat-fishery models is that the production
function (1) takes the Cobb–Douglas form, h = AE aS b, where E is some aggregate
measure of total effort in the off-shore fishery and S is coastal wetland habitat area.
It follows that the optimal cost function of a cost-minimizing fishery is:
C * = C(h, w, S ) = wA−1/ah1/aS −b/a (2)
where w is the unit cost of effort. Assuming an iso-elastic market demand function,
P = p(h) = khη, η = 1/ε < 0 , then the market equilibrium for catch of the open access
fishery occurs where the total revenues of the fishery just equals cost, or price equals
average cost, i.e. P = C */h, which in this model becomes:
khη = wA−1/ah1−a/aS −b/a (3)
which can be rearranged to yield the equilibrium level of fish harvest:
a/β
w
A −1/β S −b/β , β = (1 + η ) a − 1
h= (4)
k
It follows from (4) that the marginal impact of a change in wetland habitat is:
a/β
b w
dh
A −1/β S −(b+β )/β
=− (5)
βk
dS
The change in consumer surplus, CS, resulting from a change in equilibrium
harvest levels (from h0 to h1) is:
VALUING ECOSYSTEM SERVICES 191
h1
k[( h1 )η+1 – ( h 0 )η+1 ]
∆CS = − k[( h1 )η+1 – ( h 0 )η+1 ]
p( h )dh − [ p1h1 − p 0h 0 ] =
η +1 (6)
h0
η[ p h − p h ]
11 00
=− .
η +1
By utilizing (5) and (6) it is possible to estimate the new equilibrium harvest and
price levels and thus the corresponding changes in consumer surplus associated with
a change in coastal wetland area, for a given demand elasticity, γ.
Figure 1 is the diagrammatic representation of the welfare measure of a change in
wetland area on an open access fishery corresponding to Equation (6). As shown in
the figure, a change in wetland area that serves as a breeding ground and nursery for
an open access fishery results in a shift in the average cost curve, AC, of the fishery.
The welfare impact is the change in consumer surplus (area P*ABC).
3.3.2. Habitat-fishery linkages: dynamic approaches. If the stock effects of a
change in coastal wetlands are significant, then valuing such changes in terms of the
impacts on current harvest and market outcomes is a flawed approach. To overcome
this shortcoming, a dynamic model of coastal habitat-fishery linkage incorporates
the change in wetland area within a multi-period harvesting model of the fishery. The
standard approach is to model the change in coastal wetland habitat as affecting the
biological growth function of the fishery (Barbier, 2003). As a result, any value
impacts of a change in this habitat-support function can be determined in terms of
changes in the long-run equilibrium conditions of the fishery. Alternatively, the
welfare analysis could be conducted in terms of the harvesting path that approaches
this equilibrium or the path that is moving away from initial conditions in the fishery.
Figure 1. The economic value effects of increased wetland area on an open
access fishery
Notes: AC: average cost; D: demand curve; P*: price per tonne; h*: fish catch in tonnes after change; P*ABC:
change in consumer and producer surplus.
Source: Adapted from Freeman (1991).
192 EDWARD B. BARBIER
Most attempts to value habitat-fishery linkages via a dynamic model that incorporates
stock effects have assumed that the fishery affected by the habitat change is in a long-run
equilibrium. Such a model has been applied, for example, in case studies of valuing
habitat fishery linkages in Mexico (Barbier and Strand, 1998), Thailand (Barbier et
al., 2002; Barbier, 2003) and the United States (Swallow, 1994). Similar ‘equilibrium’
dynamic approaches have been used to model other coastal environmental changes,
including the impacts of water quality on fisheries in the Chesapeake Bay (Kahn and
Kemp, 1985; McConnell and Strand, 1989) and the effects of mangrove deforestation
and shrimp larvae availability on aquaculture in Ecuador (Parks and Bonifaz, 1997).
However, valuing the change in coastal wetland habitat in terms of its impact on
the long-run equilibrium of the fishery raises additional methodological issues. First,
the assumption of prevailing steady state conditions is strong, and may not be a
realistic representation of harvesting and biological growth conditions in the near-
shore fisheries. Second, such an approach ignores both the convergence of stock and
harvest to the steady state and the short-run dynamics associated with the impacts of
the change in coastal habitat on the long-run equilibrium. The usual assumption is
that this change will lead to an instantaneous adjustment of the system to a new
steady state, but this in turn requires local stability conditions that may not be supported
by the parameters of the model.
There are examples of pure fisheries models that assume that the dynamic system
is not in equilibrium but is either on the approach to a steady state or is moving away
from initial fixed conditions. The latter approach has proven particularly useful in the
case of open access or regulated access fisheries (Bjørndal and Conrad, 1987; Homans and
Wilen, 1997). The following model shows how this approach can be adopted here to
the case of valuing a change in wetland habitat in terms of the dynamic path of an
open access fishery.
Defining Xt as the stock of fish measured in biomass units, any net change in
growth of this stock over time can be represented as:
∂ 2F ∂F
X t − X t −1 = F ( X t −1, St −1 ) − h( X t −1, Et −1 ), > 0, > 0. (7)
∂X t −1 ∂St −1
2
Thus, net expansion in the fish stock occurs as a result of biological growth in the
current period, F (Xt−1, St−1), net of any harvesting, h(Xt−1, Et−1), which is a function of
the stock as well as fishing effort, Et −1. The influence of the wetland habitat area, St −1,
as a breeding ground and nursery habitat on growth of the fish stock is assumed to
be positive, ∂F/∂St−1 > 0, as an increase in wetland area will mean more carrying
capacity for the fishery and thus greater biological growth.
As before, it is assumed that the near-shore fishery is open access. The standard
assumption for an open access fishery is that effort next period will adjust in response
to the real profits made in the past period (Clark, 1976; Bjørndal and Conrad, 1987).
Letting p(h) represent landed fish price per unit harvested, w the unit cost of effort
and φ > 0 the adjustment coefficient, then the fishing effort adjustment equation is:
VALUING ECOSYSTEM SERVICES 193
∂p( ht −1 )
Et − Et −1 = φ[ p( ht −1 ) h( X t −1, Et −1 ) − wEt −1 ],< 0. (8)
∂ht −1
Assume a conventional bioeconomic fishery model with biological growth character-
ized by a logistic function, F (Xt−1, St−1) = rXt−1[1 − Xt−1/K (St−1)], and harvesting by a
Schaefer production process, ht = qXtEt, where q is a ‘catchability’ coefficient, r is the
intrinsic growth rate and K (St) = α ln St, is the impact of coastal wetland area on
carrying capacity, K, of the fishery. The market demand function for harvested fish is
again assumed to be iso-elastic, i.e. p(h) = khη, η = 1/ε < 0. Substituting these
expressions into (7) and (8) yields:
X t −1
X t = rX t −1 1 − − ht −1 + X t −1 (9)
α ln St −1
Et = φRt −1 + (1 − φw )Et −1, Rt −1 = kht1−1η .
+
(10)
Both Xt and Et are predetermined, and so (9) and (10) can be estimated independently
(see Homans and Wilen, 1997). Following Schnute (1977), define the catch per unit
effort as ct = ht/Et = qXt. If Xt is predetermined so is ct. Substituting the expression
for catch per unit effort in (9) produces:
ct − ct −1 r ct −1
=r− − qEt −1. (11)
qα ln St −1
ct −1
Thus Equations (10) and (11) can also be estimated independently to determine the
biological and economic parameters of the model. For given initial effort, harvest and
wetland data, both the effort and stock paths of the fishery can be determined for
subsequent periods, and the consumer plus producer surplus can be estimated for
each period. Alternative effort and stock paths can then be determined as wetland
area changes in each period, and thus the resulting changes in consumer plus pro-
ducer surplus in each period are the corresponding estimates of the welfare impacts
of the coastal habitat change.4
3.4. Replacement cost and cost of treatment
In circumstances where an ecological service is unique to a specific ecosystem and is
difficult to value, then economists have sometimes resorted to using the cost of replac-
ing the service as a valuation approach.5 This method is usually invoked because of
the lack of data for many services arising from natural ecosystems.
For example, the presence of a wetland may reduce the cost of municipal water
treatment because the wetland system filters and removes pollutants. It is therefore
4
As along its dynamic path the open access fishery is not in equilibrium, producer surpluses, or losses, are relevant for the
welfare estimate of a change in coastal wetland habitat.
5
Such an approach to approximating the benefits of a service by the cost of providing an alternative is not used exclusively in
environmental valuation. For example, in the health economics literature this approach is referred to as ‘cost of illness’ (Dickie, 2003).
This involves adding up the costs of treating a patient for an illness as the measure of the benefit to the patient of staying disease-free.
194 EDWARD B. BARBIER
tempting to use the cost of an alternative treatment method, such as the building and
operation of an industrial water treatment plant, to represent the value of the wetland’s
natural water treatment service. Such an approach does not measure directly the
benefit derived from the wetland’s waste treatment service; instead, the approach is
estimating this benefit with the cost of providing the ecosystem service that people
value. Herein lies the main problem with the replacement cost method: it is using
‘costs’ as a measure of economic ‘benefit’. In economic terms, the implication is that
the ratio of costs to benefit of an ecological service is always equal to one.
The problems posed by the replacement cost method are illustrated in Figure 2,
in the case of waste water treatment service provided by an existing wetland ecosys-
tem. The cost of the waste water treatment service provided by the wetlands is ‘free’
and thus corresponds to the horizontal axis, MCS. Given the demand curve for water,
Q 1 amount of water is consumed. However, if the wetland is destroyed the marginal
cost of an alternative, human-built waste treatment facility is MCH. Thus, the ‘replacement
cost’ of using the treatment facility to provide Q 1 amount of water in the absence of
the wetlands is the difference between the two supply curves, or area 0BDQ 1.
However, this overestimates the benefit of having the wetlands provide the waste
treatment service. The true benefit of this ecosystem service is the demand curve, or total
willingness to pay, for Q 1 amount of water less the costs of providing it, or area 0ACQ 1.
For these reasons, economists consider that the replacement cost approach should
be used with caution. Shabman and Batie (1978) suggested that this method can
provide a reliable valuation estimation for an ecological service if the following con-
ditions are met: (1) the alternative considered provides the same services; (2) the
alternative compared for cost comparison should be the least-cost alternative; and (3)
there should be substantial evidence that the service would be demanded by society
if it were provided by that least-cost alternative. In the absence of any information
on benefits, and a decision has to be made to take some action, then treatment costs
become a way of looking for a cost-effective action.
Figure 2. Replacement cost estimation of an ecosystem service
Source: Adapted from Ellis and Fisher (1987).
VALUING ECOSYSTEM SERVICES 195
One of the best-known examples of a policy decision based on using the ‘replace-
ment cost’ method to assess the value of an ecosystem service is the provision of clean
drinking water by the Catskills Mountains for New York City (Heal et al., 2005). In
1996, New York City faced a choice: either it could build water filtration systems to
clean its water supply or the city could restore and protect the Catskill watersheds to
ensure high-quality drinking water. Because estimates indicated that building and
operating the filtration system would cost $6–8 billion whereas protecting and restoring
the watersheds would cost $1–1.5 billion, New York chose to protect the Catskills. In
this case, it was sufficient for the policy decision simply to demonstrate the cost-
effectiveness of restoring and protecting the ecological integrity of the Catskills watersheds
compared to the alternative of the human-constructed water filtration system. Thus,
clearly this is an example where the criteria established by Shabman and Batie (1978)
apply.
The main reason why economists have resorted to replacement cost approaches to
valuing an ecosystem service, however, is that there is often a lack of data on the
linkage between the initial ecological function, the processes and components of
ecosystems that facilitate this function, and the eventual ecological service that
benefits humans. The lack of such data makes it extremely difficult to construct
reliable hypothetical scenarios through stated preference surveys and similar methods
to elicit accurate responses from individuals about their willingness to pay for
ecological services. As an illustration, in the Catskills case study, a stated preference
survey may have elicited an estimate of the total willingness-to-pay by New York City
residents for the amount of freshwater provided – for example, the total demand for
freshwater Q 1 in Figure 2 – but it would have been very difficult to obtain a measure
of the willingness-to-pay to avoid losses in the water treatment service that occur
through changes in the land use in Catskills watershed that affect the free provision of
this ecological service.
Similarly, as pointed out by Chong (2005), it is very difficult to use stated preference
methods in tropical developing areas to assess the benefits to local communities
of the storm protection service of mangrove systems. Although there is sufficient
scientific evidence suggesting that such a service occurs, there is a lack of ecological
data on how loss of mangroves in specific locations will affect their ability to
provide storm protection to neighbouring communities. To date, the few studies
that have attempted to value the storm prevention and flood mitigation services
of the ‘natural’ storm barrier function of mangrove systems have employed the
replacement cost method by simply estimating the costs of replacing mangroves
with constructed barriers that perform the same services (Chong, 2005). Unfortu-
nately, such estimates not only make the classic error of estimating a ‘benefit’ by
a ‘cost’ but also may yield unrealistically high estimates, given that removing all the
mangroves and replacing them with constructed barriers is unlikely to be the least-
cost alternative to providing storm prevention and flood mitigation services in coastal
areas.
196 EDWARD B. BARBIER
3.5. Expected damage function approach
For some ecological services, an alternative to employing replacement cost methods
might be the expected damage function (EDF) approach.6
The EDF approach, which is a special category of ‘valuing’ the environment as
‘input’, is nominally straightforward; it assumes that the value of an asset that yields
a benefit in terms of reducing the probability and severity of some economic damage
is measured by the reduction in the expected damage. The essential step to imple-
menting this approach, which is to estimate how changes in the asset affect the
probability of the damaging event occurring, has been used routinely in risk analysis
and health economics, for example, as in the case of airline safety performance (Rose,
1990); highway fatalities (Michener and Tighe, 1992); drug safety (Olson, 2004); and
studies of the incidence of diseases and accident rates (Cameron and Trivedi, 1998;
Winkelmann, 2003). Here we show that the EDF approach can also be applied,
under certain circumstances, to value ecological services that also reduce the prob-
ability and severity of economic damages.
Recall that one of the special features of many regulatory and habitat services of
ecosystems is that they may protect nearby economic activities, property and even
human lives from possible damages. As indicated in Table 1, such services include
storm protection, flood mitigation, prevention of erosion and siltation, pollution con-
trol and maintenance of beneficial species. The EDF approach essentially ‘values’
these services through estimating how they mitigate damage costs.
The following example illustrates how the expected damage function (EDF ) methodology
can be applied to value the storm protection service provided by a coastal wetland,
such as a marshland or mangrove ecosystem. The starting point is the standard
‘compensating surplus’ approach to valuing a quantity or quality change in a non-
market environmental good or service (Freeman, 2003).
Assume that in a coastal region the local community owns all economic activity
and property, which may be threatened by damage from periodic natural storm
events. Assume also that the preferences of all households in the community are
sufficiently identical so that it can be represented by a single household. Let
m(px, z, u0) be the expenditure function of the representative household, that is, the
minimum expenditure required by the household to reach utility level, u0, given the
vector of prices, px, for all market-purchased commodities consumed by the household,
the expected number or incidence of storm events, z0.
Suppose the expected incidence of storms rises from z0 to z1. The resulting
expected damages to the property and economic livelihood of the household, E[D(z)],
translates into an exact measure of welfare loss through changes in the minimum
expenditure function:
6
The expected damage function approach predates many of the PF methods discussed so far, and has been used extensively to
estimate the risk of health impacts from pollution (Freeman, 1982, chs. 5 and 9).
VALUING ECOSYSTEM SERVICES 197
E[D(z)] = m(px, z1, u0) − m(px, z0, u0) = c(z) (12)
where c(z) is the compensating surplus. It is the minimum income compensation that
the household requires to maintain it at the utility level u0, despite the expected
increase in damaging storm events. Alternatively, c(z) can be viewed as the minimum
income that the household needs to avoid the increase in expected storm damages.
However, the presence of coastal wetlands could mitigate the expected incidence
of damaging storm events. Because of this storm protection service, the area of coastal
wetlands, S, may have a direct effect on reducing the ‘production’ of natural disasters,
in terms of their ability to inflict damages locally. Thus the ‘production function’ for
the incidence of potentially damaging natural disasters can be represented as:
z = z(S ), z′ < 0, z ″ > 0. (13)
It follows from (12) and (13) that ∂c(z)/∂S = ∂E[D(z)]/∂S < 0. An increase in wetland
area reduces expected storm damages and therefore also reduces the minimum
income compensation needed to maintain the household at its original utility level.
Alternatively, a loss in wetland area would increase expected storm damages and
raises the minimum compensation required by the household to maintain its welfare.
Thus, we can define the marginal willingness to pay, W(S ), for the protection services
of the wetland in terms of the marginal impact of a change in wetland area on
expected storm damages:
∂D
∂E[ D (z (S ))]
W (S ) = − = − E z ′, W ′ < 0. (14)
∂S ∂z
The ‘marginal valuation function’, W(S ), is analogous to the Hicksian compensated
demand function for marketed goods. The minus sign on the right-hand sign of (14)
allows this ‘demand’ function to be represented in the usual quadrant, and it has the
normal downward-sloping property (see Figure 3). Although an increase in S reduces
z and thus enables the household to avoid expected damages from storms, the addi-
tional value of this storm protection service to the household will fall as wetland area
increases in size. This relationship should hold across all households in the coastal
community. Consequently, as indicated in Figure 3, the marginal willingness to pay
by the community for more storm protection declines with S.
The value of a non-marginal change in wetland area, from S0 to S1, can be
measured as:
S1
− W (S )dS = E[ D (z (S ))] = c(S ). (15)
S0
If there is an increase in wetland area, then the value of this change is the total
amount of expected damage costs avoided. If there is a reduction in wetland area, as
shown in Figure 3, then the welfare loss is the total expected damages resulting from
the increased incidence of storm events. As indicated in (15), in both instances the
198 EDWARD B. BARBIER
Figure 3. Expected damage costs from a loss of wetland area
valuation would be a compensation surplus measure of a change in the area of
wetlands and the storm protection service that they provide.
As indicated in (14), an estimate of the marginal impact of a change in wetland
area on expected storm damages has two components: the influence of wetland area
on the expected incidence of economically damaging natural disaster events, z′, and
some measure of the additional economic damage incurred per event. Thus the right-
hand expression in (14) can be estimated, provided that there are sufficient data on
past storm events, and preferably across different coastal areas, and some estimate of
the economic damages inflicted by each event. The most important step in the
analysis is the first one, using the data on the incidence of past natural disasters and
changes in wetland area in coastal areas to estimate z(S ). One way this analysis can
be done is through employing count data models.
Count data models explain the number of times a particular event occurs over a
given period. In economics, count data models have been used to explain a variety
of phenomenon, such as explaining successful patents derived from firm R&D expen-
ditures, accident rates, disease incidence, crime rates and recreational visits (Cameron
and Trivedi, 1998; Greene, 2003, ch. 21; Winkelmann, 2003). Count data models
could be used to estimate whether a change in the area of coastal wetlands, S, reduces
the expected incidence of economically damaging storm events. The basic methodology
for such an application of count data models is described further in the appendix.
However, applying the EDF method to estimating the storm protection value of
coastal wetlands raises two additional measurement issues.
First, as the 2004 Asian tsunami and recent hurricanes in the United States have
demonstrated, the risks to vulnerable populations living in coastal areas from the
economic damages of storm events can be very large. This suggests that coastal
populations will display a degree of risk aversion to such events, in the sense that they
would like to see the least possible variance in expected storm damages. Applying
standard techniques, such as the capital-asset pricing model, this implies in turn that
VALUING ECOSYSTEM SERVICES 199
there should be a ‘risk premium’ attached to the storm protection value of coastal
wetlands that reduces the variance in expected economic damages from storm events
(Hirshleifer and Riley, 1992).
Second, estimating how coastal wetlands affect the expected number of economic
damaging events from the count data model and then multiplying the effect by the
average economic damages across events could be misleading under some extreme
circumstances. For instance, suppose a loss in wetland area is associated with a situ-
ation in which there is a change in the incidence of storms from one devastating
storm to two relatively minor storms per year. The count data model would then be
interpreted as not providing evidence against the null that the change in the wetland
area increases expected storm damages. Clearly, there needs to be a robustness check
on the count data model to ensure that such situations do not dominate the applica-
tion of the EDF approach.
4. CASE STUDY OF MANGROVE ECOSYSTEMS IN THAILAND
This section illustrates the application of the PF approach and the EDF approach to
valuation of ecological services with a case study of mangrove ecosystems in Thailand.
The two services of interest are the provision of a breeding and nursery habitat for
fisheries and the storm protection service of mangroves.
Both the dynamic and static PF approaches are used to estimate the value of the
mangrove-fishery habitat service. The EDF approach to estimating the storm protection
service of mangroves is contrasted with the replacement cost method.
4.1. Case study background
Many mangrove ecosystems, especially those in Asia, are threatened by rapid
deforestation. At least 35% of global mangrove area has been lost in the past two
decades; in Asia, 36% of mangrove area has been deforested, at the rate of 1.52%
per year (Valiela et al., 2001). Although many factors are behind global mangrove
deforestation, a major cause is aquaculture expansion in coastal areas, especially the
establishment of shrimp farms (Barbier and Cox, 2003). Aquaculture accounts for
52% of mangrove loss globally, with shrimp farming alone accounting for 38% of
mangrove deforestation; in Asia, aquaculture contributes 58% to mangrove loss with
shrimp farming accounting for 41% of total deforestation (Valiela et al., 2001).
Mangrove deforestation has been particularly prevalent in Thailand. Some estimates
suggest that over 1961–96 Thailand lost around 2050 km2 of mangrove forests, or
about 56% of the original area, mainly due to shrimp aquaculture and other coastal
developments (Charuppat and Charuppat, 1997). Since 1975, 50– 65% of Thailand’s
mangroves have been converted to shrimp farms (Aksornkoae and Tokrisna, 2004).
Figure 4 shows two long-run trend estimates of mangrove area in Thailand. In
1961, there were approximately 3700 km2 of mangroves, which declined steadily to
200 EDWARD B. BARBIER
Figure 4. Mangrove area (km2) in Thailand, 1961–2004
Notes: FAO estimates from FAO (2003). 2000 and 2004 data are estimated from 1990 –2000 annual average
mangrove loss of 18.0 km2. Thailand estimates from various Royal Thailand Forestry Department sources
reported in Aksornkoae and Tokrisna (2004). 2000 and 2004 data are estimated from 1993 –96 annual average
mangrove loss of 3.44 km2.
Sources: Based on FAO (2003), Aksornkoae and Tokrisna (2004) and author’s estimates.
around 2700 to 2900 km2 by 1980. Since then, mangrove deforestation has continued,
although there are disagreements over the rate of deforestation. For example, FAO
estimates based on long-run trend rates suggest a slower rate of decline, and indicate
that there may be almost 2400 km2 of mangroves still remaining. However, estimates
based on Thailand’s Royal Forestry Department studies suggest that rapid shrimp
farm expansion during the 1980s and early 1990s accelerated mangrove deforestation,
and as a consequence, the area of mangroves in 2004 may be much lower, closer to
1,645 km2.
Mangrove deforestation in Thailand has focused attention on the two principal
services provided by mangrove ecosystems, their role as nursery and breeding habitats
for offshore fisheries and as natural ‘storm barriers’ to periodic coastal storm events,
such as wind storms, tsunamis, storm surges and typhoons. In addition, many coastal
communities exploit mangroves directly for a variety of products, such as fuelwood,
timber, raw materials, honey and resins, and crabs and shellfish. One study estimated
that the annual value to local villagers of collecting these products was $88 per
hectare (ha), or approximately $823/ha in net present value terms over a 20-year
period and with a 10% discount rate (Sathirathai and Barbier, 2001).
4.1.1. Breeding and nursery habitat for fisheries. An extensive literature in
ecology has emphasized the role of coastal wetland habitats in supporting neighbouring
marine fisheries (for a review, see Mitsch and Gosselink, 1993; World Conservation
Monitoring Center; World Resources Institute 1996). Mangroves in Thailand also
provide this important habitat service (Aksornkoae et al., 2004).
VALUING ECOSYSTEM SERVICES 201
Thailand’s coastline is vast, stretching for 2815 km, of which 1878 km is on the
Gulf of Thailand and 937 km on the Andaman Sea (Indian Ocean) (Kaosa-ard and
Pednekar, 1998). Since 1972, the 3 km offshore coastal zone in southern Thailand
has been reserved for small-scale, artisanal marine fisheries. The Gulf of Thailand is
divided into four such major zones, and the Andaman Sea comprises a fifth zone.7
The mangroves along these coastal zones are thought to provide breeding grounds
and nurseries in support of several species of demersal fish and shellfish (mainly crab
and shrimp) in Thailand’s coastal waters.8 The artisanal marine fisheries of the five
major coastal zones of Thailand depend largely on shellfish but also some demersal
fish. For example, in 1994 shrimp, crab, squid and cuttlefish alone accounted for 67%
of all catch in the artisanal marine fisheries, and demersal fish accounted for 5.3%
(Kaosa-ard and Pednekar, 1998).
The coastal artisanal fisheries of Thailand are characterized by classic open access
conditions (Kaosa-ard and Pedneker, 1998; Wattana, 1998). Since the 1970s, there
have been approximately 36 000 –38 000 households engaged in small-scale fishing
activities. Although there are 2500 fishing communities scattered over the 24 coastal
provinces of Thailand, 90% of the artisanal fishing households are concentrated in
communities spread along the Southern Gulf of Thailand and Andaman Sea coasts.
While the number of households engaged in small-scale fishing has remained fairly
stable since 1985, the use of motorized boats has increased by more than 30%
(Wattana, 1998). Gill nets still remain the most common form of fishing gear used
by artisanal fishers. Although a licence fee and permit are required for fishing in
coastal waters, officials do not strictly enforce the law and users do not pay. Currently,
there is no legislation for supporting community-based fishery management (Kaosa-
ard and Pednekar, 1998).
4.1.2. Storm protection. The 26 December 2004 Indian Ocean tsunami disaster
has focused attention on the role of natural barriers, such as mangroves, in protecting
vulnerable coastlines and populations in the region from such storm events (UNEP,
2005; Wetlands International, 2005). Mangrove wetlands, which are found along
sheltered tropical and subtropical shores and estuaries, are particularly valuable in
minimizing damage to property and loss of human life by acting as a barrier against
tropical storms, such as typhoons, cyclones, hurricanes and tsunamis (Chong, 2005;
Massel et al., 1999; Mazda et al., 1997). Evidence from the 12 Indian Ocean countries
affected by the tsunami disaster, including Thailand, suggests that those coastal areas
7
The four Gulf of Thailand zones consist of the following coastal provinces: Trat, Chantaburi and Rayong (Zone 1); Chon
Buri, Chachoengsao, Samut Parkakan, Samut Sakhon, Samut Songkhram, Phetchaburi, Prachaup Khiri Khan (Zone 2);
Chumphon, Surat Thani, Nakhon Si Thammarat (Zone 3); and Songkhla, Patthani, Narathiwart (Zone 4). The fifth zone on
the Indian Ocean (Andaman Sea) consists of the following coastal provinces: Ranong, Phangnga, Phuket, Krabi, Trang and
Satun (Zone 5).
8
Mangrove-dependent demersal fish include those belonging to the Clupeidae, Chanidae, Ariidae, Pltosidae, Mugilidae, Lujanidae and
Latidae families. The shellfish include those belonging to the families of Panaeidae for shrimp and Grapsidae, Ocypodidae and Portnidae
for crab.
202 EDWARD B. BARBIER
that had dense and healthy mangrove forests suffered fewer losses and less damage
to property than those areas in which mangroves had been degraded or converted to
other land uses (Dahdouh-Guebas et al., 2005; Harakunarak and Aksornkoae, 2005;
Kathiresan and Rajendran, 2005; UNEP, 2005; Wetlands International, 2005).
In Thailand, the Asian tsunami affected all six coastal provinces along the Indian
Ocean (Andaman Sea) coast: Krabi, Phang Nga, Phuket, Ranong, Satun and Trang.
In Phang Nga, the most affected province, post-tsunami assessments suggest that
large mangrove forests in the north and south of the province significantly mitigated
the impact of the Tsunami. They suffered damage on their seaside fringe, but
reduced the tidal wave energy, providing protection to the inland population (UNEP,
2005; Harakunarak and Aksornkoae, 2005). Similar results were reported for those
shorelines in Ranong Province protected by dense and thriving mangrove forests. In
contrast, damages were relatively extensive along the Indian Ocean coast where
mangroves and other natural coastal barriers were removed or severely degraded
(Harakunarak and Aksornkoae, 2005).
With the overwhelming evidence of the storm protection service provided by intact
and healthy mangrove systems, since the tsunami disaster increased emphasis has
been placed on replanting degraded and deforested mangrove areas in Asia as a
means to bolstering coastal protection. For example, the Indonesian Minister for
Forestry has announced plans to reforest 600 000 hectares of depleted mangrove
forest throughout the nation over the next 5 years. The governments of Sri Lanka
and Thailand have also stated publicly intentions to rehabilitate and replant man-
grove areas (UNEP, 2005; Harakunarak and Aksornkoae, 2005).
Although the Asian tsunami has called attention to the storm protection service
provided by mangroves, the benefits of this service extends to protection against many
types of periodic coastal natural disaster events. As one post-tsunami assessment
noted: ‘It is important to recognize that any compromising of mangrove “protection
function” is relevant to a wide variety of storm events, and not just tsunamis. Whereas
the Indian Ocean area counted “only” 63 tsunamis between 1750 and 2004, there
were more than three tropical cyclones per year in roughly the same area’ (Dahdouh-
Guebas et al., 2005, pp. 445– 6).
The EM-DAT International Disaster Database shows that the number of coastal
natural disasters in Thailand has increased in both the frequency of occurrence and
in the number of events per year (see Figure 5). Over 1975–87, Thailand experienced
on average 0.54 coastal natural disasters per year, whereas between 1987–2004 the
incidence increased to 1.83 disasters per year. Thus, a recent World Bank report
identified the coastal and delta areas of Thailand as potentially high fatality (more
than 1000 deaths per event) and other damage ‘hotspots’ at risk from storm surge
events (Dilley et al., 2005, pp. 101–3).
The EM-DAT database also calculates the economic damage incurred per event.
Figure 6 plots the damages per coastal natural disaster in Thailand for 1975–2004.
The 2004 Asian tsunami with estimated damages of US$240 million (1996 prices)
VALUING ECOSYSTEM SERVICES 203
Figure 5. Coastal natural disasters in Thailand, 1975–2004
Notes: Over 1975–2004, coastal natural disasters included wave/surge (tsunami and tidal wave), wind storm
(cyclone/typhoon and tropical storm) and flood (significant rise of water level in coastal region. In order for
EM-DAT (2005) to record an event as a disaster, at least one or more of the following criteria must be fulfilled:
10 or more people reported killed; 100 people reported affected; declaration of a state of emergency; call for
international assistance.
Source: EM-DAT (2005). EM-DAT: The OFDA/CRED International Disaster Database. www.em-dat.net –
Université Catholique de Louvain, Brussels, Belgium.
Figure 6. Real damages per coastal natural disaster in Thailand, 1975–2004
Notes: The EM-DAT (2005) estimate of the economic impact of a disaster usually consists of direct (e.g. damage
to infrastructure, crops, housing) and indirect (e.g. loss of revenues, unemployment, market destabilization)
consequences on the local economy. However, the estimate of ‘zero’ economic damages may indicate that no
economic damages were recorded for an event. The estimates of economic damages are in thousands of US$
and converted to 1996 prices using Thailand’s GDP deflator.
Source: EM-DAT (2005). EM-DAT: The OFDA/CRED International Disaster Database. www.em-dat.net –
Université Catholique de Louvain, Brussels, Belgium.
was not the most damaging event to occur in Thailand. In fact, although the inci-
dence of coastal damages has increased since 1987, in recent years the real damages
per event has actually declined. For example, from 1979 to 1996, the economic
damages per event were around US$190 million whereas from 1996 to 2004, real
damages per event averaged US$61 million.
In sum, over the past two decades the rise in the number and frequency of coastal
natural disasters in Thailand (Figure 5) and the simultaneous rapid decline in coastal
mangrove systems over the same period (Figure 4) is likely to be more than a
204 EDWARD B. BARBIER
coincidence. Natural disasters occur when large numbers of economic assets are
damaged or destroyed during a natural hazard event. Thus an increase in the inci-
dence of coastal disasters is likely to have two sets of causes: the first is the natural
hazards themselves – tsunamis and other storm surges, tidal waves, typhoons or
cyclones, tropical storms and floods – but the second set is the increasing vulnerability
of coastal populations, infrastructure and economic activities to being harmed or
damaged by a hazard event.9 The widespread loss of mangroves in coastal areas
of Thailand may therefore have increased the vulnerability of these areas to more
incidences of natural disasters.
4.2. Valuation of habitat-fishery linkage service
This subsection compares and contrasts the static and dynamic approaches outlined
in Section 3.3 to valuing the habitat-fishery support service of mangroves in Thai-
land. As discussed above, in Thailand the near-shore artisanal fisheries supported by
this ecological service are not optimally managed but largely open access.
To conduct the static production function analysis of the mangrove-fishery linkage,
the methodology of Section 3.3.1 is applied to the same shellfish, demersal fishery
and mangrove data over 1983–96 as in Barbier (2003). These comprise pooled time-
series and cross-sectional data over the 1983–96 period for Thailand’s artisanal and
shellfish fisheries, as well as the extent of mangrove area, corresponding to the five
coastal zones along the Gulf and Thailand and Indian Ocean (Andaman Sea). Evi-
dence from domestic fish markets in Thailand suggest that the demand for fish is fairly
inelastic, and an elasticity of −0.5 was assumed for the iso-elastic market demand function.
Thus the static analysis calculation of the marginal impact of a change in wetland
area in Equation (5) requires specifying the unknown parameters of the Cobb –Douglas
production function for the fishery, h = AE aS b. Section A1 in the appendix explains
the approach used to estimate the unknown parameters (A, a, b) of the log-linear version
of the Cobb–Douglas production function and reports the resulting preferred estimations
(see Table A1). Using these results in Equations (5) and (6) allows calculation of the
welfare impacts of mangrove deforestation on Thailand’s two artisanal fisheries. The
results are depicted in Table 3, which displays both the point estimates and the
95% confidence bounds on these estimates through use of the standard errors.
All price and cost data for the fisheries used in the welfare analysis are in 1996 real
terms.
9
This view that natural disasters should not be viewed solely as ‘acts of God’ but clearly have an important anthropogenic
component to their cause is reflected in much of the current expert opinion on natural disaster management. This is summa-
rized succinctly by Dilley et al. (2005, p. 115): ‘Hazards are not the cause of disasters. By definition, disasters involve large human
or economic losses. Hazard events that occur in unpopulated areas and are not associated with losses do not constitute disasters.
Losses are created not only by hazards, therefore, but also by the intrinsic characteristics of the exposed infrastructure, land
uses, and economic activities that cause them to be damaged or destroyed when a hazard strikes. The socioeconomic contribu-
tion to disaster causality is potentially a source of disaster reduction. Disaster losses can be reduced by reducing exposure or
vulnerability to the hazards present in a given area.’
VALUING ECOSYSTEM SERVICES 205
Table 3. Valuation of mangrove-fishery linkage service, Thailand, 1996–2004 (US$)
Production function approach Average annual mangrove loss
FAO (18.0 km2)a Thailand (3.44 km2)b
Static analysis:
Annual welfare loss 99 004 (12 704–814 504) 18 884 (2425–154 307)
Net present value 570 167 108 756
(10% discount rate) (55 331– 4 690 750) (10 563–888 657)
Net present value 527 519 100 621
(12% discount rate) (52 233 – 4 339 883) (9972–822 186)
Net present value 472 407 90 108
(15% discount rate) (48 080 –3 886 476) (9179–736 288)
Dynamic analysis:
Net present value 1 980 128 373 404
(10% discount rate) (403 899–2 390,728) (164 506–691 573)
Net present value 1 760 374 331 995
(12% discount rate) (357 462–2 104 176) (147 571–614 058)
Net present value 1 484 461 279 999
(15% discount rate) (299 411–1 747 117) (126 178–516 691)
Notes: All valuations are based on mangrove-fishery linkage impacts on artisanal shellfish and demersal fisheries
in Thailand at 1996 prices. The demand elasticity for fish is assumed to be −0.5. Figures in parentheses
represent upper and lower bound welfare estimates based on the standard errors of the estimated parameters
in each model (see Section A1 in the appendix).
a
FAO estimates from FAO (2003). 2000 and 2004 data are estimated from 1990–2000 annul average mangrove
loss of 18.0 km2.
b
Thailand estimates from various Royal Thailand Forestry Department sources reported in Aksornkoae and
Tokrisna (2004). 2000 and 2004 data are estimated from 1993–96 annual average mangrove loss of 3.44 km 2.
Sources: Author’s calculations.
As Figure 4 shows, there are two different estimates of the 1996 –2004 annual
mangrove deforestation rates in Thailand, namely the FAO estimate of 18.0 km2 and
the Royal Thai Forestry Department estimate of 3.44 km2. For the welfare impacts
arising from the FAO estimates of annual average mangrove deforestation rates in
Thailand over 1996 –2004, the static analysis suggests that the annual loss in the
habitat-fishery support service is around US$99 000 ($13 000 to 815 000 with 95%
confidence). The net present value of these losses over the entire period is between
US$0.47 and 0.57 million ($48 000 to 4.7 million with 95% confidence). For the
much lower Thailand deforestation estimates, the annual welfare loss is just under
$19 000 ($2400 to 154 000 with 95% confidence) and the net present value of these
losses over the 1996 –2004 period is US$90 000 to 108 000 ($9000 to 0.9 million with
95% confidence).
Following the methodology of Section 3.3.2, we can also apply a dynamic produc-
tion function model to mangrove-fishery linkages in Thailand. As explained in the
section, this approach involves estimating the parameters of the dynamic mangrove-
fishery model, and then using these parameters to simulate the dynamic path of the
fishery and the corresponding consumer and surplus changes resulting from
mangrove deforestation. Because there are no data on the biomass stock, Xt, for
206 EDWARD B. BARBIER
Thailand’s near-shore fisheries, the appropriate dynamic model is the version indicating
the change over time in fishing effort, Et, and catch per unit effort, ct, i.e. Equations
(10) and (11). To compare with the static analysis, we use the same shellfish,
demersal fisheries and mangrove data, as well as assume the same iso-elastic demand,
from Barbier (2003) to estimate Equations (10) and (11) (see Section A2 in the
appendix). For example, the estimated parameters in the appendix correspond to
the following parameters of the dynamic production function model: b0 = r, b1 = −r/
qα, b2 = −q, b0/(b1*b2) = α, a1 = φ, a2 = 1−φw and −(a2 − 1)/a1 = w. These estimated
parameters are then employed to simulate the dynamic effort and stock paths (9) and
(10) of each fishery, starting from an initial level of effort, catch per unit effort and
mangrove area, and assuming a constant elasticity of demand of −0.5.10 By using
1996 data as the initial starting point in the simulation, i.e. for X0, E0, S0 and h0,
the dynamic paths yield effort, stock and harvest for each subsequent year from
1996–2004.
In the base case dynamic simulation, mangrove area is held constant at 1996 levels.
Two alternative paths for stock, effort and thus harvest are then also simulated,
corresponding to the two different estimates of the 1996–2004 annual mangrove
deforestation rates in Thailand, namely the FAO estimate of 18.0 km2 and the Royal
Thai Forestry Department estimate of 3.44 km2 respectively. The resulting changes
in consumer plus producer surpluses in each year over 1996 –2004, between each
deforestation simulation and the base case, provide the estimates of the welfare
impacts of the decline in the mangrove-fishery support service. That is, the changes
in consumer and producer surplus resulting from mangrove deforestation in
each subsequent year of the simulation are discounted to obtain a net present value
estimate of the resulting welfare loss. As in the static analysis, the discount rate
is varied from 10% to 15% (see Table 3). The standard errors for the parameters
of the model estimated from Equations (10) and (11) were also used to construct
both lower and upper confidence bounds on the simulation paths, and thus also
on the welfare estimates of the impacts of deforestation on the mangrove-fishery
linkage.
The results for the dynamic mangrove-fishery linkage analysis are also depicted in
Table 3, which indicates the welfare calculations associated with both the FAO and
Thailand deforestation estimates over 1996–2004. The table reports calculations
arising from the simulations based on the point estimates of the parameters of the
dynamic mangrove-fishery model. The ranges of values indicated in parentheses for
the dynamic analysis represent the lower and upper bound confidence intervals
10
Although there are no reliable stock data for Thailand’s near-shore fisheries, the Schaefer harvesting function, ht = qEtXt,
assumed in the model allows stock to be determined from catch per unit effort for a given estimated parameter q. That is,
Xt = ct/q, where ct = ht/Et. See Schnute (1977) for further details. The procedure employed here is to use the known harvest
and effort levels, as well as the estimated parameter q, for each fishery in the initial year 1996 to estimate the initial unknown
stock level, X0. Equations (9) and (10) were then used to simulate the dynamic path for Xt and Et in the subsequent years (1997–2004),
as well as the subsequent harvest, ht = qEt Xt. The dynamic simulation approach employed here is standard for an open access
fishery model (see Bjørndal and Conrad, 1987; Clark, 1976; Homans and Wilen, 1997).
VALUING ECOSYSTEM SERVICES 207
derived from the standard errors of the estimated model parameters (see Section A2
in the appendix). If the FAO estimate of mangrove deforestation over 1996–2004 is
used, then the net present value of the welfare loss ranges from around US$1.5 to
2.0 million ($0.3 to 2.4 million in the upper and lower bound simulation estimates).
In contrast, the lower Thailand deforestation estimation for 1996 –2004 suggests that
the net present value welfare loss from reduced mangrove support for fisheries is
around US$0.28 to 0.37 million ($0.13 to 0.69 million in the upper and lower bound
simulation estimates).
The welfare estimates in Table 3 indicate that the losses in the habitat-fishery
support service caused by mangrove deforestation in Thailand over 1996–2004 are
around three times greater for the dynamic production function approach compared
to the static analysis. In addition, the confidence bounds on the welfare estimates
produced with the static analysis are significantly larger, suggesting that the static
approach yields much more variable estimates of the welfare losses. Given the disparity
in estimates between the two approaches, a legitimate question to ask is whether
or not one approach should be preferred to the other in valuing habitat-fishery linkages.
It has been argued in the literature that, on the methodological grounds, the
‘dynamic’ PF approach is more appropriate for valuing how coastal wetland habitats
support offshore fisheries because this service implies that fish populations are more
likely to be affected over time (Barbier, 2000). If this is the case, then the environ-
mental ‘input’ of mangroves serving as breeding and nursery habitat for near-shore
fisheries should be modelled as part of the growth function of the fish stock. In
contrast, the static analysis, by definition, ignores stock effects and focuses exclusively
on the impact of changes in mangrove area on fishing effort and costs in the same
period in which the habitat service changes. The comparison of the dynamic and
static analysis in the Thailand case study of mangrove-fishery linkages confirms that,
by incorporating explicitly the multi-period stock effects resulting from mangrove loss,
the dynamic model produces much larger estimates for the value of changes in the
habitat-fishery support service. Since in this case study at least these stock effects
appear to be considerable, then they are clearly an important component of the
impacts of mangrove deforestation on the habitat-fishery service in Thailand.
In sum, the Thailand case study suggest caution in using the static analysis in
preference to the dynamic production function approach in valuing the ecological
service of coastal wetlands as breeding and nursery habitat for offshore fisheries. As
Table 3 indicates, the static approach could underestimate the value of this service as
well as yield more variable estimates. This may prove misleading for policy analysis,
particularly when considering options to preserve as opposed to convert coastal
wetlands. Certainly, the perception among coastal fishing communities throughout
Thailand is that the habitat-fishery service of mangroves is vital, and local fishers in
these communities have reported substantial losses in coastal fish stocks and yields,
which they attribute to recent deforestation (Aksornkoae et al., 2004; Sathirathai
and Barbier, 2001).
208 EDWARD B. BARBIER
4.3. Valuation of storm protection service
To date, the most prevalent method of valuing the storm protection service provided
by coastal wetlands is the replacement cost approach (Chong, 2005). This paper has
suggested the use of an alternative methodology, the EDF approach. The purpose of
the following subsection is to compare and contrast both approaches, using the
Thailand case study.
Sathirathai and Barbier (2001) employed the replacement cost method to estimate
the value of coastal protection and stabilization provided by mangroves in southern
Thailand. The same approach and data will be employed here. According to the
Harbor Department of the Royal Thai Ministry of Communications and Transport,
the unit cost of constructing artificial breakwaters to prevent coastal erosion and
damages from storm surges is estimated to be US$1011 (in 1996 prices) per metre
of coastline. Based on this estimate, the authors calculate the equivalent cost of
protecting the shoreline with a 75-metre width stand of mangrove is approximately
US$13.48 per m2, or US$134 801 per ha (1996 prices). Over a 20-year period and
assuming a 10% discount rate, the annualized value of this cost amounts to $14 169
per ha. This is the ‘replacement cost’ value of the storm protection function per ha
of mangrove.
The analysis for this paper uses this replacement cost value to calculate the annual
and net present value welfare losses associated with the two mangrove deforestation
estimates for Thailand over 1996–2004. The results are depicted in Table 4.
For the FAO mangrove deforestation estimate of 18.0 km2 per year over 1996–
2004, the annual welfare loss in storm protection service is around US$25.5 million,
and the net present value of this loss over the entire period ranges from US$121.7 to
146.9 million. For the Thailand deforestation estimation of 3.4 km2 per year, the
annual welfare loss in storm protection is about US$4.9 million, and the net present
value of this loss over the entire period ranges from US$23.2 to 28 million.
Section 3.5 describes the methodology for the EDF approach to estimating the
value of the storm protection service of coastal wetlands such as mangroves. As
emphasized in the appendix, the key step to this approach is to estimate the influence
of changes in coastal wetland area on the expected incidence of economically
damaging natural disaster events. The application of the EDF approach here employs
a count data model for this purpose. The details of the estimation are contained in
Section A3 of the appendix.
The analysis for Thailand over 1979–96 shows that loss of mangrove area in
Thailand increases the expected number of economically damaging natural disasters
affecting coastal provinces. Using this estimated ‘marginal effect’ (−0.00308), it is
possible to estimate the resulting impact on expected damages of natural coastal
disasters. For example, EM-DAT (2005) data show that over 1979–96 the estimated
real economic damages per coastal event per year in Thailand averaged around
US$189.9 million (1996 prices). This suggests that the marginal effect of a one-km2
VALUING ECOSYSTEM SERVICES 209
Table 4. Valuation of storm protection service, Thailand, 1996–2004 (US$)
Valuation approach Average annual mangrove loss
FAO (18.0 km2)a Thailand (3.44 km2)b
Replacement cost method:c
Annual welfare loss 25 504 821 4 869 720
Net present value (10% discount rate) 146 882 870 28 044 836
Net present value (12% discount rate) 135 896 056 25 947 087
Net present value (15% discount rate) 121 698 392 23 236 280
Expected damage function approach:
Annual welfare loss 3 382 169 645 769
( 2 341 686–5 797 339) (447 106–1 106 905)
Net present value 19 477 994 3 718 998
(10% discount rate) (13 485 827–33 387 014) (2 574 894–6 374 694)
Net present value 18 021 043 3 440 818
(12% discount rate) (12 477 089–30 889 671) (2 382 292–5 897 868)
Net present value 16 138 305 3 081 340
(15% discount rate) (11 173 553–27 662 490) (2 133 404–5 281 692)
Notes: Figures in parentheses represent upper and lower bound welfare estimates based on the 95% confidence
interval for the estimated coefficients in the model (see Section A3 in the appendix).
a
FAO estimates from FAO (2003). 2000 and 2004 data are estimated from 1990–2000 annual average mangrove
loss of 18.0 km2.
b
Thailand estimates from various Royal Thailand Forestry Department sources reported in Aksornkoae and
Tokrisna (2004). 2000 and 2004 data are estimated from 1993–96 annual average mangrove loss of 3.44 km 2
c
Re-calculated based on Sathirathai and Barbier (2001).
Sources: Author’s calculations.
loss of mangrove area is an increase in expected storm damages of about US$585 000
per km2. In Table 4, this latter calculation is combined with the FAO and Thailand
estimates of the average annual rates of deforestation to compute the welfare losses
in storm protection service for Thailand over 1996–2004. The table shows the wel-
fare calculations based both on the point estimates of the count data regression and
on using the standard errors to construct 95% confidence bounds on these estimates.
Table 4 shows that, for the FAO mangrove deforestation estimate of 18.0 km2 per
year over 1996–2004, the EDF approach estimates the annual welfare loss in storm
protection service to be around US$3.4 million ($2.3 to 5.8 million with 95% confi-
dence), and the net present value of this loss over the entire period ranges from
US$16.1 to 19.5 million ($11.2 to 33.4 million with 95% confidence). For the Thai-
land deforestation estimation of 3.4 km2 per year, the annual welfare loss in storm
protection is over US$0.65 million ($0.45 to 1.1 million with 95% confidence), and
the net present value of this loss over the entire period ranges from US$3.1 to 3.7
million ($2.1 to 6.4 million with 95% confidence).
Comparing the EDF approach and the replacement cost method of estimating the
welfare impacts of a loss of the storm protection service due to mangrove deforestation
confirms that the replacement cost method tends to produce extremely high
estimates – almost 4 times greater than even the largest upper-bound estimate
210 EDWARD B. BARBIER
calculated using the EDF approach. This suggests that the replacement cost method
should be used with caution, and when data are available, the EDF approach may
provide more reliable values of the storm protection service of coastal wetlands.
4.4. Land use policy implications
Valuation of the ecosystem services provided by mangroves are important for two
land use policy decisions in Thailand. First, although declining in recent years, con-
version of remaining mangroves to shrimp farm ponds and other commercial coastal
developments continues to be a major threat to Thailand’s remaining mangrove
areas. Second, since the December 2004 tsunami disaster, there is now considerable
interest in rehabilitating and restoring mangrove ecosystems as ‘natural barriers’ to
future coastal storm events.
To illustrate how improved and more accurate valuation of ecosystems can help
inform these two policy decisions, Table 5 compares the per hectare net returns to
shrimp farming, the costs of mangrove rehabilitation and the value of mangrove
services. All land uses are assumed to be instigated over 1996–2004 and are valued
in 1996 US dollars. The net economic returns to shrimp farming are based on non-
declining yields over a 5-year period of investment, with the pond abandoned in
subsequent years (Sathirathai and Barbier, 2001). These returns to shrimp aquaculture
are estimated to be $1078 to $1220 per ha. In comparison, the costs rehabilitating
mangrove ecosystems on land that has been converted to shrimp farms and then
abandoned are $8812 to $9318 per ha. Thus valuing the goods and services of
mangrove ecosystems can help to address two important policy questions: Do the net
economic returns to shrimp farming justify further mangrove conversion to this economic
activity, and is it worth investing in mangrove replanting and ecosystem rehabilitation
in abandoned shrimp farm areas?
As indicated in Table 5, if the older methods of valuing habitat-fishery linkages
with the static approach and storm protection with the replacement cost method are
employed, then mangrove ecosystem benefits are considerably higher than the net
economic returns to shrimp farming and the costs of replanting and rehabilitating
mangroves in abandoned farm areas. However, the static analysis undervalues the
habitat-fishery linkage of mangroves whereas the replacement cost method over-
inflates storm protection. The replacement cost method estimates storm protection
at $67 610 to 81 602 per ha, which is 99% of the value of all mangrove ecosystem
benefits. In contrast, the net income to local coastal communities from collected forest
products and the value of habitat-fishery linkages total to only $730 to $881 per ha,
which suggests that these two benefits of mangroves are insufficient on their own to
justify either halting conversion to shrimp farms or replanting and rehabilitating these
ecosystems on abandoned pond land.
If improved methods of valuing habitat-fishery linkages by the dynamic approach
and storm protection by the expected damage function method are employed, then
VALUING ECOSYSTEM SERVICES 211
Table 5. Comparison of land use values per hectare, Thailand, 1996–2004 (US$)
Land use Net present value per hectare
(10–15% discount rate)
Shrimp farming:a
Net economic returns 1078–1220
Mangrove ecosystem rehabilitation:b
Total cost 8812–9318
Ecosystem goods and services Older methods Improved methods
Net income from collected forest productsc 484–584 484–584
246–297d 708–987e
Habitat-fishery linkage
67 610–81 602f 8966–10 821g
Storm protection service
Total 68 341–82 484 10 158–12 392
a
Based on Sathirathai and Barbier (2001), updated to 1996 US$.
b
Based on costs of rehabilitating abandoned shrimp farms, replanting mangrove forests and maintaining and
protecting mangrove seedlings. From Sathirathai and Barbier (2001), updated to 1996 US$.
c
Based on Sathirathai and Barbier (2001), updated to 1996 US$.
d
Based on marginal value per ha based on the static analysis of this study (see Section 4.2).
e
Based on average value per hectare over 1996–2004 based on the dynamic analysis of this study and assuming
the estimated Thailand deforestation rate of 3.44 sq km per year (see Section 4.2).
f
Based on average value per hectare of replacement cost method of this study (see Section 4.3).
g
Based on marginal value per hectare of expected damage function approach of this study (see Section 4.3).
Sources: Author’s calculations.
the outcome is somewhat different. Although the total value of mangrove ecosystem
services is lowered to $10 158 to $12 392 per ha, it still exceeds the net economic
returns to shrimp farming. Storm protection service is still the largest benefit of
mangroves, but it no longer dominates the land use value comparison. The net
income to local coastal communities from collected forest products and the value of
habitat-fishery linkages total to $1192 to $1571 per ha, which now are greater than
the net economic returns to shrimp farming. The value of the storm protection,
however, is critical to the decision as to whether or not to replant and rehabilitate
mangrove ecosystems in abandoned pond areas. As shown in Table 5, storm protec-
tion benefits make mangrove rehabilitation an economically feasible land use option.
5. CONCLUSIONS
The case study of valuing mangroves in Thailand illustrates the potential use of the
PF approach to modelling key ecological regulatory and habitat services. The study
also indicates the importance of choosing the appropriate PF method for modelling
the key ecological-economic linkages underlying each service.
For example, the case study confirms that, if coastal wetlands such as mangroves
serve as a breeding and nursery habitat for a variety of near-shore fisheries, then it
seems more appropriate to model this environmental input as part of the growth
function of the fish stock. In comparison, not accounting for the stock effects of a
change in coastal nursery and breeding grounds may lead to an underestimation of
the value of this habitat-fishery linkage. The case study also illustrates how the EDF
212 EDWARD B. BARBIER
approach can be applied to valuing the storm protection service provided by mangroves,
and demonstrates why this method should be preferred to the less-reliable replacement
cost method, which has been used extensively in the literature to date (Chong, 2005).
The case study also points to some important policy implications for Thailand. In
recent decades, considerable mangrove deforestation has taken place in Thailand, mainly
as a result of shrimp farm expansion and other coastal economic developments (see Figure 4).
Over this period, mangrove conversion for these development activities was system-
atically encouraged by government land use policies (Aksornkoae and Tokrisna, 2004;
Barbier, 2003; Sathirathai and Barbier, 2001). Such policies were designed without
consideration of the value of the ecological services provided by mangroves, such as
their habitat support for coastal fisheries and storm protection. The case study of valuing
these ecological services for Thailand illustrates that their benefits are significant, and
should certainly not be ignored in future mangrove land management decisions.
The case study applications in this paper of valuing coastal storm protection and
habitat services have policy implications beyond Thailand as well. Even before
Hurricanes Katrina and Rita devastated the central Gulf Coast of the United States
in 2005, the US Army Corps of Engineers had proposed a $1.1 billion multi-year
programme to slow the rate of wetland loss and restore some wetlands in coastal
Louisiana. In the aftermath of these hurricanes, the US Congress is now considering
expanding the programme substantially to a $14 billion restoration effort (Zinn,
2005). As noted in Section 4, in the wake of the 2004 Asian tsunami, mangrove
restoration projects for enhanced coastal protection are underway in many countries
throughout the region. International donor groups are also supporting mangrove
restoration projects in Asia, especially in countries and regions devastated by the
tsunami (Check, 2005). In addition, there is mounting scientific evidence that near-
shore fisheries throughout the world are undergoing rapid decline, with loss of coastal
habitat and nursery grounds for these fisheries a contributory cause (Jackson et al.,
2001; Myers and Worm, 2003). Valuing the storm protection and habitat services of
coastal wetlands, as illustrated by the Thailand case study in this paper, can therefore
play a vital role in current and future debates about the state of coastal ecosystems
worldwide and the assessment of the costs and benefits of restoring these vital ecosystems.
Thus, valuing the non-market benefits of ecological regulatory and habitat services
is becoming increasingly important in assisting policymakers to manage critical
environmental assets. However, further progress applying production function
approaches and other methods to value ecological services faces two challenges.
First, for these methods to be applied effectively to valuing ecosystem services, it is important
that the key ecological and economic relationships are well understood. Unfortunately,
our knowledge of the ecological functions, let alone the ecosystem processes and
components, underlying many of the services listed in Table 1 is still incomplete.
Second, natural ecosystems are subject to stresses, rapid change and irreversible
losses, they tend to display threshold effects and other non-linearities that are difficult
to predict, let alone model in terms of their economic impacts. These uncertainties
VALUING ECOSYSTEM SERVICES 213
can affect the estimation of values from an ex ante (‘beforehand’) perspective. The
economic valuation literature recognizes that such uncertainties create the conditions
for option values, which arise from the difference between valuation under conditions
of certainty and uncertainty (e.g., see Freeman, 2003 and Just et al., 2004). The
standard approach recommended in the literature is to estimate this additional value
separately, through various techniques to measure an option price, that is, the amount
of money that an individual will pay or must be compensated to be indifferent from
the status quo condition of the ecosystem and the new, proposed condition. However,
in practice, estimating separate option prices for unknown ecological effects is very
difficult. Determining the appropriate risk premium for vulnerable populations
exposed to the irreversible ecological losses is also proving elusive. These are problems
currently affecting all economic valuation methods of ecosystem services, and not just
the production function approach. As one review of these studies concludes: ‘Given
the imperfect knowledge of the way people value natural ecosystems and their
goods and services, and our limited understanding of the underlying ecology and
biogeochemistry of aquatic ecosystems, calculations of the value of the changes
resulting from a policy intervention will always be approximate’ (Heal et al., 2005, p. 218).
Finally, Section 3 noted recent attempts to extend the production function
approach to the ecosystem level through integrated ecological-economic modelling.
This allows the ecosystem functioning and dynamics underlying the provision of
ecological services to be modelled and can be used to value multiple rather than
single services. For example, returning to the Thailand case study, it is well known
that both coral reefs and sea grasses complement the role of mangroves in providing
both the habitat-fishery and storm protection services. Thus full modelling of the
integrated mangrove–coral reef–sea grass system could improve measurement of the
benefits of both services. As we learn more about the important ecological and
economic role played by such services, it may be relevant to develop multi-service
ecosystem modelling to understand more fully what values are lost when such
integrated coastal and marine systems are disturbed or destroyed.
Discussion
Carlo A. Favero
IGIER, Bocconi University and CEPR
The objective of the paper is to apply a production function (and expected damage)
approach to ‘valuing the environment as input’, with an application to a mangrove
ecosystem in Thailand. I shall concentrate my discussion only on the production
function approach, but the main methodological points raised are naturally extended
to the expected damage function approach.
214 EDWARD B. BARBIER
An environmental good or service essentially serves as a factor input into production
that yields utility. The fundamental problem of the empirical application is the
evaluation of the following value function:
T
Bt + j
Vt = Et ∑
(1 + r ) j
j =0
where B is the social benefits in any time period of the mangrove ecosystem, and r is
the discount rate.
There are two fundamental questions:
1. What is the appropriate discount rate?
2. How to evaluate B?
The first question is not explicitly addressed and different scenarios on the discount
rate are adopted; the production function approach is the answer to the second question.
In theory the production function approach can be described as follows:
• Specification of a dynamic intertemporal optimization problem, where one of the
constraints is the production function relating the input of interest to the measurable
output.
• Solution of the model.
• Identification and estimation (or, whenever estimation is not possible, calibration) of
the technology and preference parameters of the model and of the auxiliary parameters.
• Dynamic stochastic simulation of the model to derive Et Bt+j and the associated
confidence intervals.
In practice two alternative approaches are considered: a static one and a dynamic
one. I shall not comment on the static approach because I find this inappropriate to
the very nature of the problem at hand, which is, by definition, dynamic.
In the dynamic approach a model is postulated to determine the dynamics of the
stock of fish measured in biomass units, Xt, the fishing effort, Et, the landed fish price
per unit harvested, p, and the harvest, h. The adopted model is described as follows:
X t +1 − X t = F ( X t , St ) − h( X t , Et )
Et +1 − Et = φ[ pt ht − wt Et ]
Xt
F ( X t , St ) = rX t 1 −
α ln St
ht = qX t Et
pt = khtη
where F(Xt, St) is biological growth in the current period, which is a function of St, the
mangrove area, h(Xt, Et), harvesting is a function of the stock as well as fishing effort,
Et. Fishing effort is modelled as a partial adjustment model in which the equilibrium
value is determined by fish price per unit harvested and the unit cost of effort, w.
VALUING ECOSYSTEM SERVICES 215
The model is estimated and then simulated keeping w exogenous and taking alter-
native scenarios for S, that by consequence is taken as exogenous. Using some
assumption for the discount rate the present value of a reduction in the mangrove
area is then computed.
The results are interesting but there are a number of important questions that the
modelling strategy leaves open:
• In the dynamic model S is exogenous and no law of motion for S is specified. The
model is not capable of explaining the reduction in S that we observe in the data.
In fact, if agents were acting following this model we would have never observed
a reduction in S, because a reduction in S has only costs and no benefits.
Macroeconometricians might see the applicability of the Lucas critique to this
model as an immediate consequence of the assumption of exogeneity of S.
• Expectations do not explicitly enter the model.
• What are the costs incurred in omitting from the model the dynamics of w?
• What is the performance of this model when evaluated in sample by dynamic
simulation?
• How is uncertainty added for estimation and more importantly for dynamic
simulation? The result reported in Table 4 seems to take account of only coefficient
uncertainty while, given the modelling choices, the fluctuations in the relevant variables
not explained by the adopted model are likely to be the main source of uncertainty.
I think that the answer to this set of open questions could further enhance the
potential of the interesting methods for valuing ecosystem services very well discussed
in this paper.
Omer Moav
Hebrew University, University of London Royal Holloway, Shalem Center, and CEPR
Edward Barbier demonstrates how basic micro theory can be implemented to
estimate the value of ecological services for human welfare. In particular, two methods
are developed: the production function approach and the expected damage function
approach. Both methods utilize exogenous variation in the size of the ecosystem
service, such as the size of a mangrove forest, on a beneficial outcome. According to
the former the outcome is welfare gained from the decline in the price of a consumption
good, such as fish, that utilizes the ecosystem as an input in its production process.
In the latter it is the economic value of the reduction in damage, arising, for instance,
from storms, that is reduced by a larger ecosystem.
The theory is rather straightforward. It is the availability of the data that the
estimation depends on, and it is not clear that for most practical problems there exists
sufficient exogenous variation in the ecosystem, allowing for a reliable assessment.
Nevertheless, Barbier convincingly illustrates that despite the difficulties, these methods
have the potential to provide important information about the value of the ecosystem
216 EDWARD B. BARBIER
and thereby the value of preventing its disappearance. This information can become
critical for policymakers, and might, even if only in marginal cases, generate the
crucial political force to reverse processes of natural habitat destruction.
The value of the functions of the ecosystem include, as stated by Barbier, ‘climate
stability, maintenance of biodiversity and beneficial species, erosion control, flood
mitigation, storm protection, groundwater recharge and pollution control’. This
statement reveals another limitation of the estimation methods. It focuses on a limited
set of benefits, implying a potentially huge underestimation of the value of the
ecosystem. First, due to information problems regarding most functions, it is difficult
to identify the size of the impact and/or its welfare value. For instance, most likely
exogenous variety in the ecosystem is not sufficient to estimate its effect on climate
stability, and the welfare value of biodiversity is a question hard to answer.
Second, the cost of preserving the ecosystem – giving up the benefits of its
alternative use – is paid by the local population, while many of its services extend
beyond that. As is well known, preserving natural ecosystems is a problem with large
externalities that go beyond borders. In other words, who cares? Do we expect the
poor fisherman in Thailand, or their government, to allocate a significant weight in
its welfare function to biodiversity? In fact, it is the population of the developed world
that cares, and this population’s willingness to pay a compensating price, could be
above and beyond the benefit of the ecosystem to the local population.
A more technical comment on the estimation process regards the open access
assumption and, in particular, the implicit assumption that changes in the habitat are
sufficiently small, relative to the economy, such that the producer’s surplus is
unchanged. Welfare gains from a larger ecosystem emerge only from the reduction
in consumer goods prices. This assumption adds to the bias in the estimation,
reducing the value of the ecosystem. To see this point, suppose that prices of the
consumption good are also given (traded good in an open economy). In this case
there is zero welfare gain from preserving the environment.
A final comment about the estimation method regards the implicit assumption of
stability of the steady-state equilibrium. However, non-monotonic convergence to the
steady state might characterize the dynamics of the ecosystem. For instance, the
population of a species might converge to its steady state in oscillations, implying that
a negative shock to the ecosystem might, once it is sufficiently fragile, result in extinction
of a species rather than a proportional reduction in the size of the natural population.
Beyond the problems of estimating the direct value of the ecological services for
human welfare, lies a somewhat deeper question regarding the long-run effects of the
utilization of natural resources for the benefit of mankind in the production process.
Maintaining natural habitats and benefiting their production services, or destroying
them and benefiting from their alternative land use for agriculture, might have different
long-run consequences on demographic variables, institutional development, and, in
particular, human capital promoting institutions (e.g., public schools, loans, and child
labour regulation), and the resulting accumulation of human capital.
VALUING ECOSYSTEM SERVICES 217
Natural resources, according to many studies, are a hurdle for the process of
development, in particular the accumulation of human capital. (e.g. Gylfason, 2001).
But to the best of my knowledge, we do not know yet how to make a distinction in
that regard between an open-access preserved ecosystem and agricultural land.
Therefore, depleting resources or increasing the size of agricultural land on the
account of the ecosystem, could have a significant impact on the economy.
Moreover, the transition from an open access ecosystem into private owned farmland
might have an impact on wealth inequality, in particular inequality in the ownership
of such land. Deninger and Squire (1998) show that inequality in land ownership has
a negative impact on economic growth. Engerman and Sokoloff (2000) provide evidence
that wealth inequality, brought about oppressive institutions (e.g., restricted access to
the democratic process and to education). They argue that these institutions were
designed to maintain the political power of the elite and to preserve the existing
inequality. Galor et al. (2005) provide evidence that inequality in the ownership of
agricultural land has a negative effect on public expenditure on education, and argue
that the elite of landowners might prevent public schooling, despite the support of the
owners of capital and the working class.
On the other hand, if the destruction of the ecosystem increases farmland and
thereby possibly promoting industrial development, and if the process does not
generate large wealth inequality, the return to human capital will most likely rise.
This could trigger a process of development stemming from reduced fertility and
increased investment in education. This brings us back to the main problem of
preventing the distractions of an ecosystem: the externality. Each small economy
might be better-off destroying the ecosystem, giving rise to an inefficient equilibrium.
The analysis suggested by Barbier, could, at least, highlight the benefits of preserving
natural habitats for the local economy.
APPENDIX: APPLICATION TO THAILAND CASE STUDY
This appendix outlines the econometric estimations for valuing habitat-fishery linkages
and the storm protection service of mangroves in the Thailand case study of Section 4.
A1. Static valuation of habitat-fishery linkage
To apply the static analysis of habitat-fishery linkages of Section 3.3.1 to the Thailand
case study, it is necessary to estimate the unknown parameters (A, a, b) of the log-
linear version of the Cobb–Douglas production function:
ln hit = A0 + a ln Eit + b ln Mit + µit (A1)
where i = 1, . . . , 5 zones, t = 1, . . . , 14 years (1983–96) and A0 = ln A.
Equation (A1) was estimated using the pooled data on demersal fisheries, shellfish
and mangrove area from Barbier (2003). These were the data on harvest, hit, and
218 EDWARD B. BARBIER
effort, Eit, for Thailand’s shellfish and demersal fisheries, as well as mangrove area,
Mit across the five coastal zones of Thailand and over the years1983–96. Various
regression procedures for a pooled data set were utilized and compared, including:
(i) ordinary least squares (OLS); (ii) one- and two-way panel analysis of fixed and
random effects; and (iii) a maximum likelihood estimation by an iterated generalized
least squares (GLS) procedure for a pooled time series and cross-sectional regression,
which allows for correction of any groupwise heteroscedasticity, cross-group correlation
and common or within-group autocorrelation. Table A1 indicates the best regression
model for the shellfish and demersal fisheries respectively, and the relevant test statistics.
For demersal fisheries, the preferred model shown in Table A1 is the GLS estimation
allowing for groupwise heteroscedasticity and correcting for both cross-group and
common autocorrelation. For the panel analysis of the demersal fisheries, the likelihood
ratio tests of the null hypothesis of zero individual and time effects across all five zones
and fourteen time periods were significant, thus rejecting the null hypothesis. In
addition, the Breusch–Pagan Lagrange multiplier (LM) statistic was also significant at
the 95% confidence level for both the one-way and two-models, which suggests
rejection of the null hypothesis of zero random disturbances. The Hausman test
statistic was also significant at the 99% confidence level, suggesting that the fixed
effects specification is preferred to the random effects. However, in both the one- and
two-way fixed effects model the t-test on the estimated parameter for a in Equation
(A1) was insignificant, suggesting the null hypothesis that a = 0 cannot be rejected.
As indicated in Table A1, from the pooled time series cross-sectional GLS regression
for demersal fisheries, the likelihood ratio (LR) test statistic of the null hypothesis for
homoscedasticity based on the least squares regression was computed to be 24.64,
which is statistically significant. Although not shown in the table, the alternative Wald
test for homoscedasticity is also statistically significant and confirms rejection of the
null hypothesis. Thus the GLS model with correction of groupwise heteroscedasticity
is preferred to the OLS regression. The LM statistic of 14.43 also reported in Table
A1 for demersal fisheries is a test of the null hypothesis of zero cross-sectional correlation,
which proves to be statistically significant. Although not indicated in the table, the
LR test statistic for groupwise heteroscedasticity as a restriction on cross-group
correlation was estimated to be 23.26, which is also statistically significant. Thus the
null hypothesis of zero cross-group correlation in the demersal fisheries regression can
be rejected. The common autocorrelation coefficient across all five zones was estimated
to be 0.484, and as shown in Table A1, once the GLS model for demersal fish was
corrected for this common autocorrelation, the null hypothesis that the coefficient a
= 0 is now rejected.
For shellfish, as indicated in Table A1 the preferred estimation of Equation (A1) is
the GLS estimation allowing for groupwise heteroscedasticity and correcting for
cross-group correlation, with A0 restricted to zero. For the panel analysis of shellfish,
the likelihood ratio tests and Breusch–Pagan LM tests of the null hypothesis of no
individual and time effects were significant, thus rejecting the null hypothesis. The
VALUING ECOSYSTEM SERVICES 219
Table A1. Estimates of Equation (A1) for Thailand’s shellfish and demersal
fisheries
Demersal fisherya Shellfish fisheryb
Coefficient
A0 11.213 (24.568)** –
A 0.341 (4.992)** 1.688 (38.254)**
B 0.100 (2.763)** 0.196 (3.693)**
Log-likelihoodc 5.401 –71.517
Likelihood ratio statisticd 24.643** 35.076**
Lagrange multiplier statistice 14.426* 21.304**
Notes: t-statistics are shown in parentheses.
a
Preferred model is groupwise heteroscedastic and correlated GLS, corrected for common autocorrelation.
b
Preferred model is groupwise heteroscedastic and correlated GLS, with A0 restricted to zero.
In the demersal fishery regression, correction of cross-group correlation Cov[eit,ejt] = σij leads to a positive
c
log-likelihood.
d
Tests the null hypothesis of homoscedasticity based on OLS.
e
Tests the null hypothesis of zero cross-group correlation based on OLS.
* Significant at 95% confidence level.
** Significant at 99% confidence level.
Sources: Author’s estimations.
Hausman test statistic was significant, suggesting that the fixed effects specification is
preferred to the random effects. However, in both the one- and two-way fixed effects
model the t-test on the estimated parameters for a and b in Equation (A1) was
insignificant, suggesting the null hypothesis a = b = 0 cannot be rejected. As indicated
in Table A1, from the pooled time series cross-sectional GLS regression of shellfish,
the LR test statistic of the null hypothesis for homoscedasticity based on the least
squares regression is 35.08, which is statistically significant. Although not shown in
the table, the alternative Wald test for homoscedasticity is also statistically significant
and confirms rejection of the null hypothesis. Thus the GLS model with correction
of groupwise heteroscedasticity is preferred to the OLS regression. The LM statistic
of 21.30 also reported in Table A1 for the shellfish regression is a test of the null
hypothesis of zero cross-sectional correlation, which proves to be statistically signifi-
cant. Although not indicated in the table, the LR test statistic for groupwise hetero-
scedasticity as a restriction on cross-group correlation was estimated to be 43.90,
which is also statistically significant. Thus the null hypothesis of zero cross-group
correlation in the shellfish regression can be rejected. As shown in Table A1, once
the GLS model of shellfish was corrected for groupwise heteroscedasticity and corre-
lation, the null hypotheses that a = 0 and b = 0 are now rejected.
The estimations of Equation (A1) for Thailand’s shellfish and demersal fisheries
were used in conditions (5) and (6) to calculate the welfare impacts of mangrove
deforestation over 1996–2004 on Thailand’s artisanal fisheries. The analysis uses the
same iso-elastic demand function as in Barbier (2003), with a demand elasticity, ε,
of −0.5. The results are reported in Table 3, which shows welfare calculations for
both the point estimates and upper and lower bounds on these estimates based on
the standard errors of the regression coefficients reported in Table A1.
220 EDWARD B. BARBIER
A2. Dynamic valuation of habitat-fishery linkage
The dynamic habitat-fishery modelling approach to valuing the habitat-fishery link-
age is outlined in Section 3.3.2. The main difficulty in applying this approach to
valuing mangrove-fishery linkages in Thailand is that data do not exist for the bio-
mass stock, Xt, of near-shore fisheries. Thus the appropriate system of equations to
estimate comprises (10) and (11). Because Et and ct are predetermined, both of these
equations can be estimated independently (Homans and Wilen, 1997). For both the
shellfish and demersal fisheries, the estimated equations are:
Eit = a0 + a1Rit −1 + a2 Eit −1 + µit −1 (A2)
cit − cit −1 c
= b0 + b1 it −1 + b2 Eit −1 + µit −1, (A3)
cit −1 ln M it −1
where i = 1, . . . , 5 zones, t − 1 = 1, . . . , 13 years (1983–95), Rit −1 = khit+η , a1 = φ, a2 =
1
−1
(1 − φw), b0 = r, b1 = −r/αq and b2 = −q. Both equations were estimated using the
pooled data on demersal fisheries, shellfish and mangrove area from Barbier (2003).
These were the data on harvest, hit−1, and effort, Eit−1, for Thailand’s shellfish and
demersal fisheries, as well as mangrove area, Mit−1 across the five coastal zones of
Thailand and over the years 1983–96. In addition, to calculate Rit−1 from hit−1 the
elasticity η = 1/ε = −2 was assumed as in the static analysis. Various regression
procedures for a pooled data set were utilized and compared, including: (i) OLS; (ii)
one- and two-way panel analysis of fixed and random effects; and (iii) a maximum
likelihood estimation by an iterated GLS procedure for a pooled time series and
cross-sectional regression, which allows for correction of any groupwise heteroscedas-
ticity, cross-group correlation and common or within-group autocorrelation. Tables A2
and A3 indicate the best regression models of Equations (A2) and (A3) for the shellfish
and demersal fisheries respectively, and the relevant test statistics.
For demersal fisheries, the preferred model for the effort equation (A2) is the GLS
estimation allowing for groupwise heteroscedasticity and corrected for common
autocorrelation. For the panel analysis, the likelihood ratio and Breusch–Pagan LM
tests of the null hypothesis of no individual and time effects were not significant; thus,
the null hypothesis cannot be rejected. However, as indicated in Table A2, from the
pooled time series cross-sectional regression of Equation (A2) for demersal fisheries,
the LR test statistic of the null hypothesis of homoscedasticity based on the OLS
regression is computed to be 93.22, which is statistically significant. Although not
shown in the table, the alternative Wald test for homoscedasticity is also statistically
significant and confirms rejection of the null hypothesis. Thus the GLS model with
correction of groupwise heteroscedasticity is preferred to the OLS regression. The
test statistics for the null hypothesis of zero cross-group correlation are mixed. The
LM statistic of 12.24 indicated in Table A2 is significant, whereas the LR test statistic
of 12.56 is not. When the GLS regression is corrected for groupwise correlation,
VALUING ECOSYSTEM SERVICES 221
however, the constant term a0 is no longer significant. The common autocorrelation
coefficient across all five zones is estimated to be 0.242, and although slight, correc-
tion of this autocorrelation improved the overall robustness of the GLS estimation.
As shown in Table A2, the preferred model for the effort equation (A2) for shellfish
is the one-way random effects estimation corrected for heteroscedasticity. The LR
and Wald tests of the pooled time series cross-sectional regressions of Equation (A2)
for shellfish indicated that the null hypothesis of homoscedasticity can be rejected.
Thus the GLS model with correction of groupwise heteroscedasticity is preferred to
the OLS regression. However, in all versions of the GLS regression the coefficient a1
was negative and statistically insignificant. The LR test for the presence of individual
effects is statistically significant, thus rejecting the null hypothesis of no such effects,
and although not shown, the equivalent F-test of the null hypothesis is also statistically
significant. Neither the Breusch–Pagan LM test of the null hypothesis of random
provincial-level disturbances nor the Hausman test of the random versus the fixed
effects specification is statistically significant. Although these results are somewhat
contradictory, they suggest that, if individual effects are present, they are likely to be
random. The LR test and F-test of the presence of time effects is not significant,
suggesting that the one-way is preferred to the two-way specification. Correction of
heteroscedasticity improves the robustness of the one-way random effects estimation
Table A2. Estimates of Equation (A2) for Thailand’s shellfish and demersal
fisheries
Demersal fisherya Shellfish fisheryb
Coefficient
a0 22.365 (2.254)* 808.720 (2.661)**
a1 0.00004 (4.375)** 0.000003 (0.233)
a2 0.84855 (21.703)** 0.70470 (8.183)**
Log-likelihood –380.903 –520.513
Likelihood ratio statistic for 93.223** –
homoscedasticityc
Likelihood ratio statistic for correlationd 12.552 –
Lagrange multiplier statistice 12.241* –
Likelihood ratio statistic for individual effectsf – 16.285**
Breusch–Pagan Lagrange multiplier statisticg – 0.04
Hausman test statistich – 1.88
Notes: t-statistics are shown in parentheses.
a
Preferred model is groupwise heteroscedastic GLS, corrected for common autocorrelation.
b
Preferred model is one-way random effects corrected for heteroscedasticity.
c
Tests the null hypothesis of homoscedasticity based on OLS.
d
Tests the null hypothesis of zero cross-group correlation based on OLS.
e
Tests the null hypothesis of zero cross-group correlation based on OLS.
f
Tests the null hypothesis of zero individual effects.
g
Tests the null hypothesis of zero random disturbances based on OLS.
h
Tests the null hypothesis of correlation between the individual effects and the error (i.e. random effects is
preferred to fixed effects estimation).
* Significant at 95% confidence level.
** Significant at 99% confidence level.
Sources: Author’s estimations.
222 EDWARD B. BARBIER
without affecting the parameter estimates. Although not shown in the table, the
preferred model displayed a very low estimated autocorrelation of 0.022.
As indicated in Table A3, the preferred model for the growth in catch per unit
effort equation (A3) for demersal fisheries is the GLS estimation allowing for group-
wise heteroscedasticity. For the panel analysis, the LR and Breusch–Pagan LM tests
of the null hypothesis of no individual and time effects were not significant; thus, the
null hypothesis cannot be rejected. However, from the pooled time series cross-
sectional regression, both the LR and Wald test statistics of the null hypothesis of
homoscedasticity are also statistically significant. Thus the GLS model with correction of
groupwise heteroscedasticity is preferred to the OLS regression. The test statistics for
the null hypothesis of zero cross-group correlation are mixed. The LM statistic of
11.03 indicated in Table A3 is significant, whereas the LR test statistic of 18.56 is
not. However, correcting the GLS regression for groupwise correlation does not affect
the estimation significantly. Although not shown in the table, the preferred model
displayed a very low estimated autocorrelation of −0.006.
Table A3 displays the preferred model for the growth in CPE equation (A3) for
shellfish, which is the GLS estimation allowing for groupwise and correlated hetero-
scedasticity and corrected for common autocorrelation. For the panel analysis, the
LR and Breusch–Pagan LM tests of the null hypothesis of no individual and time
effects were not significant; thus, the null hypothesis cannot be rejected. However,
from the pooled time series cross-sectional regression, both the LR and Wald test
statistics of the null hypothesis of homoscedasticity are also statistically significant.
Thus the GLS model with correction of groupwise heteroscedasticity is preferred to
the OLS regression. Although the LR and LM test statistics for the null hypothesis
Table A3. Estimates of Equation (A3) for Thailand’s shellfish and demersal
fisheries
Demersal fisherya Shellfish fisheryb
Coefficient
b0 0.4896 (2.908)** 0.2997 (2.371)*
b1 –0.000187 (–2.368)* –0.000201 (–2.354)*
b2 –0.000204 (–2.637)** –0.000060 (–2.007)*
Log-likelihood –22.337 –30.350
Likelihood ratio statistic for 24.627** 109.342**
homoscedasticityc
Likelihood ratio statistic for correlationd 18.235 11.434
Lagrange multiplier statistice 11.026* 8.491
Notes: t-statistics are shown in parentheses.
a
Preferred model is groupwise heteroscedastic GLS.
b
Preferred model is groupwise heteroscedastic and correlated GLS, corrected for common autocorrelation.
c
Tests the null hypothesis of homoscedasticity based on OLS.
d
Tests the null hypothesis of zero cross-group correlation based on OLS.
e
Tests the null hypothesis of zero cross-group correlation based on OLS.
* Significant at 95% confidence level.
** Significant at 99% confidence level.
Sources: Author’s estimations.
VALUING ECOSYSTEM SERVICES 223
of zero cross-group correlation are not significant, correcting the GLS regression for
groupwise correlation improves the significance confidence level of the estimated
parameter b2 from 90 to 95%. The common autocorrelation coefficient across all five
zones is estimated to be 0.147, and although slight, correction of this autocorrelation
improved the overall robustness of the GLS estimation.
Using the estimated parameters for Equations (A2) and (A3) for Thailand’s shellfish
and demersal fisheries allows simulation of the welfare impacts of mangrove deforestation
over 1996–2004 on Thailand’s artisanal fisheries. Again, the same demand function
with elasticity of −0.5 as in Barbier (2003) is employed. The results are reported in
Table 3, which shows welfare calculations for both the point estimates and upper and
lower bounds on these estimates based on the standard errors of the regression
coefficients reported in Tables A2 and A3.
A3. Expected damage function valuation of storm protection service
As discussed in Section 3.5, the key step in applying the expected damage function
approach to valuing the storm protection service of a coastal wetland such as mangroves
is to estimate how a change in mangrove area influences the expected incidence of
economically damaging natural disaster events.
Suppose that for a number of coastal regions, i = 1, . . . , N, and over a given period
of time, t = 1, . . . , T the ith coastal region could experience in any period t any number
of zit = 0, 1, 2, 3 . . . economically damaging storm event incidents. A common assumption
in count data models is that the count variable zit has a Poisson distribution, in which
case the expected number of storm events in each region per period is given by:
∂E[zit| it , xit ]
S
E[zit|sit , xit ] = λit = e α +β S +β ′x , = λit βS (A4)
i S it it
∂Sit
where as before, Sit is the area of wetlands, xit, are other factors, and αi accounts for
other possible ‘unobserved’ effects on the incidence of disasters specific to each
coastal region. Estimation of βS, along with an estimate of the conditional mean λit,
allows ∂Ζ/∂S in Equation (13) to be determined. One drawback of the Poisson dis-
tribution (Equation (A4)) is that it automatically implies ‘equidispersion’, that is, the
conditional variance of zit is also equal to λit. To test whether this is the case, the
Poisson method of estimating (A4) should be compared to other techniques, such as
the Negative Binomial model, which do not assume equidispersion in the variance of zit.
For the Thailand case study, the estimation of (A4) is:
ln E[zit|Mit, xit ] = ln λit = αi + βS Mit + β′xit + µit, (A5)
where i = 1, . . . , 21 coastal provinces, t = 1, . . . , 18 years (1979 –96). The EM-DAT
(2005) International Disaster Database contains data on the number of coastal disas-
ters occurring in Thailand since 1975 and the approximate location and date of its
impacts. From these data it is possible to determine zit, the number of economically
224 EDWARD B. BARBIER
damaging coastal natural disasters that occurred per province per year over 1979–
96. Mangrove area, Mit, is measured in terms of the annual mangrove area in square
kilometres for each of the 21 coastal provinces of Thailand over 1979–96. Two
control variables were included as the additional factors, xit, which may explain the
incidence of economically damaging coastal disasters, the population density of a
province and a yearly time trend variable. The inclusion of the population density
variable reflects the prevailing view in the natural disaster management literature that
‘hazard events that occur in unpopulated areas and are not associated with losses do
not constitute disasters’ (Dilley et al., 2005, p. 115).11 The yearly time trend was
included as a control because the number of coastal natural disasters seems to have
increased over time in Thailand (see Figure 5).12
Various regression procedures for a panel data set for count data models were
utilized and compared, including: (1) Poisson models assuming equidispersion, i.e.
equality of the conditional mean and the variance; (2) maximum likelihood estimation
of Negative Binomial models allowing for unequal dispersion; and (3) comparing
provincial to zonal fixed effects. Table A4 reports the best count data model for
estimating Equation (A5) and the relevant test statistics.
As shown in Table A4, the preferred specification of the count data model is the
Negative Binomial model with zonal fixed effects. In both the Poisson and Negative
Binomial panel models, the zonal fixed effects specification (with coastal zone 5 as
the default) is preferred to individual province effects, which is verified by LR tests of
the two specifications. Although the parameter estimates for the zonal fixed effects
are not shown, these estimated effects were significant at the 95% or 99% confidence
levels. As indicated in Table A4, two standard tests were employed for the null
hypothesis of equidispersion of the conditional mean and variance of the Poisson
specification of the count data model (Cameron and Trivedi, 1998; Greene, 2003).
Both the LM statistic and the t-test for equidispersion based on the residuals of the
Poisson regression are significant, indicating that the null hypothesis can be rejected,
and the Negative Binomial model that does not assume equidispersion is preferred
to the Poisson specification. The LR statistic reported in the table tests the null
hypothesis that the coefficients of the regressors are zero; as the statistic is significant,
the hypothesis is rejected.
The results displayed in Table A4 for the preferred model show that a change in
mangrove area has a significant influence on the incidence of coastal natural disasters
in Thailand, and with the predicted sign. The point estimate for βS indicates that a
1 km2 decline in mangrove area increases the expected number of disasters by 0.36%.
11
This view is also reflected in the criteria used in the International Disaster Database to decide which hazard events should
be recorded as ‘natural disasters’. In order for EM-DAT (2005) to record an event as a disaster, at least one or more of the
following criteria must be fulfilled: 10 or more people reported killed; 100 people reported affected; declaration of a state of
emergency; call for international assistance. The simple correlation between population density and mangrove area for the
sample is relatively low (−0.389).
12
This is a procedure recommended by Rose (1990), when such a trend effect is suspected.
VALUING ECOSYSTEM SERVICES 225
Table A4. Negative binomial estimation of Equation (A5) with zonal fixed effects
Parameter estimatea Marginal effectb
Variable
Mangrove area (Mit) –0.0036 (–4.448)** –0.0031 (–2.745)**
Population density (POPDENit) –0.0005 (–1.079) –0.0004 (–0.894)
Annual time trend (YRTRNit) 0.0781 (5.558)** 0.0669 (2.615)**
Dispersion parameter (αit) 0.0001
Estimated conditional mean (λ) 0.8559
Log-likelihood – 373.66
Lagrange multiplier statisticc 39.967**
Regression t-testd – 5.385**
Likelihood ratio statistice 74.919**
Notes: t-statistics shown in parentheses.
a
Parameter estimates for the zonal fixed effects are not shown. Zone 5 is the default and the fixed effects for
zones 1 to 4 were negative and significant at the 95% or 99% confidence levels.
Estimate of λitβ3 (see Equation (A4)).
b
c
Tests the null hypothesis of equidispersion in the Poisson model.
d
A regression-based test of the null hypothesis of equidispersion in the Poisson model.
e
Tests the null hypothesis that the restricted regression without the explanatory variables Mit, POPDENit and
YRTRNit is the preferred Negative Binomial model with zonal fixed effects.
* Significant at 95% confidence level.
** Significant at 99% confidence level.
Sources: Author’s estimations.
It is likely that the mangrove loss in Thailand, especially since the mid-1970s (see
Figure 4), has increased the expected number of economically damaging coastal
natural disasters per year. The estimated marginal effect corresponding to βS of a
change in mangrove area on coastal natural disasters (−0.0031) can be employed to
estimate the resulting impact of mangrove deforestation over 1979–96 in Thailand
on expected damages of natural coastal disasters. This is described further in Section
4.3 and shown in Table 4.
As discussed in Section 3.5, an underlying hypothesis of the expected damage
function methodology is that, if coastal wetland loss increases the incidence of natural
disaster per year, then wetland loss is also associated with increasing storm damages.
However, under certain circumstances the results of a count data model could provide a
misleading test of this null hypothesis. For instance, suppose a loss in wetland area is
associated with a change in the incidence of storms from one devastating storm to
two relatively minor storms per year. The count data model would then be inter-
preted as not providing evidence against the null that the change in the wetland area
increases expected storm damages, when what has actually happened is that total
storm damages have declined over time with wetland loss. This suggests the need for
a robustness check on the count data model, such as Equation (A5) in the Thailand
case study, to ensure that such situations do not dominate the application of the EDF
approach.
One possible robustness check is to test the null hypothesis directly; that is, are total
damages from storm events increasing with coastal wetland loss? In the Thailand case
study, the relevant estimation is
226 EDWARD B. BARBIER
Dit = αi + βS Mit + β′xit + µit (A6)
where the dependent variable, Dit, is total real damages from all storm events per
province per year over 1979–96. The EM-DAT (2005) database provides data on the
total economic damages per province per year in Thailand, and these data were
deflated using the 1996 GDP deflator. The standard regression procedures for the
panel analysis of Equation (A6) were performed, including comparing OLS with
fixed and random effects. Table A5 reports the OLS and random effects specifica-
tions for the preferred version of Equation (A6).
The preferred model in Table A5 is the pooled weighted least squares estimation
with correction for heteroscedasticity. The LR test of the null hypothesis of zero
individual effects across all 21 provinces is not statistically significant. Although not
shown in the table, an alternative F-test of the null hypothesis is also not significant.
Neither the Breusch–Pagan LM test of the null hypothesis of random provincial-level
disturbances nor the Hausman test of the random versus the fixed effects specification
is statistically significant. These tests confirm that in the panel analysis of Equation
(A5) of the weighted OLS regression is more efficient than either the random or fixed
effects models.
The weighted least squares regression in Table A5 indicates that, over 1979–96
and across the 21 coastal provinces of Thailand, total real storm damages increased
with mangrove loss. The point estimate suggests that a 1 km2 decline in mangrove
area increases real storm damages by around $52 per province per year. The regres-
sion also confirms that, for the Thailand case study, the null hypothesis that storm
damages increase with mangrove loss cannot be rejected.
Table A5. Panel estimation of Equation (A6) for total storm damages, Thailand
Variablea Pooled OLSb Random effectsb
Mangrove area (Mit) – 51.527 (–1.976)* –52.378 (–1.563)
Population density (POPDENit) – 12.896 (–0.343) –18.723 (–0.395)
Annual time trend (YRTRNit) 965.325 (2.058)* 983.653 (2.100)*
Constant 1 3748.820 (1.275) 1 4728.707 (1.153)
Log-likelihood –4598.425
Likelihood ratio statisticc 14.173
Lagrange multiplier statisticd 2.24
Hausman teste 1.04
Notes: t-statistics shown in parentheses.
a
Parameter estimates for the zonal fixed effects for Zone 1 and Zone 4 are not shown. Although neither
parameter was statistically significant, their inclusion improved the robustness of the overall regression.
b
Weighted least squares with robust covariance matrix to correct for heteroscedasticity.
c
Tests the null hypothesis of no fixed provincial effects.
d
Tests the null hypothesis of no random provincial effects.
e
Tests the null hypothesis that the random effects specification is preferred to the fixed effects. Test was
performed excluding the zonal fixed effect for Zone 4.
* Significant at 95% confidence level.
Sources: Author’s estimations.
VALUING ECOSYSTEM SERVICES 227
REFERENCES
Acharya, G. and E.B. Barbier (2000). ‘Valuing groundwater recharge through agricultural
production in the Hadejia-Jama’are wetlands in northern Nigeria’, Agricultural Economics, 22,
247–59.
Aksornkoae, S. and R. Tokrisna (2004). ‘Overview of shrimp farming and mangrove loss in
Thailand’, in E.B. Barbier and S. Sathirathai (eds.), Shrimp Farming and Mangrove Loss in
Thailand, Edward Elgar, London.
Aksornkoae, S., R. Tokrisna, W. Sugunnasil and S. Sathirathai (2004). ‘The importance of
mangroves: Ecological perspectives and socio-economic values’, in E.B. Barbier and S.
Sathirathai (eds.), Shrimp Farming and Mangrove Loss in Thailand, Edward Elgar, London.
Barbier, E.B. (1994). ‘Valuing environmental functions: Tropical wetlands’, Land Economics,
70(2), 155–73.
— (2000). ‘Valuing the environment as input: Applications to mangrove-fishery linkages’,
Ecological Economics, 35, 47–61.
— (2003). ‘Habitat-fishery linkages and mangrove loss in Thailand’, Contemporary Economic
Policy, 21(1), 59–77.
Barbier, E.B. and M. Cox (2003). ‘Does economic development lead to mangrove loss? A
cross-country analysis’, Contemporary Economic Policy, 21(4), 418–32.
Barbier, E.B. and S. Sathirathai (eds.) (2004). Shrimp Farming and Mangrove Loss in Thailand.
Edward Elgar, London.
Barbier, E.B. and I. Strand (1998). ‘Valuing mangrove-fishery linkages: A case study of
Campeche, Mexico’, Environmental and Resource Economics, 12, 151–66.
Barbier, E.B., I. Strand and S. Sathirathai (2002). ‘Do open access conditions affect the
valuation of an externality? Estimating the welfare effects of mangrove-fishery linkages’,
Environmental and Resource Economics, 21(4), 343–67.
Bjørndal, T. and J.M. Conrad (1987). ‘The dynamics of an open access fishery’, Canadian
Journal of Economics, 20, 74–85.
Boscolo, M. and J.R. Vincent (2003). ‘Nonconvexities in the production of timber, biodiversity,
and carbon sequestration’, Journal of Environmental Economics and Management, 46, 251–68.
Cameron, C.A. and P. Trivedi (1998). Regression Analysis of Count Data, Cambridge University
Press, Cambridge.
Carlsson, F., P. Frykblom and C. Lijenstolpe (2003). ‘Valuing wetland attributes: An applica-
tion of choice experiments’, Ecological Economics, 47, 95–103.
Carpenter, S.R., D. Ludwig and W.A. Brock (1999). ‘Management of eutrophication for lakes
subject to potentially irreversible change’, Ecological Applications, 9(3), 751–71.
Charuppat, T. and J. Charuppat (1997). The Use of Landsat-5 (TM) Satellite Images for Tracing the
Changes of Mangrove Forest Areas of Thailand, Royal Forestry Department, Bangkok.
Check, E. (2005). ‘Roots of recovery’, Nature, 438, 910–11.
Chong, J. (2005). Protective Values of Mangrove and Coral Ecosystems: A Review of Methods and Evidence,
IUCN, Gland, Switzerland.
Clark, C. (1976). Mathematical Bioeconomics, John Wiley and Sons, New York.
Dahdouh-Guebas, F., L.P. Jayatissa, D. Di Nitto, J.O. Bosire, D. Lo Seen and N. Koedam (2005). ‘How
effective were mangroves as a defence against the recent tsunami?’ Current Biology, 15(12), 443–47.
Daily, G. (ed.) (1997). Nature’s Services: Societal Dependence on Natural Ecosystems, Island Press,
Washington DC.
De Groot, R.S., M.A. Wilson and R.M.J. Boumans (2002). ‘A typology for the classification,
description and valuation of ecosystem functions, goods and services’, Ecological Economics,
41, 393–408.
Deninger, K. and L. Squire (1998). ‘New ways of looking at old issues: Inequality and growth’,
Journal of Development Economics, 57, 259–87.
Dickie, M. (2003). ‘Defensive behavior and damage cost methods’, in P. Champ, K.J. Boyle
and T.C. Brown (eds.), A Primer on Nonmarket Valuation, Kluwer, Boston, MA.
Dilley, M., R.S. Chen, U. Deichmann, A.L. Lerner-Lam and M. Arnold (2005). Natural Disaster
Hotspots: A Global Risk Analysis. Disaster Risk Management Series, Hazard Management
Unit, The World Bank, Washington, DC.
Ellis, G.M. and A.C. Fisher (1987). ‘Valuing the environment as input’, Journal of Environmental
Management, 25, 149–56.
228 EDWARD B. BARBIER
EM-DAT (2005). EM-DAT: The OFDA/CRED International Disaster Database. www.em-dat.net,
Université Catholique de Louvain, Brussels, Belgium.
Engerman, S. and Sokoloff, K.L. (2000). ‘Factor endowment, inequality, and paths of development
among new world economies’, UCLA.
FAO (2003). ‘Status and trends in mangrove area extent worldwide’ (by M.L. Wilkie and S.
Fortuna) Forest Resources Assessment Working Paper No. 63, Forest Resources Division,
Food and Agricultural Organization of the United Nations, Rome.
Finnoff, D. and J. Tschirhart (2003). ‘Harvesting in an eight-species ecosystem’, Journal of
Environmental Economics and Management, 45, 589–611.
Freeman, A.M. III. (1982). Air and Water Pollution Control: A Benefit-Cost Assessment, John Wiley,
New York.
— (1991). ‘Valuing environmental resources under alternative management regimes’, Ecological
Economics, 3, 247–56.
— (2003). The Measurement of Environmental and Resource Values: Theory and Methods, 2nd edn,
Resources for the Future, Washington DC.
Galor, O., O. Moav and D. Vollrath (2005). ‘Inequality in land ownership, the emergence of
human capital promoting institutions, and the great divergence’, Brown University.
Greene, W.H. (2003). Econometric Analysis, 5th edn, Prentice-Hall, Englewood Cliffs, NJ.
Gren, I-M., K. Elofsson and P. Jannke (1997). ‘Cost-effective nutrient reductions to the Baltic
Sea’, Environmental and Resource Economics, 10, 341–62.
Gylfason, T. (2001). ‘Natural resources, education, and economic development’, European
Economic Review, 45, 847–59.
Harakunarak, A. and S. Aksornkoae (2005). ‘Life-saving belts: Post-tsunami reassessment of
mangrove ecosystem values and management in Thailand’, Tropical Coasts, July, 48–55.
Heal, G.M., E.B. Barbier, K.J. Boyle, A.P. Covich, S.P. Gloss, C.H. Hershner, J.P. Hoehn,
C.M. Pringle, S. Polasky, K. Segerson and K. Shrader-Frechette (2005). Valuing Ecosystem Services:
Toward Better Environmental Decision Making, The National Academies Press, Washington DC.
Hirshleifer, J. and J.G. Riley (1992). The Analytics of Uncertainty and Information, Cambridge
University Press, Cambridge.
Homans, F.R. and J.E. Wilen (1997). ‘A model of regulated open access resource use’, Journal
of Environmental Economics and Management, 32, 1–21.
Jackson, J.B.C., M.X. Kirby, W.H. Berger, et al. (2001). ‘Historical overfishing and the recent
collapse of coastal ecosystems’, Science, 293, 629–38.
Just, R.E., D.L. Hueth and A. Schmitz (2004). The Welfare Economics of Public Policy: A Practical
Approach to Project and Policy Evaluation, Edward Elgar, Cheltenham, UK.
Kahn, J.R., and W.M. Kemp (1985). ‘Economic losses associated with the degradation of an
ecosystem: The case of submerged aquatic vegetation in Chesapeake Bay’, Journal of Envi-
ronmental Economics and Management, 12, 246–63.
Kaiser, B. and J. Roumasset (2002). ‘Valuing indirect ecosystem services: The case of tropical
watersheds’, Environment and Development Economics, 7, 701–14.
Kaosa-ard, M. and S. Pednekar (1998). ‘Background Report for the Thai Marine Rehabilitation
Plan 1997–2001’, Report submitted to the Joint Research Centre of the Commission of the
European Communities and the Department of Fisheries, Ministry of Agriculture and
Cooperatives, Thailand, Thailand Development Research Institute, Bangkok.
Kathiresan, K. and N. Rajendran (2005). ‘Coastal mangrove forests mitigated tsunami’,
Estuarine Coastal and Shelf Science, 65, 601–606.
Knowler, D., E.B. Barbier, and I. Strand (2001). ‘An open-access model of fisheries and
nutrient enrichment in the Black Sea’, Marine Resource Economics, 16, 195–217.
Krutilla, J.V. and A.C. Fisher (1985). The Economics of Natural Environments: Studies in the Valuation
of Commodity and Amenity Resources, Resources for the Future, Washington, DC.
Lynne, G.D., P. Conroy and F.J. Prochaska (1981). ‘Economic value of marsh areas for marine
production processes’, Journal of Environmental Economics and Management, 8, 175–86.
Mardle, S., C. James, C. Pipitone and M. Kienzle (2004). ‘Bioeconomic interactions in an
established fishing exclusion zone: The Gulf of Castellammare, NW Sicily’, Natural Resource
Modeling, 17(4), 393–447.
Massel, S.R., K. Furukawa and R.M. Brinkman (1999). ‘Surface wave propagation data in
mangrove forests’, Fluid Dynamics Research, 24, 219–49.
Mazda, Y., E. Wolanski, B. King, A. Sase, D. Ohtsuka and M. Magi (1997). ‘Drag force due
to vegetation in mangrove swamps’, Mangroves and Salt Marshes, 1, 193–99.
VALUING ECOSYSTEM SERVICES 229
McConnell, K.E. and I.E. Strand (1989). ‘Benefits from commercial fisheries when demand and
supply depend on water quality’, Journal of Environmental Economics and Management, 17, 284–92.
Michener, R. and C. Tighe (1992). ‘A Poisson regression model of highway fatalities’, American
Economic Review, 82(2), 452–56.
Millennium Ecosystem Assessment (2003). Ecosystems and Human Well-being: A Framework for
Assessment, Island Press, Washington DC.
Mitsch, W.J. and J.G. Gosselink (1993). Wetlands, 2nd edn, Van Norstrand Reinhold, New York.
Myers, R.A. and B. Worm (2003). ‘Rapid worldwide depletion of predatory fish communities’,
Nature, 423, 280–83.
Naylor, R. and M. Drew (1998). ‘Valuing mangrove resources in Kosrae, Micronesia’, Environ-
ment and Development Economics, 3, 471–90.
Olson, M.K. (2004). ‘Are novel drugs more risky for patients than less novel drugs?’ Journal of
Health Economics, 23, 1135–58.
Pagiola, S., K. von Ritter and J. Bishop (2004). How Much is an Ecosystem Worth? Assessing the
Economic Value of Conservation, The World Bank, Washington DC.
Parks, P. and M. Bonfaz (1997). ‘Nonsustainable use of renewable resources: Mangrove
deforestation and mariculture in Ecuador’, Marine Resource Economics, 9, 1–18.
Ready, R.C. (1995). ‘Environmental valuation under uncertainty’, in D.W. Bromley (ed.), The
Handbook of Environmental Economics, Blackwell Publishers, Cambridge, MA, pp. 568–93.
Ricketts, T.H., G.C. Daily, P.R. Ehrlich and C.D. Michener (2004). ‘Economic value of tropical
forest to coffee production’, Proceedings of the National Academy of Science, 101(304), 12579–82.
Rodwell, L.D., E.B. Barbier, C.M. Roberts and T.R. McClanahan (2002). ‘A model of tropical
marine reserve-fishery linkages’, Natural Resource Modeling, 15(4), 453–86.
Rose, N.L. (1990). ‘Profitability and product quality: Economic determinants of airline safety
performance’, Journal of Political Economy, 98(5), 944–64.
Sathirathai, S. and E.B. Barbier (2001). ‘Valuing mangrove conservation in Southern
Thailand’, Contemporary Economic Policy, 19, 109–22.
Schnute, J. (1977). ‘Improved estimates of the Schaefer Production Model: Theoretical
considerations’, Journal of the Fisheries Research Board of Canada, 34, 583–603.
Settle, C. and J.F. Shogren (2002). ‘Modeling native-exotic species within Yellowstone Lake’,
American Journal of Agricultural Economics, 84(5), 1323–28.
Shabman, L.A. and S.S. Batie (1978). ‘Economic value of natural coastal wetlands: A critique’,
Coastal Zone Management Journal, 4(3), 231–47.
Sugunnasil, W. (1998). ‘Fishing communities and coastal resource management in Southern
Thailand’, mimeo, Department of Social Sciences, Prince of Songkla University, Pattani, Thailand.
Sumaila, U.R. (2002). ‘Marine protected area performance in a model of a fishery’, Natural
Resource Modeling, 15(4), 439–51.
Swallow, S.K. (1994). ‘Renewable and nonrenewable resource theory applied to coastal
agriculture, forest, wetland and fishery linkages’, Marine Resource Economics, 9, 291–310.
UNEP (United Nations Environment Program) (2005). After the Tsunami: Rapid Environmental
Assessment Report, UNEP, Nairobi, 22 February; www.unep.org/tsunami/reports.
Valiela, I., J.L. Bowen and J.K. York (2001). ‘Mangrove forests: One of the world’s threatened
major tropical environments’, BioScience, 51(10), 807–15.
Wattana, S. (1998). ‘Fishing Communities and Coastal Resource Management in Southern Thai-
land’. Mimeo. Department of Social Sciences, Prince of Songkla University, Pattani, Thailand.
Wetlands International (2005). Natural Mitigation of Natural Disasters. Assessment Report to
Ramsar STRP12, Wetlands International, Jakarta, 2 February.
Winkelmann, R. (2003). Econometric Analysis of Count Data, 4th edn, Springer-Verlag, Berlin.
World Conservation Monitoring Centre (WCMC) (1992). ‘Wetlands’, in WCMC (ed.), Global
Biodiversity: Status of the Earth’s Living Resources, IUCN, Gland, Switzerland.
World Resources Institute (WRI) (1996). World Resources 1996–7, World Resources Institute,
Washington DC.
— (2001). World Resources 2000–2001. People and Ecosystems: The Fraying Web of Life, World
Resources Institute, Washington DC.
Wu, J., K. Skelton-Groth, W.G. Boggess and R.M. Adams (2003). ‘Pacific salmon restoration: Trade-offs
between economic efficiency and political acceptance’, Contemporary Economic Policy, 21(1), 78–89.
Zinn, J. (2005). Hurricanes Katrina and Rita and the Coastal Louisiana Ecosystem Restoration. CRS
Report for Congress, The Congressional Research Service, The Library of Congress,
Washington DC, 26 September.