A Global Meta-Analysis of the Value of Ecosystem Services Provided by Lakes

Abstract

This study presents the first meta-analysis on the economic value of ecosystem services delivered by lakes. A worldwide data set of 699 observations drawn from 133 studies combines information reported in primary studies with geospatial data. The meta-analysis explores antagonisms and synergies between ecosystem services. This is the first meta-analysis to incorporate simultaneously external geospatial data and ecosystem service interactions. We first show that it is possible to reliably predict the value of ecosystem services provided by lakes based on their physical and geographic characteristics. Second, we demonstrate that interactions between ecosystem services appear to be significant for explaining lake ecosystem service values. Third, we provide an estimation of the average value of ecosystem services provided by lakes: between 106 and 140 USD$2010 per respondent per year for non-hedonic price studies and between 169 and 403 USD$2010 per property per year for hedonic price studies.

Highlights

• •We conduct the first meta-analysis on the value of ecosystem services provided by lakes.
• •We analyze results of 133 studies, offering a total of 699 values.
• •We show that antagonisms and synergies across ecosystem services are perceived and valued by people.

1. Introduction

Out of all the surface freshwater on Earth, about 90% is contained in natural and artificial lakes (Shiklomanov and Rodda, 2003). Being one of the most important sources of water for human and for economic use, lakes provide many services. Some of them are directly valued by humans (water supply, flood damage reduction) whereas others have positive impacts mainly on the environment (e.g. improved wildlife habitat). Since most of these services are not traded on markets, assessing their economic value is not straightforward. As a result, a wide non-market valuation literature has developed with some recent empirical applications to lakes, see for example Artell (2014). Due to the wide range of valuation methods, characteristics of lakes and value estimates, it is still difficult to assess whether general results emerge from this literature. This is the main issue we address here.

We propose to conduct a meta-analysis of the economic value of ecosystem services provided by lakes. We wish to identify if there exists a valuation function that relates the ecosystem service value of a lake to its physical, economic and geographic characteristics. Our analysis relies on the most extensive global database of non-market and market valuations of ecosystem services provided by artificial and natural lakes (699 values extracted from 133 studies).

We argue that the results of this meta-analysis might be useful for decision-making. First, there remain substantial debates on the economic value of ecosystem services provided by lakes (Magat et al., 2000, Viscusi et al., 2008, Sander and Polasky, 2009). A good understanding of the physical, economic and geographic characteristics of lakes upon their economic value may inform decisions related to their use, conservation or restoration. Second, it is not clear that the relationships obtained with the existing meta-analyses for other water bodies (e.g. rivers, wetlands, transitional and coastal waters) may be used for lakes. Since some services provided by lakes are quite specific, we may indeed expect specific economic values for this type of water body (see Supporting Information S1).

To our best knowledge, our meta-analysis is the first one focusing on a large set of ecosystem services specifically provided by lakes. From a methodological point of view, our meta-analysis combines information reported in primary studies with geospatial data from geographic information system (GIS) data layers and other external sources. In addition, we explore how antagonisms and synergies between ecosystem services are valued by respondents. This is the first meta-analysis to incorporate simultaneously these two characteristics (external GIS data and ecosystem service interactions).

The remainder of this article is organized as follows. Section 2 is devoted to the presentation of the meta-database. In Section 3, we conduct the econometric analysis of economic values provided by lake ecosystem services. Section 4 presents our main results.

2. Meta-Database on Values of Ecosystem Services Provided by Lakes

2.1. Building of the Meta-Database

The scientific references have been selected through systematic searches of the keywords “Valuation and Lake”, “Value and Lake”, “Willingness to pay or WTP and Lake”, “Stated preferences and Lake”, on search engines and on the websites of major publishers of academic journals (Scopus, Science Direct, Wiley, Web of knowledge, RepEc, AgEconSearch). Similar searches were also conducted on databases specialized in environmental valuation. Lastly, the “grey literature” was searched using various search engines (including Google Scholar and Science.gov). This is important to reduce the influence of a potential publication bias in the meta-regression analysis. In total, the literature search process took about one year (December 2013–December 2014) and it led us to examine over 300 studies.

Based on their abstract, studies have been first classified as irrelevant, potentially relevant and relevant. Irrelevant studies (studies without any reference to one or several lakes or those which did not report any economic valuation results) where disregarded at this first step. Further investigations were then conducted on potential relevant studies in order to reclassify them either as irrelevant or relevant. Lastly, all studies considered as relevant were downloaded and an additional screening process was conducted to decide if they had to be included or not in the final database.

This selection procedure led us to retain 133 studies, see Supporting Information S2 for the full reference list. A vast majority of the database is made of peer-reviewed articles (110 studies), the second category the most represented being institutional reports (11 studies). Studies are quite recent on average. Among the 133 studies of the database, 58 have been published after 2010, 52 between 2000 and 2010 and the remaining before 2000.

All continents are represented in our database, with an over-representation of North-America. North-America ranks first with 71 studies (68 studies deal with United States). The second continent the most represented is Europe with 30 studies. As already mentioned, United States are by far the country for which we have the most of studies. This may result from a selection bias since our systematic searches for lake valuation studies have been done in English. It may also reflect the fact that hedonic price approaches have been extensively used in this country for valuing housing amenities.

A single study may report multiple lake values, either because several lakes are considered or because of the use of several valuation methods or scenarios. Due to multiple values per study, we have 699 observations (i.e. lake ecosystem service values) in our final sample located all over the world, see Fig. 1. This represents on average a little bit more than 5.2 observations per study. Again, United States rank first with 376 observations. They are followed by Norway (66 observations), Turkey (26 observations), New-Zealand (25 observations), Finland (23 observations) and Japan (19 observations).

Fig. 1.

Location of lakes and number of observations per country in the meta-database.

2.2. Ecosystem Services Provided by Lakes

For each primary valuation study (or for each observation in case of multiple observations per study) we have identified the ecosystem services provided by the considered lake. In total, we have gathered economic values for 12 different ecosystem services provided by lakes, see Fig. 2. These services belong to three categories of ecosystem services (provisioning services, regulation and maintenance services, cultural services) presented in Supporting Information S3.

Fig. 2.

Ecosystem services provided by lakes in the meta-database.

Not surprisingly, the vast majority of ecosystem services for which a lake value is associated with belongs to the cultural service category. We have categorized the cultural services in our database slightly differently compared to the list presented in Supporting Information S3. In particular, the “recreation service” has been split into several sub-services: fishing, boating, swimming, camping, sightseeing and unspecified recreational services. In addition, the “amenity” sub-service has been created for studies based on the hedonic price (HP) approach. As explained in Lansford and Jones (1995), an HP study of shoreline and “near-the-lake” properties captures an important component of the recreational and aesthetic values that are provided by the existence of such a lake. There is however no direct correspondence between these amenities and the cultural service category defined in Supporting Information S3. In our database, among the cultural service category, the “amenity” service ranks first (244 observations) followed by the different recreational services such as “fishing” (265 observations) and “boating” (183 observations).

For provisioning services, we have only 25 observations for the “water for drinking” and the “water for non-drinking purposes” services. Lastly, we have 206 observations of economic values for regulation and maintenance services. The majority (193 observations) refers to the “maintaining populations and habitats” service, whereas the remaining observations deal with the “flood protection” service and the “erosion prevention” service.

Some studies value only one particular lake ecosystem service, but a significant number of them provides values for two or more services, Fig. 3. The number of ecosystem services valued in each study varies from 1 to 7, with an average a little bit higher than 2. This raises an interesting identification issue since, in most cases, a direct correspondence between a particular service and its associated economic value does not exist. This identification issue might be particularly relevant to address in case of complementarity or substitutability relationships among services. Indeed, in all previous meta-analyses on water ecosystem services, it has been assumed that the economic value of a water body is a linear function of the ecosystem services provided. We argue that such a specification could be questioned in case of tradeoffs or synergies between ecosystem services. Since there are complex relationships among ecosystem services (Fu et al., 2011, Raudsepp-Hearne et al., 2010) , the value for a specific ecosystem service might depend upon the other ecosystem services provided. Not introducing interactions across ecosystem services may then lead to biased estimates in the meta-analysis. It also raises some concerns vis-a-vis the common use of a “value catalog approach” for transfer of ecosystem service values.

Fig. 3.

Number of ecosystem services valued in each primary study.

2.3. Reconciling Lake Values

Lake values have been reported in the literature in many different metrics (i.e. willingness to pay per unit of area or per person, marginal value, capitalized value), using different currencies and for different periods of time. In order to enable a comparison across studies, all these values must be standardized. As explained by Ghermandi et al. (2010) or by Londoño and Johnston (2012), the standardization of different and heterogenous metrics used to value ecosystem services is a difficult and controversial task. But such adjustments are required to reconcile variable definitions across studies (commodity consistency requirement), and are nearly universal in meta-analyses of ecosystem service values (Johnston and Rosenberger, 2010, Nelson and Kennedy, 2009). We explain here how ecosystem service values from the original studies have been normalized.

2.3.1. Accounting for Heterogeneity in Space and in Time

In our meta-database, lake values have been obtained for different countries (30 countries) and for different periods of time (from 1972 to 2012). This requires some normalization procedures. Following Ghermandi et al. (2010), differences in purchasing power among countries have been accounted for by using the Purchasing Power Parity (PPP) index provided by the PennWorld Table. As a result, all currencies have been converted in USD PPP. Second, the problem of having different years of observation is usually solved by using appropriate price deflators, see Ghermandi and Nunes (2013) for a recent example. Values reported for price levels other than 2010 have been converted using national customer price indexes (CPI) provided by the International Monetary Fund (World Economic Outlook 2014). As a result, all ecosystem service values have been expressed in 2010 US$.

2.3.2. Normalized Value for Ecosystem Services Provided by Lakes

One specific issue with HP studies is that they give a capitalized value whereas other valuation methods typically provide a value estimate per unit of time. Additional data management is then required for making HP estimates more comparable. In our case, the capitalized values obtained from HP studies have been annualized assuming constant value per year, using the country-specific 30-year fixed mortgage rate as a discount factor (for the year of the study) and considering a 30-year time horizon. A similar procedure has been used by Woodward and Wui (2001) or by Ghermandi and Nunes (2013). In our meta-analysis, all values obtained from HP studies have then been normalized and expressed in monetary unit per sold property and per year.

Lake values reported in studies which are not based on an HP approach are expressed in different metrics including monetary unit per unit of lake area per unit of time, or monetary value per household/person/trip per unit of time. Rationalizing the use of a particular normalized value is not straightforward in this case. Some previous studies have used a normalized value expressed in monetary unit per unit of area per unit of time (Woodward and Wui, 2001, Ghermandi et al., 2010, Brander et al., 2012, Ghermandi and Nunes, 2013). When an aggregated value for the investigated ecosystem is provided in the primary study, such a normalization procedure is easy to implement but when no aggregated value is provided, the study must be disregarded. This is why some meta-analyses of ecosystem service values have excluded studies in which values are estimated per unit of area (Londoño and Johnston, 2012). In that case, values are usually expressed in monetary unit per visit per unit of time (Brander et al., 2007, Johnston and Rosenberger, 2010) or in monetary unit per household/respondent per unit of time (Brouwer et al., 1999, Johnston et al., 2005, Johnston et al., 2006, Londoño and Johnston, 2012, Ge et al., 2013). We opted for this later normalization procedure: all values from primary studies using non-hedonic price (NHP) methods are expressed in monetary unit per household/respondent per year. In some cases, the results reported in the primary study needed to be adapted to fit the required format. For example, values per person per visit had to be transformed into values per person per year using data on number of visits and/or duration of visits.

2.3.3. A Preliminary Statistical Analysis of Lake Values

Considering the full sample, we find a mean value for lake ecosystem services equal to 315.1 USD$2010 per year (per respondent for NHP studies and per property for HP studies). The median value is 77.6 USD$2010 per year, showing that the distribution of values is skewed with a long tail of high values. On average, the values we find for ecosystem services provided by lakes are higher than the ones reported by Brouwer et al. (1999) for wetlands, 134 USD$2010 per respondent and per year.

We discuss now the breakdown of lake values according to a number of possible explanatory factors. Mean lake values have been calculated (1) by countries, (2) by lake size classes and (3) by ecosystem services. Results are presented separately for studies using an HP approach and for studies using another type of valuation method.

Fig. 4 presents mean annual lake values per country (in USD$2010 per property per year for studies using an HP approach and in USD$2010 per respondent per year for studies using a NHP valuation method). When considering HP studies, United States rank first with a mean annual value of lakes per property equal to 442.5 USD$2010. The following countries in terms of mean value are New-Zealand, Ireland, Finland, and Canada with 392.2, 310.9, 265.9 and 166.1 USD$2010, respectively. When considering studies using another type of valuation method, Switzerland ranks first with a mean annual value of lakes per respondent equal to 765 USD$2010. The following countries are France, United States and Chile with respectively 644.7, 491.4 and 465 USD$2010 for the mean annual value of lakes per respondent. For countries where lake values are available both with an HP approach and with another valuation approach (i.e. Canada, China, England, Finland, Netherlands, New-Zealand and United States), we observe some significant differences across lake values depending upon the method of valuation. This indicates that the valuation method used in the primary study might have an impact on the estimated value.

Fig. 4.

Mean annual value of lakes per country and per valuation method.

Another lake characteristic that we may expect to determine its value is its size (area). There is no particular expectation of the sign of this relationship. On the one hand, there may be diminishing marginal returns for most lake services as lake size increases, but on the other hand some ecological functions may require some minimum thresholds of habitat area which suggests that lake values may increase with size (Brander et al., 2006). For HP studies, no monotonic relationship seems to emerge between the lake value per respondent and its size. This is consistent with previous findings reporting constant returns to scale with respect to size for some ecosystem service values. Indeed, both Brander et al., 2006, Woodward and Wui, 2001 conclude that the economic value of wetland services is not significantly influenced by the size (area) of wetlands, i.e. that wetland values exhibit constant returns to scale. When considering studies not relying on HP, the picture is quite different. We find in that case a positive relationship between the size of the lake and its value. This may indicate the presence of increasing return to scale but the result may also be related to the fact that the biggest lakes in our meta-database are located in the United States (Lake Michigan, Lake Erie and Lake Ontario), a country for which we would expect a priori high values for lake ecosystem services.

In Fig. 5, we have split the annual value of a lake according to the presence or not of a specific ecosystem service (and still according to the valuation method used in the primary study). Values reported in this figure should not be interpreted as the value for the considered service since each lake in our meta-database provides often more than one service (on average 2.2 different services per observation). This figure calls for a few comments. In our meta-database, cultural ecosystem services provided by lakes are highly valued. This is particularly true for the “spiritual and symbolic appreciation” and for the “amenity” services. Interestingly, the two regulation and maintenance services in our meta-database (i.e. “flood protection” and “maintaining populations and habitats”) provide relatively high economic values. This is consistent with previous findings on wetland ecosystem services. Indeed, in their meta-analysis of values for wetland ecosystem services, Brouwer et al. (1999) report that the wetland function which generates the highest value is flood control, followed by wildlife habitat provision and landscape structural diversity. More recently, Brander et al. (2006) also report high values for biodiversity, amenity and flood protection services of wetlands.

Fig. 5.

Mean annual value of lakes per ecosystem services and per valuation method.

3. Meta-Analysis Specification and Results

The above analysis of the available data in the lake valuation literature does not allow for interactions between the various potential explanatory variables. In order to obtain marginal effects, given the interference of potentially relevant intervening characteristics, we use a meta-regression analysis.

3.1. Empirical Specification

One critical issue when conducting a meta-analysis is the high level of heterogeneity and the potentially non-comparability of studies pooled in the meta-data. As a good practice, studies included in the meta-analysis should satisfy a criterion of minimal consistency for the dependent variable across observations (Smith and Pattanayak, 2002). This commodity consistency criterion requires in particular a minimal level of uniformity for the definition of the good that is valued. By conducting a meta-analysis on lake ecosystem service values, we assume that these ecosystem services satisfy the commodity consistency criterion. This is a common assumption in meta-analysis of ecosystem service values (e.g. forest, wetland, coral reefs) where each ecosystem service is usually identified in the meta-regression using a binary variable (see Supporting Information S1). Our assumption is also supported by methodological works which have demonstrated that it is possible to pool in a single meta-analysis dissimilar commodities such as a wide range of recreational activities (Moeltner et al., 2007).

In our meta-analysis, the commodity consistency criterion is addressed in several ways.

First, we estimate two different meta-regression models: one for hedonic price (HP) studies and another for non-hedonic price (NHP) studies. The rational is that ecosystem services which are capitalized in housing values may be quite different from ecosystem services considered in studies using NHP methods.¹

Second, we control for observed characteristics of lakes included in the primary studies. Lakes can be either natural (477 observations) or artificial, and this distinction may matter since ecosystem services differ depending on these two categories of lakes. The size of lakes varies very significantly from 0.001 km² (Lago Paione Inferiore, Italy) to 58,000 km² (Lake Michigan, United States) with an average equal to 4520 km². All these physical characteristics of lakes have been extracted from the primary studies in order to be introduced as moderators in the meta-analysis.

Third, since some services may depend on water quality (e.g. drinking water) whereas others may rely more on the quantity of water available in the lake (e.g. flood control), we include as moderator the type of scenario used in the primary study to infer lake values. In particular, we make the distinction between a scenario of change in lake water quantity and a scenario of change in lake water quality.

The dependent variable in our regression equation is the natural logarithm of lake ecosystem service values in USD$2010, labelled lny. For HP studies, this value must be understood per property and per year whereas for NHP studies this value is per respondent and per year (Table 1).

Table 1.

Definition and description of variables used in the meta-analysis.

Variable	Definition	Mean	Min	Max
Dependent variable
lnyij	NHP studies: value of ecosystem services provided by a lake (lnUSD$2010 per respondent per year)	3.92	−3.67	9.18
	HP studies: value of amenity service provided by a lake (lnUSD$2010 per property and per year)	4.30	−8.06	9.25
Ecosystem services
DrinkWater	Drinking water service is provided (=1)	0.35	0	1
Fishing	Fishing service is provided (=1)	0.38	0	1
Swimming	Swimming service is provided (=1)	0.25	0	1
Boating	Boating service is provided (=1)	0.26	0	1
Camping	Camping service is provided (=1)	0.05	0	1
Sightseeing	Sightseeing service is provided (=1)	0.22	0	1
UnspecRec	Another recreational service (unspecified) is provided (=1)	0.31	0	1
PopHabitat	Maintaining populations and habitats service is provided (=1)	0.28	0	1
Spiritual	Spiritual service is provided (=1)	0.03	0	1
Amenity	Amenity service is provided (=1)	0.35	0	1
Characteristics of the study
MethodCE	Choice experiment method is used (=1)	0.14	0	1
MethodTC	Travel cost method is used (=1)	0.20	0	1
MethodCV	Contingent valuation method is used (=1)	0.28	0	1
MethodHP	Hedonic price method is used (=1)	0.34	0	1
MethodOT	Another valuation method is used (=1)	0.04	0	1
Peer reviewed	Study has been published in a refereed journal (=1)	0.67	0	1
Scenario improve	The scenario is an improvement compared to the current situation (=1)	0.70	0	1
Scenario location	The scenario is a change of location (=1)	0.10	0	1
Scenario quality	The scenario is a change of water quality (=1)	0.47	0	1
Scenario quantity	The scenario is a change of water quantity (=1)	0.08	0	1
Scenario fish	The scenario is a change of fish quantity (=1)	0.04	0	1
Scenario ecological	The scenario is a change of ecological conditions (=1)	0.08	0	1
Scenario view	The scenario is a change of lake view (=1)	0.03	0	1
Scenario other	Another type of scenario (=1)	0.08	0	1
Substitute included	Lake substitute is included in the study (=1)	0.16	0	1
Characteristics of the study site
Natural	Lake natural (=1)	0.70	0	1
<1 km2	Lake area smaller than 1 km2 (=1)	0.08	0	1
[1, 20] km2	Lake area between 1 and 20 km2 (=1)	0.34	0	1
[20, 1000] km2	Lake area between 20 and 1000 km2 (=1)	0.27	0	1
>1000 km2	Lake area larger than 1000 km2 (=1)	0.30	0	1
Unesco heritage	Lake is/belongs to an Unesco World Heritage site (=1)	0.03	0	1
Special area	Lake belongs to a special area (protected park, Ramsar site, etc.) (=1)	0.22	0	1
External and geospatial variables
lnGDP capita	Logarithm of GDP per capita (per country and year, World-Bank)	10.40	5.96	11.51
Water stress	Total annual water withdrawals expressed as a percentage of the total annual available blue water (per river basin, World Resources Institute)	0.50	0.00	6.30
Water variability	Variation in water supply between years (per river basin, World Resources Institute)	0.43	0.14	1.51
Drought index	Average length of drought times the dryness of the droughts from 1901 to 2008 (per river basin, World Resources Institute)	25.37	0.00	46.03
ln Lake abundance	Logarithm of the number of distinct lakes within a 20 km radius from the primary site (own GIS computation based on Global Lakes and Wetlands Database)	−5.14	−7.94	0
RegEurope	Lake located in Europe (=1)	0.25	0	1
RegNorthAmerica	Lake located in North-America (=1)	0.56	0	1
RegPacificAsia	Lake located in Pacific or Asian region (=1)	0.13	0	1
RegOther	Lake located in another region (=1)	0.06	0	1

For NHP studies, the meta-analytical regression model is specified as follows:

$$\ln y_{ij}=\gamma E{S}_{ij}+{\beta }^b{X}_{ij}^b+{\beta }^s{X}_{ij}^s+{\beta }^c{X}_{ij}^c+{\zeta }_{ij}$$ (1)

where ES includes the ecosystem services provided by the lake (with potential interactions across ecosystem services), X^b is a vector describing the water body characteristics (i.e. type of water body, size of water body), X^s is a vector describing the study characteristics (i.e., survey method, payment vehicle, elicitation format) and X^c includes context-specific explanatory variables. In Eq. (1), subscript i takes values from 1 to the number of studies and subscript j takes values from 1 to the number of observations, ζ_j is the usual error term and the vectors β^b, β^s, β^c and γ contain coefficients to be estimated for the explanatory variables in X^b, X^s, X^c and ES, respectively.

For HP studies, we have created an “amenity” service which captures the recreational and aesthetic values that are provided by the presence of a lake. Since this service is present in all HP studies it cannot be included in the HP meta-regression. For HP studies we use the following simplified specification of the meta-regression model:

$$\ln y_{ij}={\beta }^b{X}_{ij}^b+{\beta }^s{X}_{ij}^s+{\beta }^c{X}_{ij}^c+{\zeta }_{ij}$$ (2)

where notations are similar to the ones used in Eq. (1). The predicted lake value from Eq. (2) should be considered as the value for the “amenity” service.

3.2. Specification of Independent Variables

For NHP studies, the first group of moderators, ES, consists of a set of dummy variables representing all ecosystem services provided by the lake under consideration.² We also include dummy variables when two ecosystem services are jointly provided by a lake. Since it is not possible to estimate an interaction effect for all pairs of ecosystem services, we limit our analysis to the ones the most often observed in our data set.

The second group of moderator variables, X^s, controls for specific characteristics of primary studies (peer reviewed studies, valuation method used, existence of a lake substitute). Studies also differ depending upon the type of scenario used for inferring lake values. To ensure some minimum level of commodity consistency, we distinguish between scenarios of change in location (respondents are proposed some changes in location with respect to the lake), in lake water quantity, in lake water quality, in fish availability, and in lake view.

The third set of moderators, X^b, includes some characteristics of the water body (natural lake, lake size, special area or Unesco World Heritage site).

The last group of moderator variables X^c, includes spatial context variables supposed to have an influence on the willingness to pay, but which were not directly available from the primary studies. Johnston et al. (2016) provide a recent example of this type of “augmented” meta-analysis for the case of water quality in the United States. Our choice of additional spatial variables has been primary guided by economic theory (which predicts which factors are supposed to influence willingness to pay) and by data availability at the global scale. Since we expect a relationship between household wealth and willingness to pay, we has first introduced in the meta-regression the GDP per capita. Second, since willingness to pay has been shown to be related to water resource pressure and to water scarcity we have introduced three geospatial variables measuring water stress, variability in water and risk of drought. These variables have been extracted from GIS layers developed within the Aqueduct project of the World Resources Institute, and they are defined at the river basin level (Masutomi et al., 2009). Third, we expect the economic value of an ecosystem service provided by a lake to be related to the geospatial availability of lake substitutes (Schaafsma et al., 2012) . Using the Global Lakes and Wetlands Database, we computed a measure of lake abundance defined as the number of distinct lakes within a 20 km radius from the primary site.³ Finally, we have introduced some spatial controls for lakes respectively in Europe, North-America, Pacific or Asia, and in the rest of the world. The rational for including these spatial controls is that for cultural reasons ecosystem services might not be valued in the same way everywhere in the world.

3.3. Econometric Model

The non-independence of estimates from primary studies has been recognized as a crucial methodological issue in the meta-analysis literature. This is due to the fact that single studies often produce multiple values, and the observations drawn from the same study may therefore suffer from within-study correlation (Nelson and Kennedy, 2009). In addition, the primary studies may not be independent of each other, which implies between-study autocorrelation.

Some possible remedies for within-study correlation include the use of OLS with standard errors adjusted for clustering of observations within studies (Chaikumbung et al., 2016) or the selection of only one observation per study (Ghermandi et al., 2010), We explicitly take into account the hierarchical structure of the data by estimating panel data models (fixed-effect and random-effect).⁴

In Eqs. (1) and (2), the error term ζ_ij is split in two components: μ_j which is an error term at the second (study) level and ϵ_ij which is an error term at the first (observation) level. We assume that μ_j and ϵ_ij follow a normal distribution with means equal to zero and that they are uncorrelated, so that it is sufficient to estimate their variances, ${\sigma }_{\mu }^2$ and ${\sigma }_{ϵ}^2$, respectively. Some Hausman specification tests are then used to determine the most appropriate panel-data model.

3.4. Results of the Meta-Regressions

For NHP studies, the results obtained for different specifications of the basic meta-regression model described in Eq. (1) are presented in Table 2.

Table 2.

Estimates of the meta-regression models with random-effects: non hedonic price studies.

	ML1		ML2		ML3		ML4		ML5
	Coeff.	Std. err.	Coeff.	Std. err.	Coeff.	Std. err.	Coeff.	Std. err.	Coeff.	Std. err.
Ecosystem services
DrinkWater	0.78	0.66	0.64	0.65	0.51	0.58	0.55	0.59	0.79	0.58
Fishing	0.87***	0.26	0.90***	0.27	0.21	0.24	−0.04	0.24	−0.03	0.25
Swimming	0.72***	0.28	1.03***	0.32	0.61**	0.27	0.29	0.27	0.32	0.27
Boating	0.54	0.35	0.40	0.36	0.03	0.30	−0.23	0.31	−0.24	0.31
Camping	−0.01	0.76	0.47	0.76	−0.88	0.67	−0.81	0.65	−0.95	0.63
Sightseeing	1.22**	0.50	1.11**	0.57	0.29	0.47	0.48	0.45	0.68	0.45
UnspecRec	0.96***	0.30	1.87***	0.39	0.80**	0.34	0.71**	0.34	0.67**	0.33
PopHabitat	0.93***	0.30	1.69***	0.38	0.20	0.40	−0.10	0.41	0.02	0.41
Spiritual	1.29*	0.67	1.83***	0.67	0.77	0.55	0.39	0.60	0.43	0.61
Ecosystem service interactions
PopHabitat  × Fishing			0.85	1.05	1.05	0.82	1.08	0.78	0.93	0.77
PopHabitat  × Swimming			−1.56**	0.65	−1.07**	0.54	−0.72	0.53	−0.66	0.52
PopHabitat  × Boating			0.98	1.55	2.01*	1.16	1.89*	1.06	1.84*	1.02
PopHabitat  × Sightseeing			−0.09	1.24	0.26	0.98	0.04	0.91	−0.47	0.88
PopHabitat  × UnspecRec			−2.45***	0.69	−0.91	0.57	−0.94*	0.56	−0.79	0.56
Characteristics of the study
MethodCE					0.79***	0.24	0.72***	0.23	0.78***	0.23
MethodOT					−0.98*	0.52	−1.04**	0.50	−1.15**	0.49
MethodTC					1.77***	0.26	1.53***	0.25	1.43***	0.25
Peer reviewed					2.84***	0.40	1.65***	0.46	1.48***	0.57
Scenario improve					−0.12	0.26	−0.20	0.26	−0.18	0.25
Scenario quality					0.02	0.41	−0.13	0.41	−0.29	0.41
Scenario quantity					−0.56	0.42	−0.59	0.41	−0.62	0.41
Scenario fish					−0.46	0.47	−0.82*	0.46	−0.83*	0.46
Scenario ecological					0.15	0.55	−0.06	0.55	−0.15	0.54
Scenario other					1.21**	0.48	0.86*	0.48	0.71	0.49
Substitute included					−0.36	0.43	−0.40	0.40	−0.09	0.42
Characteristics of the study site
Natural							−0.32	0.44	−0.28	0.46
[1, 20] km2							1.29***	0.36	1.22***	0.39
[20, 1000] km2							2.03***	0.39	1.84***	0.42
>1000 km2							3.08***	0.52	2.85***	0.54
Unesco heritage							−0.25	1.02	0.12	0.98
Special area							−0.28	0.43	−0.09	0.45
External and geospatial variables
ln GDP capita									−0.17	0.13
Water stress									0.20	0.15
Water variability									−0.28	0.68
Drought index									−0.02	0.02
lnLake abundance									−0.14**	0.07
RegEurope									1.55	0.94
RegNorthAmerica									2.44***	0.89
RegPacificAsia									0.73	0.83
N	456		456		456		446		446
Log likelihood	−847.66		−838.03		−747.43		−709.66		−700.50
Prob. Wald test	0.00		0.00		0.00		0.00		0.00

***, **, and * aresignificant at the 1, 5, and 10% levels, respectively.

Panel data random-effects models estimated by using generalized least squares with STATA software.

The meta-regression model we have considered is semi-logarithmic with exception of some of the context variables, which are included in logarithmic form. The Hausman specification tests we conducted reveal that the random-effects model is always preferred.

The first model provides some indications on how people value lake ecosystem services. Six ecosystem services (among the nine specified) appear to be significant namely fishing, swimming, sightseeing, unpecified recreational, maintenance of populations and habitats and spiritual or symbolic appreciation. A particularly high value is found for the spiritual or symbolic appreciation service (+3.6 USD$2010 per year) whereas a low value is documented for the swimming service (+2.1 USD$2010 per year).⁵

In general, ecosystem services are not independent from one another. In Model ML2, we have then introduced some interactions across ecosystem services. This allows us to investigate the existence of antagonisms (the provision of one service is enhanced at the cost of reducing the provision of another service) or on contrary synergies (the provision of one service is enhanced when it is produced with one or more other services) across ecosystem services. An interesting result from ML2 is that some interactions between ecosystem services appear to be significant. In ML2, the interaction between the “maintenance of populations and habitats” service and the “swimming” service appears to be significant at 5%, with a negative sign. A similar finding is found for the interaction between “maintenance of populations and habitats” and “unspecified recreational” services at 1%. The value for a specific lake ecosystem service is shown to depend upon other ecosystem services provided by this lake. The negative sign for the coefficient of the interaction term suggests some form of substitutability among some ecosystem services. We then demonstrate that people perceive and value antagonisms across ecosystem services (in our case across regulation and cultural services).

Our results are in line with previous findings which have documented the existence of tradeoffs across ecosystem services. Working at the landscape level, Raudsepp-Hearne et al. (2010) report a pattern of antagonisms between provisioning services and both regulating and cultural services. Maes et al. (2012) have conducted a spatial assessment of the relationships between biodiversity, ecosystem services, and conservation status of protected habitats at the European scale. Similarly, their results suggest that provisioning services create tradeoffs with regulating services and with cultural services.

Some characteristics of the studies introduced in model ML3 appear to be consistently significant. In particular we find a significant impact of the valuation method used to obtain lake values. Compared to the reference method (contingent valuation), the value obtained with a travel cost study or with a discrete choice experiment approach results in a higher lake value. The impact of the valuation method has already been documented in some previous meta-analyses of ecosystem service values, but with contrasting results. For example, for the value of wetlands, Woodward and Wui (2001) have found positive and significant coefficients for replacement costs and hedonic prices, while Brander et al. (2006) have reported a positive coefficient for contingent valuation studies. Peer reviewed studies tend also to produce higher lake values which suggests a publication bias and supports the inclusion of grey literature in the meta-analysis. Our result contradicts Chaikumbung et al. (2016) who report lower values for estimates of ecosystem services published in journal articles, but is in line with results provided by some other studies (Ghermandi and Nunes, 2013). Lastly, studies differ depending upon the type of scenario used for inferring the lake value but the impact of these scenarios on estimated lake values remains very limited.

In model ML4 and ML5, we have added some characteristics describing the type of water body (natural versus artificial, size of the water body) and we have augmented the database with external and geospatial data extracted from GIS layers. We do not find any specific premium for natural lakes compared to artificial ones. Although some artificial lakes may have been constructed without fully replacing natural lake ecological functions or without fully supporting recreational activities or aesthetic services, it appears that they are valued by people similarly to natural lakes. We demonstrate the existence of a size effect, larger lakes being more valued than smaller lakes. One explanation could be that some ecological functions may require minimum thresholds of habitat resulting in increasing lake values with size (Brander et al., 2006). We do not find any specific premium for a lake belonging to a special area (regional or national park, Ramsar site) or for lakes considered as a Unesco World Heritage. Lake abundance (our measure for the existence of substitute lakes) is significant at 5% with a negative sign which is in line with previous results (Johnston et al., 2016). Lake ecosystem service value tends to decrease with the existence of substitute lakes. Lastly, we find significant regional effects. The economic value of a lake is significantly higher in North-America at 1%.

For HP studies, the estimates of different specifications for the meta-regression model described in Eq. (2) are presented in Table 3. The Hausman specification tests reveal that the random-effects model should be preferred as well.

Table 3.

Estimates of the meta-regression models with random-effects: hedonic price studies.

	ML1		ML2		ML3
	Coeff.	Std. err.	Coeff.	Std. err.	Coeff.	Std. err.
Characteristics of the study
Peer reviewed	2.55***	0.72	1.32*	0.76	−0.78	0.76
Scenario improve	0.97***	0.33	0.92***	0.35	0.56	0.35
Scenario location	0.31	0.61	−0.02	0.62	−1.43**	0.70
Scenario quality	1.51**	0.73	1.18	0.75	−0.94	0.85
Scenario quantity	−4.02***	1.13	−4.46***	1.13	−5.45***	1.13
Scenario ecological	1.43	1.34	1.36	1.33	−0.16	1.26
Scenario view	3.87***	0.55	3.77***	0.54	3.18***	0.55
Scenario other	2.43**	1.06	2.35**	1.05	2.00*	1.03
Substitute included	1.30	1.32	1.76	1.30	1.80*	1.08
Spatial model	0.12	0.35	0.12	0.34	0.16	0.33
Characteristics of the study site
Natural			0.03	0.39	−0.45	0.40
[1, 20] km2			0.35	0.42	−0.20	0.44
[20, 1000] km2			1.04*	0.53	0.44	0.55
>1000 km2			2.55***	0.77	1.56**	0.75
Unesco heritage			−1.59	1.92	−1.63	2.08
Special area			0.89	0.61	0.83	0.58
External and geospatial variables
ln GDP capita					0.42	0.78
Water stress					0.36	0.35
Water variability					−3.41***	1.32
Drought index					0.05	0.05
lnLake abundance					0.19	0.12
RegEurope					1.90	8.78
RegNorthAmerica					1.40	8.71
RegPacificAsia					1.58	7.71
N	233		224		224
Log likelihood	−432.82		−410.24		−396.62
Prob. Wald test	0.00		0.00		0.00

***, **, and * aresignificant at the 1, 5, and 10% levels, respectively.

Panel data random-effects models estimated by using generalized least squares with STATA software.

Model ML1 only includes characteristics of the study as moderators whereas we introduce characteristics of the lake in model ML2 and ML3. Model ML3 also incorporates external and geospatial data. Globally the results obtained using HP studies are consistent with those obtained with NHP studies. Similarly to the case of NHP studies we find a significant positive effect of being peer reviewed (ML1 and ML2), suggesting the existence of a publication bias. The use of a scenario corresponding to an improvement compared to the current situation also results in an increased value of the lake (ML1 and ML2). Considering the different scenarios, estimates suggest that a lake view is highly capitalized in property prices, a result in line with the existing literature (Bourassa et al., 2004). We document a positive lake size effect: being located near a large lake is associated with a high property price. When introduced in ML3, lake level variability is highly significant with a negative sign: a high variation between years of lake water levels is associated with a low property value.

The different models presented in Tables 2 and 3 allow also to provide some estimates of the average lake value, see Table 4. When considering NHP studies and focusing on models ML3, ML4 and ML5, the average predicted values of a lake in our sample are 106, 120 and 140 USD$2010 per respondent per year. We get higher values when considering HP studies, respectively 169, 193 and 403 USD$2010 per property per year. This suggests that the ecosystem services capitalized in property prices are not equivalent to those valued using NHP approaches. Whatever the type of valuation method considered, our results demonstrate that lakes provide substantial monetary benefits to people.

Table 4.

Predicted lake values from meta-regressions.

	Mean	Sd. dev.	Min	Max
Predicted lake value for NHP studies (in USD$2010 per person per year)
Model ML1	41.77	67.56	1.00	407.98
Model ML2	50.77	74.71	1.00	346.55
Model ML3	105.56	156.96	0.69	868.66
Model ML4	120.31	133.28	1.55	766.12
Model ML5	140.50	212.83	0.89	1080.50
Predicted lake value for HP studies (in USD$2010 per property per year)
Model ML1	169.06	433.72	0.05	2193.36
Model ML2	193.13	591.05	0.04	5379.41
Model ML3	403.33	1331.02	0.21	9494.27

This table provides statistics on estimated values for lakes using meta-regression models presented in Tables 2 and 3.

4. Conclusion

While there is a widespread recognition that lakes provide valuable ecological services, there remain substantial debates on their economic value. This lack of consensus may come from the fact that lake valuation studies are extremely diverse in terms of type of value estimated (i.e use and non-use values), type of lake considered and valuation method used. As a result, lake ecosystems are often undervalued in decisions related to their use, conservation or restoration.

In this paper, estimates of values attached to different ecosystem services provided by lakes have been compared and synthesised in a meta-analysis using the most extensive global database of non-market and market valuations of ecosystem services provided by lakes. The meta-analysis provides insights into the factors that have to be considered when attempting to transfer lake values on the basis of the valuation studies. From a methodological point of view, instead of relying solely on information retrieved from the primary studies, we have introduced spatial context variables expected to have an influence on lake values. More specifically, we have augmented our meta-database with external spatial variables extracted from geographic information systems. One of the key results of our meta-regression analysis is the relevance of some of these spatial variables (i.e water variability or lake abundance) in explaining variation in lake values.

In terms of empirical results, we have provided an estimation of the average value of ecosystem services provided by lakes (between 106 and 140 USD$2010 per respondent per year for NHP studies and between 169 and 403 USD$2010 per property per year for HP price studies) and we have offered insights on how people value different lake ecosystem services. An interesting result is that some interactions between ecosystem services appear to be significant for explaining lake values. This reflects antagonisms between some ecosystem services as already documented for landscape by Fu et al. (2011), or Raudsepp-Hearne et al. (2010).

Our results show promise for benefit transfer since they suggest that it may be possible to reliably predict the value of ecosystem services provided by lakes based on their physical and geographic characteristics. This opens the door to benefit transfer at large scales (Brander et al., 2012, Chaikumbung et al., 2016) . The challenges of meta-analysis and value transfer when working at the global scale should not be understated. First, due to cultural or societal differences the data-generating process for lake values might not be the same everywhere. Using a single value function might then be questionable. Second, although our data cover a large variety of countries some regions (Africa or Latin America) are under-represented. This makes the global benefit transfer difficult. Finally, a difficulty of large scale value transfer is related to the need to aggregate individual benefits in order to derive relevant policy implications from the valuation exercise. As the population who benefits from an improvement of aquatic ecosystem services may be spread across a wide geographical area, one of the key parameters for aggregating benefits of improved water ecosystem quality is the spatial distribution of these benefits. One of the main difficulties is then to identify the benefiting population (beneficiaries). This issue is important since aggregate benefits depend on estimates of both individual benefits and of the number of beneficiaries (Bateman et al., 2006)

Appendix A. Supplementary data

Supporting Information S1, S2 and S3.