Economic Benefits of Nature-Based Solutions for Climate Risk

Search Strategy

We performed a systematic review of peer-reviewed literature, following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (Page et al. 2021) approach. The selection of studies included in the analysis was based on Scopus. In addition, the ESVD (Brander et al. 2024) was reviewed because it compiles diverse articles assessing ecosystem services from NBSs. The final set of studies was updated on June 10, 2024, and the software Rayyan³ was used for the screening process and for managing and organizing the database of selected papers. After uploading the Scopus search results, Rayyan compiled the abstracts and allowed searching for keywords and placing tags based on the inclusion and exclusion criteria. We submitted the review protocol to Open Science Framework Registries (Staccione et al. 2023). The methods for evaluating NBS co-benefits, based on the same review process, are detailed in the accompanying analysis (Staccione et al. n.d.). All included studies and their sample sizes are available in Appendix A.

The studies targeted by the search had to present the design, implementation, or use of different nature-based approaches aimed primarily to reduce the risks and impacts of climate-related hazards and disasters at different scales and in different ecosystems. Furthermore, included studies needed to clearly address the co-benefits generated by the NBS investigated, other than the primary goal of the solution. We categorized the ecosystem services in our database following the economics of ecosystem and biodiversity (TEEB 2011) framework. The framework groups such services into four broad categories: provisioning services (e.g., food, water, raw materials), regulating services (e.g., local climate regulation, air quality regulation, erosion control), habitat or supporting services (e.g., maintenance of diversity), and cultural services (e.g., recreation, aesthetic appreciation, tourism). This classification provided a standard structure to build the database, and we used the individual services as explanatory variables in the different model specifications.

We acknowledge the inherent subjectivity in categorizing co-benefits. We based our coding on how these benefits were explicitly presented in each SP study. These were usually quite clear in choice experiments because they were categorized as different attributes. Whereas local climate regulation, air quality, or carbon storage were usually straightforward to categorize, recreation and tourism were sometimes more interchangeable. Recreation was typically framed in terms of improved access or availability of recreational opportunities at the site, and tourism refers to an increase in visitor numbers. Finally, studies should present models, methods, and metrics to assess the effectiveness of co-benefits, prioritizing quantitative evaluation. Therefore, the keywords selected for the reviews refer to five main concepts: NBSs, risk reduction, disasters, co-benefits, and methods (see the list of keywords in Appendix B). We used the keywords listed in these five domains in a single search query, and they were connected by the AND operator.

The review of journal articles was limited to those published in English between January 2005 and June 2024. The time span was defined according to the first definition of ecosystem services as benefits that people can obtain from nature (Millennium Ecosystem Assessment 2005) and the successive development of related concepts, including NBS and co-benefits. Although other SP studies have examined ecosystem services before 2005, the lack of formalized jargon would have created an additional challenge in identifying relevant studies. We initially included all papers that explored NBSs designed for disaster risk reduction. However, we retained only those providing a monetary assessment of co-benefits, excluding those assessing only risk-reduction performance or addressing co-benefits without monetization. We also kept only studies providing a comparable monetary value that can be converted to US dollars per hectare per year.

Coding of Key Variables

The dependent variable, similar to previous related meta-analyses, is the 2022 US dollar value prices of NBS co-benefits per hectare per year (Brander and Koetse 2011; Bockarjova, Botzen, and Koetse 2020; Filho et al. 2021). This is possible because most studies provide some indication of the price value, size of the site, and time component. This approach allows for comparisons of our results with previous meta-analyses. This process depends on the information available in the original studies, and it leads to the methodological choices and assumptions explained here. Missing data from the selected studies were complemented with information from official sources, such as the World Bank Open Data (e.g., exchange rates, GDP per capita, population density). First, to derive the dependent variable, all WTP values were converted to 2022 US dollars through official exchange rates and GDP deflators retrieved from the World Bank.⁴

Regarding the spatial component, if a study reported the value of the site as a whole, we divided the total value by the number of hectares. We retrieved this figure from either the study or external sources of information to convert the value to US dollars per hectare. When studies provided WTP per visit instead, the preferred option was to convert it to an annual basis by multiplying it by the annual number of visitors (Brander and Koetse 2011). We derived the aggregate WTP value by multiplying the “per year per household value” by the number of households in the area. When the value was reported per person instead of per household, we used the average household occupation from official sources.

We retrieved population density from the most accurate and granular data available (town/city, region/state, country)—the most granular data from government sources and, whenever needed, national averages from World Development Indicators. We estimated the number of households using the population totals in the area and the number of households for the region. Whenever the number of households was unavailable, we used the average number of people per household in the town (or country when data were unavailable) to estimate the number of households in the area. Table 2 presents an example of how the challenges of standardizing the data were dealt with through a systematic protocol.

In the regressions, we performed a log transformation of the dependent variable to deal with the WTP skewness. The main independent variables of interest in our meta-regression were the different types of NBSs and the co-benefits or ecosystem services. The co-benefits were coded as binary variables (taking the value of one if the site provides that ecosystem service and zero otherwise). Notably, one site could provide several ecosystem services. Similarly, NBS types were also coded as dummy variables taking a value of one when the site matched a certain type, and zero otherwise. We included the size of the nature site, geographical location, and other socioeconomic variables, such as population density and respondents’ income, in the meta-regression. Although some papers included information about respondents’ income, we retrieved this information from the World Bank data for consistency. The GDP per capita at the lowest level was used as a proxy for respondents’ income. We converted these variables to 2022 US dollar values using the purchasing power parity exchange rates to US dollars to account for differences across countries. Geographical location was determined using continent dummy variables, and the size of the nature site was measured in hectares. Population density is defined as the number of people per square kilometer.

Last, with the objective of analyzing the impact of the previous land uses on the monetary valuation, we classified all studies according to the land use change of the NBS intervention in the following categories: conservation or maintenance of an existing nature site, enhancement or restoration of an existing site, and policies that proposed transforming a site from its current land use to an NBS. The last option was divided depending on the original use: urban (street/concrete or building), undeveloped nature site, and agricultural land. An overview on the coding of the variables used in this analysis can be found in Appendix D.

Econometric Specification

The dependent variable is the log of the monetary value (US$/hectare/year) attributed to NBS co-benefits. The log transformation addresses the skewness of our dependent variable and residuals. The set of covariates (X) varies per specification. The simplest version (model 1) includes only GDP per capita, NBS size, and population density. This model allows for testing whether our results align with those of previous works, such as Koetse and Brander (2011) and Bockarjova, Botzen, and Koetse (2020). In addition, we control for study type by creating a dummy variable that takes the values one for contingent valuation (CV) studies and zero for choice experiment (CE) studies because previous studies have found that the elicitation format can influence the valuation outcome (Bockarjova, Botzen, and Koetse 2020). CV studies are also more likely to be susceptible to the payment vehicle (e.g., bidding games, dichotomous choices) and can yield a lower valuation compared with CE because they refer to bundled services instead of payment for individual improvements on NBS attributes.

The following models add complexity by including NBS types, geographical binary variables, land use changes, and co-benefit specific variables: model 2 includes the continent binary variables to control for geographical location; models 3 and 4 add the NBS type and the co-benefits, respectively; and model 5 combines the variables introduced in each previous iteration and adds the land use covariates.

Ordinary least squares (OLS) has limitations when addressing heteroskedasticity in meta-analysis data (Stanley 2013). To correct for these issues, we used weighted least squares (WLS) models. Unrestricted WLS has been demonstrated to be superior to OLS when there is publication or small-sample bias (Stanley and Doucouliagos 2014). WLS corrects for different observations having different error variances, which is the issue that violates the OLS assumption. In WLS, studies with smaller standard errors are given higher weights. Because it is a common challenge that has been encountered in previous meta-analyses, not all studies report standard errors. In case we opted to drop these observations, we would lose 71% of the observations of our meta-analysis since only 64 observations included standard errors. As an alternative WLS approach, the number of observations of the selected studies has been commonly in previous meta-analyses used as a proxy weight to account for the precision with which the effect sizes are estimated (Dalhuisen et al. 2003; Travisi and Nijkamp 2004; Nelson and Kennedy 2009; Nobel et al. 2020). This is the preferred approach for our analysis to prevent arriving at unreliable models based on a low number of observations. Each term in the WLS criterion includes an additional weight determining how much each observation in the dataset influences final parameter estimates:

1

2

The study with the largest sample was used as the baseline. The remaining WTP values from the other studies, our dependent variable, were weighted according to the largest sample (Wmax). The WTP values were multiplied by their own sample size (Wi) and then divided by the largest sample (equation [1]). Equation [1] indicates the new dependent variable included in our model. The interpretation remains the same as before. Equation [2] was the model used in our WLS meta-regressions. The OLS models were also tested for the robustness of our results and whether the chosen method significantly influenced them. To account for within-study correlation of error terms, we clustered the errors at study ID level for studies that contributed more than one valuation to the database.

In Section 4, the coefficients of the meta-regression are applied to create a value transfer function, which is then used in two case studies to illustrate its utility. Notably, the outcome in terms of economic value per year refers to the NBS co-benefits, which means that the risk reduction benefits also must be considered when producing a cost-benefit analysis. For the value transfer function, we can use both the OLS and the WLS models. We focused on deriving a value transfer function employing the results of WLS because of its superior properties. We used model 5 for its superior performance in explaining the variance. Owing to collinearity concerns, model 5 does not include the log of GDP per capita and the continent dummies together, as flagged by the variance inflation factor. Equation [3] shows the value transfer function applied to the chosen case studies:

3

The coefficients from the meta-regression (model 5) are substituted for the unknown coefficients in the value transfer function (Table 3). Simultaneously, the variables were substituted by the policy site characteristics. We took the exponent of the value of the right side since our dependent variable was originally measured in natural logs. To obtain the total value per year, we needed to multiply the result of that equation, after inserting the site characteristics, by the size of the policy site. The results are in Section 3.

Descriptive Statistics

After the screening phase, the final database consisted of 216 observations across 61 studies (see Appendix A). This higher number of observations compared with those from similar studies is due mainly to new evidence published since 2022. The main reason for excluding studies was the lack of a comparable monetary value. Most studies provide more than one valuation because choice experiments usually elicit preferences for multiple co-benefits of a particular NBS or ecosystem. Sample sizes also vary considerably across the studies, with 120 as the lowest number of respondents and 6,172 as the highest. Table 4 shows the summary statistics. In total, the meta-analysis is based on 49,596 respondents, combining all studies together.

The final set of papers indicates that most of the SP studies were conducted in either Europe or North America, accounting for 40% and 25% of the total number of observations, respectively. This distribution aligns with that recorded not only in previous meta-analyses regarding co-benefits of NBSs (Bockarjova, Botzen, and Koetse 2020) but also in systematic reviews of the risk reduction aspect of NBSs (Sudmeier-Rieux et al. 2021; Debele et al. 2023). Regarding the publication year, we conclude that, as expected, the amount of evidence on the NBS co-benefits has been increasing exponentially in recent years: 77% of the included studies were published after 2017 and 52% since 2020. Regarding the NBS type, around 50% of our observations related to urban NBSs, mainly green roofs and urban parks. This result is also in line with those of Debele et al. (2023), who found that 48% of the studies they identified focused on urban NBSs. Wetlands and forests are also quite prominent in the literature, featuring in 15% and 16% of works, respectively. Similarly, flood risk and heat mitigation are the two most common objectives of NBSs for our selected studies; recreation, aesthetic appreciation, and biodiversity support were the more frequent co-benefits assessed.

As expected, the monetary valuations differ considerably per study. The wide variations in population affected and hectares of the NBSs implemented lead to considerable variations in valuation per hectare per year. The median valuation per hectare per year was US$5,543 at 2022 prices. This valuation is similar to those of previous studies, reporting US$12,000 per hectare per year for urban NBSs such as parks, whereas the valuation of other types of urban nature ranged between US$1,250 and US$2,800 (Bockarjova, Botzen, and Koetse 2020). The average GDP per capita in the study was US$40,348 at 2022 prices, whereas the median was US$42,337. According to World Bank data, the average GDP per capita in 2022 at the global scale was US$12,730, affirming that the studies valuing NBS co-benefits are mostly located in wealthier countries. More evidence is certainly needed in the most vulnerable countries because they will be impacted by climate change–related risks in the upcoming years. The median area of the NBSs in our database was 1,700 ha, and the median population density was 605 people per square kilometer. As expected, this figure is considerably higher than the world average of 59.24 people per square kilometer: unsurprising, given that this figure accounts for uninhabited areas and that almost 50% of our observations are from urban settings.

Table 5 distinguishes between two types of urban NBSs: The first encompasses those to be developed on existing buildings, such as green roofs or green walls. Ultimately, these urban NBSs may not have a wider impact on co-benefits such as recreation and would not require changing the economic purpose of land in an urban setting for a different use (Urban−building). The second category consists of urban NBSs that would be implemented at the street level or would mean to change the use of open urban spaces for NBS creation (Urban−street). As Table 5 shows, regarding the land use change variables, urban NBSs located at the street level (e.g., parks) are highly valued, with a median WTP per hectare of US$208,916. On the other hand, increasing the conservation efforts or improving the maintenance of existing urban nature to increase protection against climate risks has a lower median value of US$1,192 per hectare.

Approximately 9% of the studies sampled did not specify the land use change the intervention would entail. As expected, NBSs situated in urban settings had a higher median WTP (US$36,747 per hectare) than nonurban NBSs (US$785 per hectare). This outcome is in line with previous findings since urban areas are usually densely populated and have a higher GDP per capita, two variables associated with a higher WTP in previous studies.

Meta-Regression Results

The following meta-regressions consist of a series of WLS models where the dependent variable is the log of the monetary value (US$/ha/year) attributed to NBS co-benefits. The log transformation addresses the skewness of our dependent variable and residuals. In other words, the lack of symmetry around the mean caused by the outliers shown in the descriptive statistics. The five models presented below include several key variables by groups, starting from the basic specification in model 1 to the full specification in model 5 (Table 3). The final models show a higher adjusted R-squared compared with the basic specifications (from 0.47 in model 1 to 0.768 in model 5). Therefore, there is a higher percentage of the dependent variable variation explained by models with the co-benefits, continent dummies, land uses, and the types of NBSs. Compared with previous meta-analyses, our regressions show a relatively high adjusted R-squared since models in Filho et al. (2021) and Brander and Koetse (2011) produced R-squareds of 0.48 and 0.13, respectively.

Model 1 represents the basic specification, where we can observe that our findings align with those in the literature (Brander and Koetse 2011; Bockarjova, Botzen, and Koetse 2020). First, the NBS size coefficient shows diminishing returns. The larger the NBS size, the lower the economic value per hectare. Since this is a log-log coefficient, it represents an elasticity. In other words, a 1% increase in size is associated on average with a 0.53% decrease in the value per hectare, keeping everything else constant. This finding is also common across all specifications with a similar effect size, including the OLS models used as robustness checks (Appendix C). As in Bockarjova, Botzen, and Koetse (2020), CV studies estimate a lower average WTP compared with choice experiments.

Model 3 indicates that urban NBSs are strongly correlated with a higher value per hectare. Regarding the co-benefits, model 4 indicates that heat reduction, maintenance of genetic diversity, and aesthetic appreciation have positive and statistically significant coefficients. Therefore, NBSs providing these benefits will be, on average, more valued by respondents. Hence, respondents are willing to accept a higher tax (or other payment) to develop these NBSs or restore/maintain NBSs that provide these benefits.

On the other hand, air carbon storage has a negative coefficient while being statistically significant. This indicates a lower average per hectare value associated with this co-benefit while keeping everything else constant. A plausible interpretation is that respondents are more likely to be willing to pay for co-benefits that they can enjoy daily, such as reduced heat or recreation areas, whereas air quality or carbon sequestration cannot be perceived as valuable, despite the long-term negative health consequences. Tantiwat, Gan, and Wang (2021) found a positive relationship between awareness of current air quality problems and potential health consequences and higher valuation for air quality improvements. This could explain our findings since it is possible that the lack of awareness could also have influenced our results.

In addition, the Netherlands Bureau for Economic Policy Analysis found that air quality has a small impact on property prices compared with noise pollution or availability of accessible green and water amenities within close distance to one’s house (CPB Netherlands Bureau for Economic Analysis 2022). This is consistent with our observed coefficients. Carbon sequestration can also be perceived as a global service that does not only affect the local area and the residents who are making the monetary contribution and, hence, should be considered a lower priority. Its contribution on a global scale may be seen as less significant by respondents. The null effect of flood risk reduction is another interesting finding, which is in line with some previous meta-analyses (Bockarjova, Botzen, and Koetse 2020). This could be related to potentially uncaptured disamenities (Dundas et al. 2017).

Similarly, an increase in tourism is also associated with lower WTP in model 4, which can be linked to several reasons, including noise, congestion, pollution, or crowding, potentially leading to decreased recreational quality (Gössling 2002; Sanz-Blas, Buzova, and Garrigos-Simon 2024). Ojijo and Steiger (2024) found that residents place high importance on tourism not altering land use change and on limiting the number of tourism accommodations. In all our models, local climate regulation is strongly correlated with a higher WTP on average, indicating a strong preference for NBSs that address this issue.

Last, we tested the correlation of previous land use with NBS valuation. It is important to note that the relevance of land use changes depends partially on how this is explained in the original studies because it may have not been explicitly presented in the choices. Therefore, our findings regarding previous land use should be interpreted as correlational and not necessarily a direct preference of respondents for land use change itself. Previous land use is relevant for policy-making, as significant differences in valuation may indicate which NBSs could be prioritized to maximize value for local communities. For example, we would expect that NBSs planned in urban settings with virtually no green areas would be well received by residents.

The main distinction between conservation and new developments is also insightful, as people may be used to the nature around them, lowering the marginal benefits of additional or improved nature. We excluded NBSs proposing conservation or maintenance measures as our baseline for these dummies. The adjusted R-squared increases considerably compared with model 1, and it increases more when included with co-benefits in model 5. In line with our previous findings, respondents highly valued repurposing urban land to create new green or blue spaces (urban lakes, ponds), particularly those at the street level rather than on top of buildings. In addition, model 5 shows that developing NBSs on agricultural land is a land use change that respondents respond to favorably and correlates with higher WTP. In Appendix C, our OLS models, without correcting the WTP variable by paper, point in the same direction as these models.