Valuing urban nature through life satisfaction: The consistency of GIS and survey indicators of nature

This study estimates the relationships between green and blue nature and the life satisfaction (LS) of residents in the six largest cities in the Netherlands to monetize the value of urban nature. The analysis uses both survey and geographic information system (GIS) data on the availability of nature to examine the influence of this methodological choice on the valuation outcomes. The main findings are that different indicators of the availability of nature consistently reveal positive relationships with reported LS, which implies substantial marginal willingness-to-pay (MWTP) values for urban nature. Valuation results based on the survey data indicate higher MWTPs compared to GIS data on nature availability, which may be explained by the more disaggregated data from surveys on the availability and use of nature.


Introduction
Urban nature is highly relevant to local residents due to various positive effects that enhance the quality of life (Bolund and Hunhammar, 1999).The presence of nature can offer psychological benefits by helping with recovery from stress and increasing self-reported positive emotions (Ulrich et al., 1991).Attention to nature is also a key mechanism to restore mental fatigue stemming from work situations that require long, effortful, and directed attention (Kaplan and Kaplan, 1989).Moreover, the presence of nature positively influences people's affective, cognitive, and psychological responses, which can occur from merely viewing nature through a window (Hartig et al., 2003;Van den Berg, Maas, Verheij, and Groenewegen, 2010).In addition to psychological effects, the presence of nature also has direct effects on health, such as higher self-reported health and greater longevity (De Vries, Verheij, Groenewegen, and Spreeuwenberg, 2003;Maas et al., 2006;Takano et al., 2002).Furthermore, urban nature provides other ecosystem services to humans, such as aesthetics and recreation, which can be valued differently by individuals (Buchel and Frantzeskaki, 2015).Understanding people's preferences for urban nature can inform city planning regarding nature creation (Derkzen et al., 2017).
The economic valuation of nature can provide important insights for policymakers who make decisions about the protection and development of nature.For instance, cost-benefit analysis (CBA) has been used to guide environmental decision-making (Dehnhardt, 2014).CBA allows for a transparent, rational, and simplified evaluation of environmental policy alternatives.This use of CBA is contingent on clearly quantifying the costs and benefits of policy alternatives in comparable monetary units (Turner, 2007).However, benefits of nature do not have a directly observed market value due to the absence of an obvious trade (Liu, Costanza, Farber, and Troy, 2010).For that reason, researchers have commonly used revealed preference methods to estimate direct-use values of nature that are indirectly captured by market prices, mainly housing price variations in hedonic price analysis (Liu et al., 2010).Moreover, stated preference methods have often been applied to elicit non-use values of nature in terms of an individual's willingness to pay (WTP) derived from contingent valuation questions and choice experiments in surveys (Koetse, Verhoef, and Brander, 2017;Bockarjova et al., 2020).Each of these methods has its strengths and limitations.For instance, hedonic pricing studies are considered to have a high degree of validity by being based on observed market prices but capture only direct-use values 1 of nature.Stated preference surveys also allow for measuring non-use values 2 but may be associated with strategic behavior or ignorance in answering survey questions, resulting in under-or overestimations of WTP (Garrod and Willis, 1999;Di Tella and MacCulloch, 2006).
The life satisfaction (LS) approach is a relatively new alternative method for environmental valuation and may be able to overcome the issue of cognitive effort associated with stated preference valuation surveys while still being broadly applicable to measure combined use and non-use values (Frey, Luechinger, and Stutzer, 2010;Di Tella & MacColloch, 2006;Welsch and Kühling, 2009).This approach estimates the relationship between environmental conditions and LS 3 as well as how LS relates to income.Based on these two relationships, an implicit WTP for environmental goods is estimated by deriving the amount of income needed to compensate for a change in environmental conditions while keeping LS constant (Bertram and Rehdanz, 2015;Ferreira and Moro, 2010).However, the LS valuation approach also presents methodological challenges, including how to measure available nature in a respondent's living environment.Most urban nature valuation studies have either only used geographic information system (GIS) data as an objective indicator of available nature or survey data as a perceived measure of nature without examining the influence of this methodological choice on the valuation results (Ambrey andFleming, 2014, 2011;Bertram and Rehdanz, 2015;Wang, Kang, and Yu, 2017).
This study uses the LS approach to value urban nature in the six largest cities in the Netherlands, for which this value, to our knowledge, has not yet been studied.By applying both GIS and survey data types regarding nature availability and different nature variables, this study examines how they affect LS and influence the related monetary valuation estimates.While the GIS and survey data in this research provide information about the amount of green and blue nature in the respondents' area, the survey also contains additional information on accessibility, visit frequency, and types of nature.Hence, these nature indicators are not directly comparable but can reveal different insights into how nature relates with LS.We show that the different indicators of nature consistently have a positive relationship with LS, with the strongest relationship for frequent use of nature.
The remainder of this article is structured as follows.Section 2 reviews related literature.Section 3 describes the data collection and statistical methods used in this study.Section 4 presents the results.Section 5 discusses the main findings in relation to the literature, and Section 6 concludes the text.Kahneman and Sugden (2005) have noted that the use of LS as an interpretation for utility received little attention until the 1990 s.However, Edgeworth's idea of the "hedonimeter" (1881/1967, p. 101), an imaginary psychophysical device able to continuously measure an individual's level of pleasure as a metric of one's happiness for a given period, has received renewed interest.Indeed, a strand of economics research is trying to reintroduce this idea as experienced utility (Bruni and Porta, 2005;Clark, Frijters, and Shields, 2008;Frank, 2001;Frey and Stutzer, 2010;Van Praag and Baarsma, 2005).Conventional ideas of utility do not always reflect true preferences due to the bounded rationality most people display when making choices.Therefore, the use of LS data as a proxy for experienced utility may prove valuable in eliciting information on consumer preferences and social welfare (Kahneman and Krueger, 2006;MacKerron, 2012).Clark et al. (2008) have further examined this relationship between happiness data and utility and discussed whether the implicit tradeoff corresponds to choice behavior.They have found that most studies indicate a positive relationship of LS with income, marriage, job status, health, and religion (Layard, 2005).Krueger and Schkade (2008) and Welsch and Küling (2009) have added that LS lends itself well to modeling individual utility due to its reliability.

Using LS for environmental valuation
More specifically, Kahneman and Sugden (2005) have focused on using experienced utility for environmental valuation, and they provide a strong case for its application because it avoids some of the anomalies associated with stated preference methods.Unfamiliarity with the nature valuation question plays no issue, unlike the case with stated preference methods, since respondents only have to report LS (Frey et al., 2010;Di Tella & MacColloch, 2006).This reduces the cognitive load.Moreover, respondents might implicitly value certain amenities in LS of which they are not explicitly aware, such as air quality and nature provision (Frey et al., 2010).Welsch and Kühling (2009) have further reasoned that experienced utility is more direct than revealed methods since it relies on surveys about people's well-being, whereas it is less direct than stated preference methods because it does not ask to value specific environmental conditions.
The LS approach circumvents a key limitation seen with revealed preferences: the need for competitive markets and the inability to  (De Groot, Wilson and Boumans, 2002).
3 LS is also commonly referred to as "subjective well-being" and "happiness", essentially meaning the "subjective enjoyment of life" (Veenhoven, 2012).For consistency, this article uses LS throughout.capture non-use values (Bertram and Rehdanz, 2015;Ferreira and Moro, 2010).That is because the LS approach distinctly reflects contributions to individual welfare without the presence of markets (Frey et al., 2010).

Applications of the LS approach to value urban nature
Urban nature consists of parks, trees, waterways, community gardens, and so forth.These are prime examples of where the LS approach can be a suitable method through which to value urban nature by examining how the differences in availability of these urban nature types vary with changes in individual LS.Nevertheless, to date, only a few studies have valued urban nature with an LS approach.A recent study has valued parks and green spaces through four questions to which respondents from a representative UK sample indicate their well-being on several dimensions (Field in Trust, 2018).While controlling for various demographic characteristics, average LS scores are compared between users and non-users of green parks.The findings show individuals should be compensated between GBP 8.40 and GBP 22.83 per reduced visit to maintain their level of well-being, giving a total value for parks in the United Kingdom of between GBP 34 billion (EUR 38 billion) and GBP 92 billion (EUR 102 billion). 4ertram and Rehdanz (2015) and Krekel, Kolbe, and Wüstemann (2016) have estimated the relationships between urban nature and LS, together with the implicit WTP, in Berlin.The former study uses GIS data and survey data on green space use, combined with demographic controls.The relationship between nature and well-being follows an inverted U-shape, indicating a high marginal return for places with little green space.Their derived WTP for an average availability of green space per month equals EUR 27.The latter study focusing on Berlin finds an estimate of EUR 23 per person to pay for a 1 ha increase of an average green space area.The study is broader in scope since it uses longitudinal survey data.Wang et al. (2017) have studied the city of Dalian, China.They use survey data on respondents' well-being but, contrary to the previously discussed studies, also on the experienced environmental conditions.The data on green spaces still shows a strong and positive relationship with LS.They estimate a WTP of CNY 24,580 (EUR 3158)5 per year for a one-unit increase in green space quality.It is important to note that this unit of increase in quality differs from the interpretations in the aforementioned studies, where the WTP corresponds to an increase in area.
A study by Ambrey and Fleming (2011) has also used surveys with LS questions and income to value the quality of scenic amenities elicited by questions in another survey.Because both surveys have information on the location of respondents, they integrate the data to arrive at an implicit WTP of AUD 14,251 (EUR 11,187) for a scenic quality improvement of green spaces in Southeast Queensland, Australia.This information on scenic amenity is visualized on maps using GIS software.Both nature valuations in Dalian and Queensland arrive at an amount approximately equal to 28% of annual household income (Wang et al., 2017).
A subsequent study by Ambrey and Fleming (2014) has employed satellite data as an objective indicator of availability of green space.Increasing the area of green space in the major Australian cities corresponds with an implicit WTP AUD 12,800 (EUR 8627.20), 6hich is derived from LS data using surveys.Although the interpretation of the nature variable differs from the two former studies, these similar results provide further corroboration of using survey data as a proxy for objective land use data in deriving WTP estimates from LS information.Tsurumi and Managi (2015) and Tsurumi et al. (2018) have applied the LS approach to estimate the value of green nature in Japan's largest urban areas.The former study combines survey data including income and LS data and also assesses perceived quality of nature by including questions about respondents' preferences for green spaces.These data are complemented with GIS data on land use.They determine a WTP of JPY 93,714 (EUR 707) and JPY 160,065 (EUR 1208)7 for 1% increases of green within 100-300 m and 300-500 m, respectively.
The latter study by Tsurumi et al. (2018) differs from previous work on the topic by using higher-resolution satellite data integrated in GIS.This uniquely allows for measuring "tree-level" greenery data in the metropolitan wards of Tokyo, which means different types of urban green spaces can be identified.In addition to demographic and income information, the conducted survey contains various indices of LS.The authors find varying estimates for different types of greenery at several distances, with the largest values for types that affect mental health the most.
Finally, Kopmann and Rehdanz (2013) have employed the LS approach for 31 European countries using European surveys on LS and land cover data.Due to the large scope of the research, geographic and climate controls have been added to account for omitted variable bias.The heterogeneous effects across all regions make straightforward interpretation difficult, but the main findings show that LS is positively related with increases in natural land cover.Importantly, marginal WTP is highest for land cover types that are relatively scarce and vice versa.This indicates a nonlinear specification of natural land cover with decreasing marginal WTP.
In summary, the review of the literature provides results for comparison, corroborates the use of the LS approach for nature valuation, and indicates the need to analyze the methodological differences between using GIS or survey data on urban nature.The reason is that most studies only used one of these nature indicator types, while it would be relevant to understand whether or not both GIS and survey approaches result in consistent significant relationships with LS.

Data collection and variable definitions
This study uses survey data collected between October 21 and 31, 2020.The survey was answered by 2009 residents from six major cities in the Netherlands: Amsterdam, Den Haag, Eindhoven, Groningen, Rotterdam, and Utrecht.It was administered online by an ISO-certified company using a panel of households.8A representative sample was drawn for each city stratified by age (all 18 +), gender, and education level.9Table 1 presents the main descriptive statistics of the sample (see Table 1 of appendix A for the descriptive statistics of all variables).Table 2 describes how the variables are coded and derived from the survey questions. 10  The dependent variable of self-reported LS was elicited on an 11-point Likert scale ranging from 0 "not satisfied at all" to 10 "completely satisfied," which asked respondents, "In general, how satisfied are you with your life?" as is common in the literature (Bertram and Rehdanz, 2015;Diener et al., 2013;Krekel et al., 2016).
A range of variables was used to operationalize the environmental conditions, each of which contained different types of information.The questionnaire provided information about the amount of urban nature in the vicinity (abundant, little, gray), the types of urban nature most present in the area, 11 urban nature visit frequency (daily, weekly, monthly, few times a year, never), and the time it takes to get to a patch of urban nature (within 10 min or over).
The survey measurements of nature availability were complemented with objective indicators derived from GIS.For this purpose, a database by Statistics Netherlands called CBS Bestand Bodemgebruik 12 was used, which contains information on land cover use in the Netherlands in 2015 with a resolution of 25 m * 25 m.Together with a database on postcode locations called CBS Postcode statistieken, 13 GIS data on nature availability can be constructed and related to the LS measures at the postcode level.The relevant GIS data consists of recreational terrain, 14 forests and open natural terrain, 15 and inland waterways. 16Following Bertram and Rehdanz (2015), the study creates an objective measure of urban nature via buffer zones with a radius of 1 km around the centroid of the respondents' postcode and calculating the area (ha) of urban green and blue within this region. 17Areas that do not fall within urban green or blue are coded as "other."Given that the 1 km radius covers much of the postcode area, it captures the urban nature most relevant to the respondent.Moreover, this radius is approximately consistent with the survey measure of urban nature that can be reached within about 10 min' walking distance.Hence, this GIS variable is more easily comparable to the survey indicator of nature availability within 10 min walking distance than the survey indicators of the amounts and types of nature availability.Considering the spatial resolution of 25 m * 25 m, respondents' perceptions of available nature may diverge from this GIS measurement due to urban nature not being captured in the CBS data (such as street trees, green walls, gardens, and so forth).Hence, the GIS-and survey-based measure of nature availability can provide distinct but complementary information.Whereas the GIS-based measures are objective indicators of nature availability, the survey measures indicate how respondents perceive and use nature.
Our main models use dummy variables of income levels to allow us to model the separate relationships between each income level and LS (except for the excluded baseline income level).Furthermore, because income is a self-reported categorical value with income bands, the study uses the midpoints of these categories to create an approximation to a continuous variable, as is common in the LS valuation literature (Bhat, 1994;Welsch and Kühling, 2009;Ferreira and Moro, 2010;Stutzer, 2003;Von Hippel et al., 2017;Alberini et al., 2018;Schmitt et al., 2018;Sanduijav et al., 2021).This income variable facilitates the WTP estimation (Section 3.2).The last category is an open interval of having an income over EUR 6000 a month, for which we must derive a proxy in creating an approximation of a continuous variable since the omission of these observations could lead to selection bias.Based on income distribution data of the Dutch population from Netherlands Statistics, we approximate the highest income category as EUR 7000. 18 The sociodemographic control variables are dummy variables of homeownership (with being a renter as the excluded baseline), living in a houseboat or a detached or semi-detached house or apartment (with row houses as the excluded baseline), male gender, age with various categories (with 60 + as the excluded baseline), having children between 0 and 12 or 13 and 17 years old, having a multiperson household (without kids below 18 years old, and living alone as the excluded baseline), employment status (unemployed, parttop 5% of the income distribution of our respondents.Netherlands Statistics reports that the 9th income decile corresponds to an average income of EUR 6417, which suggests that EUR 7000 is a reasonable approximation of the average income for our highest income category.As a robustness check, we also examined how sensitive our results are to using EUR 6000 and EUR 8000 for this value.This only very slightly changed the income coefficient, and also did not affect the coefficients of the environmental variables in a substantial manner, which means our results are robust to this alternative coding of the highest income category.

S.P. de Vries et al.
time work, and full-time employment as the excluded baseline), and education levels. 19Moreover, dummy variables were included of living in the city center and the city in which the respondent lives (with Utrecht as the excluded baseline).

Statistical methods
A general specification of an empirical LS function, as adapted from Frey et al. (2010), is as follows: Here, LS i,j stands for self-reported LS by individual i in city j.The constant is given by μ 0 .Xi,j is a vector of environmental conditions reported by respondent i in city j with parameter vector β.Income, Y i,j , is a vector of income dummies in the main regressions of this paper with parameter vector γ.To estimate the WTP, we employ the income variable that was approximated to be continuous (Frey et al., 2010).Z i,j is a vector of individual micro-level characteristics to control for with parameter vector δ.ρ j accounts for systemic differences in the constant term among the six cities.Lastly, ε i,j captures the error term.We created six clusters for the error term, one per city j, to allow for dependence of error terms within a city.Our base model using the survey data is: Eq. 2 can easily be extended with variables on distance of nature, greenery type, and so forth.θ i,j contains the type and location of housing, gender, age category, family situation, employment status, and educational attainment.
The model using GIS data resembles Eq. 2, except for the nature variables: LS can be modeled by means of an OLS regression or by means of an ordered discrete choice model such as a logit or probit regression (Bertram & Rehdanz, 2013;Ferrer-i-Carbonell and Frijters, 2004;Tsurumi et al., 2018;Welsch and Kühling, 2009).Due to the ordinal nature of the happiness data, a discrete choice model would be a natural choice.However, Ferrer-i-Carbonell and Frijters, Notes: Dutch equivalents for the education levels are a LBO/VBO/VMBO; b MAVO, first three years of HAVO, VWO, and VMBO; c HAVO and VWO from fourth year onwards; d MBO; e HBO or university bachelor degree; f university master degree.
2004 have found that assuming cardinality 20 makes little difference and that standard least-squares can be applied.Our study uses an ordered probit model for the main regression because of the ordered characteristic of the dependent variable (for specifications of ordered probit models, see Greene andHensher, 2010 andLitchfield et al., 2012).However, previous LS studies have found that the OLS results are generally similar to an ordered probit model (Bertram & Rehdanz, 2013;Tsurumi et al., 2018).We give the OLS results in Table 2 in appendix D, and they are indeed largely similar in significance and sign to the ordered probit model results reported in the main paper text, although there is a slight variation in the size effects. 21One of the features of an ordered probit model is latency of the dependent variable (LS).The latent variable represents the underlying, unobserved continuous variable that determines the ordered categorical outcome variable.The observed ordinal outcome variable is related to the latent variable through a set of threshold parameters.These thresholds divide the range of the latent variable into different categories or levels of the ordinal outcome variable.We  20 The cardinality assumption requires the distances between the well-being ratings are the same (for instance, an increase from 0 to 1 is valued equally to an increase from 9 to 10). 21The significance and signs of the OLS results are largely similar to the findings of the ordered probit models, but the OLS coefficients can be more easily interpreted since they represent marginal effects.
S.P. de Vries et al.
assume that the discrete dependent variable, which is on a 0-10 scale, represents a continuous latent variable.However, we note that the latent variable is assumed to follow a standard normal distribution.An implicit economic value of urban nature is estimated by taking the ratio of the coefficients of the nature and income variables from Eqs. 2 and 3. A comparison of Eq. 2's base specification and model 3 can provide insight into how the two types of environmental data result in different relationships with LS.A comparison with the dummy for the access to nature within 10 min and the 1 km radius buffer is most suited for a direct comparison of these models since the range and time correspond well and capture access to approximately the same urban nature.A difference in marginal willingness to pay (MWTP) results would indicate a different valuation of objectively available nature versus how respondents perceive and use nature.Lastly, we combine the survey and GIS data in the same model to observe the potential changes in the relationship, both in significance and effect size, between our main explanatory variables.
The variance inflation factor is used to assess potential issues with multicollinearity (Table 1, appendix C). 22 Because the education dummies indicate a relatively high inflated variance, we tested a model using only dummies for three levels of education.This reduces the model fit and still provides insignificant results, which is why the final model involves six education dummies.Results of the correlation table (not shown) motivate the use of only the city center dummy as a location indicator without separate dummies for the rest of the city or living in a village to further mitigate multicollinearity concerns.

Relationships between nature and LS
Table 3 shows the results of the model specifications using the abundance of urban nature in the vicinity as an environmental variable.Model 1 indicates a positive and significant coefficient both for income and having abundant nature in the vicinity versus the excluded baseline of living in a gray area.The observed relationship between the income dummies and LS is positive and generally reflects diminishing marginal utility with respect to income, 23 which is consistent with previous findings in the available literature (Layard, 2005).All income bands have a statistically significant relationship with LS, including the dummy that indicates the respondent did not report their income.The lowest income band (lower than EUR 1500 per month) operates as the excluded baseline.After addition of the sociodemographic control variables in specification 2, the effects remain significantly positive but reduce in size.
The coefficient signs of control variables on LS are largely consistent with the literature, which provides confidence in our results.Owning compared with renting a house is positively related to LS in a significant manner, as also found by Ambrey and Fleming (2014) and Zang, Zang, and Hudson (2018).However, the coefficient of the type of housing is insignificant.The coefficient of education is insignificant, which is not uncommon (MacKerron, 2012;Powdthavee, Lekfuangfu, and Wooden, 2015).Gender, too, is insignificant, consistent with previous studies (Zweig, 2015).People in the 60 + age category (the excluded baseline dummy) show the highest LS, with adults between 40 and 49 being associated with the lowest levels of LS.The approximated U-shaped relationship between age and LS 24 is in accordance with earlier research on this relationship (Blanchflower and Oswald, 2004).Living with children below 12 (and a spouse) has a positive coefficient.This is in contrast to other studies that have observed a negative relationship for this variable (Ambrey andFleming, 2014, 2011;Bertram and Rehdanz, 2015;Wang, Kang, and Yu, 2017).Our positive relationship between LS and children may be explained by the cofounding of having children with the positive effect of marriage on LS, for which we do not control separately.Living in a multi-person household (either with a spouse or in student housing, for example) corresponds to higher LS compared with living alone, as previous studies have also found (Evans and Kelley, 2004;Stutzer and Frey, 2006).Unemployment is significantly associated with lower LS compared with full employment at a 10% level, as has been commonly concluded (Frijters, Haisken-DeNew, and Shields, 2004), although this relationship is not universally reflected in the LS approach literature (Ambrey and Fleming, 2011;Bertram and Rehdanz, 2015;Krekel, Kolbe, and Wüstemann, 2016).Finally, there is a positive coefficient of living in Amsterdam, Rotterdam, and Groningen compared with living in Utrecht.
Table 4 shows models using the following other nature variables: nature visit frequency, time to get to nature, amount of green and water in hectares, and a combination of frequency and proximity indicators.All models indicate results for income similar to specification 2. Compared with never making use of nature in the vicinity, daily use positively relates to LS.The coefficients are positive and significant regardless of the frequency of visits but increase in size the more frequent the visits are.Model 4 specifies whether the nearest urban nature is within 10 min.This is significantly associated with a higher LS compared with urban nature located farther away than 10 min.Using the green area in a circle with a radius of 1 km around the centroid of each postcode in model 5 reveals that each hectare increase has a positive relationship with LS.Model 6 combines the frequency and abundance of nature.The significance of the area of water does not change considerably, although the size decreases.The frequency variables of nature visits remain statistically significant (p < 0.05 for weekly and daily visits), and the coefficient decreases slightly.The pseudo R-squared shows that model 22 Only the tables for models 2 and 3 in the main text show the variance inflation factor (VIF).Other models have a lower (better) mean VIF and indicate similar results. 23Although the coefficients increase in size with the income level, this is not the case with the highest income category, which still has a positive albeit lower coefficient (see also Table 2 in Appendix D for the OLS results).An explanation may be that this income group is very small and hence has a too low number of observations (3.9% of the total sample) to estimate this effect reliably. 24The reason is that the coefficient of the 50-59 age group is only slightly lower than the excluded baseline of 60 years and older, the coefficient is the lowest for the 40-49 age group, and for the subsequent lower age groups the coefficient are again less negative (see also  6 fits the data slightly better than the other models, although the fit is largely similar across models 3-6.The Akaike's and Bayesian information criteria are shown in Table 2 in appendix C. The values are similar across all specifications, with model 6, which combines the different GIS and survey nature indicators, displaying the lowest value in both cases; this means it provides the best fit according to these criteria.Models 5 and 6 have fewer observations due to missing postcode information from people living outside the city boundaries.

Nature type specific relationships with LS
Table 5 shows the results of the nature types in the surroundings that were reported by respondents in the survey, with living in a gray area (i.e., not having any of these nature types) as the excluded baseline dummy.The results of the main explanatory variables are consistent with model specifications 2-6.Model 7 shows that a garden in the vicinity is also significantly associated with higher LS.This applies to green roofs and parks as well.The relationship between water (lakes, ponds) and LS is most pronounced compared with not living near such natural bodies.Other types of nature show no statistically significant coefficient.Model 8 adds the frequency of use variables to the specification shown in model 7.The frequency variables remained statistically significant, which is consistent with models 5 and 6.Whereas being surrounded by the sea, a lake, a garden, and green roofs maintains the level of significance with a small drop in the coefficients, living close to a park becomes statistically insignificant.One possible explanation can be that parks are the most visited type of nature, which may be the reason adding the frequency variables reduces the significance of living close to one.

Economic valuation results
The MWTP estimates are provided in Table 6 and are expressed in euros per household per month.These are estimated from the ordered probit regression in Model 5 of Appendix D (Table 1), which employs the continuous approximation to the income variable using the midpoints of the income bins.The MWTP for living in an area with abundant urban nature compared with living in a gray area, while keeping LS constant, is EUR 3940 per month.Living in the vicinity of little urban nature compared to living in an environment without urban nature is associated with an MWTP of EUR 2761 per month.Daily use of nature compared with never using urban nature is associated with an MWTP of EUR 7149 per household per month while keeping LS constant.Finally, a 1 ha increase of urban green or water in their 1 km radius buffer zone is associated with an MWTP of EUR 11.19 and EUR 18.81 per month, respectively.
It is of interest to compare how this WTP estimate derived from the GIS indicator of nature availability relates to the WTP results obtained from the survey indicators of nature to obtain insights into the effect of this methodological choice on valuation outcomes.For this purpose, the MWTP estimate based on the survey indicator of nature within 10 min is most appropriate since this is approximately consistent with the walking distance of the 1 km buffer zone around the postcode's centroid.The average size of the natural area is 59.15 ha per postcode buffer for the latter variable.Multiplying this average with the average MWTP for green and blue nature of EUR 15 results in an MWTP of EUR 887.25 per month based on the GIS estimate.This is slightly less than one-fourth of the estimate derived when using the survey data (EUR 3940 per month).This difference could be explained by the survey indicator capturing a higher perceived supply of nature in the following two ways.First, the survey data represents more types and small plots of greenery and water, which may not be fully reflected by the spatial resolution of the GIS maps of 25 m by 25 m.Moreover, the survey data only captures areas of nature of which respondents are actually aware, whereas the GIS data may also partly include natural areas of which people are not aware, meaning this has a smaller association with LS.
In summary, although the GIS and survey indicators of nature are not directly comparable, we do consistently find sizable MWTP values for either approach to measure nature.Moreover, the higher MWTP values of the frequent use of nature compared with the availability of nature measured with both the survey and GIS indicators show that the usage of nature is positively related with LS.

Relationships between urban nature and LS
The results regarding the relationships between nature and LS meet the a priori expectations and are largely similar to the findings in the existing literature.The presence of urban nature in the vicinity, whether little or abundant, is statistically significant and positively associated with higher LS compared to living in a gray area.This is a finding that is robust in the LS literature (Ambrey andFleming, 2014, 2011;Bertram and Rehdanz, 2015;Kopmann and Rehdanz, 2013;Krekel, Kolbe, and Wüstemann, 2016;Tsurumi and Managi, 2015;Tsurumi, Imauji, and Managi, 2018;Wang, Kang, and Yu, 2017).Moreover, we obtained the expected result that a more frequent use of nature has a positive association with LS.This is in line with Fields in Trust ( 2018), which also revealed a significant positive relationship for users of urban parks compared with people who never utilize them.
Our findings of positive associations of LS and areas of green and blue spaces as measured with the GIS indicators are also comparable with various studies.For instance, our coefficient size of green area is consistent with an estimate by Ambrey and Fleming (2014) of the effect of green spaces on LS in Southeast Queensland, Australia. 25Moreover, Krekel, Kolbe, and Wüstemann (2016) used a similar buffer area to the one in this study by creating a GIS measure of green nature availability in Berlin and have found a similar coefficient size as ours. 26Presence of urban blue spaces is associated with a substantially higher LS in our study, which is also comparable with Krekel, Kolbe, and Wüstemann (2016).Dissimilarities in the use of ecosystem services in regard to blue nature may potentially explain this difference.
We use objective and subjective proximity measures in our models: the self-reported 10-minute commute time to nature dummy in model 4 is most comparable to the GIS measure in model 5.Both measures are comparable as, at an average speed, a person can walk approximately 1 km in 10 min, which is the radius of the buffer zone used in creating the nature availability variable in GIS.We note that the subjective measure of nature availability within 10 min may capture more urban nature than the objective GIS estimate because the latter's spatial resolution does not represent all small nature elements.Model 6 includes both the proximity to nature and frequency of use variables.All variables, except little abundance of nature in the vicinity, remain statistically significant and display a positive coefficient.
Finally, our analysis allowed for examining whether specific nature types relate to LS the most, which seems to be the case for water (lakes, ponds) and, to a lesser extent, green roofs and gardens.Given the scarcity of highly disaggregated nature data, few other studies have estimated the relationships between individual nature types and LS.An exception is MacKerron and Mourato (2013), who have found that individuals report a 0.6-point higher LS near marine and coastal areas.Tsurumi et al. (2018) have determined that the relationships with LS that affect mental well-being the most are observed for greenery types.Moreover, De Vries, Nieuwenhuizen, Farjon, Van Hinsberg, and Dirkx (2021) have shown that individuals reported higher LS when nature was within window view due to scenic beauty and "fascinatingness."This could also provide a possible explanation for the higher LS found in our study for the aforementioned nature types, which, in theory, are viewable from indoors.percentage point increase in greenspace area corresponds to 1.85 ha in their study.Dividing 0.0032 with this amount results in a coefficient of 0.00173, which is very similar to our estimate in model 6. 26 A direct comparison of the green urban nature coefficient found by Krekel, Kolbe and Wüstemann (2016) is not possible due to their incorporation of a non-linear effect.However, a green area of 27.7 ha in their study results in the same effect on LS, with lower area amounts giving a larger effect, and higher area amounts a lower effect compared to the results in this study.Given that a green area of 27.7 ha is only roughly less than 10 ha below the mean green area of this study, the results from the GIS estimate are comparable.
S.P. de Vries et al.

Monetary valuation using LSA
Since the MWTP estimates derived using the self-reported survey data are not directly comparable to studies using GIS estimates, we only compare our MWTP derived with the GIS measure to the existing literature here.Our estimated MWTP of EUR 15 per household per month using the GIS indicator of urban green proximity is lower than the adjusted result27 of Ambrey and Fleming (2014), who arrived at an MWTP of AUD 52.80 (EUR 35.9028 ) for a 1 ha increase of public greenspace per household per month.The authors had hypothesized that the magnitude of the relationship between nature and LS satisfaction would be smaller than in comparable studies due to the relative abundance of urban nature in Australian cities.Given that less than 10% of respondents classify their area as "gray" in this study, the same could be said for Dutch cities. Bertram and Rehdanz (2015) have arrived at MWTP of EUR 26.82 per month for a 1 ha increase of public greenspace.Krekel, Kolbe, and Wüstemann (2016) have found a similar result to Bertram and Rehdanz (2015).While estimates are somewhat higher than ours, they prove the same order of magnitude and suggest a broad comparability of the MWTP for urban nature when using GIS measures.
A body of literature has emerged that uses the LS approach to value nature and the environment.The advantages and limitations of this approach to date for environmental valuation have been discussed in detail in review articles (e.g., Welsch, 2009;Welsch and Kühling, 2009;Frey et al., 2010).It has been argued that a main advantage of the LS approach is that it entails a less cognitively demanding task for respondents compared with other survey-based methods, such as the stated-preference valuation method.
A limitation is that our study is limited to correlational associates between LS and nature.For instance, happier people may more regularly visit nature sites, or they could be happier due to these regular visits.Nevertheless, the consistent significant positive coefficients of GIS and survey-based nature proximity indicators reported in our study between various models point at a strong association between the presence of nature and subjective LS.Moreover, the causality assumption may be problematic in the presence of omitted variable bias.Especially in cross-sectional surveys, it may be more challenging to control for all individual and spatial heterogeneity in factors that influence the relationships of interest.Such unobserved factors may be better accounted for in future research using panel data analysis for which repeated surveys of the same sample are required, which is in practice more challenging to collect.However, although panel data analysis would provide a solution for the issue with unobserved factors discussed above, it would still not solve the simultaneity issue.
Moreover, future research can examine how the order of the LS and nature-related questions in the questionnaire influences the results.In our questionnaire, the LS question was asked after the questions about nature, which may imply that the answers to the nature questions potentially unconsciously influence answers to the LS question.Nevertheless, our approach may be motivated by literature showing that reported order effects in LS research, if significant, are generally small in size (Diener et al., 2013).This finding may depend on the context of the study, and future research could examine the importance of such order effects in valuing urban nature using the LS approach.A final suggestion for future research is that including more bands in the income question could result in a more precise measurement of the income variable (Ferreira and Moro, 2010).

Conclusion
Nature is increasingly viewed as a solution for multiple problems in urban areas, such as health issues and heat stress, but there is still an insufficient understanding of its positive association with human well-being.This study contributes to emerging literature that values the benefits of urban nature through its relationship with LS.We examined how these valuation outcomes relate to the use of various indicators of nature derived from GIS and survey questions.
The main findings are that different indicators of the availability of nature consistently reveal positive relationships with reported LS, including the survey indicators of the abundance of nature in the vicinity, daily use of nature, and availability of nature within 10 min' distance as well as GIS indicators of available green and blue nature.Derived MWTP values for urban nature based on these relationships imply that people place substantial value on urban nature.
Moreover, the higher MWTP values of the frequent use of nature compared with the availability of nature measured with both the survey and GIS indicators show that, especially, the usage of nature is positively related with LS.The difference between the availability of nature within 10 min' distance and the GIS indicators of nature availability with similar walking distance can be attributed to the more disaggregated data on the availability and use of nature from surveys.An implication for research in this field is that survey indicators on availability and use of nature may offer more detailed insights into the relationship between nature and LS for people who actively use and are aware of nature in their surroundings.GIS data on nature availability can provide more aggregate information on how nature relates with LS of both users and non-users, which gives more conservative MWTP values.An implication that can be derived from our results for city planners and other policymakers is that there are substantial benefits to additional urban nature.Our estimated MWTP values could therefore serve as input in cost-benefit analyses to assess the economic desirability of green projects.
Future LS research can examine the difference in valuation outcomes between survey and GIS indicators of nature in other countries and contexts where more detailed GIS information on nature may be available.Covering six major cities in the Netherlands could be considered a caveat to this study.Despite this research being representative by age, education, gender, and income, the inclusion of smaller towns could ensure a broader representation of the Dutch urban population.Finally, our results are based on cross-sectional relationships between LS and income as well as urban nature indicators, which may not reflect temporal dynamics.Future research can apply panel data analysis to examine how nature availability and its use may be associated with LS and valuation outcomes over time.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.Notes: These summary statistics are based on the sample of respondents who reported their income level.This is the sample used for estimating the models with the continuous income variable on which the MWTP estimates are derived.

Appendix B
Fig. 1.Buffer zone example.Created using QGIS.GIS data point example.Using the centroid of one of the post code boundaries in Amsterdam, a buffer zone with a radius of 1 km is created.Land use data is coded in urban green, urban blue, and other.For each buffer zone the area of the land use categories is calculated.This information can subsequently be merged using the associated post code information.

Table 2
Coding of key variables used in our regression models.
How often do you make use of a patch of nature in the neighbourhood like a green promenade, park, or forest to, for instance, stroll, run, or cycle?= 'not satisfied at all'-10 = 'completely satisfied' In which class does your net monthly income fall as a household? 1 = 'less than €1500 per month'-5 = 'Over €6000 per month' Table 2 in Appendix D).
S.P.de Vries et al.

Table 3
Relationship between urban nature and LS of nature in vicinity on LS (ordered probit).

Table 4
Relationships between different nature variables and LS (ordered probit).

Table 5
Relationships between different nature types and LS (ordered probit).

Table 2
English translation of relevant survey question.

Table 2
Akaike's and Bayesian Information Criteria.

Table 1
Relationships between urban nature and LS.(ordered probit, continuous approximation of income using midpoints, full models).

Table 2
Relationships between urban nature and LS.(OLS, income dummies, full models).

Table 2
(continued ) (continued on next page) S.P. de Vries et al.