Human well-being and natural capital indictors for Great Lakes waterfront revitalization

Abstract

Revitalization of natural capital amenities at the Great Lakes waterfront can result from sediment remediation, habitat restoration, climate resilience projects, brownfield reuse, economic redevelopment and other efforts. Practical indicators are needed to assess the socioeconomic and cultural benefits of these investments. We compiled U.S. census-tract scale data for five Great Lakes communities: Duluth/Superior, Green Bay, Milwaukee, Chicago, and Cleveland. We downloaded data from the US Census Bureau, Centers for Disease Control and Prevention, Environmental Protection Agency, National Oceanic and Atmospheric Administration, and non-governmental organizations. We compiled a final set of 19 objective human well-being (HWB) metrics and 26 metrics representing attributes of natural and seminatural amenities (natural capital). We rated the reliability of metrics according to their consistency of correlations with metric of the other type (HWB vs. natural capital) at the census-tract scale, how often they were correlated in the expected direction, strength of correlations, and other attributes. Among the highest rated HWB indicators were measures of mean health, mental health, home ownership, home value, life success, and educational attainment. Highest rated natural capital metrics included tree cover and impervious surface metrics, walkability, density of recreational amenities, and shoreline type. Two sociodemographic covariates, household income and population density, had a strong influence on the associations between HWB and natural capital and must be included in any assessment of change in HWB benefits in the waterfront setting. Our findings are a starting point for applying objective HWB and natural capital indicators in a waterfront revitalization context.

Introduction

Waterfronts in the Laurentian Great Lakes are spaces with current or historical ecological, economic, or cultural connections to the adjacent Great Lake, tributary river, connecting channel, estuary, or harbor. On the terrestrial side, the waterfront may include industrial, residential, or commercial areas, built infrastructure, brownfields, beach, and natural green spaces including riparia and wetlands (Angradi et al., 2019). Many of these aquatic and riparian spaces have a legacy of environmental degradation from their industrial past (Hartig and Wallace, 2015; Hartig et al., 2020). Cleaning up and revitalizing Great Lakes waterfronts has recently gained momentum through federal programs including the Great Lakes Areas of Concern program (AOC; Environment Canada, 2014; USEPA, 2021), and the Great Lakes Restoration Initiative (GLRI, 2021; Liesch and Graziano, 2021; Jurjonas et al., 2022).

Waterfront revitalization may also be called waterfront redevelopment, renewal, reimagining, or regeneration (Hoyle 2000; Sairinen and Kumplulainen, 2006). We prefer the term revitalization (latin: re-vita, “return to life”) to avoid the implication that economic or real estate development is necessarily the salient goal. We have proposed an inclusive and holistic definition of revitalization as policies or projects which promote equitable and sustainable improvements in human physical, mental, cultural, and socioeconomic well-being, while also protecting or improving local or regional natural capital (defined below) including habitat for native species (after Angradi et al., 2019).

Revitalization can describe efforts specifically designed to create multifunctional spaces that improve communities through protection, improvement, or restoration of the natural environment (e.g., access to the environment, green space, green infrastructure, native vegetation, biodiversity), or through creation of environmental amenities, especially health-promoting recreational amenities. Revitalization benefits may also be realized through efforts to improve public safety, conserve cultural heritage, memorialize local history, improve sanitation, enhance public use, increase aesthetic quality, or increase commercial activity through entrepreneurial governance (Sairinen and Kumpulainen, 2006). Disinvestment in industry, and passive or non-managment of vegetation can also increase green space and biodiversity at the waterfront. In this paper, we emphasize aspects of waterfront revitalization centered on natural capital, the stock of natural assets – the geology, soil, air, water, and all living things that comprise the biodiversity, structure, and function of natural ecosystems – that provide humans with the goods and services that make life possible and worth living (Costanza and Daly, 1992; WFNC, 2018).

The economic benefits of ecological restoration and community revitalization in the Great lakes are significant and well documented (Isely et al., 2018; Hartig et al., 2020; IAGLR, 2021). Improvement in human well-being (HWB) resulting from the restoration or revitalization of waterfront spaces is presumed but not often confirmed empirically (Villanueva et al., 2015; Martin and Lyons, 2018; Roe et al., 2019). In a recent review, we found few examples of revitalization efforts in the Great Lakes for which indicators of non-economic benefits were used to assess outcomes (Angradi et al., 2019).

The definition of HWB is unresolved (Ryan and Deci, 2001; Smith et al., 2013; King et al., 2014). The Millennium Ecosystem Assessment (after Reid et al., 2005, page v.) defines HWB as including a secure and adequate livelihood, enough food, shelter, clothing, access to goods, health, a healthy physical environment, good social relations, mutual respect, and the ability to help others and provide for children, personal safety, security from disasters, and freedom of choice and action. Sources of variation in HWB can be difficult to generalize because it is individualized, culturally rooted, and dynamic, varying with age and life experience of every person (Ryan and Deci, 2001; Yocom et al., 2016). Well-being includes objective and subjective dimensions (Summers et al., 2018). Socioeconomic status, educational attainment, mental and physical health, and social connections are objective – they reflect quantifiable reality regardless of how they are perceived by the individual. Subjective components of HWB include individuals’ perception of their objective life circumstances, their spiritual and cultural fulfillment, life satisfaction, happiness, affective state, and more (Smith et al., 2013; King et al., 2014).

Collecting subjective data for individuals is expensive, time consuming, and can require intensive oversight. Subjective well-being data are available for a limited number of places, and not at all at the local scale for coastal Great Lakes cities. Standardized subjective well-being data are generally only available at scales of county, region, or larger and can rarely be aggregated to a specific location (King et al., 2014). There are, however, several sources of objective HWB and natural capital data collected nationally or for larger U.S. urban areas at the census tract or census block-group scale. Indicators based on these data do not depend on the self-reported well-being of individuals; they are based on objective measurements of the biophysical environment and the socioeconomic and health status of a census tract’s residents. This paper explores the use of those indicators in waterfront revitalization applications.

Our rationale for attempting to link natural capital indicators to HWB indicators is based on a large literature (e.g., Peen et al., 2010; Lee and Maheswaran, 2011; Jackson et al., 2013; Keniger et al., 2013; Hartig et al., 2014; Chawla, 2015; Sandifer et al., 2015; Jennings et al., 2016; Seymour, 2016; Frumkin et al., 2017; Gascon et al., 2017; Browning and Rigolon, 2018; Kondo et al., 2018; Pun et al., 2018; Bratman et al., 2019; and Dadvand and Nieuwenhuijsen, 2019). Many of the referenced studies report plausible and statistically significant (albeit often weak) associations between nature (e.g., ecosystem services, biodiversity, green space, natural capital) and well-being (Gascon et al., 2018; Pearson et al., 2019). The accumulated evidence from this research suggests that increasing or improving natural capital or access to natural capital in waterfront spaces is likely to positively impact the well-being of people living in and using these spaces.

Methods

Research approach and questions

This is a residential exposure or indirect contact study (White et al., 2021), wherein the HWB benefits of a census tract’s natural capital accrue to the census-tract residents. Residency in the tract is a proxy for the cumulative exposure to natural capital from living and recreating in the census tract. Our presumption is that if associations currently exist between variation among census tracts in natural capital amenities and variation among tracts in HWB benefits realized by residents, then plausible forecasts are possible for how a comparable future change in waterfront census-tract amenities (intentional or passive) may be associated with future change in HWB. This is an observational study of a Great Lake waterfronts at a point in time; we make no case for cause-and-effect relationships between natural capital and HWB (Siegel and Stenson, 1999).

Using publicly available data for five coastal Great Lakes cities, we addressed four questions:

1. Which census-tract scale waterfront natural capital amenities are most consistently, unambiguously, and strongly associated with indicators of HWB?

2. What covariates confound or help explain the associations between natural capital and well-being metrics?

3. Do any combinations of metrics and demographic covariates have power to predict human well-being at the census-tract scale?

4. Can an observational (cross-sectional) study based on available census-tract-scale data support the hypothesis that attributes of natural capital amenities at the waterfront influence the objective well-being (economic, social, and health) of waterfront residents?

Case study communities

We selected five Great Lakes communities for study: Duluth, Minnesota and Superior, Wisconsin (combined); Green Bay, Wisconsin; Milwaukee, Wisconsin; Chicago, Illinois and northeast Indiana (combined, Fig. 1; maps of all study locations in Appendix A, Part 1); and Cleveland, Ohio. We selected these communities because we had census-tract scale indicator data on multiple natural capital amenities for them from the U.S. Environmental Protection Agency’s EnviroAtlas program (USEPA, 2020). The boundaries for each community were based on the EnviroAtlas community boundaries. Waterfront census tracts were defined as current census tracts that bordered a Great Lake, river, or harbor shoreline. For rivers, the upstream extent of what was considered waterfront was subjective if an EnviroAtlas community boundary was not reached. We obtained data from 221 waterfront census tracts, although not all metrics were available for all tracts (see Appendix A, Part 5). Each of the five communities contained a Great Lakes Area of Concern: the St. Louis River, Minnesota and Wisconsin; Lower Green Bay and Fox River, Wisconsin; Milwaukee Estuary, Wisconsin; Waukegan Harbor, Illinois; Grand Calumet River, Indiana (Chicago area); and Cuyahoga River, Ohio (USEPA, 2021)

Fig 1.

Waterfront census tracts for Chicago, Illinois, U.S.A. area including northwest Indiana. Maps for all communities are in Appendix A, Part 1.

U.S. census tracts have a population size between 1,200 and 8,000 people, with an optimum size of 4,000 people. The area of census tracts varies widely depending on population density (USCB, 2021a). Tracts are intended to be homogeneous units with respect to population characteristics, economic status, and living conditions at the time of establishment. In urban areas with dense populations, census tracts may be analogous to neighborhood units which have an ideal population size of 5,000–6,000 (Grover et al., 2016). Our analysis is thus at approximately the urban neighborhood scale (Seymour, 2016) but is not necessarily of recognized neighborhoods.

Indicators and metrics

We downloaded objective HWB and natural capital metrics (or data to derive metrics) from federal, academic, and commercial sources. For more information and links to data sources see Appendix A, Part 2. We identified metrics which we considered potentially responsive to intentional or passive biophysical changes at the waterfront associated with natural capital improvement projects including, for example, remediation, brownfield reuse, habitat restoration, or socioeconomic revitalization. We attempted to compile indicators that addressed a range of natural capital attributes of waterfront census tracts including shorelines, land cover, green space (vegetated land, including agriculture, lawns, forests, wetlands, and gardens), open space, waterbodies, and the atmosphere. In this paper, we refer to categories of metrics related to a benefit (e.g., access to water) as indicators, and things that can be measured, counted, or otherwise quantified as metrics (e.g., density of water access sites as count/km²).

Except for the EnviroAtlas and the Neighborhood Atlas data (Appendix A, Part 2), all the data from which we extracted or derived indicators a were available at the census-tract scale. Census block-group scale data from the EnviroAtlas and the Neighborhood Atlas were aggregated to the census-tract scale as population-weighted block group means. We used population rather than area weighting because block group population can vary widely from about 300 to 6000 (USCB, 2021a) so that at the census-tract scale the natural capital attributes (including per capita attributes) are on average nearer to each resident than they would be based on area-weighting. We did not interpolate census-tract-scale data from coarser geographic scales.

For HWB indicators, we included objective metrics from multiple HWB domains (Smith et al., 2013) including education, health, living standards, social cohesion, leisure time, and safety and security. Into health we lumped mental and physical health and unhealthy behaviors (Table 1a). Not included due to lack of data are metrics from primarily subjective domains: connection to nature, life satisfaction and happiness, and spiritual and cultural fulfillment.

Table 1a.

Candidate human well-being (HWB) metrics and covariates of associations included in this study in alphabetical order by metric label.

Candidate HWB metric	Expectation	Metric label	Data source	HWB domain	Variable name in source data
National percentile of block group ADI score (0–100; higher is worse)	n	ADInational	ADI	Living standards	ADI_NATRANK
State-specific decile of block group ADI score (0–100; higher is worse)	n	ADIstate	ADI	Living standards	ADI_STATERNK
% binge drinking ages ≥18 y in last 30d	n	Binge_p	500c	Health	BINGE_CrudePrev
Child age dependency ratio (<18/18–64y) in 2019	Amb	Childage_r	ACS	Social cohesion & living standards	S0101_C01_036E
% educational attainment >25 y with BA College in 2019	p	College_p	ACS	Education	S1501_C02_015E
% divorces in people older than 15 y in 2019	n	Divorce_p	ACS	Social cohesion	S1201_C04_001E
Employment ratio (percent of population 20 to 64y employed in 2019)	p	Employment_r	ACS	Living standards	S2301_C03_021E
% adults ≥18y with hypertension	n	Highbp_p	500c	Health	BPHIGH_CrudePrev
Median census-tract home value in 2019	p	Homevalue	ACS	Living standards	B25077_001M
Median census-tract household income in 2019 (covariate)	p	Houseincome	ACS	Living standards	S1903_C03_001E
Fraction children incarcerated in 2010 in census tract	n	Jail_p	OA	Safety and security	Incarceration_rate_rP_gP_pall
Life expectancy at birth in years in census tract	p	Lifeexpect	USLEEP	Health	e (0)
% adults ≥18y with no leisure time exercise in last 30d	n	Noexercise_p	500c	Leisure time	LPA_CrudePrev
Percent of residents identifying as nonwhite in the census tract (covariate)	na	Nonwhite_p	ACS	na	Derived from ACS data
% obesity among adults ≥18 y in the census tract	n	Obesity_p	500c	Health	Obesity_CrudePrev
Old age dependency ratio (≥65/18–64y) in the census tract	Amb	Oldage_r	ACS	Social cohesion & living standards	S0101_C01_035E
% owner occupied housing in the census tract	p	Owned_p	ACS	Living standards	DP04_0046PE
Mean of several strongly intercorrelated percentage metrics of poor health	n	Poorhealth_p	500c	Health	Derived. See Appendix A, Part 4
% adults ≥18 y in census tract with poor mental health ≥14 days in last 30d	n	Poormental_p	500c	Health	MHLTH_CrudePrev
% adults ≥18 y poor physical health ≥14 days in last 30d in the census tract	n	Poorphysical_p	500c	Health	PHLTH_CrudePrev
% adults≥18 y in the census tract Sleeping <7 h/night in last 30d	n	Poorsleep_p	500c	Health	Sleep_CrudePrev
Census-tract population density in 2019 (number per mile2) (covariate)	na	Popdensity	ACS	na	Derived in GIS; multiply by 2.59 to get number per km2
% families below the poverty level in the census tract	n	Poverty_p	ACS	Living standards	S1702_C02_001E
Current % smoking among adults aged ≥18 y in the census tract	n	Smoking_p	500c	Health	CSMOKING_CrudePrev
% of children in 2015 living in a tract they grew up in	Amb	Staying_p	OA	Social cohesion & living standards	%_Staying_in_Same_Tract_as_Adults_rP_gP_pall
Fraction of children that grew up in the tract with a high income as an adult in their mid-thirties	p	Success_p	OA	Living standards	Frac. _in_Top_20%_Based_on_Indiv_Income_rP_gP_pall
Overall social vulnerability index of the census tract (percentile ranking from 0–1; higher is worse)	n	SVIall	SVI	Safety and security	RPL_themes
Socio-economic vulnerability index of the census tract (percentile ranking from 0–1; higher is worse)	n	SVIse	SVI	Safety and security	RPL_theme1

Final metrics in bold type. All metrics are at the census-tract scale. Data sources refer to codes in Appendix A, Part 2. Expectation is the expected direction of the correlation between the metric and a positive (p) or negative (n) relationship with a positive metric of the other type; see text for explanation. For metric suffixes, _p = percent, _r = ratio. Amb = ambiguous metric (discussed in text); ADI=Area Deprivation Index; HWB=human well-being; ACS = American Community Survey; OA = Opportunity Atlas; 500c=500 Cities database; USLEEP = U.S. Small Area Life Expectancy Estimates Project; SVI = Social Vulnerability index; na = not applicable (covariate). HWB domains from Smith et al. (2013).

We included two multimetric HWB indices, the Social Vulnerability Index (SVI), and the Area Deprivation Index (ADI) that have used in applied research for assessment of community resilience to disaster (SVI, Flanagan et al., 2011) and neighborhood socioeconomic deprivation (ADI; UW, 2021; see Appendix A, Part 2 for more details). We have not included as metrics any of the several human well-being indices (also known as health and welfare indices, quality of life indices, or urban health indicator tools) in our study because they usually include a combination of economic, social, health, and environmental (or ecosystem service) metrics (Smith et al., 2013; Pineo et al., 2018), so they inherently confound the association between natural capital and HWB (ADI and SVI do not include natural capital indicators).

Metric expectations

Human well-being metrics may be positively or negatively related to natural capital metrics (Table 1a). Likewise, natural capital metrics can indicate positive or negative attributes of the waterfront (Table 1b). We assumed percent impervious surface (Impervious_p), per capita impervious surface (Impervious_pc), flood probability, (Flood_p), minimal tree views (Mintreeview_p), distance to parks (Parkdist, Park>500m_p), and percent artificial shorelines (Artificialshore_p) were negative metrics because they are associated with risk or the absence of or distance to natural capital attributes. Brownfield and Superfund metrics (BFcount and SFcount) are ambiguous natural capital metrics because this designation can indicate both past contamination and land potentially restored to beneficial use. For this report we presumed that counts of brownfields and Superfund sites are negative attributes of census tracts with respect to human well-being. We assumed the remaining metrics were positive.

Table 1b.

Candidate natural capital (natural capital) metrics included in this study in alphabetical order by metric label.

Candidate natural capital metric	Expectation	Metric label	Data source	Variable name in source data
% artificial shoreline in census tract	n	Artificialshore_p	NOAA	Type code 190
% beach shoreline in census tract	p	Beachshore_p	NOAA	Type codes 140, 150;
Number of brownfield sites in census tract	n	BFcount	ACRES	Derived
Bikeability index (0–100); higher is better	p	Bikeability	Walk	Bike_score
Reduction in daytime temperature by trees (C°)	p	Coolingday	EnviroAtlas	Maxtempreduction
Reduction in nighttime temperature by trees (C°)	p	Coolingnight	EnviroAtlas	Maxtempreductionnight
% CO removed by trees in census tract	p	COremoved_p	EnviroAtlas	COAQYr
% green space in census tract	p	Greenspace_p	EnviroAtlas	Green_p
Green space per capita (m 2 /person)	p	Greenspace_pc	EnviroAtlas	Green_pc
% land area with 1% chance of annual flooding	n	Flood_p	EnviroAtlas	FP1_land_p
% impervious surface in census tract	n	Imperv_p	EnviroAtlas	Imp_p
Impervious surface per capita (m 2 /person) in census tract	n	Imperv_pc	EnviroAtlas	Imp_pc
% population with minimal views of trees in the census tract	n	Mintreeview_p	EnviroAtlas	WVT_pct
% natural shore other than beach in the census tract	p	Naturalshore_p	NOAA	Type codes 100–180 excluding beach
% area of green amenities in the census tract	p	Greenarea_p	OSM	Derived; see Appendix A, Part 3
Hybrid (seminatural) recreational or interactional amenity density (count/km2)	p	Hybridrec_d	OSM	Derived; see Appendix A, Part 3
Stream density (km/km 2 ) in the census tract	p	Streams_d	OSM	Derived; see Appendix A, Part 3
Trail density (km/km 2 ) in the census tract	p	Trails_d	OSM	Derived; see Appendix A, Part 3
% of population in the census tract >500m from park entrance	n	Park>500m_p	EnviroAtlas	BWDP_pct
Average walking distance to Park entrance (m) in census tract	n	Parkdist	EnviroAtlas	ParkProxD
% of particulate matter >2.5 microns and <10 microns removed annually by trees in the census tract	p	PM 10 removed_p	EnviroAtlas	P10AQyr
% of particulate matter <2.5 microns removed annually by trees in the census tract	p	PM 2.5 removed_p	EnviroAtlas	P25AQYr
% tree cover in riparia within 50 m buffer in the census tract	p	Ripariantree_p	EnviroAtlas	RB50_ForP
Number of Superfund sites in census tract	n	SFcount	EnviroAtlas	Derived in GIS from EnviroAtlas data
% tree cover in the census tract	p	Treecover_p	EnviroAtlas	Mfor_p
Per capita tree cover (m 2 /person)	p	Treecover_pc	EnviroAtlas	Mfor_pc
Census tract Walk Score (0–100; higher is better)	p	Walkability	Walk	Walk_score
Water access amenity density in the census tract (count/km2)	p	Wateraccess_d	OSM	Derived; see Appendix A, Part 3
% population in the census tract with view of water within 50 m of home	p	Waterview_p	EnviroAtlas	WVW_pct
% wetlands in the census tract	p	Wetland_p	EnviroAtlas	Wet_p

Final metrics in bold type. All metrics are at the census tract scale. Data sources refer to codes in Appendix A, Part 2. Expectation is the expected direction of the correlation between the metric and a positive (p) or negative (n) relationship with a positive metric of the other type. For metric suffixes, _p = percent, _pc = per capita, _d = density. OSM = OpenStreetMap; NOAA = National Oceanic and Atmospheric Agency; ACRES = (EPA) Brownfields Grantee Assessment Cleanup and Redevelopment Exchange System; Walk = walk score.

The Area Deprivation Index (ADI) and the Social Vulnerability Index (SVI, ATSDR, 2021) are formulated as negative HWB metrics (i.e., a higher value indicates worse social conditions). Fraction of incarcerated residents (Jail_p), rate of poverty (Poverty_p), and divorce rate (Divorce_p) are negative metrics.

A high child (Childage_r) or old age (Oldage_r) dependency ratio indicates that the economy faces a greater burden to support and provide the social services for dependent children and older residents (UN, 2021). We therefore initially considered these to be negative community-scale HWB metrics. The HWB metrics Childage_r, Oldeage_r, and percent of residents living in the tract they grew up in (Staying_r), are conceptually ambiguous, however, because the expected direction of correlations depends on other factors (e.g., in the developing world versus the developed world) or how the indicator is defined (e.g., as a social burden indicator or as a social cohesion indicator). For these three HWB metrics we report the strength, consistency, and ambiguity of the metrics both as positive and as negative metrics.

Two HWB metrics with negative expectation (as natural capital improves, metric value decreases), percent adults ≥18 years old reporting poor mental health for ≥14 days in the last 30d (Poormental_p), and percent adults ≥18 years old reporting poor physical health for ≥14 days in the last 30 d (Physical_p), are based on self-reporting (CDC, 2020).

When we refer to natural capital amenities in this paper, we mean amenities on a gradient from entirely natural (e.g., forest, wetland) to seminatural and built environmental amenities (after Childers et al., 2019), among which we include pedestrian infrastructure, public open spaces (e.g., sport fields, parks, plazas), and developed access amenities such as beaches, boat landings, or piers. We include these “hybrid” features on our gradient of natural capital amenities because they provide access to a critical natural capital amenity, water, that would otherwise provide fewer benefits.

The metric Walkability (Walk Score, 2021) is an index based on walking routes to nearby amenities (e.g., schools, parks, natural, and cultural amenities). Points are awarded based on the distance to amenities in each category. It also integrates road metrics such as block length and intersection density. The vendor characterizes a walkability score of 90–100 as a “walker’s paradise” where daily errands to not require a car, and a score of 0–24 as car dependent (Walk Score, 2021). We included this metric because it has been hypothesized that residents of neighborhoods with high walkability engage in more healthy physical activity than residents of less walkable neighborhoods (Frank et al., 2010).

Analytical approach

We combined downloaded datasets into a single table using Esri ArcGIS Pro software (Esri, 2021). We compared Spearman rank correlation coefficients (r_s) among natural-capital and among HWB metrics to determine if any were sufficiently highly correlated (r_s>0.9) to warrant dropping them from the analysis as redundant to reduce the analytical burden. We created three equal-sized classes for three covariate demographic metrics, population density (Popdensity), percent of residents that were nonwhite (Nonwhite_p), and median household income (Houseincome) using percentiles (33.3 and 66.6) of the population. We used the Chi-square test of independence among covariates. We compared metric means among communities and among covariate classes using the Kruskal-Wallis test for Wilcoxon scores (alpha=0.05). We computed r_s for correlations between natural capital and HWB metrics for all census tracts and by covariate class. We identified which associations between natural capital and HWB metrics were significant, relatively strong (|r_s | ≥0.5; according to Cohen, 2013, an r value of ≥0.5 is a “large” association for data of this type), and in the expected direction (see Table 1a–b). Preliminary plots revealed several of the associations had outliers and many were nonlinear. Spearman rank correlations are more robust to outliers and capture the strength of monotonic but nonlinear relationships better than Pearson correlation (Schober et al., 2018). We used locally weighted scatterplot smoothing (LOWESS) to highlight nonlinear associations between metrics.

We calculated a consistency value for each final (non-redundant) HWB and natural capital metric as the percent of metrics of the other type (e.g., HWB vs. natural capital) for which the correlation was significant and in the expected direction. This value is a measure of the general responsiveness of each HWB metric to the final set of natural capital metrics and a measure of the general relevance of each natural capital metric to the final set of HWB metrics. To quantify ambiguity in associations between metrics, we calculated an ambiguity ratio for each metric as 1 – (the number of correlations that were significant and in the expected direction / the number of correlations that were significant). The ratio ranges from 0 (all the significant correlations were in the expected direction) to 1 (none of the significant correlations were in the expected direction). We made these calculations for all correlations and by covariate class.

We examined biplots for all correlations between natural capital and HWB metrics where |r_s | ≥0.5 for outliers, repeated zero values, or multi-model relationships likely to confound the interpretation and render the metrics less reliable for application.

We created multiple linear regression models to determine if combinations of natural capital metrics and covariates had any power to predict HWB metrics. We examined normal probability plots, skewness, and other diagnostic parameters for transformed and untransformed metrics. Based on this, percentage-based metrics were arcsine-square root transformed y=(arcsine(√(x/100)); count metrics were square-root transformed (y= (x+0.05)), and amount and density metrics were log₁₀-transformed (y=log₁₀(x +1)). For each HWB metric, we calculated Akaike’s information criterion (AIC; DeLeeuw, 1992) for every possible subset of natural capital and covariate variables. We used the AIC to select the “best” regression model (with up to 10 predictors) for each dependent variable. For each selected model we report significance and fit statistics (F, R²), parameter estimates, and standardized β coefficients (± 95% confidence limits), which provide a measure of effect size for each predictor independent of the units of measure. Statistical analyses were performed using software licensed to the U.S. Environmental Protections Agency, including SAS for Windows V. 9.4 (SAS, 2016), and Sigmaplot for Windows V. 14 (SYSTAT, 2017).

Results

Metric screening

Several metrics for poor health or unhealthy behaviors from the 500 Cities dataset were strongly rank correlated with each other (r_s>0.9, Appendix A, Part 4). We created a summary metric, Poorhealth_p, which replaced health metrics Poorsleep_p, Smoking_p, Poorphysical_p, Obesity_p, and Noexercise_p with their mean. We dropped two other HWB metrics, a version of the Area Deprivation Index based on state ranking (ADIstate) which was highly correlated with ADI based on a national ranking (ADInational; rs=0.95) and the socioeconomic version of the Social Vulnerability Index (SVIse) which was highly correlated with overall SVI (SVIall; r_s=0.90).

We dropped four natural capital metrics from the analysis because of strong correlations with other natural capital metrics. We dropped the EnviroAtlas metrics Coolingday and Coolingnight, which were both highly correlated (r_s ≥0.98) with percent tree cover (Treecover_p), from which the cooling metrics are derived. We dropped percent CO removed by trees (COremoved_p), which was highly correlated (r_s=0.98) with particulate air pollution removed by trees (PM₁₀removed_p). Walkability and census-tract bike score (Bikeability) were correlated (r_s=0.91); we dropped Bikeability because it is calculated using data irrelevant to natural capital (i.e., hills, bike sharing). All the significant rank correlations causing metrics to be dropped were between metrics from the same data source. After metric screening there were 19 final HWB metrics and 26 final natural capital metrics.

Variation in metric means among communities and covariate classes

Variation in natural capital and HWB among communities and among classes of demographic attributes (population density, racial composition, household income), provides context for the regional scale of this study and reveals associations between demography and natural capital and HWB. Figure 2 shows covariate class means for a selection of natural capital and HWB metrics representative across HWB domains and natural capital amenity types; see Appendix A, Part 5 for a complete listing of metric means.

Fig. 2.

Mean (±95% confidence intervals) for selected natural capital (top row) and HWB (bottom row) metrics by covariate class; “ns” indicates that the effect of class on the metric was not significant. Complete statistics are available in Appendix A, Part 5

The mean number of brownfields (BFcount) per waterfront census tract (BFcount) was highest in Green Bay and lowest in Chicago-area waterfront census tracts. Parks were furthest from residents (Parkdist, Park>500m_p) in Cleveland and closest in Chicago. Mean tree cover was highest in Duluth-Superior and lowest in Green Bay. Mean walkability was highest in Chicago and lowest in Duluth/Superior census tracts. Covariate means varied among communities. Mean population density, percent nonwhite residents, and household income were all highest in the Chicago area. Population density was lowest in Green Bay. Mean percent nonwhite residents and household income were lowest in Duluth-Superior.

In most cases, metrics varied among covariate classes in the expected or normative direction. Class means of natural capital metrics indicating “greener,” amenities were higher in census tracts with low population density (except parks which were more distant), fewer nonwhite residents, and higher household income. The effect of household income on mean walkability was not significant. HWB metrics indicating positive well-being were higher in census tracts with low population density, fewer nonwhite residents, and higher household income (Fig. 2). Exceptions were poor health (Poorhealth_p, Poormental_p), which were higher, and College_p, which was lower in low density tracts.

The population density of census tracts was not independent of the racial composition of census tracts. Tracts with a higher population density corresponded to census tracts with higher nonwhite populations more often than expected due to chance (p<0.05). Racial composition was also not independent of household income (p<0.05). Household income class was independent of population density class. Affluent tracts were not more likely to have a high or low population density than expected due to chance (p>0.10). These categorical findings were corroborated by rank correlations. Population density was significantly correlated with percent nonwhite residents in census tracts (Fig. 3a; r_s=0.36, p<0.05), and Houseincome was inversely correlated with percent nonwhite residents (Nonwhite_p; Fig. 3b; rs=-0.61, p<0.05), but Houseincome was not correlated with population density (Popdensity, Fig. 3c; p=0.14).

Fig. 3.

Correlations between covarying metrics influencing associations between natural capital and HWB metrics. Values on plots are the Spearman rank correlation coefficients (r_s). Reference lines are the high, medium, and low covariate class boundaries. Popdensity is shown on a log scale.

Because of the unequal sample sizes among communities (range N=14 to 88) and the significant effects of other factors generalizable across communities such as population density and household income, we dropped community as a covariate. Because variation among tracts in percent of the residents that were nonwhite was related to variation in both population density and household income, we dropped Nonwhite_p as a covariate metric.

Correlations between HWB and natural capital metrics

For all census tracts, 214 of 572 correlations (37%) between the final set of 19 HWB and 26 natural capital metrics (the number is higher than 26 * 19 because some metrics were analyzed as both positive and negative metrics) were statistically significant (p<0.05). One hundred and fifty-three correlations (27%) were both statistically significant and varied with metrics of the other type in the expected direction (expectations given in Table 1a–b).

Across covariate classes, HWB metrics most consistently correlated in the expected direction with natural capital metrics were Owned_p (mean consistency=41%), College_p (28%), Homevalue (25%), and ADInational (25%) (Table 2). HWB metrics that were rarely correlated with natural capital metrics in the expected direction included residents living in the tract they grew up in (Staying_p, 5%), Binge_p, (6%), and Highbp_p (7%). The most consistent natural capital metrics were BFcount (42%), Hybridrec_d, (33%), and Imperv_pc (32%). Least consistent (least often correlated with HWB indicators, Table 2) natural capital metrics were percent green landcover from OpenStreetMap data (Greenarea_p, 4%); percent wetland land cover (Wetland_p, 6%), trail density (Trails_p, 8%) and Waterview (8%).

Table 2.

Mean consistency, mean ambiguity ratio, number of strong rank correlations (r_s>0.50) across all six covariate classes (see Appendix A, Part 6 for a complete listing), and metric rating for each metric.

Metric	Mean consistency (%)	Mean ambiguity ratio	Number of strong rank correlations	Metric rating	Regression model total R2
HWB metrics
Owned_p	41	0.30	16	1	0.78
College_p	28	0.31	12		0.85
Poorhealth_p	17	0.23	9		0.88
Poormental_p	16	0.38	8		0.76
Homevalue	25	0.25	8		0.82
Success_p	23	0.29	7		0.71
ADInational	25	0.30	7		0.78
Houseincome	23	0.01	5		0.32
Jail_p	24	0.27	3	2	0.67
Highbp_p	7	0.36	3		0.71
Lifeexpect	19	0.13	2		0.50
SVIall	20	0.03	1		0.46
Poverty_p	16	0.07	1		0.66
Childage_r	12 (23)	0.68 (0.32)	5 (5)	3	0.36
Oldage_r	9 (22)	0.55 (0.46)	0 (1)		0.23
Binge_p	6	0.51	2		0.63
Staying_p	5 (16)	0.77 (0.23)	0 (0)		0.22
Employment_r	14	0.39	0		0.35
Divorce_p	14	0.38	0		0.29
Natural capital metrics
Walkability	25	0.43	16	1	na
Imperv_pc	32	0.10	10		na
Ripariantree_p	20	0.71	8		na
Treecover_p	21	0.38	8		na
Imperv_p	18	0.47	7		na
Artificialshore_p	25	0.32	6		na
Hybridrec_d	33	0.15	6		na
PM 10 removed_p	27	0.26	6		na
BFcount	42	0.14	6	2	na
Treecover_pc	12	0.60	5		na
Flood_p	16	0.30	4		na
Mintreeview_p	12	0.17	3		na
Parkdist	15	0.22	2		na
Greenspace_pc	10	0.67	3	3	na
Greenspace_p	13	0.72	1		na
Wateraccess_d	19	0.68	1		na
Trails_d	8	0.28	1		na
PM 2.5 removed_p	9	0.41	1		na
Beachshore_p	13	0.17	0		na
Naturalshore_p	22	0.36	0		na
Greenarea_p	4	0.63	0		na
Streams_d	10	0.37	0		na
Park500m_p	13	0.50	0		na
SFcount	33	0.15	0		na
Waterview_p	8	0.55	0		na
Wetland_p	6	0.62	0		na

A consistency of 50% means that half the correlations with metrics of the other type (i.e., HWB vs. natural capital) were significant and in the expected direction. An ambiguity ratio of 1 means that none of the significant correlations with metrics of the other type were in the expected direction. Values in parentheses for Staying_p apply if this metric is considered a negative HWB indicator. Values in parentheses for Oldage_r and Childage_r apply if these metrics are considered positive HWB indicators. Regression results are described in the next section. Na= not applicable.

Human well-being metrics with the fewest ambiguous correlations across covariate classes were Houseincome (mean ambiguity ratio=0.01; Table 2), SVIall (0.03), Poverty_p (0.07) and Lifeexpect (0.13). The most ambiguous HWB metrics were Staying_p (0.77), Childage_r (0.68), Oldage_r (0.63), and Binge_p (0.51%). The least ambiguous natural capital metrics were Imperv_pc (0.10), BF_count (0.14), SF_count (0.15), and Hybridrec_d (0.15). The most ambiguous natural capital metrics were, Greenspace_p (0.72), riparian tree cover (Ripariantree_p, 0.71), Wateraccess_d (0.68), and per capita green space (Greenspace_pc, 0.67).

Across covariate classes, ninety-four correlations were significant in the expected direction and had an |r_s| value ≥0.5 (Appendix B). The largest number of strong correlations (35) was for census tracts in the low population density class. The fewest (5) was for the low household income class. The HWB metrics for which the most correlations with natural capital metrics were strong and in the expected direction were Owned_p, (16 correlations), College_p (12), Poorhealth_p (9), Poormental_p (8), and Homevalue (8). The natural capital metrics with the strongest correlations with HWB in the expected direction were Walkability (16 correlations), Imperv_pc (10), Treecover_p (8), Ripariantree_ p (8), and Imperv_p (7).

Biplots of the strong (|r_s| ≥0.5) rank correlations showed the relationships were often non-linear (Fig. 4). A recurring trend across covariate classes and HWB metrics were that the association between Walkability and HWB was strongest (steepest slope) at higher levels of Walkability, even in the most urban (high population density) census tracts (Appendix B). For example, across all tracts, the correlation between Walkability and Poorhealth_p was weak at a Walkability value <60; (Fig. 4a, see Appendix A, Part 7 for all biplots). For census tracts with a high population density, the association between Walkability and HWB metrics was weak for census tracts with a Walkability value <80 (Fig. 4b–c).

Fig. 4.

Selected plots of rank correlations between natural capital (x) and HWB indicators (y). Rank correlation coefficient (r_s) provided for each correlation. Where possible, lines were fit using locally weighted scatterplot smoothing (LOWESS) for a first-degree polynomial with a sampling proportion of 0.5. Biplots of all correlations with |r_s|≥0.5 available in Appendix A, Part 7 See Table 1a–b for metric definitions. Complete set of plots given in Appendix A, Part 7.

There was an analogous pattern for Imperv_p and Imperv_pc and HWB metrics. Increases in Imperv_p beyond about 30% and increases in Imperv_pc beyond about 400 m² generally had little effect on HWB metric values, although data were fewer in that range (Fig. 4d–f).

In census tracts with a low population density, particulates removed by trees (PM₁₀removed_p), Ripariantree_p, and Treecover_p were often correlated with HWB metrics (Fig. 4g–i). HWB metrics College_p and Success_p, were high only in tracts with no brownfield sites (Fig. 4j–k). Natural capital metrics based on distance to a park (Parkdist, Park>500m_p) were rare among strong correlations with HWB metrics (Appendix B). In census tracts with a medium income, distance to a park entrance was inversely correlated with College_p (Fig. 4l).

Metric rating

Based on the foregoing analyses we rated the relative reliability of HWB and natural capital metrics for potential application in waterfront revitalization design and assessment. The rating is based on metric consistency and ambiguity of correlations, number of strong correlations, and number of zero values. We rated 8 of 19 HWB metrics as likely to be reliable (rating=1) as indictors of HWB benefits (Table 2). These metrics had moderate to high consistency, moderate to low ambiguity, multiple (≥5) strong significant correlations with HWB metrics, and few zero values. We rated six HWB metrics as potentially useful (rating=2). These metrics had moderate to low consistency, moderate to high ambiguity, and <5 strong correlations. Five HWB metrics were rated as likely unreliable (rating=3) for the census tracts in our study. They had high ambiguity and low consistency and/or ≤1 strong correlation with natural capital metrics. An exception was Childage_r, which we rated as not reliable in the current context because it had an equal number (5) of correlations as a negative social burden metric and as a positive social cohesion metric.

We rated 8 of 26 natural capital metrics as likely reliable (rating=1, Table 2). These metrics had low to moderate ambiguity, moderate to high consistency, at least 5 strong correlations and were not dominated by zero values. Two metrics Ripariantree_p and Treecover_p had high ambiguity overall but had multiple strong correlations in low population density tracts and are likely to be reliable for those census tracts. We rated 5 metrics as potentially reliable (rating=2). These metrics had moderate to low ambiguity, moderate to low consistency, and <4 strong correlations. We rated BFcount a 2 despite 6 strong correlations and high consistency, because most (73%) census tracts had no brownfields. We rated 13 metrics not reliable (rating=3) in the waterfront setting. These metrics had low to high ambiguity and low consistency, and ≤3 strong correlation with HWB metrics. Two of the metrics rated 3, SFcount and tributary stream density (Streams_d), were mostly zero values (80% and 70% respectively). Natural capital metrics that we scored a 3 (Table 2) were excluded as predictors from regression models.

Predicting HWB

Combinations of transformed (see methods) natural capital metrics and the two covariates, Houseincome and Popdensity, predicted several HWB metric values quite well (R² ≥0.70), including, Poorhealth_p, Homevalue, College_p, ADInational, Owned_p, Poormental_p, Highbp_p, and Success_p (See Appendix A, Part 8 for regression estimates). The best model was for Poorhealth_p (R²=0.88), which was predicted by Houseincome, Walkability, Imperv_pc, Parkdist, Mintreeview_p, and Flood_p.

The HWB metrics Jail_p, Poverty_p, Binge_p, and Lifeexpect were moderately well predicted (R²=0.50–0.67). The metrics SVIall, Childage_r, Employment_r, Divorce_p, Oldage_r, Houseincome, and Staying_p, were poorly predicted (R² < 0.46). For 12 of 19 models, Houseincome was the predictor with the largest effect size (Fig. 5). In the remaining models, Treecover_p (4 models), PM₁₀removed (2 models), and Imperv_pc (1 model) were the most influential predictors. Across all models, the most frequently included predictors were Houseincome (16 models), Treecover_p (12 models), PM₁₀removed (12 models), Imperv_p (12 models), Imperv_pc (11 models), and Mintreeview (10 models).

Fig. 5.

Multiple regression results for HWB metrics plotted as effect sizes for independent predictors. Effect size values are standardized β coefficients with 95% confidence intervals. Coefficients with open symbols were not significantly different from zero. Symbols farthest from the dashed zero line had the largest effect on the model. R² values show on plots; only models with R² ≥0.5 shown; for all models and additional parameter estimates see the Appendix A, Part 8.

The regression results generally corroborate our ratings for the HWB metrics (Table 2). Natural capital metrics explained more of the variation in HWB metrics rated 1 or 2 than HWB metrics rated 3. The exception was the metric Houseincome, the model for which did not include the covariate Houseincome and was poorly predicted by natural capital metrics alone (R²=0.22).

The sign of the larger effects (i.e., β furthest from the zero line in Fig. 5) was consistent with our original expectations for HWB and natural capital metrics. For example, Houseincome was a positive effect for positive HWB metrics and a negative effect for negative HWB metrics (i.e., it always predicted good outcomes, with one exception, Binge_p, discussed below). Walkability, a positive metric, generally varied in the same direction with Houseincome, except for home ownership (Owned_p). In that case, higher walkability predicted lower home ownership. Another strong predictor, Treecover_p, generally predicted beneficial outcomes.

Discussion

Associations between natural capital and HWB

The HWB metrics that we rated as most reliable for application to waterfront revitalization (rating of 1) were mostly from the living standards (Houseincome, Owned_p, Success_p), health (Poorhealth_p, Poormental_p), and education (College_p) domains (Table 1a). Several low-rated metrics (rating of 3) were from the social cohesion domain including Staying_p, Divorce_p, and Childage_r.

For several HWB metrics, it was initially not clear if they were positive metrics of social cohesions (e.g., Staying_p as an indicator of social connections) or negative metrics (e.g., Staying_p as an indicator of limited social mobility). Likewise, Childage_r and Oldage_r may be negative indicators of the social burden of caring for children and senior residents but may also be positive indicators of social cohesion resulting from multigenerational households. Based on the correlations with natural capital metrics (Table 2), our results suggest that Childage_r and Oldage_r were positive (i.e., social cohesion metrics) HWB metrics and Staying_p was a negative (i.e., social mobility) metric for the waterfront census tracts included in our analyses. Elsewhere or for other socioeconomic conditions, the expectation for these metrics may be different.

The self-reported rate of binge drinking (Binge_p) seems to us an unreliable health metric at the census- tract scale. Associations between Binge_p and the natural capital metrics Flood_p and Mintreeview_p were unconvincing and sample size was small due to missing values for some covariate classes (Appendix A, Part 5). We speculate that the influence of census tract natural capital on the rate of binge drinking is confounded by the age of the population, since this behavior is more prevalent among young adults (Patrick et al., 2019; Binge_p was also negatively correlated with hypertension, Highbp_p, a condition most prevalent in people ≥60y; Yoon et al., 2015). Also, the number of nearby alcohol vendors or local restrictions, which we did not quantify, could be an important source of among-tract variation in binge drinking.

The HWB metrics Divorce_p and Employment_r were poorly rated (Table 2) and not strongly correlated with any natural capital metrics. Common reasons cited for divorce (e.g., Amato and Preveti, 2003) generally do not include attributes of local natural capital, although indirects links are certainly plausible. Plausible indirect connections between natural capital and employment rate at the census-tract scale are also possible, but it seems likely that both divorce and employment rate are more strongly related to social and economic drivers at other spatial scales.

The highest rated natural capital metrics included both positive (tree cover metrics, Hybridrec_d, Walkability) and negative metrics (impervious surface metrics, BFcount, Artificialshore_p, Mintreeview_p, Parkdist). Several natural capital metrics had high ambiguity (they were frequently correlated with HWB metrics in the unexpected direction). Greenspace metrics from the EnviroAtlas dataset, Greenspace_p and Greenspace_pc, and Greenarea_p from OpenStreetMap, had high ambiguity ratios which we speculate is related to the lack of specificity in the metric. Waterfront greenspace can include forest, lawn, gardens, industrial parks, waste ground, verges, wetlands, and a variety of covers types that may vary in how and how much they contribute to well-being. Several authors have noted that quality of green spaces can confound positive beneficial effects (Giles-Corti et al., 2005; Jennings et al., 2016; Ekkel and De Vries, 2017; Markevyche et al., 2017).

The ecological and often the aesthetic inverse of green space, impervious surfaces (Imperv_p, Imperv_pc, and Artificialshore_p), were highly rated negative metrics, strongly correlated with HWB metrics and included in multiple regression models. We suspect that impervious surfaces, and their perceived “quality,” is inherently less variable than green spaces. For example, there are only three categories of impervious surface for the developed land cover class (>20% impervious surface) and at least twelve vegetated National Land Cover types (<20% Impervious surface; MRLC, 2021).

Metrics representing more specific attributes of green space, particulate air pollution removed by trees (PM₁₀removed_p), and tree cover (Treecover_p, Ripariantree_p) were strongly correlated with HWB and included as significant predictors in regression models. Percent wetland (Wetlands_p) was not strongly correlated with any HWB metrics. Across all census tracts, the percent of census tract area classified as wetland was <1 percent (Appendix A, Part 5), too small an area to influence HWB. In other settings where wetlands provide important ecosystems services (e.g., water quality, shoreline protection), including cultural ecosystem services (e.g., hunting, birdwatching) or disservices (e.g., insect pests, disease), they may be more closely linked to the well-being of residents (de Jesus Crespo and Fulford, 2018).

Natural capital metrics with a high frequency of zero values, including SF_count, Wateraccess_d, and Streams_d, were rated as low and were infrequently strongly correlated with HWB metrics. The exception was the number of brownfields in census tracts (BFcount) for which 73% of census tracts had none, but it was nevertheless strongly correlated with HWB metrics. Plots of BFcount against HWB metrics (Fig. 4) show that the negative correlation between HWB and BFcount was a function of the presence of at least one brownfield rather than the total number of brownfields in the census tract. All high values for HWB were in census tracts with no brownfields. In most cases, brownfields make up a small fraction of the area of the census tract, and the apparent association with HWB may owe more to urban degradation or other negative attributes of waterfront census tracts that included brownfields than to the brownfields themselves.

Many of the significant correlations between natural capital and HWB metrics were non-linear (Fig. 4, Appendix A, Part 7). The relationship between the natural capital metric and the HWB metric varied over the plotted range of the natural capital metric or over the range of another related variable such as relative affluence or population density. The shape of these relationships can be used to infer important thresholds or limitations for the application of indicators.

Importance of covariates

The rank correlation and regression results show the strong effects of two covariates, population density and household income on the relationship between natural capital and HWB in waterfront census tracts. For example, the access-related amenity Walkability was only strongly correlated with HWB metrics in census tracts with a high population density (>5,129 residents/mile²). These are highly urbanized areas where amenities and intersections are concentrated and walkability is high (Walkability=81; see Appendix A, Part 5). The positive association between Walkability and health was weaker in less populous residential or rural census tracts where mean Walkability was low (<40; Figure 4a) indicating “car dependence” (Walk Score, 2021).

Our findings highlight the importance of covariates (also known as “effects modifiers”) for understanding and applying metrics. Figure 6 illustrates the relationships among a class 1 HWB metric, Poorhealth_p, a class 1 natural capital metric, Walkability, and the socioeconomic covariate Houseincome. Health was highest (the value of the negative metric Poorhealth_p was lowest) in census tracts with high income. Health was poorest in low income, unwalkable census tracts. In this case, there was also a significant interaction effect between the covariate and the predictor. The positive effect in the model of walkability on health was much higher in low-income census tracts than in more affluent tracts (compare the slope of Walkability versus Poorhealth_p at high and low values of Houseincome). This could, in part, reflect a bias wherein affluent residents are more likely to own cars and use them for more of their transportation needs than less affluent residents; walkability may therefore be less relevant to their well-being (Diez Roux and Mair, 2010). Failing to account for this interaction might undermine equitable waterfront design if the different pedestrian infrastructure needs of affluent and less affluent residents are ignored.

Figure 6.

Regression prediction for a HWB model including the interaction term Houseincome * Walkability. Plotted ranges correspond to actual data ranges across census tracts. All variables transformed. The curved contours on the Poorhealth_p response surface indicate an interaction between Houseincome and Walkability. See Appendix A, Part 8 for regression details.

Census-tract affluence, as indicated by the metric Houseincome, explained much and often most of the variation among census tracts in HWB indicators. Because it is a significant covariate of many natural capital metrics, including an affluence covariate in predictive models or otherwise stratifying analysis by income levels is essential for assessing project outcomes. This affluence effect may be due to reverse causation if affluent people chose to live in areas with more or better natural capital amenities (Diez Roux and Mair, 2010; Pun et al., 2018), and, because they are affluent, they may have a well-being advantage through better health care access, more education, healthier diet, more leisure time for exercise (in nature or otherwise), and other health-promoting factors.

Comparable research

Studies of the effects of residential exposure to natural capital in the form of green space, impervious surface, tree cover, and other amenities on human well-being at the waterfront are few. White et al. (2021) found an association between the mental health of residents and the amount of nearby green and blue space. Residents of greener and coastal neighborhoods reported a higher positive well-being, but the effect was confounded by the frequency of recreational visits to green spaces.

We found one study with a connection to the Great Lakes. Pearson et al. (2019) examined associations between anxiety/mood disorder hospitalizations and proximity to a Great Lake (“blue space”) at the ZIP- code scale for the state of Michigan. They found a small protective effect of proximity to a Great Lake on mental health. Confirmation of causation awaits examination of individual-level blue space exposures and mental health outcomes.

Tsai et al. (2020) used a tree cover metric and an impervious surface metric derived from the same dataset we used (EnviroAtlas; USEPA, 2020) to examine the effects of nature proximity on the self-reported general health (SRGH) of a large sample of U.S. women. They found a small beneficial effect of residential nature on SRGH. They point out that even a small positive effect of nature, if it applies to tens of millions of women, is an important public health benefit. These results corroborate our finding that the natural capital metrics percent tree cover and impervious surface are consistent correlates of HWB.

Larson et al. (2016) examined the effects of public parks on well-being in a study that was similar in approach to ours but different in scale. They related city-level indices of HWB to existing data on park attributes and sociodemographic factors including income and population density in 44 U.S. cities. Their best-fitting model (R²=0.58) for predicting overall well-being included percent of the city covered by parks, a score reflecting climate and geography in each city, percent unmarried people, recent population change, and income. They concluded, as did we, that even when variation in income is controlled, there is evidence that green space, parks in their case, areas with tree cover (and without impervious surface) in our case, are positively related to human well-being and urban quality of life.

Using publicly available city-level data, Gallagher et al. (2013) examined associations among environmental, health and sociodemographic metrics for 50 large U.S. cities. Most notably, they found that access to parks and recreational/interactional opportunities was significantly correlated with educational attainment (r=0.39), household income (r=0.51), and percent living in poverty (r=−0.55), which corroborates our results.

We showed that combinations of natural capital predictors and socioeconomic covariates could explain much of the variation in HWB at the census-tract scale that would result from waterfront revitalization. Summers et al. (2016) used county-scale data to forecast how community decisions, including investment in natural and built capital, would impact community wellbeing across domains (i.e., Smith et al., 2013). Their findings and ours support the proposition that the relationships between capital and HWB can be used to forecast outcomes of community decisions including unintended and undesirable consequences.

In a study tangentially related to ours, but interesting nonetheless, Roe et al. (2019) conducted an experiment in which they increased the comfort amenities (shade, seating) and added a visual element to increase the “inherent fascination” of a waterfront promenade. They measured subjective mood and other HWB indicators before and after exposure to the intervention and a control. They reported small but significant positive effects on HWB from exposure to the intervention. Experimental studies of the effects of biophysical changes at the waterfront on HWB are rare. Targeted, small scale experiments like Roe et al. (2019) would be useful for comparing HWB benefits among project design alternatives.

Limitations of our study

We explored potential usefulness of natural capital and HWB metrics in the waterfront revitalization context. We were not interrogating causal environmental drivers of variation in human well-being, which requires a different approach, usually an experiment or a longitudinal study, in which a single population cohort or community is tracked through time using HWB data for individuals (Kondo et al., 2018). Our approach does have some advantages however: it is less expensive (Siegel and Stenson, 1999), less time consuming, does not require follow-on sampling, and does not require the oversight required when data for individual subjects are collected. Many empirical studies, both cross-sectional and longitudinal, include a single HWB indictor such as mental health, and a single indicator of nature, often greenness based on the normalized difference vegetation index (NDVI; Rugel et al., 2017). In our study we examined associations between 19 objective HWB metrics from multiple domains (health, living standards, education, etc.) and 26 measures of natural capital (e.g., tree cover, green space, and partly built amenities). Of these, about 30% were not useful, suggesting that relying on a small number of HWB or natural capital metrics for assessing revitalization may miss significant associations.

Our study has shortcomings related to scale. We did not delineate the waterfront as a polygon separate from where people live within census tracts which would allow isolation of the natural capital amenities and benefits specific to the waterfront. Rather, we compared amenities and benefits at the only practical scale, the census tract. We could not otherwise unambiguously link waterfront attributes to residential areas. We included, where possible, natural capital metrics characteristic of the waterfront such as green and blue space attributes, shoreline type, and water access (Table 1b). Restricting our analysis to waterfront census tracts was consistent with our objective of evaluating metrics relevant to waterfront revitalization which in the Great Lakes can be catalyzed by sediment remediation and habitat restoration in Areas of Concern at or near the aquatic-terrestrial interface (Angradi et al., 2019). Most of our natural capital indicators are not specific to the waterfront, however, and our findings may be generalizable to non-waterfront census tracts in Great Lakes urban communities.

The “uncertain geographic context” problem (Kwan, 2012, 2018; also see Matthews, 2016) is relevant for our study. We interpret this as the uncertainty over whether the census-tract scale at which objective HWB data are nationally available is the best scale for examining associations between natural capital and HWB at the waterfront (Diez-Roux and Mair, 2010). For example, the well-being of census tract residents may be more associated with attributes of natural capital at a smaller (residents’ yard or street) or larger spatial scale than the census tract (such as a scale that includes residents place of work or school). Also, residents of highly urbanized, densely populated (and therefore relatively small) census tracts may be more likely to derive most HWB benefits from a census tract within which they do not reside than are residents of rural or residential census tracts. There is no remedy for this uncertainty in our cross-sectional approach. We did not have finer-scale HWB “response” data, nor did we have the behavioral data to assess where people actually benefit from natural capital. An assumption of our design was that natural capital in the census tract is a reasonable proxy for the exposure to nature of census tract residents and at least partly accounts for the variation in well-being attributable to residents cumulative or total exposure to nature.

We did not examine location effects (e.g., Green Bay vs. Cleveland) so we could not assess variation among census tracts attributable to variation among communities not accounted for by household income or population density. There is evidence from the Great Lakes region of variation among community types in how HWB elements are prioritized. Fulford et al. (2015) showed this through classification of communities based on socioeconomic factors and the availability of and dependence on local natural capital. Where data are available at the proper scale, community classification can be used to refine the sociodemographic context of assessments beyond population density, race, or affluence.

The reliability of metrics we examined varied among data sources. As noted previously, brownfield and Superfund sites were often absent from census tracts. There was, on average, about one property per waterfront tract for brownfields and one site for about every three tracts for Superfund (Appendix A, Part 5). We suspect that the census tract is not the best scale for studying relationships between brownfields and HWB. Brownfields are an important consideration for revitalization, however. In some highly urbanized areas, brownfields are among the only spaces where new green amenities can be created or allowed to develop passively through vegetative succession (Mathey et al., 2015; Mathey and Rink, 2020).

We did not find the metrics derived from crowdsourced OpenStreetMap (OSM) data to be strongly related to HWB. The exception was the density of recreational amenities (Hybridrec_d), which was significantly rank correlated with several HWB metrics and was a weak predictor of HWB in several regression models. The other OSM metrics (Streams_d, Wateraccess_d, Greenarea_p, and Trails_d) underperformed. Several OSM metrics had many zero values or had much higher mean metric values for Chicago and Milwaukee than the other communities in our study. For example, the density of water access amenities (Wateraccess_d), including boat ramps, fishing piers, and beaches was 50 times higher in Milwaukee waterfront census tracts than tracts in Cleveland, Green Bay, or Duluth-Superior. This suggests to us an uneven distribution of volunteer mapping effort among communities, for which we had no remedy. Researchers should not abandon this potential data source for HWB research however, especially at intra-urban or local scales where effort is likely to be less variable.

Among shoreline type metrics, only Artificialshore_p was strongly correlated with any HWB metrics or explained any error in regression models. We are not satisfied that our shoreline metrics represented the range of attributes of shoreline natural capital including accessibility, safety, aesthetic quality, naturalness, riparian, and aquatic habitat – all of which are potentially in play for waterfront interventions associated with revitalization (e.g., Dyson and Yocom, 2015; Cordell et al., 2017; Toft, 2017). Additional shoreline metrics could doubtless be extracted and synthesized from available digital data.

Applying these findings

Our findings constitute empirical support for the use of existing census tract HWB and natural capital indicators in community decision-making. This approach can be used for assessing, at the census-tract scale, the change in HWB resulting from revitalization or related efforts (e.g., urban greening, habitat restoration) that result in biophysical change at the waterfront. Where there is evidence of robust local associations between natural capital metrics and HWB outcomes, measures of natural capital can used as proxy indicators of HWB benefits. This is useful because up to date, high resolution, fine scale natural capital indicator data (land cover, park attributes, walkability) are likely easier and less expensive to obtain before and after revitalization than like-scaled HWB data. Specific findings can also be used to inform local revitalization design. For example, our findings that tree cover and impervious surface had stronger associations with HWB than generic green space, that the positive effect of percent tree cover on HWB diminished above 30%, and that walkability had a stronger positive effect on HWB in more urban and less affluent census tracts, are all relevant to project planning and design.

Cole et al. (2019, 2021) have shown that the benefits of urban greenspace on residents’ health and well-being can be influenced by both blue and green gentrification and white privilege. Relatively little research has examined social or environmental justice issues around revitalization planning or implemnentation (Avni and Fischler, 2020). Several of our sources include data for multiple time periods, including the decennial U.S. Census, the quinquennial American Community Survey, and the regularly updated National Land Cover Database (MLRC, 2021). We encourage researchers to use sociodemographic and natural capital data from these and similar sources to create indicators to detect gentrification, marginalization, or displacement of longtime residents (Freeman, 2005; Gould and Lewis, 2016; Richardson et al., 2019). Such indicators can be used by communities, local governments, grantees, or other entities to confirm positive HWB benefits and to detect unintended social consequences of waterfront revitalization.

Conclusion

We answered three of our four research questions. As for our fourth question: “can a cross-sectional analysis of available data support the hypothesis that attributes of natural capital amenities at the waterfront influence the objective well-being of waterfront residents?” our findings are highly suggestive but are observational only, are scale specific, and confounded by at least affluence and population density. Needed are composite studies, including studies with longitudinal designs, that include objective (e.g., attributes of natural capital) and subjective (e.g., individuals’ perceptions of accessible nature) indictors along with detailed socioeconomic data including indicators of social change.

Although there has been much research on the specific causal mechanisms and pathways linking natural capital, especially urban natural capital, to human well-being, much remains unknown or unclear (Kuo, 2015; Frumkin et al., 2017). The complexities of nature-health pathways challenge even the best datasets and most sophisticated models (Diez Roux and Mair, 2010). We are optimistic that this international program of research will succeed, and these causal pathways can be translated into refined operational design guidance and indictors for public health interventions (e.g., WHO, 2017).

Our research goal was more immediate: identify and assess indicators of waterfront natural capital and HWB that can be extracted from public databases and used now. Among the highest rated HWB indicators from our analyses were measures of health, income, home value, life success, life expectancy, and educational attainment. Highest rated natural capital metrics included tree cover and impervious surface metrics, walkability, density of recreational amenities, and shoreline type.

In the Great Lakes region, waterfront revitalizations and similar projects are probably being implemented faster than human well-being outcomes can be thoroughly assessed. Revitalization is a complex endeavor that often involves community engagement, political action, land acquisition, public and private investment, permitting, as well as engineering and construction challenges. Selecting and measuring indicators of project outcomes must not be an overlooked aspect of revitalization, however. Social and biophysical research on the selection, validation, and application of reliable and inexpensive, if imperfect, indictors is essential (Wellman et al., 2014; Ekkel and De Vries, 2017). If nothing is measured, nothing can be learned about how and how much communities are benefiting from revitalization investments.

About 80% of the residents of the eight U.S. Great Lakes states live in urban communities (USCB, 2021b) where waterfront space is valued at a premium (Kokot, 2008; Samant and Brears, 2017; Avni and Teschner, 2019). Nevertheless, the modern public waterfront should sustainably and equitably benefit all its residents (and visitors), affluent and otherwise (Avni and Fischler, 2020; Taufen and Yocom, 2021). Reliable, evidence-based indicators of different types are needed to inform the design and assess the revitalization of these complex multiuse, multispecies, and multigenerational spaces.