Types and spatial contexts of neighborhood greenery matter in associations with weight status in women across 28 U.S. communities

Abstract

Excess body weight is a risk factor for many chronic diseases. Studies have identified neighborhood greenery as supportive of healthy weight. However, few have considered plausible effect pathways for ecosystem services (e.g., heat mitigation, landscape aesthetics, and venues for physical activities) or potential variations by climate. This study examined associations between weight status and neighborhood greenery that capture ecosystem services most relevant to weight status across 28 U.S. communities.

Weight status was defined by body mass index (BMI) reported for 6,591 women from the U.S. Sister Study cohort. Measures of greenery within street and circular areas at 500m and 2000m buffer distances from homes were derived for each participant using one-meter land cover data. Street area is defined as 25m-wide zone on both sides of street centerlines multiplied by the buffer distances, and circular area is the area of the circle centered on a home within each of the buffer distances. Measures of street greenery characterized the pedestrian environment to capture physically and visually accessible greenery for shade and aesthetics. Circular greenery was generated for comparison. Greenery types of tree and herbaceous cover were quantified separately, and a combined measure of tree and herbaceous cover (i.e., aggregate greenery) was also included. Mixed models accounting for the clustering at the community level were applied to evaluate the associations between neighborhood greenery and the odds of being overweight or obese (BMI > 25) with adjustment for covariates selected using gradient boosted regression trees. Analyses were stratified by climate zone (arid, continental, and temperate).

Tree cover was consistently associated with decreased odds of being overweight or obese. For example, the adjusted odds ratio [AOR] was 0.92, 95% Confidence Interval [CI]: 0.88 – 0.96, given a 10% increase in street tree cover at the 2000m buffer across the 28 U.S. communities. These associations held across climate zones, with the lowest AOR in the arid climate (AOR: 0.74, 95% CI: 0.54 – 1.01). In contrast, associations with herbaceous cover varied by climate zone. For the arid climate, a 10% increase in street herbaceous cover at the 2000m buffer was associated with lower odds of being overweight or obese (AOR: 0.75, 95% CI: 0.55 – 1.03), whereas the association was reversed for the temperate climate, the odds increased (AOR: 1.19, 95% CI: 1.05 – 1.35).

Associations between greenery and overweight/obesity varied by type and spatial context of greenery, and climate. Our findings add to a growing body of evidence that greenery design in urban planning can support public health. These findings also justify to further define the mechanism that underlies the observed associations.

1. Introduction

Excess body weight is a risk factor for leading causes of global mortality such as cardiovascular disease, hypertension, type 2 diabetes, and cancers, (Knight, 2011; Ritchie and Roser, 2020; WHO, 2018). Nearly 39% of adults worldwide are overweight and 13% are obese (WHO, 2018). The global economic impact of obesity was estimated at 2 trillion USD in 2014, equivalent to 2.8% of the global gross domestic product (Dobbs et al., 2014). In the United States (U.S.), more than two thirds of adults are overweight or obese (Flegal et al., 2016; Fryar et al., 2020).

The causes of excess body weight are complex but the environment is a known determinant (Mackenbach et al., 2014). Nature is often considered to provide salutogenic effects (Silva et al., 2018; van den Bosch and Ode Sang, 2017), which are largely bestowed by ecosystem services. The benefits human beings obtain from ecosystems derive from supporting, regulating, provisioning, and cultural services(Millennium Ecosystem Assessment, 2005). Ecosystem benefits to human health are delivered through multiple pathways (Bratman et al., 2019; Dzhambov et al., 2020; Markevych et al., 2017). Those that support healthy weight by promoting and/or facilitating physical activity include shade for heat mitigation, air purification for pollutant filtration, landscape views for aesthetics, and venues for activities, social interaction, and engagement with nature (Millennium Ecosystem Assessment, 2005).

Though significant findings on the relationships between neighborhood greenery and weight status are not always observed (e.g., Hobbs et al., 2019; Jimenez et al., 2020; Slater et al., 2019), several longitudinal studies support a relationship between healthy body weight and greenery in the residential environment. An eight-year study in Finland (Halonen et al., 2014) reported that subjects living more than 750m from a green space (e.g., park, sport field, nature conservation area) throughout the study period had 50% greater odds of being overweight (odds ratio [OR]: 1.50, 95% confidence interval [CI]: 1.07, 2.11) than those living within 250m of a green space. Study subjects who relocated further than 250m from a green space also had about 50% greater odds of being obese (OR: 1.49, 95% CI: 1.08, 2.06). Feng and Astell-Burt (2019) studied postpartum weight tracked biennially for 15 years and the quantity and perceived quality of public green space within Australian census units. They found that body mass index (BMI) was at least 0.43 kg/m² lower for those who lived in areas with ≥ 6% green space than those with ≤ 5% green space. Furthermore, the greatest BMI reduction (−0.86 kg/m²) was observed for residents living in areas with 21 – 40% green space. A nine-year study in Sweden (Persson et al., 2018) reported an incidence rate ratio of 0.88 (95% CI: 0.79, 0.99) for obesity in women per interquartile range (IQR) increase in average greenness (measured by Normalized Difference Vegetation Index, NDVI; IQR was 0.07) within 500m from residences. However, no significant association was found for men.

While this longitudinal evidence supports a protective role for various types of greenery in healthy body weight, studies define greenery using disparate and often coarse methods. For example, NDVI is an aggregate measure that combines all types of greenery; its use obscures the individual effect pathways of beneficial ecosystem services. Some ecosystem services are produced by specific types of greenery (Livesley et al., 2016). For instance, trees can provide shade for thermal comfort, herbaceous cover (e.g., grass lawn) can provide venues for activities, and both can provide aesthetic values (Bowler et al., 2010; Säumel et al., 2016).

The spatial context of neighborhood greenery is also relevant, although not widely studied. A key factor for promoting human health through greenery is accessibility (Ekkel and de Vries, 2017; Grunewald et al., 2017). Road networks are an essential feature to promote active lifestyles in built environments (Frank et al., 2006; Rundle et al., 2016; Saelens and Handy, 2008; Sundquist et al., 2011). Greenery along this network can further enhance accessibility to destinations, as well as use of the route itself. However, many studies, including those on weight status and physical activity, commonly characterize neighborhood greenery within circular buffers only. These measures do not distinguish private from public land, whereas measures based on the neighborhood road network (i.e., network buffer) target an important class of physically and visually accessible greenery. Ecosystem services provided along transportation routes may demonstrate stronger beneficial relationships to healthy weight than those provided across the larger landscape (Säumel et al., 2016).

Everyday engagement with the natural environment may be determined largely by human values (Kaplan and Kaplan, 1989; Ward Thompson, 2013). With a focus on adults, an experimental study conducted by Jiang et al. (2015) found that street scenes with higher tree density were preferred by people in four U.S. midwestern cities (Champaign-Urbana, St. Louis, Indianapolis, and Springfield); however, no preference was found for herbaceous cover. Other observational studies also found varying associations between greenery types and health outcomes among adults. Tsai et al. (2019b) distinguished trees from herbaceous cover in examining their relationships with active transportation (e.g., walking and biking), and further explored spatial contexts (i.e., sidewalk, street, and circular buffer areas) of greenery in two U.S. midwestern cities (Green Bay and Milwaukee, WI). At the largest neighborhood extent (1250m from home) in the study, only tree cover within sidewalk and street buffers, not circular buffers, was associated with increased odds of participation in active transportation. Herbaceous cover was inversely associated with active transportation, and aggregate greenery (a combined measure of tree and herbaceous covers) showed no significant effect.

Climate plays a critical role in vegetation species and distribution (Stephenson, 1990) and may play a role in human behavior. Data on vegetation species are commonly not available and were not available for this study. It is possible that the varying associations between greenery types and health outcomes may also differ across climate zones. Tsai et al. (2019a) reported that more street tree cover was consistently associated with lower odds of being overweight or obese in two U.S. cities with very distinct climate conditions, hot and dry Phoenix, AZ (arid), and cool and rainy Portland, OR (temperate). Street herbaceous cover and healthy weight status were associated positively in Phoenix but inversely in Portland. In addition to the contrasting effects of herbaceous cover in these two climate zones, differences in cutoff values for associations with healthy weight were observed. Residents in the arid climate had lower odds of overweight/obesity if they had a minimum of 6% street tree cover. This cutoff values for street tree cover was 16% in the temperate climate.

Findings from the studies focused on greenery along transportation routes suggest that type and spatial context of greenery may affect healthy weight through multiple pathways of ecosystem services, and reveal that variations exist by climate. However, evidence across a large and diverse geographic scale is lacking.

Here, we address a knowledge gap concerning potential pathways for greenery to deliver ecosystem services that may be more relevant to healthy weight-related outcomes, and test previously-documented relationships between greenery and weight status across a large geographic scale. We assess the impact of type and spatial context of greenery on weight status among women in 28 communities across the U.S.

2. Materials and Methods

2.1. Study population and health outcome

We included participants in the Sister Study, a prospective cohort of 50,884 U.S. women aged 35 to 74 years old without a history of breast cancer but with at least one sister diagnosed with breast cancer at enrollment during 2003 - 2009 (Sandler et al., 2017). Data collection included biological, sociodemographic, lifestyle, and health-related factors. Participants have been contacted every two to three years for detailed follow-up since 2008. Data used in this analysis were from the Sister Study Data Release 6.0.

To match the timeframe of the exposure variables (i.e., greenery datasets described in the Section 2.2), we estimated BMI using self-reported height and weight information from the questionnaires collected during 2012 – 2014. Participants’ residential locations were geocoded based on the addresses they provided at enrollment. We used high-resolution land cover data for communities from the U.S. Environmental Protection Agency’s EnviroAtlas project (EnviroAtlas hereafter, www.epa.gov/enviroatlas) (Figure 1). At the time of this analysis, there were 30 EnviroAtlas featured communities, covering 1,450 cities and towns in the U.S. (Pilant et al., 2020). A total of 11,059 study participants resided within the EnviroAtlas community boundaries. We excluded participants (2,206) who reported having moved since enrollment, those with missing values for BMI (1,164) and those with any missing covariates considered in the analysis (1,098). The final analysis included complete cases, 6,591 participants residing within 28 EnviroAtlas communities (no eligible participants resided in Woodbine, IA or Paterson, NJ). Since the proportion of participants that were underweight (<18.5 kg/m²) was very small (1.2 %), we dichotomized BMI values as “Neither overweight nor obese” (BMI < 25 kg/m²; the referent; normal weight hereafter) and “Overweight or obese” (BMI >= 25 kg/m²) (Centers for Disease Control Prevention, 2018).

Figure 1.

Study area – EnviroAtlas communities.

We performed several sensitivity analyses. First, we tested relationships with greenery using “not obese” (BMI <30; referent) versus “obese” (BMI >= 30) as several other studies have done (e.g., Dempsey et al., 2018; Ghimire et al., 2017; Villeneuve et al., 2018). Second, we examined the relationship between greenery and overweight/obesity based on the datasets using multiple imputation for missing covariates. Lastly, we restricted analysis to the 86.0% of complete cases living in urban areas (5,669) and to the 85.9% of complete cases who were non-Hispanic white (5,663).

2.2. Greenery measures

Greenery was quantified using the Meter-Scale Urban Land Cover (MULC, 1meter x 1meter pixels) data developed by the EnviroAtlas project and partners, and processed by EnviroAtlas for consistency (Pilant et al., 2020). MULC data for each community were developed at different times (from years 2010 to 2016) using the best available data. These years bracket the date of the Sister Study survey and land cover does not usually change rapidly across communities. Urban studies have reported that less than 2% of the land cover changed in Baltimore, MD from 1999 to 2004 (Zhou et al., 2008) and overall tree cover loss was 0.6% in Syracuse, NY from 1994 to 2009 (Nowak et al., 2016). A study (Nowak and Greenfield, 2012) across 20 U.S. cities reported an average decrease in tree cover of 1.5% over a period of 3 to 5 years.

Most participants (62.6%) resided in communities with MULC data coverage within the time period when the health outcome (BMI) was collected, and the rest of the participants resided in communities with MULC data ± 2 years from this period. The base layer for MULC data was derived from the USDA National Agriculture Imagery Program (NAIP) aerial photography, supplemented with Light Detection and Ranging (LiDAR) data. NAIP data were leaf-on and usually collected in summer months. Additional data used in MULC classification included agriculture, wetlands, other hydrology, building footprints, and roads. The MULC classes are trees and forest (trees hereafter), grass and herbaceous cover (herbaceous hereafter), shrubs, woody wetland, emergent wetland, impervious surfaces, water, soil and barren lands, agriculture, and orchards. For this study, we considered three types of greenery: tree cover, herbaceous cover, and aggregate greenery. ”Tree cover” was defined as the total percentage of trees and woody wetland. “Herbaceous cover” was defined as the total percentage of herbaceous, shrubs, and emergent wetland. “Aggregate greenery” was defined as the total percentage of “tree cover” and “herbaceous cover”.

Greenery was evaluated using two spatial context measures. The first was based on the road network to capture the public pedestrian environment using the methodology described in Tsai et al. (2019b). The public pedestrian environment (street area hereafter) refers to areas that are physically and visually accessible. These were characterized using data for road segments with speed limits of less than 55 miles per hour and the attribute of pedestrian access (walkable roads hereafter) based on the NAVTEQ Streets database (Nokia-HERE, 2011). Street area was defined as a 25m-wide zone around the road centerline, which resulted in a total area of 50m multiplied by the length of the roads. Street buffer distance were examined by the widths of right-of-way and walkable roads. Since these widths vary, a buffer distance of 25m on the both sides of street centerlines are likely to capture road medians, right-of-way, and people’s front yards without capturing back yards (Tsai et al, 2019b). Measures of greenery within this area were intended to capture the ecosystem provision of shade and aesthetics along physically- and visually- accessible areas.

The second context measure of greenery was evaluated as the overall environment (circular area hereafter). This metric was defined as total area within a Euclidean distance from residences. Circular areas were calculated as πr². Measures of greenery within these areas reflected broader-scale ecosystem services such as ambient air filtration and habitat provision.

Two buffer sizes, 500m and 2000m, were applied to the network-based street area and the Euclidean-based circular area for each participant. These two buffer sizes were selected to represent close and distant greenery following the suggestion from Browning and Lee (2017) to use buffer sizes of up to 2000m for exploring relationships between greenery and fitness-related outcomes.

In addition to assessing greenery through continuous measures, we also applied tertiles to evaluate the associations between discrete levels of greenery and weight status. A total of twelve continuous greenery measures (labelled as Continuous in Figure 2 for street area and Supplemental Figure 1 for circular area) and another twelve categorical greenery measures (labelled as Lower, Middle, Upper Tertiles in Figure 2 for street area and Supplemental Figure 1 for circular area) were calculated for each participant.

Figure 2.

Adjusted odds ratio (AOR) of being overweight or obese given a 10% increase in continuous street greenery (labelled as Continuous, 10-unit increase) and in different levels of categorical street greenery (labelled as Lower Tertile [Ref. stands for referent], Middle Tertile, and Upper Tertile) determined by tertiles for the full study population. X-axis is on log-scale. Models controlled for individual-level age, income, race/ethnicity, education, marital status, alcohol consumption, use of depression medication, urbanicity, survey season, park proximity and intersection density.

2.3. Covariates

Covariates used in this analysis included individual-level socioeconomic and demographic information, lifestyle factors, health conditions, built environment characteristics, survey year, and survey season. These covariates derived from the same survey period as the health outcome, unless otherwise stated (Table 2). Socioeconomic and demographic variables were age (categorical, <55 as referent vs. 55 – 64, ≥65), household income (categorical: <20k as referent vs. 20 – 50k, 50 – 100k, 100 – 200k, and >200k), employment status (categorical: unable to work [i.e., medical leave, disability] as referent vs. unemployment, working, and unpaid/other occupation [i.e., retiree, homemaker, and student]), marital status (categorical: never as referent vs. married, divorced/separated, and widow), and three covariates measured at baseline - educational attainment (categorical: high school or less as referent vs. associate college or some college, bachelor degree, and graduate degree), race/ethnicity (binary: non-Hispanic White as referent vs. other races/ethnicities), and self-reported urbanicity (binary: rural [rural or small town] as referent vs. urban [suburban or urban]).

Table 2.

Characteristics of street greenery measures by weight status for the full population and in the arid, continental, and temperate climates.

	Overweight or obese
	No	Yes	Total
Full Population (N = 6,591)	Mean (sd)	Mean (sd)	Mean (sd)
500m
Street Tree	32.48 (17.34)	29.43 (16.03)	30.69 (16.65)
Street Herbaceous	21.14 (11.88)	21.34 (11.04)	21.26 (11.39)
Street Aggregate Greenery	53.63 (19.17)	50.77 (18.49)	51.95 (18.83)
2000m
Street Tree	30.99 (14.31)	28.52 (13.28)	29.54 (13.77)
Street Herbaceous	21.49 (10.42)	21.84 (9.87)	21.70 (10.10)
Street Aggregate Greenery	52.48 (17.33)	50.36 (16.50)	51.24 (16.88)
Arid Climate (N = 686)
500m
Street Tree	13.25 (8.01)	11.96 (7.12)	12.49 (7.51)
Street Herbaceous	10.14 (6.12)	9.36 (6.22)	9.67 (6.19)
Street Aggregate Greenery	23.39 (12.41)	21.32 (11.45)	22.16 (11.88)
2000m
Street Tree	12.33 (6.25)	11.31 (5.51)	11.73 (5.84)
Street Herbaceous	10.10 (5.78)	9.50 (5.86)	9.75 (5.83)
Street Aggregate Greenery	22.44 (10.65)	20.81 (9.78)	21.47 (10.17)
Continental Climate (N = 3,925)
500m
Street Tree	33.21 (16.04)	30.81 (14.86)	31.81 (15.41)
Street Herbaceous	24.38 (12.26)	24.37 (10.83)	24.37 (11.45)
Street Aggregate Greenery	57.58 (16.63)	55.18 (15.72)	56.18 (16.15)
2000m
Street Tree	32.02 (12.88)	29.94 (11.84)	30.80 (12.33)
Street Herbaceous	24.82 (10.31)	25.02 (9.15)	24.94 (9.65)
Street Aggregate Greenery	56.84 (14.55)	54.96 (13.25)	55.74 (13.83)
Temperate Climate (N=1,980)
500m
Street Tree	37.67 (17.76)	32.82 (16.67)	34.80 (17.28)
Street Herbaceous	18.38 (9.40)	19.58 (9.41)	19.09 (9.42)
Street Aggregate Greenery	56.05 (16.53)	52.40 (16.09)	53.89 (16.36)
2000m
Street Tree	35.34 (14.14)	31.74 (13.44)	33.21 (13.84)
Street Herbaceous	18.69 (8.02)	19.92 (8.36)	19.42 (8.24)
Street Aggregate Greenery	54.03 (13.85)	51.66 (13.29)	52.63 (13.56)

Lifestyle factors were alcohol consumption (categorical: never/former as referent vs. current social, and current regular), tobacco smoking (categorical: never as referent vs. current and former), and dog ownership (binary: no as referent vs. yes). Health conditions included individual-level menopausal status (binary: pre-menopause as referent vs. post-menopause), use of depression medication (binary: no as referent vs. yes), and use of anxiety medication (binary: no as referent vs. yes). Two other individual-level variables from the Sister Study considered as covariates were survey season (categorical: winter as referent vs. spring, summer, and fall) and survey year (categorical: 2012 as referent vs. 2013 and 2014).

Built environment characteristics included Estimated Intersection Density of Walkable Roads (intersection density hereafter) (EPA, 2016a) and Estimated Walking Distance to a Park Entrance (park proximity hereafter) (EPA, 2016b). Intersection density is a commonly-used measure of street connectivity which is associated with walkability and active living for fitness-related outcomes (Grasser et al, 2013). A threshold of 54 intersections per square kilometer is used to define whether a neighborhood is walkable, based on the U.S. Green Building Council’s guidelines for Leadership in Energy and Environmental Design for Neighborhood Development (LEED-ND) (Stangl and Guinn, 2011; Talen et al., 2013). Intersection density was derived at 10m resolution using the total number of intersections of walkable roads as the input layer for the Kernel Density function. This function fits a smoothly curved surface over each point where its apex is at the analysis centroid and values diminish along the analysis distance to reach to zero at the defined search radius (i.e., the buffer sizes used in this analysis). Intersection density based on residential location was also categorized into quartiles.

Although many studies have considered park proximity as a measure of greenery (e.g., Coombes et al., 2010; Dadvand et al., 2014; Halonen et al., 2014), it is possible that nearby parks represent a discrete pathway for physical activity and that their presence may correlate with street greenery. Therefore, we used this measure as a covariate. Park proximity refers to the distance along walkable roads between any point and the closest park entrance. Park boundaries and entrances were identified using multiple sources including PAD-US (Protected Areas Database of the United States, USGS, 2016), Google Maps, and information from county and municipal parks and recreation departments. Walking distance to the nearest park entrance was calculated at 10m resolution across each EnviroAtlas community using Cost Distance function and then interpolated for extents up to 5km beyond the community boundary using Natural Neighbor function. Park proximity was extracted for all study participants based on their residential locations and categorized into six levels: ≥2000m as referent vs. <250, 250 – 500, 500 – 750, 750 – 1000, and 1000 – 2000m. All geoprocessing was conducted using ArcGIS Desktop 10.4.

2.4. Statistical Analysis

We analyzed the associations between weight status (response) and multiple greenery metrics (predictors) across the 28 EnviroAtlas communities using a generalized linear mixed effect model with a random intercept to account for dependency between individuals within community (cluster). We applied gradient boosted regression trees (GBRT) to determine the final set of covariates for inclusion. The inclusion criterion for covariates was a relative influence value greater than one. Covariates selected for the final analysis included age, race/ethnicity, income, education, marital status, alcohol consumption, use of depression medication, urbanicity, survey season, park proximity, and intersection density. Effect modification was examined by incorporating multiplicative interaction terms for all the selected covariates and greenery measures and was observed by race/ethnicity and climate zone. We performed a sensitivity analysis for race/ethnicity (see Section 2.1) and stratified analysis by climate zone. Mixed-effect models were performed using SAS 9.4 Proc Glimmix (SAS Institute, Cary NC, USA); GBRT for covariate selection was performed using the dismo package in R 3.6.1.

Finally, primary associations were stratified by climate zone. Climate zones used in this analysis were based on the Köppen climate classification for the conterminous United States (Idaho State Climate Services, 1999), which divided the U.S. into arid, continental, temperate, and tropical zones. Since the tropical zone is a very small portion of the U.S. (the everglades and the keys of Florida, accounting for only 0.2% of the total area of the continuous U.S.), we reclassified the tropical zone as part of the temperate zone. Consequently, a total of three climate zones (arid, continental, and temperate) were used for climate-stratified analysis (Figure 1). The Köppen climate classification considers temperature and precipitation among many other factors. Arid climate is characterized primarily by low precipitation. The other two climate zones are characterized mainly by monthly average temperature – at least one month below −3 °C (26.6 °F) and at least one month above 10 °C (50 °F) for continental climate, and the coldest temperature within a range between −3 °C (26.6 °F) and 18 °C (64.4 °F) for temperate climate (Kottek et al., 2006). Since the amount of greenery may vary by climate zone, we categorized greenery into tertiles separately for each zone (labelled as Lower, Middle, and Upper Tertile in Figure 3 for Arid, Continental, and Temperate Climate) in addition to exploring the variations in the relationships between greenery and weight status by climate zone using continuous greenery measures. We expected that the cutoff values for each tertile would be lower for the arid climate where it is less vegetated.

Figure 3.

Adjusted odds ratios (AORs) of being overweight or obese given a 10% increase in continuous street greenery (labelled as Continuous, 10-unit increase) and in different levels of categorical street greenery (labelled as Lower Tertile [Ref. stands for referent], Middle Tertile, and Upper Tertile) determined by tertiles in the arid, continental, and temperate climates. X-axis is on log-scale. Models controlled for individual-level age, income, race/ethnicity, education, marital status, alcohol consumption, use of depression medication, urbanicity, survey season, park proximity and intersection density.

3. Results

3.1. Characteristics of study participants by weight status

Population characteristics by weight status are detailed in Table 1. Surveys were completed primarily during spring (30.1%) and fall (37.4%). Almost 60% of the participants were overweight (31.4%) or obese (27.3%). Mean age was 60.3 years old, with almost 75% of the participants above 55 years old. Most participants were non-Hispanic white (85.9%), had an annual household income greater than $50,000 (78.6%), had a bachelor’s or graduate degree (59.7%), and were married (69.6%). More than half of the participants were current regular drinkers (64.2%) and did not use depression medication (80.2%). Most of the study participants reported living in urban areas (86.0%) and in climate zones that are more forested (59.6% in the continental climate and 30.0% in the temperate climate). Only around 10.4% of participants lived in dry areas (i.e., arid climate zone).

Table 1.

Participants’ characteristics by weight status. Variables were obtained mostly from the same survey period (2012 - 2014 follow-up) as the health outcome from the Sister Study, unless otherwise stated.

		Overweight or Obese
Health Outcome		No (BMI < 25)	Yes (BMI >= 25)	Total Number
Total Number		2723 (41.31%)	3868 (58.69%)	6591
Body Mass Index (BMI)	Continuous, Mean (sd)	22.25 (1.80)	30.96 (5.38)	27.36 (6.06)
Age	Continuous, Mean (sd)	60.51 (9.01)	61.30 (8.65)	60.97 (8.81)
	< 55	741 (45.18%)	899 (54.82%)	1640 (24.88%)
	55 – 64	1059 (40.30%)	1569 (59.70%)	2628 (39.87%)
	>= 65	923 (39.73%)	1400 (60.27%)	2323 (35.25%)
Race/Ethnicity (Baseline)	Non-Hispanic white	2480 (43.79%)	3183 (56.21%)	5663 (85.92%)
	Other	243 (26.19%)	685 (73.81%)	928 (14.08%)
Income (unit: USD)	< 20,000	76 (32.62%)	157 (67.38%)	233 (3.54%)
	20,000-50,000	353 (29.92%)	827 (70.08%)	1180 (17.90%)
	50,000-100,000	879 (37.29%)	1478 (62.71%)	2357 (35.76%)
	100,000-200,000	981 (46.56%)	1126 (53.44%)	2107 (31.97%)
	>= 200,000	434 (60.78%)	280 (39.22%)	714 (10.83%)
Education (Baseline)	High School or less	249 (30.55%)	566 (69.45%)	815 (12.37%)
	Associate/College or so	621 (33.73%)	1220 (66.27%)	1841 (27.93%)
	Bachelor	927 (47.01%)	1045 (52.99%)	1972 (29.92%)
	Graduate	926 (47.17%)	1037 (52.83%)	1963 (29.78%)
Marital Status	Never	146 (30.48%)	333 (69.52%)	479 (7.27%)
	Married	2041 (44.47%)	2549 (55.53%)	4590 (69.64%)
	Divorced/Separated	340 (34.03%)	659 (65.97%)	999 (15.16%)
	Widowed	196 (37.48%)	327 (62.52%)	523 (7.94%)
Alcohol consumption	Never/Former	453 (34.40%)	864 (65.60%)	1317 (19.98%)
	Current Social	359 (34.39%)	685 (65.61%)	1044 (15.84%)
	Current Regular	1911 (45.18%)	2319 (54.82%)	4230 (64.18%)
Use of Depression Medication	No	2289 (43.29%)	2999 (56.71%)	5288 (80.23%)
	Yes	434 (33.31%)	869 (66.69%)	1303 (19.77%)
Urbanicity (baseline)	Rural	352 (38.18%)	570 (61.82%)	922 (13.99%)
	Urban	2371 (41.82%)	3298 (58.18%)	5669 (86.01%)
Survey Season	Spring	836 (42.16%)	1147 (57.84%)	1983 (30.09%)
	Summer	237 (38.41%)	380 (61.59%)	617 (9.36%)
	Fall	987 (40.06%)	1477 (59.94%)	2464 (37.38%)
	Winter	663 (43.42%)	864 (56.58%)	1527 (23.17%)
Park Proximity (unit: meters)	< 250	493 (42.32%)	672 (57.68%)	1165 (17.68%)
	250 – 499	548 (41.33%)	778 (58.67%)	1326 (20.12%)
	500 – 749	403 (38.42%)	646 (61.58%)	1049 (15.92%)
	750 – 999	257 (38.30%)	414 (61.70%)	671 (10.18%)
	1000 – 1999	583 (41.70%)	815 (58.30%)	1398 (21.21%)
	>= 2000	439 (44.70%)	543 (55.30%)	982 (14.90%)
Intersection Density at 500m (unit: intersections/km2)	Continuous, Mean (sd)	46.83 (29.51)	48.71 (26.99)	47.93 (28.07)
	0 - 29.36	738 (44.81%)	909 (55.19%)	1647 (24.99%)
	29.37 - 44.92	699 (42.42%)	949 (57.58%)	1648 (25.00%)
	44.93 - 61.11	632 (38.35%)	1016 (61.65%)	1648 (25.00%)
	61.12 - 338.21	654 (39.68%)	994 (60.32%)	1648 (25.00%)
Intersection Density at 2000m (unit: intersections/km2)	Continuous, Mean (sd)	39.49 (22.37)	40.39 (20.77)	40.01 (21.45)
	0 - 25.15	745 (45.23%)	902 (54.77%)	1647 (24.99%)
	25.16 - 38.19	664 (40.29%)	984 (59.71%)	1648 (25.00%)
	38.20 - 52.42	653 (39.62%)	995 (60.38%)	1648 (25.00%)
	52.43 - 198.71	661 (40.11%)	987 (59.89%)	1648 (25.00%)
Climate Zone	Arid	279 (40.67%)	407 (59.33%)	686 (10.41%)
	Continental	1637 (41.71%)	2288 (58.29%)	3925 (59.55%)
	Temperate	807 (40.76%)	1173 (59.24%)	1980 (30.04%)

Approximately 65% of the participants lived within 1000m walking distance to a park entrance, but less than half of the participants had intersection density considered beneficial for walkability (i.e., at least 54 intersections per square kilometer) within 500m.

3.2. Characteristics of greenery measures by weight status and climate zones

Mean values for 500m and 2000m neighborhood street greenery are given in Table 2, including by weight status for the full population and by climate zones. Circular areas generally contained more greenery than street areas and those results are provided in Supplemental Table 2.

For the full population, mean values of street tree cover were 31% for the 500m and 30% for the 2000m neighborhoods. Participants in the normal weight group had higher mean values of tree cover compared to those in the overweight/obese group for both 500m and 2000m neighborhoods. Mean values of street herbaceous cover were 21% for the 500m and 22% for the 2000m neighborhoods. In contrast to tree cover, participants in the normal weight group had less herbaceous cover than the overweight/obese group for both 500m and 2000m neighborhoods, though significant differences in mean values were not observed.

For the continental and temperate climate zones, we observed similar mean values of greenery and mean differences in greenery between normal weight and overweight/obesity groups as those observed for the full population. For the arid climate zone, mean values of tree and herbaceous cover were around 10% for both neighborhoods. Higher mean values of tree cover were observed for the normal weight group compared to those in the overweight/obese group. In contrast to the findings observed for the full population and other climate zones, participants in the normal weight group had higher mean values of herbaceous cover compared to those in the overweight/obese group in the arid climate, though the mean differences were not significant.

3.3. Relationships between greenery and weight status

For the full population (Figure 2), a 10% increase (labelled as Continuous, 10-unit increase) in street aggregate greenery was associated with reduced odds of overweight/obesity for both 500m (AOR: 0.93, 95% CI: 0.90 – 0.97) and 2000m (AOR: 0.94, 95% CI: 0.89 – 0.99) neighborhoods. The upper tertiles of street aggregate greenery were also associated with reduced odds of overweight/obesity for the 500m (AOR: 0.77, 95% CI: 0.66 – 0.91) and the 2000m (AOR: 0.78, 95% CI: 0.65 – 0.93) neighborhoods compared to the first tertile. Overall, similar but weaker associations were observed for circular greenery (See Supplemental Figure 1).

Results for aggregate greenery obscured the contrasting effects of tree and herbaceous cover when analyzed separately. A 10% increase in street tree cover was associated with reduced odds of overweight/obesity, with AORs of 0.94 [95% CI: 0.90 – 0.97] for the 500m and 0.92 [95% CI: 0.88 – 0.96] for the 2000m neighborhoods. The upper tertile of street tree cover was associated with reduced odds of overweight/obesity as well, for both 500m (AOR: 0.77, 95% CI: 0.67 – 0.89) and 2000m (AOR: 0.81, 95% CI: 0.70 – 0.94) neighborhoods.

In contrast, a 10% increase in street herbaceous cover was associated with increased odds of overweight/obesity, with AORs of 1.04 [95% CI: 0.98 – 1.10] for the 500m and 1.07 [95% CI: 1.00 – 1.14] for the 2000m neighborhoods. The upper tertile of street herbaceous cover was also associated with increased odds of overweight/obesity, with AORs of 1.18 [95% CI: 1.01 – 1.36] for the 500m and 1.22 [95% CI: 1.05 – 1.42] for the 2000m neighborhoods compared to the first tertile.

Results were similar for all sensitivity analyses (Supplemental Table 1) compared to those from the main study.

3.4. Stratified analysis

Effect modification was observed by climate zone. Associations between aggregate greenery and weight status generally showed weaker associations than those between trees or herbaceous cover alone, so aggregate greenery is not discussed here (results are provided in Supplemental Figure 2).

For the arid climate (Figure 3), a 10% increase in street herbaceous cover had marginally significant associations with reduced odds of overweight/obesity for the 500m (AOR: 0.78, 95% CI: 0.58 – 1.04) and 2000m (AOR: 0.75, 95% CI: 0.55 – 1.03) neighborhoods. Only the upper tertile of street herbaceous cover for the 500m neighborhood was associated with reduced odds (AOR: 0.64, 95% CI: 0.41 – 0.99) of overweight/obesity.

For the continental climate, a 10% increase in street tree cover for the 2000m neighborhood was associated with reduced odds (AOR: 0.93, 95% CI: 0.88 – 0.99]) of overweight/obesity. The upper tertile of street tree cover for the 500m neighborhood was associated with reduced odds of overweight/obesity, with AOR of 0.82 [95% CI: 0.69 – 0.98]. Herbaceous cover generally showed positive associations with odds of overweight/obesity, though significant associations were not observed.

For the temperate climate, a 10% increase in street tree cover for the 500m and 2000m neighborhoods was associated with reduced odds of overweight/obesity (AORs: 0.90, 95% CI: 0.85 – 0.96 and 0.89, 95% CI: 0.82 – 0.96, respectively). Similarly, the upper tertiles of street tree cover for the 500m (AOR: 0.69, 95% CI: 0.54 – 0.89) and 2000m (AOR: 0.67, 95% CI: 0.52 – 0.87) neighborhoods were associated with reduced odds of overweight/obesity. In contrast to tree cover, a 10% increase in street herbaceous cover was associated with increased odds of overweight/obesity, with AORs of 1.15 [95% CI: 1.04 – 1.28] for the 500m and 1.19 [95% CI: 1.05 – 1.35] for the 2000m neighborhoods. The upper tertile of street herbaceous cover was also associated with increased odds of overweight/obesity for the 500m (AOR: 1.42, 95% CI: 1.12 – 1.81) and for the 2000m (AOR: 1.47, 95% CI: 1.15 – 1.88) neighborhoods.

4. Discussion

There is growing evidence that greenery provides salutogenic influences on human health and wellbeing (Dadvand and Nieuwenhuijsen, 2019; Markevych et al., 2017; van den Bosch and Ode Sang, 2017). However, ecological processes and ecosystem services for human health and wellbeing differ by type and spatial context of greenery (Alberti, 2010; Lamy et al., 2016). Few studies have considered plausible effect pathways between human health and ecosystem services through types and spatial contexts of greenery. Moreover, vegetation morphology and phenology are greatly influenced by climate. The potential varying effects of greenery by climate on human health are under-explored. Our analysis addressed these research gaps by examining various neighborhood metrics of greenery including type (trees, herbaceous cover), spatial context (street and circular areas), and extent (500m and 2000m) across 28 U.S. communities using a national cohort of women. The broad geographic distribution of the study population allowed us to identify potentially different effects of greenery types across climate zones (arid, continental, and temperate climates).

4.1. Greenery types and weight status across the 28 U.S. communities and in different climates

Our findings for aggregate greenery are consistent with previous literature (Feng and Astell-Burt, 2019; Halonen et al., 2014; Persson et al., 2018). Additionally, our analyses across the full study population reveal that tree cover drives the beneficial ecosystem services when it comes to weight status. Street and circular tree cover were both inversely associated with odds of being overweight or obese at the 500m and 2000m neighborhoods across the 28 U.S. communities. This association was seen across climate zones although magnitude of the point estimates and their statistical significance (likely driven by sample size) varied by climate.

In contrast, street and circular herbaceous cover were both positively associated with the odds of being overweight across the 28 U.S. communities. However, this association differed by climate. Herbaceous cover had inverse associations with odds of being overweight or obese in the arid climate but had positive associations in the continental and temperate climate zones. It is possible that trees are rare in an arid climate, herbaceous cover may provide higher relative values as a natural resource compared to herbaceous cover in a more forested region, such as temperate and continental climate. This finding is consistent with a previous study (Tsai et al., 2019a) that reported inverse associations between herbaceous cover and the odds of overweight/obesity in the arid climate but positive associations in the temperate climate. That study also found that street trees had consistently inverse associations with overweight/obesity across climate zones. Furthermore, Tsai et al. (2019a) also found that street aggregate greenery was negatively associated with being overweight or obese, and that these associations were weaker than those for street tree cover alone.

Studies of neighborhood greenery and active transportation may parallel findings on weight status and thereby may offer a partial explanation. For instance, an increase in active transportation led to a reduction in overweight/obesity (Bassett et al., 2008). A study (Tsai et al., 2019b) conducted in the continental climate reported an inverse association between participation in active transportation and street herbaceous cover but a positive association with street tree cover.

These related studies support that trees and herbaceous cover have different associations with health outcomes, and our findings further reveal that variations in associations can differ by climate across a large geographic scale.

4.2. Spatial contexts of greenery and weight status across the 28 U.S. communities and in different climates

Although tree cover within both street and circular areas was associated with lower odds of being overweight or obese, these associations were stronger for street than for circular tree cover, especially at the 2000m neighborhood, across the 28 U.S. communities and in the continental and temperate climates. The varying associations of herbaceous cover also had different magnitudes in street versus circular areas in different climates. Stronger and positive associations with overweight/obesity were found for circular herbaceous cover in the temperate climate, whereas stronger and negative associations were observed for street herbaceous cover in the arid climate.

Our findings on the magnitudes of greenery associations with BMI within street and focal areas are supported by previous research (Tsai et al., 2019b). Only street, not circular, tree cover was associated with increased odds of participating in active transportation at buffer sizes greater than 750 meters, whereas circular herbaceous cover had stronger negative associations with the odds of participating in active transportation compared to street herbaceous cover. Since street areas were intentionally designed to capture ecosystem provision of shade from trees and landscape aesthetics from tree and herbaceous cover along pedestrian routes, results from our spatial context measures suggest that these physically- and visually-accessible ecosystem services may encourage people to engage in activities over a greater distance from home and consequently facilitate better weight status.

4.3. Categorical greenery and weight status across the 28 U.S. communities and in different climates

Our results indicate that categorical greenery generally had similar trends as those using continuous greenery but additionally suggest potential cutoff values of neighborhood greenery for facilitating healthy weight status. These findings correspond to a previous study (Villeneuve et al., 2018) reporting that the upper tertile (cutoff values were not reported) of residential greenery (derived using 30m resolution data) was associated with reduced odds of obesity for women across the United States.

While researchers attempt to understand if there is a minimal amount of greenery for promoting human health, an Australian study (Feng and Astell-Burt, 2019) conducted in cities with a temperate climate suggested a cutoff value of 21% aggregate greenery by census unit for helping reduce postpartum weight gain. In our study, we observed differences in cutoff values of neighborhood greenery across climate zones. The observed cutoff values of street tree cover at the distant neighborhood were 15.1% in the arid climate and 40.7% in the temperate climate in this analysis. These findings may be particularly important for greenery planning in different climate zones.

In comparison, Tsai et al. (2019a) reported these cutoff values as 5.6% in the arid climate and 16.7% in the temperate climate, using the same street greenery measures. Since the characteristics of the populations and study designs in these two studies are different in many ways, including gender (only women in the current analysis versus both genders in the previous study), age structure (mean age of 60 and mostly above 55 years old versus mean age of 39 and all under 55 years old), and overall neighborhood greenery (e.g., mean street tree cover at 2000m of 12% versus 9% in the arid climate, and 33% versus 20% in the temperate climate), it is likely that the higher cutoff values and weaker associations observed in the current study are due to these differences. Despite the differences in cutoff values for potentially facilitating healthy weight across studies, both the current and previous studies observed a lower cutoff value of greenery in the arid climate relative to the other climate zones. These findings suggest that greenery in the arid climate where vegetation and water resources are both limited may bring a comparable level of benefits to human health as observed in forested climate zones with greater greenery. This is not only important for promoting human health through greenery but also for environmental resource management.

4.4. Strengths and Limitations

Study strengths include the use of individual-level data from a well-characterized large national cohort (Sandler et al., 2017) and finely resolved data for greenery and built environment measures (Pickard et al., 2015; Pilant et al., 2020) across multiple U.S. urban communities with diverse climate characteristics. The Sister Study collects data on multiple dimensions that allow us to build a robust model with a comprehensive consideration of covariates. The street greenery measures were designed to capture accessible ecosystem services and derived from high spatial resolution (1 m²) land cover data, which allow us to quantify small but important urban natural features (e.g., street trees, front yards, gardens). The built environment measures for park proximity and intersection density enable us take additional important urban features into account for the analysis.

This research has several limitations. First, the results are associative rather than causal due to the cross-sectional design. Second, our study population is relatively homogeneous. All participants are female, and most are aged above 55, non-Hispanic white, well-educated, residing in urban areas, and have a sister diagnosed with breast cancer. Therefore, our results may not apply to the general population. Third, the outcome measure is derived based on self-reported height and weight, though previous research (Lin et al., 2012) observed that the agreement between self-reported and examiner-measured height and weight was high in the Sister Study, with respective Kappa coefficients of 0.84 and 0.92. Another study (Krul et al., 2010) also found that BMI categories based on self-reported height and weight had more than 85% matching rate with objectively measured BMI categories in North America.

We were able to build a parsimonious model using gradient boosted regression trees to account for many individual-level covariates but not all. While physical activity levels influence body weight (Hill, 1997) and have been associated with neighborhood greenery (Hogendorf et al., 2019; Vich et al., 2018; Vich et al., 2019), we do not account for this variable in the current analysis. We do not have information about participants’ preferences for residential locations; therefore, self-selection bias cannot be explored. Neighborhood safety (Kondo et al., 2017) is another factor we were unable to address. Likewise, area-level population, housing density, and socioeconomic measures may confound associations between neighborhood greenery and weight status. However, we observed significant correlations between area- and individual-level covariates. We were able to account for these factors to some extent based on the inclusion of individual-level urbanicity, intersection density, and socioeconomic variables as covariates in our analysis, and sensitivity analysis limited to urban areas yielded similar results.

Greenery was quantified from leaf-on imagery primarily from a single year but may not coincide with the year/season that the Sister Study variables were collected. In addition, greenery was measured based on a fixed location (i.e., primary residence). Residential greenery metrics can under- or overestimate true exposure based on human mobility patterns (Hirsch et al., 2016; Kwan, 2018; Vich et al., 2018). However, greenery measures in this analysis were based on meter-scale land cover data, which permitted more accurate estimates, especially for urban areas, than more widely-available but coarser-resolution data.

Another limitation related to greenery measures is the determination of thresholds for categorical greenery. Some studies have examined potential non-linear relationships between greenery and health outcomes by adding spline terms, and did not observe categorical differences in models without spline terms (James et al., 2016; Tsai et al., 2020). One research team in particular has applied fixed thresholds, commonly at 10%, 20%, 30%, 40%, 60%, and 80% (e.g., Astell-Burt and Feng, 2019a; Astell-Burt and Feng, 2019b; Feng and Astell-Burt, 2019). Our tertile breaks were determined by the distribution of the study population and may vary substantially for different populations. As a result, these thresholds may not be generalizable. Future research should consider applying fixed thresholds in addition to those determined based on data distribution.

Despite limitations, our findings are consistent with previous studies (Feng and Astell-Burt, 2019; Halonen et al., 2014; Persson et al., 2018; Tsai et al., 2019a; Villeneuve et al., 2018) on neighborhood greenery and weight status across several countries and populations. Additionally, this study contributes to emerging information on the importance of spatial contexts of greenery to healthy weight status in women, and on how these associations vary across climate zones.

5. Conclusion

The relationship between neighborhood greenery and weight status is complex. We found stronger beneficial relationships between healthy weight and street tree cover at 2000m in the temperate climate. Results from this study contribute to the existing literature by linking the potential ecosystem services from specific types of neighborhood greenery to healthy weight status across different climate zones. Our findings support considerations of urban greenery accessibility and types by climate for promoting public health and wellbeing. While this analysis links street greenery to healthy weight status, prioritizing greenery along walkable streets may also bring co-benefits for the environment and society through reduced air pollution, increased social interaction, and neighborhood safety. We recommend that future research strengthen the evidence of this study by intervention designs with considerations of spatial contexts and greenery types that are suitable for the local climate or longitudinal approaches so causality can be ascertained.