The study was approved by the Michigan State University's Institutional Review Board (IRB Approval #STUDY00000587; date 03/21/2019 and modified for COVID conditions 05/26/2020). Informed consent was obtained in writing from all participants.

This study involves a subset of participants in a larger, longitudinal study of the benefits of park restoration on health (see Pearson et al., 2020 for more details). Sampling for the parent study was conducted in two stages. First, neighborhoods (n = 9) were selected that contain a park not currently maintained by the Detroit Parks and Recreation Department as a conventional park. Second, to recruit participants, we mailed postcards and conducted recruitment activities (e.g., information booths) in each selected neighborhood (defined as 500 m² with a park as the center). Participants were recruited from within a 16-cell grid (120 m²/cell) around each park. Longitudinal data collection for the parent study is scheduled to occur every May–October until 2023.

In August–October 2019 (late funding led to a truncated field season), field staff visited homes in each study neighborhood to brief potential participants on the study, request participation, and screen for inclusion. We recruited only one English-speaking male or female (≥ 18y) without mobility issues per household, which was at the household's discretion (n = 145 participants). In 2020, however, due to IRB restrictions during the COVID-19 pandemic, no new recruitment took place. Instead, we modified our protocol to collect contactless data. In 2020, we contacted and obtained data for a subset of 86 of the participants from the 2019 wave (52%) from June 15th to July 29th. To retain participants between 2019 and 2020, we employed a set of strategies including: (1) sending holiday cards with neighborhood-level study results; (2) sending birthday cards; and (3) adding news items and updates through the study website (www.stand-detroit.org).

In 2019, a paper survey was given to participants at enrollment, and anthropometrics were taken at a scheduled office health appointment. At the appointment, height was measured twice using a stadiometer (SECA Corp), and weight was measured twice using a scale with bioelectric impedance capability (Tanita TBF-300). Measures were then averaged, and BMI was calculated as a ratio of weight and height (kg/m²). The paper survey was returned during the health appointment.

In 2020, a paper survey was provided in a packet along with a sanitized scale, both placed on a participant's front porch/apartment entry door, while a field staff member waited in a car nearby and watched as the participant retrieved the packet and took their weight on the scale. The participant then phoned or texted their weight and went indoors while the field staff member retrieved and sanitized the scale. Staff returned 3–5 days later to retrieve the paper survey.

Throughout this study, the paper survey was self-administered and completed in the privacy of participants' own homes. The survey included basic demographic data information (age, sex, ethnicity, employment, household composition, length of residence), income, perceptions of the neighborhood (Saelens et al., 2003; Forsyth et al., 2009; Schroeder and Wilbur, 2013; Prouse et al., 2015; Ramos et al., 2016), disease and prescription medication history [National Eye Institute, 2000; Cantor et al., 2009; Audiometry Questionnaire (AUQ), 2018; NHANES, 2018], diet (NHANES, 2010; Nebeling et al., 2017), perceived stress or PSS (Cohen et al., 1983), anxiety and depressive symptoms (Cella et al., 2007; Hays et al., 2009), and attitude toward nature (Nature-relatedness 6, NR-6) (Nisbet and Zelenski, 2013).

In 2020, we also included pandemic-related questions including before and during COVID-19 outdoor physical activities, value of nature, perceived access to nature, and social and economic effects of COVID-19. These questions were registered in the COVID-19 PhenX Toolkit (Community Access module at https://www.phenxtoolkit.org/toolkit_content/PDF/MSU_CWC_Community.pdf). Using a similar approach to Ettman et al. (2020), we generated a “stressor score” which summed all positive agreement for the effects of COVID-19 on life conditions into one value (variables included in Table 1), which ranged from low stressors (0) to high stressors (Ettman et al., 2020) (mean = 1.4, sd = 1.5).

The viewshed, or area visible from participants' homes, was assessed objectively using previously described methods (Nutsford et al., 2015). In brief, we first defined the visibility analysis study area (VASS) as the city boundary of the city of Detroit plus a 12 km buffer. We clipped the VASS using the US/Canada border so that only the area within the US was kept for our analysis, due to data availability (note the clipping did not affect viewshed). A 1 m digital elevation model (DEM) was downloaded from USGS, mosaicked, and clipped to the VASS. To account for the influence of building structures on visibility, building footprint data were downloaded from SEMCOG (https://semcog.org/) and checked against project field maps and Google Earth to make sure building footprints were up to date. Building footprints were then rasterized and added to the DEM to create VASS elevation data. To reduce computation load in visibility analysis, we resampled the VASS elevation data to 3-m resolution. The observer locations were generated from the front door of participants' home locations using building footprints. To locate green space in VASS, we used all parks in the metro-Detroit area, and rasterized them to a 1 m-resolution binary raster where a value of 1 indicated parks. Only parks were used due to concerns that vacant lots vs. parks may have different effects on mental health (Nisbet and Zelenski, 2013). Figure 1 illustrates how the viewshed from a participant's home was assessed using these digital data.

Visibility analysis was conducted in ArcGIS Pro 2.6. The observer offset was set to 1.67 m, which is the average height of participants in our study. To quantify the visibility of parks in viewsheds from a human perspective (Nutsford et al., 2015), we calculated a Vertical Visibility Index (VVI) from each observer location, whereby we accounted for vertical dimensions of visibility (slope, aspect and elevation). The VVI of parks is calculated as the sum of the weighted pixel values of each visible park cell in the viewshed by adjusting for the vertical degrees of visibility between the eye-level of a human and the top and bottom point of each visible park cell. These values were then grouped into tertiles, to represent each 33% higher values in visibility of greenspace in our sample (1 = none or low visibility; 3 = high visibility).

To understand our sample, we first conducted descriptive statistics on the demographics, self-reported effects of COVID-19 on life conditions, and health status of participants, using data from 2019 and 2020. Only data from participants who participated in both years were included in the analysis (n = 86). To assess potential differences in the full sample from 2019 (including those who did not participate in 2020) and the sample from 2020, we conducted t-tests for differences in demographic and health characteristics. We did not detect significant differences in the proportion of females, income, number of children living at home, length of residence, BMI, PSS scores, anxiety or depressive symptoms across the two samples. We did find significantly higher nature-related scores for the full 2019 sample (mean = 3.37 vs. mean = 3.03, p = 0.019) compared to the 2020 sample.

Next, to explore self-reported use and perceived value of nature changes during, vs. prior to, the pandemic, we tested for significant differences in values over time using a paired t-test for continuous data and McNemar's test for paired proportions. We then calculated descriptive statistics on the use and value of nature before (typical month before COVID-19) and during the pandemic (last month) and tested for significant differences using a paired t-test.

To explore whether perceptions of access to and use of nature were associated with mental health measures, we first evaluated correlations between our measures and selected only those with r < 0.6 for model inclusion. Thus, we fitted separate linear regression models for each mental health outcome and included (i) perceived use of neighborhood sidewalks and parks; (ii) perceived accessibility of parks; and (iii) perceived accessibility of sidewalks/shared spaces in the neighborhood. We adjusted for sex, age, married/partner status 2020, stressor score, and the previous year's mental health value (for PSS, anxiety or depression symptoms, respectively).

To explore whether views of greenspace from home were related to mental health outcomes, we first looked descriptively at word clouds of an open response to the survey prompt, “If you have experienced a major stressor in the past 12 months, please let us know what it was,” stratified by whether participants had any view of greenspace from home or none. Next, we fitted separate linear regression models for each of the three mental health outcomes and included visibility of greenspace as the independent variable of interest. These models were adjusted for sex, age, married/partner status in 2020, stressor score, and the previous year's mental health value. We selected this small subset of potential confounders because of the small sample size and due to their well-established associations with mental health (Kessler et al., 2007; Scott et al., 2010). All statistical analyses were performed using Stata v16 (Statacorp, College Station, TX, USA).

When exploring changes in self-reported outdoor physical activities and passive enjoyment of nature before COVID-19 and during the last month (during COVID-19), we observed significant decreases in active transport, riding a bicycle for leisure, walking in the neighborhood for leisure, use of neighborhood sidewalks and parks, use of home-based activity, and smelling rain or plants (Table 2). We observed an average increase in value of nature (mean = 3.4, sd = 1.1, not shown in tabular form) and a significant increase in attitudes toward nature (Table 1).

When evaluating associations between perceived access to nature and mental health, none of the perceived access measures significantly predicted PSS, anxiety or depressive symptoms (Table 3). Being female, older, and having a higher stressor score were significantly associated with higher depressive symptoms. Similar to Table 4 results, PSS, anxiety and depression scores in 2019 all significantly predicted these same outcomes in 2020. We note however that the association for anxiety and depression (in Table 3) is somewhat stronger than for PSS, possibly reflecting the more ‘trait’ like measures of anxiety and depression which tend to remain more stable over time than the more state like measure of PSS which may be more influenced by short-term and recent issues and thus less stable over time.

When evaluating associations between visibility of greenspace from home and health outcomes, we found that higher visibility of greenspace was associated with higher perceived stress (Table 4 Model 1, β = 1.54, p = 0.079) and anxiety (β = 3.97, p = 0.025). Females also had significantly higher depressive symptoms. Across all three models, the previous year's mental health was a significant predictor of mental health measures in 2020.

Prior to the pandemic, among both those with and without views of greenspace, death and issues with family were common major stressors (Figure 2). After the onset of the pandemic, we can see a clear addition of COVID-19 as a major stressor among those with and without views of greenspace. Those without views of greenspace also showed a clear addition of fear as a major stressor. In contrast, those with views of greenspace had more consistent trends in other major stressors including birth, children, and specific family members (especially mother, brother, and sister).

Strengths of this study include its longitudinal design in low-income, minority populations and ability to collect high-quality data in a timely manner during the pandemic. We obtained multiple mental health measures and objectively measured views of greenspace, utilizing geospatial techniques. However, several limitations warrant discussion.

First, our small sample size (largely due to pandemic conditions) limits our ability to adjust for multiple potential confounders and to detect significant associations. Also, we did not obtain objective measures of stress (e.g., blood pressure) in 2020, due to restrictions on contact with participants. We also did not collect objective measures of physical activity changes over time (including where activities occurred or how frequently) using accelerometers and GPS devices, due to research restrictions during the pandemic and mobility restrictions placed on residents in efforts to control disease transmission. Our visibility analyses are sensitive to designation of the viewpoint. We used the front door of participants' apartments/houses as the viewpoint. However, even though the analysis involves the 360-degree possible viewshed, when using the front door as the viewpoint, views behind the house may be blocked. In reality, the back of the building could be just within a few steps where a park may be visible. In other words, the visibility analysis calculates the visibility from a single observation point, but in reality, participants' viewing locations can vary (e.g., different sides of their home, facing of windows, etc.). There may be seasonality effects that influence our results in two ways. First, the survey included questions about activities and perceptions during the past month and during a typical month pre-COVID. As such, one may have considered November to be a typical month pre-COVID, while the last month during the survey was May 2020. Environmental conditions, including birdsong, may vary during different months of the year. Second, the 2019 study wave was conducted August–October, while the 2020 study wave was conducted June–July, which may affect responses. Last, we only quantified views of parks, rather than all forms of greenspace, including street trees or vacant lots. We decided not to capture all greenspace using metrics such as NDVI because, in theory, we were concerned that vacant lots vs. parks may have different effects on mental health (Sivak et al., 2021). But, given our findings related to the negative effects of visibility of parks on mental health, future research could specifically examine quality of greenspaces within view.

Already researchers are grappling with the ways in which COVID-19 might re-shape the ways we use public spaces, city planning and equity into the future (Honey-Roses et al., 2020). Many countries and regions have begun discussions about pedestrian- or bicycle-centered roads, designation of more city greenspaces, or the funding of “green new deals” focused on pandemic economic recovery in areas that could also improve the natural environment and public health. For example, in New Zealand, in May 2020, the government committed to over a billion dollars for “nature jobs,” including over 10,000 jobs in environmental work, biodiversity and environmental conservation. In Europe, the Green Deal, aimed at improving the economy in ways that better serve people and bolster resilience, had already been adopted at the time of the pandemic. Its implications go beyond environmental regulation and include social justice and public health issues: “[The deal] aims to protect, conserve and enhance the EU's natural capital, and protect the health and well-being of citizens from environment-related risks and impacts. At the same time, this transition must be just and inclusive” (European Commission, 2019). However, both the EU Green Deal and pandemic economic recovery call for thinking about longer term trends, re-thinking hegemonic sources of funding and investment, and fostering global interdependence and governance. Given the findings of our study, it is important for policy makers to consider heterogeneity in the effects of COVID-19, neighborhood characteristics and cascading life conditions based on sex, ethnicity and income. It is important to remember that poor mental health is not a new issue and that efforts to leverage the pandemic's changes to our everyday lives need to consider core causes of marginalization and income inequality, while advancing efforts to improve environmental conditions and access to high quality greenspace, particularly in racially or ethnically minoritized and lower income neighborhoods.