The Green Schoolyards Project consisted of two separate studies: a) a serial cross-sectional study focused on physical sites within parks, and b) a prospective cohort study focused on students. The cross-sectional study was designed to examine the associations between heat index, children’s physical activity, and interaction with green features for multiple sites (e.g., playground, track, and soccer field) per park on multiple times per day. The cohort study followed a selection of students from each school affiliated with these parks to assess the impact of green features—the amount of which differs per park—on heat index and student’s physical activity levels during recess, along with their connection to nature, social-emotional learning skills, standardized test scores, and disciplinary behavior.
The research design of the Green Schoolyards Project was based on the social-ecological model of health behavior, which states multiple levels of influence—individual, social, environmental, and policy—impact health behaviors (e.g., physical activity), and these influences interact across levels to impact behavior (19). Effective health interventions focus on behavior-specific influences, and intervene at multiple levels of influence. A principal goal of the project was to learn if green features moderate the relation between ambient heat and physical activity of children, which includes factors within the individual level (i.e., sex, age, race, and ethnicity); environmental level (i.e., green features and heat index); and policy level (i.e., policies that impact park temperatures and park use) of the socio-ecological model.
Three elementary school parks within a school district in Central Texas, US, were used for this study. The project was a comparative analysis between similar parks with different levels of green features. Three schools met our initial selection criteria: serving populations greater than 85% economically disadvantaged Latinx; located in zip codes with low Nature Factor scores; joint-use agreements between the school district and the city Parks and Recreation Department permitting the surrounding community to use parks after school hours; and equivalent park features (e.g., playgrounds, soccer fields, running tracks, and basketball courts).
The selection criterion of Nature Factor is defined as the sum of Nature Factor Ratings of all parks within a zip code (20). Nature Factor Rating is the sum of four park-level ratings: park acreage rating, Trust for the Public Land land use rating, National Recreation and Parks Association park status rating, and tree canopy rating (21). High values for Nature Factor Ratings (e.g., high park acreage, designed lands, open park status, and high levels of tree canopy) correspond to high Nature Factor scores (i.e., higher levels of nature present in that zip code). The three selected schools are in zip codes with Nature Factor scores of 121, 118, and 198, respectively, which are relatively low compared to those of other zip codes (n = 53; min. = 0; max = 712; mean = 150; standard deviation = 143).
The three school parks were characterized by different profiles of green features: the “intervention park” had added green features (i.e., trees, two gardens, and a nature trail); the “low-green park” had relatively low amounts of historical green features (i.e., trees); and the “high-green park” had relatively high amounts of historical green features (i.e., trees and a garden). The school of the intervention park participated in the Green School Parks pilot project, a district-parks department partnership in which green features were installed at elementary school parks. In August 2017, the intervention park received an outdoor classroom, a 3,000-gallon leaky water cistern, two rain gardens with a 1,208 m2 drainage area, and a 76 meter-long nature trail. The community also planted over 100 trees in the park, in October 2017.
For each school park, we calculated tree canopy cover using i-Tree Canopy, a publicly available tool that uses random point sampling to estimate the percentage of tree canopy cover for a predefined area (22). Although trees at the intervention park were more abundant and evenly distributed than trees at the parks at the other schools, the tree canopy cover was only 8.5% (standard error = 1.97) because trees planted were saplings. The low-green park had 11.5% (standard error = 2.26) tree canopy cover, with most trees clustered in the far northwest corner. The high-green park had 22.5% (standard error = 2.95) tree canopy cover, with relatively large trees on the periphery of the park.
We designed data collection to take place on 18 days over the fall semester in 2019, which will be duplicated in 2020 for a total of 36 study days. Each year, study days consist of two September weeks (i.e., five weekdays and one weekend day per week) and one November week (i.e., five weekdays and one weekend day). We selected September and November because these months have historically high and moderate temperature conditions, respectively: the weather station at the city’s major airport recorded monthly mean average air temperatures of 26.4 °C in September and 15.3 °C in November from 2009 to 2018 (23). Prior to undertaking any project activities, we received approval of project protocols by the institutional review board at The University of Texas Health Science Center at Houston (HSC-SPH-19-0502) and the school district. We also received informed consent from participants’ parents and written assent from study participants.
Measurement of Heat Index
We measured heat index—the combination term for air temperature and relative humidity that captures what the temperature feels like (24)—by semi-permanently installing 10 HOBO MX2302A external air temperature/relative humidity sensor data loggers (Onset Computer Corporation, MA) at each park. Previous studies have used comparable networks of in situ sensors to monitor microclimatic conditions of a given area (25–27). Measurement of near-surface air temperatures is advantageous over the use of land surface temperatures as a proxy for air temperatures, as research has shown land surface temperatures are not directly comparable to air temperatures (28, 29). In situ measurement of air temperatures has been found to be more useful for estimating short-term, actual temperature exposures than using land surface temperatures or the percentage of impervious surface (30). Designed for outdoor use, the MX2302A model collects air temperature (± 0.2 °C from 0 to 70 °C) and relative humidity data (± 2.5% from 10–90%), and is configured to wirelessly link with the free HOBOmobile app on a cell phone or tablet (31), permitting efficient collection of air temperature and relative humidity data by project members.
Before deploying HOBO sensors, we encased each sensor in an RS3-B solar radiation shield (Onset Computer Corporation, MA), which results in improved temperature measurement accuracy by protecting the sensor from absorption of incoming solar radiation and resultant heat gain. In another attempt to improve temperature measurement accuracy, we attached the radiation shield (encasing the external sensor) and data logger to a 2 × 2” piece of weather-treated lumber, which serves as a physical buffer minimizing heat transfer between the sensor and the installation surface in the park, such as a metal swing set pole (Fig. 1). We programmed the sensors to record air temperature and relative humidity every five minutes, consistent with previous studies (27, 32, 33).
Figure 1. HOBO MX2302A external air temperature/relative humidity sensor data logger on swing set at low-green park.
For each park, we selected 10 sites for HOBO sensors based on land cover (e.g., grass, pavement, and mulch); land use (e.g., soccer field, basketball court, and playground); comparability across parks; and even spatial coverage (Fig. 2). To include a highly impervious area for comparison within park sites, we installed one of the 10 sensors at each park’s parking lot, just outside park boundaries. To capture air temperature and relative humidity experienced by humans, we installed sensors at two meters above ground level, similar to previous studies (26, 27, 32). To promote community awareness of our project and deter vandalism, we attached a small laminated tag with a description of the sensor and our contact information to each sensor. Sensors were installed the day before a study week, and removed at the end of each study week.
Measurement of Green Features
We identified the location, type, and quantity of green features using four-band, 60 cm orthoimagery taken in November 2018 by the US Department of Agriculture’s National Agriculture Imagery Program (34). Within a geographic information system (ArcGIS 10.6.1, ESRI, Redlands, CA, USA), we digitized polygons of trees, gardens, and nature trails, an established technique deemed appropriate for the relatively small park areas (i.e., intervention = 21,448 m2; low-green = 27,923 m2; high-green = 16,187 m2) (35).
Direct Observation of Parks
The cross-sectional study utilized the System for Observing Play and Recreation in Communities (SOPARC), a validated direct observation tool for assessing the conditions and users of park sites (36). Following SOPARC protocol, we divided each park into target areas intended for physical activity, such as basketball courts and soccer fields (Fig. 2). On study days during school (i.e., 7:00 and 12:00) and after school (i.e., 16:00 and 18:00), study staff—in pairs for interrater reliability—administered SOPARC by walking from target area to target area and recording what they observed.
We adapted SOPARC to measure physical activity levels of children aged 1–12 years old and these children’s interactions with green features. Although previous research has used direct observation to examine the influence of nature on children’s play (37), no research employs SOPARC to quantify the number of children’s interactions with different green features at multiple park sites. On study days for each target area, trained staff recorded target area conditions, scanned for the physical activity levels of female and male children, and then scanned for the number of female and male children interacting with green features (i.e., no interaction, under tree canopy or touching tree, interacting with garden, and on nature trail). Staff used the iSOPARC application on an electronic tablet for scan counts and recorded data on a data collection form (see Additional File 1).
Figure 2. HOBO sensors and SOPARC target areas at (A–B) intervention, (C–D) low-green, and (E–F) high-green parks.
Cohort Study Sample
For the cohort study, we recruited 40 3rd and 40 4th grade students per school over two years, to achieve a final sample size of 30 students per grade, after accounting for attrition. From mid-August 2019 through early September 2019, we recruited participants by convenience sampling. Participant incentives were a total of $35 US dollars/year worth of supermarket gift cards (i.e., $10 for each September study week and $15 for each November study week).
Measurement of Geographic Location and Physical Activity
On study days during recess (i.e., 30-minute period of unstructured play under teacher supervision), the cohort sample wore elastic belts around their waist equipped with a Qstarz BT-Q1000XT Global Positioning System (GPS) device (Qstarz Intl Co., Taipei, Taiwan) and an Actigraph wGT3X-BT accelerometer (ActiGraph LLC, FL) to measure geographic location and physical activity levels, over time (38, 39). We set sampling rates of 15 seconds for GPS devices and accelerometers (40, 41). For the separate recess periods per grade, belt distribution began five minutes before recess start, and belt collection occurred once teachers signaled recess end.
Time-Matching of Geographic Location, Physical Activity, and Heat Index
Data from GPS devices, accelerometers, and HOBO sensors will be time-matched, allowing us to know a student’s location, student’s physical activity intensity level, and heat index at that location at 15-second intervals throughout recess. The location of green features will be joined to the time-matched device data, within GIS. Although previous studies have matched children’s geographic location and physical activity levels over time (38–41), this study enables assessment of a child’s experienced heat index and physical activity level at any particular location.
Measurement of Outcomes Following Park Use and School Policies
For baseline data collection in November 2019, we collected data on the cohort sample and school policies from three sources. First, we administered aloud a written survey—in both English and Spanish language—to the cohort sample, asking them about their connection to nature using two adapted instruments: Inclusion of Nature with Self and Connection to Nature Index (see Additional File 2)(42–44). Second, the school district provided student-level data on sociodemographic characteristics, social-emotional learning skills (from a student climate survey), disciplinary behavior, and standardized test scores. Lastly, we distributed an annual survey—adapted from a previous study (45)—to ask school principals about policies impacting park access, greening at school parks, and student physical activity.
Statistical Analysis of Baseline Data
For baseline findings shared within, we performed the analysis over several steps. To develop a thermal profile for each park, we first calculated heat index—from air temperature and relative humidity data recorded by HOBO sensors—using a set of validated equations utilized by the US National Weather Service (46, 47). Heat index is measured in degrees Fahrenheit, rather than Celsius (48). Within GIS, we used inverse distance weighting to create a spatially continuous thermal profile for each park. A common interpolation method in urban heat island measurement (49), inverse distance weighting permits estimation of unsampled heat index values between HOBO sensors by averaging the values of sampled heat index values from sensors surrounding each prediction location. We used SOPARC data to understand how children interact with trees during time periods with different temperature conditions, summing observed counts of children under tree canopy or touching trees by sex of child, park, and study period.