**4. Discussion**

We examined if the amount of green space such as parks and vegetation level are associated with human mobility under pandemic mitigation policies due to COVID-19 in Maryland and California. A novel result of this study is that the decline of mobility during the COVID-19 pandemic appeared to be lower in regions with green space such as state parks in Maryland and local parks in California. This may imply that people sought green spaces or beach areas to cope with the stress of the pandemic and to perform outdoor activities with social distancing. Further, our results demonstrated that types and scales of parks (e.g., state, local) may impact the effects of parks on mobility changes in different US states potentially due to differences in size of administrative regions (e.g., county subdivision), population density, land cover, or scales of green space (e.g., parks, forest, beach). Our study does

not provide data for an association between mobility and COVID-19 transmission. Nonetheless, the potential impact of green space on mobility during a pandemic shown in our study and other recent literature [7,31,32] implies the need for future design and planning for green space and open spaces where disease control measures such as physical distancing can be performed during an outbreak [31,44,45]. Our results suggest that different plans for sheltering in place in relation to local built environment at regional levels are required. Furthermore, this study is in the scope of sustainable development for resilience and preparedness with nature against future pandemics [46]. Currently, there is an urgent need for studying how mental health consequences can be mitigated in a pandemic [24]. Previous work suggested that a long quarantine duration can be a major stressor causing emotional distress and increased risk of psychiatric illness, unhealthy behaviors, and noncompliance with mandate public health guidelines [47]. A study conducted in Italy provided evidence that the COVID-19 pandemic was significantly associated with increased risks of developing depression, anxiety, and sleeping disorders during lockdown [48]. Given the impact of the pandemic on mental health, and the benefits of green space to mental health, understanding the relationship between the pandemic and green space is paramount.

Many studies have utilized the vegetation index to assess amount of local green space in relation to provision of green space [49,50] and health effects of green space [51]. Given the negative correlation between EVI and urbanization (e.g., population density, impervious areas) in our study, the differences in mobility reduction by EVI levels reflect that people traveled less to more populated areas for safety. This was seen for the graph of daily mobility changes by EVI levels in California. Our data also showed that the presence of parks did not corresponded with higher vegetation level at the MCD (CCD) level in Maryland and California, which indicates that parks were more located in urbanized areas. To control for the effect of urbanization, we examined two types of urbanization indicators (i.e., population density, percent impervious area) in our statistical models. Our results found that the presence of parks and other types of natural areas for recreational activities (e.g., fishing, hunting) was more significantly associated with mobility patterns than was vegetation level. This was particularly due to the characteristics of land cover in CCDs in California where state and national parks were located. These large-scale parks were located in CCDs in central eastern parts of California, where there are also deserts such as the Mojave Desert bounded by the Tehachapi Mountains, San Gabriel and San Bernardino Mountains. These areas are also bounded by Arizona and Nevada where the land cover is largely desert at the borders with California. Due to these characteristics, EVI tended to be lower in these areas despite the presence of state and national parks. On the other hand, lower effects of EVI could be related with the nature of the vegetation index as it does not fully represent the quality of greenness or the actual accessibility (e.g., presence of entry points for people) to green space, although the vegetation index has been used extensively to measure greenness [52]. Study results for the best greenness metrics for assessing health effects are inconsistent. Thus, we suggest that future studies on green space and health consider types and volume of green space in addition to the vegetation index to better understand the functions of green space for outdoor activities and in relation to health, and that research on the characteristics of green space that are most relevant for health be conducted.

Studies suggested that transmission of COVID-19 can be mitigated with further local control measures including suspending public transport, closing entertainment venues, and banning public gatherings in addition to travel bans [1,53]. In response to the pandemic, the Maryland Department of Natural Resources postponed and/or canceled programs and events and closed areas in parks where the public may congregate, such as visitor centers, administrative buildings, and shelters [54]. Several state parks and golf courses were closed to the public during our study period. Gatherings of more than 10 persons were prohibited, and state residents were encouraged to keep distance when outside, not participate in team sports, and avoid touching surfaces that may be handled by others (e.g., playground equipment and benches). Access to most parks in Maryland as of April 2020 was not restricted due to COVID-19. Parks and trails remained open for activities such as hiking, biking, or walking, although state park beach areas were closed on 30 March [55]. Thus, we assumed that

the closure of some parks due to COVID-19 did not affect our analysis on mobility and green space, although uncertainty remains. Similarly, California Department of Parks and Recreation announced its first order for suspension of tours to some California State Parks in 14 March in an effort to protect public health from COVID-19 and announced additional temporary full closure of state parks in 3 April [56]. Since then, access to California State Parks has continuously changed according to the severity of disease spread and compliance with state and local public health ordinances of communities. It is important to note that having different regulations for access to green space hinders direct comparisons of the relationships between green space and human mobility among different US states. This may justify analysis for individual states or administrative units with similar scales.

While our study focused on the increases in visits to green space at an early stage of the pandemic, which was also identified in other literature [7,31,32], COVID-19 has deteriorated the ability of people to utilize urban green space resources (e.g., parks, playgrounds) in many communities [57]. Hall et al. suggested that efforts should be made to assess the lasting effects of pandemics on physical activities along with closures of open green spaces [57]. In addition to such call for research, our empirical results on the relationships between green space and human mobility suggest the need for further understanding and study directions about the way people use green space. Although crowding in urban parks is likely to contribute to transmission of COVID-19, timely data for the number of visitors in green space are not readily available. Further, it is unknown if and how people perform safety measures such as wearing masks and practicing physical distancing in public green spaces. While numerous studies have provided evidence of health benefits of green space on psychological and physical health through stress relief, enhanced physical activities, and social cohesion [25,27], there is a lack of information on which benefits occur when people interact with green space during the pandemic with social distancing policies, and how these relate to the health detriments of increased disease transmission. It is unclear if the frequency or purposes of visiting green space have changed due to the COVID-19 pandemic and, if so, how those changes and resultant health effects differ by subpopulation (e.g., race/ethnicity, socioeconomic status). Future studies will need to address the complex relationships among COVID-19, greens space, and health effects in order to better understand how the pandemic affects our interaction with green space and its health impacts.

Our study has several strengths. To our best of our knowledge, this study is the first to examine the effect of amount of green space on human mobility during the COVID-19 pandemic in Maryland and California. We used novel social media app-based mobility data and considered the density of basic social assets (e.g., number of food retail establishments and hospitals) within county subdivisions (MCDs, CCDs) as an indirect indicator of people's travel outside county subdivisions for basic services (e.g., grocery shopping, health care services). Our results provide timely information relevant for the policies of control measures of a pandemic in relation to green spaces.

This study has some limitations. The mobility identified based on app-based location services can only be observed when the app is active and is thereby may be affected by situations for which people are likely to use a phone (e.g., searching directions, connecting with friends, posting photos). We were not able to consider user's movement within a Bing tile. We could not consider all county subdivisions in Maryland and California due to the lack of mobility data in some subregions based on the privacy protection and lack of Facebook app users. Facebook Data for Good provides daily mobility changes, only since March 2020 with February as the reference period, which prevents comparisons of mobility patterns across years (e.g., the same day one year before) and prevents analysis of mobility patterns during the pre-pandemic period. As a result, we could not disentangle mobility changes due to the state-of-emergency declaration from a potential seasonal pattern of mobility though February–April. A recent study argued that seasonality rather than the COVID-19 pandemic caused the reported increases in park visitations in Google Mobility Reports for some Western US counties [58]. We note that our study area likely shares the same seasonality of mobility across county subdivisions, therefore, the potential impact of seasonality does not likely explain differences in mobility across study subregions. We also note that our main question was the impact of green space on mobility

patterns during the COVID-19 pandemic rather than the impact of the state-of-emergency declaration or specific policies or guidance, which may vary locally, on mobility patterns during the lockdown period. Another limitation is that we could not consider Baltimore city in our regression analysis for Maryland as it was an extreme outlier in terms of population size and number of food retail establishment and hospitals compared to other MCDs, which affected our assumption for the linear relationships for mobility changes and the covariates in the model. Although we used commonly applied datasets for food retail establishments in Maryland, these measures do not include every type of food supply store (e.g., farmer's markets, food trucks). Therefore, there is an uncertainty regarding weather such results would be robust when the choices of types of food retail or health care facilities are different. In addition, food retail data were not available for California in this study. Thus, a cautious interpretation is required of the study results and their comparisons between the study states. We note that our findings of the relationship between presence of parks and mobility changes during the study period may have limited generalizability beyond the sub-urban or peripheral urban areas. This leaves a question for the impact of urbanicity on the use of green space and requires future studies comparing highly urbanized cities in the US. Only Maryland and California were examined in our analysis. The Facebook Data for Good has gradually published open data for mobility for more expended regions since February 2020. However, those updated mobility data have a discrepancy for observation period with previous datasets. We considered Maryland and California, which had the same reference period for daily estimating mobility changes across the US.

Due to the unprecedented nature of the pandemic and the critical need for scientific evidence, many research studies related to COVID-19 have been conducted in an urgent manner in spite of methodological challenges, which leads to some degrees of uncertainty. For the sake of public health against the novel impact of COVID-19, our study examined a unique hypothesis for the relationship between green space, COVID-19, and mobility by combining social media data and satellite remote sensing technology. However, future work is needed to confirm the findings presented here and to investigate relevant questions. Here we discuss suggestions for future studies. First, higher resolution of mobility data would aid understanding of neighborhood-level mobility patterns and their relationships with green space. Individual-level data could illuminate differences in patterns by population characteristics, such as by socioeconomic status and race/ethnicity, which is important given the higher health burden faced by these groups. However, ethical issues of privacy should be considered and continuous efforts to find the best methodological approach for mobility data with lower spatial resolution and with individual-level data. Discussions on finding a scientifically and socially acceptable trade-off between privacy and scientific data needs might support future research. Second, there are various types of mobility data available. While Facebook Data for Good only provides mobility data of Facebook users, mobility data are obtainable through mobile phone networks (i.e., cellular networks). Mobile phone networks are composed of geographic zones (called 'cell') around a phone tower and each mobile phone can be located by identifying the geographical location and the associated cell of its transmitting phone tower [59]. This type of mobility data will incorporate any people who use mobile phone services for a given geographical zone. Facebook's mobility data still have strengths of researcher-friendly pre-generated outcomes by the Facebook Data for Good team, whereas the data based on mobile phone networks from phone towers may require the researcher's own computation logics and detailed understanding of such data. Another type of mobility data is specifically available for "point-of-interest (POI)." Mobility to POI data can be considered in assessing the mobility patterns to each destination of green space including parks and they may be less sensitive to challenges from low spatial resolutions of mobility estimation. Some datasets are freely shared for research through an existing collaborative consortium between the data company and research organizations. Considering this type of datasets based on POI would be helpful to understand the health effects of mobility to green space. Third, there are inconsistencies in the definition of green space among different countries and disciplines, and while various datasets on green space provide information, none portray the rich characteristics of heterogeneity in green space such as different

types of vegetation, park access, park features, etc. Differences in scales, features, and quality in green and open space may result in difficulties in comparing study results among different regions and could obscure important relationships between mobility and green space. It is imperative to understand sizes of geographic divisions, environmental characteristics (e.g., land cover), green space features (e.g., parking availability), and urbanization in study regions to investigate the effects of green space on human mobility. Fourth, more information should be produced for human behavior patterns for using green space. The pattern of using green space (e.g., purposes of visiting) would vary among communities by urbanicity, culture, or safety level, but less is known for these patterns. Studies, possibly with survey of local residents, would be helpful for future research. In spite of these limitations, our work aims to inform future studies of urban sustainability and public health in relation to green space under the threats of global pandemic.
