UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Leaf the kids outdoors : approaches and enquiries in quantifying natural environments for health Davis, Zoë 2020

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata

Download

Media
24-ubc_2020_november_davis_zoe.pdf [ 7.47MB ]
Metadata
JSON: 24-1.0391841.json
JSON-LD: 24-1.0391841-ld.json
RDF/XML (Pretty): 24-1.0391841-rdf.xml
RDF/JSON: 24-1.0391841-rdf.json
Turtle: 24-1.0391841-turtle.txt
N-Triples: 24-1.0391841-rdf-ntriples.txt
Original Record: 24-1.0391841-source.json
Full Text
24-1.0391841-fulltext.txt
Citation
24-1.0391841.ris

Full Text

Leaf the kids outdoors: approaches and enquiries in quantifying natural environments for health  by Zoë Davis  B.S., The University of Vermont, 2015  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE  in  The Faculty of Graduate and Postdoctoral Studies (Forestry)  The University of British Columbia (Vancouver)   June 2020 © Zoë Davis, 2020 ii The following individuals certify that they have read, and recommend to the Faculty of Graduate and Postdoctoral Studies for acceptance, the thesis entitled:  Leaf the kids outdoors: approaches and enquiries in quantifying natural environments for health  submitted by Zoë Davis  in partial fulfillment of the requirements for the degree of Master of Science in Forestry  Examining Committee: Dr. Matilda van den Bosch, Assistant Professor, Forest and Conservation Sciences, UBC Supervisor  Dr. Lorien Nesbitt, Assistant Professor, Forest Resource Management, UBC Supervisory Committee Member  Dr. Martin Guhn, Assistant Professor, Human Early Learning Partnership, UBC Supervisory Committee Member Dr. Hugh Davies, Associate Professor, Occupational and Environmental Health, UBC Additional Examiner      iii Abstract Over the past few decades there has been an increased interest in how urban natural environments (NEs) affect health, particularly with the migration of people into cities and subsequent densification. Recently, there has been an increase in studies on associations between NEs and childhood health and development, but, to date, no systematic assessment of evidence level exists, especially not within the context of different NE exposure measurements. The focus of this research is to determine which NE measurements have been used in childhood health and development research and to assess if associations and the relative level of evidence changes with different NE metrics. Additionally, this thesis seeks to evaluate possible risk of misclassification of NE exposure due to temporal alignment issues between NE exposure and health data by analyzing general and local vegetation change in Metro Vancouver over a 15-year period.   The most common NE metrics identified in the systematic review (Chapter 2) were remote sensing derived metrics, such as the normalized difference vegetation index (NDVI) and land use/land cover datasets. These metrics were also most consistently associated to health outcomes, particularly birth outcomes and decreased ADHD symptoms. Overall, there was considerable heterogeneity within NE metrics, suggesting that more research is needed before conclusions on evidence level can be made.  Through modeling NDVI over time with the Theil-Sen estimator, it was found that vegetation remained relatively stable throughout the Metro Vancouver region, an exception being in localized areas of the southeast. Additionally, large amounts of variation were found in NDVI between years, suggesting that direct comparison of NDVI between years is not advisable. To minimize the risk of misclassification in exposure, datasets should be aligned temporally.  Ultimately, this thesis discusses and highlights the importance of careful consideration of measurement selection for NE exposure in health studies. While confirming some evidence for childhood health benefits through NE exposure, the thesis concludes that, to date, research has insufficiently considered optimal selection of NE metric and the impact of vegetation change over time. These are important consideration for improving evidence-based decisions for healthy urban planning and should be considered in future studies.   iv Lay Summary Exposure to natural environments (NEs) has been shown to promote children’s health. However, there has been little exploration on how different methods for measuring NEs are associated with different health outcomes. This thesis explores methods for measuring NE in childhood mental health and development research and the possible bias introduced through misalignment of NE metrics and health outcomes. The most common NEs used in childhood health were derived from remote sensing imagery. These metrics showed the strongest association to birth outcomes and decreased ADHD symptoms. Results from the evaluation of temporal misalignment of NE indicate vegetation had remained stable in large parts of Metro Vancouver over time, but year-to-year differences were significant and localized change was identified; suggesting temporal alignment between datasets is important for reducing exposure misclassification. This research highlights the importance of careful NE exposure selection in health research to achieve reliable results and improve evidence.   v Preface The research questions and objectives of this thesis were created through conversations between me, my supervisor, and my committee. Portions of this thesis appear as co-authored, peer-reviewed journal articles.  In Chapter 2, Matilda van den Bosch helped me define the research questions. I, Zoë Davis, carried out the database searches and identified the articles to be included in the review and led the data extraction and quality assessment process. This work would not be possible without the help of Dr. Martin Guhn, Ingrid Jarvis, Dr. Michael Jerrett, Dr. Lorien Nesbitt, Dr. Tim Oberlander, Dr. Hind Sbihi, Dr. Jason Su, and Dr. Matilda van den Bosch, who assisted in the data quality and analysis of evidence level. Dr. Jason Su provided the outcomes for the pooled analysis. A version of this chapter has been submitted to a peer review journal:  • Davis, Z., Guhn, M., Jarvis, I., Jerrett, M., Nesbitt, L., Oberlander, T., Sbihi, H., Su, J., van den Bosch, M. (2020). The relation between natural environments and childhood mental health and development: A systematic review and assessment of different exposure measurements.  The work presented in Chapter 3 was the result of exploring the data that is to be used in the Born to be Wise Project. My supervisor and committee helped define the research questions and aided in the interpretation of the results. A version of this chapter has been submitted to a peer reviewed journal: • Davis, Z., Nesbitt, L., Guhn, M., van den Bosch, M. (2020). Change in greenness - when to worry about temporal alignment of environmental exposures in health studies?   vi Table of Contents Abstract -------------------------------------------------------------------------------------------------------- iii Lay Summary -------------------------------------------------------------------------------------------------- iv Preface ---------------------------------------------------------------------------------------------------------- v Table of Contents --------------------------------------------------------------------------------------------- vi List of Tables ------------------------------------------------------------------------------------------------- viii List of Figures ------------------------------------------------------------------------------------------------- ix Abbreviations --------------------------------------------------------------------------------------------------- x Acknowledgements ------------------------------------------------------------------------------------------- xii Dedication ---------------------------------------------------------------------------------------------------- xiii Chapter One: Cities, health, and natural environments ------------------------------------------------------ 1 1.1 Introduction ---------------------------------------------------------------------------------------------- 1 1.2 An overview of theories and pathways behind the association between natural environments and health -------------------------------------------------------------------------------------------------------- 2 1.3 NEs and children’s health ------------------------------------------------------------------------------- 3 1.4 Measuring natural environments ----------------------------------------------------------------------- 5 1.5 Research objectives and thesis overview -------------------------------------------------------------- 9 Chapter Two: The relationship between NEs and childhood mental health and development: A systematic review --------------------------------------------------------------------------------------------- 11 2.1 Introduction -------------------------------------------------------------------------------------------- 11 2.1.2 A deeper look at measuring NEs ------------------------------------------------------------------ 11 2.1.3 Research gaps and objectives -------------------------------------------------------------------- 13 2.2 Methods ------------------------------------------------------------------------------------------------- 13 2.2.1 Search strategy and methods --------------------------------------------------------------------- 13 2.2.2 Study eligibility ------------------------------------------------------------------------------------ 14 2.2.3 Data extraction ------------------------------------------------------------------------------------ 15 2.2.4 Quality assessment and classification of evidence ---------------------------------------------- 15 2.2.5 Pooled analysis of effect size --------------------------------------------------------------------- 16 2.3 Results -------------------------------------------------------------------------------------------------- 17 2.3.1 Natural environment measurements ------------------------------------------------------------- 18 2.3.2 Natural environment measures and relation to health outcomes ------------------------------ 19 2.3.3 Birth outcomes ------------------------------------------------------------------------------------ 20 2.3.4 Academic achievement and absenteeism -------------------------------------------------------- 22 2.3.5 Diagnosed mental health and developmental disorders --------------------------------------- 22 2.3.6 Cognitive development, attention, and social functioning (CDAS) ------------------------------ 28 2.3.7 ADHD/ADD symptoms ----------------------------------------------------------------------------- 30 2.3.8 Memory -------------------------------------------------------------------------------------------- 31 2.3.9 General Measures of Behavior – Internalizing and externalizing behaviors ------------------- 31 2.3.10 Social Functioning -------------------------------------------------------------------------------- 32 2.3.11 Development and well-being -------------------------------------------------------------------- 33 2.4 Discussion ---------------------------------------------------------------------------------------------- 34 2.4.1 Strengths and limitations of included studies --------------------------------------------------- 37 2.5 Conclusions --------------------------------------------------------------------------------------------- 38 Chapter Three: Change in greenness or change in NDVI? --------------------------------------------------- 39 3.1 Introduction -------------------------------------------------------------------------------------------- 39 3.1.1 Research gaps and objectives --------------------------------------------------------------------- 40 3.2 Materials and methods -------------------------------------------------------------------------------- 40  vii 3.2.1 Study area ------------------------------------------------------------------------------------------ 40 3.2.2 NDVI data ------------------------------------------------------------------------------------------ 41 3.2.3 Statistical Analyses -------------------------------------------------------------------------------- 42 3.3 Results -------------------------------------------------------------------------------------------------- 43 3.3.1 Descriptive statistics ------------------------------------------------------------------------------ 43 3.3.2 Temporal change of NDVI ------------------------------------------------------------------------- 46 3.3.3 Identifying areas with change -------------------------------------------------------------------- 47 3.3.4 Sensitivity analysis -------------------------------------------------------------------------------- 50 3.4 Discussion ---------------------------------------------------------------------------------------------- 52 3.5 Conclusions --------------------------------------------------------------------------------------------- 55 Chapter Four: Conclusions ------------------------------------------------------------------------------------ 57 4.1 Overview and key findings ----------------------------------------------------------------------------- 57 4.2 Limitations ---------------------------------------------------------------------------------------------- 58 4.3 Future research ----------------------------------------------------------------------------------------- 59 4.4 Closing thoughts --------------------------------------------------------------------------------------- 61 References ----------------------------------------------------------------------------------------------------- 62 Appendix A ---------------------------------------------------------------------------------------------------- 88 Appendix B ---------------------------------------------------------------------------------------------------- 98 Appendix C --------------------------------------------------------------------------------------------------- 125      viii List of Tables Table 2-1: Characteristics of different NE metrics. ------------------------------------------------12 Table 2-2: Search terms, key words, and MeSH terms used in the search. Each topic was adapted for each database search. ‘?’ denotes wildcard in queries. Full reproducible search can be found in B1 of the Appendix B. ------------------------------------------------------------14 Table 2-3: NE metric categories and rates included in the studies. Count refers to the number of studies (significant findings) that employed the metric and rate. See Table 2-1 for details of metrics and rates.-----------------------------------------------------------------------------------19 Table 2-4: Number of NE metrics used for each health outcome. Note: For studies that used multiple NE metrics, each significant finding was counted separately, therefore, the total n exceeds number of studies evaluated.------------------------------------------------------------20 Table 2-5: Associations between NE and outcome for each study. ↑ indicates positive associations between NE exposure and health outcome, i.e., NE improves health; ↓indicates an inverse association between NE exposure and health outcome, i.e., NE worsens health; ↔ indicates no significant association between NE exposure and health outcome. ----------24-27 Table 2-6: Specific tests and subscales used in each CDAS category. ---------------------------29 Table 2-7: Summarized results of the associations between health outcomes and NE metrics. --------------------------------------------------------------------------------------------------------34 Table 3-1: Descriptive statistics for NDVI within the Metro Vancouver for each buffer and year. ---------------------------------------------------------------------------------------------------44-45 Table 3-2: Descriptions of the Theil-Sen model slope characteristics. --------------------------48 Table 3-3: Descriptions of the OLS model slope characteristics. --------------------------------50           ix List of Figures Figure 1-1: Example pathway model in which NE can influence health. --------------------------4 Figure 1-2: The electromagnetic spectrum. Image created by Philip Ronan, Gringer / CC BY-SA (https://creativecommons.org/licenses/by-sa/3.0). The NIR band is between 750 to 2500 nm.---------------------------------------------------------------------------------------------------------6 Figure 2-1: An example of how NEs are represented in Euclidean buffers around an address of interest (point). The satellite imagery from Planet Labs (2016) (a), displayed in a false color composite, was used to calculate the NDVI layer (b). Green represents high NDVI values (closer to 1) and red representing values closer to -1. For each buffer (100 m, 250 m, 500 m, and 1000 m) the average NDVI was taken for each buffer (rate). For the LULC dataset (Williams et al., 2018) (c) the proportion of NEs was taken for each of the buffers. -----------------------13  Figure 2-2: PRISMA diagram of the procedure for article selection. -----------------------------17 Figure 2-3: Summarized birth weight changes per 1-IQR increase in NDVI (30 m Landsat) within varying buffer sizes for five studies. Color of the line represents individual studies, and solid gray line represents the pooled effect size for the median of all effect sizes across buffers; dotted gray line represents the confidence interval. --------------------------------------------21 Figure 3-1: Extent of Metro Vancouver with constituent municipalities. ------------------------41 Figure 3-2: Boxplots of NDVI values by year and buffer size at the postal code level. ---------46 Figure 3-3: Pearson’s correlation matrix of NDVI values for all buffer (point values and 100 m - 1000 m) and year (1999 - 2014) combinations. Scale on the right denotes the correlation coefficient of each combination. ------------------------------------------------------------------47 Figure 3-4: Direction and steepness of the slope for each Theil-Sen model for each grid cell. Orange indicates more positive changes in NDVI over time, purple represents more negative changes in NDVI over time, and yellow indicates no change. -----------------------------------49 Figure 3-5: Direction and strength of the slope for each OLS linear model for each grid cell. Orange indicates more positive changes in NDVI over time; purple represents more negative changes in NDVI over time. ------------------------------------------------------------------------51 Figure 3-6: Theoretical NDVI values over time for a set of postal codes. The points represent hypothetical postal code NDVI values over time. ------------------------------------------------54     x Abbreviations ADHD – attention deficit/hyperactivity disorder AEDC – Australian Early Development Census ART – attention restoration theory CANUE – Canadian Urban Environmental Health Consortium CDAS – cognitive development, attention, and social functioning DSM-IV – Diagnostic and Statistical Manual of Mental Disorders, fourth edition EM – electromagnetic  g – grams GEE – Google Earth Engine GPS – geographical positioning system HRQOL – health related quality of life HRT-SE –hit reaction time – standard error HS – Hind Sbihi IARC – International Agency on Research on Cancer IJ – Ingrid Jarvis IQR – interquartile range JS – Jason Su K-CPT – Conners’ Kiddie Continuous Performance Test LC – land cover LN – Lorien Nesbitt LU – land use LULC – land use/land cover m – meter MAUP – modifiable areal unit problem  xi MeSH – medical subject headings MG – Martin Guhn min – minute MJ – Michael Jerrett mm – millimeter MODIS – moderate image spectroradiometer  MvdB – Matilda van den Bosch NA – not applicable NCD – non-communicable disease NDVI – normalized difference vegetation index  NE(s) – natural environment(s) NIR – near-infrared  NSI – natural space index OLS – ordinary least squares OPEC – outdoor play environment categories POSDAT – public open space desktop auditing tool PRISMA – Preferred Reporting Items for Systematic Reviews and Meta-Analyses SDQ – Strengths and Difficulties Questionnaire SRT – stress reduction theory TO – Tim Oberlander TOA – top of atmosphere USA – United States of America xii Acknowledgements I would like to express my gratitude to my supervisor, Matilda, for giving me a chance to be a part of this project. Her guidance and support throughout this process was inspirational and I am thrilled to have worked with her. I would also like to thank Lorien Nesbitt, Hind Sbihi, and Martin Guhn for the meetings, coding sessions, and conversations that helped guide this research.  Thank you to Marie and James for providing an amazing home and to be kind enough to share their cat, Joe, with me.  I am grateful for the friends I have made here at UBC and the ones who kept me sane from afar. Special thanks to Ingrid Jarvis, Susan Winters, Stefanie Lane, Joe Egnot, Sam Grubinger, Sunny Tseng, Yue Yu, Ella Stephens, Adam Zylka, and Sarah Leidinger, for always being there.  Thank you to my parents, especially my mom, for moving me across the continent and never telling me that I couldn’t do anything. Finally, thank you for Sam Herniman, for all the love and support through it all.  This project was part of the Born to be Wise project, funded through the Canadian Institutes of Health Research.        xiii Dedication To coffee, for without it, this work would not have been possible.   1 Chapter One: Cities, health, and natural environments 1.1 Introduction Currently, more than half the world’s population lives in cities (United Nations, 2018) and in Canada, over 80% of the population is urban (Statistics Canada, 2018). Cities provide many people with a place to live and work; additionally, they have become hubs for governments, commerce, and social capital (United Nations Department of Economic and Social Affairs, 2016). Through this shift towards urban life, more and more children now grow up in urban environments (UNICEF, 2012). Children living in cities can benefit from accessible health care, cultural resources, and educational opportunities (UNICEF, 2012). However, they are also exposed to negative environmental and cultural factors, such as changes in food access and nutrition, increases in harmful environmental exposures (e.g., air and noise pollution), reduction of opportunities for physical activity, and disconnection from nature (Gracey, 2007; Moore et al., 2003). These factors may be contributing to a rise in non-communicable diseases (NCDs), such as childhood obesity and diabetes (Allender et al., 2011; Pirgon & Aslan, 2015), particularly in vulnerable urban groups such as the socially disadvantaged (Chiabai et al., 2018; World Health Organization, 2011). As a response, there has been a growing interest in how to improve the health of people in cities through progressive urban management and planning strategies, for example, nature-based solutions (European Commission Directorate-General, 2015; Northridge & Sclar, 2003). Nature-based solutions implement natural environments (NEs) or natural elements to address a number of environmental, social, and economic problems (European Commission Directorate-General, 2015), and are one technique municipalities are adopting to achieve ‘livable’ city goals. Livable city definitions allow for city planners to focus attention on areas of weaknesses and strengths of city centers, thus enabling the creation of better places to live (Balsas, 2004). However, livable is a difficult term to define, and is highly contested, as it can measure a variety of factors that are all indicative of the quality of urban life (Ruggeri et al., 2018; Zanella et al., 2015). For instance, livability may be based on the cost of living and job opportunities; however, livability may also be defined by access to amenities for culture or leisure (Parker & Simpson, 2018; Zanella et al., 2015). While both of these definitions are correct, in this thesis, access to natural environments will be primarily focused on. NEs encompass a wide array of urban landscapes, such as parks, woodlands, street trees,  2 agriculture, and waterfronts and provide a wide range of benefits to urban communities through ecosystem services (ES) (Coutts & Hahn, 2015; Konijnendijk et al., 2005; Millennium Ecosystem Assessment, 2005). ES encompass a wide range of services provided by natural environments, such as cultural spaces, regulating of environments, and disease prevention (van den Bosch & Nieuwenhuijsen, 2017). For children, ES include the systems and areas that promote health and well-being, such as regulating climates and providing places to play and socialize (European Commission Directorate-General, 2015; van den Bosch & Nieuwenhuijsen, 2017). There is a diverse and growing literature base exploring the association between NEs and health as a basis to provide recommendations for healthy living.   1.2 An overview of theories and pathways behind the association between natural environments and health Studies focusing on the associations between nature and human health began in the late 20th century with descriptive models and hypotheses to explain the relationships between NEs and health (Seymour, 2016). The most common of these theoretical frameworks include the restorative effects of NEs, most notably the stress reduction theory (SRT, Ulrich et al., 1991) and the attention restoration theory (ART, Kaplan & Kaplan, 1989; Kaplan, 1995). SRT proposes that particular landscape features create immediate reactions, such as likes and dislikes, without conscious processing (van den Berg & Staats, 2018). Landscapes which contain NEs may create positive emotions, thereby be ‘soothing’ to individuals and reduce the individual’s stress. Studies that have tested this hypothesis have found faster and more complete stress recovery in individuals who viewed scenes of NEs after a stressful experience compared with individuals that viewed urban environments (Ulrich, 1984; Ulrich et al., 1991).  The attention restoration theory (ART) assumes people only have a limited capacity to direct their attention to tasks that require cognitive energy, and with use, this capacity becomes depleted (Kaplan, 1995; van den Berg & Staats, 2018). The ART proposes that exposure to restorative environments allows for the restoration of attention, and cumulative repeated exposures are essential for maintaining full attention (Kaplan & Kaplan, 1989; Kaplan, 1995). ART proposed four characteristics of restorative environments: the feeling of being away from daily life; a sense of extent; fascination with the environment; and a compatibility of the environment to individual interests (Kaplan, 1995). While other environments (e.g., museums) can be restorative for some groups of people, NEs more commonly contain these elements  3 (van den Berg & Staats, 2018). ART therefore concludes that being in nature can restore attention.  In addition to benefits associated with direct contact with NEs, additions to the framework include multi-pathed and indirect pathways (e.g., Hartig et al., 2014; Lachowycz & Jones, 2013; Markevych et al., 2017; Villanueva et al., 2013). Indirect pathways include the benefits to health that result from NE influence on another factor; for example, the health benefits from participating in physical activity as a result of access to NEs. Additionally, some of these pathways can occur simultaneously, through many paths. For example, a reduction in heat due to NEs could allow for more physical activity because the space is cooler. In some situations, NEs can be harmful to health, for example pollen and volatile organic compounds produced by plants that are irritant to health (Livesley et al., 2016). While these side effects are important when considering NE in urban planning, in general, the net-benefits of NEs outweigh the negative effects (Braubach et al., 2017; van den Bosch & Nieuwenhuijsen, 2017). 1.3 NEs and children’s health Today, children interact less with NEs than they did in previous generations due to urbanization and shifts in lifestyles towards more sedentary and indoor activities (Louv, 2008; Radesky & Christakis, 2016). Additionally, urban life exposes children to harmful social and physical environments, such as higher levels of pollution and noise, closer proximity to toxic sites, higher rates of crime, and a potential lack of social cohesion within the community (Woolf & Aron, 2013). These changes in environment may be one of contributing factors to increases in childhood NCDs, including obesity, diabetes, and mental illnesses like Attention Deficit and Hyperactivity Disorder (ADHD) (Anderson & Butcher, 2006; McMartin et al., 2014; Patel et al., 2007). Children are particularly susceptible to negative environmental exposures, due to factors such as body size-to-dose ratio, immature metabolic pathways, and sensitive developmental systems (Suk et al., 2003).  The theory of developmental origins of health and disease (Gluckman et al., 2016) suggests that exposures during early life may create permanent changes to the physiology, metabolism, and structure of the body, which may influence health and disease development later in life (Gluckman et al., 2016). Potential mechanisms that may be relevant to children are highlighted in Figure 1-1. Here, NEs influence health through two main categories: active  4 contact with NEs, and passive exposures to NEs. Active contact involves activities where children can interact with NEs, such as through physical activity, exploration of new places, contact with elements of nature such as microbes, and through social connection created through visits with friends and family. These exposures also allow for the restoration of attention as proposed by the ART hypothesis. Passive exposures include the benefits that happen through daily exposure, such as through reduction of harmful exposures and the reduction of stress through passive viewing of NEs. Additionally, other factors affect the health of children, such as prenatal exposures (Suzuki, 2018).   Figure 1-1: Example pathway model in which NE can influence health. Several of these environmental factors are modifiable and therefore possible to address through policy interventions. For example, it is possible to reduce traffic-related air pollution by closing roads in the vicinity of schools and creating public parks in residential neighborhoods to create areas for play and socializing. Studies supporting NEs as a way to promote health have increased in the last decade (Kondo et al., 2018; Markevych et al., 2017; Rojas-Rueda et al., 2019). Recent reviews have found mostly small effects of improved birth outcomes (Dzhambov et al., 2014), increased physical activity (Gray et al., 2015), mental health and well-being (McCormick, 2017; Vanaken & Danckaerts, 2018), and academic achievement (Browning & Rigolon, 2019) with exposure to NEs. However, despite the number of reviews conducted on childhood health outcomes, only one review (Browning & Rigolon, 2019) assessed the relationship between different NE metrics and health outcomes. Subsequently, little is known about how different methods of NEs measurements relate to specific health outcomes of interest.     5 1.4 Measuring natural environments The pathway models of NE exposure and health operate under the assumption that individual people are nested within community determinants of health (e.g., physical environment, lifestyle, and local community, etc.). In current research, focus has been on intermediate and distal determinants of health, such as physical activity, social capital, and stress (Coutts & Hahn, 2015). These assumptions influence the ways in which NEs are measured and what methods are used for determining individual and population level exposures to NEs.  Typically, measurements of exposure to NEs are based on certain spatial units of analysis, for example individual, neighborhood, and city scales (Labib et al., 2020). Each unit represents a level of exposure that may affect an individual’s health via different pathways (Bratman et al., 2019). For example, an exposure occurring within the immediate vicinity of a person (e.g., ≤ 100 m) may correspond to a pathway related to stress reduction, as proposed by SRT, whereas an exposure measured at a larger scale (e.g., ~1000 m) may correspond with pathways related to physical activity (Markevych et al., 2017). Many studies utilize multiple measurements to capture possible pathways linking NE and the chosen health outcome. Distances can be measured via various methods, such as through the use of Euclidean buffers around an address of interests (e.g., residential or school address) or by creating boundaries around a neighborhood defined by municipal boundaries, or boundaries determined by participants. In this thesis, an individual’s exposure to NE is referred to as the ‘rate’ of exposure.   In addition to different rates that can be used to estimate individual NE exposure, there are also multiple metrics used to estimate NEs with an area. Common ways of measuring NE include remote sensing derived datasets, such as vegetation indices and land use/land cover (LULC) datasets, data collected by experts (e.g., surveys on quality completed by trained personnel), surveys completed by participants, and measures of use by participants (Labib et al., 2020). In this thesis, the combination of NE metric and rate is referred to as an NE measurement, but both metrics and rates are discussed separately.      Remote sensing data are collected by capturing solar radiation that is reflected off the Earth’s surface using a sensor (Lavender & Lavender, 2016). Commonly, sensors are mounted  6 on satellites, planes, or drones (Canadian Centre for Remote Sensing, 2016; Pettorelli et al., 2018), and are typically developed to be sensitive to particular areas of the electromagnetic spectrum (EM; Figure 1-2), referred to as bands. Bands are spaced across the EM spectrum in sections that are useful for identifying types of land cover or other features of interest. For applications involving vegetation, bands typically include areas of the visible wavelengths (red, green, blue) but also bands in the non-visible spectrum that are sensitive to properties of vegetation, such as near-infrared (NIR).   Figure 1-2: The electromagnetic spectrum. The NIR band is between 750 to 2500 nm. Image created by Philip Ronan, Gringer / CC BY-SA (https://creativecommons.org/licenses/by-sa/3.0).   The earliest available satellite imagery was collected by the Landsat satellite in 1984 and continued operating to present day (Markham et al., 2004). Since then, moderate resolution sensors such as MODIS (1999 – present) and Sentinel (2014 – present), as well as high resolution sensors such as Planet Labs, have become publicly available (Lavender & Lavender, 2016). Each of these sensors have particular advantages and drawbacks, which make some a better choice than others for particular research questions (Lavender & Lavender, 2016). Due to the variety of sensors available and the scope of imagery across many decades, remotely sensed imagery has become a common choice for measuring NEs in epidemiological research (Hay, 2000). The limitations of remote sensing data are determined by their resolutions: spatial, spectral, temporal, and radiometric (Hay, 2000; Pettorelli et al., 2018). Spatial resolution refers to the area a pixel represents on the ground – the smaller the pixel size, the higher the spatial resolution, and the smaller the ground area represented by each pixel. In the context of NE- 7 health research, higher spatial resolution allows for better identification of NEs in the area of interest. Spectral resolution is the range of the EM spectrum to which the sensor is sensitive. Particular configurations of bands are useful for identifying different types of ground cover, for example, vegetation. The return time of the sensor to a certain ground location on Earth is the temporal resolution. Finally, radiometric resolution is the amount of information that can be recorded in a pixel by a sensor. The higher the radiometric resolution (bits), the larger the range of values that can be recorded for any particular band. As an example, an 8-bit blue band will have 16 times more possible blue values than a 4-bit blue band, and consequently would be able to detect smaller differences between values of blue. Each sensor is limited by these resolutions; most of the time one resolution is maximized while another is minimized (Canadian Centre for Remote Sensing, 2016; Lavender & Lavender, 2016). Therefore, these trade-offs determine which sensors are best for a particular research question (Hay, 2000). Additionally, the historic archives of a sensor are another important aspect. For instance, high spatial resolution imagery from a sensor such as Planet Scope (3 m) may be useful in terms of identifying small NE features, but access to imagery over many years is limited because Planet Scope first launched in 2016. Thus, a sensor with a larger historical scope, such as Landsat (launched in 1984), may be a better choice, particularly when following cohorts over time, despite the lower spatial resolution (30 m).  The normalized difference vegetation index (NDVI) is one of the most common methods for assessing exposure to natural environments via remotely sensed imagery (Labib et al., 2020). NDVI is calculated as the ratio between the visible red band (red) and the near infrared band (NIR) of the respective sensor, as NDVI = (NIR-red)/(NIR + red) (eq. 1) (Tucker, 1979). NDVI was originally developed to estimate vegetation health, with values closer to 1 representing healthy photosynthetic plants; values near zero representing unhealthy vegetation; and values near -1 representing water and inanimate objects (Deering & Rouse, 1975). For epidemiological work, NDVI has been widely used as a proxy for human exposure to NEs (e.g., Dadvand et al., 2015), and it is assumed that NDVI can represent the amount of NEs that people come into contact with, typically within a circular buffer (rate) around a home addresses or other address of interest (e.g., schools). NDVI may be a common choice because it is easy to calculate and shows strong correlations to the amounts of NEs (Gascon et al., 2016).   8 Additionally, multiple data sources, such as lidar, municipal boundaries, and satellite imagery, can be combined to create LULC datasets (MacFaden et al., 2012). These datasets represent discrete classifications of features on the ground (Foody, 2002). For land cover (LC), these classes represent the true ground cover, such as urban NEs (possibly distinguishing between different types of NEs, such as grass, trees, or shrubs), built-up surfaces, and water. Conversely, the classes in land use (LU) datasets represent the intended development or areas with a defined purpose (e.g., sports and leisure facilities, urban green spaces, forests, European Environment Agency, 2017) and are typically categorized by zoning boundaries created by municipalities (Stefanov & Netzband, 2010). The quality of both LC and LU datasets are assessed by their spatial resolution and the number of pixels that are accurately classified in the dataset (Finegold et al., 2016). In general, LULC datasets have high accuracy, making them ideal candidates for identification of small areas of NEs. In comparison with NDVI, they offer the ability to discretely categorize an area as ‘green’ or ‘not green’. However, due to the intensive process of gathering, processing, and creating these data, LULC datasets are generally only available for fewer points in time, making direct alignment to health data more difficult.  Other ways of measuring NEs include the use of on-site evaluations, such as the Public Open Space Desktop Auditing Tool (POSDAT) (Edwards et al., 2013) and the Natural Space Index (NSI) (Rugel et al., 2017), surveys or audits conducted by professionals (e.g., Wells, 2000), self-reported exposure (e.g., Amoly et al., 2014), and monitoring use of NEs through geographic positioning systems (GPS) (e.g., Ward et al., 2016). These methods allow for individual and on-site evaluations of exposure, however, they come at the cost of replication difficulties, both in terms of the number of participants and the collection of data over time. There is some evidence that perceived greenness by lay people poorly correlates to NDVI, suggesting survey metrics may be a more accurate, or alternative, measure of people’s actual exposure to NEs (Leslie et al., 2010).   A rarely considered limitation of NE metrics is the alignment to the health dataset on a temporal scale. This means that the data for exposure assessment is collected from a different time period than the collection of health data, which increases the risk of exposure misclassification. To counter this issue, a common praxis has been to assume the study area has not changed between the NE measurement capture date and the date of the health data,  9 or in other words to assume vegetation stability (e.g., NDVI, Helbich, 2019). The validity of this assumption has rarely been tested (Helbich, 2019). If significant changes in vegetation have occurred, as is likely in rapidly urbanizing and developing areas, this praxis could lead to inaccurate exposure assignments.    1.5 Research objectives and thesis overview The relationship between NEs and children’s health and well-being has been explored more frequently in recent years. However, better evidence is still needed before recommendations can be made on how NEs influence childhood health, particularly when it comes to the way NEs are measured. Currently, there are many NE metrics used in epidemiological research (Labib et al., 2020), however, there is a need to know which metrics demonstrate the strongest relationships to certain health outcomes. Knowledge of this may help support the use of certain NE metrics when modeling particular pathways, thereby providing support for hypothesized theories.  The use of remote sensing metrics has allowed for a common, and relatively quick technique to assess NE exposure. However, studies implementing remote sensing data have not always assured the alignment of data in time to health outcomes. This may increase the risk of exposure misclassifications due to changes in the landscape that occurred during the intervening years.  The objective of this thesis is to assess the possible relationships by which NEs are measured, both in the context of associations to different childhood health outcomes and in the potential addition of bias due to temporal misalignment of data. Better evidence is needed to assess the relative relation of different health outcomes to NE and the ways in which NE is measured.  Chapter 2 explores the relative association between different NE metrics and childhood mental health and development through a systematic review. The objectives of this chapter are two-fold: Firstly, to identify which NE metrics are most commonly used in childhood health studies; and secondly, to assess the relative strength of association between NE metrics and their respective health outcomes and to assess if any of the associations differ depending on the metric used. Specific objectives for this chapter are:  10 • Identify the most common metrics used in childhood development mental health research.  • Assess the level of evidence for the association between childhood mental health and development and NE exposure. • Assess how potential association may differ depending on the metric used in the context of childhood development and mental health. Chapter 3 explores the common methodology of using NE exposure and health data that do not correspond to the same time period. A clear understanding of how these data change over time is essential for accurate assignment of NE exposures. Using Metro Vancouver as a case study, the specific objectives of Chapter 3 were to: • Develop a method for analyzing change in NDVI over time. • Examine NDVI at localized postal codes and at aggregated postal code geographies to determine how vegetation may have changed over time.         11 Chapter Two: The relationship between NEs and childhood mental health and development: A systematic review 2.1 Introduction Over a billion children now live in urban places (UNICEF, 2012). While plans for designing resilient and livable cities offers people access to healthy environments, children are often insufficiently considered in such plans, particularly in relation to children’s health and development (Ataol et al., 2019; Bishop & Corkery, 2017; Karsten, 2005).   Considering this, there is a growing body of research that suggests an association between exposure to NEs and childhood health. A review by Browning & Rigolon (2019) found weak evidence between NE measurements and academic achievement in large buffer zones. Dzhambov et al. (2014) found a weak, positive association between birth weight and NE within 100 m NDVI, however, the heterogeneity of measurements made it difficult to compare studies. In general, children who spent more time outside, also spent more time being physically active, and less time sedentary (Gray et al., 2015). Additionally, general improved mental health, overall well-being, and cognitive development were associated with higher access to NE (McCormick, 2017; Vanaken & Danckaerts, 2018). Despite the positive findings from most of these reviews, only one study assessed the association between a health outcome (academic achievement) and different types of NE measurements (Browning & Rigolon, 2019). Evidence suggesting which NE metrics are most strongly associated with different types of health outcomes may help inform particular pathways of interest. However, the relationship between different NE metrics and various health outcomes remains mostly unexplored.  2.1.2 A deeper look at measuring NEs There are many ways of measuring exposure to NEs; Table 2-1 describes some of the common metrics used for NE exposure assessments and their respective spatial and temporal characteristics. For NE metrics derived from remote sensing data, rate is often determined by buffers or polygons surrounding a location of interest (e.g., residential address), wherein a distance-to-address or proportion of NE within the area can be determined. In other cases, such as GPS tracking or surveys, rate is determined by the amount of time within NEs.   12 Table 2-1: Characteristics of different NE metrics. NE Metric Source (sensor) Spatial resolution Temporal Characteristics NDVI Landsat 30 m  • Imagery available since 1984 (Landsat 5) • 16-day revisit time NDVI MODIS 250 m • Imagery available since 2000  • 16-day composite imagery (from daily collection) NDVI RapidEye 5 m • Imagery available from 2009 • 5.5-day revisit time LULC Multiple  (e.g., imagery, lidar, municipal layers, etc.) Depends on source imagery (< 10 m most common) • Availability depends on the agency (e.g., municipality) responsible for production On-site evaluations Multiple  (e.g., survey conducted by professional, sky-view index) NA • Varies depending on metric Self-reports Multiple  (e.g., survey conducted by participant) NA • NA Use of NE Multiple  (e.g., survey, GPS monitoring) NA • Availability depends on the method of collection, for instance, GPS measurements could be for a week, and surveys could cover a year Typically, for NE measurements determined by NDVI or LULC datasets, Euclidean buffers are used to approximate exposures. In the example in Figure 2-1, the rate is determined by a circular, concentric buffer around a residential address. For NDVI datasets (Figure 2-1b), all cells within the buffer are averaged to provide a single aggravated value of NDVI for each buffer. This NDVI dataset can be based on a single image captured during a specific date, or it can be an aggregated exposure, such as a yearly average of NDVI. For LULC datasets, the proportion of the buffer classified as a NE (e.g., tree, grass/shrub, recreation facility, etc.) is calculated (Figure 2-1c).   13 Figure 2-1: An example of how NEs are represented in Euclidean buffers around an address of interest (point). The satellite imagery from Planet Labs (2016) (a), displayed in a false color composite, was used to calculate the NDVI layer (b). Green represents high NDVI values (closer to 1) and red represents values closer to -1. For each buffer (100 m, 250 m, 500 m, and 1000 m), the average NDVI was taken for each buffer (rate). For the LULC dataset (Williams et al., 2018) (c) the proportion of NEs was taken for each of the buffers. 2.1.3 Research gaps and objectives Past studies have shown positive association between health outcomes and exposure to NEs, but there is little evidence evaluating the methods with which NE is measured and how this may influence the association to health. The overall goal of this chapter is to make a comparison and determine the relative associations between NE metrics and childhood health outcomes. The three objectives focused on in this chapter are: (i) identify the most common NE measurements used in childhood mental health and development research, (ii) identify which NE measurements are most consistently related to childhood mental health and development, and (iii) determine the level of evidence for an association between different NE measurements and respective health outcomes. This chapter was conducted with the help of Martin Guhn (MG), Ingrid Jarvis (IJ), Michael Jerrett (MJ), Lorien Nesbitt (LN), Tim Oberlander (TO), Hind Sbihi (HS), Jason Su (JS) and Matilda van den Bosch (MvdB). Explanations of their contributions is included in the Preface section (p. v) 2.2 Methods 2.2.1 Search strategy and methods Prior to conducting the review, a protocol was created to define the design, strategy, and methods (Appendix A). Using the guidelines determined in the protocol, a search for relevant articles was carried out on July 31st, 2018 in four databases: EMBASE, MEDLINE (National Library of Medicine), Web of Science, and PsycINFO (American Psychological Association)  14 using Medical Subject Headings (MeSH) and/or keywords (Table 2-2). The keywords for children’s health and development were derived from a search filter curated by the University of Alberta (Tjosvold et al., 2015) and modified for queries in the Web of Science and PsycINFO. The full reproducible searches can be found in Table B1 of Appendix B. The search was limited to studies published in English between January 1, 2000 and July 31, 2018. The initial screening considered title and abstract. Studies deemed potentially valid were retrieved for full article screening and “snow-balling” of bibliographies for identification of additional articles. The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) protocol (Moher et al., 2009) was used for determining eligible articles to include. Table 2-2: Search terms, key words, and MeSH terms used in the search. Each topic was adapted for each database search. ‘?’ and ‘*’ denotes wildcard in queries. Full reproducible search can be found in Table B1 of the Appendix B.  Topic Search Terms NE exposure MeSH Terms: City Planning, Forests, TREES, Environmental Design, Urban Health, Environmental Planning Keywords: green space?, greenspace?, green?ess, blue space?, urban forest?, open space?, natural space?, green, public park?, vegetation, tree?, natur? NE metric MeSH Terms: geographic information system, remote sensing, remote sensing technologies Keywords: geographic information system?, GIS, metric, remote sens?, land cover, landcover, normali?ed difference vegetation index, NDVI, green view, GVI, EVI, enhanced vegetation index, OBIA, object based image analysis, lidar, landsat, sky view index, vertical visibility index, VVI, spatial, tree canopy change, google street view, GSV, accelerometry, global position, ecological momentary assessment, EMA Childhood mental health and development exp child/ or exp "congenital, hereditary, and neonatal diseases and abnormalities"/ or exp infant/ or adolescent/ or exp pediatrics/ or child, abandoned/ or exp child, exceptional/ or child, orphaned/ or child, unwanted/ or minors/ or (pediatric* or paediatric* or child* or newborn* or congenital* or infan* or baby or babies or neonat* or pre-term or preterm* or premature birth* or NICU or preschool* or pre-school* or kindergarten* or kindergarden* or elementary school* or nursery school* or (day care* not adult*) or schoolchild* or toddler* or boy or boys or girl* or middle school* or pubescen* or juvenile* or teen* or youth* or high school* or adolesc* or pre-pubesc* or prepubesc*).mp. or (child* or adolesc* or pediat* or paediat*).jn 2.2.2 Study eligibility The selection criteria were as follows: (a) the study was an original research article published in a peer-reviewed journal in English; (b) the population consisted of children from birth to the age of 12, including prenatal; (c) the study used quantitative methods of reporting mental health and development (e.g., birth outcome, disease classification, validated scales, or  15 school achievement); and (d) environmental exposure included an NE measurement. Case studies and qualitative studies were excluded. Birth outcomes were included among the eligible outcomes, considering its predictive capacity of childhood development (Suzuki, 2018). This is one of the more common childhood health outcomes studied in relation to NE and the inclusion of these studies provided more strength to the assessment of relative associations depending on NE measurement. The selection of studies to be included was determined by my supervisor (MvdB) and I. Disagreements was resolved by discussion until consensus. For details on inclusion and exclusion criteria, see Appendix A, Section 2.  2.2.3 Data extraction Following similar approaches to previous reviews (e.g., Gascon et al., 2015), each study was reviewed for a number of characteristics and pertinent information was extracted. Extracted information included author(s), title, publishing journal, year of publication, study design, study country, population, sample size, NE measurement (metric and rate), mental health and development outcome (measure and collection method), main results (including effect sizes), statistical methods, confounding factors/covariates, reported sources of bias, reported strengths and limitations, and other pertinent information (Appendix B, Table B2). Initially, I worked independently to extract this information. Subsequently, a co-author (MG, IJ, MJ, LN, TO, HS, JS, or MvdB) reread the articles and reviewed the extracted data for control and accuracy.  2.2.4 Quality assessment and classification of evidence Many different approaches have been suggested for reaching a composite score of study quality in systematic reviews (Russell et al., 2009). The National Institute of Health (2018) has suggested using a combination of a numerical score and an expert overall appraisal to determine the quality of selected studies. The quality assessment criteria to derive a numerical score per study were adapted from a previous systematic review (Gascon et al., 2015) and included quality of remote sensing product (e.g., spatial and temporal resolution), expert or lay-person NE evaluation, quality of health outcome measure (e.g., self-reported or objective test or diagnose), study design, and risk of bias (Appendix B, Table B2). The numerical score was used to support the expert overall appraisal. Thus, no specific cut-off was used for the quality scores but relied on a combined assessment for the final rating. Additionally, criteria that were not applicable for the study were not considered, as to not double count or penalize studies for different study designs and use of NE measurements.  16 This means that each study had a different number of possible maximum points, ranging from 21 to 25 points. This method resulted in a final three-grade quality assessment of the study – poor, moderate, or good. I initially conducted the quality assessment, and a second reviewer (either MG, IJ, MJ, LN, TO, HS, JS, or MvdB) completed the same assessment independently. For each study, the reviewers provided a numerical score for each criterion item and an expert overall appraisal (Appendix B, Table B3). A final decision of the quality score was made after iterative reviews and discussion between me, MvdB and the other reviewer until consensus and established the degree of evidence for relationships between separate measurements of NE and childhood mental health and development. The combination of numeric and expert scores allowed for studies with similar numeric scores to receive different final quality score. Similar to previous reviews (e.g., Gascon et al., 2015), an adapted version of the International Agency for Research on Cancer (IARC) definitions to describe causal relationships was used to create categories for the level of evidence for an association between the selected health outcomes and NE exposure (International Agency for Research on Cancer, 2019). While acknowledging that most of the studies included in this review would not allow for establishing causal relationships, the following evidence categories for commonly observed association between exposure and outcome were included – sufficient, limited, insufficient, and lack of evidence. Sufficient evidence indicates significant relationships were observed between NE and improved health outcome in at least three studies, and at least half of those studies were of moderate or good quality. Limited evidence indicates that significant relationships were observed in studies of varying quality, but more studies are required to rule out bias and confounding. Insufficient evidence suggests that there is a lack of good quality studies, lack of replication, and/or lack of statistical power, which prevents a conclusion of evidence to be made. Finally, lack of evidence is defined by several good quality studies which indicate no association between outcome and exposure.  2.2.5 Pooled analysis of effect size Where applicable, the effect sizes of the studies were pooled. Pooling only occurred if three or more studies utilized exactly the same method of quantifying NE exposure and assessed the same health outcome. Effect pooling was conducted by using fixed effect models in the meta package in R (Balduzzi et al., 2019).  17 2.3 Results 2,050 articles were identified in EMBASE, 2,769 articles in MEDLINE, 6,808 articles in Web of Science and 1,364 articles in PsycINFO. 2,989 duplicates were identified and removed. After screening titles, 117 articles were selected for an abstract search. Additionally, 56 articles were identified from other sources and passed the title screen. After evaluating the abstract of each paper and 65 were selected for full-text screening. The bibliography search yielded eight additional papers. During the full-text screening, 18 studies were eliminated. Thus, 55 articles were included in the final review (Figure 2-2). Table B4 in Appendix B presents the extracted data from each paper, including study design, population, sample size, NE metric and rate, health outcome, and main results and effect sizes.  A majority of the studies had a cross-sectional design (n = 46), of these, ten had an ecological design. Six studies were longitudinal, two had a pseudo-experimental design, and one had a case-control design (Appendix B, Table B4).   Figure 2-2: PRISMA diagram of the procedure for article selection. The 55 studies were conducted in 12 different countries. A majority of the studies were conducted in the United States (n = 26). The other studies were conducted in Spain (n = 6), the  18 United Kingdom (n = 5), Canada (n = 3), Germany (n = 3), New Zealand (n = 3), Australia (n = 2), Lithuania (n = 2), Sweden (n = 2), Denmark (n = 1), France (n = 1), and Israel (n = 1). Most of the studies were published after 2009 (n = 50) and only five studies were published prior to 2004. The sample sizes ranged from 17 (Faber Taylor & Kuo, 2009; Wells, 2000) to 3,026,603 (Cusack et al., 2017b). The sample sizes in studies of ecological design ranged between 11 schools (Mårtensson et al., 2009) and 1,772 schools (MacNaughton et al., 2017). 29 studies received a composite score of poor quality; 20 of moderate quality, and six of good quality (Appendix B, Table B4).  2.3.1 Natural environment measurements Multiple types of measurements were used to evaluate NEs, including remote sensing products such as NDVI and LULC datasets, information collected via surveys or by experts, exposures assigned to participants by experts, and participants’ use of NEs (Table 2-3). Table 2-3 describes the number of times a NE metric was used regardless of significant findings. NDVI was the most common metric used (n = 33). In particular, NDVI derived from 30 m Landsat was the most common (n = 17), followed by NDVI from 250 m MODIS (n = 7). LULC layers were the second most common (n = 19); thirteen of these datasets had high accuracy (> 80%) and were derived from moderate to high-resolution imagery (5 m to 30 m imagery). Six studies had moderate accuracy (60-80%). When the accuracy level of the LULC data was not reported in the article, it was tracked the original source.             19 Table 2-3: NE metric categories and rates included in the studies. This count refers to the metrics which were used to assess NEs in each of the studies, regardless of whether or not they were associated with significant findings. See Table 2-1 for details of metrics and rates. Metric   Rate  NDVI (n = 33) • NDVI summarized (e.g., mean, median) within specified area around residence (e.g., buffer or polygon) (n = 27). Buffer sizes varying between 50 m and 2000 m  • NDVI as a proxy for exposure during commute (n = 1) • Exposure to road adjacent trees (n = 3) LULC (n = 32) • Proportion of classes considered NE within a specified area around a residence (e.g., buffer or polygon) (n = 20) • Distance to nearest NE from residence (continuous or discrete) (n = 11) •  Landscape distribution of NE patches (n = 1) Survey (n = 5) • Parent or guardian reported amount of “greenness” of surroundings (n = 1) • Parent or guardian reported “quality” of NE around residence (n = 4) Expert Measures  (n = 6) • On-site measurements of NE around address (e.g., surveys conducted by professionals, OPEC*, sky-view index) (n =3) • Expert assignments of participant exposure to pre-selected environment (e.g., a walk through an urban or natural environment pre- and post- health assessment) (n = 3) Use (n = 6) • Survey on child’s use of NE completed by parents (n = 5) • Time child spent in NE measured by GPS tracts (n = 1) * OPEC: Outdoor Play Environment Categories  2.3.2 Natural environment measures and relation to health outcomes Positive associations between NE and childhood mental health and development were found in 76 cases, inverse association in 11 cases, and no significant association in three cases (Table 2-4). In many cases, multiple rates were used for the same NE metric (Table 3; Appendix B, Table B4). As such, they were considered separately in this analysis. The numbers do not sum up to the number of studies included (n = 55) because outcomes were evaluated separately, and some studies analyzed more than one health outcome. Four categories of health outcomes were created: birth outcomes (n = 19), academic achievement (n = 8), mental disorders (n = 3) and cognitive development, attention and social functioning (CDAS) (n = 25). NE measurements were highly variable across studies and health outcomes (Table 2-3). The CDAS category was the most heterogeneous health group and had the greatest variety of NE measurements, while the other health outcome categories used only objective metrics derived from remote sensing products (Table 2-4). When available, effect sizes for the respective health outcomes are reported in Table B4 in Appendix B.     20 Table 2-4: Number of NE metrics used for each health outcome. Note: for studies that used multiple NE metrics, each significant finding was counted separately, therefore, the total n exceeds number of studies evaluated.   Birth Outcome Academic Achievement Mental Disorders CDAS Sum NDVI 55 31 2 37 125 LULC 11 18 4 31 64 Survey 0 0 0 11 11 Expert measures 0 0 0 8 8 Use 0 0 0 25 25 2.3.3 Birth outcomes Nineteen studies evaluated some aspect of neonatal health, such as birth weight, gestational age, infant mortality, and head circumference (Table 2-5; Appendix B, Table B4). All of these used data collected from hospital or vital records. A majority of studies (n = 15) showed positive association between NE exposure during pregnancy and birth weight and gestational age. The most common NE measurements were NDVI from Landsat (within 50, 100, 250, 500, and 1000 m Euclidian buffers) and from MODIS (within 250, 500, 1000 and 2000 m Euclidian buffers) around residential addresses. Land cover datasets were common when evaluating distance to the nearest NE (within 300 or 500 m) and for evaluating the proportion of NEs within 500 m buffers or polygons determined by municipal boundaries. Three studies suggested an inverse association between NE exposure and birth weight (Dadvand et al., 2012a; Grazuleviciene et al., 2015) and gestational age (Abelt & McLafferty, 2017; Grazuleviciene et al., 2015). The NE measurements used in these three studies included NDVI (30 m Landsat) within buffers of 100, 250, and 500 m and assessments of access defined as NE (from LULC dataset) within 500 m from residence. Birth weight was the only objective outcome in our review that consistently used the same NE measurement: one interquartile range (IQR) increase in NDVI (30 m Landsat), allowing for comparison between studies (Figure 2-3) (Balduzzi et al., 2019; Borenstein et al., 2011; Cuijpers, 2016). For five studies (Agay-Shay et al., 2014; Dadvand et al., 2012a, 2014b; Laurent et al., 2013; Markevych et al., 2014a), the median effect size was taken for all the buffers and a pooled analysis was performed, resulting in a mean effect size of 10.46 g (CI: 7.59, 13.34) increase in birth weight per 1-IQR increase of NDVI. Additionally, a pooled analysis was performed on the 250 m buffer. The mean effect (4  21 combined studies: Agay-Shay et al., 2014; Dadvand et al., 2014b, 2012a; Markevych et al., 2014a) was an increase of 21.12g (CI: 16.01, 26.23) for an increase in 1-IQR NDVI within a 250 m buffer.   Figure 2-3: Summarized birth weight changes per 1-IQR increase in NDVI (30 m Landsat) within varying buffer sizes for five studies. Color of the horizontal lines represents individual studies, and solid vertical gray line represents the pooled effect size for the median of all effect sizes across buffers; dotted vertical gray line represents the confidence interval.  The results from the compilation of studies on birth outcomes suggest greater effects are observed within larger buffer sizes. Glazer et al. (2018) examined the relationship between blue space and birth weight in Rhode Island, USA. They found a 7.4 g increase in birth weight when the mother lived within 500 m of a freshwater body, compared with mothers who lived further away. Kihal-Talantikite et al. (2013) found that an interaction effect of socioeconomic deprivation and NE, as measured by a LULC dataset, was associated with an increased risk of infant mortality in a cluster of neighborhoods of Lyon, France (Table 2-5; Appendix B, Table B4). Head circumference was only examined by a single study (Dadvand et al., 2012b) (Table 2-5). The results suggested an association between a 1-IQR increase of NDVI (30 m Landsat), and an increase in head circumference of 1.2 mm, 1.4 mm, and 1.7 mm within a 100 m, 250 m, and 500 m buffer, respectively.   22 Level of evidence: Based on the compiled assessment of the quality of the studies (Table 2-5; Appendix B, Table B4), the evidence of association between NEs, as measured by NDVI (30 m Landsat) or LULC datasets, and birth weight and gestational age was classified as sufficient (International Agency for Research on Cancer, 2019) (Table 2-7). Since only one study analyzed the association to blue space, this evidence was classified as insufficient (Table 2-7). The evidence of association between NE exposure and excess infant mortality was classified as insufficient due to the lack of other studies for comparison. Finally, due to the lack of similar studies, the evidence of association between NE exposure and head circumference was classified as insufficient (Table 2-7). 2.3.4 Academic achievement and absenteeism Academic achievement was analyzed by the success of students (ages 5-18) on standardized tests or school attendance (Table 2-5; Appendix B, Table B4) (n = 8). Most students were aged 8-9 (grade three) (n = 5). All studies assessed NE at the school, either by estimating the spatial rate of NEs within school property or within the neighborhood that the school served. Results for academic achievement were mixed (Table 2-5) – five studies showed positive associations between NE and various measures of academic achievement (Hodson & Sander, 2017; Kuo et al., 2018; Kweon et al., 2017; Sivarajah et al., 2018; Wu et al., 2014), two studies showed inverse associations between NE and the outcomes (Beere and Kingham, 2017; Browning et al., 2018). One study analyzed the association between chronic absenteeism and school surrounding NE (MacNaughton et al., 2017). They found a 2.6% decrease in absenteeism with each 1-IQR increase of NDVI (250 m MODIS). Level of evidence: Given the ecological study design of these articles and the mixed results, the evidence between NE, as measured by NDVI and LULC datasets, and academic achievement was classified as limited. Due to the lack of similar studies, the level of evidence for an association between NE as measured by NDVI and absenteeism was classified as insufficient (Table 2-7). 2.3.5 Diagnosed mental health and developmental disorders Three studies analyzed prevalence of doctor-diagnosed mental disorders: schizophrenia (Engemann et al., 2018), ADHD (Markevych et al., 2018), and autism (Wu & Jackson, 2017) (Table 2-5; Appendix B, Table B4). All three studies found that NE correlated with lower disease prevalence, while using different NE measurements. Engemann et al. (2018) and Markevych et  23 al. (2018) used NDVI, derived from 30 m Landsat and 250 m MODIS respectively. Engemann et al. (2018) found a higher risk of schizophrenia onset in Danish neighborhoods with the fewest NEs compared to neighborhoods with the most NE. For every 0.1 increase in NDVI, Markevych et al. (2018) found a decrease in the rate of ADHD diagnosis in Germany. Wu and Jackson (2017) utilized a LULC dataset and found a decrease in autism prevalence with increase in proportion of NE within school districts in California, USA.  Level of evidence: Due to the limited number of studies, we classified the evidence for an association between NE exposure and each of the mental health and developmental disorders as insufficient (Table 2-7).  24 Table 2-5: Associations between NE and outcome for each study. ↑ indicates positive associations between NE exposure and health outcome, i.e., NE improves health; ↓indicates an inverse association between NE exposure and health outcome, i.e., NE worsens health; ↔ indicates no significant association between NE exposure and health outcome.  Health Outcome Article Metric Sensor/Source Rate Direction of association Birth outcomes Birth weight Agay-Shay et al., 2014 NDVI Landsat 250 m ↑  Dadvand et al., 2014b NDVI Landsat 50 m, 100 m, 250 m, 500 m ↑  Dadvand et al., 2012a NDVI Landsat 500 m              ↓  Dadvand et al., 2012b NDVI Landsat 100 m, 250 m, 500 m ↑  Grazuleviciene et al., 2015 NDVI Landsat 500 m              ↓  Hystad et al., 2014 NDVI Landsat 100 m ↑  Laurent et al., 2013 NDVI Landsat 50 m ↑  Cusack et al., 2017b NDVI Landsat 50 m, 250 m, 1000 m ↑  Markevych et al., 2014 NDVI Landsat 100 m, 250 m, 500 m, 800 m ↑  Dadvand et al., 2012a LULC Landsat 500 m ↑           Cusack et al., 2017a NDVI MODIS 250 m ↑  Fong et al., 2018 NDVI MODIS 250 m ↑  Dadvand et al., 2014 NDVI MODIS (Adjacent trees) 200 m           ↔  Cusack et al., 2017b LULC Adjacent trees 26 m  ↑  Cusack et al., 2017b LULC Proportion of NE 1000 m ↑  Ebisu et al., 2016 LULC Proportion of NE 250 m ↑  Agay-Shay et al., 2014 LULC Distance to NE 300 m ↑  Glazer et al., 2018 LULC Distance to NE (water) 500 m ↑  Hystad et al., 2014 LULC Distance to NE 300 m ↑ Gestational age Abelt and McLafferty, 2017 NDVI Landsat 250 m, 500 m              ↓  Dadvand et al., 2012 NDVI Landsat 100 m ↑  Grazuleviciene et al., 2015 NDVI Landsat 500 m              ↓  Grazuleviciene et al., 2015 NDVI Landsat 500 m              ↓  Hystad et al., 2014 NDVI  Landsat 100 m ↑  Hystad et al., 2014 NDVI  Landsat 100 m ↑  25 Health Outcome Article Metric Sensor/Source Rate Direction of association  Casey et al., 2016 NDVI MODIS 250, 1250 m ↑  Fong et al., 2018 NDVI MODIS 250 m ↑  Donovan et al., 2011 LULC Proportion of NE 50 m ↑  Ebisu et al., 2016 LULC Proportion of NE 250 m ↑  Nichani et al., 2017 LULC Proportion of NE Polygon ↑  Dadvand et al., 2012 LULC Distance to NE 500 m              ↓ Infant mortality Kihal-Talantikite et al., 2013 LULC Proportion of NE Polygon              ↓ Head circumference Dadvand et al., 2012b NDVI Landsat 100 m, 250 m, 500 m ↑ Academic achievement Standardized tests Browning et al., 2018 NDVI MODIS 250 m, 500 m, 1000 m, 2000 m              ↓  Wu et al., 2014 NDVI MODIS 250 m, 500 m, 1000 m, 2000 m ↑  Beere and Kingham, 2017 LULC Proportion of NE School ground, school zone              ↓  Hodson and Sander, 2017 LULC Proportion of NE Polygon, attendance zones ↑  Kuo et al., 2018 LULC Proportion of NE Polygon, school zone ↑  Kweon et al., 2017 LULC Proportion of NE Polygon, school boundary ↑  Sivarajah et al., 2018 LULC Proportion of NE Polygon, school boundary ↑ Absenteeism MacNaughton et al. NDVI MODIS 250 m ↑ Mental health and development disorders Schizophrenia Engemann et al., 2018 NDVI Landsat 210 m ↑ ADHD Markevych et al., 2018 NDVI MODIS Polygon ↑ Autism Wu and Jackson, 2017 LULC Proportion of NE Polygon, school boundary ↑ Cognitive development, attention, and social functioning (CDAS) General behavior Amoly et al., 2014† Use of NE Parent/guardian survey Hours/year ↑  Amoly et al., 2014† NDVI Landsat 100 m, 250 m, 500 m ↑  Balseviciene et al., 2014 NDVI Landsat 300 m ↑  McEachan et al., 2018 NDVI Landsat 100 m, 300 m, 500 m ↑  Feng and Astell-Burt, 2017 LULC Proportion of NE polygon ↑  Balseviciene et al., 2014 LULC Distance to NE Address-to-NE-distance ↑  Markevych et al., 2014b LULC Distance to NE 500 m              ↓  26 Health Outcome Article Metric Sensor/Source Rate Direction of association  Feng and Astell-Burt, 2017 Quality of NE Survey NA ↑  Mårtensson et al., 2009† Expert Measure On-site  NA ↑  Flouri et al., 2014† Use of NE Parent/guardian survey Day/week ↑  Richardson et al., 2017† Use of NE Parent/guardian survey Private garden ↑ Well-being Amoly et al., 2014† NDVI Landsat 500 m ↑  Kim et al., 2016 LULC Proportion of NE 400 m ↑  Kim et al., 2016 LULC Patches 400 m, 800 m ↑  Larson et al., 2018 LULC Proportion of NE Polygon ↑   Tillmann et al., 2018 LULC Proportion of NE (park) 500 m ↑  Tillmann et al., 2018 LULC Proportion of NE (water, grass) 500 m              ↓  Christian et al., 2017 LULC Distance to NE Address-to-NE-distance              ↓  Kim et al., 2016 LULC Distance to NE 400 m, 800 m ↑  Söderström et al., 2013 Expert Measure On-site NA             ↔  Wells and Evans, 2003 Expert Measure On-site NA ↑  Amoly et al., 2014† Use of NE Survey Hours/year ↑  McCracken et al., 2016† Use of NE Parent/guardian survey Day/week ↑  Ward et al., 2016† Use of NE GPS For 7 days ↑ ADHD/ADD Amoly et al., 2014† NDVI  Landsat 100 m ↑  Dadvand et al., 2017 NDVI Landsat 100 m, 300 m, 500 m ↑  Dadvand et al., 2015 NDVI RapidEye 250 m ↑  Faber Taylor and Kuo, 2011† Survey Parent/guardian survey NA ↑  Faber Taylor et al., 2001 Survey Parent/guardian survey NA ↑  Faber Taylor et al., 2002 Survey Parent/guardian survey NA ↑  Kuo and Faber Taylor, 2004 Survey Parent/guardian survey NA             ↔  Faber Taylor and Kuo, 2009† Expert measure Assigned walk  20 min walk ↑  Mårtensson et al., 2009† Expert Measure On-site  NA ↑  Wells, 2000 Expert Measure On-site NA ↑ Social functioning Amoly et al., 2014† NDVI Landsat 250 m ↑  Balseviciene et al., 2014 NDVI Landsat 300 m              ↓  27 Health Outcome Article Metric Sensor/Source Rate Direction of association  Balseviciene et al., 2014 LULC Distance to NE Address-to-NE-distance ↑  Richardson et al., 2017† LULC Proportion of NE 500 m ↑  Tillmann et al., 2018 LULC Proportion of NE 500 m ↑  Christian et al., 2017 LULC Distance to NE Address-to-NE-distance ↑  Wells and Evans, 2003 Expert Measure On-site NA ↑  Amoly et al., 2014† Use of NE Survey Hours/year ↑  Flouri et al., 2014† Use of NE Parent/guardian survey Days/week ↑  McCracken et al., 2016† Use of NE Parent/guardian survey Days/week ↑  Richardson et al., 2017† Use of NE Parent/guardian survey Private garden ↑ Memory Dadvand et al., 2015 NDVI RapidEye 100 m  ↑  Dadvand et al., 2015 NDVI RapidEye  50 m  ↑  Schutte et al., 2017† Expert measure Assigned walk 20 min walk ↑   Note: The spatial resolutions of the sensors are as follows: Landsat, 30 m; MODIS, 250 m; and RapidEye, 5 m.  † Indicates that the study assessed temporal rate, or the amount of time spent in NEs. For more information, see Appendix A, Table A3.  28  2.3.6 Cognitive development, attention, and social functioning (CDAS) The CDAS category included a wide range of health outcomes and was therefore further divided into five subcategories (Table 2-6): (1) self-reported ADHD/ADD symptoms (in contrast to doctor-diagnosed); (2) memory function; (3) general behavioral outcomes; (4) social functioning, and (5) childhood development and well-being. For details of which tools and scales were used to assess respective outcomes, see Appendix B, Table B4. Due to the diversity of tests, specific tests and subscales were evaluated separately (Table 2-6).                          29  Table 2-6: Specific tests and subscales used in each CDAS category. CDAS Category Health outcome (Tool) ADHD/ADD symptoms • ADHD severity (ADHD/DSM IV, survey) • Inattention (ADHD/DSM IV, K-CPT and HRT-SE response time) • Concentration • Impulse inhibition • Delay of gratification • Self-discipline • Directed attention capacity Memory • Spatial working memory • Working memory (n back test) • Superior working memory (n back test) General behavior – internalizing and externalizing • SDQ difficulties (SDQ) • Hyperactivity (SDQ) • Internalizing difficulties (SDQ) • Externalizing difficulties (SDQ) • School functioning (HRQOL) • Aggressive behavior Social functioning • Social competence (AEDC) • Peer relationship problem (SDQ) • Pro-social behavior (SDQ) • Social functioning (HRQOL) • Psychosocial functioning (HRQOL) • Psychosocial health (Lewis life) • Friends (Kid KINDL) Development and well-being • Physical health and well-being (AEDC) • Total HRQOL • Emotional functioning (HRQOL) • Life satisfaction • Happiness • Well-being index • Global self-worth • Anxiety (autism) • Self-esteem (Kid KINDL) Abbreviations: ADHD/DSM IV = Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition  AEDC = Australian Early Development Census HRQOL = health-related quality of life HRT-SE = hit-reaction time – standard error K-CPT = Conners Kiddie Continuous Performance test SDQ = strengths and difficulties questionnaire    30  2.3.7 ADHD/ADD symptoms Ten studies analyzed symptoms of attention (ADHD/ADD and inattention) (Table 2-5; Appendix B, Table B4). Three studies used NDVI (30 m Landsat) within 100, 250, 300, and 500 m buffers around residential or school addresses (Amoly et al., 2014; Dadvand et al., 2015, 2017) and all found an inverse association to symptoms of the condition (i.e., fewer symptoms in areas with more NEs). Amoly et al. (2014) found an approximately 6% decrease in symptoms with 1-IQR increase in NDVI. Dadvand et al. (2017) found a decrease in omission errors and lower reaction time standard error (consistent with better attention) with increased NE exposure. The same research group found a decrease in inattention with a 1-IQR increase in school yard NE (Dadvand et al., 2015).  Surveys of NE exposure, as reported by parents, guardians or teachers, were used in four studies, all conducted by the same research group (Faber Taylor et al., 2001, 2002; Faber Taylor & Kuo, 2011; Kuo & Faber Taylor, 2004) (Appendix B, Table B4). In all cases, they found inverse associations to symptoms of the condition. Two studies analyzed self-reported ADHD symptoms (rather than ADHD diagnosis). Faber Taylor et al. (2001) found that children who played in areas with higher amounts of NEs had fewer symptoms than children who played indoors or in built environments (Table 2-5). Similarly, children had more severe ADHD symptoms when playing “deep indoors” compared to “open grass” or “big trees and grass” (Faber Taylor & Kuo, 2011) (Table 2-5). Kuo and Faber Taylor (2004) used parent-reported survey data to assess ADHD symptoms in after-school or weekend activities and found no significant association to NE (Table 2-5). Finally, Faber Taylor et al. (2002) found that girls who had more views of NE from their residence had better concentration, impulse inhibition, and self-discipline compared to those with more limited views (Table 2-5). Three studies quantified NE exposure by expert measures (Faber Taylor & Kuo, 2009; Mårtensson et al., 2009; Wells, 2000). These measures included an assessment of the quality of preschool play areas (quality defined by an expert tool, assigned walks, and surveys complete by experts in participant’s homes). Faber Taylor and Kuo (2009) found participants performed better on a puzzle task after they had gone for a 20-minute walk in a park compared to after they had taken a walk in a downtown area (Table 2-5; Appendix B, Table B4). Mårtensson et al. (2009) used the Outdoor Play Environment Categories (OPEC) and the amount of visible sky to determine the amount of NEs on school properties and found a  31  decrease in inattention for children in schools with higher OPEC ratings (Table 2-5; Appendix B, Table B4). Lastly, Wells (2000) followed children who moved from a neighborhood with less NE to one with more and found the children’s directed attention capacity increased after the move.  Level of evidence: Due to the similar research designs, consistent findings, and quality of the studies, we classified the evidence of association between decreased ADHD/ADD symptoms and NDVI (30 m Landsat) as sufficient (Table 2-7). However, the evidence of association for expert measures and surveys of NE was found to be limited due to the quality and heterogeneity of the studies (Table 2-7).  2.3.8 Memory Two papers analyzed associations between memory function and NE exposure (Dadvand et al., 2015; Schutte et al., 2017) (Table 2-5; Appendix B, Table B4). Dadvand et al. (2015) used NDVI (30 m Landsat) to quantify exposure over a year and observed an increase in working memory and superior working within the schools with more NE compared with schools with less NE.  Schutte et al. (2017) assigned 20-minute walks in different environments prior to and after completing a memory test and found children had more accurate spatial working memory after a walk in nature compared to a walk in an urban environment (Table 2-5; Appendix B, Table B4).  Level of evidence: Due to the differences in NE measurements and limited number of studies, there is insufficient evidence in regards to the association between NE exposure and memory (Table 2-7).  2.3.9 General Measures of Behavior – Internalizing and externalizing behaviors General behavior was assessed in eight studies (Amoly et al., 2014; Balseviciene et al., 2014; Feng & Astell-Burt, 2017; Flouri et al., 2014; Markevych et al., 2014b; Mårtensson et al., 2009; McEachan et al., 2018; Richardson et al., 2017) (Table 2-5; Appendix B, Table B4), applying three measurements: use of NE, NDVI (30 m Landsat), LULC datasets, and expert measures. Three studies assessed use of NE through surveys to identify the amount of time (e.g., hours/year, days/week) children spent in NEs (Amoly et al., 2014; Flouri et al., 2014; Richardson et al., 2017) and all three found decreased behavioral problems with increased NE exposure. Two studies (Amoly et al., 2014; McEachan et al., 2018) used the total difficulties score from the Goodman’s  32  strengths and difficulties (SDQ) questionnaire and NDVI (30 m Landsat) within buffers of 100 m, 250 m, 300 m, and 500 m of the residential address to quantify exposure to NEs, identifying a positive association to behavior. Finally, three studies (Balseviciene et al., 2014; Feng & Astell-Burt, 2017; Markevych et al., 2014b) used LULC layers to determine the relative distance from either residential address or census block of residence to the nearest NE or whether the NE existed within a municipal boundary and consistently found positive associations between NE exposure and improved behavior measurement. Mårtensson et al. (2009) found a decrease in hyperactivity of children whose play areas had a higher OPEC score.  Level of evidence: Following assessment of study quality, the evidence of association between use of NE (determined by parent surveys) was classified as limited due to the small number of studies. The evidence of association between NE assessed by NDVI (100 m, 250 m, and 500 m buffer zones), LULC datasets, and expert measures was classified as insufficient due to the quality of the studies and lack of replication (Table 2-7).  2.3.10 Social Functioning Eight studies analyzed various aspects of social functioning (Amoly et al., 2014; Balseviciene et al., 2014; Christian et al., 2017; Flouri et al., 2014; McCracken et al., 2016; Richardson et al., 2017; Tillmann et al., 2018; Wells & Evans, 2003) (Table 2-5; Appendix B, Table B4). Seven of these demonstrated positive associations between social functioning and NE exposure. LULC datasets were the most commonly used exposure measure, assessing spatial rate through the distance to the nearest NE from a residential address (Balseviciene et al., 2014; Christian et al., 2017; McCracken et al., 2016; Richardson et al., 2017; Tillmann et al., 2018) (Table 2-5; Appendix B, Table B4). Tillmann et al. (2018) found inverse associations between social outcomes and spatial rate, as determined by proportion of NE within 500 m of a residential address. Parent reported use of NE (e.g., hours/year, days/week) was included in three studies (Amoly et al., 2014; Flouri et al., 2014; Richardson et al., 2017). NDVI (Landsat 30 m) was used in two studies (Amoly et al., 2014; Balseviciene et al., 2014) within 250 and 300 m buffers. Finally, Wells and Evans (2003) quantified vegetation by assessing views from windows using surveys conducted by professionals.  Level of evidence: The level of evidence of association between NE exposure as measured by use of NE was classified as limited following the overall positive results and quality of the  33  studies (Table 2-7). Additionally, evidence of association for NE exposure determined by LULC was classified as limited. The evidence of association for NDVI (Landsat) and expert measures was classified as insufficient due to the quality and small number of studies (Table 2-7). 2.3.11 Development and well-being Eight studies assessed the association between NEs and childhood development and well-being (Amoly et al., 2014; Christian et al., 2017; Kim et al., 2016; Larson et al., 2018; McCracken et al., 2016; Tillmann et al., 2018; Ward et al., 2016; Wells & Evans, 2003) (Table 2-5). Four NE measurements were used: use of NE as defined by parent survey or GPS tracking (Amoly et al., 2014; McCracken et al., 2016; Ward et al., 2016); LULC datasets for assessing the relative distance to a park and for determining the proportion of NE within areas surrounding a residence (Christian et al., 2017; Kim et al., 2016; Larson et al., 2018; Tillmann et al., 2018); NDVI (30 m Landsat) (Amoly et al., 2014); and expert evaluations of the amount of vegetation visible from inside participants’ homes (Wells & Evans, 2003). Six studies found positive associations between NE exposure and outcome, while two studies, both using LULC datasets to quantify the proportion of land cover types, found negative associations (Larson et al., 2018; Tillmann et al., 2018).  Level of evidence: The evidence of association between NE exposure and childhood development and well-being was classified as insufficient due to the varying quality of studies and the heterogeneity in NE measurements used (Table 2-7).              34  Table 2-7: Summarized results of the associations between health outcomes and NE metrics. Outcome NE metric Evidence of association Birth weight NDVI (Landsat), LULC Sufficient  Distance to blue space (LULC) Insufficient Gestational age NDVI (Landsat), LULC Sufficient Excess infant mortality LULC Insufficient Head circumference NDVI (Landsat) Insufficient Academic achievement NDVI (MODIS), LULC Limited Absenteeism NDVI (MODIS) Insufficient Prevalence of mental disorders (ADHD diagnosis, schizophrenia, autism) NDVI (Landsat, MODIS), LULC Insufficient Attention (ADHD symptoms) NDVI (Landsat) Sufficient  Expert measures, surveys Limited Memory NDVI (Landsat), expert measure Insufficient General behavior Use of NE Limited  NDVI (Landsat), LULC, expert measures Insufficient Social functioning LULC, use of NE  Limited  NDVI (Landsat), expert measures Insufficient Well-being NDVI, LULC, use of NE, expert measures Insufficient  2.4 Discussion  The results from this chapter suggest that the associations between childhood health outcomes and NEs depend on choice of NE measurement to some extent. There was sufficient evidence of improved birth weight and gestational age by NE exposures, as measured by NDVI (30 m Landsat) and LULC datasets within buffers between 50 and 1000 m. Additionally, there was sufficient evidence for an association to self-reported ADHD/ADD symptoms in studies applying NDVI (30 m Landsat). There was limited evidence for a relationship between academic achievement and NE, as measured by NDVI (both 30 m Landsat and 250 m MODIS derived) and LULC datasets. There was limited evidence between improved general behavior and social functioning and use of NE; additionally, there was limited evidence of association between social functioning and LULC datasets. Limited evidence of association also existed between self-reported ADHD symptoms and expert measures and surveys. Insufficient evidence exists to establish relationships between NE measurements and the prevalence of mental disorders, memory function, and childhood development and well-being (Table 2-7).  The metrics most consistently associated birth outcomes and self-reported ADHD/ADD symptoms were Landsat-derived NDVI (30 m resolution) (n = 33) and LULC datasets (n = 32).  35  However, it is possible that this relationship appears significant due to the common use of these metrics and not that NDVI and LULC datasets are the most optimum measurement for assessing exposure to NE. Measurements based on surveys (n = 5) and expert measures (n = 6), for example, are less common and use a variety of different scales and tools. Therefore, it is difficult to determine whether association to a given health outcome is consistent across studies or not. This does not necessarily mean that surveys and expert measures would be less useful or appropriate, rather, that efforts to improve such measurements for increased application and consistency should be considered. Similarly, measures of NE use, which also had limited evidence, may be biased by the relatively small number of studies using this metric (n = 6). The goal of assessing the level of evidence between NE metric and health outcome was to identify which NE metrics may be the most associated with childhood health outcomes, with the possibility of providing recommendations for future studies. However, given the inconsistencies in associations and level of evidence between different health outcomes, it was not possible to say which NE metric “best” represents NE exposure for children.  For example, there was sufficient evidence for an association between birth weight and NEs as measured by NDVI (30 m Landsat), but the evidence for an association between the same metric and head circumference, prevalence of mental disorders, memory, general behavior, and social functioning was insufficient. This may be due to the lower number of studies analyzing these outcomes. It is also possible that certain metrics are more appropriate for measuring NE exposure than others depending on health outcome and proposed pathway. For example, birth outcomes could be influenced by general NE exposure during pregnancy, for instance through heat reduction or air pollution (Carolan-Olah & Frankowska, 2014; Shah & Balkhair, 2011) and this could arguably be adequately measured by NDVI. On the other hand, outcomes such as social functioning may be more related to NE, as measured by metrics of perception or experience through surveys or on-site evaluations.  Additionally, this review identified a large number of different scales or indicators for the various health outcomes. Some health outcomes, such as birth weight, were assessed by similar measurements across studies, possibly partly explaining the higher consistency in findings for this outcome. However, in the majority of cases, even though scales of high validity and reliability were used, heterogeneous measures were applied across studies and  36  thus made the assessment of consistency and evidence level less straight-forward. This suggests that a better understanding of NE measurements and their relationship to health can be achieved through increased collaboration between researchers and replication of studies using the same measurements. In regard to spatial rate, the most frequently used buffer sizes in studies using NDVI metrics were 100 m, 250 m, and 500 m. Associations to various health outcomes were, in general, consistent across buffer sizes, although for birth outcomes the association seemed to be most consistently found in larger buffer sizes. In studies using LULC datasets, spatial rate was determined by polygons, buffers, or distance from residential addresses to nearest NE. No conclusions can be drawn regarding LULC spatial rate indicators due to the large inconsistencies and relatively small number of studies per each rate. Eight studies evaluated the time spent in NE (temporal rate) by parent/guardian surveys or mobile devices, but at this point, it is impossible to determine an “optimal” time-frame for NE exposure in relation to health.  In general, the findings of this chapter for an association between NE exposure and childhood development and mental health are consistent with previous reviews (Browning & Rigolon, 2019; Dzhambov et al., 2014; McCormick, 2017). For instance, Dzhambov et al. (2014) suggest a weak positive association between birth weight, which was also observed. Additionally, the results suggested improved mental well-being, overall health, and cognitive development with greater exposure to NE, which was also described by McCormick (2017). Only Browning and Rigolon (2019) also explored the difference in association based on NE metric, however, most of the positive findings they highlight are for college-level examination and not for elementary and middle school aged children. While other reviews note the heterogeneity of NE metrics and health outcomes, this review builds upon the past literature and identifies the NE metrics that show the strongest relationship to different childhood health outcomes. While this review cannot provide a single recommendation, it is a first step to understanding optimal measurement strategies for childhood health. Additionally, NDVI and LULC were identified as the most common metrics for childhood health and development research; this was also found in a recent review by Labib et al. (2020), that concluded LULC, NDVI, or a combination of the two, were the most common NE metrics in a wide range on health outcomes and age groups.   37  2.4.1 Strengths and limitations of included studies In general, mostly high-quality studies, but some moderate quality studies, made efforts to include more than one NE measurement, such as NDVI in combination with high accuracy LULC (e.g., Cusack et al., 2017b; Dadvand et al., 2012a; Markevych et al., 2014a, 2014b). These studies also considered temporal alignment (i.e., that the time of health outcome data collection coincided with the time of NE data collection) which prevents exposure misclassification due to changes in the environment over time (Helbich, 2019). In urbanizing areas with rapid development of commercial and residential areas, substantial land cover changes can occur over short periods of time (Aguilera et al., 2011).   A few studies included aspects of quality of NE (Feng & Astell-Burt, 2017; Kim et al., 2016; Mårtensson et al., 2009; McEachan et al., 2018; Söderström et al., 2013), including landscape play quality (Mårtensson et al., 2009), which would be important for assessing recreational and psychological benefits. The studies that included quality aspects suffered from other quality shortcomings (e.g., temporal misalignment of data, and poor control on confounding factors), which resulted in a final moderate quality. A common characteristic for studies of high quality was the accountancy of exposure assignment. In other words, that the NE exposure was linked to the child per accurate home address, rather than on aggregated neighborhood scale, another important aspect for avoiding exposure misclassification. Related to this is whether the exposure is assigned per the child’s activity space and use of NE. This was considered in one of the high-quality studies (Dadvand et al., 2015), though it was accounted for in several studies that got an overall score of moderate quality (Amoly et al., 2014; Flouri et al., 2014; Mårtensson et al., 2009; McEachan et al., 2018; Söderström et al., 2013). Other strengths that characterized some of the high-quality studies were appropriate control for a number of adequate confounders (Cusack et al., 2017a; Dadvand et al., 2012a, 2017; Markevych et al., 2014b), and application of longitudinal or case-control study designs (Dadvand et al., 2015, 2017; Richardson et al., 2017). A common limitation of the included studies was the risk of bias, for example, residential selection bias effects, mostly a consequence of cross-sectional study design without adjustment for neighborhood self-selection. This results in a risk that some of the positive associations may be a result of residential-choice processes so that neighborhoods with an abundance of NE tend to be more expensive, thus attracting residents of higher income and education and who already have healthy habits and lifestyles (Yu & Zhu, 2015). Other limitations include self-reported  38  outcome variables, limited control of confounders, lack of effect estimates, suboptimal statistical approaches (e.g., failure to control for spatial autocorrelation), and failure to account for the actual time period a child had resided in the environment. Finally, for some of the health categories, many studies were conducted by the same research group, possibly leading to bias.  2.5 Conclusions This review highlights the variety and inconsistencies of NE metrics currently used in childhood mental health and development research. Since NDVI, LULC, and use metrics were demonstrated to have the strongest evidence for association for childhood health outcomes, future studies should consider utilizing them for comparison sake, at the very least in sensitivity analysis. Additionally, it draws attention to the need for consistent collaborative work on this subject, if we are to fully understand and realize what factors contribute to the health and the importance of NE in urban areas in different regions of the world. Continued research in the NE-child health realm needs to be prioritized if we are to provide sufficient evidence for decisions around healthy urban planning for promoting healthy childhoods. 39  Chapter Three: Change in greenness or change in NDVI? 3.1 Introduction The availability of remote sensing data, both temporally and geographically, has made it appealing for use in epidemiological studies; consequently, becoming one of the most common NE metrics used. Population cohort data can be spatially and temporally linked to remotely sensed NE metrics through geo-based information (e.g., through residential address), thus enabling assessments of associations between health outcomes and NEs. However, remote sensing data processing can be intensive, and many considerations must be taken into account, such as how the data were collected, availability of data for a particular area, adjustment needed during preprocessing, and the added preparation necessary when following cohorts over time (Flouri et al., 2014).  NDVI has become a common choice for determining population-level NE exposures due to its ability to objectively measure exposure over large areas. Labib et al. (2020) identified it as one of the most common NE metrics used in epidemiological studies. In these studies, NDVI is generally considered an indicator of the amount of vegetation; so that higher NDVI values corresponds with more vegetation. Similarly, some studies suggest cut-offs for identifying types of NEs, such as grass, shrubs, or trees, using NDVI (e.g., Gascon et al., 2016; Weier & Herring, 2000). However, this can be misleading when considering that NDVI was originally developed as a method for evaluating photosynthetic production of landscapes (Deering & Rouse, 1975). Thus, these assumptions of amount of vegetation or type, may not accurately reflect vegetation in urban landscapes. A common issue when dealing with remote sensing datasets is the introduction of stochasticity to the data (Hird & McDermid, 2009) (i.e., noise) from, for example, the atmosphere, weather, and climate conditions during image collection, spectral bands of interest, and the choice of sensor (Schott et al., 2016). Efforts to reduce local areas of noise (e.g., errors caused by atmosphere) have primarily been through local and temporal averaging (Bradley et al., 2007; Hird & McDermid, 2009), as modeling NDVI reduces noise by removing outliers and smoothing trends in data (Hird & McDermid, 2009). This noise can obscure signals and influence the data, however, this has yet to be explored in epidemiological studies.   40  Ideally, both exposure and health data would be from the same year, ensuring the exposures adequately correspond to the outcome. However, this is not always possible given how data becomes available and how some datasets are created. For example, LULC data are generally considered accurate to a single year despite most datasets utilizing many years’ worth of data. A solution to this used by some studies has been to assume that the vegetation in the study area remained constant over time, particularly in order to ease the processing of data in longitudinal studies (e.g., Markevych et al., 2016). This issue is of increasing concern in rapidly developing and densifying cities as changes in vegetation cover, and consequently exposure to NE, happen in these areas (Government of Canada, 2015; Weng, 2007). Development takes place heterogeneously across urban landscapes and often the largest changes occur in expanding neighborhoods along the periphery of the metropolitan area (Hammer et al., 2004; Jin et al., 2019). Particularly for these areas, the temporal misalignment may lead to greater error in the correlation between NE and health outcomes.  3.1.1 Research gaps and objectives With the rapid development and densification taking place in cites, the temporal alignment of exposure and outcome datasets is important for assigning accurate exposures to NEs in epidemiological studies. However, it is unclear on the best approach to take in order to identify areas of change and determine if NEs remain the same across time. While methods have been developed for a pixel approach (Ji et al., 2006), there has been relatively little research on the buffered units commonly used in epidemiological research (Helbich, 2019). Using Metro Vancouver as a case study, this chapter aims to evaluate the change in NDVI between 1999 and 2014 with the goals of (i) examining NDVI at localized postal codes and at aggregated postal code geographies to determine which direction vegetation may have changed over time, and (ii) developing a method for analyzing change in NDVI over time. 3.2 Materials and methods 3.2.1 Study area This analysis was conducted in the Vancouver metropolitan area, British Columbia, Canada. Metro Vancouver is home to 2.3 million people and is the largest urban center in Western Canada (Metro Vancouver, 2019b). It covers 2,865 square kilometers and consists of 21 municipalities, one Electoral Area, and one Treaty First Nation (Metro Vancouver, 2019a). To the west, Metro Vancouver is bordered by the ocean (i.e., the Georgia Straight); the Coast  41  Mountains to the north; the municipal boundary of Abbotsford to the east; and the Canadian border with the USA in the south (Figure 3-1). Over the last 20 years, the area has undergone a substantial transformation through the development of agricultural land into residential subdivisions, and expansion and densification of urban centers (Wang et al., 2019).     Figure 3-1: Extent of Metro Vancouver with constituent municipalities. 3.2.2 NDVI data NDVI datasets available from the Canadian Urban Environmental Health Research Consortium (CANUE, https://canue.ca/) were used in this chapter. This curated dataset was created using Google Earth Engine (GEE) functions to generate cloud-free annual growing season Landsat composites and mask water for the study area (“CanMap Postal Code Suite v2015.3 [computer file],” 2015; “Landsat 5 TM Annual Greenest-Pixel TOA Reflectance Composite, 1984 to 2012,” 2017; “Landsat 8 Annual Greenest-Pixel TOA Reflectance Composite, 2013 to 2015,” 2017; “USGS Landsat 5 TM TOA Reflectance (Orthorectified), 1984 to 2011,” 2017; “USGS Landsat 8 TOA Reflectance (Orthorectified), 2013 to 2018,” 2017; Gorelick et al., 2017). Landsat 7 was not included due to the scan-line correction failure. Landsat 5 and 8 imagery were used to generate yearly data between 1999 and 2014, except 2012 where neither satellite was functioning (Markham et al., 2004).  42  Analysis was conducted at the 6-digit postal code level, which represents a small geographic area corresponding to the size of a half city block and comprises approximately 35 residents for a typical cohort (Khan, 2018). This is in alignment with most epidemiological studies where linkages to aggregated health outcome data usually occur at the postal code or similar level. All postal codes in Metro Vancouver (n = 59,381) were used in this analysis. Around each postal code, average NDVI values were calculated for four circular buffer zones (radii equal to 100 m, 250 m, 500 m, and 1000 m) for each year of imagery. For each buffer, an average of NDVI for all the pixels within the buffer was taken. Additionally, the NDVI value for the pixel which intersected the postal code centroid (point) was available for analysis (point value). Buffer sizes that have been used in previous epidemiological studies (Amoly et al., 2014; Hystad et al., 2014) and that were large enough to provide a representation of the area in proportion to the spatial resolution of Landsat (30 m) were used. In this study, the smallest buffer (100 m) contained an average of 85 pixels and the largest buffer (1000 m) contained an average of 6699 pixels.   3.2.3 Statistical Analyses Descriptive statistics were computed for each buffer and year at the postal code level. To assess if regional postal code NDVI had changed over the study period, two-tailed paired t-tests were used to determine which years were significantly different from each other. P-values were adjusted using Bonferroni’s method (α = 0.01) to account for multiple testing. Correlations of NDVI values between years and buffers were checked using Pearson’s correlation coefficient on a random sample of 10,000 postal codes.  To identify the areas of change and the direction of NDVI change over time, postal codes were grouped into cells using a grid overlay. Each cell in the grid was 748 m by 630 m, representing a 100 by 100 cell grid across the study area. The grid approach was chosen for calculation purposes, as it provided a larger set of data with which to create models and provided values of relevance for urban planning. The size of a grid cell would correspond relatively well to an individual’s residential activity space (Harding et al., 2014) and planning models such as “super-blocks” for healthy cities (Mueller et al., 2020). Postal codes were assigned to the grid cell that contained the centroid for that postal code. Theil-Sen models were created for each cell to estimate NDVI over time (Therneau, 2018). The Theil-Sen estimator fits a linear model from the median of the slopes between all point pairs in the dataset. Thus, it is more robust  43  against outliers (Fernandes & Leblanc, 2005) and has been a common choice when assessing remotely sensed data, such as NDVI, in monitoring phenological trends over time (Kovalskyy et al., 2012). The direction and steepness of the slope were used to estimate the fluctuations in vegetation cover (as measured by annual average NDVI). For a sensitivity analysis, these results were compared to ordinary least squares (OLS) linear models. OLS models, while the more common choice, may not be appropriate for remotely sensed data with known measurement error (Fernandes & Leblanc, 2005). Therefore, the output of these models were compared to Theil-Sen models to test the use of OLS in this scenario. Analyses were carried out in R, using the deming, ggplot2, leaflet, rNaturalEarth, and sf packages (Cheng et al., 2019; Pebesma, 2018; R Core Team, 2018; Ram & Wickham, 2019; South, 2017; Therneau, 2018; Wickham, 2016). 3.3 Results 3.3.1 Descriptive statistics There were 59,381 unique postal codes in Metro Vancouver during the study period (1999 to 2014). However, the number of postal codes for any one year fluctuated depending on the number of postal codes that were retired, and the number of new ones created (Table 3-1). Despite masking water bodies prior to calculating NDVI, a small number of negative values for NDVI were found within some buffers. This could be due to a poor fit of the water mask, such as positional inaccuracies between the imagery and mask (Donchyts et al., 2016).            44  Table 3-1: Descriptive statistics for NDVI within the Metro Vancouver for each buffer and year. Buffer (m) Year n Minimum Mean Median Maximum Standard deviation PV 1999 55055 -0.103 0.362 0.375 0.782 0.138  2000 55436 -0.217 0.341 0.354 0.754 0.153  2001 55871 -0.111 0.391 0.4 0.816 0.141  2002 56204 -0.154 0.353 0.361 0.813 0.145  2003 56507 -0.275 0.302 0.306 1 0.130  2004 56826 -0.100 0.362 0.361 0.821 0.138  2005 57049 -0.168 0.357 0.367 1 0.137  2006 57524 -0.268 0.353 0.361 0.826 0.133  2007 57940 -0.091 0.363 0.367 0.833 0.135  2008 58215 -0.086 0.366 0.373 0.798 0.138  2009 58478 -0.65 0.323 0.323 1 0.130  2010 58673 -0.369 0.327 0.333 1 0.138  2011 58825 -0.391 0.333 0.333 0.824 0.138  2013 59311 -0.2 0.351 0.354 0.814 0.155  2014 59091 -0.167 0.380 0.382 0.823 0.147 100 1999 55055 -0.023 0.376 0.383 0.698 0.117  2000 55436 -0.038 0.357 0.363 0.707 0.128  2001 55871 0.005 0.406 0.414 0.770 0.125  2002 56204 -0.032 0.371 0.376 0.768 0.128  2003 56507 -0.017 0.316 0.321 0.688 0.113  2004 56826 0.003 0.379 0.38 0.762 0.122  2005 57049 -0.013 0.372 0.386 0.726 0.120  2006 57524 0.006 0.369 0.377 0.741 0.116  2007 57940 -0.012 0.383 0.391 0.720 0.121  2008 58215 0.003 0.385 0.391 0.768 0.124  2009 58478 0.001 0.343 0.342 0.707 0.115  2010 58673 0.001 0.345 0.352 0.707 0.118  2011 58825 0.01 0.351 0.35 0.729 0.123  2013 59311 -0.089 0.373 0.377 0.772 0.139  2014 59091 -0.003 0.399 0.409 0.758 0.131 250 1999 55055 -0.012 0.387 0.396 0.695 0.110  2000 55436 -0.044 0.369 0.381 0.69 0.121  2001 55871 -0.018 0.420 0.431 0.761 0.119  2002 56204 -0.026 0.385 0.395 0.756 0.123  2003 56507 -0.023 0.327 0.33 0.677 0.109  2004 56826 0.005 0.392 0.395 0.754 0.118  2005 57049 -0.018 0.385 0.394 0.726 0.115  2006 57524 0.012 0.383 0.39 0.699 0.111  2007 57940 -0.002 0.397 0.407 0.701 0.116  2008 58215 0.006 0.400 0.406 0.725 0.120  45  Buffer (m) Year n Minimum Mean Median Maximum Standard deviation  2009 58478 0.016 0.358 0.356 0.704 0.112  2010 58673 -0.005 0.360 0.359 0.695 0.114  2011 58825 0.012 0.366 0.369 0.734 0.121  2013 59311 -0.059 0.391 0.399 0.771 0.135  2014 59091 -0.066 0.415 0.426 0.766 0.124 500 1999 55055 0.009 0.395 0.399 0.666 0.105  2000 55436 -0.007 0.377 0.386 0.687 0.116  2001 55871 0.033 0.429 0.437 0.722 0.114  2002 56204 -0.009 0.394 0.4 0.7 0.120  2003 56507 -0.005 0.335 0.333 0.648 0.106  2004 56826 0.026 0.400 0.4 0.701 0.115  2005 57049 0.003 0.392 0.398 0.68 0.110  2006 57524 0.023 0.392 0.395 0.678 0.107  2007 57940 0.014 0.406 0.411 0.691 0.111  2008 58215 0.034 0.409 0.412 0.721 0.116  2009 58478 0.02 0.367 0.363 0.683 0.108  2010 58673 0.013 0.368 0.365 0.664 0.110  2011 58825 0.028 0.374 0.373 0.697 0.118  2013 59311 -0.026 0.401 0.407 0.757 0.132  2014 59091 -0.023 0.425 0.431 0.744 0.120 1000 1999 55055 0.046 0.402 0.404 0.659 0.100  2000 55436 0.028 0.384 0.39 0.661 0.110  2001 55871 0.069 0.437 0.442 0.714 0.109  2002 56204 0.039 0.401 0.408 0.681 0.114  2003 56507 0.039 0.341 0.338 0.629 0.102  2004 56826 0.072 0.409 0.409 0.706 0.110  2005 57049 0.057 0.400 0.404 0.663 0.105  2006 57524 0.066 0.400 0.402 0.666 0.102  2007 57940 0.06 0.414 0.418 0.678 0.105  2008 58215 0.073 0.418 0.419 0.699 0.111  2009 58478 0.071 0.376 0.37 0.644 0.104  2010 58673 0.079 0.377 0.371 0.647 0.104  2011 58825 0.054 0.382 0.378 0.67 0.114  2013 59311 0.005 0.411 0.415 0.721 0.126  2014 59091 0.076 0.433 0.439 0.725 0.115     46  Small differences in mean NDVI were observed over the study period, but generally, the means remained similar (Figure 3-2). Across all the buffers, 2003 had the lowest mean NDVI and 2001 the highest (Table 3-1). Point value had the largest variation in NDVI values compared with the other buffer sizes.   Figure 3-2: Boxplots of NDVI values by year and buffer size at the postal code level. Pearson’s correlations were evaluated between buffers and years. A majority (n = 2061) of the year – buffer combinations were highly (>0.7) and significantly (p<0.01) correlated with each other (Figure 3-3). As expected, the strongest correlations were observed amongst the larger buffers and between years that were closer together.  3.3.2 Temporal change of NDVI Results from the two-tailed, paired t-tests showed that all combinations and year pairs were significantly different from one another (Appendix C, Table C1).  47   Figure 3-3: Pearson’s correlation matrix of NDVI values for all buffer (point values and 100 m - 1000 m) and year (1999 - 2014) combinations. Scale on the right denotes the correlation coefficient of each combination.  3.3.3 Identifying areas with change On average, 631 postal codes were included in each cell of the grid overlay created for the Theil-Sen models. Table 3-2 presents slope ratios, indicating that the mean change in NDVI per each buffer zone was close to zero across the region over the time period. The largest changes were seen for point values with maximum slope ratios of 0.028 and -0.035, increase and decrease, respectively.      48  Table 3-2: Descriptions of the Theil-Sen model slope characteristics Buffer Minimum Mean Median Maximum Stand. Dev.  Point Value -0.035 -0.00015 0 0.028 0.0044 100 m -0.032 0.00013 0.00043 0.027 0.0038 250 m -0.032 0.00012 0.00040 0.021 0.0034 500 m -0.023 0.00016 0.00036 0.024 0.0030 1000 m -0.015 0.00023 0.00027 0.022 0.0024   The spatial distribution of the change is demonstrated in Figure 3-4. A majority of the metropolitan area had stable values (white on the map), but small pockets of change were found in the southeast including areas with both increases (orange) and decreases (purple) in NDVI (vegetation). Similar patterns across the region were observed in all buffers, although larger buffers appeared to attenuate the steepness of the slope (Table 3-2).   49   Figure 3-4: Direction and steepness of the slope for each Theil-Sen model for each grid cell. Orange indicates more positive changes in NDVI over time, purple represents more negative changes in NDVI over time, and yellow indicates no change.   50  3.3.4 Sensitivity analysis In order to test the difference between the Theil-Sen models and OLS linear models, the same analysis was conducted substituting the Theil-Sen models for OLS linear models. Most of the metropolitan region exhibited stable vegetation trends (white on the map), with a small pocket of change in the southeast (orange and purple, indicating increase and decrease, respectively) (Figure 3-5). The results were almost identical between the two models, although the OLS models (Table 3-3) had larger slope ratios in general compared with the Thiel-Sen models (Table 3-2).  Table 3-3: Descriptions of the OLS model slope characteristics. Buffer Minimum Mean Median Maximum Stand. Dev.  Point Value -0.036 -0.00022 0.00023 0.03158 0.00464 100 m -0.030 0.00011 0.00049 0.02766 0.00392 250 m -0.031 0.00019 0.00053 0.02721 0.00348 500 m -0.024 0.00029 0.00049 0.02679 0.00303 1000 m -0.014 0.00042 0.00051 0.02411 0.00237   51   Figure 3-5: Direction and strength of the slope for each OLS linear model for each grid cell. Orange indicates more positive changes in NDVI over time; purple represents more negative changes in NDVI over time. White denotes changes close to zero.  52  3.4 Discussion This chapter analyzed how vegetation, as measured by average annual NDVI, changed between 1999 and 2014 in Metro Vancouver. The variation in NDVI at the postal code level was significantly different from year to year. However, after spatial aggregation and modeling using the Theil-Sen estimator, most neighborhoods exhibited small changes in NDVI over time, the exception being areas in the southeast, which showed larger changes. This suggests that vegetation has remained relatively stable for Metro Vancouver over the study period, at least on a larger neighborhood scale. Additionally, the expected results that larger buffers reduced the variances found within postal code-level NDVI were found. During a sensitivity analysis, the outcomes of Theil-Sen estimator were compared to the OLS linear models and were found to be similar; indicating that either model is able to approximate vegetation change.  The average change in NDVI per year (Theil-Sen models) ranged between -0.00015 and 0.00028 across all buffer sizes. Since NDVI is a ratio of red and near infrared reflectance, its real-world value is difficult to conceptualize since it is often dependent on the image that was used to create the index, and subsequently is difficult to compare across time (White et al., 2014). While some sources (e.g., Weier & Herring, 2000) propose NDVI ranges are indicative of certain habitat structures, these ranges may not be suitable for urban use, between different sensors, or at different scales. For example, NDVI calculated at global scales may reflect the type of ecosystem fairly well but may not be as representative of vegetation at city-level scales (Jiang et al., 2006). Additionally, different sensors record unique ranges of the electromagnetic spectrum and at varying spatial scales (i.e., pixel size) which could perform differently over heterogeneous landscapes. Therefore, the NDVI values of change identified have little to no real-world value as they stand alone and can only be useful when comparing to other values calculated in a similar way.  Furthermore, while Rhew et al. (2011) suggest that there is a correlation between NDVI and expert evaluations of street-level greenness, other studies have found little agreement between NDVI and perceived greenness by lay people or land cover maps (Gascon et al., 2016; Leslie et al., 2010). Similarly, studies comparing greenness indicators developed from Google Street View, which captures a vertical perspective of greenness, find only small to moderate correlation to NDVI values (Nesbitt et al., 2018; Villeneuve et al., 2018). Since little is known  53  about what NDVI looks like on the ground, the small changes in NDVI detected in this chapter, may be undetectable by humans. Therefore, context is important when considering NDVI and its relationship to health. NDVI is representative of photosynthetic productivity, therefore, values for the same area can change over time regardless of whether or not there was actual change in the amount of vegetation. As such, NDVI values are generally used in comparative scenarios, both geographically and temporally, rather than as direct measure of vegetation health. Additionally, multispectral imagery is susceptive to spectral saturation in areas where vegetation covers more than 60% of the area (Duncanson et al., 2010). In other words, NDVI becomes an insufficient measure to tell the difference between areas with high biomass. In order to properly interpret what unit of change in NDVI (temporally and/or spatially) actually represents in the context of health, further investigation is required.  To my knowledge, Helbich (2019) is the only other study to propose methods for assessing vegetation stability for epidemiological studies. In that study, Helbich (2019) found significant year-to-year differences in a sample of Dutch postal codes and concluded vegetation had not remained stable over time. The same trends were found at the postal code level in Metro Vancouver; however, the changes were assessed over time at aggregated spatial units using Theil-Sen and OLS models, the differences disappeared. One possible reason for this is the amount of variation that was included in the postal code-level data. Many sources of variation, including environmental conditions (e.g., atmosphere, weather, climatic processes) or sensor characteristics (e.g., angle of image collection) could be present in this dataset. This becomes problematic with summarized NDVI, for instance, yearly average NDVI, because there are few datapoints over the time period, making it difficult to detect signal from noise. While summarized data makes it easier to calculate exposures, it also makes it difficult to determine outlier datapoints. Additionally, the inability to detect true signals made the modeling process impossible at a postal code level, which is why the grid overlay was essential for this process to work. This suggests that future studies should consider the use of more time sensitive data, for instance monthly values instead of yearly, in order to model NDVI change on smaller spatial scales. Subsequently, the grouping of postal codes into a grid may have led to additional loss of variance because the grouping smoothed the data. A possible solution for this is to conduct pixel-based analyses, rather than the buffered NDVI data. This would allow for detection of outlier pixels at the source rather than later on when one cannot be sure of the source of the variation.     54  There is no consensus on methods for determining temporal stability and the need for temporal alignment has been sparsely considered in past research. With some exceptions (e.g., Engemann et al., 2018), many studies have assumed the vegetation in an area has remained stable over the study period. Findings from this study highlighted some areas where change in greenness has taken place. While, these changes are relative to both the geographic area and the time period assessed, the identification of these areas suggests caution should be taken when directly comparing NDVI values over time. This reasoning is illustrated in Figure 3-6. Here, theoretical NDVI-values for four different years are given for three postal codes (circles) and the direction of the change between years (line). For t1 through t3, all postal codes share the same relative greenness, or in other words, these postal codes maintain the same order of greenness over time although each separate year displays different NDVI-values per each postal code. For datasets that exhibit this pattern, the chosen year for NDVI would be of less importance when analyzing health differences associated with difference in greenness exposure because the relative order in greenness value between the postal codes stays the same from year to year, despite the differences in NDVI value. However, for t0, the relative order has changed, and we would thus expect a different order of association to health outcomes on a spatial scale. This suggests that for cross-sectional studies, temporal alignment is important to consider, and direct comparison of NDVI values between years could be misleading.   Figure 3-6: Theoretical NDVI values over time for a set of postal codes. The points represent hypothetical postal code NDVI values over time.   55  The choice of neighborhood level exposures (distance and buffer size around postal code or residence) for analyzing associations between green spaces and health outcomes has been debated in earlier literature (Ekkel & de Vries, 2017). In our analysis, buffers sizes were chosen that are representative of previous health studies (Amoly et al., 2014; Hystad et al., 2014) and that were also considerate of the underlying imagery (Landsat, 30 m). Nevertheless, high variation was found within the point values, which highlights the possibility of incorrect or extreme NDVI values when using point values and buffers with radii that are smaller than or equal to the spatial resolution of the imagery (e.g., point values).  Similar to Helbich (2019), lower sensitivity to temporal influence was found in larger buffers. This is not surprising given that larger buffers contain more pixels and allow for an averaging of outliers. However, aggregation of this type could mask important changes in the data and therefore would be more sensitive to strong changes in NDVI values, particularly in the larger buffers. Additionally, the use of nested buffers may also make these studies susceptible to issues outlined in the modifiable areal unit problem (MAUP) (Higgs, 2009). MAUP is a statistical bias related to the sensitivity of results depending on the spatial unit of aggregation (Nelson & Brewer, 2017; Openshaw, 1984) For instance, larger buffers are more likely to show significant relationships with health (Labib et al., 2020; Nelson & Brewer, 2017), which may be a result of the ability of larger buffers to encompass all NEs immediately surrounding a home, in which smaller buffer cannot (Browning & Lee, 2017). Due to this, the use of aggregation (e.g., buffers) should be considered with regard to the research question, particularly when interpreting studies where results differ depending on buffer size. For instance, it may be beneficial to use non-overlapping buffers to determine the association with NEs at different distances, rather than as a cumulative effect (Browning & Lee, 2017). This could result in studies that have significant results, for example in larger buffers, but do not necessarily make sense when considering the proposed pathways.  3.5 Conclusions This study assessed how vegetation, as measured by NDVI from Landsat, fluctuates temporally and geographically within the Metro Vancouver region. The results showed that NDVI values at the postal code level were significantly different from year to year. However, overall NDVI did not change at an aggregated geographic level, although areas where special attention should be given were identified. This suggests that although temporal stability of  56  Metro Vancouver can be assumed, understanding the trends of vegetation is an important step in determining how green space influences human health.  The issue of temporal misalignment in epidemiological studies has been posed as a possible issue that may result in inconsistent results concerning green space exposure and human health. This study suggests that inaccurate exposures could have compounding effects on results and concludes that direct temporal alignment between health and NDVI data is important for accurate comparison between neighborhoods, but that time bound changes in health cannot be directly related to variation in NDVI values.   57  Chapter Four: Conclusions 4.1 Overview and key findings This research contributes to the field of childhood environmental epidemiology through a systematic review (Chapter 2) and an exploration of common practices related to data alignment and temporal stability of vegetation (Chapter 3). The aim of Chapter 2 was to provide insight on the methods used to measure NE in the context of childhood development and mental health and to assess how potential associations may differ depending on the metric used. Chapter 3 explored the commonly used practice of assuming stability of vegetation through an investigation of the temporal trends of Landsat-derived NDVI data in Metro Vancouver and the possible implications of exposure misclassification in health studies. Together these two chapters highlight the current methods for measuring NEs for health research and also provides recommendations for reducing possible errors in NE measurements.   Chapter 2 presents a systematic review of the NE metrics used in childhood health and development research, with a particular focus on birth outcomes, academic achievement, mental health diagnoses, and cognitive development and social functioning. Remotely sensed datasets, specifically NDVI and LULC, were the most common NE metrics used in childhood health and development studies. These metrics showed the strongest associations to increased birth weight and gestational age, decreased self-reported ADHD symptoms. The evidence level for these outcomes was assessed as sufficient following a quality evaluation of the included studies. Additionally, metrics that measured the use of NE was associated with improved social function in moderate quality studies, although use was not a widely used metric. The implications from this chapter revolve around the need for future studies to use comparable NE metrics and health outcomes if we are to improve the evidence of how NE relate to health through particular pathways. Nevertheless, given sufficient evidence for a few childhood health outcomes, the results from the study can also be used to advocate for more NE in children’s daily lives. Chapter 3 demonstrated the usability of Theil-Sen estimator for modeling NDVI over a 15-year time period in Metro Vancouver. Small trends were observed in most of the metropolitan area, however, isolated areas in the south-east exhibited larger changes. It could therefore  58  be assumed that, in the larger region, vegetation had remained more or less stable over the time period. Contextually, the changes in NDVI observed in this study were very small and not necessarily identifiable by humans. Additionally, while the overall change was marginal, there was a significant year-to-year variation in NDVI values, which suggests that it is difficult to identify true changes in NDVI. These results suggest that direct comparison of NDVI values between years should be avoided in order to reduce error caused by systematic differences between collection dates due to factors such as environmental conditions. Furthermore, to reduce the chance of exposure misclassification, health and NDVI data should be temporally aligned.  Together, these chapters provide guidance on how to better measure NE exposure in health studies and provide an assessment of the quality of research conducted to date. The use of comparative measures needs to be deployed in future studies. This will help strengthen the evidence of association and provide the opportunity for comparison between studies. Chapter 2 noted the heterogeneity in NE metrics used and how this prevents comparability between studies also considering the same health outcome. Additionally, errors may be introduced through inaccurate assignment of exposures due to temporal misalignment between NE and health datasets. Therefore, trends of NE metrics (e.g., NDVI) over time should be considered for longitudinal datasets in lieu of assuming temporal stability of an area. The methods detailed in Chapter 3 could be used to identify how areas have changed over time and how this may relate to changes in health. Building on this thesis, future research can improve the knowledge of the association between childhood health and NE, including how children interact with the environment and provide recommendations for urban planners to prioritize NEs in city development. Additionally, with careful selection of NE metrics, we may be able to uncover specific measurements that relate to individual pathways, which in turn can influence planning to provide humans with opportunities to be healthier. 4.2 Limitations While Chapter 2 was a comprehensive literature review, gray literature was not assessed. This may have resulted in the exclusion of some studies, especially for geographic regions not represented in the review, as economically disadvantaged areas may not have access to the  59  peer-reviewed literature needed to support and justify their own research. Additionally, only articles published in English were assess. While effect sizes were pooled where possible, this was only available for one health outcome and NE measurement due to the heterogeneous nature of the metrics. This also contributed to the inability to create a funnel plot or other methods to control for publication bias (Mavridis & Salanti, 2014); therefore, the results may have been skewed towards more positive associations. Additionally, the quality assessment and evidence scoring used in this review inherently have an element of subjectivity; this was attempted to be reduced by replicating methods from previous reviews and quality assessment tools (Gascon et al., 2015; Higgins & Green, 2011; International Agency for Research on Cancer, 2019) and by having each article evaluated by myself and another author independently, aiming for as high reliability as possible.  Results from Chapter 3 could have been different if the data was derived from a sensor other than Landsat, such as SPOT. While Landsat is a common choice due to its availability and low cost, the 30 m spatial resolution comes at a disadvantage when assessing small patches of vegetation within cities. Comparisons between high, moderate, and low-resolution sensors have consistently found low and moderate resolution sensors greatly underestimate the amount of vegetation in urban areas compared with high resolution imagery (Qian et al., 2015). However, due to cost and limited availability, high resolution imagery remains a barrier to epidemiological research. Furthermore, Landsat 5 and 7 have temporal resolutions of 16 days, making the number of images available for each year lower than for other sensors with higher temporal resolutions. Thus, the average growing season dataset used may have different numbers of images for each year due to the availability of high-quality, cloud free images. Other sensors have higher daily revisit times, such as MODIS, which could allow for more images to be considered for averaged datasets (Masuoka et al., 1998). However, the spatial resolution is lower for MODIS (250 m) than for Landsat (30 m) or any high-resolution sensor. Finally, modeling on a grid has an inherent component of approximation. Nevertheless, the common use of these kinds of data in similar research, would allow for this method to be tested in other studies. 4.3 Future research As more research is conducted in this field, a few steps could be taken to reduce the bias and errors associated with some of the NE measurements used in the past. The existing research  60  has highlighted the relationship between NEs and childhood health and development. However, gaps in evidence on which NE metrics are most consistently related to different health outcomes and how changes in urban landscapes affect NDVI measurements still exist. In order to fill these gaps, future research should use LULC datasets to reduce the likelihood of some NEs being over-represented due to vegetation health, which is a problem with NDVI datasets. Additionally, LULC datasets offer an advantage over NDVI in that they create discrete classifications, which are not dependent on the properties of the input image. These common classification schemes allow for comparison between different areas and across time. With the expansion of high-resolution sensors, particularly for the use of research, the cost of production of future datasets and the advancement of algorithms and computer processing will allow for automated solutions for LULC dataset creation. However, currently the cost of production for such datasets is a barrier for creating multiple datasets for an area in quick succession. Another possible method for obtaining LC data is un-mixed pixel datasets. Un-mixing pixels is a process which takes the spectral signatures of low and moderate resolution imagery and determines an estimate of ground cover within each pixel. This may allow for the creation of pseudo-landcover for historic datasets, such as Landsat. However, the application of such data is, to date, untested in the context of epidemiology.  Identifying the human “use” of NEs may help with distinguishing between pathways of interest. For example, in Chapter 2 children’s use of NEs via parent reporting or GPS monitoring found an increase in social functioning and evidence for improved behavior. However, measures such as NDVI and LULC datasets did not find the same evidence. In order to identify how NEs interact with health, research needs to focus on actual exposures to NEs. This could be accomplished through GPS monitoring of activity to create individualized exposures of NE. This has been done in a few studies (e.g., Dadvand et al., 2015; Ward et al., 2016) however, large scale and repeated studies will help provide evidence for which NEs influence human health. Additionally, quality measures such as POSDAT or NSI may be useful in identifying NEs that are important for particular populations. For children, this may include playgrounds or sports fields; features that are not included in measurements like NDVI. Additionally, since people are highly mobile, the use of dynamic and individual exposures can help identify exposures more accurately (Dadvand et al., 2015; Helbich, 2018).   61  Lastly, if NDVI continues to be used, future studies should investigate the temporal stability of the datasets and either exclude areas of high change or utilize them as case studies. Chapter 3 provided a simple way of modeling NDVI overtime and could serve as a blueprint for assessing temporal stability in future research. If temporal stability can be determined, it allows for the use of other NE metrics that were developed at a different time, such as LULC datasets.  4.4 Closing thoughts If we are to plan for healthier cities, particularly with children in mind, we need to know how best to measure NE exposure for particular populations and which NE measurements are most appropriate for assessing association with health outcomes. This thesis highlights NEs as important environments for the healthy development of children, but before more conclusive evidence can be provided, research needs to take methods into consideration for accurately measuring exposure and changes in exposure over time.  (Planet Labs, 2016; Williams et al., 2018) for the citation in Chapter 3.  62  References  Abelt, K., & McLafferty, S. (2017). Green Streets: Urban Green and Birth Outcomes. International Journal of Environmental Research and Public Health, 14(7). https://doi.org/10.3390/ijerph14070771 Agay-Shay, K., Peled, A., Crespo, A. V., Peretz, C., Amitai, Y., Linn, S., … Nieuwenhuijsen, M. J. (2014). Green spaces and adverse pregnancy outcomes. Occupational and Environmental Medicine, 71(8), 562–9. https://doi.org/10.1136/oemed-2013-101961 Aguilera, F., Valenzuela, L. M., & Botequilha-Leitão, A. (2011). Landscape metrics in the analysis of urban land use patterns: A case study in a Spanish metropolitan area. Landscape and Urban Planning, 99(3), 226–238. https://doi.org/10.1016/j.landurbplan.2010.10.004 Allender, S., Wickramasinghe, K., Goldacre, M., Matthews, D., & Katulanda, P. (2011). Quantifying Urbanization as a Risk Factor for Noncommunicable Disease. Journal of Urban Health, 88(5), 906–918. https://doi.org/10.1007/s11524-011-9586-1 Amoly, E., Dadvand, P., Forns, J., López-Vicente, M., Basagaña, X., Julvez, J., … Sunyer, J. (2014). Green and Blue Spaces and Behavioral Development in Barcelona Schoolchildren: The BREATHE Project. Environmental Health Perspectives. https://doi.org/10.1289/ehp.1408215 Anderson, P. M., & Butcher, K. F. (2006). Childhood Obesity: Trends and Potential Causes. The Future of Children, 16(1), 19–45. https://www.jstor.org/stable/3556549 Ataol, Ö., Krishnamurthy, S., & van Wesmael, P. (2019). Children’s Participation in Urban Planning and Design: A Systematic Review. Children, Youth and Environments, 29(2), 20.  63  Balduzzi, S., Rücker, G., & Schwarzer, G. (2019). How to perform a meta-analysis with R: A practical tutorial. Evidence-Based Mental Health. https://doi.org/DOI: 10.1136/ebmental-2019-300117 Balsas, C. J. L. (2004). Measuring the livability of an urban centre: An exploratory study of key performance indicators. Planning Practice and Research, 19(1), 101–110. https://doi.org/10.1080/0269745042000246603 Balseviciene, B., Sinkariova, L., Grazuleviciene, R., Andrusaityte, S., Uzdanaviciute, I., Dedele, A., & Nieuwenhuijsen, M. J. (2014). Impact of residential greenness on preschool children’s emotional and behavioral problems. International Journal of Environmental Research and Public Health, 11(7), 6757–70. https://doi.org/10.3390/ijerph110706757 Beere, P., & Kingham, S. (2017). Assessing the relationship between greenspace and academic achievement in urban New Zealand primary schools. New Zealand Geographer, 73(3), 155–165. https://doi.org/10.1111/nzg.12155 Bishop, K., & Corkery, L. (Eds.). (2017). Designing cities with children and young people: Beyond playgrounds and skate parks. New York: Routledge, Taylor & Francis Group. Borenstein, M., Hedges, L. V., Higgins, J. P., & Rothstein, H. R. (Eds.). (2011). Introduction to meta-analysis. Chichester, U.K: John Wiley & Sons. Bradley, B. A., Jacob, R. W., Hermance, J. F., & Mustard, J. F. (2007). A curve fitting procedure to derive inter-annual phenologies from time series of noisy satellite NDVI data. Remote Sensing of Environment, 106(2), 137–145. https://doi.org/10.1016/j.rse.2006.08.002 Bratman, G. N., Anderson, C. B., Berman, M. G., Cochran, B., de Vries, S., Flanders, J., … Daily, G. C. (2019). Nature and mental health: An ecosystem service perspective. Science Advances, 5, 14. Braubach, M., Egorov, A., Mudu, P., Wolf, T., Ward Thompson, C., & Martuzzi, M. (2017). Effects of Urban Green Space on Environmental Health, Equity and Resilience. In N. Kabisch, H.  64  Korn, J. Stadler, & A. Bonn (Eds.), Nature-Based Solutions to Climate Change Adaptation in Urban Areas: Linkages between Science, Policy and Practice (pp. 187–205). Cham: Springer International Publishing. https://doi.org/10.1007/978-3-319-56091-5 Browning, M. H. E. M., Kuo, M., Sachdeva, S., Lee, K., & Westphal, L. (2018). Greenness and school-wide test scores are not always positively associated – A replication of “linking student performance in Massachusetts elementary schools with the ‘greenness’ of school surroundings using remote sensing.” Landscape and Urban Planning, 178, 69–72. https://doi.org/10.1016/j.landurbplan.2018.05.007 Browning, M. H. E. M., & Rigolon, A. (2019). School Green Space and Its Impact on Academic Performance: A Systematic Literature Review. International Journal of Environmental Research and Public Health, 16(3), 429. https://doi.org/10.3390/ijerph16030429 Browning, M., & Lee, K. (2017). Within What Distance Does “Greenness” Best Predict Physical Health? A Systematic Review of Articles with GIS Buffer Analyses across the Lifespan. International Journal of Environmental Research and Public Health, 14(7), 675. https://doi.org/10.3390/ijerph14070675 Canadian Centre for Remote Sensing. (2016). Fundamentals of Remote Sensing (p. 258). Natural Resources Canada. Retrieved from https://www.nrcan.gc.ca/sites/www.nrcan.gc.ca/files/earthsciences/pdf/resource/tutor/fundam/pdf/fundamentals_e.pdf CanMap Postal Code Suite v2015.3 [computer file]. (2015). Markham: DMTI Spatial Inc. Carolan-Olah, M., & Frankowska, D. (2014). High environmental temperature and preterm birth: A review of the evidence. Midwifery, 30(1), 50–59. https://doi.org/10.1016/j.midw.2013.01.011  65  Casey, J. A., James, P., Rudolph, K. E., Wu, C.-D., & Schwartz, B. S. (2016). Greenness and Birth Outcomes in a Range of Pennsylvania Communities. International Journal of Environmental Research and Public Health, 13(3). https://doi.org/10.3390/ijerph13030311 Cheng, J., Karambelkar, B., & Xie, Y. (2019). leaflet: Create interactive web maps with the JavaScript “Leaflet” Library (Version 2.0.3). Retrieved from https://CRAN.R-project.org/package=leaflet Chiabai, A., Quiroga, S., Martinez-Juarez, P., Higgins, S., & Taylor, T. (2018). The nexus between climate change, ecosystem services and human health: Towards a conceptual framework. Science of The Total Environment, 635, 1191–1204. https://doi.org/10.1016/j.scitotenv.2018.03.323 Christian, H., Ball, S. J., Zubrick, S. R., Brinkman, S., Turrell, G., Boruff, B., & Foster, S. (2017). Relationship between the neighbourhood built environment and early child development. Health & Place, 48, 90–101. https://doi.org/10.1016/j.healthplace.2017.08.010 Coutts, C., & Hahn, M. (2015). Green Infrastructure, Ecosystem Services, and Human Health. International Journal of Environmental Research and Public Health, 12(8), 9768–9798. https://doi.org/10.3390/ijerph120809768 Cuijpers, P. (2016). Meta-analysis in mental health: A practical guide. Amsterdam, Netherlands: Pim Cuijpers Uitgeverij. Cusack, L., Larkin, A., Carozza, S. E., & Hystad, P. (2017a). Associations between multiple green space measures and birth weight across two US cities. Health & Place, 47, 36–43. https://doi.org/10.1016/j.healthplace.2017.07.002  66  Cusack, L., Larkin, A., Carozza, S., & Hystad, P. (2017b). Associations between residential greenness and birth outcomes across Texas. Environmental Research, 152, 88–95. https://doi.org/10.1016/j.envres.2016.10.003 Dadvand, P., de Nazelle, A., Figueras, F., Basagana, X., Su, J., Amoly, E., … Nieuwenhuijsen, M. J. (2012a). Green space, health inequality and pregnancy. Environment International, 1, 110–115. https://doi.org/10.1016/j.envint.2011.07.004 Dadvand, P., Nieuwenhuijsen, M. J., Esnaola, M., Forns, J., Basagana, X., Alvarez-Pedrerol, M., … Sunyer, J. (2015). Green spaces and cognitive development in primary schoolchildren. Proceedings of the National Academy of Sciences of the United States of America, 112(26), 7937–42. https://doi.org/10.1073/pnas.1503402112 Dadvand, P., Ostro, B., Figueras, F., Foraster, M., Basagaña, X., Valentín, A., … Nieuwenhuijsen, M. J. (2014a). Residential Proximity to Major Roads and Term Low Birth Weight: The Roles of Air Pollution, Heat, Noise, and Road-Adjacent Trees. Epidemiology, 25(4), 518–525. https://doi.org/10.1097/EDE.0000000000000107 Dadvand, P., Sunyer, J., Basagana, X., Ballester, F., Lertxundi, A., Fernandez-Somoano, A., … Nieuwenhuijsen, M. J. (2012b). Surrounding greenness and pregnancy outcomes in four Spanish birth cohorts. Environmental Health Perspectives, 120(10), 1481–1487. https://doi.org/10.1289/ehp.1205244 Dadvand, P., Tischer, C., Estarlich, M., Llop, S., Dalmau-Bueno, A., Lopez-Vicente, M., … Sunyer, J. (2017). Lifelong Residential Exposure to Green Space and Attention: A Population-based Prospective Study. Environmental Health Perspectives, 125(9), 097016. https://doi.org/10.1289/EHP694 Dadvand, P., Wright, J., Martinez, D., Basagana, X., McEachan, R. R. C., Cirach, M., … Nieuwenhuijsen, M. J. (2014b). Inequality, green spaces, and pregnant women: Roles of  67  ethnicity and individual and neighbourhood socioeconomic status. Environment International, 71(du1, 7807270), 101–8. https://doi.org/10.1016/j.envint.2014.06.010 Deering, D., & Rouse, J. (1975). Measuring “Forage Production” of Grazing Units from Landsat MSS Data. 10th International Symposium on Remote Sensing of Environment, 1169–1178. Donchyts, G., Schellekens, J., Winsemius, H., Eisemann, E., & van de Giesen, N. (2016). A 30 m Resolution Surface Water Mask Including Estimation of Positional and Thematic Differences Using Landsat 8, SRTM and OpenStreetMap: A Case Study in the Murray-Darling Basin, Australia. Remote Sensing, 8(5), 386. https://doi.org/10.3390/rs8050386 Donovan, G. H., Michael, Y. L., Butry, D. T., Sullivan, A. D., & Chase, J. M. (2011). Urban trees and the risk of poor birth outcomes. Health & Place, 17(1), 390–393. https://doi.org/10.1016/j.healthplace.2010.11.004 Duncanson, L. I., Niemann, K. O., & Wulder, M. A. (2010). Integration of GLAS and Landsat TM data for aboveground biomass estimation, 36(2), 14. Dzhambov, A. M., Dimitrova, D. D., & Dimitrakova, E. D. (2014). Association between residential greenness and birth weight: Systematic review and meta-analysis. Urban Forestry & Urban Greening, 13(4), 621–629. https://doi.org/10.1016/j.ufug.2014.09.004 Ebisu, K., Holford, T. R., & Bell, M. L. (2016). Association between greenness, urbanicity, and birth weight. Science of The Total Environment, 542, 750–756. https://doi.org/10.1016/j.scitotenv.2015.10.111 Edwards, N., Hooper, P., Trapp, G. S. A., Bull, F., Boruff, B., & Giles-Corti, B. (2013). Development of a Public Open Space Desktop Auditing Tool (POSDAT): A remote sensing approach. Applied Geography, 38, 22–30. http://dx.doi.org/10.1016/j.apgeog. 2012.11.010.  68  Ekkel, E. D., & de Vries, S. (2017). Nearby green space and human health: Evaluating accessibility metrics. Landscape and Urban Planning, 157, 214–220. https://doi.org/10.1016/j.landurbplan.2016.06.008 Engemann, K., Pedersen, C. B., Arge, L., Tsirogiannis, C., Mortensen, P. B., & Svenning, J.-C. (2018). Childhood exposure to green space – A novel risk-decreasing mechanism for schizophrenia? Schizophrenia Research, 199, 142–148. https://doi.org/10.1016/j.schres.2018.03.026 European Commission Directorate-General. (2015). Towards an EU research and innovation policy agenda for nature-based solutions and re-naturing cities: Final report of the Horizon 2020 expert group on “Nature-based solutions and re-naturing cities.” Luxembourg: 2020 Expert Group on Nature-based Solutions and Re-Naturing Cities. European Environment Agency. (2017, May). Copernicus Land Monitoring Service—Local Componets: Urban Atlas. Retrieved from https://land.copernicus.eu/user-corner/publications/ua-flyer/at_download/file Faber Taylor, A., & Kuo, F. E. (2009). Children with Attention Deficits Concentrate Better After Walk in the Park. Journal of Attention Disorders, 12(5), 402–409. https://doi.org/10.1177/1087054708323000 Faber Taylor, A., & Kuo, F. E. M. (2011). Could Exposure to Everyday Green Spaces Help Treat ADHD? Evidence from Children’s Play Settings: Everyday green space and ADHD symptoms. Applied Psychology: Health and Well-Being, 3(3), 281–303. https://doi.org/10.1111/j.1758-0854.2011.01052.x Faber Taylor, A., Kuo, F. E., & Sullivan, W. C. (2001). Coping with ADD: The Surprising Connection to Green Play Settings. Environment and Behavior, 33(1), 54–77. https://doi.org/10.1177/00139160121972864  69  Faber Taylor, A., Kuo, F. E., & Sullivan, W. C. (2002). Views of Nature and Self-discipline: Evidence from Inner City Children. Journal of Environmental Psychology, 22(1–2), 49–63. https://doi.org/10.1006/jevp.2001.0241 Feng, X., & Astell-Burt, T. (2017). Residential Green Space Quantity and Quality and Child Well-being: A Longitudinal Study. American Journal of Preventive Medicine, 53(5), 616–624. https://doi.org/10.1016/j.amepre.2017.06.035 Fernandes, R., & Leblanc, S. G. (2005). Parametric (modified least squares) and non-parametric (Theil–Sen) linear regressions for predicting biophysical parameters in the presence of measurement errors. Remote Sensing of Environment, 14. Finegold, Y., Ortmann, A., Lindquist, E., d’Annunzio, R., & Sandker, M. (2016). Map Accuracy Assessment and Area Estimation: A practical guide (National forest monitoring assessment No. working paper No. 46/E) (p. 69). Rome, Italy: Food and Agriculture Organization of the United Nations. Retrieved from http://www.fao.org/3/a-i5601e.pdf Flouri, E., Midouhas, E., & Joshi, H. (2014). The role of urban neighbourhood green space in children’s emotional and behavioural resilience. Journal of Environmental Psychology, 40, 179–186. https://doi.org/10.1016/j.jenvp.2014.06.007 Fong, K., Kloog, I., Coull, B., Koutrakis, P., Laden, F., Schwartz, J., & James, P. (2018). Residential Greenness and Birthweight in the State of Massachusetts, USA. International Journal of Environmental Research and Public Health, 15(6), 1248. https://doi.org/10.3390/ijerph15061248 Foody, G. M. (2002). Status of land cover classification accuracy assessment. Remote Sensing of Environment, 80(1), 185–201. https://doi.org/10.1016/S0034-4257(01)00295-4 Gascon, M., Cirach, M., Martinez, D., Dadvand, P., Valentin, A., Plasencia, A., & Nieuwenhuijsen, M. (2016). Normalized difference vegetation index (NDVI) as a marker of surrounding  70  greenness in epidemiological studies: The case of Barcelona City. Urban Forestry & Urban Greening, 19, 88–94. https://doi.org/10.1016/j.ufug.2016.07.001 Gascon, M., Triguero-Mas, M., Martínez, D., Dadvand, P., Forns, J., Plasència, A., & Nieuwenhuijsen, M. (2015). Mental Health Benefits of Long-Term Exposure to Residential Green and Blue Spaces: A Systematic Review. International Journal of Environmental Research and Public Health, 12(4), 4354–4379. https://doi.org/10.3390/ijerph120404354 Glazer, K. B., Eliot, M. N., Danilack, V. A., Carlson, L., Phipps, M. G., Dadvand, P., … Wellenius, G. A. (2018). Residential green space and birth outcomes in a coastal setting. Environmental Research, 163(ei2, 0147621), 97–107. https://doi.org/10.1016/j.envres.2018.01.006 Gluckman, P. D., Buklijas, T., & Hanson, M. A. (2016). The Developmental Origins of Health and Disease (DOHaD) Concept. In the Epigenome and Developmental Origins of Health and Disease (pp. 1–15). Elsevier. https://doi.org/10.1016/B978-0-12-801383-0.00001-3 Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., & Moore, R. (2017). Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment, 202, 18–27. https://doi.org/10.1016/j.rse.2017.06.031 Government of Canada, S. C. (2015, April 13). Canada goes urban. Retrieved September 9, 2019, from https://www150.statcan.gc.ca/n1/pub/11-630-x/11-630-x2015004-eng.htm Gracey, M. (2007). Child health in an urbanizing world. Acta Paediatrica, 91(1), 1–8. https://doi.org/10.1111/j.1651-2227.2002.tb01629.x Gray, C., Gibbons, R., Larouche, R., Sandseter, E., Bienenstock, A., Brussoni, M., … Tremblay, M. (2015). What Is the Relationship between Outdoor Time and Physical Activity, Sedentary Behaviour, and Physical Fitness in Children? A Systematic Review.  71  International Journal of Environmental Research and Public Health, 12(6), 6455–6474. https://doi.org/10.3390/ijerph120606455 Grazuleviciene, R., Danileviciute, A., Dedele, A., Vencloviene, J., Andrusaityte, S., Uzdanaviciute, I., & Nieuwenhuijsen, M. J. (2015). Surrounding greenness, proximity to city parks and pregnancy outcomes in Kaunas cohort study. International Journal of Hygiene and Environmental Health, 218(3), 358–65. https://doi.org/10.1016/j.ijheh.2015.02.004 Hammer, R. B., Stewart, S. I., Winkler, R. L., Radeloff, V. C., & Voss, P. R. (2004). Characterizing dynamic spatial and temporal residential density patterns from 1940–1990 across the North Central United States. Landscape and Urban Planning, 69(2–3), 183–199. https://doi.org/10.1016/j.landurbplan.2003.08.011 Harding, C., Patterson, Z., Miranda-Moreno, L. F., & Zahabi, S. A. (2014). A spatial and temporal comparative analysis of the effects of land-use clusters on activity spaces in three Quebec cities. Environment and Planning B: Planning and Design, 41(6), 1044–1062. https://doi.org/10.1068/b130068p Hartig, T., Mitchell, R., de Vries, S., & Frumkin, H. (2014). Nature and Health. Annual Review of Public Health, 35(1), 207–228. https://doi.org/10.1146/annurev-publhealth-032013-182443 Hay, S. I. (2000). An Overview of Remote Sensing and Geodesy for Epidemiology and Public Health Application. Advances in Parasitology, 47, 1–35. Helbich, M. (2018). Toward dynamic urban environmental exposure assessments in mental health research. Environmental Research, 161, 129–135. https://doi.org/10.1016/j.envres.2017.11.006 Helbich, M. (2019). Spatiotemporal Contextual Uncertainties in Green Space Exposure Measures: Exploring a Time Series of the Normalized Difference Vegetation Indices.  72  International Journal of Environmental Research and Public Health, 16(5), 852. https://doi.org/10.3390/ijerph16050852 Higgins, J. P., & Green, S. (2011, March). Cochrane Handbook for Systematic Reviews of Interventions. John Wiley & Sons Inc. Retrieved from http://handbook-5-1.cochrane.org/ Higgs, G. (2009). The role of GIS for health utilization studies: Literature review. Health Services and Outcomes Research Methodology, 9(2), 84–99. https://doi.org/10.1007/s10742-009-0046-2 Hird, J. N., & McDermid, G. J. (2009). Noise reduction of NDVI time series: An empirical comparison of selected techniques. Remote Sensing of Environment, 113(1), 248–258. https://doi.org/10.1016/j.rse.2008.09.003 Hodson, C. B., & Sander, H. A. (2017). Green urban landscapes and school-level academic performance. Landscape and Urban Planning, 160, 16–27. https://doi.org/10.1016/j.landurbplan.2016.11.011 Hystad, P., Davies, H. W., Frank, L., Van Loon, J., Gehring, U., Tamburic, L., & Brauer, M. (2014). Residential Greenness and Birth Outcomes: Evaluating the Influence of Spatially Correlated Built-Environment Factors. Environmental Health Perspectives, 122(10), 1095–1102. https://doi.org/10.1289/ehp.1308049 International Agency for Research on Cancer, I. A. for R. on C. (2019, January). IARC Monographs on the Identification of Carcinogenic Hazards to Humans: Preamble. Retrieved November 11, 2019, from https://monographs.iarc.fr/wp-content/uploads/2019/01/Preamble-2019.pdf Ji, W., Ma, J., Twibell, R. W., & Underhill, K. (2006). Characterizing urban sprawl using multi-stage remote sensing images and landscape metrics. Computers, Environment and Urban Systems, 30(6), 861–879. https://doi.org/10.1016/j.compenvurbsys.2005.09.002  73  Jiang, Z., Huete, A. R., Chen, J., Chen, Y., Li, J., Yan, G., & Zhang, X. (2006). Analysis of NDVI and scaled difference vegetation index retrievals of vegetation fraction. Remote Sensing of Environment, 101(3), 366–378. https://doi.org/10.1016/j.rse.2006.01.003 Jin, J., Gergel, S. E., Lu, Y., Coops, N. C., & Wang, C. (2019). Asian Cities are Greening While Some North American Cities are Browning: Long-Term Greenspace Patterns in 16 Cities of the Pan-Pacific Region. Ecosystems. https://doi.org/10.1007/s10021-019-00409-2 Kaplan, R., & Kaplan, S. (1989). The Experience of Nature. Cambridge University Press. Kaplan, S. (1995). The restorative benefits of nature: Toward an integrative framework. Journal of Environmental Psychology, 15(3), 169–182. https://doi.org/10.1016/0272-4944(95)90001-2 Karsten, L. (2005). It all used to be better? Different generations on continuity and change in urban children’s daily use of space. Children’s Geographies, 3(3), 275–290. https://doi.org/10.1080/14733280500352912 Khan, S. (2018). Positional accuracy of geocoding from residential postal codes versus full street addresses. Health Reports, 29(82), 9. Kihal-Talantikite, W., Padilla, C. M., Lalloue, B., Gelormini, M., Zmirou-Navier, D., & Deguen, S. (2013). Green space, social inequalities and neonatal mortality in France. BMC Pregnancy & Childbirth, 1, 191. https://doi.org/10.1186/1471-2393-13-191 Kim, J.-H., Lee, C., & Sohn, W. (2016). Urban Natural Environments, Obesity, and Health-Related Quality of Life among Hispanic Children Living in Inner-City Neighborhoods. International Journal of Environmental Research and Public Health, 13(1), 121. https://doi.org/10.3390/ijerph13010121 Kondo, M. C., Fluehr, J. M., McKeon, T., & Branas, C. C. (2018). Urban Green Space and Its Impact on Human Health. International Journal of Environmental Research and Public Health, 15(3). https://doi.org/10.3390/ijerph15030445  74  Konijnendijk, C., Nillsson, K., Randrup, T. B., & Schipperijn, J. (Eds.). (2005). Urban Forests and Trees: A reference book. Springer Science + Business Media B.V. Retrieved from https://link.springer.com/content/pdf/10.1007%2F3-540-27684-X.pdf Kovalskyy, V., Roy, D. P., Zhang, X. Y., & Ju, J. (2012). The suitability of multi-temporal web-enabled Landsat data NDVI for phenological monitoring – a comparison with flux tower and MODIS NDVI. Remote Sensing Letters, 3(4), 325–334. https://doi.org/10.1080/01431161.2011.593581 Kuo, F. E., & Faber Taylor, A. (2004). A Potential Natural Treatment for Attention-Deficit/Hyperactivity Disorder: Evidence from a National Study. American Journal of Public Health, 94(9), 1580–1586. Kuo, M., Browning, M. H. E. M., Sachdeva, S., Lee, K., & Westphal, L. (2018). Might School Performance Grow on Trees? Examining the Link Between “Greenness” and Academic Achievement in Urban, High-Poverty Schools. Frontiers in Psychology, 9. https://doi.org/10.3389/fpsyg.2018.01669 Kweon, B.-S., Ellis, C. D., Lee, J., & Jacobs, K. (2017). The link between school environments and student academic performance. Urban Forestry & Urban Greening, 23, 35–43. https://doi.org/10.1016/j.ufug.2017.02.002 Labib, S. M., Lindley, S., & Huck, J. J. (2020). Spatial dimensions of the influence of urban green-blue spaces on human health: A systematic review. Environmental Research, 180, 108869. https://doi.org/10.1016/j.envres.2019.108869 Lachowycz, K., & Jones, A. P. (2013). Towards a better understanding of the relationship between greenspace and health: Development of a theoretical framework. Landscape and Urban Planning, 118, 62–69. https://doi.org/10.1016/j.landurbplan.2012.10.012 Landsat 5 TM Annual Greenest-Pixel TOA Reflectance Composite, 1984 to 2012. (2017, Accessed). Retrieved from  75  https://explorer.earthengine.google.com/#detail/LANDSAT%2FLT5_L1T_ANNUAL_GREENEST_TOA Landsat 8 Annual Greenest-Pixel TOA Reflectance Composite, 2013 to 2015. (2017, Accessed). Retrieved from https://explorer.earthengine.google.com/#detail/LANDSAT%2FLC8_L1T_ANNUAL_GREENEST_TOA Larson, L. R., Barger, B., Ogletree, S., Torquati, J., Rosenberg, S., Gaither, C. J., … Schutte, A. (2018). Gray space and green space proximity associated with higher anxiety in youth with autism. Health & Place, 53, 94–102. https://doi.org/10.1016/j.healthplace.2018.07.006 Laurent, O., Wu, J., Li, L., & Milesi, C. (2013). Green spaces and pregnancy outcomes in Southern California. Health & Place, 1, 190–195. https://doi.org/10.1016/j.healthplace.2013.09.016 Lavender, S., & Lavender, A. (2016). Practical Handbook of Remote Sensing. Boca Raton, FL USA: Taylor & Frances Group. Leslie, E., Sugiyama, T., Ierodiaconou, D., & Kremer, P. (2010). Perceived and objectively measured greenness of neighbourhoods: Are they measuring the same thing? Landscape and Urban Planning, 95(1), 28–33. https://doi.org/10.1016/j.landurbplan.2009.11.002 Livesley, S. J., McPherson, G. M., & Calfapietra, C. (2016). The Urban Forest and Ecosystem Services: Impacts on Urban Water, Heat, and Pollution Cycles at the Tree, Street, and City Scale. Journal of Environment Quality, 45(1), 119. https://doi.org/10.2134/jeq2015.11.0567 Louv, R. (2008). Last Child in the Woods: Saving Our Children from Nature-Deficit Disorder. New York, United States: Algonquin Books of Chapel Hill. Retrieved from http://ebookcentral.proquest.com/lib/ubc/detail.action?docID=3419101  76  MacFaden, S. W., O’Neil-Dunne, J. P. M., Royar, A. R., Lu, J. W. T., & Rundle, A. G. (2012). High-resolution tree canopy mapping for New York City using LIDAR and object-based image analysis. Journal of Applied Remote Sensing, 6(1), 063567–1. https://doi.org/10.1117/1.JRS.6.063567 MacNaughton, P., Eitland, E., Kloog, I., Schwartz, J., & Allen, J. (2017). Impact of Particulate Matter Exposure and Surrounding “Greenness” on Chronic Absenteeism in Massachusetts Public Schools. International Journal of Environmental Research and Public Health, 14(2). https://doi.org/10.3390/ijerph14020207 Markevych, I., Fuertes, E., Tiesler, C. M. T., Birk, M., Bauer, C.-P., Koletzko, S., … Heinrich, J. (2014a). Surrounding greenness and birth weight: Results from the GINIplus and LISAplus birth cohorts in Munich. Health & Place, 1, 39–46. https://doi.org/10.1016/j.healthplace.2013.12.001 Markevych, I., Schoierer, J., Hartig, T., Chudnovsky, A., Hystad, P., Dzhambov, A. M., … Fuertes, E. (2017). Exploring pathways linking greenspace to health: Theoretical and methodological guidance. Environmental Research, 158, 301–317. https://doi.org/10.1016/j.envres.2017.06.028 Markevych, I., Standl, M., Sugiri, D., Harris, C., Maier, W., Berdel, D., & Heinrich, J. (2016). Residential greenness and blood lipids in children: A longitudinal analysis in GINIplus and LISAplus. Environmental Research, 151(ei2, 0147621), 168–173. https://doi.org/10.1016/j.envres.2016.07.037 Markevych, I., Tesch, F., Datzmann, T., Romanos, M., Schmitt, J., & Heinrich, J. (2018). Outdoor Air Pollution, Greenspace, and Incidence of ADHD: A Semi-Individual Study. Science of The Total Environment, 642(15), 1362–1368. https://doi.org/10.1016/j.scitotenv.2018.06.167  77  Markevych, I., Tiesler, C. M. T., Fuertes, E., Romanos, M., Dadvand, P., Nieuwenhuijsen, M. J., … Heinrich, J. (2014b). Access to urban green spaces and behavioural problems in children: Results from the GINIplus and LISAplus studies. Environment International, 71, 29–35. https://doi.org/10.1016/j.envint.2014.06.002 Markham, B. L., Storey, J. C., Williams, D. L., & Irons, J. R. (2004). Landsat sensor performance: History and current status. IEEE Transactions on Geoscience and Remote Sensing, 42(12), 2691–2694. https://doi.org/10.1109/TGRS.2004.840720 Mårtensson, F., Boldemann, C., Söderström, M., Blennow, M., Englund, J.-E., & Grahn, P. (2009). Outdoor environmental assessment of attention promoting settings for preschool children. Health & Place, 15(4), 1149–1157. https://doi.org/10.1016/j.healthplace.2009.07.002 Masuoka, E., Fleig, A., Wolfe, R. E., & Patt, F. (1998). Key characteristics of MODIS data products. IEEE Transactions on Geoscience and Remote Sensing, 36(4), 1313–1323. https://doi.org/10.1109/36.701081 Mavridis, D., & Salanti, G. (2014). Exploring and accounting for publication bias in mental health: A brief overview of methods. Evidence-Based Mental Health, 17(1), 11–15. https://doi.org/10.1136/eb-2013-101700 McCormick, R. (2017). Does Access to Green Space Impact the Mental Well-being of Children: A Systematic Review. Journal of Pediatric Nursing, 37, 3–7. https://doi.org/10.1016/j.pedn.2017.08.027 McCracken, D. S., Allen, D. A., & Gow, A. J. (2016). Associations between urban greenspace and health-related quality of life in children. Preventive Medicine Reports, 3, 211–221. https://doi.org/10.1016/j.pmedr.2016.01.013 McEachan, R. R. C., Yang, T. C., Roberts, H., Pickett, K. E., Arseneau-Powell, D., Gidlow, C. J., … Nieuwenhuijsen, M. (2018). Availability, use of, and satisfaction with green space, and  78  children’s mental wellbeing at age 4 years in a multicultural, deprived, urban area: Results from the Born in Bradford cohort study. The Lancet Planetary Health, 2(6), e244–e254. https://doi.org/10.1016/S2542-5196(18)30119-0 McMartin, S. E., Kingsbury, M., Dykxhoorn, J., & Colman, I. (2014). Time trends in symptoms of mental illness in children and adolescents in Canada. CMAJ: Canadian Medical Association Journal, 186(18), E672–E678. https://doi.org/10.1503/cmaj.140064 Metro Vancouver. (2019a). Member Municipalities. Retrieved January 20, 2019, from http://www.metrovancouver.org/about/municipalities/Pages/default.aspx Metro Vancouver. (2019b). Regional Profile. Retrieved January 29, 2019, from http://www.metrovancouver.org/services/regional-planning/regional-economy/regional-profile/Pages/default.aspx Millennium Ecosystem Assessment (Ed.). (2005). Ecosystems and human well-being: Synthesis. Washington, DC: Island Press. Moher, D., Liberati, A., Tetzlaff, J., & Altman, D. G. (2009). Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement. PLoS Medicine, 6(7), 6. Moore, M., Gould, P., & Keary, B. S. (2003). Global urbanization and impact on health. International Journal of Hygiene and Environmental Health, 206(4–5), 269–278. https://doi.org/10.1078/1438-4639-00223 Mueller, N., Rojas-Rueda, D., Khreis, H., Cirach, M., Andrés, D., Ballester, J., … Nieuwenhuijsen, M. (2020). Changing the urban design of cities for health: The superblock model. Environment International, 134, 105132. https://doi.org/10.1016/j.envint.2019.105132 National Institute of Health, N. H., Lung and Blood Institute. (2018). Study Quality Assessment Tools—Quality Assessment Tool for Observational Cohort and Cross-sectional Studies. Retrieved July 15, 2019, from https://www.nhlbi.nih.gov/health-topics/study-quality-assessment-tools  79  Nelson, J. K., & Brewer, C. A. (2017). Evaluating data stability in aggregation structures across spatial scales: Revisiting the modifiable areal unit problem. Cartography and Geographic Information Science, 44(1), 35–50. https://doi.org/10.1080/15230406.2015.1093431 Nesbitt, L., Andreani, M., Jarvis, I., Li, X., Ratti, C., Seiferling, I., … Bosch, M. van den. (2018). How Green Is Green? Modeling Urban Greenness Exposure in Environmental Health Research. Environmental Health Perspectives. Retrieved from https://ehp.niehs.nih.gov/doi/10.1289/isesisee.2018.O01.03.32 Nichani, V., Dirks, K., Burns, B., Bird, A., Morton, S., & Grant, C. (2017). Green space and pregnancy outcomes: Evidence from Growing Up in New Zealand. Health & Place, 46, 21–28. https://doi.org/10.1016/j.healthplace.2017.04.007 Northridge, M. E., & Sclar, E. (2003). A Joint Urban Planning and Public Health Framework: Contributions to Health Impact Assessment. American Journal of Public Health, 93(1), 118–121. https://doi.org/10.2105/AJPH.93.1.118 Openshaw, S. (1984). The Modifiable Areal Unit Problem. Norwich: Geo Books. Parker, J., & Simpson, G. D. (2018). Public Green Infrastructure Contributes to City Livability: A Systematic Quantitative Review. Land, 7(4), 161. https://doi.org/10.3390/land7040161 Patel, V., Flisher, A. J., Hetrick, S., & McGorry, P. (2007). Mental health of young people: A global public-health challenge. The Lancet, 369(9569), 1302–1313. https://doi-org.ezproxy.library.ubc.ca/10.1016/S0140-6736(07)60368-7 Pebesma, E. (2018). Simple Features for R: Standardized Support for Spatial Vector Data. The R Journal, 10(1), 439–446. https://doi.org/10.32614/RJ-2018-009 Pettorelli, N., Schulte to Bühne, H., Shaprio, A., & Glover-Kapfer, P. (2018). Satellite remote sensing for conservation (4) (p. 125). Switzerland: [WWF] World Wildlife Fund Conservation Technology Series.  80  Pirgon, Ö., & Aslan, N. (2015). The Role of Urbanization in Childhood Obesity. Journal of Clinical Research in Pediatric Endocrinology, 7(3), 163–167. https://doi.org/10.4274/jcrpe.1984 Planet Labs. (2016, October). Planet Imagery Product Specification: Planetscope and RapidEye. Retrieved February 9, 2019, from https://www.planet.com/products/satellite-imagery/files/1610.06_Spec%20Sheet_Combined_Imagery_Product_Letter_ENGv1.pdf Qian, Y., Zhou, W., Yu, W., & Pickett, S. T. A. (2015). Quantifying spatiotemporal pattern of urban greenspace: New insights from high resolution data. Landscape Ecology, 30(7), 1165–1173. https://doi.org/10.1007/s10980-015-0195-3 R Core Team. (2018). R: A language and environment for statistical computing. Vienna, Austria. Retrieved from https://www.R-project.org/ Radesky, J. S., & Christakis, D. (2016). Increased Screen Time Implications for Early Childhood Development and Behavior. Pediatric Clinics of North America, 63(5), 827–839. https://doi.org/10.1016/j.pcl.2016.06.006 Ram, K., & Wickham, H. (2019). wesanderson: A Wes Anderson Palette Generato. Retrieved from https://CRAN.R-project.org/package=wesanderson Rhew, I. C., Vander Stoep, A., Kearney, A., Smith, N. L., & Dunbar, M. D. (2011). Validation of the Normalized Difference Vegetation Index as a Measure of Neighborhood Greenness. Annals of Epidemiology, 21(12), 946–952. https://doi.org/10.1016/j.annepidem.2011.09.001 Richardson, E. A., Pearce, J., Mitchell, R., & Kingham, S. (2013). Role of physical activity in the relationship between urban green space and health. Public Health, 127(4), 318–324. https://doi.org/10.1016/j.puhe.2013.01.004 Richardson, E. A., Pearce, J., Shortt, N. K., & Mitchell, R. (2017). The role of public and private natural space in children’s social, emotional and behavioural development in  81  Scotland: A longitudinal study. Environmental Research, 158, 729–736. https://doi.org/10.1016/j.envres.2017.07.038 Rojas-Rueda, D., Nieuwenhuijsen, M. J., Gascon, M., Perez-Leon, D., & Mudu, P. (2019). Green spaces and mortality: A systematic review and meta-analysis of cohort studies. The Lancet Planetary Health, 3(11), e469–e477. https://doi.org/10.1016/S2542-5196(19)30215-3 Rugel, E. J., Henderson, S. B., Carpiano, R. M., & Brauer, M. (2017). Beyond the Normalized Difference Vegetation Index (NDVI): Developing a Natural Space Index for population-level health research. Environmental Research, 159, 474–483. https://doi.org/10.1016/j.envres.2017.08.033 Ruggeri, D., Harvey, C., & Bosselmann, P. (2018). Perceiving the Livable City: Cross-Cultural Lessons on Virtual and Field Experiences of Urban Environments. Journal of the American Planning Association, 84(3–4), 250–262. https://doi.org/10.1080/01944363.2018.1524717 Russell, R., Chung, M., Balk, E. M., Atkinson, S., Giovannucci, E. L., Ip, S., … Lau, J. (2009). Issues and Challenges in Conducting Systematic Review to Support Development of Nutrient Reference Values: Workshop Summary: Nutrition Research Series, Vol. 2. Agency for Healthcare Research and Quality (US). Retrieved from https://www.ncbi.nlm.nih.gov/books/NBK44088/ Schott, J. R., Gerace, A., Woodcock, C. E., Wang, S., Zhu, Z., Wynne, R. H., & Blinn, C. E. (2016). The impact of improved signal-to-noise ratios on algorithm performance: Case studies for Landsat class instruments. Remote Sensing of Environment, 185, 37–45. https://doi.org/10.1016/j.rse.2016.04.015  82  Schutte, A. R., Torquati, J. C., & Beattie, H. L. (2017). Impact of Urban Nature on Executive Functioning in Early and Middle Childhood. Environment and Behavior, 49(1), 3–30. https://doi.org/10.1177/0013916515603095 Seymour, V. (2016). The Human–Nature Relationship and Its Impact on Health: A Critical Review. Frontiers in Public Health, 4. https://doi.org/10.3389/fpubh.2016.00260 Shah, P. S., & Balkhair, T. (2011). Air pollution and birth outcomes: A systematic review. Environment International, 37(2), 498–516. https://doi.org/10.1016/j.envint.2010.10.009 Sivarajah, S., Smith, S. M., & Thomas, S. C. (2018). Tree cover and species composition effects on academic performance of primary school students. PLOS ONE, 13(2), e0193254. https://doi.org/10.1371/journal.pone.0193254 Söderström, M., Boldemann, C., Sahlin, U., Mårtensson, F., Raustorp, A., & Blennow, M. (2013). The quality of the outdoor environment influences children’s health—A cross-sectional study of preschools. Acta Paediatrica, 102(1), 83–91. https://doi.org/10.1111/apa.12047 South, A. (2017). rnaturalearth: World Map Data for Natural Earth (Version 0.1.0). Retrieved from https://CRAN.R-project.org/package=rnaturalearth Statistics Canada. (2018, May 17). Canada goes urban. Government of Canada. Retrieved from https://www150.statcan.gc.ca/n1/pub/11-630-x/11-630-x2015004-eng.htm Stefanov, W., & Netzband, M. (2010). Characterization and Monitoring of Urban/Peri-urban Ecological Function and Landscape Structure Using Satellite Data. In T. Rashed & C. Jürgens (Eds.), Remote sensing of urban and suburban areas (1st ed.). Springer Science + Business Media. Suk, W. A., Murray, K., & Avakian, M. D. (2003). Environmental hazards to children’s health in the modern world. Mutation Research/Reviews in Mutation Research, 544(2–3), 235–242. https://doi.org/10.1016/j.mrrev.2003.06.007  83  Suzuki, K. (2018). The developing world of DOHaD. Journal of Developmental Origins of Health and Disease, 9(3), 266–269. https://doi.org/10.1017/S2040174417000691 Therneau, T. (2018). deming: Deming, Theil-Sen, Passing-Bablock and Total Least Squares Regression (Version 1.4). Mayo Clinic. Tillmann, S., Clark, A. F., & Gilliland, J. A. (2018). Children and Nature: Linking Accessibility of Natural Environments and Children’s Health-Related Quality of Life. International Journal of Environmental Research and Public Health, 15(6). https://doi.org/10.3390/ijerph15061072 Tjosvold, L., Campbell, S., & Dorgan, M. (2015, November 24). Filter to Retrieve Pediatric Articles in the OVID Medline Database. John W. Scott Health Sciences Library, University of Alberta. Tucker, C. J. (1979). Red and Photographic Infrared Linear Combinations for Monitoring Vegetation. Remote Sensing of Environment, 8, 127–150. Ulrich, R. (1984). View through a window may influence recovery from surgery. Science, 224(4647), 420–421. https://doi.org/10.1126/science.6143402 Ulrich, R. S., Simons, R. F., Losito, B. D., Fiorito, E., Miles, M. A., & Zelson, M. (1991). Stress recovery during exposure to natural and urban environments. Journal of Environmental Psychology, 11(3), 201–230. https://doi.org/10.1016/S0272-4944(05)80184-7 United Nations, E. & S. A. (2018). World Urbanization Prospects: The 2018 Revision. Retrieved from https://esa.un.org/unpd/wup/Publications/Files/WUP2018-KeyFacts.pdf UNICEF. (2012, February). The State of the World’s Children 2012: Children in an Urban World. Executive Summary. United Nations Children’s Fund. Retrieved from http://www.unicef.org/sowc2012  84  United Nations Department of Economic and Social Affairs. (2016). The World’s Cities in 2016. UN. https://doi.org/10.18356/8519891f-en USGS Landsat 5 TM TOA Reflectance (Orthorectified), 1984 to 2011. (2017, Accessed). Retrieved from https://explorer.earthengine.google.com/#detail/LANDSAT%2FLT5_L1T_TOA. USGS Landsat 8 TOA Reflectance (Orthorectified), 2013 to 2018. (2017, Accessed). Retrieved from https://explorer.earthengine.google.com/#detail/LANDSAT%2FLC8_L1T_TOA van den Berg, A. E., & Staats, H. (2018). How nature can affect health—Theories and mechanisms. In M. van den Bosch & W. Bird (Eds.), How nature can affect health—Theories and mechanisms, chapter in Nature and Public Health (1st ed.). Oxford, England: Oxford University Press. Retrieved from https://ebookcentral.proquest.com/lib/ubc/detail.action?docID=5205550. van den Bosch, M., & Nieuwenhuijsen, M. (2017). No time to lose – Green the cities now. Environment International, 99, 343–350. https://doi.org/10.1016/j.envint.2016.11.025 Vanaken, G.-J., & Danckaerts, M. (2018). Impact of Green Space Exposure on Children’s and Adolescents’ Mental Health: A Systematic Review. International Journal of Environmental Research and Public Health, 15(12), 2668. https://doi.org/10.3390/ijerph15122668 Villanueva, K., Pereira, G., Knuiman, M., Bull, F., Wood, L., Christian, H., … Giles-Corti, B. (2013). The impact of the built environment on health across the life course: Design of a cross-sectional data linkage study. BMJ Open, 3(1), e002482. https://doi.org/10.1136/bmjopen-2012-002482 Villeneuve, P. J., Ysseldyk, R. L., Root, A., Ambrose, S., DiMuzio, J., Kumar, N., … Rainham, D. (2018). Comparing the Normalized Difference Vegetation Index with the Google Street View Measure of Vegetation to Assess Associations between Greenness, Walkability, Recreational Physical Activity, and Health in Ottawa, Canada. International Journal of  85  Environmental Research and Public Health, 15(8), 1719. https://doi.org/10.3390/ijerph15081719 Wang, J., Larocque, H., & Stats Canada. (2019, February 11). Long-term population density change in Toronto and Vancouver, 1971 to 2016 [Stats Canada]. Retrieved February 4, 2020, from https://www150.statcan.gc.ca/n1/pub/16-508-x/16-508-x2019001-eng.htm Ward, J. S., Duncan, J. S., Jarden, A., & Stewart, T. (2016a). The impact of children’s exposure to greenspace on physical activity, cognitive development, emotional wellbeing, and ability to appraise risk. Health & Place, 40, 44–50. https://doi.org/10.1016/j.healthplace.2016.04.015 Weier, J., & Herring, D. (2000, August 30). Measuring Vegetation (NDVI & EVI) [Text.Article]. Retrieved January 20, 2020, from https://earthobservatory.nasa.gov/features/MeasuringVegetation/measuring_vegetation_1.php Wells, N. M. (2000). At Home with Nature: Effects of “Greenness” on Children’s Cognitive Functioning. Environment and Behavior, 32(6), 775–795. https://doi.org/10.1177/00139160021972793 Wells, N. M., & Evans, G. W. (2003). Nearby Nature: A Buffer of Life Stress among Rural Children. Environment and Behavior, 35(3), 311–330. https://doi.org/10.1177/0013916503035003001 Weng, Y.-C. (2007). Spatiotemporal changes of landscape pattern in response to urbanization. Landscape and Urban Planning, 81(4), 341–353. https://doi.org/10.1016/j.landurbplan.2007.01.009 White, J. C., Wulder, M. A., Hobart, G. W., Luther, J. E., Hermosilla, T., Griffiths, P., … Guindon, L. (2014). Pixel-Based Image Compositing for Large-Area Dense Time Series Applications  86  and Science. Canadian Journal of Remote Sensing, 40(3), 192–212. https://doi.org/10.1080/07038992.2014.945827 Wickham, H. (2016). ggplot2: Elegant Graphics for Data Analysis. New York: Springer-Verlag. Retrieved from https://ggplot2.tidyverse.org Williams, D. A. R., Matasci, G., Coops, N. C., & Gergel, S. E. (2018). An Object-Based Urban Landcover Mapping Methodology Using High Spatial Resolution Imagery and Airborne Laser Scanning. Journal of Applied Remote Sensing. Woolf, S. H., & Aron, L. (Eds.). (2013). Physical and Social Environmental Factors. In U.S. Health in International Perspective: Shorter Lives, Poorer Health (Vol. 181, pp. 945–946). Washington, DC: The National Academies Press. Retrieved from https://academic.oup.com/milmed/article/181/9/945-946/4159859 World Health Organization (Ed.). (2011). Health co-benefits of climate change mitigation: Housing sector. Geneva, Switzerland: Public Health & Environment Department, Health Security & Environment Cluster, World Health Organization. Wu, C.-D., McNeely, E., Cedeño-Laurent, J. G., Pan, W.-C., Adamkiewicz, G., Dominici, F., … Spengler, J. D. (2014). Linking Student Performance in Massachusetts Elementary Schools with the “Greenness” of School Surroundings Using Remote Sensing. PLoS ONE, 9(10), e108548. https://doi.org/10.1371/journal.pone.0108548 Wu, J., & Jackson, L. (2017). Inverse relationship between urban green space and childhood autism in California elementary school districts. Environment International, 107, 140–146. https://doi.org/10.1016/j.envint.2017.07.010 Yu, C.-Y., & Zhu, X. (2015). Impacts of Residential Self-Selection and Built Environments on Children’s Walking-to-School Behaviors. Environment and Behavior, 47(3), 268–287. https://doi.org/10.1177/0013916513500959  87  Zanella, A., Camanho, A. S., & Dias, T. G. (2015). The assessment of cities’ livability integrating human wellbeing and environmental impact. Annals of Operations Research, 226(1), 695–726. https://doi.org/10.1007/s10479-014-1666-7  88  Appendix A  A protocol for Chapter 2 Protocol for “the relation between natural environments and childhood mental health and development: A systematic review and assessment of different exposure measurements” 1. Objectives • Identify which natural environment (NE) measurements have most commonly been used in studies on the relation between NE and childhood mental health and development • Identify which NE measurement, including exposure rate (aspects of buffer size or other metric specific features that determine exposure rate) are most consistently related to mental health and development in children 2. Inclusion/ Exclusion criteria  2.1 PICOS  We will base the study characteristics off of several sources, including Hannes et al. (2007), Porritt et al. (2014), and The Joanna Briggs Institute (2014). Table A1 includes a list of the inclusion/exclusion criteria that will be used to determine the search criteria and determine study eligibility.  Table A1: Inclusion/Exclusion criteria following the PICOS protocol. Item (PICOS) Criteria Population • Infants, children, young children • Age 0-12, including prenatal  • Gender, male and female Intervention • Exposure to NE as measured by defined metrics, such as NDVI, land cover, audits, etc. • Exposure rate, as measured by various aspects of the metric, e.g. buffer, percent green/blue space, accessibility, time, etc. Comparison (if any) • Indicators of non-NE (built-environments) Outcomes • Childhood mental health and development, incl. birth outcome, as defined by disease classification, validated scales, or school achievement Study Design considerations • Randomized trials • Observational studies (case-control, longitudinal cohort, and cross-sectional)   89   2.2 Inclusion - Publication characteristics  • Primary study published in a peer-reviewed journal • Published between January 1, 2000 and July 31, 2018 • Written in English  2.3 Exclusion criteria • NE exposure only by pictures or other virtual displays • NE not assessed by defined measurement • Articles not written in English • Case studies or qualitative studies • Not primary research study • Studies not focused on childhood mental health or development outcome  3. Search Strategy 3.1 Limits Applied to Search • Articles published between January 1, 2000 and July 31, 2018.  • Limit to children ages 0 - 12, including prenatal  • Published in English   3.2 Databases and other sources • National Center for Biotechnology Information (MEDLINE) • Elsevier’s Excerpta Medicia Database (EMBASE) • Web of Science (WOS) • American Psychological Association (PyscINFO) • Snow-balling  90  • Additional sources, such as known articles, will also be included  3.3 Search terms Search terms will be generated from reviewing literature and choose similar MeSH terms (Table A2). MEDLINE and EMBASE databases were screened for MeSH terms that fit areas of interest. University of Alberta’s curated search filters (Tjosvold et al., 2015) will be used for children’s health and development search terms and modified to the other databases. Exact keywords will be used in all database searches.  Table A2: List of MeSH term and other keywords used in queries. Topic MeSH Term Keywords/Other terms Natural environments City Planning, Forests, TREES, Environmental Design, Urban Health, Environmental Planning  green space?, greenspace?, green?ess, blue space?, urban forest?, open space?, natural space?, green, public park?, vegetation, tree?, natur? GIS geographic information system, remote sensing, remote sensing technologies  geographic information system?, GIS, metric, remote sens?, land cover, normali?ed difference vegetation indes, NDVI, green view, GVI, EVI, enhanced vegetation index, OBIA, object based image analysis, lidar, landsat, sky view index, vertical visibility index, VVI, spatial, tree canopy change, google street view, GSV, landcover, accelerometery, global position, ecological momentary assessment, EMA  Children’s Health and Development  exp child/ or exp "congenital, hereditary, and neonatal diseases and abnormalities"/ or exp infant/ or adolescent/ or exp pediatrics/ or child, abandoned/ or exp child, exceptional/ or child, orphaned/ or child, unwanted/ or minors/ or (pediatric* or paediatric* or child* or newborn* or congenital* or infan* or baby or babies or neonat* or pre-term or preterm* or premature birth* or NICU or preschool* or pre-school* or kindergarten* or kindergarden* or elementary school* or nursery school* or (day care* not adult*) or schoolchild* or toddler* or boy or boys or girl* or middle school* or  91  pubescen* or juvenile* or teen* or youth* or high school* or adolesc* or pre-pubesc* or prepubesc*).mp. or (child* or adolesc* or pediat* or paediat*).jn  3.4 PRISMA diagram Following the PRISMA protocol (Liberati et al., 2009; Moher et al., 2009) we will identify and screen relevant articles for inclusion in the review. A PRISMA diagram will be used to demonstrate the article selection process.  4. Method review 4.1 Details of methods • Initial screening of titles and abstracts: ZD, back-up MvdB • Final selection of articles to be included: ZD and MvdB • One reviewer (ZD) will extract data (see 5.2. for details) from included articles, which will be reviewed by co-author (MG, IJ, MJ, LN, HS, TO, JS, or MvdB)  • ZD plus senior reviewer (MG, IJ, MJ, LN, HS, TO, JS, or MvdB) will separately conduct a quality assessment (section 4.2) • Disagreement in data extraction will be discussed with additional reviewer and resolved by consensus, a process that was similar to Ohly et al. (2016) 4.2 Quality assessment Following the STROBE Protocol (von Elm et al., 2007), two authors will independently conduct a quality assessment for each article. Criteria will be based on previous reviews (de Keijzer et al., 2016; Gascon et al., 2015) and modified to fit this review. Disagreement in quality assessment will be discussed with additional reviewers and resolved by consensus. For a full list of criteria, see Table A3. Quality will be determined by both the numeric score of each paper score and an overall judgement of the study’s quality (National Institute of Health, 2018). 92  Table A3: Criteria for quality assessment of the studies.   Criteria Possible scores Description  Number of NE metrics 0 = one metric 1 = multiple metrics considered One or multiple data sources used to determine NE exposure    Multiple buffer distances 0 = no 1 = yes Multiple distances from origin (residential address/school, etc.) Categories for assessing “green space” Choose 1 – 3 categories  On-site evaluation 0 = self-reported 1 = audit by expert 2 = expert deployed on-site measure Was “greenness” of the NE evaluated by a participant survey, or an audit, survey or site location chosen by an expert Remotely Sensed Indices 0 = remotely sensed (low resolution, >30 m) 1 = remotely sensed (moderate resolution, > 5 and ≤30m) 2 = remotely sensed (high resolution, ≤5m) Spatial resolution of satellite or aerial imagery used to produce related indices or measures. Multi-source land use/ land cover 0 = Low accuracy assessment (<60% accurate) or cannot be found 1 = Moderate accuracy assessment (60 – 80%) 2 = High accuracy assessment (>80%) Accuracy assessment provided in the background literature on production of dataset.   Temporal Resolution 0 = Not considered 1 = Considered Time was considered in the analysis in the form averaged exposure data or by acquisition dates corresponding with cohort  Quality of Green Space considered 0 = no 1 = yes (e.g. POSDAT) Quality was assessed by survey/appraisal/land use classification   Use of green space 0 = not measured and/or not included in the analysis 1 = measured and included in analysis Measures of health were based on the child being in or not being in green space. Includes GPS tracking, parent surveys.   93   Quality of Neighborhood Scale 0 = larger than a city block 1 = postcode/neighborhood block/ census tract 2 = individual address/school address Scale at which the initial exposure rate was calculated (e.g. home address vs nearest neighborhood boundaries)  Outcome assessment 0 = self-reported 1 = interviews conducted by expert or by an objective measure Process by which the health data was obtained  Study Design  0 = Observational Cross-sectional Studies • Cross-sectional 1 = Longitudinal Studies • Case-control studies • Longitudinal cohort • Before-and-after studies 2 = Non-randomized controlled trials • Quasi-randomized controlled trials • Controlled before-after studies (CBAs) • Interrupted time series studies (ITSs) 3 = Randomized Control Trials (RCTs) Informed by the Cochrane Study Guide protocol (Ryan et al. 2013) and Deeks et al. (2003)  Potential Confounding Factors 0 = no confounding factors considered 1 = some confounding factors considered but some key factors omitted 2 = careful consideration of confounders Study made clear efforts consider all possible confounders   Potential Bias  0 = study design/methods may have led to serious bias 1 = study design/methods may have led to some bias 2 = study design/methods may have led to little bias As pertaining to selection bias, performance bias, detection bias, attrition bias, and reporting bias using the Cochrane guidelines (Higgins and Green, 2011; Ryan et al., 2013)  Statistical Methods Used 0 = inappropriate statistical testing or interpretation of test 1 = appropriate tests conducted and interpretation of results   Effect Size 0 = incomplete information 1 = complete information (estimate and standard error/confidence interval)   94   Directness of study (NE) 0 = NE exposure one of many variables tested 1 = NE exposure primary variable tested NEs were the primary environmental exposure considered in the study. Moderation effects of NE on pollution sources not included.   Directness of study (mental health/development) 0 = mental health/development one of many outcome variables tested 1 = mental health/development primary outcome considered Mental health or development was the primary health outcome considered in the study   Participants have been living at least 1 year in the studied area 0 = no or not clearly specified 1 = yes Study specified if participants had lived in residence for at least 1 year.   95  5. Data extraction 5.1 Process • A citation manager (e.g., Zotero) will be used to document titles selected during the search. • A data extraction template will be developed and used in Excel • A quality assessment template will be developed and used in Excel  5.2 Data extracted from each study For each study, the information found in Table A4 will be extracted by one co-author (ZD) and this information will be reviewed and verified by another co-author (MG, IJ, MJ, LN, HS, TO, JS, MvdB).  Table A4: List and description of data extracted from each study. Item Description Authors Authors of study Title Study title Year published Year of publication Study design Design of study Study location Location of the study, as specific as possible Scale of study Description of how the NE exposure was defined Population Number of participants and age/age range of participants. Description of control population if applicable NE measures Including: • Primary sensor/source of the metric (e.g., Landsat) • Primary exposure parameters (e.g., NDVI) • Primary exposure rate (e.g., buffer sizes) • Other NE measures and rates included in study Health or Development outcome Including: • Measures considered (e.g., birth weight) • Collection methods (e.g., hospital records) Main results Including: • Reported significant estimated effects, adjusted effects included if possible • A descriptive narrative of main results  Statistical models used Type of model used Confounding factors List of confounding factors used in study, if reported Potential bias Potential sources of bias listed in study Strengths and limitations Strengths and limitations of the study as listed in the study Other relevant information For example, the use of defined cohorts  96   6. Data Synthesis • By NE metric  • By exposure rate • By health/development measure  7. Data Analysis • Main findings/strength of evidence for each main outcome by exposure metric • General interpretation of results (associative strength of each exposure metric) 8. Strengths and Weaknesses Limitations at study and outcome level (risk of bias) and at review level (incomplete retrieval of identified research, reporting bias)         97  References de Keijzer, C., Gascon, M., Nieuwenhuijsen, M.J., Dadvand, P., 2016. Long-Term Green Space Exposure and Cognition Across the Life Course: a Systematic Review. Curr. Environ. Health Rep. 3, 468–477. https://doi.org/10.1007/s40572-016-0116-x Deeks, J., Dinnes, J., D’Amico, R., Sowden, A., Sakarovitch, C., Song, F., Petticrew, M., Altman, D., 2003. Evaluating non-randomised intervention studies. Health Technol. Assess. 7. https://doi.org/10.3310/hta7270 Gascon, M., Triguero-Mas, M., Martínez, D., Dadvand, P., Forns, J., Plasència, A., Nieuwenhuijsen, M., 2015. Mental Health Benefits of Long-Term Exposure to Residential Green and Blue Spaces: A Systematic Review. Int. J. Environ. Res. Public. Health 12, 4354–4379. https://doi.org/10.3390/ijerph120404354 Hannes, K., Claes, L., The Belgian Campbell Group, 2007. Learn to Read and Write Systematic Reviews: The Belgian Campbell Group. Res. Soc. Work Pract. 17, 748–753. https://doi.org/10.1177/1049731507303106 Higgins, J.P., Green, S., 2011. Cochrane Handbook for Systematic Reviews of Interventions. Liberati, A., Altman, D.G., Tetzlaff, J., Mulrow, C., Gøtzsche, P.C., Ioannidis, J.P.A., Clarke, M., Devereaux, P.J., Kleijnen, J., Moher, D., 2009. The PRISMA Statement for Reporting Systematic Reviews and Meta-Analyses of Studies That Evaluate Health Care Interventions: Explanation and Elaboration. PLoS Med. 6, e1000100. https://doi.org/10.1371/journal.pmed.1000100 Moher, D., Liberati, A., Tetzlaff, J., Altman, D.G., 2009. Preferred Reporting Items for Systematic Reviews and Meta-Analyses: The PRISMA Statement. PLoS Med. 6, 6. National Insitute of Health, 2018. Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies [WWW Document]. URL https://www.nhlbi.nih.gov/health-topics/study-quality-assessment-tools Ohly, H., White, M.P., Wheeler, B.W., Bethel, A., Ukoumunne, O.C., Nikolaou, V., Garside, R., 2016. Attention Restoration Theory: A systematic review of the attention restoration potential of exposure to natural environments. J. Toxicol. Environ. Health Part B 19, 305–343. https://doi.org/10.1080/10937404.2016.1196155 Porritt, K., Gomersall, J., Lockwood, C., 2014. SYSTEMATIC REVIEWS, Step by Step. Syst. Rev. 114, 6. Ryan, R., Hills, S., Prictor, M., McKenzie, S., Cochrane Consumers and Communication Review Group, 2013. Study Quality Guide. Cochrane Consumers and Communication Review Group. The Joanna Briggs Institute, 2014. Joanna Briggs Institute Reviewers’ Manual: 2014 Edition. The Joanna Briggs Institute. Tjosvold, L., Campbell, S., Dorgan, M., 2015. Filter to Retrieve Pediatric Articles in the OVID Medline Database. von Elm, E., Altman, D.G., Egger, M., Pocock, S.J., Gøtzsche, P.C., Vanderbroucke, J.P., 2007. The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) Statement: guidelines for reporting observational studies. Lancet 370, 1453–57. https://doi.org/10.1016/S0140-6736(07)61602-X   98  Appendix B Supporting Tables for Chapter 2 Table B1: Search terms used organized by database. (S1a - EMBASE search strategy; S1b – MEDLINE search strategy; S1c – PsychINFO search strategy; S1d – Web of Science search strategy) a. EMBASE Database: Embase <1974 to 2018 August 10> Search Strategy:  Search term 1 city planning/ (1910) 2     forest/ (16847) 3  "tree"/ (27487) 4 environmental planning/ (9250) 5 (green space? or greenspace? or green?ess or blue space? or urban forest? or park? or natural environment? or open space? or natural space? or green or public park or vegetation or tree? or natur?).tw,kw. (776346) 6 urban health/ (478) 7 1 or 2 or 3 or 4 or 5 or 6 (803970) 8 geographic information system/ (8463) 9 remote sensing/ (5709) 10 (geographic information system? or GIS or metric or remote sens? or satellite image? or land cover or normali?ed difference vegetation index or NDVI or green view or GVI or EVI or enhanced vegetation index or OBIA or object based image analysis or lidar or landsat or sky view index or vertical visibility index or VVI or spatial or tree canopy change or google street view or GSV or landcover or accelerometry or global position or ecological momentary assessment or EMA).tw,kw. (315755) 11 8 or 9 or 10 (321629) 12 exp child/ or exp "congenital, hereditary, and neonatal diseases and abnormalities"/ or exp infant/ or adolescent/ or exp pediatrics/ or child, abandoned/ or exp child, exceptional/ or child, orphaned/ or child, unwanted/ or minors/ or (pediatric* or paediatric* or child* or newborn* or congenital* or infan* or baby or babies or neonat* or pre-term or preterm* or premature birth* or NICU or preschool* or pre-school* or kindergarten* or kindergarden* or elementary school* or nursery school* or (day care* not adult*) or schoolchild* or toddler* or boy or boys or girl* or middle  99  school* or pubescen* or juvenile* or teen* or youth* or high school* or adolesc* or pre-pubesc* or prepubesc*).mp. or (child* or adolesc* or pediat* or paediat*).jn. (4536610) 13 7 and 11 and 12 (2050) b. MEDLINE Database: Ovid MEDLINE(R) and Epub Ahead of Print, In-Process & Other Non-Indexed Citations and Daily <1946 to August 10, 2018> Search Strategy:  Search terms 1 City Planning/ (1989) 2 Forests/ (5343) 3 TREES/ (23443) 4 Environment Design/ (5635) 5 Urban Health/ (17163) 6 (green space? or greenspace? or green?ess or blue space? or urban forest? or park? or natural environment? or open space? or natural space? or green or public park? or vegetation or tree? or natur*).tw,kw. (1203451) 7 1 or 2 or 3 or 4 or 5 or 6 (1237470) 8 Geographic Information Systems/ (6882) 9 Remote Sensing Technology/ (2151) 10 (geographic information system? or GIS or metric or remote sens? or satellite image? or land cover or normali?ed difference vegetation index or NDVI or green view or GVI or EVI or enhanced vegetation index or OBIA or object based image analysis or lidar or landsat or sky view index or vertical visibility index or VVI or spatial or tree canopy change or google street view or GSV or landcover or accelerometry or global position or ecological momentary assessment or EMA).tw,kw. (299672) 11 8 or 9 or 10 (303252) 12 exp child/ or exp "congenital, hereditary, and neonatal diseases and abnormalities"/ or exp infant/ or adolescent/ or exp pediatrics/ or child, abandoned/ or exp child, exceptional/ or child, orphaned/ or child, unwanted/ or minors/ or (pediatric* or paediatric* or child* or newborn* or congenital* or infan* or baby or babies or neonat* or pre-term or preterm* or premature birth* or NICU or preschool* or pre-school* or kindergarten* or kindergarden* or elementary school* or nursery school* or (day care* not adult*) or  100  schoolchild* or toddler* or boy or boys or girl* or middle school* or pubescen* or juvenile* or teen* or youth* or high school* or adolesc* or pre-pubesc* or prepubesc*).mp. or (child* or adolesc* or pediat* or paediat*).jn. (4699199) 13 7 and 11 and 12 (2769) c. PsychInfo Search terms S1 DE “Urban Planning” S2 DE “Nature (Environment)” S3 DE “Environment Planning” S4 DE “Recreation Areas” S5 “green space#” or “greenspace#” or “green#ess” or “urban forest#” or “park#” or “natural environment#” or “open space#” or “natural space#” or “green” or “public park#” or “vegetation” or “tree#” or “natur#” S6 S1 OR S2 OR S3 OR S4 or S5 S7 DE “Information Systems” S8 DE “Geography” S9 “geographic information system#” or “GIS” or “metric” or “remote sens#” or “satellite image#” or “land cover” or “landcover” or “normali#ed difference vegetation index” or “NDVI” or “green view” or “GVI” or “EVI” or “enhanced vegetation index” or “OBIA” or “object based image analysis” or “lidar” or “landsat” or “vertical visibility index” or “VVI” or “spatial” or “tree canopy change” or “google street view” or “GSV” or “accelerometry” or “global position” or “ecological momentary assessment” or “EMA” S10 S7 OR S8 OR S9 S11 “child#” or “infan#” or “adolescen#” or “minor#” or “paediatric” or “newborn” or “baby” or “babies” or “neonat#” or “pre-term” or “preterm” or “premature birth” or “NICU” or “preschool#” or “elementary school” or “nursery school” or “kindergarten” or “kindergarden” or “day care” or “schoolchild#” or “toddler” or “boy” or “boys” or “girl” or “girls” or “middle school#’ or “pubescen#” or “juvenile#” or “teen” or “youth” or “high school#” or “adolesc#” or “pre-pubesc#” S12 S6 AND S10 AND S11 d. Web of Science Search terms # 1 (TS=(“city planning” OR “urban planning” OR “environmental planning” OR “nature” OR “environment” OR “green space” OR “greenspace” OR “greenspace” OR “green spaces” OR “greenspaces” OR “greenness” OR “blue space” OR “blue spaces” OR “urban forest” OR “urban forests” OR “park” OR “parks” OR “natural environment” OR “natural environments”  101  OR “open space” OR “open spaces” OR “green” OR “public park” OR “public parks” OR “vegetation” OR “tree” OR “trees” OR “natural” OR “recreation area” OR “recreation areas” OR “forest”)) # 2 (TI=(“city planning” OR “urban planning” OR “environmental planning” OR “nature” OR “environment” OR “green space” OR “greenspace” OR “greenspace” OR “green spaces” OR “greenspaces” OR “greenness” OR “blue space” OR “blue spaces” OR “urban forest” OR “urban forests” OR “park” OR “parks” OR “natural environment” OR “natural environments” OR “open space” OR “open spaces” OR “green” OR “public park” OR “public parks” OR “vegetation” OR “tree” OR “trees” OR “natural” OR “recreation area” OR “recreation areas” OR “forest”)) # 3  #2 OR #1 # 4 (TS=(“information system” OR “information systems” OR “geography” OR “geographic information systems” OR “geographic information system” OR “GIS” OR “metric” OR “remote sensing” OR “remote sense” OR “satellite image” OR “satellite images” OR “satellite imagery” OR “land cover” OR “landcover” OR “normalized difference vegetation index” OR “normalized difference vegetation index” OR “NDVI” OR “green view” OR “GVI” OR “EVI” OR “enhanced vegetation index” OR “OBIA” OR “object based image analysis” OR “lidar” OR “landsat” OR “vertical visibility analysis” OR “VVI” OR “spatial” OR “tree canopy change” OR “google street view” OR “GSV” OR “landcover” OR “accelerometry” OR “global position” OR “ecological momentary assessment” OR “EMA”))  # 5 (TI=(“information system” OR “information systems” OR “geography” OR “geographic information systems” OR “geographic information system” OR “GIS” OR “metric” OR “remote sensing” OR “remote sense” OR “satellite image” OR “satellite images” OR “satellite imagery” OR “land cover” OR “landcover” OR “normalized difference vegetation index” OR “normalized difference vegetation index” OR “NDVI” OR “green view” OR “GVI” OR “EVI” OR “enhanced vegetation index” OR “OBIA” OR “object based image analysis” OR “lidar” OR “landsat” OR “vertical visibility analysis” OR “VVI” OR “spatial” OR “tree canopy change” OR “google street view” OR “GSV” OR “landcover” OR “accelerometry” OR “global position” OR “ecological momentary assessment” OR “EMA”)) # 6 #5 OR #4 # 7 (TS=(“child” OR “children” OR “infant” OR “infancy” OR “infants” OR “adolescence” OR “adolescent” OR  102  “pediatric” OR “pediatrics” OR “paediatric” OR “paediatrics” OR “newborn” OR “baby” OR “babies” OR “neonatal” OR “neonatally” OR “pre-term” OR “preterm” OR “premature birth” OR “NICU” OR “preschool” OR “preschooler” OR “elementary school” OR “elementary schools” OR “preschools” OR “kindergarten” OR “kindergarden” OR “day care” OR “schoolchild” OR “schoolchildren” OR “toddler” OR “boy” OR “girl” OR “girls” OR “juvenile” OR “juveniles” OR “teen” OR “youth” OR “high school” OR “high schooler” OR “high schools” OR “pre-pubescent” OR “pre-pubescence” OR “prepubescent” OR “prepubescence”)) # 8 (TI=(“child” OR “children” OR “infant” OR “infancy” OR “infants” OR “adolescence” OR “adolescent” OR “pediatric” OR “pediatrics” OR “paediatric” OR “paediatrics” OR “newborn” OR “baby” OR “babies” OR “neonatal” OR “neonatally” OR “pre-term” OR “preterm” OR “premature birth” OR “NICU” OR “preschool” OR “preschooler” OR “elementary school” OR “elementary schools” OR “preschools” OR “kindergarten” OR “kindergarden” OR “day care” OR “schoolchild” OR “schoolchildren” OR “toddler” OR “boy” OR “girl” OR “girls” OR “juvenile” OR “juveniles” OR “teen” OR “youth” OR “high school” OR “high schooler” OR “high schools” OR “pre-pubescent” OR “pre-pubescence” OR “prepubescent” OR “prepubescence”)) # 9 #8 OR #7 # 10 #9 AND #6 AND #3  103  Table B2: Criteria for quality assessment of the studies. These criteria provided a numerical score for each study and was used in determining the quality of each study along with an overall judgement of quality based on the reviewer’s expertise. Criteria were modified from Gascon et al. (2015). For some studies, not all of the criteria were applicable, thus those criteria were removed from the numeric score as not to over count duplicate criteria.  Criteria Possible Scores Description   Number of NE metrics 0 = one metric 1 = multiple metrics considered One or multiple data sources used to determine NE exposure    Multiple buffer distances 0 = no 1 = yes Multiple distances from origin (residential address/school, etc.) Categories for assessing NE Choose 1 – 3 categories On-site evaluation 0 = self-reported 1 = audit by expert 2 = expert deployed on-site measure Was “greenness” evaluated by a participant survey (0), or an audit, survey or site location chosen by an expert (1), expert determined “greenness” while on location (2)  Remotely Sensed Indices 0 = remotely sensed (> 30 m) 1 = remotely sensed (> 5 and ≤ 30m) 2 = remotely sensed (≤ 5m) Spatial resolution of satellite or aerial imagery used to produce related indices or measures. Low resolution (0), medium resolution (1), or high resolution (2) Multi-source land use/ land cover 0 = Low accuracy assessment (< 60% accurate) or not traceable 1 = Moderate accuracy assessment (60 – 80%) 2 = High accuracy assessment (> 80%) Accuracy assessment provided in the background literature on production of dataset.   Temporal Resolution 0 = Not considered 1 = Considered Time was considered in the analysis in the form averaged exposure data or by acquisition dates corresponding with cohort  Quality of Green Space considered 0 = no 1 = yes (e.g., POSDAT) Quality was assessed by survey/appraisal/land use classification   Use of green space 0 = not measured and/or not included in the analysis 1 = measured and included in analysis Measures of health were based on the child being in or not being in NE. Includes GPS tracking, parent surveys.   Quality of Neighborhood Scale 0 = larger than a city block 1 = postcode/neighborhood block/ census tract 2 = individual address/school address Scale at which the initial exposure rate was calculated (e.g., home address vs nearest neighborhood boundaries)  Outcome assessment 0 = self-reported 1 = interviews conducted by expert or by an objective measure Process by which the health data was obtained  104   Study Design  0 = Observational Cross-sectional Studies • Cross-sectional 1 = Longitudinal Studies • Case-control studies • Longitudinal cohort • Before-and-after studies 2 = Non-randomized controlled trials • Quasi-randomized controlled trials • Controlled before-after studies (CBAs) • Interrupted time series studies (ITSs) 3 = Randomized Control Trials (RCTs) Informed by the Cochrane Study Guide protocol Ryan et al. (2013) and Deeks et al. (2003)   Potential Confounding Factors 0 = no confounding factors considered 1 = some confounding factors considered but some key factors omitted 2 = careful consideration of confounders Study made clear efforts to consider all possible confounders   Potential Bias (defined by Cochrane (Ryan et al., 2013)) 0 = study design/methods may have led to serious bias 1 = study design/methods may have led to some bias 2 = study design/methods may have led to little bias As pertaining to selection bias, performance bias, detection bias, attrition bias, and reporting bias (Higgins and Green, 2011; Ryan et al., 2013)  Statistical Methods Used 0 = inappropriate statistical testing or interpretation of test 1 = appropriate tests conducted and interpretation of results Studies that did not consider multicollinearity or spatial autocorrelations were adjudicated low score  Effect Size 0 = incomplete information 1 = complete information (estimate and standard error/confidence interval) Effect size was reported with confidence intervals in the study.   Directness of study (green space) 0 = green space exposure one of many variables tested 1 = green space exposure primary variable tested Green space was the primary environmental exposure considered in the study. Moderation effects of NE on pollution sources not included.   Directness of study (mental health/development) 0 = mental health/development one of many outcome variables tested 1 = mental health/development primary outcome considered Mental health or development was the primary health outcome considered in the study. Note that the study is not necessarily of lower quality for including more than one health outcomes, but for the purpose of this review, we considered the directness of the study as part of the determinants for assessing level of evidence.    Participants have been living at least 1 year in the studied area 0 = no or not clearly specified 1 = yes Study specified if participants had lived in residence for at least 1 year.     105  Table B3: Compiled quality rating for each study. Authors Number of NE metrics (0-1) Multiple buffer/distances  (0-1)  On- site evaluation (0-2)  Remotely Sensed Indices (0-2)  Multi -source land use/ land cover (0-2)  Temporal Resolution (0-1)  Quality of NE Considered (0- 1) Use of NE (0- 1) Quality of Neighborhood Scale (0-2) Outcome Assessment (0 -1) Study Design (0-3)  Confounding Factors (0-2)  Potential Bias (0-2)  Statistical Methods Used (0-1) Effect Size (0-1)  NE (0-1)  Mental Health/ Development (0-1)  Duration of residence (0-1) Score (absolute) Score (%) Overall Assessment (Good, Moderate, Poor) (Abelt & McLafferty, 2017) 1 1 NA 1 1 1 0 0 1 1 0 1 1 0 1 1 1 0 12 48 Poor (Agay-Shay et al., 2014) 1 1 NA 1 2 1 0 0 1 1 0 1 1 1 1 1 1 0 14 56 Moderate (Amoly et al., 2014) 1 1 NA 1 2 1 0 1 2 0 0 1 1 1 1 1 1 0 15 60 Moderate (Balseviciene et al., 2014) 1 1 NA 1 0 1 0 0 2 0 0 1 0 1 0 1 1 0 10 40 Poor (Beere & Kingham, 2017) 0 1 NA NA 2 1 0 0 0 1 0 0 0 1 1 1 1 0 9 38 Poor (Browning et al., 2018) 0 1 NA 0 NA 1 0 0 0 1 0 1 1 1 0 1 1 0 8 33 Poor (Casey et al., 2016) 0 1 NA 0 NA 1 0 0 2 1 0 1 2 1 1 1 1 0 12 50 Moderate (Christian et al., 2017) 1 0 NA NA 2 1 0 0 0 0 0 1 1 0 1 0 1 1 9 38 Poor (Cusack et al., 2017a) 1 1 NA 1 2 0 0 0 2 1 0 2 1 1 1 1 1 0 15 60 Good (Cusack et al., 2017b) 0 0 NA 0 NA 1 0 0 2 1 0 2 1 1 1 1 1 0 11 46 Moderate (Dadvand et al., 2012a) 1 1 NA 1 2 1 0 0 2 1 0 2 1 1 1 1 1 0 16 64 Good (Dadvand et al., 2015) 1 1 NA 2 NA 1 0 1 2 1 1 1 2 1 1 1 1 1 18 82 Good  106  Authors Number of NE metrics (0-1) Multiple buffer/distances  (0- 1)  On-site evaluation (0- 2)  Remotely Sensed Indices  (0-2)  Multi -source land use/ land cover (0-2)  Temporal Resolution (0-1)  Quality of NE Considered (0-1)  Use of NE (0-1)  Quality of Neighborhood Scale (0-2) Outcome Assessment (0 -1) Study Design (0- 3) Confounding Factors (0-2) Potential Bias (0-2) Statistical Methods Used (0-1)  Effect Size (0-1) NE (0- 1) Mental Health/ Development (0- 1) Duration of residence (0-1)  Score (absolute) Score (%)  Overall Assessment (Good, Moderate, Poor) (Dadvand et al., 2014a) 0 0 NA 0 NA 0 0 0 2 1 0 2 1 1 1 0 1 0 9 38 Moderate (Dadvand et al., 2012b) 0 1 NA 1 NA 1 0 0 2 1 0 1 1 1 1 1 1 0 12 50 Moderate (Dadvand et al., 2017) 1 1 NA 1 NA 1 0 0 2 1 1 2 1 1 1 1 1 1 16 67 Good (Dadvand et al., 2014b) 1 1 NA 1 2 0 0 0 2 1 0 2 1 1 1 1 1 0 15 60 Moderate (Donovan et al., 2011) 0 1 NA NA 0 1 0 0 2 1 0 1 1 0 1 1 1 0 10 42 Poor (Ebisu et al., 2016) 0 1 NA NA 1 1 0 0 2 1 0 1 1 1 1 1 1 0 12 50 Moderate (Engemann et al., 2018) 1 1 NA 1 NA 1 0 0 1 1 1 1 1 1 1 1 1 1 14 58 Moderate (Faber Taylor & Kuo, 2009) 0 NA 0 NA NA 1 0 1 2 1 2 0 0 1 0 1 1 NA 10 43 Poor (Faber Taylor & Kuo, 2011) 0 NA 0 NA NA 1 0 1 2 0 0 0 0 1 0 1 1 0 7 30 Poor (Faber Taylor et al., 2001) 0 NA 0 NA NA 1 0 1 1 0 0 0 0 1 0 1 1 0 6 26 Poor (Faber Taylor et al., 2002) 0 NA 0 NA NA 1 0 0 2 1 0 0 0 1 0 1 1 0 7 30 Poor (Feng & Astell-Burt, 2017) 0 0 NA NA 2 0 1 0 1 0 1 1 1 1 1 1 1 0 11 46 Moderate (Flouri et al., 2014) 0 0 NA NA 1 0 0 1 1 0 1 2 1 1 1 1 1 0 11 46 Moderate  107  Authors Number of NE metrics (0-1) Multiple buffer/distances  (0- 1)  On-site evaluation (0- 2)  Remotely Sensed Indices  (0-2)  Multi -source land use/ land cover (0-2)  Temporal Resolution (0-1)  Quality of NE Considered (0-1)  Use of NE (0-1)  Quality of Neighborhood Scale (0-2) Outcome Assessment (0 -1) Study Design (0- 3) Confounding Factors (0-2) Potential Bias (0-2) Statistical Methods Used (0-1)  Effect Size (0-1) NE (0- 1) Mental Health/ Development (0- 1) Duration of residence (0-1)  Score (absolute) Score (%)  Overall Assessment (Good, Moderate, Poor) (Fong et al., 2018) 0 0 NA 0 NA 1 0 0 2 1 0 2 0 1 1 1 1 0 10 42 Moderate (Glazer et al., 2018) 1 1 NA 1 2 0 0 0 2 1 0 1 0 1 1 1 1 0 13 52 Moderate (Grazuleviciene et al., 2015) 1 1 NA 1 0 0 0 0 2 1 0 2 1 1 1 1 1 1 12 48 Moderate (Hodson & Sander, 2017) 1 0 NA NA 1 1 0 0 0 1 0 1 0 1 1 1 1 0 9 38 Poor (Hystad et al., 2014) 1 1 NA 1 NA 1 0 0 1 1 0 1 0 1 1 1 1 0 10 42 Moderate (Kihal-Talantikite et al., 2013) 0 0 NA NA 0 0 0 0 0 1 0 1 0 1 0 1 1 0 5 21 Poor (Kim et al., 2016) 1 1 NA NA 1 0 1 0 2 0 0 0 0 1 0 1 0 0 8 33 Poor (Kuo & Faber Taylor, 2004) 0 NA 0 NA NA 1 0 1 2 0 0 1 0 1 1 1 1 0 9 39 Poor (Kuo et al., 2018) 1 1 NA NA 2 1 0 0 2 1 0 1 1 1 1 1 1 0 14 58 Moderate (Kweon et al., 2017) 1 0 NA NA 2 1 0 0 2 1 0 1 1 0 0 1 1 0 11 46 Moderate (Larson et al., 2018) 1 0 NA NA 1 1 0 0 1 1 0 2 1 1 1 1 1 0 12 50 Moderate (Laurent et al., 2013) 0 1 NA 1 NA 0 0 0 2 1 0 2 1 1 1 1 1 0 12 50 Moderate (MacNaughton et al., 2017) 0 1 NA 0 NA 1 0 0 2 1 0 1 1 0 1 0 1 0 9 38 Poor  108  Authors Number of NE metrics (0-1) Multiple buffer/distances  (0- 1)  On-site evaluation (0- 2)  Remotely Sensed Indices  (0-2)  Multi -source land use/ land cover (0-2)  Temporal Resolution (0-1)  Quality of NE Considered (0-1)  Use of NE (0-1)  Quality of Neighborhood Scale (0-2) Outcome Assessment (0 -1) Study Design (0- 3) Confounding Factors (0-2) Potential Bias (0-2) Statistical Methods Used (0-1)  Effect Size (0-1) NE (0- 1) Mental Health/ Development (0- 1) Duration of residence (0-1)  Score (absolute) Score (%)  Overall Assessment (Good, Moderate, Poor) (Markevych et al., 2014b) 1 1 NA 1 2 1 0 0 2 0 0 2 1 1 1 1 1 1 16 64 Good (Markevych et al., 2018) 0 0 NA 0 NA 1 0 0 0 1 1 2 1 1 1 1 1 1 11 46 Moderate (Markevych et al., 2014a)  1 1 NA 1 2 0 0 0 2 1 0 2 1 1 1 1 1 0 15 60 Moderate (Mårtensson et al., 2009) 1 NA 2 2 NA 1 1 1 2 0 0 1 1 1 0 1 1 0 15 63 Moderate (McCracken et al., 2016) 1 0 NA NA 2 0 0 1 1 0 0 1 1 1 1 1 1 0 11 46 Poor (McEachan et al., 2018) 0 1 NA 1 NA 1 1 1 2 0 0 2 1 1 1 1 1 0 14 58 Moderate (Nichani et al., 2017) 0 NA NA NA 2 0 0 0 0 1 0 2 1 1 1 1 1 1 11 48 Moderate (Richardson et al., 2017) 1 0 NA NA 2 0 0 0 2 0 1 1 1 1 1 1 1 1 13 58 Good (Schutte et al., 2017) 0 NA 1 NA NA 1 0 0 2 1 2 1 1 1 0 1 1 NA 12 55 Moderate (Sivarajah et al., 2018) 1 0 NA NA 2 0 0 0 2 1 0 1 1 1 0 1 1 0 11 46 Moderate (Söderström et al., 2013) 0 NA 2 NA NA 1 1 1 2 1 0 1 0 1 0 1 1 0 12 50 Moderate (Tillmann et al., 2018) 1 0 NA 1 2 1 0 0 2 0 0 2 1 1 0 1 1 0 13 52 Moderate (Ward et al., 2016) 0 NA 0 NA 2 1 0 1 2 1 0 0 1 1 0 1 0 0 10 42 Poor  109  Authors Number of NE metrics (0-1) Multiple buffer/distances  (0- 1)  On-site evaluation (0- 2)  Remotely Sensed Indices  (0-2)  Multi -source land use/ land cover (0-2)  Temporal Resolution (0-1)  Quality of NE Considered (0-1)  Use of NE (0-1)  Quality of Neighborhood Scale (0-2) Outcome Assessment (0 -1) Study Design (0- 3) Confounding Factors (0-2) Potential Bias (0-2) Statistical Methods Used (0-1)  Effect Size (0-1) NE (0- 1) Mental Health/ Development (0- 1) Duration of residence (0-1)  Score (absolute) Score (%)  Overall Assessment (Good, Moderate, Poor) (Wells, 2000) 0 NA 1 NA NA 1 0 0 2 1 0 1 1 1 0 1 1 0 10 43 Poor (Wells & Evans, 2003) 0 NA 1 NA NA 1 0 0 2 0 0 1 1 1 0 1 1 0 9 39 Poor (Wu et al., 2014) 0 1 NA 0 NA 1 0 0 0 1 0 2 1 1 1 1 1 0 10 42 Poor (Wu & Jackson, 2017) 1 1 NA NA 1 1 0 0 0 1 0 2 1 1 1 1 1 0 12 50 Moderate  110  Table B4: Description of the main characteristics of NE exposure and rate and mental health or development outcome with main, significant results divided into health outcome categories, birth outcomes, academic achievement, mental disorder and cognitive development, attention and social functioning.  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Birth Outcomes Abelt and McLafferty, 2017 (USA) Cross sectional 0 (infants) 103,484 children Greenness around population-weighted centroid of mother's census tract  Number of street trees around population-weighted centroid of mother's census tract  Distance to nearest major green space  Distance to publicly accessible waterfront Average NDVI (Landsat 5 TM pre-calculated NDVI layer from USGS derived from single image) within 250 m and 500 m buffers  Number of street trees (2005-06 NYC Street Tree Census) within 100 m, 150 m and 500 m buffers  Binary variable indicating residences was within 800 m network distance of a green space greater than 5000 m2 (MapPLUTO 16v1)  Binary variable indicating if residence was within 800 m network distance to a publicly accessible waterfront (NYC Waterfront Parks and Publicly Accessible Waterfront Spaces, PAWS) • Birth weight (term, low, SGA) • Gestational age (preterm) SGA (250 m): OR = 1.71 (CI:1.11, 2.64) Increased risk of SGA in deprived areas and NDVI.  SGA (500 m): OR = 2.02 (CI:1.21, 3.35) Increased risk of SGA in deprived areas and NDVI Agay-Shay et al., 2014 (Israel) Cross sectional 0 (infants) 39,132 children Greenness around center point of residential street  Distance to nearest major green space Average NDVI (from single Landsat 7 ETM+ image) within 100 m, 250 m and 500 m buffers  Binary variable indicating if address was within 300 m of a green space larger than 5000 m2 (OpenStreetMap) • Birth weight (low birth weight, very low birth weight) • Gestational age (preterm, very preterm) Birth weight: a 1-IQR increase in surrounding greenness (NDVI, 250 m) was associated with increased birth weight (19.2g, CI: 13.3, 25.1) and decreased risk for low birth weight (OR=0.84, CI: 0.78, 0.90)   Birth weight: Living within 300 m of green space larger than 5000 m2 associated with increased birth weight (18.1g CI: 8.7, 27.6) and decreased risk of low birth weight (OR=0.89, CI:0.8, 1.01). In the main analyses no other significant findings in adjusted models.                       Stratification showed stronger effect in low SES groups for increased birth weight and decreased risk of low birth weight.                                                        111  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Casey et al., 2016 (USA) Cross sectional 0 (infants) 16,913 mother-infant pairs Greenness around residential address Average NDVI (MODIS composite) within 250 m and 1250 m buffers of residential address, weighted for seasons during pregnancy • Birth weight (term, SGA) • Gestational age (preterm birth) • Low 5 min APGAR score Preterm birth: OR=0.78 (CI:0.61, 0.99) for greenness tertiles 2-3 vs 1 SGA: OR=0.73 (CI:0.58, 0.97) for greenness tertiles 2-3 vs 1 Cusack et al., 2017a (USA) Cross sectional 0 (infants) 3,026,603 Greenness around maternal residential address at time of birth Average NDVI (MODIS 16-day composite images over pregnancy) within 250 m buffer • Birth weight (full term, SGA) • Gestational age (preterm birth) Term birth weight (exposure during first trimester): β=2.7 (CI:1.2, 4.3) comparing Q2 to Q1;  β=2.5 (CI: 0.9, 4.2) comparing Q3 to Q1;  β=1.9 (CI: 0.1, 3.7) comparing Q4 to Q1. Term birth weight (exposure during third trimester): β=2.2 (CI: 0.6, 3.8) comparing Q2 to Q1;  β=2.8 (CI: 1.0, 4.5) comparing Q3 to Q1;  β=2.3 (CI: 0.4, 4.2) comparing Q4 to Q1.  *Some similar and significant but slightly inconsistent results when stratifying per different cities and income/ethnicity/etc. Cusack et al., 2017b (USA) Cross sectional 0 (infants) 88, 807 (TX) 90,265 (OR) Greenness around residential address Average NDVI (Annual average Landsat 5) within 50 m, 100 m, 250 m and 500 m buffers  Percent green space and percent tree canopy  within 50 m,  100 m, 250 m and 500 m buffers (EPA EnviroAtlas)  Percent street trees were assessed by the proportion of tree canopy with 26 m of a busy road  Binary variable indicating if residences was within 300 m of a park (EPA EnviroAtlas) • Birth weight (term) Term birth weight (Portland): β=12.6 (CI: 3.6, 21.4) comparing Q4 to Q1 NDVI (50 m buffer);  β=12.9 (CI: 4.4, 21.4) comparing Q4 to Q1 % green space  Term birth weight (Austin): β=-8.5 (CI: -16.9, -0.01) comparing Q4 to Q1 NDVI (250 m buffer);  β=13.6 (CI: -22.5, -4.6) comparing Q4 to Q1 NDVI (1000 m);  β=-14.7 (CI: -23.0, -6.3) comparing Q4 to Q1 % tree cover (1000 m buffer)  *After stratification results changed somewhat, in general for Portland: results remained significant for NDVI (50 m) in higher income and education groups, white, non-smokers, and in areas with higher population density. In general, for Austin the results remained significant and negative in lower education, non-white, non-smokers, and lower population density.  Dadvand et al., 2012a (Spain) Cross sectional 0 (infants) 8,246 Greenness around maternal residential address  Distance to NE Average NDVI (single Landsat 7 ETM+ image) within 100 m buffer  Binary variable indicating if residence was within 500 m of a major green space (Ecologic map of Barcelona) • Birth weight Birth weight (in group with no education completed): 436.3g (CI: 43.1, 829.5) per 10% increase of NDVI in 100 m buffer zones; 189.8g (CI:23.9, 355.7) if living within 500 m from park.   Birth weight (in group with secondary school completed): -36.5g (CI: -72.2, -0.7) if living within 500 m from park.   112  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Dadvand et al., 2012b (Spain) Cross sectional 0 (infants) 2,393 Greenness around residential address at time of delivery Average NDVI (single Landsat 4-5 TM images for the cities of Asturias, Gipuzkoa, Sabadell, and Valencia) within 100 m, 150 m and 500 m buffers • Birth weight • Gestational age • Head circumference Birth weight: 36.1g (CI: 16.4, 55.7), 38.3g (CI: 17.1, 59.5), and 44.2g (CI: 20.2, 68.2) more per an IQR increase in average NDVI in 100 m, 250 m, and 500 m respectively.  Smaller increase when adjusting for NO2 (28.5, 34.4g).  Head circumference: 1.2mm (CI: 0.4, 2.0), 1.4mm (CI: 0.4, 2.3), 1.7mm (CI: 0.5, 2.9) more per an IQR increase in average NDVI in 100 m, 250 m, and 500 m respectively. Almost the same when adjusting for NO2 (1.2-1.6mm). Dadvand et al., 2014a (Spain) Cross sectional 0 (infants) 6,438 Road-adjacent tree cover around residential address Road adjacent tree coverage was calculated as the average percent of tree coverage within 200 m of each side of the road (vegetative continuous field, MODIS derived) • Birth weight (low birth weight)  • Small for gestational age No consistent or significant finding of a buffering effect of street trees on harmful effect of proximity to major road.  Dadvand et al., 2014b (UK) Cross sectional 0 (infants) 10,780 Greenness around maternal residential address  Proximity to a major green space Average NDVI (single Landsat 4-5 TM image) within 50 m, 100 m, 250 m, 500 m and 1000 m buffers  Residential proximity was defined as living within 300 m of a major park (Urban Atlas) •  Birth weight Birth weight: in fully adjusted, non-stratified models (including ethnicity and SES indicators - individual + neighborhood) 14.0g (CI:-0.8, 28.7), 15.8g (CI:1.1, 30.6), 16.2g (CI:1.7, 30.8), 15.8g (CI:0.9, 30.7), 12.7 (CI: -1.8, 27.2) higher birth weight per an IQR increase in average NDVI in 50 m, 100 m, 250 m, 500 m, 1000 m buffer zones, respectively.  *Summary of stratified analyses suggest that white British get significant effects, but not Pakistani or other. Only most deprived get significant effects.  Donovan et al., 2011 (USA) Cross sectional 0 (infants) 5,696 Tree canopy around maternal residential address Percent of tree canopy within 50 m, 100 m and 200 m buffers (Metro-Portland land cover) • Preterm birth • Gestational age (less than 37 weeks) • Small for gestational age Small for gestational age: a higher percentage of tree canopy cover within 50 m from residence decrease the risk of SGA (OR: 0.99, CI: 0.98, 1.0). Marginal effect: 10% increase in tree canopy cover reduced the number of SGA by 1.42 per 1000 births.  Ebisu et al., 2016 (USA) Cross sectional 0 (infants) 239,811 Proportion of NE around maternal address at time of birth Proportion of urban, urban-open or green land use within 250 m buffer (National Land Cover Database, NLCD) •  Birth weight • Low birth weight • Small for gestational age small for gestational age Birth weight: 3.2 g (CI: 0.4, 6.0) more per an IQR increase in green space.  LBW: 7.6% (CI: -12.4, -2.6) less per an IQR increase in green space.  SGA: 2.6% (CI: -5.1, -0.1) less per an IQR increase in green space.  Fong et al., 2018 (USA) Cross sectional 0 (infants) 780,435 Greenness around maternal residential address Average NDVI (MODIS) from pixel containing residential address during pregnancy duration •  Birth weight • Term low birth weight  • Small for gestational age Birth weight: 6.92 g (CI:5.42, 8.42) more per 0.1 increase in NDVI in the NDVI range 0.25-0.50 Birth weight: 2.28 g (CI: 0.18, 4.38) more per 0.1 increase in NDVI in the NDVI range 0.50-0.75 TLBW: OR=0.98 (CI: 0.97, 1.00) per 0.1 increase in NDVI SGA: OR=0.98 (CI: 0.98, 0.99) per 0.1 increase in NDVI  113  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Glazer et al., 2018 (USA) Cross sectional 0 (infants) 61,640 mother-infant pairs Greenness around maternal residential address  Distance to NE and water Average NDVI (from single Landsat 7 scene) within 150 m, 200 m and 500 m buffers  Binary variable indicating if residences was within 500 m of a recreational facility (Statewide Comprehensive Outdoor Recreation Plan - SCORP)  Binary Variable indicating if residence was within 500 m of inland freshwater body (RI state water layer)  Binary variable indicating if residence was within 500 m or 1000 m of the coast (RI state water layer) • Birth weight • Gestational age • Preterm birth • Small for gestational age Birth weight: 7.4 g (CI:0.4, 14.4) more if living within 500 m of a freshwater body (compared to further away) Grazuleviciene et al., 2015 (Lithuania) Cross sectional 0 (infants) 3,292 mothers-infant pairs Greenness around maternal residential address  Distance to NE Average NDVI (single Landsat 5 TM image) within 100 m, 300 m and 500 m buffers  Binary variable indicating if residence was within 300 m of a city park (Urban Atlas for Kaunas)  Relative distance to city park (Urban Atlas for Kaunas) from each maternal address • Birth weight • Low birth weight • Preterm birth • Gestational age • Small for gestational age Low birth weight: increased risk, OR= 2.23, (CI:1.20, 4.15) when NDVI in 500 m buffer zones was less than or equal to the median NDVI and distance to city park greater than 1000 m Term low birth weight: increased risk, OR= 2.97 (CI: 1.04, 8.45) when NDVI in 500 m buffer zones less than or equal to the median NDVI and distance to city park was greater than 1000 m Preterm birth: increased risk OR =1.77, 1.10–2.81)when NDVI in 500 m buffer zones less than or equal to the median NDVI and distance to city park was greater than 1000 m Hystad et al., 2014 (Canada) Cross sectional 0 (infants) 64,705 Greenness around 6-digit postal code that contained the maternal residential address Average NDVI (annual average Landsat 7 ETM+ image) within 100 m and 250 m buffers   Walkability of neighborhood and distance to parks from a walkability index • Very preterm birth weight • Moderately preterm birth weight • Term birth weight • Small for gestational age Very preterm birth: OR=0.91 (CI: 0.77, 1.07) per 0.1 unit increase in NDVI (100 m buffer zone); OR = 0.80 (CI: 0.55, 1.18) in highest quartile NDVI (100 m) compared to lowest. Moderately preterm birth: OR= 0.95 (CI: 0.86, 1.06) in Q 2 and OR = 0.95(CI: 0.86, 1.06) in Q3 Term birth weight: β= 20.6g (CI: 16.5, 24.7) per 0.1 unit increase; β= 44.6g (CI: 34.8, 54.4) in highest IQR; β= 5.3g (CI:2.7, 7.9) per 300m distance to park. SGA: OR=0.97 (0.94-1.00) per 0.1 unit increase in NDVI (100 m); OR = 0.95 (CI:0.88, 1.03) in highest quartile   114  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Kihal-Talantikite et al., 2013 (France) Cross sectional 0 (infants) 715 deaths Proportion of NE within census area Proportion of green space including parks and forest within the census block containing the residential address (Spatial Land Cover Dataset for Lyon) • Excess infant mortality Excess infant mortality: there is some evidence that there is a relationship between greenness level and clustering in infant mortality (log LL=10.6, compared to 12, p = 0.01) Laurent et al., 2013 (USA) Cross sectional 0 (infants) 81,186 Greenness around maternal residence Average NDVI (Landsat 5 and 7) within 50 m, 100 m and 150 m buffers • Birth weight • Preterm deliveries • Preeclampsia Birth weight: 4.83g - 6.64g increase in models adjusted for various pollutants and traffic density (9 different models), all significant values, CI in these two models: (1.93, 7.72 and 3.27, 10.00 respectively) per an IQR increase in NDVI (50 m buffer zones) Very small increases and fewer significant values in 100 and 150 m buffer zones.  Markevych et al., 2014a (Germany) Cross sectional 0 (infants) 3,203 Greenness around maternal address at birth Average NDVI (single Landsat 5 TM image) within 100 m, 150 m, 500 m and 800 m buffers  Area of neighborhood green space (parks and forest) within 500 m buffer (Bavarian Land Use Dataset) • Birth weight Birth weight: 22.8 g (CI: 0.8, 44.8) increase per an IQR increase in NDVI (100 m buffer), adjusted for population density Birth weight: 18.3 - 31.9g increase per an IQR increase in NDVI (250 m buffer) adjusted for pollutants and population. Density, also in 500 m and 800 m buffers significant increase 17.6 g and up to 38.3g, depending on how the models are adjusted. Highest effect in low maternal education level in buffer zone 500 m: 58.2 g increase (CI: 2.0, 114.4) per IQR NDVI. Nichani et al., 2017 (New Zealand) Cross sectional 0 (infants) 6,853 Proportion of greenness within census area of maternal address Proportion of green space (parks, beach, urban parklands/open spaces, forest, grasslands, croplands, and other green areas) within census area unit (CAU) (New Zealand Land Cover Database LCDB) • Birth weight • Gestational age Gestational age: Increase 0.33 of a week (CI: 0.11, 0.56) per an IQR increase green space (land cover) in neighborhood among mothers of low education.  Academic Achievement Beere and Kingham, 2017 (New Zealand) Cross sectional Ecological Age up to grade 6 838 schools Greenness within school enrollment boundaries Proportion of NE within 2653 m (average school zone) of school address (Land cover Database LCDB3.3) • Ministry of Education (MOE) Standards Math: β = -0.145 (CI: -0.2-0.09) (P<0.001) within school ground ß = -0.075 (CI: -0.1-0.05) (P<0.001) within total in-zone  Reading: ß = -0.145 (CI: -0.22-0.07) (p <0.001) within school ground ß = -0.046 (CI:-0.09-0.01) (p=0.022) within total in-zone  Writing: ß = -0.077 (CI:-0.13-0.03) (p=0.004) within school ground  115  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Browning et al., 2018 (USA) Cross sectional Ecological 8-9 years old 404 schools Greenness around school address Average NDVI (MODIS composite) within 250 m buffer • Standardized test scores (MA) Both reading and math were negatively associated with NDVI (buffer sizes 250-2000 m), β varying from -0.027 to -0.051, p<0.01, depending on buffer size and month.  Hodson and Sander, 2017 (USA) Cross sectional Ecological 8 - 9 years old (third graders) 222 schools Proportion of NE within school attendance areas  Proportion of water within school attendance areas Proportion of grass, shrub and impervious cover in each school attendance area (SAA) (National Land Cover Dataset, NLCD)  Proportion of water within each school attendance area (USGS National Hydrography Dataset) • State comprehensive standardized test (MN) Reading: β=0.12 (p<0.01) with a linear increase in average percent canopy cover Reading: β=0.10 (p<0.05)with a linear increase in average percent impervious surface Percentage of students exceeding basic reading standard: β=0.27 (p,0.05) with a linear increase in average percent canopy cover Percentage of students exceeding basic reading standard: β=0.23 (p,0.05) with a linear increase in average percent impervious surface Kuo et al., 2018 (USA) Cross sectional Ecological 8-9 years old  (third graders) 318 schools Proportion of NE within school boundaries and catchment areas Proportion of grass and tree cover within school catchment (polygon of school catchment where student lives), school (school boundary with a 25 m buffer) and neighborhood (the catchment  area minus  school)  (Chicago Urban Tree Canopy Assessment)  •  Standardized test (ISAT, IL) Math: β=0.22 (p<0.05) with increased percent of school trees.  Kweon et al., 2017 (USA) Cross sectional Grades 2-10 (ages 7-16) 219 schools Proportion of NE within school boundary Proportion of tree cover, grass and shrub cover, bare soil, paved surfaces and buildings within each school parcel (Land Cover-Washington DC) • DC Comprehensive Assessment System Percent students being proficient/advanced in math: β=0.23 (p<0.01) with increased percent of trees Percent students being proficient/advanced in reading test: β=0.22 (p<0.01) with increased percent of trees MacNaughton et al., 2017 (USA) Cross sectional Ecological All public elementary and secondary schools (5-18) 1,772 schools Greenness around school address Average NDVI (composite MODIS) for the pixel of school location •  School Absenteeism (MA) Chronic absenteeism: Effect size 2.6% lower rate per an IQR increase in NDVI (β= -1.68, p<0.001, SE: 0.085)   116  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Sivarajah et al., 2018 (Canada) Cross sectional Ecological Grades 3 and 6 (ages 8-12) 387 schools Proportion of green space/ tree canopy around school Proportion of each school boundary covered by total soft surface, tree canopy, and the ratio of tree canopy to total land area that could hold vegetation (Tree Inventory (TDSB Neighbourhood Works) and Urban Tree Canopy Assessment for Toronto) • Education Quality and Accountability Office records on student performance, students above the provincial standard (Ontario) Writing: β = 16.25 (p˂0.05) increase in grade 6 students passing provincial standards per percent tree cover increase All test mean: β = 15.95 (p=0.05) increase in grade 6 students passing provincial standards per percent tree cover increase Wu et al., 2014 (USA) Cross sectional Ecological Grade 3 8-9 years old 905 schools Greenness around school address Average NDVI (MODIS) within 250 m, 500 m, 1000 m and 2000 m buffers  • Standardized tests for reading and math scores (MA Comprehensive Assessment System) English skills: associated with greenness across buffer zones (250, 500, 1000, 2000 m) from β = 0.19 (CI: 0.16, 0.21) to β = 0.42 (CI: 0.38, 0.46) in March. β = 0.04 (CI: 0.01, 0.08) to β = 0.06 (CI: 0.04, 0.08) in July. β = -0.12 (CI: -0.21, -0.14) to β = -0.06 (CI: -0.08, -0.03) in October.   Math skills: associated with greenness across buffer zones (250, 500, 1000, 2000 m) from β = 0.20 (CI: 0.16, 0.23) to β = 0.32 (CI: 0.27, 0.37) in March. β = 0.05 (CI: 0.02, 0.09) to β = 0.09 (CI: 0.07, 0.12) in July. β = -0.11 (CI: -0.15, -0.07) to β = -0.04 (CI: -0.10, -0.01) (P-value < 0.05) in October.    Mental Disorder Engemann et al., 2018 (Denmark) Longitudinal 10 years old to onset of disease, death, or emigration  943,027 Greenness around residential address Average NDVI (Landsat 4/5 TM, 8 OLI) within 210 m, 330 m, 570 m and 930 m square buffers • Onset of schizophrenia Schizophrenia prevalence: Increased risk (IRR = 1.43, CI: 1.28, 1.60) in lowest compared to highest green space decile NDVI (calculated within 210x210m from residence) (adjusted for SES) Markevych et al., 2018 (Germany) Longitudinal 10-14 years old 66,823 Greenness within census area of residence NDVI (MODIS) was calculated within postal code areas • ADHD diagnoses prevalence (ICD-10-GM F90) ADHD prevalence: RR=0.82 (CI:0.68, 0.98) per 0.1 unit increase in NDVI  117  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Wu and Jackson, 2017 (USA) Cross sectional Ecological Kindergarten - Grade 5 5-12 years old 543 public elementary schools Proportion of green space/ tree canopy around school General land cover composition (forest, grassland and urban land), general tree canopy and near road tree canopy (50 m buffer) around each school district (NLCD) • Autism diagnosis Autism prevalence: Rate ratios (RR) for forest, RR= 0.90 (CI: 0.84, 0.95); for grassland RR =0.90 (CI: 0.83, 0.97); for average tree canopy, RR =0.89 (CI: 0.83, 0.95), and for near-road tree canopy, RR =0.81 (CI: 0.73, 0.91). All per 10% increase in unit Cognitive Development / Attention / Stress (CDAS) Amoly et al., 2014 (Spain) Cross sectional 7-10 years old 2,111 Greenness around residential address  Distance to nearest major green space  Amount of time each child spent playing outside Average NDVI (from single Landsat 5 TM image) within 100 m, 250 m and 500 m buffers  Binary variable if residence was within 300 m of a green space larger than 0.5 km2 (Ecological Map of Barcelona)  Annual number of hours spent playing in green spaces compiled from a questionnaire filled out by parents  • SDQ (emotional symptoms, conduct problems, hyperactivity/inattention, peer relationship problems, prosocial behavior) completed by parents  • ADHD/DM-IV (inattention, hyperactivity-impulsivity symptoms) completed by teachers TDS (SDQ): 4.8% (CI: -8.6, -0.9) less with an IQR increase in green space playing time, 3.9% (CI: -7.2, -0.4) less with an IQR increase in annual beach attendance.  Emotional Symptoms (SDQ): 8.2% (CI: -13.9, -2.2) less with an IQR increase in green space playing time.  Peer relationship problems (SDQ): 15.4% (CI:-22.7, -7.4) less with an IQR increase in green space playing time, 16.8% (CI: -23.4, -9.7) less with an IQR increase in annual beach attendance. Pro-social behavior (SDQ): 1.1% (CI: 0.0, 2.2) more with an IQR increase in annual beach attendance.  TDS (SDQ): 3.6% (CI: -6.6, -0.6), 3.8% (CI: -6.4, -1.2), 4.0% (CI: -6.7, -1.2) less with an IQR increase in average NDVI in 100, 250, and 500 m buffer zones respectively.  Hyperactivity/inattention (ADHD/DM-IV): 5.0% (CI: -8.2, -1.6), 4.5% (CI: -7.4, -1.6), 4.1% (CI: -7.1, -1.0) less with an IQR increase in average NDVI in 100, 250, and 500 m buffer zones respectively.  Emotional symptoms (SDQ): 4.3% (CI: -8.1, -0.1) less with an IQR increase in average NDVI in 500 m buffer zone.  Conduct problems (SDQ): 4.8% (CI: -9.4, 0.2) and 3.6% (CI: -7.8, 0.7) less with an IQR increase in average NDVI in 100 and 250 m buffer zones respectively.  Peer relationship problems (SDQ): 4.9% (CI: -10.4, 0.9) less with an IQR increase in average NDVI in 250 m buffer zones.  ADHD (ADHD/DM-IV): 6.0% (CI: -11.3, -0.2) less with an IQR increase in average NDVI in 100 m buffer zones.  Inattention (ADHD/DM-IV): 6.2% (CI: -11.6, -0.4) less with an IQR increase in average NDVI in 100 m buffer zones.   118  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Balseviciene et al., 2014 (Lithuania) Cross sectional 4-6 years old 1,468 Greenness around residential address (kindergarten address if residential address not available)  Distance from participant's residential address to the nearest park larger than 1 ha Average NDVI (from single Landsat 7 ETM+ image) within 300 m buffer  Distance to major green space from residential address (Kaunas Land cover data set) • S-PSI/SF (child domain score, parent domain score, child-parent interaction score, total stress score)  • SDQ (prosocial behavior, emotional and behavioral problems, peer relationship problems) TDS (SDQ): β = 0.069 (p<0.05) for lower education group with increasing distance to park. Peer Problems (SDQ): β = 0.023 (p<0.05) for lower education group with increasing distance to park Conditional problems (SDQ): β = 0.026 (p < 0.05) for lower education use with increasing distance to park Hyperactivity (SDQ): β = 0.026 (p<0.05) for lower education group with increasing distance to park Prosocial behavior (SDQ): β = -0.029 (p<0.05) for lower education group with increasing distance to park Conditional problems (SDQ): β = 0.901 (p<0.05) for higher education group with increase in NDVI within 300 m of home Prosocial behavior (SDQ): β = -1.104 (p<0.05) for higher education group with increase in NDVI within 300 m of home Christian et al., 2017 (Australia) Cross sectional mean age: 5.3 years old 23,295 Distance to nearest park Distance to nearest park, attractive park or pocket park (POS layer) from population weighted census area (SA1) • Australian Early Development Census (AEDC) domains: physical health and wellbeing, social competence and emotional maturity Physical health and wellbeing: OR = 0.989 (CI: 0.976 – 0.998) per 100 m increase in distance to nearest park. Social competence:  OR = 0.99 (CI: 0.978 – 0.999) per 100 m increase in distance to nearest park.  β=0.996 (CI: 0.993 – 0.999) per 100 m increase in distance to nearest attractive park. Emotional maturity: OR = 0.989 (CI: 0.977 – 0.988) per 100 m increase to nearest park Dadvand et al., 2015  (Spain) Case control 7-10 years old 2,593 Greenness around residential address, during commute, and around school address (composite greenness) Residential NE was the average NDVI (single RapidEye image) within a 250 m buffer  Commuting NE assessed  as the average NDVI (RapidEye) within a 50 m buffer of the shortest distance from residence to school via parent reported transportation mode  School NE was calculated as average NDVI (RapidEye) within a 50 m buffer of the school boundaries  Total surrounding greenness was an calculated by averaging all exposures weighted by time of exposure • Working, superior memory (computerized n-back test)  • Attention (computerized attentional network test, ANT) Working memory: Δβ= 9.8 (CI: 5.2, 14.0) per an IQR change in NDVI within school; Δβ= 9.5 (CI: 4.2, 15.0) per an IQR change in NDVI surrounding school; Δβ= 4.9(CI: 1.0, 8.8) per an IQR change in NDVI commuting; Δβ= 9.8 (CI: 5.0, 15.0) per an IQR change in total surrounding NDVI.    Superior working memory: Δβ= 6.9 (CI: 3.4, 10.0) per an IQR change in NDVI within school; Δβ= 6.3 (CI: 2.3, 10.0) per an IQR change in NDVI surrounding school; Δβ= 6.7 (CI: 2.8, 11.0) per an IQR change in total surrounding NDVI. Inattentiveness: Δβ= -3.4 (CI: -6.6, -0.2) per an IQR change in NDVI within school; Δβ= -3.7 (CI: -7.3, -0.1) per an IQR change in NDVI surrounding school; Δβ= -3.9 (CI: -7.4, -0.4) per an IQR change in surrounding greenness index. *Results partly attenuated when adjusting for elemental carbon   119  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Dadvand et al., 2017 (Spain) Longitudinal 6 months - 7 years old 1,527 (at birth)  1,119 (at 4-5 year follow up)  1,044 (at 7 year follow up) Greenness at residential address, assessed at 12 wk, 20 wk, 32 wk of pregnancy and at birth, when the children were 6 mo, 1,  2, 4, 5 and 7 years old Average NDVI (Landsat) within 100 m, 300 m and 500 m buffers  Vegetation Continuous Field (VCF), measure of woody vegetation cover within 100 m, 300 m and 500 m buffers • Attention (Conners' Kiddie Continuous Performance Test, K-CPT for children aged 4-7 years old, and the computerized Attentional Network Task, ANT for children older than 6) Omission error (K-CPT): Decreased mean ratios 0.90 (CI: 0.85-0.96), 0.88 (CI: 0.82, 0.94), and 0.88 (0.81, 0.95) per an IQR increase of average NDVI in 100, 300, and 500 m buffer zones respectively.  HRT-SE (K-CPT): 1 ms (CI:-2.0, -0.1), 1.3 ms (CI:-2.5, -0.2), 1.3ms (CI: -2.5, -0.1) shorter per an IQR increase of average NDVI in 100, 300, and 500 m buffer zones respectively.  HRT-SE (ANT): 7.9 ms (CI:-15.1, -0.8) shorter per an IQR increase of average NDVI in 500 m buffer zones.   HRT-SE = hit reaction time-standard error Faber Taylor and Kuo, 2009 (USA) Pseudo-experimental 7-12 years old (mean age: 9.23 years old) 17 Expert assigned walk Participants completed a puzzle and then were taken for a walk in either a downtown, park, or neighborhood area. They were tested again after the walk. This was repeated for each of the environmental exposures. • Digit Span Backwards (DSB)  • Stroop Color-Word test  • Symbol Digit Modalites (SDM)  • Vigilance Task of the Gordon Diagnostic System Model (VT) Attention in children with ADHD: Cohen's d = 0.52 (p=0.023) for walk in park compared to downtown and Cohen's d=0.77 (0.007) for walk in park compared to neighborhood walk Faber Taylor and Kuo, 2011 (USA) Cross sectional 5-18 years old 72 children aged 5-7 years  164 children aged 8-10 years,  54 children aged 14-18 years  Survey reported child behavior when compared to sample photographs of typical play environments Survey filled out by parents/guardians that described severity of symptoms in relation to typical play environments • ADD/ADHD symptom severity from survey completed by parents/guardians  ADHD symptoms: more severe when playing "deep indoors" compared to in 'open grass' (d = .57, p < .001) and compared to big trees and grass (d = .25, p < .05).   Children playing in built outdoors had more severe symptoms compared to playing in open grass (d = .64, p = .001), and Big Trees & Grass (d = .31, p=0.05). Faber Taylor et al., 2001 (USA) Cross sectional 7-12 years old 96 Parent assessed environmental greenness during mildest and worst symptoms, researcher determined activity areas with photos Survey filled out by parents/guardian that described the child's behavior in different locations • ADD/ADHD symptom severity from survey completed by parents/guardians Post activity attention function among kids with ADHD: after activity in green space better functioning compared to indoor setting d = .30, p < .0001, (M = 3.53 versus 3.22, respectively) and compared to built outdoor activities, d = .28, p < .0001, (M = 3.53 versus 3.24, respectively).  120  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Faber Taylor et al., 2002 (USA) Cross sectional 7-12 years old 169 families Green view from residence window  Survey filled out by parents/guardian • Concentration tasks • Inhibition of initial impulses • Delay of gratification  *All tests administered by a trained professional  Concentration (girls): Linear relation to view of nature: F(1,76) = 10.9, p<0.01. β=0.23 Impulse inhibition (girls): Linear relation to view of nature: F(1, 76) = 3.8, p=0.05. β=0.17 Delay of gratification: Linear relation to view of nature: F(1, 76) = 12.7, p<0.001. β=0.42 Self-discipline (girls): Linear relation to view of nature: F(1, 76) = 19.4, p<0.0001. β=0.27 Feng and Astell-Burt, 2017 (Australia) Longitudinal 4-5 years old  4,968 Proportion of park space within census area of residence Percentage of land use classified as "parkland" within each census area (SA2) that contained the residential address  Parent reported perceived park quality • Strength and Difficulties Questionnaire (SDQ) Score (total difficulties score, internalizing subscale, externalizing subscale) TDS total: β = -0.54 (CI: -0.86, -0.22) for green space quantity 21-40% (compared to 0.5%). Non-significant results for ≥41% TDS total: β = 0.53 (CI:0.26, 0.81) for strongly disagree about quality of parks, compared to strongly agree.  Flouri et al., 2014 (UK) Longitudinal birth (9 months), 3, 4, 7 years old 6,384  Proportion of NE in census area measured at 3, 5 and 7 years old Percent of green space within the Lower Layer Super Output Area (LSOA) contain the residential address (Generalised Land Use Dataset - GLUD) • Strengths and Difficulties Questionnaire (SDQ) Score (hyperactivity, emotional symptoms, conduct problems and peer problems) • Parent survey on use of NEs Conduct problems:  β= 0.036 (p<0.05) with less park or playground use Hyperactivity:  β=0.052 (p<0.05) with less park or playground use Peer problems:  β= 0.047 (p<0.05) with less park or playground use  Adjusting for Socioeconomic disadvantage (SED), Life Adversity (ALE), and Neighborhood disadvantage (IMD): Conduct problems:  β= 0.047 (p<0.01) with less park or playground use Hyperactivity:  β=0.063 (p<0.01) with less park or playground use Peer problems:  β= 0.047 (p<0.01) with less park or playground use Kim et al., 2016 (USA) Cross sectional 9 - 11 years old (4th, 5th graders) 92 Landscape patch characteristics around home Unsupervised classification with 3 classes (grass, trees/forest and developed/impervious) based on NAIP 1 m Orthophotos DOQQ  Landscape characteristics (percentage landscape, number of patches, mean patch size, mean shape index, mean nearest neighbor distance, patch cohesion index) calculated using FRAGSTATS within 1/4 mile and 1/2 mile buffers • Health-related Quality of Life (PedsQL) self-reported questionnaire  HRQOL: β=0.30 (p˂0.01) for park existence within ~ 400 m, β=0.25 (p˂0.05) for park existence within ~800 m  HRQOL:  β=0.36 (p<0.05) with higher percentage green in landscape (~ 400 m buffer)  HRQOL: β=0.38 (p<0.05) with higher number of green patches (400 and 800 m buffer)  HRQOL: β=0.61 (p<0.01) with longer distance between patches (400 m buffer),  β=0.54 (p<0.01) with longer distance between patches (800 m buffer)  121  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Kuo and Faber Taylor, 2004 (USA) Cross sectional 5-18 years old 452 Greenness of area determined by parent/guardian performing survey Parent reported after effects of common after-school or weekend activities assessing child's symptoms related to ADHD in different environments • ADHD symptoms from parent reported questionnaire No Significant Results Larson et al., 2018 (USA) Cross sectional 6-17 years old 53,609 total  1,501 with ASD  Proportion of NE within ZIP code Proportion of impervious surface and tree canopy within zip code areas (NLCD) • Anxiety amongst ASD children reported by Maternal and Child Health Bureau of the Health and Resources and Services Administration funded National Survey of Children’s Health Anxiety among ASD children: Increased risk per percent increase in impervious surface (OR = 1.03, CI: 1.01, 1.05) Anxiety among ASD children: Increased risk per percent increase in tree canopy cover (OR = 1.03, CI: 1.01, 1.05)  Markevych et al., 2014b (Germany) Cross sectional 10 years old 1,218 Distance to NE from child's residential address Relative distance to child's residence to a green space (cemetery, garden, park, or plant nursery) (Bavarian Land Use Dataset) • Strengths and Difficulties Questionnaire (SDQ) (total difficulties, behavioral problems, prosocial behavior) Hyperactivity/inattention: OR = 1.20 (CI: 1.01, 1.42) per 500 m increase in distance to nearest urban green space.  Behavioral problems: OR = 1.41 (CI: 1.06, 1.87) if living further than 500 m compared to living within 500 m distance from urban green space.  Mårtensson et al., 2009 (Sweden) Cross sectional Ecological 4.5-6.5 years old 11 schools Composite score comprised of the size, layout, features of play areas, and tree canopy cover  Aerial photographs to assess size and layout of play areas  Outdoor Play Environment Categories (OPEC) to assess features of play areas  Sky-view factor to assess proportion of tree canopy cover • ADHD Symptoms measured by the  Early Childhood Attention Deficit Disorders Evaluation Scale (ECADDES) Inattention: F = -7.38 (p<0.05) for schools with high OPEC (least mean squares) all schools F=-10.50 (p<0.05) with high OPEC (least mean square) schools with standard outdoor time Hyperactivity: F=-10.60 (p<0.05) with high OPEC (least mean square) schools with standard outdoor time McCracken et al., 2016 (UK) Cross sectional 8-11 years old 254-276 (minor instances of missing data) Proportion of greenness around residential address Percentage of each type of greenspace and total area of greenspace that was within a 500 m buffer of residential address (Central Scotland Green Network) • Health-related Quality of Life measured by the Kid-KINDL questionnaire • Use by parent/guardian report Composite Kid-KINDL score: β = 0.21 (p<0.01) with increased use of green space (R2 = 0.03) Self-esteem (subscale):  β = 0.28 (p<0.01) with increased use of green space (R2 = 0.07) Friends (subscale): β = 0.23 (p<0.01) with increased use of green space (R2 = 0.04)  122  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) McEachan et al., 2018 (UK) Cross sectional 4 years old 2,594 Greenness around residential address  Use of greenness by children reported by parents Average NDVI (single Landsat 5 TM image) within 100 m, 300 m and 500 m buffers  Distance to nearest major green space  Subset reported a survey on their satisfaction with, and use of local green spaces • Strengths and Difficulties Questionnaire (SDQ) (total difficulties, internalizing subscale, externalizing subscale) Total behavioral difficulties: β= - 4.27 (CI: -7.65, -0.90) per unit increase in NDVI (100m); β= - 5.22 (CI: -8.91, -1.54) per unit increase in NDVI (300 m);   β= - 4.82 (CI: -8.57, -1.07) per unit increase in NDVI (500 m) Internalizing difficulties:  β= - 2.35 (CI: -4.20, -0.50) per unit increase in NDVI (100m);   β= - 3.15 (CI: -5.18, -1.13) per unit increase in NDVI (300 m);   β= - 2.85 (CI: -4.91, -0.80) per unit increase in NDVI (500 m) *Results only significant among South Asian kids.  Richardson et al., 2017 (UK) Longitudinal 4-6 years old 3,833 Greenness around child's residence     NE access  Area of public parks and natural space within 500 m of child's residence (Scotland's Green space map)  Questionnaire asking if child had access to a private garden • Strengths and Difficulties Questionnaire (SDQ) (total difficulties, hyperactivity problems, emotional problems, peer problems, conduct problems and prosocial behavior) Hyperactivity: No access to garden β=0.52 (CI: 0.20, 0.84) Peer problems: No access to garden β=0.23 (CI: 0.04, 0.41) Conduct problems:  No access to garden β=0.27 (CI: 0.09, 0.46) Total difficulties:  No access to garden β=1.15 (CI: 0.52, 1.78) Prosocial behavior: β = 0.08 (CI: 0.02, 0.14) increase per an IQR increase in natural space.  *Several stratification and interaction analyses also conducted, not reported in this summary.  Schutte et al., 2017 (USA) Pseudo-experimental 4-8 years old 17 children aged 4 16 children aged 5  17 children aged 7 17 children aged 8 Expert assigned walk Participants completed a puzzle and then taken for a walk in either and urban or nature setting. They were then tested after the walk. This was repeated with the other walk a few days later. • Spatial memory (directional and distance errors) • Go/No task (d' and mean reaction time on correct trials) • CPT (d' and mean reaction time on correct trials) • DSB (longest correct span)  All tests administered by trained professionals Reaction time for correct trials: faster response after nature walk F(1, 62) = 4.54, p = .037, ηp2 = .07 compared to urban walk. (Mean reaction time = 665 ms, SD=81 ms, compared to 687 ms, SD: 85 ms) Spatial working memory (constant distance error): more accurate (M=0.01cm) after nature walk F(1, 62) = 15.25, p < .001,d = .27 compared to after urban walk (M=0.42 cm)   Söderström et al., 2013 (Sweden) Cross sectional Ecological 3-5.9 years old 169 Quality of play area Amount and quality of the play area in preschools were assessed using the Outdoor Play Environment Categories (OPEC) specifically looking at the total outdoor area, amount of trees, shrubbery and hilly terrain, and the integration between vegetation, open areas and play structures • AM cortisol, PM-AM cortisol • BMI, waist circumference • Parent reported health diary • Well-being diary No Effect Size reported  123  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Tillmann et al., 2018 (Canada) Cross sectional 8-14 years old 851 Greenness around residential address  Access to green space Used NDVI (Landsat 7 and 8) to categorize physical environment: grass/shrubbery, dense vegetation  Access to nature was calculated as within a 500 m buffer  Ratio of parks to water within 500 m of residential address (CanVec layer) • Health-related Quality of Life (PedsQL) Total HRQL (urban): β=0.18 (p<0.05) per percent park in 500 m buffer Total HRQL (urban): β=-1.13 (p<0.0001) per percent water in 500 m buffer Total HRQL (urban): β=-0.28 (p<0.05) per percent grass & shrubbery in 500 m buffer Psychosocial health (urban): β=- 1.09 (p<0.001) per percent water in 500 m buffer Psychosocial health (urban): β=-0.27 (p<0.05) per percent grass & shrubbery in 500 m buffer Emotional functioning (urban): β=-1.26 (p<0.01) per percent water in 500 m buffer Social functioning (urban): β=0.27 (p<0.05) per percent park in 500 m buffer Social functioning (urban): β=-1.43 (p<0.0001) per percent water in 500 m buffer Social functioning (urban): β=-0.31 (p<0.05) per percent grass & shrubs in 500 m buffer School functioning (urban): β=-0.40 (p<0.01) per percent grass and shrubs in 500 m buffer Social functioning (rural): β=0.48 (p<0.05) per percent grass & shrubs in 500 m buffer Social functioning (rural): β=0.41 (p<0.05) per percent dense vegetation in 500 m buffer Ward et al., 2016 (New Zealand) Cross sectional 11-14 years old 118; 108 with full criteria Location, frequency, duration and intensity of play in green areas Children wore GPS trackers for seven days and recorded the times they were playing in parks.   Participants also completed surveys on mental health and behavior • Emotional Wellbeing (Life Satisfaction Scale) • Sensation seeking (Short Form Sensation Seeking) • Risk taking (Balloon Analogue Risk Task) • Cognitive development (CNS Vital Signs)  Completed in survey by parents/guardians Life satisfaction: Positive association to green space exposure: β=0.86, (CI: 0.46, 1.26) per 1% increase Life satisfaction: Positive association to green space exposure adjusted for physical activity: β=0.66, (CI: 0.37, 0.96) per 1% increase Happiness: positive association to green space exposure β=0.45, (CI: 0.18, 0.71) per 1% increase Happiness: positive association to green space exposure adjusted for physical activity β=0.36, (CI: 0.17, 0.55) per 1% increase Wellbeing index: positive association to green space exposure β=3.18, (CI: 1.49, 4.87) per 1% increase Wellbeing index: positive association to green space exposure adjusted for physical activity β=2.67, (CI: 1.29, 4.05) per 1% increase Wells, 2000 (USA) Cross sectional 7-12 years old 17 Green view from residence windows and material of yard from pre and post move residences 10 item scale administered by professional at each residential location • ADHD Symptoms  as measured by the Attention Deficit Disorder Evaluation Scale (ADDES) Directed attention capacity (DAC): After increased naturalness in children's residence, a larger proportion in the predictive model of was explained by naturalness F(1, 14) = 9.22, p < .01, R2 before = 0.50, R2 after = 0.70  124  Author(s), Year (Country) Study Design Population age N NE Exposure Method and Rate of NE Exposure Outcome Measured Tool Main Results  (significant results in adjusted models) Wells and Evans, 2003 (USA)     Cross sectional Grades 3-5 (mean age: 9.2 years old) 337 Green view from residence window  Amount of vegetation (plants) in home  Outdoor/yard material Survey administered by a professional • Stressful life events (Lewis Stressful Life Events Scale) • Psychological Distress (Rutter Child Behavior Questionnaire) • Self-perception and psychological well-being (Global Self-Worth subscale of the Harter Competency Scale) Psychological distress: explained to some extent by nearby nature, F(2, 335) = 6.27, p˂ 0.05. R2 = 0.15.   Global self-worth: explained to some extent by nearby nature, F(2, 298) = 5.05, p ˂0.05         Abbreviations: ADD = attention deficit disorder ADHD = attention deficit hyperactivity disorder AEDC = Australian Early Development Census ANT = attentional network test ASD = autism spectrum disorder CI = confidence interval DC = [USA] District of Columbia DOQQ = [USA] digital orthrophoto quarter quadrangle  EPA = [USA] Environmental Protection Agency g = gram GLUD = [UK] generalised land use dataset ha = hectare HRT-SE = hit reaction time – standard error IL = [USA] Illinois IQR = interquartile range IRR = increased risk ratio km = kilometer LBW = low birth weight LCDB = [New Zealand] Land Cover Database LSOA = [UK] lower super output area m = meter MA = [USA] Massachusetts MN = [USA] Minnesota MOE = [New Zealand] Ministry of Education ms = millisecond NAIP = [USA] National Agriculture Imagery Program NDVI = normalized difference vegetation index NLCD = [USA] National Land Cover Database NYC = [USA] New York City OPEC = Outdoor Play Environment Categories OR = odds ratio POS = [Australia] public open space RI = [USA] Rhode Island RR = rate ratio SA1 = [Australia] Statistical Area 1 SA2 = [Australia] Statistical Area 2 SAA = school attendance area SDQ = Strengths and Difficulties Questionnaire SGA = small for gestational age TDS = Total Difficulties Score TLBW = term low birth weight USGS = United States Geological Survey VCF = vegetation continuous field wk = week  125  Appendix C Supporting Tables for Chapter 3 Table C1: Test statistics for t-test between years. The paired test is reported, year 1 and year 2 along with mean and standard deviation for each year. Degrees of freedom, t-statistic, and p-value are also reported for each buffer. Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value Point Value  1999 1999 0.372 0.372 0.116 0.116 537693 NA NA 1999 2000 0.369 0.351 0.115 0.120 469704 227.6632 0 1999 2001 0.372 0.379 0.115 0.117 525522 -103.092 0 1999 2002 0.372 0.362 0.115 0.118 530237 143.4942 0 1999 2003 0.372 0.330 0.115 0.111 532380 527.3845 0 1999 2004 0.372 0.370 0.115 0.111 531920 25.45624 7.34E-143 1999 2005 0.372 0.369 0.115 0.112 530838 52.56741 0 1999 2006 0.372 0.368 0.115 0.108 530158 60.09218 0 1999 2007 0.372 0.379 0.114 0.109 529383 -95.1674 0 1999 2008 0.372 0.376 0.114 0.112 529068 -55.1612 0 1999 2009 0.372 0.347 0.114 0.105 528133 343.0114 0 1999 2010 0.372 0.347 0.114 0.108 527598 315.925 0 1999 2011 0.372 0.349 0.114 0.110 527153 273.5197 0 1999 2013 0.373 0.375 0.114 0.132 526403 -25.4287 1.48E-142 1999 2014 0.373 0.386 0.114 0.126 525843 -141.341 0 2000 2000 0.351 0.351 0.121 0.121 474848 NA NA 2000 2001 0.351 0.384 0.120 0.117 473234 -451.874 0 2000 2002 0.351 0.360 0.120 0.117 472519 -133.33 0 2000 2003 0.351 0.322 0.120 0.109 471802 344.397 0 2000 2004 0.351 0.369 0.120 0.111 471387 -251.929 0 2000 2005 0.351 0.368 0.120 0.111 470370 -224.076 0 2000 2006 0.351 0.367 0.120 0.108 469705 -198.189 0 2000 2007 0.351 0.378 0.120 0.109 469035 -332.695 0 2000 2008 0.351 0.374 0.120 0.112 468735 -295.359 0 2000 2009 0.351 0.343 0.120 0.105 467940 94.36414 0 2000 2010 0.351 0.344 0.120 0.108 467510 70.7311 0 2000 2011 0.351 0.349 0.120 0.110 467210 23.45791 1.29E-121 2000 2013 0.352 0.372 0.119 0.130 466525 -199.392 0 2000 2014 0.352 0.385 0.119 0.126 465980 -332.111 0  126  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2001 2001 0.380 0.380 0.118 0.118 536752 NA NA 2001 2002 0.380 0.362 0.118 0.119 535977 272.49 0 2001 2003 0.380 0.330 0.118 0.111 535185 545.0267 0 2001 2004 0.380 0.371 0.118 0.111 534715 150.8278 0 2001 2005 0.380 0.369 0.117 0.112 533598 155.8132 0 2001 2006 0.380 0.369 0.117 0.109 532913 154.3396 0 2001 2007 0.380 0.379 0.117 0.110 532163 15.02399 5.24E-51 2001 2008 0.380 0.377 0.117 0.113 531838 50.87718 0 2001 2009 0.380 0.348 0.117 0.106 530878 375.5832 0 2001 2010 0.380 0.348 0.117 0.109 530388 361.3013 0 2001 2011 0.380 0.350 0.117 0.111 529993 362.9406 0 2001 2013 0.381 0.377 0.117 0.132 529238 37.4658 8.42E-307 2001 2014 0.380 0.388 0.116 0.127 527848 -83.5068 0 2002 2002 0.363 0.363 0.119 0.119 544867 NA NA 2002 2003 0.363 0.331 0.119 0.112 544040 478.7245 0 2002 2004 0.363 0.371 0.119 0.112 543570 -154.805 0 2002 2005 0.363 0.370 0.119 0.113 542428 -117.895 0 2002 2006 0.363 0.369 0.119 0.110 541738 -97.5735 0 2002 2007 0.364 0.380 0.119 0.111 540933 -256.084 0 2002 2008 0.364 0.377 0.119 0.113 540603 -241.262 0 2002 2009 0.364 0.348 0.119 0.107 539618 217.7847 0 2002 2010 0.364 0.348 0.119 0.110 539048 191.5159 0 2002 2011 0.364 0.350 0.119 0.111 538623 191.7978 0 2002 2013 0.364 0.377 0.118 0.133 537838 -158.154 0 2002 2014 0.364 0.388 0.118 0.127 536338 -296.806 0 2003 2003 0.331 0.331 0.113 0.113 551000 NA NA 2003 2004 0.331 0.371 0.113 0.112 550515 -640.827 0 2003 2005 0.331 0.369 0.112 0.113 549353 -574.308 0 2003 2006 0.331 0.369 0.112 0.110 548653 -569.491 0 2003 2007 0.331 0.379 0.112 0.111 547838 -675.081 0 2003 2008 0.331 0.377 0.112 0.114 547493 -619.825 0 2003 2009 0.332 0.348 0.112 0.107 546498 -238.248 0 2003 2010 0.332 0.348 0.112 0.110 545923 -213.79 0 2003 2011 0.332 0.350 0.112 0.111 545458 -228.05 0 2003 2013 0.332 0.377 0.112 0.133 544668 -468.066 0 2003 2014 0.332 0.387 0.112 0.127 543108 -569.135 0 2004 2004 0.371 0.371 0.113 0.113 554400 NA NA 2004 2005 0.371 0.369 0.112 0.114 553198 44.56346 0 2004 2006 0.371 0.369 0.112 0.110 552463 51.86312 0  127  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2004 2007 0.371 0.379 0.112 0.111 551608 -160.11 0 2004 2008 0.371 0.377 0.112 0.114 551258 -111.421 0 2004 2009 0.371 0.348 0.112 0.107 550258 390.3895 0 2004 2010 0.371 0.348 0.112 0.110 549683 335.1671 0 2004 2011 0.371 0.350 0.112 0.111 549208 332.3143 0 2004 2013 0.372 0.377 0.112 0.134 548413 -60.4105 0 2004 2014 0.371 0.387 0.112 0.128 546848 -193.22 0 2005 2005 0.369 0.369 0.114 0.114 556193 NA NA 2005 2006 0.369 0.368 0.114 0.110 555413 8.177746 2.90E-16 2005 2007 0.369 0.379 0.114 0.111 554523 -203.44 0 2005 2008 0.369 0.377 0.114 0.114 554168 -154.051 0 2005 2009 0.369 0.348 0.114 0.107 553148 352.7897 0 2005 2010 0.369 0.348 0.113 0.110 552563 323.1743 0 2005 2011 0.369 0.350 0.113 0.111 552083 296.2431 0 2005 2013 0.369 0.377 0.113 0.134 551283 -85.4325 0 2005 2014 0.369 0.387 0.113 0.128 549713 -223.866 0 2006 2006 0.367 0.367 0.112 0.112 566141 NA NA 2006 2007 0.367 0.378 0.112 0.113 565071 -228.21 0 2006 2008 0.367 0.375 0.112 0.116 564681 -173.835 0 2006 2009 0.367 0.346 0.112 0.109 563606 383.7191 0 2006 2010 0.367 0.346 0.112 0.112 562916 345.9288 0 2006 2011 0.367 0.348 0.112 0.113 562376 289.4362 0 2006 2013 0.367 0.375 0.112 0.135 561461 -92.3791 0 2006 2014 0.367 0.385 0.112 0.129 559831 -230.702 0 2007 2007 0.377 0.377 0.114 0.114 570851 NA NA 2007 2008 0.377 0.375 0.114 0.117 570451 53.55191 0 2007 2009 0.378 0.346 0.114 0.109 569361 587.4441 0 2007 2010 0.378 0.346 0.114 0.112 568646 522.3801 0 2007 2011 0.378 0.348 0.114 0.113 568071 491.019 0 2007 2013 0.378 0.376 0.113 0.136 567141 25.95378 2.01E-148 2007 2014 0.378 0.385 0.113 0.130 565456 -100.978 0 2008 2008 0.374 0.374 0.117 0.117 573671 NA NA 2008 2009 0.375 0.346 0.117 0.110 572536 543.9807 0 2008 2010 0.375 0.346 0.117 0.112 571806 465.8829 0 2008 2011 0.375 0.348 0.117 0.113 571226 434.2502 0 2008 2013 0.375 0.375 0.116 0.136 570271 -5.35549 8.54E-08 2008 2014 0.375 0.385 0.116 0.130 568566 -146.876 0 2009 2009 0.346 0.346 0.110 0.110 575536 NA NA 2009 2010 0.346 0.346 0.110 0.112 574806 -1.80749 0.070686  128  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2009 2011 0.346 0.348 0.110 0.114 574176 -35.8214 1.05E-280 2009 2013 0.346 0.375 0.110 0.136 573201 -342.646 0 2009 2014 0.346 0.385 0.109 0.130 571486 -507.014 0 2010 2010 0.346 0.346 0.113 0.113 576986 NA NA 2010 2011 0.346 0.348 0.113 0.114 576326 -30.2476 8.07E-201 2010 2013 0.346 0.375 0.113 0.136 575276 -315.627 0 2010 2014 0.346 0.385 0.112 0.130 573556 -480.57 0 2011 2011 0.348 0.348 0.114 0.114 578731 NA NA 2011 2013 0.348 0.375 0.114 0.137 577666 -335.22 0 2011 2014 0.348 0.385 0.114 0.130 575941 -462.452 0 2013 2013 0.375 0.375 0.137 0.137 583657 NA NA 2013 2014 0.375 0.384 0.137 0.131 581892 -111.213 0 2014 2014 0.384 0.384 0.131 0.131 585915 NA NA 100 m buffer  1999 1999 0.372 0.372 0.116 0.116 537693 NA NA 1999 2000 0.369 0.351 0.115 0.120 469704 227.6632 0 1999 2001 0.372 0.379 0.115 0.117 525522 -103.092 0 1999 2002 0.372 0.362 0.115 0.118 530237 143.4942 0 1999 2003 0.372 0.330 0.115 0.111 532380 527.3845 0 1999 2004 0.372 0.370 0.115 0.111 531920 25.45624 7.34E-143 1999 2005 0.372 0.369 0.115 0.112 530838 52.56741 0 1999 2006 0.372 0.368 0.115 0.108 530158 60.09218 0 1999 2007 0.372 0.379 0.114 0.109 529383 -95.1674 0 1999 2008 0.372 0.376 0.114 0.112 529068 -55.1612 0 1999 2009 0.372 0.347 0.114 0.105 528133 343.0114 0 1999 2010 0.372 0.347 0.114 0.108 527598 315.925 0 1999 2011 0.372 0.349 0.114 0.110 527153 273.5197 0 1999 2013 0.373 0.375 0.114 0.132 526403 -25.4287 1.48E-142 1999 2014 0.373 0.386 0.114 0.126 525843 -141.341 0 2000 2000 0.351 0.351 0.121 0.121 474848 NA NA 2000 2001 0.351 0.384 0.120 0.117 473234 -451.874 0 2000 2002 0.351 0.360 0.120 0.117 472519 -133.33 0 2000 2003 0.351 0.322 0.120 0.109 471802 344.397 0 2000 2004 0.351 0.369 0.120 0.111 471387 -251.929 0 2000 2005 0.351 0.368 0.120 0.111 470370 -224.076 0 2000 2006 0.351 0.367 0.120 0.108 469705 -198.189 0  129  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2000 2007 0.351 0.378 0.120 0.109 469035 -332.695 0 2000 2008 0.351 0.374 0.120 0.112 468735 -295.359 0 2000 2009 0.351 0.343 0.120 0.105 467940 94.36414 0 2000 2010 0.351 0.344 0.120 0.108 467510 70.7311 0 2000 2011 0.351 0.349 0.120 0.110 467210 23.45791 1.29E-121 2000 2013 0.352 0.372 0.119 0.130 466525 -199.392 0 2000 2014 0.352 0.385 0.119 0.126 465980 -332.111 0 2001 2001 0.380 0.380 0.118 0.118 536752 NA NA 2001 2002 0.380 0.362 0.118 0.119 535977 272.49 0 2001 2003 0.380 0.330 0.118 0.111 535185 545.0267 0 2001 2004 0.380 0.371 0.118 0.111 534715 150.8278 0 2001 2005 0.380 0.369 0.117 0.112 533598 155.8132 0 2001 2006 0.380 0.369 0.117 0.109 532913 154.3396 0 2001 2007 0.380 0.379 0.117 0.110 532163 15.02399 5.24E-51 2001 2008 0.380 0.377 0.117 0.113 531838 50.87718 0 2001 2009 0.380 0.348 0.117 0.106 530878 375.5832 0 2001 2010 0.380 0.348 0.117 0.109 530388 361.3013 0 2001 2011 0.380 0.350 0.117 0.111 529993 362.9406 0 2001 2013 0.381 0.377 0.117 0.132 529238 37.4658 8.42E-307 2001 2014 0.380 0.388 0.116 0.127 527848 -83.5068 0 2002 2002 0.363 0.363 0.119 0.119 544867 NA NA 2002 2003 0.363 0.331 0.119 0.112 544040 478.7245 0 2002 2004 0.363 0.371 0.119 0.112 543570 -154.805 0 2002 2005 0.363 0.370 0.119 0.113 542428 -117.895 0 2002 2006 0.363 0.369 0.119 0.110 541738 -97.5735 0 2002 2007 0.364 0.380 0.119 0.111 540933 -256.084 0 2002 2008 0.364 0.377 0.119 0.113 540603 -241.262 0 2002 2009 0.364 0.348 0.119 0.107 539618 217.7847 0 2002 2010 0.364 0.348 0.119 0.110 539048 191.5159 0 2002 2011 0.364 0.350 0.119 0.111 538623 191.7978 0 2002 2013 0.364 0.377 0.118 0.133 537838 -158.154 0 2002 2014 0.364 0.388 0.118 0.127 536338 -296.806 0 2003 2003 0.331 0.331 0.113 0.113 551000 NA NA 2003 2004 0.331 0.371 0.113 0.112 550515 -640.827 0 2003 2005 0.331 0.369 0.112 0.113 549353 -574.308 0 2003 2006 0.331 0.369 0.112 0.110 548653 -569.491 0 2003 2007 0.331 0.379 0.112 0.111 547838 -675.081 0 2003 2008 0.331 0.377 0.112 0.114 547493 -619.825 0  130  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2003 2009 0.332 0.348 0.112 0.107 546498 -238.248 0 2003 2010 0.332 0.348 0.112 0.110 545923 -213.79 0 2003 2011 0.332 0.350 0.112 0.111 545458 -228.05 0 2003 2013 0.332 0.377 0.112 0.133 544668 -468.066 0 2003 2014 0.332 0.387 0.112 0.127 543108 -569.135 0 2004 2004 0.371 0.371 0.113 0.113 554400 NA NA 2004 2005 0.371 0.369 0.112 0.114 553198 44.56346 0 2004 2006 0.371 0.369 0.112 0.110 552463 51.86312 0 2004 2007 0.371 0.379 0.112 0.111 551608 -160.11 0 2004 2008 0.371 0.377 0.112 0.114 551258 -111.421 0 2004 2009 0.371 0.348 0.112 0.107 550258 390.3895 0 2004 2010 0.371 0.348 0.112 0.110 549683 335.1671 0 2004 2011 0.371 0.350 0.112 0.111 549208 332.3143 0 2004 2013 0.372 0.377 0.112 0.134 548413 -60.4105 0 2004 2014 0.371 0.387 0.112 0.128 546848 -193.22 0 2005 2005 0.369 0.369 0.114 0.114 556193 NA NA 2005 2006 0.369 0.368 0.114 0.110 555413 8.177746 2.90E-16 2005 2007 0.369 0.379 0.114 0.111 554523 -203.44 0 2005 2008 0.369 0.377 0.114 0.114 554168 -154.051 0 2005 2009 0.369 0.348 0.114 0.107 553148 352.7897 0 2005 2010 0.369 0.348 0.113 0.110 552563 323.1743 0 2005 2011 0.369 0.350 0.113 0.111 552083 296.2431 0 2005 2013 0.369 0.377 0.113 0.134 551283 -85.4325 0 2005 2014 0.369 0.387 0.113 0.128 549713 -223.866 0 2006 2006 0.367 0.367 0.112 0.112 566141 NA NA 2006 2007 0.367 0.378 0.112 0.113 565071 -228.21 0 2006 2008 0.367 0.375 0.112 0.116 564681 -173.835 0 2006 2009 0.367 0.346 0.112 0.109 563606 383.7191 0 2006 2010 0.367 0.346 0.112 0.112 562916 345.9288 0 2006 2011 0.367 0.348 0.112 0.113 562376 289.4362 0 2006 2013 0.367 0.375 0.112 0.135 561461 -92.3791 0 2006 2014 0.367 0.385 0.112 0.129 559831 -230.702 0 2007 2007 0.377 0.377 0.114 0.114 570851 NA NA 2007 2008 0.377 0.375 0.114 0.117 570451 53.55191 0 2007 2009 0.378 0.346 0.114 0.109 569361 587.4441 0 2007 2010 0.378 0.346 0.114 0.112 568646 522.3801 0 2007 2011 0.378 0.348 0.114 0.113 568071 491.019 0 2007 2013 0.378 0.376 0.113 0.136 567141 25.95378 2.01E-148 2007 2014 0.378 0.385 0.113 0.130 565456 -100.978 0  131  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2008 2008 0.374 0.374 0.117 0.117 573671 NA NA 2008 2009 0.375 0.346 0.117 0.110 572536 543.9807 0 2008 2010 0.375 0.346 0.117 0.112 571806 465.8829 0 2008 2011 0.375 0.348 0.117 0.113 571226 434.2502 0 2008 2013 0.375 0.375 0.116 0.136 570271 -5.35549 8.54E-08 2008 2014 0.375 0.385 0.116 0.130 568566 -146.876 0 2009 2009 0.346 0.346 0.110 0.110 575536 NA NA 2009 2010 0.346 0.346 0.110 0.112 574806 -1.80749 0.070686 2009 2011 0.346 0.348 0.110 0.114 574176 -35.8214 1.05E-280 2009 2013 0.346 0.375 0.110 0.136 573201 -342.646 0 2009 2014 0.346 0.385 0.109 0.130 571486 -507.014 0 2010 2010 0.346 0.346 0.113 0.113 576986 NA NA 2010 2011 0.346 0.348 0.113 0.114 576326 -30.2476 8.07E-201 2010 2013 0.346 0.375 0.113 0.136 575276 -315.627 0 2010 2014 0.346 0.385 0.112 0.130 573556 -480.57 0 2011 2011 0.348 0.348 0.114 0.114 578731 NA NA 2011 2013 0.348 0.375 0.114 0.137 577666 -335.22 0 2011 2014 0.348 0.385 0.114 0.130 575941 -462.452 0 2013 2013 0.375 0.375 0.137 0.137 583657 NA NA 2013 2014 0.375 0.384 0.137 0.131 581892 -111.213 0 2014 2014 0.384 0.384 0.131 0.131 585915 NA NA 250 m buffer  1999 1999 0.372 0.372 0.116 0.116 537693 NA NA 1999 2000 0.369 0.351 0.115 0.120 469704 227.6632 0 1999 2001 0.372 0.379 0.115 0.117 525522 -103.092 0 1999 2002 0.372 0.362 0.115 0.118 530237 143.4942 0 1999 2003 0.372 0.330 0.115 0.111 532380 527.3845 0 1999 2004 0.372 0.370 0.115 0.111 531920 25.45624 7.34E-143 1999 2005 0.372 0.369 0.115 0.112 530838 52.56741 0 1999 2006 0.372 0.368 0.115 0.108 530158 60.09218 0 1999 2007 0.372 0.379 0.114 0.109 529383 -95.1674 0 1999 2008 0.372 0.376 0.114 0.112 529068 -55.1612 0 1999 2009 0.372 0.347 0.114 0.105 528133 343.0114 0 1999 2010 0.372 0.347 0.114 0.108 527598 315.925 0 1999 2011 0.372 0.349 0.114 0.110 527153 273.5197 0 1999 2013 0.373 0.375 0.114 0.132 526403 -25.4287 1.48E-142  132  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 1999 2014 0.373 0.386 0.114 0.126 525843 -141.341 0 2000 2000 0.351 0.351 0.121 0.121 474848 NA NA 2000 2001 0.351 0.384 0.120 0.117 473234 -451.874 0 2000 2002 0.351 0.360 0.120 0.117 472519 -133.33 0 2000 2003 0.351 0.322 0.120 0.109 471802 344.397 0 2000 2004 0.351 0.369 0.120 0.111 471387 -251.929 0 2000 2005 0.351 0.368 0.120 0.111 470370 -224.076 0 2000 2006 0.351 0.367 0.120 0.108 469705 -198.189 0 2000 2007 0.351 0.378 0.120 0.109 469035 -332.695 0 2000 2008 0.351 0.374 0.120 0.112 468735 -295.359 0 2000 2009 0.351 0.343 0.120 0.105 467940 94.36414 0 2000 2010 0.351 0.344 0.120 0.108 467510 70.7311 0 2000 2011 0.351 0.349 0.120 0.110 467210 23.45791 1.29E-121 2000 2013 0.352 0.372 0.119 0.130 466525 -199.392 0 2000 2014 0.352 0.385 0.119 0.126 465980 -332.111 0 2001 2001 0.380 0.380 0.118 0.118 536752 NA NA 2001 2002 0.380 0.362 0.118 0.119 535977 272.49 0 2001 2003 0.380 0.330 0.118 0.111 535185 545.0267 0 2001 2004 0.380 0.371 0.118 0.111 534715 150.8278 0 2001 2005 0.380 0.369 0.117 0.112 533598 155.8132 0 2001 2006 0.380 0.369 0.117 0.109 532913 154.3396 0 2001 2007 0.380 0.379 0.117 0.110 532163 15.02399 5.24E-51 2001 2008 0.380 0.377 0.117 0.113 531838 50.87718 0 2001 2009 0.380 0.348 0.117 0.106 530878 375.5832 0 2001 2010 0.380 0.348 0.117 0.109 530388 361.3013 0 2001 2011 0.380 0.350 0.117 0.111 529993 362.9406 0 2001 2013 0.381 0.377 0.117 0.132 529238 37.4658 8.42E-307 2001 2014 0.380 0.388 0.116 0.127 527848 -83.5068 0 2002 2002 0.363 0.363 0.119 0.119 544867 NA NA 2002 2003 0.363 0.331 0.119 0.112 544040 478.7245 0 2002 2004 0.363 0.371 0.119 0.112 543570 -154.805 0 2002 2005 0.363 0.370 0.119 0.113 542428 -117.895 0 2002 2006 0.363 0.369 0.119 0.110 541738 -97.5735 0 2002 2007 0.364 0.380 0.119 0.111 540933 -256.084 0 2002 2008 0.364 0.377 0.119 0.113 540603 -241.262 0 2002 2009 0.364 0.348 0.119 0.107 539618 217.7847 0 2002 2010 0.364 0.348 0.119 0.110 539048 191.5159 0 2002 2011 0.364 0.350 0.119 0.111 538623 191.7978 0  133  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2002 2013 0.364 0.377 0.118 0.133 537838 -158.154 0 2002 2014 0.364 0.388 0.118 0.127 536338 -296.806 0 2003 2003 0.331 0.331 0.113 0.113 551000 NA NA 2003 2004 0.331 0.371 0.113 0.112 550515 -640.827 0 2003 2005 0.331 0.369 0.112 0.113 549353 -574.308 0 2003 2006 0.331 0.369 0.112 0.110 548653 -569.491 0 2003 2007 0.331 0.379 0.112 0.111 547838 -675.081 0 2003 2008 0.331 0.377 0.112 0.114 547493 -619.825 0 2003 2009 0.332 0.348 0.112 0.107 546498 -238.248 0 2003 2010 0.332 0.348 0.112 0.110 545923 -213.79 0 2003 2011 0.332 0.350 0.112 0.111 545458 -228.05 0 2003 2013 0.332 0.377 0.112 0.133 544668 -468.066 0 2003 2014 0.332 0.387 0.112 0.127 543108 -569.135 0 2004 2004 0.371 0.371 0.113 0.113 554400 NA NA 2004 2005 0.371 0.369 0.112 0.114 553198 44.56346 0 2004 2006 0.371 0.369 0.112 0.110 552463 51.86312 0 2004 2007 0.371 0.379 0.112 0.111 551608 -160.11 0 2004 2008 0.371 0.377 0.112 0.114 551258 -111.421 0 2004 2009 0.371 0.348 0.112 0.107 550258 390.3895 0 2004 2010 0.371 0.348 0.112 0.110 549683 335.1671 0 2004 2011 0.371 0.350 0.112 0.111 549208 332.3143 0 2004 2013 0.372 0.377 0.112 0.134 548413 -60.4105 0 2004 2014 0.371 0.387 0.112 0.128 546848 -193.22 0 2005 2005 0.369 0.369 0.114 0.114 556193 NA NA 2005 2006 0.369 0.368 0.114 0.110 555413 8.177746 2.90E-16 2005 2007 0.369 0.379 0.114 0.111 554523 -203.44 0 2005 2008 0.369 0.377 0.114 0.114 554168 -154.051 0 2005 2009 0.369 0.348 0.114 0.107 553148 352.7897 0 2005 2010 0.369 0.348 0.113 0.110 552563 323.1743 0 2005 2011 0.369 0.350 0.113 0.111 552083 296.2431 0 2005 2013 0.369 0.377 0.113 0.134 551283 -85.4325 0 2005 2014 0.369 0.387 0.113 0.128 549713 -223.866 0 2006 2006 0.367 0.367 0.112 0.112 566141 NA NA 2006 2007 0.367 0.378 0.112 0.113 565071 -228.21 0 2006 2008 0.367 0.375 0.112 0.116 564681 -173.835 0 2006 2009 0.367 0.346 0.112 0.109 563606 383.7191 0 2006 2010 0.367 0.346 0.112 0.112 562916 345.9288 0 2006 2011 0.367 0.348 0.112 0.113 562376 289.4362 0 2006 2013 0.367 0.375 0.112 0.135 561461 -92.3791 0 2006 2014 0.367 0.385 0.112 0.129 559831 -230.702 0  134  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2007 2007 0.377 0.377 0.114 0.114 570851 NA NA 2007 2008 0.377 0.375 0.114 0.117 570451 53.55191 0 2007 2009 0.378 0.346 0.114 0.109 569361 587.4441 0 2007 2010 0.378 0.346 0.114 0.112 568646 522.3801 0 2007 2011 0.378 0.348 0.114 0.113 568071 491.019 0 2007 2013 0.378 0.376 0.113 0.136 567141 25.95378 2.01E-148 2007 2014 0.378 0.385 0.113 0.130 565456 -100.978 0 2008 2008 0.374 0.374 0.117 0.117 573671 NA NA 2008 2009 0.375 0.346 0.117 0.110 572536 543.9807 0 2008 2010 0.375 0.346 0.117 0.112 571806 465.8829 0 2008 2011 0.375 0.348 0.117 0.113 571226 434.2502 0 2008 2013 0.375 0.375 0.116 0.136 570271 -5.35549 8.54E-08 2008 2014 0.375 0.385 0.116 0.130 568566 -146.876 0 2009 2009 0.346 0.346 0.110 0.110 575536 NA NA 2009 2010 0.346 0.346 0.110 0.112 574806 -1.80749 0.070686 2009 2011 0.346 0.348 0.110 0.114 574176 -35.8214 1.05E-280 2009 2013 0.346 0.375 0.110 0.136 573201 -342.646 0 2009 2014 0.346 0.385 0.109 0.130 571486 -507.014 0 2010 2010 0.346 0.346 0.113 0.113 576986 NA NA 2010 2011 0.346 0.348 0.113 0.114 576326 -30.2476 8.07E-201 2010 2013 0.346 0.375 0.113 0.136 575276 -315.627 0 2010 2014 0.346 0.385 0.112 0.130 573556 -480.57 0 2011 2011 0.348 0.348 0.114 0.114 578731 NA NA 2011 2013 0.348 0.375 0.114 0.137 577666 -335.22 0 2011 2014 0.348 0.385 0.114 0.130 575941 -462.452 0 2013 2013 0.375 0.375 0.137 0.137 583657 NA NA 2013 2014 0.375 0.384 0.137 0.131 581892 -111.213 0 2014 2014 0.384 0.384 0.131 0.131 585915 NA NA 500 m buffer  1999 1999 0.372 0.372 0.116 0.116 537693 NA NA 1999 2000 0.369 0.351 0.115 0.120 469704 227.6632 0 1999 2001 0.372 0.379 0.115 0.117 525522 -103.092 0 1999 2002 0.372 0.362 0.115 0.118 530237 143.4942 0 1999 2003 0.372 0.330 0.115 0.111 532380 527.3845 0 1999 2004 0.372 0.370 0.115 0.111 531920 25.45624 7.34E-143 1999 2005 0.372 0.369 0.115 0.112 530838 52.56741 0  135  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 1999 2006 0.372 0.368 0.115 0.108 530158 60.09218 0 1999 2007 0.372 0.379 0.114 0.109 529383 -95.1674 0 1999 2008 0.372 0.376 0.114 0.112 529068 -55.1612 0 1999 2009 0.372 0.347 0.114 0.105 528133 343.0114 0 1999 2010 0.372 0.347 0.114 0.108 527598 315.925 0 1999 2011 0.372 0.349 0.114 0.110 527153 273.5197 0 1999 2013 0.373 0.375 0.114 0.132 526403 -25.4287 1.48E-142 1999 2014 0.373 0.386 0.114 0.126 525843 -141.341 0 2000 2000 0.351 0.351 0.121 0.121 474848 NA NA 2000 2001 0.351 0.384 0.120 0.117 473234 -451.874 0 2000 2002 0.351 0.360 0.120 0.117 472519 -133.33 0 2000 2003 0.351 0.322 0.120 0.109 471802 344.397 0 2000 2004 0.351 0.369 0.120 0.111 471387 -251.929 0 2000 2005 0.351 0.368 0.120 0.111 470370 -224.076 0 2000 2006 0.351 0.367 0.120 0.108 469705 -198.189 0 2000 2007 0.351 0.378 0.120 0.109 469035 -332.695 0 2000 2008 0.351 0.374 0.120 0.112 468735 -295.359 0 2000 2009 0.351 0.343 0.120 0.105 467940 94.36414 0 2000 2010 0.351 0.344 0.120 0.108 467510 70.7311 0 2000 2011 0.351 0.349 0.120 0.110 467210 23.45791 1.29E-121 2000 2013 0.352 0.372 0.119 0.130 466525 -199.392 0 2000 2014 0.352 0.385 0.119 0.126 465980 -332.111 0 2001 2001 0.380 0.380 0.118 0.118 536752 NA NA 2001 2002 0.380 0.362 0.118 0.119 535977 272.49 0 2001 2003 0.380 0.330 0.118 0.111 535185 545.0267 0 2001 2004 0.380 0.371 0.118 0.111 534715 150.8278 0 2001 2005 0.380 0.369 0.117 0.112 533598 155.8132 0 2001 2006 0.380 0.369 0.117 0.109 532913 154.3396 0 2001 2007 0.380 0.379 0.117 0.110 532163 15.02399 5.24E-51 2001 2008 0.380 0.377 0.117 0.113 531838 50.87718 0 2001 2009 0.380 0.348 0.117 0.106 530878 375.5832 0 2001 2010 0.380 0.348 0.117 0.109 530388 361.3013 0 2001 2011 0.380 0.350 0.117 0.111 529993 362.9406 0 2001 2013 0.381 0.377 0.117 0.132 529238 37.4658 8.42E-307 2001 2014 0.380 0.388 0.116 0.127 527848 -83.5068 0 2002 2002 0.363 0.363 0.119 0.119 544867 NA NA 2002 2003 0.363 0.331 0.119 0.112 544040 478.7245 0  136  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2002 2004 0.363 0.371 0.119 0.112 543570 -154.805 0 2002 2005 0.363 0.370 0.119 0.113 542428 -117.895 0 2002 2006 0.363 0.369 0.119 0.110 541738 -97.5735 0 2002 2007 0.364 0.380 0.119 0.111 540933 -256.084 0 2002 2008 0.364 0.377 0.119 0.113 540603 -241.262 0 2002 2009 0.364 0.348 0.119 0.107 539618 217.7847 0 2002 2010 0.364 0.348 0.119 0.110 539048 191.5159 0 2002 2011 0.364 0.350 0.119 0.111 538623 191.7978 0 2002 2013 0.364 0.377 0.118 0.133 537838 -158.154 0 2002 2014 0.364 0.388 0.118 0.127 536338 -296.806 0 2003 2003 0.331 0.331 0.113 0.113 551000 NA NA 2003 2004 0.331 0.371 0.113 0.112 550515 -640.827 0 2003 2005 0.331 0.369 0.112 0.113 549353 -574.308 0 2003 2006 0.331 0.369 0.112 0.110 548653 -569.491 0 2003 2007 0.331 0.379 0.112 0.111 547838 -675.081 0 2003 2008 0.331 0.377 0.112 0.114 547493 -619.825 0 2003 2009 0.332 0.348 0.112 0.107 546498 -238.248 0 2003 2010 0.332 0.348 0.112 0.110 545923 -213.79 0 2003 2011 0.332 0.350 0.112 0.111 545458 -228.05 0 2003 2013 0.332 0.377 0.112 0.133 544668 -468.066 0 2003 2014 0.332 0.387 0.112 0.127 543108 -569.135 0 2004 2004 0.371 0.371 0.113 0.113 554400 NA NA 2004 2005 0.371 0.369 0.112 0.114 553198 44.56346 0 2004 2006 0.371 0.369 0.112 0.110 552463 51.86312 0 2004 2007 0.371 0.379 0.112 0.111 551608 -160.11 0 2004 2008 0.371 0.377 0.112 0.114 551258 -111.421 0 2004 2009 0.371 0.348 0.112 0.107 550258 390.3895 0 2004 2010 0.371 0.348 0.112 0.110 549683 335.1671 0 2004 2011 0.371 0.350 0.112 0.111 549208 332.3143 0 2004 2013 0.372 0.377 0.112 0.134 548413 -60.4105 0 2004 2014 0.371 0.387 0.112 0.128 546848 -193.22 0 2005 2005 0.369 0.369 0.114 0.114 556193 NA NA 2005 2006 0.369 0.368 0.114 0.110 555413 8.177746 2.90E-16 2005 2007 0.369 0.379 0.114 0.111 554523 -203.44 0 2005 2008 0.369 0.377 0.114 0.114 554168 -154.051 0 2005 2009 0.369 0.348 0.114 0.107 553148 352.7897 0 2005 2010 0.369 0.348 0.113 0.110 552563 323.1743 0 2005 2011 0.369 0.350 0.113 0.111 552083 296.2431 0 2005 2013 0.369 0.377 0.113 0.134 551283 -85.4325 0 2005 2014 0.369 0.387 0.113 0.128 549713 -223.866 0  137  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2006 2006 0.367 0.367 0.112 0.112 566141 NA NA 2006 2007 0.367 0.378 0.112 0.113 565071 -228.21 0 2006 2008 0.367 0.375 0.112 0.116 564681 -173.835 0 2006 2009 0.367 0.346 0.112 0.109 563606 383.7191 0 2006 2010 0.367 0.346 0.112 0.112 562916 345.9288 0 2006 2011 0.367 0.348 0.112 0.113 562376 289.4362 0 2006 2013 0.367 0.375 0.112 0.135 561461 -92.3791 0 2006 2014 0.367 0.385 0.112 0.129 559831 -230.702 0 2007 2007 0.377 0.377 0.114 0.114 570851 NA NA 2007 2008 0.377 0.375 0.114 0.117 570451 53.55191 0 2007 2009 0.378 0.346 0.114 0.109 569361 587.4441 0 2007 2010 0.378 0.346 0.114 0.112 568646 522.3801 0 2007 2011 0.378 0.348 0.114 0.113 568071 491.019 0 2007 2013 0.378 0.376 0.113 0.136 567141 25.95378 2.01E-148 2007 2014 0.378 0.385 0.113 0.130 565456 -100.978 0 2008 2008 0.374 0.374 0.117 0.117 573671 NA NA 2008 2009 0.375 0.346 0.117 0.110 572536 543.9807 0 2008 2010 0.375 0.346 0.117 0.112 571806 465.8829 0 2008 2011 0.375 0.348 0.117 0.113 571226 434.2502 0 2008 2013 0.375 0.375 0.116 0.136 570271 -5.35549 8.54E-08 2008 2014 0.375 0.385 0.116 0.130 568566 -146.876 0 2009 2009 0.346 0.346 0.110 0.110 575536 NA NA 2009 2010 0.346 0.346 0.110 0.112 574806 -1.80749 0.070686 2009 2011 0.346 0.348 0.110 0.114 574176 -35.8214 1.05E-280 2009 2013 0.346 0.375 0.110 0.136 573201 -342.646 0 2009 2014 0.346 0.385 0.109 0.130 571486 -507.014 0 2010 2010 0.346 0.346 0.113 0.113 576986 NA NA 2010 2011 0.346 0.348 0.113 0.114 576326 -30.2476 8.07E-201 2010 2013 0.346 0.375 0.113 0.136 575276 -315.627 0 2010 2014 0.346 0.385 0.112 0.130 573556 -480.57 0 2011 2011 0.348 0.348 0.114 0.114 578731 NA NA 2011 2013 0.348 0.375 0.114 0.137 577666 -335.22 0 2011 2014 0.348 0.385 0.114 0.130 575941 -462.452 0 2013 2013 0.375 0.375 0.137 0.137 583657 NA NA 2013 2014 0.375 0.384 0.137 0.131 581892 -111.213 0 2014 2014 0.384 0.384 0.131 0.131 585915 NA NA 1000 m buffer   138  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 1999 1999 0.372 0.372 0.116 0.116 537693 NA NA 1999 2000 0.369 0.351 0.115 0.120 469704 227.6632 0 1999 2001 0.372 0.379 0.115 0.117 525522 -103.092 0 1999 2002 0.372 0.362 0.115 0.118 530237 143.4942 0 1999 2003 0.372 0.330 0.115 0.111 532380 527.3845 0 1999 2004 0.372 0.370 0.115 0.111 531920 25.45624 7.34E-143 1999 2005 0.372 0.369 0.115 0.112 530838 52.56741 0 1999 2006 0.372 0.368 0.115 0.108 530158 60.09218 0 1999 2007 0.372 0.379 0.114 0.109 529383 -95.1674 0 1999 2008 0.372 0.376 0.114 0.112 529068 -55.1612 0 1999 2009 0.372 0.347 0.114 0.105 528133 343.0114 0 1999 2010 0.372 0.347 0.114 0.108 527598 315.925 0 1999 2011 0.372 0.349 0.114 0.110 527153 273.5197 0 1999 2013 0.373 0.375 0.114 0.132 526403 -25.4287 1.48E-142 1999 2014 0.373 0.386 0.114 0.126 525843 -141.341 0 2000 2000 0.351 0.351 0.121 0.121 474848 NA NA 2000 2001 0.351 0.384 0.120 0.117 473234 -451.874 0 2000 2002 0.351 0.360 0.120 0.117 472519 -133.33 0 2000 2003 0.351 0.322 0.120 0.109 471802 344.397 0 2000 2004 0.351 0.369 0.120 0.111 471387 -251.929 0 2000 2005 0.351 0.368 0.120 0.111 470370 -224.076 0 2000 2006 0.351 0.367 0.120 0.108 469705 -198.189 0 2000 2007 0.351 0.378 0.120 0.109 469035 -332.695 0 2000 2008 0.351 0.374 0.120 0.112 468735 -295.359 0 2000 2009 0.351 0.343 0.120 0.105 467940 94.36414 0 2000 2010 0.351 0.344 0.120 0.108 467510 70.7311 0 2000 2011 0.351 0.349 0.120 0.110 467210 23.45791 1.29E-121 2000 2013 0.352 0.372 0.119 0.130 466525 -199.392 0 2000 2014 0.352 0.385 0.119 0.126 465980 -332.111 0 2001 2001 0.380 0.380 0.118 0.118 536752 NA NA 2001 2002 0.380 0.362 0.118 0.119 535977 272.49 0 2001 2003 0.380 0.330 0.118 0.111 535185 545.0267 0 2001 2004 0.380 0.371 0.118 0.111 534715 150.8278 0 2001 2005 0.380 0.369 0.117 0.112 533598 155.8132 0 2001 2006 0.380 0.369 0.117 0.109 532913 154.3396 0 2001 2007 0.380 0.379 0.117 0.110 532163 15.02399 5.24E-51 2001 2008 0.380 0.377 0.117 0.113 531838 50.87718 0  139  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2001 2009 0.380 0.348 0.117 0.106 530878 375.5832 0 2001 2010 0.380 0.348 0.117 0.109 530388 361.3013 0 2001 2011 0.380 0.350 0.117 0.111 529993 362.9406 0 2001 2013 0.381 0.377 0.117 0.132 529238 37.4658 8.42E-307 2001 2014 0.380 0.388 0.116 0.127 527848 -83.5068 0 2002 2002 0.363 0.363 0.119 0.119 544867 NA NA 2002 2003 0.363 0.331 0.119 0.112 544040 478.7245 0 2002 2004 0.363 0.371 0.119 0.112 543570 -154.805 0 2002 2005 0.363 0.370 0.119 0.113 542428 -117.895 0 2002 2006 0.363 0.369 0.119 0.110 541738 -97.5735 0 2002 2007 0.364 0.380 0.119 0.111 540933 -256.084 0 2002 2008 0.364 0.377 0.119 0.113 540603 -241.262 0 2002 2009 0.364 0.348 0.119 0.107 539618 217.7847 0 2002 2010 0.364 0.348 0.119 0.110 539048 191.5159 0 2002 2011 0.364 0.350 0.119 0.111 538623 191.7978 0 2002 2013 0.364 0.377 0.118 0.133 537838 -158.154 0 2002 2014 0.364 0.388 0.118 0.127 536338 -296.806 0 2003 2003 0.331 0.331 0.113 0.113 551000 NA NA 2003 2004 0.331 0.371 0.113 0.112 550515 -640.827 0 2003 2005 0.331 0.369 0.112 0.113 549353 -574.308 0 2003 2006 0.331 0.369 0.112 0.110 548653 -569.491 0 2003 2007 0.331 0.379 0.112 0.111 547838 -675.081 0 2003 2008 0.331 0.377 0.112 0.114 547493 -619.825 0 2003 2009 0.332 0.348 0.112 0.107 546498 -238.248 0 2003 2010 0.332 0.348 0.112 0.110 545923 -213.79 0 2003 2011 0.332 0.350 0.112 0.111 545458 -228.05 0 2003 2013 0.332 0.377 0.112 0.133 544668 -468.066 0 2003 2014 0.332 0.387 0.112 0.127 543108 -569.135 0 2004 2004 0.371 0.371 0.113 0.113 554400 NA NA 2004 2005 0.371 0.369 0.112 0.114 553198 44.56346 0 2004 2006 0.371 0.369 0.112 0.110 552463 51.86312 0 2004 2007 0.371 0.379 0.112 0.111 551608 -160.11 0 2004 2008 0.371 0.377 0.112 0.114 551258 -111.421 0 2004 2009 0.371 0.348 0.112 0.107 550258 390.3895 0 2004 2010 0.371 0.348 0.112 0.110 549683 335.1671 0 2004 2011 0.371 0.350 0.112 0.111 549208 332.3143 0 2004 2013 0.372 0.377 0.112 0.134 548413 -60.4105 0 2004 2014 0.371 0.387 0.112 0.128 546848 -193.22 0 2005 2005 0.369 0.369 0.114 0.114 556193 NA NA  140  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2005 2006 0.369 0.368 0.114 0.110 555413 8.177746 2.90E-16 2005 2007 0.369 0.379 0.114 0.111 554523 -203.44 0 2005 2008 0.369 0.377 0.114 0.114 554168 -154.051 0 2005 2009 0.369 0.348 0.114 0.107 553148 352.7897 0 2005 2010 0.369 0.348 0.113 0.110 552563 323.1743 0 2005 2011 0.369 0.350 0.113 0.111 552083 296.2431 0 2005 2013 0.369 0.377 0.113 0.134 551283 -85.4325 0 2005 2014 0.369 0.387 0.113 0.128 549713 -223.866 0 2006 2006 0.367 0.367 0.112 0.112 566141 NA NA 2006 2007 0.367 0.378 0.112 0.113 565071 -228.21 0 2006 2008 0.367 0.375 0.112 0.116 564681 -173.835 0 2006 2009 0.367 0.346 0.112 0.109 563606 383.7191 0 2006 2010 0.367 0.346 0.112 0.112 562916 345.9288 0 2006 2011 0.367 0.348 0.112 0.113 562376 289.4362 0 2006 2013 0.367 0.375 0.112 0.135 561461 -92.3791 0 2006 2014 0.367 0.385 0.112 0.129 559831 -230.702 0 2007 2007 0.377 0.377 0.114 0.114 570851 NA NA 2007 2008 0.377 0.375 0.114 0.117 570451 53.55191 0 2007 2009 0.378 0.346 0.114 0.109 569361 587.4441 0 2007 2010 0.378 0.346 0.114 0.112 568646 522.3801 0 2007 2011 0.378 0.348 0.114 0.113 568071 491.019 0 2007 2013 0.378 0.376 0.113 0.136 567141 25.95378 2.01E-148 2007 2014 0.378 0.385 0.113 0.130 565456 -100.978 0 2008 2008 0.374 0.374 0.117 0.117 573671 NA NA 2008 2009 0.375 0.346 0.117 0.110 572536 543.9807 0 2008 2010 0.375 0.346 0.117 0.112 571806 465.8829 0 2008 2011 0.375 0.348 0.117 0.113 571226 434.2502 0 2008 2013 0.375 0.375 0.116 0.136 570271 -5.35549 8.54E-08 2008 2014 0.375 0.385 0.116 0.130 568566 -146.876 0 2009 2009 0.346 0.346 0.110 0.110 575536 NA NA 2009 2010 0.346 0.346 0.110 0.112 574806 -1.80749 0.070686 2009 2011 0.346 0.348 0.110 0.114 574176 -35.8214 1.05E-280 2009 2013 0.346 0.375 0.110 0.136 573201 -342.646 0 2009 2014 0.346 0.385 0.109 0.130 571486 -507.014 0 2010 2010 0.346 0.346 0.113 0.113 576986 NA NA 2010 2011 0.346 0.348 0.113 0.114 576326 -30.2476 8.07E-201 2010 2013 0.346 0.375 0.113 0.136 575276 -315.627 0  141  Year 1 Year 2 Year 1 mean Year 2 mean Year 1 standard deviation Year 2 standard deviation Degrees of freedom t-stat p-value 2010 2014 0.346 0.385 0.112 0.130 573556 -480.57 0 2011 2011 0.348 0.348 0.114 0.114 578731 NA NA 2011 2013 0.348 0.375 0.114 0.137 577666 -335.22 0 2011 2014 0.348 0.385 0.114 0.130 575941 -462.452 0 2013 2013 0.375 0.375 0.137 0.137 583657 NA NA 2013 2014 0.375 0.384 0.137 0.131 581892 -111.213 0 2014 2014 0.384 0.384 0.131 0.131 585915 NA NA    

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0391841/manifest

Comment

Related Items