ArticlePDF Available

Abstract and Figures

Studies have shown that natural environments can enhance health and here we build upon that work by examining the associations between comprehensive greenspace metrics and health. We focused on a large urban population center (Toronto, Canada) and related the two domains by combining high-resolution satellite imagery and individual tree data from Toronto with questionnaire-based self-reports of general health perception, cardio-metabolic conditions and mental illnesses from the Ontario Health Study. Results from multiple regressions and multivariate canonical correlation analyses suggest that people who live in neighborhoods with a higher density of trees on their streets report significantly higher health perception and significantly less cardio-metabolic conditions (controlling for socio-economic and demographic factors). We find that having 10 more trees in a city block, on average, improves health perception in ways comparable to an increase in annual personal income of $10,000 and moving to a neighborhood with $10,000 higher median income or being 7 years younger. We also find that having 11 more trees in a city block, on average, decreases cardio-metabolic conditions in ways comparable to an increase in annual personal income of $20,000 and moving to a neighborhood with $20,000 higher median income or being 1.4 years younger.
Content may be subject to copyright.
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
Neighborhood greenspace and
health in a large urban center
Omid Kardan
, Peter Gozdyra
, Bratislav Misic
, Faisal Moola
, Lyle J. Palmer
, Tomáš Paus
& Marc G. Berman
Studies have shown that natural environments can enhance health and here we build upon that work
by examining the associations between comprehensive greenspace metrics and health. We focused
on a large urban population center (Toronto, Canada) and related the two domains by combining
high-resolution satellite imagery and individual tree data from Toronto with questionnaire-based
self-reports of general health perception, cardio-metabolic conditions and mental illnesses from
the Ontario Health Study. Results from multiple regressions and multivariate canonical correlation
analyses suggest that people who live in neighborhoods with a higher density of trees on their
streets report signicantly higher health perception and signicantly less cardio-metabolic conditions
(controlling for socio-economic and demographic factors). We nd that having 10 more trees in a city
block, on average, improves health perception in ways comparable to an increase in annual personal
income of $10,000 and moving to a neighborhood with $10,000 higher median income or being 7
years younger. We also nd that having 11 more trees in a city block, on average, decreases cardio-
metabolic conditions in ways comparable to an increase in annual personal income of $20,000 and
moving to a neighborhood with $20,000 higher median income or being 1.4 years younger.
Many have the intuition that living near trees and greenspace is benecial to our health. But how much
could a tree in the street or a nearby neighborhood park improve our health? Here we set out to exam-
ine this very question by studying the relationship between health and neighborhood greenspace as
measured with comprehensive metrics of tree canopy on the street vs. tree canopy in parks and private
It is a known fact that urban trees improve air quality
, reduce cooling and heating energy use
and make urban environments aesthetically more preferable
. Importantly, several studies have shown
that exposure to greenspaces can be psychologically and physiologically restorative by promoting men-
tal health
, reducing non-accidental mortality
, reducing physician assessed-morbidity
, reducing
income-related health inequality’s eect on morbidity
, reducing blood pressure and stress levels
reducing sedentary leisure time
, as well as promoting physical activity
. In addition, greenspace may
enhance psychological and cardio-vascular benets of physical activity, as compared with other settings
Moreover, experimental research has demonstrated that interacting with natural environments can
have benecial eects – aer brief exposures - on memory and attention for healthy individuals
for patient populations
. In addition, having access to views of natural settings (e.g., from a home or a
hospital bed) have been found to reduce crime and aggression
and improve recovery from surgery
Although many studies have shown that natural environments enhance health or encourage healthy
behaviors, to our knowledge, fewer studies have quantied the relationship between individual trees
and health. In addition, studies have not separately estimated the treed area beside the streets and
other urban greenspaces and related those variables to individuals’ health in various domains, including
Department of Psychology, The University of Chicago, Chicago, IL, USA.
Institute for Clinical Evaluative Sciences,
Toronto, ON, Canada.
Indiana University, Bloomington, IN, USA.
The David Suzuki Foundation, Toronto, ON,
Translational Health Science, The University of Adelaide, Adelaide, SA, Australia.
Rotman Research
Institute, University of Toronto, Toronto, ON, Canada.
Grossman Institute for Neuroscience, Quantitative Biology,
and Human Behavior, University of Chicago. Correspondence and requests for materials should be addressed to
O.K. (email: or M.G.B. (email:
Received: 08 February 2015
Accepted: 01 June 2015
Published: 09 July 2015
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
cardio-metabolic conditions, mental disorders and general health perception. Knowing the kind of
greenspace that may be associated with health benets would be critical when deciding the type of
greenspace that should be incorporated into built environments to improve health.
e typical method for quantifying exposure to greenspace for individuals in large population studies
is to use the percentage of area covered in greenspace in an individuals neighborhood. e size of the
areas and the accuracy (and also denition) of greenspace quantication vary across dierent studies.
For example
, used data containing > 10 m
accuracy for greenspace and geographical units of 4 km
average in their study, Richardson et al. (2013) used > 200 m
accuracy for greenspace and geographical
units that averaged 5 km
, and
used the presence of public “natural” spaces in areas within a 5 km radius
from schools to quantify exposure to nature for school-aged children.
In this study, we were interested in examining greenspace with lower granularity (i.e., higher geo-
graphical resolution) and quantifying associations that are specic to exposure to trees, as opposed to
exposures to any greenspace, such as grass or shrubbery. Here, our denition of greenspace consisted of
tree canopy only and not of urban grass or bushes (or other “natural” settings). is choice is based on
the assumption that trees are the most consistent green components in an area and potentially the most
important component for having benecial eects
We also used a much higher geographical resolution for the following reasons. First, we wanted to dis-
tinguish between trees along the roads and streets versus those in domestic gardens and parks, and other
open areas. To do so, we used individual tree data from the ‘Street Tree General Data’ and tree-canopy
polygon data from the ‘Forest and Land Cover’ dataset to construct our greenspace variables. Both data-
sets came from the city of Toronto. Second, to ensure that the tree variables were less confounded by
health insurance policies as well as demographic parameters (age, sex, education, and income), we used
a single urban population (Toronto) in Canada, a country with a universal publically funded healthcare
system that, compared with the United States, guarantees access to health-care services independent of
income and/or employment status
. ese health-care equalities facilitate the interpretation of the rela-
tionships between individual urban trees and health in this urban population. Although nancial barri-
ers may not impede access to health care services in Canada, dierential use of physician services with
respect to socio-economic status persist; Canadians with lower incomes and fewer years of schooling
visit specialists at a lower rate than those with moderate or high incomes and higher levels of education
despite the existence of universal health care
. In particular, we examined the relationship between tree
canopy density beside the streets and in other areas such as parks and domestic gardens with an indi-
viduals health. e health variables that we focused on were: 1) Overall health perception; 2) Presence
of cardio-metabolic conditions such as hypertension, high blood glucose, obesity (both overweight and
obese), high cholesterol, myocardiac infarction, heart disease, stroke, and diabetes; and 3) Mental health
problems including major depression, anxiety, and addiction. Subjective self-rated health perception was
chosen as one of the health outcomes because self-perception of health has been found to be related to
morbidity and mortality rates and is a strong predictor of health status and outcomes in both clinical
and community settings
Furthermore, on the tree variable side, we distinguished tree canopy of trees beside the street from
those planted in other areas, such as parks and private backyards. A distinction of these dierent sources
of tree canopy may be helpful for urban planning policies. We hypothesized that street trees could have
stronger benecial associations with individual’s health because they may be more accessible to all resi-
dents in a given neighborhood as residents are likely exposed to street trees in their daily activities and
through views from their windows; for example see
Figure1 shows a geographic map of the individual tree data (i.e., the individual trees on the street)
and Fig.2 shows a geographic map of the satellite tree data (i.e., the amount of tree canopy) for dier-
ent neighborhoods in the city of Toronto. Both tree datasets were used to quantify the “greenness” of
the neighborhoods (see Methods). Figure3 shows the dissemination areas (i.e., Toronto neighborhood
units) that were used in our analysis. e highlighted neighborhoods are the ones that were included in
our analysis.
To uncover the relationships between neighborhood greenspace and health we performed two analy-
ses. e rst was a multiple regression of each health outcome on socio-economic, demographic and tree
density variables. e second was a canonical correlation analysis where we examined the multivariate
relationship between all health outcomes and socio-economic, demographic and tree density variables.
Our canonical correlation model is shown in Fig.4. In all of these analyses we attempted to quantify the
independent relationships of street tree canopy and non-street tree canopy on health.
Regression Results. Health Perception. Our results suggest that people who live in areas that have
more (and/or larger) trees on the streets report better health perception, aer controlling for demo-
graphic factors, such as income, age and education [p < 0.0001]. As can be seen in Table1, the regression
coecient for the street tree density variable shows that a four percent square meters (400 cm
) increase
in the treed area for every square meter of neighborhood predicts about 0.04 increased health perception
(i.e., 1% of our 1–5 health perception scale) for individuals living in that area. A 400 cm
increase in
treed area is equal to the addition of about 200 average trees (with 40 m
crown area) on the streets in a
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
dissemination area of almost average size (about 200,000 m
) in Toronto. is is approximately 10 more
trees per city block (a DA usually contains about 25 blocks). As can be seen in Table1, this increase in
health perception is equivalent to the eect of a $10,200 increase in annual household income and living
in a DA with equally (i.e., $10,200) higher median income. (Notice that for this comparison we added up
the estimates of income and area income because a hypothetical increase of income for the families in a
DA also increases the median area income in that DA to the same extent). is same increase in health
perception is also, on average, equivalent to being 7 years younger.
Figure 1. e Greenspace map of the city of Toronto constructed from the individual tree information
Street Tree General Data. is image is shown in much lower resolution compared to the real image and
the dissociation between individual trees and other areas is clearly perceivable for the zoomed-in area. Parks
are shown in dark green. is gure was created using Environmental Systems Research Institutes (ESRI)
ArcGIS soware v. 10.2.
Figure 2. e Greenspace map of the city of Toronto constructed from the Geographical Information
System (GIS) polygon data set Forest and Land Cover. e levels are shown in units of 10–15% for
display purposes only as we analyzed these data as a continuous variable. is gure was created using
Environmental Systems Research Institutes (ESRI) ArcGIS soware v. 10.2.
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
Other than street tree density, variables that independently predict better health perception in this
multiple regression are: eating more servings of vegetables and fruits in ones diet (1 more serving per
day predicts 1.2% better health perception [p < 0.0001]), being younger (10 years less age predicts 1.5%
better health perception [p < 0.0001]), being male (males have on average almost 1% better health per-
ception than females [p = 0.0004]), having higher education (belonging to one higher educational group
predicts 1.6% better health perception [p < 0.0001]), living in more auent neighborhoods (belonging to
one higher area median income group predicts 0.7% better health perception [p < 0.0001]), and having
higher household income (belonging to one higher income group predicts 1.6% better health perception
[p < 0.0001]). It should be mentioned that the associations between health perception and tree density
and other predictors reported here explain 9% of the variance in health perception. While the model
explains a signicant proportion of the variance in the data, it does not explain all of the variance of the
dependent variable. is is true of all models whose R
values are less than 1. As such the models pre-
dictions may not always hold true if the other unidentied factors that predict the remaining variability
in health perception are not controlled for.
Figure 3. e dissemination area map of the city of Toronto (2006). e colored regions show the
dissemination areas that were included in the study. is gure was created using Environmental Systems
Research Institutes (ESRI) ArcGIS soware v. 10.2.
Figure 4. e canonical correspondence model that was used in our canonical correlation analyses to
assess the relationship of the predictors (socio-economic, demographic and tree density variables) with
health factors.
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
Cardio-metabolic Conditions. Results of regressing the Cardio-metabolic conditions index on the inde-
pendent variables are shown in Table 2. Results suggest that people who live in areas that have more
(and/or larger) trees on the streets report signicantly fewer cardio-metabolic conditions. People reported
decrease of 0.04 units of cardio-metabolic conditions (0.5% of the 0–8 scale for cardio-metabolic con-
ditions) for every increase of 408 cm
in tree density. is is approximately equivalent to 11 more
average-sized trees on the streets per city block. is eect for cardio-metabolic conditions is equivalent
to a $20,200 increase in both area median income and annual household income adjusted for other
variables. is decrease in cardio-metabolic conditions is also, on average, equivalent to being 1.4 years
Other than street tree density, variables that predict fewer cardio-metabolic conditions, aer con-
trolling for other variables in this multiple regression, are: eating more servings of vegetables and fruits
in ones diet (1 more serving per day predicts 0.08% less cardio-metabolic conditions [p = 0.0129]), being
younger (10 years less age predicts 3.7% less cardio-metabolic conditions [p < 0.0001]), being female
(females report on average 3.3% less cardio-metabolic conditions than males [p < 0.0001]), having higher
education (belonging to one higher educational group predicts 0.71% less cardio-metabolic conditions
[p < 0.0001]), living in more auent neighborhoods (belonging to one higher area median income group
predicts 0.36% higher reported health perception [p < 0.0001]), and having higher household income
(belonging to one higher income group predicts 0.28% less cardio-metabolic conditions [p < 0.0001]).
In addition, we added the interaction terms of all predictors with the tree density variables and the
models R
for health perception and cardio-metabolic conditions did not improve much (Δ R
= 0.0008
for health perception, Δ R
= 0.0009 for cardio-metabolic conditions), even though there was a small
positive interaction between street tree density and age that was statistically signicant. We chose not to
include these interactions due to lack of a priori hypotheses, their small eect sizes and to preserve the
models simplicity. Again, it should be mentioned that the associations between cardio-metabolic condi-
tions and tree density and other predictors reported here explain 19% of the variance in cardio-metabolic
Varia bl e Estimate
Error t-stat p-value df
Increase FMI
Intercept 2.7794 0.0296 93.8319 < 0.0001 6202 0.0685 0.0644
Diet 0.0481 0.0024 19.7007 < 0.0001 668 0.2130 0.1781
Age –0.0059 0.0004 –16.8734 < 0.0001 10566 0.05246 0.0500
Sex 0.0374 0.0107 3.4853 0.0004 14364 0.04498 0.0432
Education 0.0663 0.0032 20.6885 < 0.0001 6647 0.06620 0.0624
Income 0.0710 0.0034 21.0145 < 0.0001 448 0.2630 0.2117
Area income 0.0278 0.0056 4.9162 < 0.0001 3664 0.08932 0.0825
Street Tree den. 0.0101 0.0015 6.6879 < 0.0001 34158 0.02915 0.0284
Other Tree den. –0.0003 0.0004 –0.7293 0.4658 25993 0.03342 0.0324
Table 1. Combined results of regression of health perception on the multiply-imputed data. R
= 0.0885,
adjusted R
= 0.0876, F (8, 7879*) = 94.6814, p < 0.0001. FMI is fraction of missing information.
*e average of estimated degrees of freedom.
Varia bl e Estimate
Error t-stat p-value df
Increase FMI
Intercept 0.1236 0.0363 3.4049 0.0008 895 0.1937 0.1643
Diet –0.0062 0.0026 –2.3217 0.0204 1206 0.1569 0.1371
Age 0.0296 0.0004 70.4279 < 0.0001 1724 0.1307 0.1166
Sex 0.2894 0.0128 22.5830 < 0.0001 857 0.1871 0.1596
Education –0.0570 0.0037 –15.2098 < 0.0001 553 0.2351 0.1932
Income –0.0240 0.0038 –6.2648 < 0.0001 168 0.4563 0.3213
Area income –0.0286 0.0066 –4.3071 < 0.0001 863 0.1864 0.1591
Street Tree den. –0.0097 0.0018 –5.4025 < 0.0001 801 0.1937 0.1643
Other Tree den. –0.0001 0.0005 –0.1196 0.9048 776 0.1970 0.1667
Table 2. Combined results of regression of cardio-metabolic conditions on the multiple-imputed data.
= 0.1920, adjusted R
= 0.1845, F (8, 871*) = 25.6089, p < 0.0001. FMI is fraction of missing information.
*e average of estimated degrees of freedoms.
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
conditions. While the model explains a signicant proportion of the variance in the data, it does not
explain all of the variance of the dependent variable. is is true of all models whose R
values are less
than 1. As such the model’s predictions may not always hold true if the other unidentied factors that
predict the remaining variability in cardio-metabolic conditions are not controlled for.
Mental Disorders and Other Disorders. Results of Mental Disorders and Other Disorders can be found
in Supplemental Tables S1 and S2. Regressing the Mental Disorders index on the independent variables
do not capture a signicant amount of variance in Mental Disorders in the data [R
= 0.0136, adjusted
= 0.0111, p = 0.1820]. We will further investigate this issue later in the canonical correlation anal-
Finally, the Other Disorders index is not a coherent variable and was not constructed to be used as
a dependent variable in the regression analyses, but mainly was constructed as a control variable for
the canonical correlation analysis. Nonetheless, results of regressing the Other Disorders index (Cancer,
Migraines, Arthritis, or Asthma) on the independent variables are shown in Table S2.
Canonical Correlation Results. Figures5–7 show the results from the canonical correlation analysis,
which nds the relationship (i.e., linear combination of weights) between two sets of variables. e height
of each bar shows the correlation of each variable with the corresponding set of canonical weights. Error
bars show ± 2 standard errors containing both between and within imputation variance calculated by
bootstrapping imputed data sets. Importantly, all canonical variates are orthogonal to one another.
e canonical correlation coecient (r) for each pair of linear composites is shown near the bidi-
rectional arrow representing the relationship between the two sets of variables (demographic and
green-space variables and health-related variables). e canonical correlation coecients for all the four
pairs of linear composites were statistically signicant (p < 0.0001 for Bartletts approximate chi-squared
statistic with Lawley’s modication).
e rst pair of linear composites (Fig. 5) is dominated by the eect of age on physical disorders
(Cardio-metabolic and Other disorders). is suggests that being older is highly correlated (r = 0.4565,
= 0.2084) with having more cardio-metabolic conditions, as well as cancer, arthritis, asthma and
e second pair of linear composites is mainly dominated by Health Perception and shows that indi-
viduals with higher annual income, higher education, higher vegetables/fruits consumption and who live
in areas with higher street tree density report the best health perception. is replicates and extends the
results found in the regression. e same group of people also reports fewer cardio-metabolic condi-
tions, although the errorbar for the loading of these conditions crosses zero (indicating a non-signicant
Figure 5. e rst pair of linear composites for the canonical correlation analysis; F (32, 114680)
= 381.2263), R
= 0.2084, p < 0.0001. Bars show correlation of each variable (canonical loadings) with the
rst set of weighted canonical scores. Error bars show ± 2 standard errors containing both between and
within imputation variance calculated by bootstrapping imputed data sets. Please notice the dierent colors
for health perception (teal) and other three health condition variables (yellow). is is to emphasize that
they have dierent coding directions in terms of a persons well-being (more health perception is positive,
more health conditions is negative).
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
eect). is is possibly due to the fact that the main part of the variability in cardio-metabolic conditions
(that was mainly due to older age) was already captured by the rst canonical loadings. e canonical
correlation for this second linear composite is of medium size (r = 0.2868, R
= 0.0822).
Figure 6. e second pair of linear composites for the canonical correlation analysis; F (21, 89297)
= 211.0480), R
= 0.0822, p < 0.0001. Bars show correlation of each variable with the second set of
weighted canonical scores. Error bars show ± 2 standard errors containing both between and within
imputation. Please notice the dierent colors for health perception (teal) and other three health condition
variables (yellow). is is to emphasize that they have dierent coding directions in terms of a persons well-
being (more health perception is positive, more health conditions is negative).
Figure 7. e third pair of linear composites for the canonical correlation analysis; F (12, 63702)
= 139.9347, R
= 0.0491, p < 0.0001. Bars show correlation of each variable with the third set of weighted
canonical scores. Error bars show ± 2 standard errors containing both between and within imputation
variance. Please notice the dierent colors for health perception (teal) and other three health condition
variables (yellow). is is to emphasize that they have dierent coding directions in terms of a persons well-
being (more health perception is positive, more health conditions is negative).
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
e third pair of linear composites has a modest eect size (r = 0.2216, R
= 0.0491) and is mainly
dominated by sex. is composite shows that females report more other disorders and more mental
disorders. is complies with the regression results and the fact that occurrence of breast cancer is more
frequent among women even at younger ages
Results from the fourth composite are shown in Supplementary Figure S1. e fourth component was
dominated by mental disorders aer much of the variability due to sex was extracted by the previous
composites (mainly third composite). Neither the demographic nor the tree density variables signicantly
correlated with the fourth canonical scores. e very small eect size (r = 0.0539, R
= 0.0029) shows that
the data and variables might not be rich enough for an analysis of mental disorders, as mentioned before
in the regression analysis. Indeed, only a non-reliable combination of demographic and tree variables
seem to be related to more mental disorders at this stage of analysis. Future studies with more detailed
data regarding mental disorders may help to test the results found for the fourth composite.
Finally, Table 3 shows the communalities for all the variables, which are computed as sum of the
squared loadings across all latent variables and represent how much of the variance in the variable has
been accounted for by the canonical correlation model. e communality results show that the canonical
variates are able to capture/reproduce at least 15% of the variance in all original variables. In conclusion,
both the regression and the canonical correlation analyses suggest that higher tree density on the streets,
in a given dissemination area, correlates with better health perception and fewer cardio-metabolic con-
ditions for people living in that dissemination area.
Results from our study suggest that people who live in areas with higher street tree density report bet-
ter health perception and fewer cardio-metabolic conditions compared with their peers living in areas
with lower street tree density. ere are two important points about our results that add to the previous
literature. First, the eect size of the impact of street tree density seems to be comparable to that of a
number of socioeconomic or demographic variables known to correlate with better health (beyond age).
Specically, if we consider two families, one earning $10,200 more annually than the other, and living
in a neighborhood with the same higher median income, it is predicted that the more auent family
who is living in the richer neighborhood perceives themselves as healthier people. Interestingly, however,
that prediction could turn out to be wrong if the less auent family lives in a neighborhood that has on
average 10 more trees beside the streets in every block. Regarding cardio-metabolic conditions, the same
scenario is expected to hold true for an income dierence of $20,200.
Ten more trees in every block is about 4% increase in street tree density in a dissemination area in
Toronto, which seems to be logistically feasible; Torontos dissemination areas have a 0.2% to 20.5%
range of street tree density and trees can be incorporated into various planting areas along local roads.
According to our ndings improving health perception and decreasing cardio-metabolic conditions by
planting 10 more trees per city block is equivalent to increasing the income of every household in that city
block by more than $10,000, which is more costly than planting the additional 10 trees. (See the “Urban
Watershed Forestry Manual, Part 3 Urban Tree Planting Guide” for estimation of urban tree planting and
maintenance costs and other considerations for urban tree planting. Generally, planting and maintenance
of 10 urban trees could annually cost between $300 to $5000). Finally, it should be mentioned that this
estimation of increased tree density being equivalent to specic increases in economic status of people is
based on respondents from Canada, which has a publically funded universal health-care system. It may
be the case that in other countries that do not have universal health care individuals’ health may be more
aected by economic status, which could cause the tree density relationship with health to be smaller-in
economic terms. is, however, is an empirical question that is certainly worthy of further investigation.
e second important nding is that the “health” associations with tree density were not found (in a
statistically reliable manner) for tree density in areas other than beside the streets and along local roads.
It seems that trees that aect people most generally are those that they may have the most contact (visual
or presence) with, which we are hypothesizing to be those planted along the streets. Another possible
explanation could be that trees on the street may be more important to reductions in air pollution gen-
erated by trac through dry deposition
. is does not indicate, however, that parks are not benecial.
Varia bl e Communality Varia bl e Communality
Age 0.9845 Str. Tree Density 0.3980
Income 0.8158 Other Tree Density 0.1317
Area Income 0.2649 Health Perception 1.0000
Sex 0.9848 Cardio-metabolic Conditions 1.0000
Education 0.5016 Mental Disorders 1.0000
Diet 0.4372 Other Disorders 1.0000
Table 3. Communalities for the variables based on the canonical correlation analysis.
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
is study only shows that planting trees along the roads may be more benecial than planting trees in
parks and private residences at least for these health measures. For example, our sample only consists of
adults and trees in parks may be more benecial to children who spend more time in such locations
Future studies need to address this possibility more thoroughly.
An important issue that is not addressed in this study is the mechanisms by which these benecial
eects of proximity to more (or larger) urban trees on health occur. Improving air quality, relieving
stress, or promoting physical activity could all be contributing factors to improved reported health. e
current study provides two pieces of information that could be useful when trying to study the underly-
ing mechanisms of the health benets attained from urban trees. First, more than proximity (tree density
in the neighborhood), it is the availability of the trees to the largest proportion of people (trees on the
roads) that is benecial. Second, the form of the relationship is linear, at least in the density range of
0 to 20% for trees on the streets found in the city of Toronto (i.e., adding the quadratic or the square
root of street tree density to the multiple regressions did not improve the models, suggesting that the
relationship of health outcomes with street tree density neither decreases (quadratic transformation),
nor increases (square root transformation) in a meaningful way at higher levels of street tree density).
ese two results imply that: 1) some of the eects may be partially related to the mere visual exposure
to trees
or to the dry deposition of air pollutants and 2) that the eects are not likely to plateau or
accelerate, in a meaningful way, as the level of tree canopy density increases.
In addition, in a post-hoc analysis, we compared the health outcomes of individuals living in areas with
more leaf-retaining versus more deciduous trees, adjusted for street and other tree density and demo-
graphic variables. Our analysis showed that people living in year-round green areas (more leaf-retaining
trees) reported less cardio-metabolic conditions (p = 0.017) than their peers, but not better health per-
ception. Again, while not conclusive, this result points to some importance regarding the types of trees
that should be planted, but it would be much too premature to favor the planting of non-deciduous vs.
deciduous trees.
Our study could benet from improvements in at least three aspects. First, we used cross-sectional
data for practical reasons; longitudinal data would provide us with much stronger inferences of causality.
Second, our health data items are self-reported, which introduces some error and potential biases in
health variables reported. ird, we are assuming that controlling for area median income accounts for
many other neighborhood variables that could aect mental and physical health in indirect ways (such
as neighborhood safety, pollution, etc.), which might not always hold true. In future research we plan to
test our current ndings in a more comprehensive manner that obviates the mentioned limitations. In
summary, our results show that street trees are associated with a signicant, independent and reliable
increase in health benets in urban populations and that small increases in the number of trees along
the street could improve health markedly and in cost-eective ways.
Materials and Methods
Canada is divided into geographical units called dissemination areas (DA), which consist of 400 to 700
inhabitants and whose boundary lines mostly follow roads. We used data from 3,202 DAs located in the
city of Toronto with an average population of 690 individuals and average physical size of 172,290 m
We combined data from three dierent sources to construct our tree, health and demographic var-
e rst source of tree canopy data came from the ‘Street Tree General Data,’ which is a Geographical
Information System (GIS) dataset that lists the locations of over 530,000 individual trees planted on pub-
lic land within the city of Toronto. is dataset comes from experts who traversed the city of Toronto
and recorded tree species and diameters at breast height. Trees in public parks are not listed as the listed
trees were only from public land that lines the streets. e set contains each trees common and botanical
names, their diameters at breast height (DBH), the street addresses and the general location reference
information. Figure1 shows the green-space map of Toronto generated from these data for illustration.
e second source of tree canopy data came from the Geographical Information System (GIS) pol-
ygon data set ‘Forest and Land Cover,’ which contained detailed areal information of tree canopies in
Toronto. In these data, the satellite imagery resolution was 0.6 m – from QuickBird Satellite imagery,
2007. e treed area was calculated using automated remote sensing soware - Ecognition. is auto-
mated land-cover map was then monitored by sta from the University of Vermont Spatial Analysis Lab
and adjusted to increase accuracy. In this dataset there is the ability to dierentiate shrub cover from
trees. ere is, however, some susceptibility to errors when dierentiating large shrubs from trees. To
validate the accuracy of the QuickBird satellite imagery, it was compared with two other methods used
to assess tree canopy cover: 1) Ocular estimates of canopy cover by eld crews during data collection in
2008; 2) 10,000 random point samples of leaf-o and leaf-on aerial orthophotos (imagery available in
required orthorecitifed format included 1999, 2005 and 2009)
. e tree canopy coverage estimates for
each of the respective approaches were: QuickBird: 28%; Ocular: 24%; and Aerial Orthophotos: 26.2%
. Because of the similarity in results, we can be condent in the accuracy of the QuickBird
satellite results. For more information on the automated classication of leaf-on tree canopy from the
2007 satellite imagery see Appendix 4 of
. Figure2 shows a map of tree canopy in each dissemination
area as generated from the QuickBird Satellite.
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
Information about individuals’ health and demographics was obtained in the context of the Ontario
Health Study ( is is an ongoing research study of adults (18 years
and older) living in the Canadian province of Ontario aimed at investigating risk factors associated with
diseases such as: cancer, diabetes, heart disease, asthma, and Alzheimers Disease. e data were collected
using two (similar) versions of a web-based questionnaire consisting of demographic and health-related
questions. ese questionnaires were completed by 94,427 residents living in the greater Toronto area
between September, 2010 and January, 2013. For this study, we used data from a subset of 31,109 resi-
dents (31,945 respondents, out of which 827 were removed during quality control for having duplicate
records and 9 were removed because of missing consent records). A record was considered a duplicate
with the following data quality checks: 1) Multiple registrations of the same Last Name, First Name and
Date of Birth 2) Multiple registrations of the same Last Name, First Name and Postal Code 3) Multiple
registrations of the same Last Name, First Name, Date of Birth and Postal Code 4) Multiple registrations
of the same email address. Additional data quality checks included several built-in checks in the online
system, which included automatic skip patterns and limited ranges for free text numerical responses such
that participant responses must be within reasonable limits. e nal sample included individuals who
resided in the 3,202 dissemination areas of the city of Toronto as individual tree data were only available
for these areas. ese dissemination areas are shown in Fig.3.
Demographic Variables. For each individual, we used sex (59% female; compared to the popula-
tion male/female ratio: Torontos population was 48.0% male and 52.0% female in 2011 according to
Statistics Canada), age (Mean = 43.8, range = 18–99; as of 2011 the mean age of residents above 19 years
of age for the entire population of Toronto is: 47.9 according to Statistics Canada), education (coded
as: 1 = none (0.0%), 2 = elementary (1.0%), 3 = high school (15.3%), 4 = trade (3.3%), 5 = diploma
(15.9%), 6 = certicate (5.9%), 7 = bachelor’s (35.3%), 8 = graduate degree (23.3%), with Mean = 6.07,
range = 1–8; According to the 2011 National Household Survey in, the distribution of
education for the entire city of Toronto is the following: 33% of all City residents 15 years and over
have a bachelor degree or higher, 69% of City residents between the ages of 25 and 64 years have a
postsecondary degree, 17% of 25–64 years old residents have graduate degrees), and annual household
income (coded as: 1 = less than $10 000, 2 = $10 000 – $24 999, 3 = $25 000 – $49 999, 4 = $50 000 – $74
999, 5 = $75 000 – $99 999, 6 = $100 000 – $149 999, 7 = $150 000 – $199 999, 8 = $200 000 or more,
with Mean = 4.67 which is equivalent to $90 806 annual income range = 1–8; compared to the entire
city of Torontos population mean household income, which was: $87,038 in 2010 according to Statistics
Canada), as well as diet (number of fruits and vegetable servings respondent consume every day, with
Mean = 2.24, range = 0–10), as potential confounding variables. In addition, for each dissemination area
we used the area median income from Statistics Canada and coded those data the same as the household
income data, with mean = 4.08, range = 2–8. Population densities in a given DA were used in the multi-
ple imputation analysis but not as a variable in the regressions or the canonical correlation analyses. e
correlations between demographic variables can be found in Figure S2 of Supplementary Information.
Our studied sample had similar demographics to the entire city of Toronto, but was slightly younger
(mean age = 43.8; Toronto population = 47.9), slightly more female (59%; Toronto population = 52%),
slightly more educated (35.3% had bachelor’s degrees vs. 33% in the Toronto population) and slightly
wealthier (mean household income = $93,399 vs. $87,038 in the entire city of Toronto).
Green-space variables. Crown area of the trees was used to calculate the density of area covered by
trees separately for the trees on the streets and the trees from greenspace in private locations and parks
in each DA. We estimated the crown area of the trees based on their diameter at breast height (DBH)
values. We obtained formulas for estimating tree crown diameter based on DBH for 8 tree types (Maple,
Locust, Spruce, Ash, Linden, Oak, Cherry, and Birch) that were derived from forestry research. Forestry
researchers have t linear and non-linear models to relate crown diameter and DBH for dierent species
of trees. ese models achieved good ts as veried by their high R
values (above 0.9)
. e formulas
that were used to estimate crown diameters from DBH for these tree types and their references can be
found in the Supplementary Equations section of the Supplementary Information. ese 8 tree species
covered 396,121 (83%) of the trees in our dataset. For the other 81,017 (17%) of the trees, we estimated
crown diameter based on the linear regression of crown diameters on DBHs obtained from the 83% of
the trees belonging to the tree types with known crown formulas. e crown areas of all the trees were
then calculated using the crown diameters and assuming that the crown areas were circular in shape.
Street tree density for each dissemination area was quantied as the total area of the crowns of trees
) beside the streets in the dissemination area over total dissemination area size (m
) multiplied by
100 to be in percentage format. e range for this variable was found to be from 0.02% in the areas with
the least street tree density to 20.5% in the areas with highest street tree density (Mean = 4.57%). Other
tree density for each dissemination area was calculated by subtracting out the area covered by crowns of
the trees on the streets (street tree area) from the total treed area (m
) in the dissemination area (from
the satellite Tree Canopy data), and then dividing that by the area size and multiplying by 100 to be in
percentage format. e range for this variable was found to be from 0.00% in the areas with almost no
trees in parks (or no parks), no domestic gardens or other open areas; to 75.4% in areas with high tree
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
density and parks (Mean = 23.5%). As mentioned above, there was limited ability to dierentiate large
shrub cover from tree cover in the satellite data. erefore, the variable “other tree density” could contain
some unwanted large shrub cover as well, especially for areas with very high other tree density.
Health variables. All of the health variables were constructed from the self-reported items in the
Ontario Health Study (OHS). Items related to disorders were based on the question “Have you ever been
diagnosed with …?” and coded with 0 = No and 1 = Yes. ese consisted of physical conditions includ-
ing high blood pressure, high cholesterol, high blood glucose, heart attack (MI), stroke, heart disease,
migraines, chronic obstructive pulmonary disorder (COPD), liver cirrhosis, ulcerative colitis, irritable
bowel disease (IBD), arthritis, asthma, cancer, and diabetes (DM), as well as mental health conditions
including addiction, depression, and anxiety. About 66.3% of all respondents reported having at least one
of the mentioned health conditions. e percentages of “Yes” responses for each of these conditions are
reported in Supplementary Table S3. Additionally, body mass index (BMI) for each person was calculated
from his/her self-reported height and weight. Our “Obesity” variable was constructed as 0 for BMI below
25, 0.5 for BMI between 25 and 30 (overweight, 26% of respondents), and 1 for BMI over 30 (obese,
13% of respondents). Other variables drawn from these data are general health perception (self-rated
health (1 = poor, 2 = fair, 3 = goo d, 4 = very good, 5 = excellent, with Mean = 3.66, range = 1–5), and
four more variables that were used in the multiple imputations to increase the accuracy of imputations:
walking (the number of days a participant has gone for a walk of at least 10 minutes in length last week,
with Mean = 5.33, range = 0–7), smoking (if participant has ever smoked 4-5 packs of cigarettes in their
lifetime, 38% Yes), alcohol consumption frequency (coded as 0 = never, 1 = less than monthly, 2 = abo ut
once a month, 3 = two to three times a month, 4 = once a week, 5 = two to three times a week, 6 = four to
ve times a week, with Mean = 3.60, range = 0–7), and alcohol binge frequency (coded as 0 = never, 1 = 1
to 5 times a year, 2 = 6 to 11 times a year, 3 = about once a month, 4 = 2 to 3 times a month, 5 = once
a week, 6 = 2 to 3 times a week, 7 = 4 to 5 times a week, 8 = 6 to 7 times a week, with Mean = 1.62,
range = 0–8).
e dependent variables related to physical and mental health were created from the multiple-imputed
data. For each complete dataset, the Cardio-metabolic Conditions index was constructed by sum-
ming the following seven variables related to cardio-metabolic health: High Blood Glucose, Diabetes,
Hypertension, High Cholesterol, Myocardial infarction (heart attack), Heart disease, Stroke, and
Obesity” with Mean = 0.89, range = 0–8. e Mental disorders index was constructed by summing
Major Depression, Anxiety, and Addiction, with Mean = 0.26, range = 0–3. e Health Perception index
was the third dependent variable in our analyses with Mean = 3.66, range = 1–5. e Other disorders
index consisted of Cancer, Migraines, Asthma, and Arthritis (Mean = 0.48, range = 0–4. is index was
constructed to be a control variable in the canonical correlation analysis. e additional variables (e.g.,
cirrhosis) were included to increase the accuracy of the imputation, but were not analyzed. e correla-
tion matrix between the health variables, the tree variables, and the demographic variables is reported
in supplementary Figure S2 of the Supplementary Information.
Multiple imputations analysis. e self-reported health data contained some missing values for
dierent variables (mainly due to “I dont know” responses). List wise deletion of the data (keeping only
participants with no missing values in any of the items) would have resulted in a loss of 73% of the
participants because the missing values in the dierent items were distributed across subjects, and was
therefore an unreasonable method of analysis. To handle the missing data problem, we assumed that the
data were missing at random (MAR), meaning that the probability of missingness for a variable was not
dependent on the variables value aer controlling for other observed variables. We then replaced the
missing values with multiple imputed data
. irty complete datasets were created from the original
dataset using the estimate and maximize (EM) algorithm on bootstrapped data implemented by the
Amelia package for R [Amelia
;]. All of the 30 imputations converged in less than 11 iterations. Variables
used in the imputations and their missing percentages are reported in Supplementary Table S4.
Regression analysis. e regression analyses were performed separately for each imputed dataset
and then combined based on Rubins rules
using the Zelig program in R
. Rubin suggested that the
mean of each regression coecient across all imputed datasets be used as the regression coecients for
the analysis. In addition, to avoid underestimation of standard errors and taking the uncertainty of the
imputed values into account, both the within imputation variance and between imputation variance of
each coecient should be used to construct the standard error for each regression coecient. Lastly
proposed using degrees of freedom estimated as a function of the within and between imputation var-
iance and the number of multiple imputations when approximating the t-statistics for each parameter.
To assess the amount of the variance in the dependent variables that is explained by the regression
model for the multiple imputed data we used the method suggested by Harel (2009) to estimate the R
and the adjusted R
values. Based on this method, instead of averaging R
values from the 30 imputa-
tions, rst the square root of the R
value (r) in each of the imputed datasets is transformed to a z-score
using Fishers r to z transformation, z = atanh(r). e average z across the imputations can then be
calculated. Finally, the mean of the z values is transformed back into an R
. e same procedure can be
used for adjusted R
values. Harel (2009) suggests that the number of imputations and the sample size
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
be large when using this method, which holds true in the current study. Also, the resulting estimates
of R
could be inated (i.e. are too large), while estimates of adjusted R
tend to be biased downwards
(i.e. are too small). erefore, we estimated both values for a better evaluation of the explained variance.
Canonical correlation analysis. To investigate further the relationship between the two sets of
variables, namely the health-related variables (Health Perception, Cardio-metabolic conditions, Mental
Disorders, and Other Disorders) and the demographic and green-space variables (Age, Sex, Education,
Income, Area income, Diet, Street Tree Density, and Other Tree Density), we performed a canonical cor-
relation analysis
. Our model is presented in the diagram shown in Fig.4. Mauchly’s test of sphericity
was performed on the average of imputations in MATLAB (Sphertest: Sphericity tests menu) and showed
that the correlation matrix of the data is signicantly dierent from the identity matrix (p < 0.0001). is
signicant departure of the data from sphericity warrants the canonical correlation analysis.
In a canonical correlation analysis, rst, the weights that maximize the correlation of the two weighted
sums (linear composites) of each set of variables (called canonical roots) are calculated. en the rst
root is extracted and the weights that produce the second largest correlation between sum scores is
calculated, subject to the constraint that the next set of sum scores is orthogonal to the previous one.
Each successive root will explain a unique additional proportion of variability in the two sets of varia-
bles. ere can be as many canonical roots as the minimum number of variables in the two sets, which
is four in this analysis. erefore, we obtain four sets of canonical weights for each set of variables, and
each of these four canonical roots have a canonical correlation coecient which is the square root of the
explained variability between the two weighted sums (canonical roots).
To obtain unbiased canonical weights for variables and canonical correlation coecients, we averaged
data values over the 30 imputations and performed canonical correlation analysis on the z-scores of the
averaged data using MATLAB (MATLAB and Statistics Toolbox Release 2014a, e MathWorks, Inc.,
Natick, Massachusetts, United States). For a more straight-forward interpretation and better characteri-
zation of the underlying latent variable, instead of using the canonical weights, we calculated the Pearson
correlation coecient (canonical loading) of each observed variable in the set with the weighted sum
scores for each of the four linear composites. is way, each canonical root (linear composite) could be
interpreted as an underlying latent variable whose degree of relationship with each of the observed var-
iables in the set (how much the observed variable contributes to the canonical variate) is represented by
the loading of the observed variable and its errorbar (see canonical correlation results).
To estimate the standard errors of the canonical loadings, we bootstrapped z-scores from each of
the 30 complete imputed data (1000 simulations for each) and performed canonical correlation analysis
30000 times using MATLAB. en, we calculated the variances of the set of loadings, which were calcu-
lated as explained above, over each completed dataset (within imputation variance). We also calculated
the variance of the 30 sets of coecients (between imputation variance). e standard errors of the
coecients were then estimated using the same Rubins rules as was done for the regression analyses.
1. Nowa, D. J., Crane, D. E. & Stevens, J. C. Air pollution removal by urban trees and shrubs in the United States. Urban forestry
& urban greening 4, 115–123 (2006).
2. Nowa, D. J., Hirabayashi, S., Bodine, A. & Greeneld, E. Tree and forest eects on air quality and human health in the United
States. Environmental Pollution 193, 119–129 (2014).
3. Abari, H., Pomerantz, M. & Taha, H. Cool surfaces and shade trees to reduce energy use and improve air quality in urban areas.
Solar energy 70, 295–310 (2001).
4. aplan, S., aplan, . & Wendt, J. S. ated preference and complexity for natural and urban visual material. Perception &
Psychophysics 12, 354-&, doi: 10.3758/bf03207221 (1972).
5. Smardon, . C. Perception and aesthetics of the urban-environment - review of the role of vegetation. Landscape and Urban
Planning 15, 85–106, doi: 10.1016/0169-2046(88)90018-7 (1988).
6. ichardson, E. A., Pearce, J., Mitchell, . & ingham, S. ole of physical activity in the relationship between urban green space
and health. Public health 127, 318–324 (2013).
7. Huynh, Q., Craig, W., Janssen, I. & Picett, W. Exposure to public natural space as a protective factor for emotional well-being
among young people in Canada. BMC public health 13, 407 (2013).
8. Villeneuve, P. J. et al. A cohort study relating urban green space with mortality in Ontario, Canada. Environmental esearch 115,
51–58, doi: 10.1016/j.envres.2012.03.003 (2012).
9. Maas, J. et al. Morbidity is related to a green living environment. Journal of epidemiology and community health 63, 967–973
10. Mitchell, . & Popham, F. Eect of exposure to natural environment on health inequalities: an observational population study.
Lancet 372, 1655–1660, doi: 10.1016/s0140-6736(08)61689-x (2008).
11. Grahn, P. & Stigsdotter, U. A. Landscape planning and stress. Urban forestry & urban greening 2, 1–18 (2003).
12. Pretty, J., Peacoc, J., Sellens, M. & Grin, M. e mental and physical health outcomes of green exercise. International journal
of environmental health research 15, 319–337 (2005).
13. Storgaard, . L., Hansen, H. S., Aadahl, M. & Glumer, C. Association between neighbourhood green space and sedentary leisure
time in a Danish population. Scandinavian Journal of Public Health 41, 846–852, doi: 10.1177/1403494813499459 (2013).
14. aczynsi, A. T. & Henderson, . A. Environmental correlates of physical activity: a review of evidence about pars and
recreation. Leisure Sciences 29, 315–354 (2007).
15. Humpel, N., Owen, N. & Leslie, E. Environmental factors associated with adults’ participation in physical activity: a review.
American journal of preventive medicine 22, 188–199 (2002).
16. Berman, M. G., Jonides, J. & aplan, S. e Cognitive Benets of Interacting With Nature. Psychological Science 19, 1207,
doi: 10.1111/j.1467-9280.2008.02225.x E (2008).
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
17. aplan, S. & Berman, M. G. Directed Attention as a Common esource for Executive Functioning and Self-egulation.
Perspectives on Psychological Science 5, 43, doi: 10.1177/1745691609356784 (2010).
18. Berto, . Exposure to restorative environments helps restore attentional capacity. Journal of Environmental Psychology 25, 249,
doi: 10.1016/j.jenvp.2005.07.001 (2005).
19. Taylor, A. F. & uo, F. E. Children With Attention Decits Concentrate Better Aer Wal in the Par. Journal of Attention
Disorders 12, 402, doi: 10.1177/1087054708323000 er (2009).
20. Berman, M. G. et al. Interacting with nature improves cognition and aect for individuals with depression. Journal of Aective
Disorders 140, 300–305, doi: 10.1016/j.jad.2012.03.012 (2012).
21. Cimprich, B. & onis, D. L. An environmental intervention to restore attention in women with newly diagnosed breast cancer.
Cancer nursing 26, 284 (2003).
22. uo, F. E. & Sullivan, W. C. Environment and crime in the inner city - Does vegetation reduce crime? Environment and Behavior
33, 343 (2001).
23. uo, F. E. & Sullivan, W. C. Aggression and violence in the inner city - Eects of environment via mental fatigue. Environment
and Behavior 33, 543 (2001).
24. Ulrich, . S. View through a window may inuence recovery from surgery. Science 224, 420–421, doi: 10.1126/science.6143402
25. Donovan, G. H. et al. e relationship between trees and human health: Evidence from the spread of the emerald ash borer.
American journal of preventive medicine 44, 139–145 (2013).
26. Lasser, . E., Himmelstein, D. U. & Woolhandler, S. Access to Care, Health Status, and Health Disparities in the United States
and Canada: esults of a Cross-National Population-Based Survey. American Journal of Public Health 96, 1300–1307,
doi: 10.2105/AJPH.2004.059402 (2006).
27. Dunlop, S., Coyte, P. C. & McIsaac, W. Socio-economic status and the utilisation of physicians’ services: results from the Canadian
National Population Health Survey. Social science & medicine 51, 123–133 (2000).
28. Mossey, J. M. & Shapiro, E. Self-rated health: a predictor of mortality among the elderly. American journal of public health 72,
800–808 (1982).
29. Idler, E. L. & Benyamini, Y. Self-rated health and mortality: a review of twenty-seven community studies. Journal of health and
social behavior 38, 21–37 (1997).
30. Engström, G., Hedblad, B. & Janzon, L. Subjective well-being associated with improved survival in smoing and hypertensive
men. European Journal of Cardiovascular is 6, 257–261 (1999).
31. Miller, A. B., Baines, C. J., To, T. & Wall, C. Canadian National Breast Screening Study: 1. Breast cancer detection and death rates
among women aged 40 to 49 years. CMAJ: Canadian Medical Association Journal 147, 1459 (1992).
32. Vardoulais, S., Fisher, B. E., Pericleous, . & Gonzalez-Flesca, N. Modelling air quality in street canyons: a review. Atmospheric
environment 37, 155–182 (2003).
33. Amoly, E. et al. Green and blue spaces and behavioral development in Barcelona schoolchildren: the BEATHE project. Environ
Health Perspect 122, 1351–1358, doi: 10.1289/ehp.1408215 (2014).
34. City of Toronto, P., Forestry and ecreation. Every Tree Counts: A Portrait of Torontos Urban Forest. Urban Forestry (2013).
35. Hemery, G., Savill, P. & Pryor, S. Applications of the crown diameter–stem diameter relationship for dierent species of
broadleaved trees. Forest Ecology and Management 215, 285–294 (2005).
36. Troxel, B., Piana, M., Ashton, M. S. & Murphy-Dunning, C. elationships between bole and crown size for young urban trees
in the northeastern USA. Urban Forestry & Urban Greening 12, 144–153 (2013).
37. ubin, D. B. Multiple imputation for nonresponse in surveys. 81 (John Wiley & Sons, 2004).
38. ubin, D. B. Inference and missing data. Biometria 63, 581–592 (1976).
39. Schafer, J. L., Olsen, M. . & Smardon, . C. Multiple imputation for multivariate missing-data problems: A data analysts
perspective Multivariate behavioral research 33, 545–571 (1998).
40. Honaer, J., ing, G. & Blacwell, M. Amelia II: A program for missing data. Journal of Statistical Soware 45, 1–47 (2011).
41. Owen, M., Imai, ., ing, G. & Lau, O. Zelig: Everyones Statistical Soware.  pacage version, 4.1-3 (2013).
42. ubin, D. B. & Schener, N. Multiple imputation for interval estimation from simple random samples with ignorable nonresponse.
Journal of the American Statistical Association 81, 366–374, doi: 10.2307/2289225 (1986).
43. Hotelling, H. elations between two sets of variates. Biometria, 28 321–377 (1936).
44. Johnson, . A., Wichern, D. W. & Education, P. Applied multivariate statistical analysis. Vol. 4 (Prentice hall Englewood Clis,
NJ, 1992).
is work was funded in part by a TKF Foundation grant to MGB, an internal grant from the University
of Chicago to MGB and the Tanenbaum Endowed Chair in Population Neuroscience at the University of
Toronto to TP. Data used for this research were made available by the Ontario Health Study (OHS), which
is funded by the Ontario Institute for Cancer Research, the Canadian Partnership Against Cancer, Cancer
Care Ontario, and Public Health Ontario. We thank the participants in the Ontario Health Study. We
also thank Kelly McDonald and arsiya Nagulesapillai for preparing the data from OHS, and Ruthanne
Henry for helping us gain access to the Toronto GIS data.
Author Contributions
L.J.P. and T.P. were involved in the collection of the health data. M.G.B., P.G. and F.M. aggregated the
greenspace data. M.G.B., O.K., P.G., T.P. and B.M. analyzed the data. P.G. prepared Figures 1–3 and O.K.
prepared Figures 4–7. All authors wrote and reviewed the manuscript.
Additional Information
Supplementary information accompanies this paper at
Competing nancial interests: e authors declare no competing nancial interests.
How to cite this article: Kardan, O. et al. Neighborhood greenspace and health in a large urban center.
Sci. Rep. 5, 11610; doi: 10.1038/srep11610 (2015).
SCIENTIFIC RepoRts | 5:11610 | DOI: 10.1038/srep11610
is work is licensed under a Creative Commons Attribution 4.0 International License. e
images or other third party material in this article are included in the article’s Creative Com-
mons license, unless indicated otherwise in the credit line; if the material is not included under the
Creative Commons license, users will need to obtain permission from the license holder to reproduce
the material. To view a copy of this license, visit
... Urban green spaces (UGS) have been shown to improve human wellbeing and to contribute to healthy communities (Hartig et al., 2014;Bratman et al., 2019;Pauleit et al., 2019). For instance, UGS abundance correlates strongly with city-wide wellbeing and physical health indicators (de Vries et al., 2013;Kardan et al., 2015;Mavoa et al., 2019). However, there is an unequal distribution of UGS, with disadvantaged communities, such as lower-income neighborhoods often with higher percentages of ethnically and racially diverse people, usually having less UGS around them (Astell-Burt et al., 2014;Rigolon, 2016). ...
... Nevertheless, there is currently a paucity of empirical data on UGS experiences and perceptions from disadvantaged communities. A key reason for this is that most UGS research is focused on distributional equity (see definition above), which does not usually collect empirical social data on UGS experiences or perceptions (Heynen et al., 2006;Hartig, 2008;Dai, 2011;Schipperijn et al., 2013;Astell-Burt et al., 2014;Kardan et al., 2015;Rigolon, 2016;Haase et al., 2017). Therefore, it is still not clear how people experience or perceive UGS in places characterized by low UGS abundance, exposure, or access. ...
Empirical assessments of the experiences and perceptions of urban green space (UGS) in a social housing context are scant. Studying UGS perception in these contexts is important to understand how people experience and derive benefits from UGS in disadvantaged communities. This short communication provides interdisciplinary and methodological guidance on how to conduct empirical assessment of UGS experience and perception in social housing neighborhoods. For conducting these types of studies, we suggest researchers go beyond simply calculating UGS abundance around communities and consider assessing UGS experiences and perceptions. We suggest designing studies that account for the mixed nature of many social housing communities and that develop useful UGS perception measures that can be associated with wellbeing and community indicators. Finally, we also recommend: (1) designing studies at the local scale and developing strong relationsihps with the social housing communities before and while conducting these studies; (2) considering safety as a key component of UGS perception measures; (3) considering the association between UGS perception measures and wellbeing and community experience indicators to strengthen existing models of UGS benefit provision.
... Trees not only make people healthier, but also make communities more livable, as urban forests remove a tremendous amount of air pollution (Nowak et al., 2018). Residents of tree-lined communities feel healthier and have fewer cardio-metabolic conditions than their counterparts in communities without trees (Kardan et al., 2015). Trees create a landscape that is attractive to recreation and residential areas, an environment where people want to live and work. ...
Full-text available
In consonance with the universal quest for sustainable development, the United Nations (UN) declared 17 global goals in 2015 known as the sustainable development goals (SDGs). Scholarly literature has highlighted the relevance of trees and forests (TFs) to this quest from various perspectives. However, the perspectives appear elitist, as they tend to exclude those of some seemingly ordinary but actually crucial actors whose activities directly affect the growth, health, and population of TFs, namely small-scale farmers, chainsaw operators, and artisanal miners operating particularly in rural areas in developing countries. The study examined the excluded perspectives of the said actors in selected rural comminutes in Ghana on the relevance of TFs to SD. The study was guided by the SDGs in its data collection, analysis, and reporting approach. Focus group discussions were held with the farmers and miners, while in-depth interviews were held with the chainsaw operators. The actors‘ perspectives generally indicated that TFs were relevant to seven, not relevant to six, and somehow relevant to four of the 17 SDGs. Although the actors saw TFs as important for SD, they (actors) were not favourably disposed to tree-planting for purposes of land restoration because they saw TFs as renewable resources with natural regenerative capacity. Stronger advocacy and more intensive sensitization by the central and local governments, forestry commission, and other pro-environmental organisations, institutions, and agencies on the importance of growing, nurturing, and conserving TFS for SD are recommended. Furthermore, the government should enforce regulations to stop the destruction of TFs by the actors to ensure a sustainable green environment for sustainable development.
... The importance of having contact with nature for physical and mental health have also been further evidenced along the COVID-19 pandemic [14,15,16,17]. Some recent studies have also reported that greater accessibility and neighbourhood scale implementations of UGI are associated with better subjective well-being and health [18,19]. Another study showed that spending at least 30 minutes in an urban square -specially so in green urban squaresreduces stress, anxiety, anger and fatigue, and can generate better cognitive performance, and increased happiness [20,21]. ...
Full-text available
Urban stress is one of today’s most critical health challenges that urban stakeholders need to urgently address. However, while the positive role of nature for mental health and stress level reduction has been widely reported, the role distinct urban green infrastructure design characteristics play in citizen stress level reduction is yet to be understood. This paper presents a novel methodology where virtual reality, eye-tracking technologies, biometric sensing for heart rate variability, and participant questionnaires have been combined, to evaluate the psychological and physiological stress level reduction capacity of urban green infrastructure. 30 subjects have participated in a virtual reality immersion of 12 urban squares. 360° images were captured for each square, and mapped on a virtual sphere surrounding the participant. For each urban square, the green view index was computed using semantic image classification, and based on the obtained green view index values, the images were organized in two blocks: green >35% and non-green <20%. The eye-tracking data showed significant differences between green and non-green blocks. Fixation counts decreased in the green image block, which is an indicator of stress reduction. The perceived restorativeness scale questionnaire also reported highly significant differences with higher global scores for the green image block.
... Previous research has been agnostic towards questions of type, yet these questions have high policy relevance within resource-and spaceconstrained city planning. Emerging research indicates tree canopy in particular may play an important role in supporting mental health [55,56,[64][65][66]. Associations between tree canopy and both lower prevalence and incidence of social loneliness reported in this study aligns with those previous findings. ...
Full-text available
Introduction Associations between green space type and social loneliness (a scarcity of people one feels they can depend on) were investigated in city-living participants in the Sax Institute’s 45 and Up Study. Methods Availability of green space, tree canopy and open grass were measured as a percentage of land-use within 1.6 km road−network distance buffers using high-resolution data. Multilevel logistic regressions adjusted for confounding tested associations between each green space indicator with the odds of social loneliness at baseline (prevalence) and follow-up (incidence), adjusted for demographic and socioeconomic confounders. Results The prevalence of social loneliness at baseline was 5.3% (n = 5627 /105,498). Incidence of social loneliness at follow-up was 3.4% (n = 1772/51,365). Adjusted regressions indicated reduced odds of prevalent (OR = 0.95, 95%CI = 0.92–0.98) and incident social loneliness with 10% more green space (OR = 0.92, 95%CI = 0.90– 0.96). Similar associations were found with a 10% increase in tree canopy for both prevalent (OR = 0.92, 95%CI = 0.88–0.95) and incident social loneliness (OR = 0.92, 95%CI = 0.88–0.97). Two-way interaction terms indicated effect modification by sex but not couple status. Among women, a 10% increase in total green space was associated with lower odds of prevalent (OR = 0.95, 0.91–0.95) and incident (OR = 0.89, 0.85–0.95) social loneliness. A 10% increase in tree canopy among women was associated with lower odds of prevalent (OR = 0.89, 085–0.92) and incident (OR = 0.85, 0.80–0.92) social loneliness. Meanwhile, a 10% increase in open grass among women was associated with higher odds of prevalent (OR = 1.08, 1.01–1.15) and incident (OR = 1.19, 1.03–1.35) social loneliness. Associations for men were statistically significant for a 10% increase in total green space (OR = 0.96, 0.92–0.99) and tree canopy (OR = 0.93, 0.90–0.97) for prevalent social loneliness only. Conclusion Urban greening and tree canopy restoration may reduce risks of social loneliness, perhaps especially in women.
... Valor psicológico El efecto del arbolado sobre la salud mental de las personas ha sido demostrado en numerosos estudios (Beyer et al., 2014;Bratman et al., 2019;Kardan et al., 2015;Soga et al., 2021). En ellos, se enfatiza cómo el arbolado viario juega un papel fundamental en la reducción del estrés y la ansiedad, el aumento de la concentración o su relación con la mejoría de las condiciones de pacientes hospitalizados (Ulrich, 1984;McDonald et al, 2017), entre otros efectos. ...
Bajo el lema "TIG al servicio de los ODS", con el XIX Congreso de Tecnologías de la Información Geográfica se ha querido contribuir a la consecución de los ODS 2030, evidenciando la potencialidad de las geotecnologías e identificando los objetivos y las metas a los que colabora la actividad de la comunidad TIG para transformar el contexto económico, social y ambiental hacia un futuro más sostenible en todas las escalas, a nivel global, regional y local. Las TIG ofrecen herramientas de representación, análisis, integración, comprensión, proyección y modelado de la realidad espacial o geográfica. Las TIG, al igual que todas las partes interesadas, están llamadas a contribuir a la consecución de la nueva Agenda y a colaborar en el proceso de seguimiento de los progresos conseguidos en el cumplimiento de los Objetivos de Desarrollo Sostenible (ODS) y las metas, aportando información y sistematizando indicadores transversales para su evaluación. La publicación incluye las contribuciones científicas presentadas en el Congreso, celebrado en Zaragoza entre los días 12 y 14 de septiembre de 2022.
... Urban open space provides a notable contribution toward improving air quality, extenuating global warming, and making urban environments more desirable to inhabitants. Also, access to green spaces improves the psychological well-being and emotional state of elderly people in addition to encouraging physical activities (Kardan et al. 2015). Abdullah and Ghassan (2020) noted that public open spaces are mostly developed by nonspecialists and lack the expertise of planners, landscape architects, and professionals with knowledge of public open space development. ...
Conference Paper
Full-text available
Old age is an important stage in life needing more care and attention in human development projects. Moreover, with the problem of population growth and the quest for sustainable growth, there is a need to improve the deficit in open space delivery. Also, elderly residents' requirements have not been efficiently incorporated into communal projects. This study aims to classify and assess the elements and amenities that are important to elderly users in a public open space, a case study of Barış Manço Park, in line to encourage elderly residents' participation in outdoor activities through improving accessibility, elderly amenities, and overall quality of open spaces. A methodical survey of 56 respondents was done in the study area and data were obtained using a structured questionnaire that highlights two different sections; the demographics and satisfaction with aspects of the park. In addition to the questionnaire survey, a site examination was also conducted guided by an organized checklist. Results from the questionnaire investigation found that the problems most respondents have with Barış Manço Park are; the streets, parking, and sidewalks (23.2%), pollution (21.4%), and accessibility problems (17.9%), and socio-cultural problems (16.1%). The questionnaire investigation also reveals that the features of the park that most respondents found satisfactory, measured on a mean satisfactory scale of one to five, are; Attraction (4.14), pavement (4.09), landscape (4.07), recreational possibilities (4.04) and urban furniture (4.04). The most notable observation from the checklist is the lack of designated parking spaces and toilets in the park. The study recommends the need for government and public open space developers to pay attention to prevalent user requirements as well as the safety of elderly users. ÖZET Yaşlılık, insani gelişme projelerinde daha fazla özen ve dikkat gerektiren yaşamın önemli bir aşamasıdır. Ayrıca nüfus artışı sorunu ve sürdürülebilir büyüme arayışı ile, açık alan kabulündeki açığı iyileştirmeye ihtiyaç vardır. Ayrıca, yaşlı sakinlerin gereksinimleri ortak projelere verimli bir şekilde dahil edilmemiştir. Bu çalışma, halka açık bir alanda yaşlı kullanıcılar için önemli olan unsurları ve kolaylıkları sınıflandırmayı ve değerlendirmeyi, Barış Manço Park'tan bir örnek olay incelemesi yapılarak, erişilebilirliği, yaşlılara sağlanan kolaylıkları ve açık alanların genel kalitesini iyileştirerek yaşlı sakinlerin açık hava etkinliklerine katılımını teşvik etmeyi amaçlamaktadır. Çalışma alanında 56 katılımcıyla metodik bir anket yapıldı ve veriler iki farklı bölümü vurgulayan yapılandırılmış bir anket kullanılarak elde edildilmiştir; parkın demografi ve memnuniyet yönleriyle. Anket çalışmasına ek olarak, organize bir kontrol listesi rehberliğinde bir saha incelemesi de yapılmıştır. Anket araştırmasının sonuçları, katılımcıların çoğunun Barış Manço Park ile yaşadığı sorunların; sokaklar, park ve kaldırımlar (%23.2), kirlilik (%21.4), erişilebilirlik sorunları (%17.9) ve sosyo-kültürel sorunlar (%16.1) olduğu bulundu. Anket araştırması ayrıca, katılımcıların çoğunun tatmin edici bulduğu, bir ile beş arasında ortalama tatmin edici bir ölçekte ölçülen park özelliklerinin; Cazibe (4.14), kaldırım (4.09), peyzaj (4.07), eğlence olanakları (4.04) ve kent mobilyaları (4.04) olduğunu buldu. Kontrol listesinden en dikkat çekici gözlem, parkta belirlenmiş park yerleri ve tuvaletlerin olmamasıdır.
Urban living is becoming increasingly predominant, with 55% of the world’s population currently living in cities and 68% projected to do so by 2050 (1). While megacities with more than 10 million inhabitants command much attention, they account for less than 10% of the world’s urban population. In contrast, nearly half of all urban residents – over 2 billion people – live in cities with populations under 500,000 (Table 9.1). It is in these smaller cities where population growth rates tend to be highest. The USA, Latin America, and Japan are very highly urbanized but both Africa as a whole and India remain well below 50% urbanized (Figure 9.1). Many low- and lower-middle-income countries in particular are projected to urbanize rapidly in coming decades, with a projected increase of 2.5 billion people in urban areas by 2050 – accounting for essentially all projected human population growth through mid-century.
This chapter describes basic principles and key findings regarding the development and maturation of the human brain, the former referring to the pre-natal and early post-natal periods, and the latter concerning childhood and adolescence. In both cases, we focus on brain structure as revealed in vivo with multi-modal magnetic resonance imaging (MRI). We begin with a few numbers about the human brain and its cellular composition, and a brief overview of a number of MRI-based metrics used to characterize age-related variations in grey and white matter. We then proceed with synthesizing current knowledge about developmental and maturational changes in the cerebral cortex (its thickness, surface area and intra-cortical myelination), and the underlying white matter (volume and structural properties). To facilitate biological interpretations of MRI-derived metrics, we introduce the concept of virtual histology. We conclude the chapter with a few notes about future directions in the study of factors shaping the human brain from conception onwards.
Healthcare providers may not be aware of increasing research evidence suggesting that exposure to nature can affect the brain. This integrative review of publications reporting evidence that nature exposure may influence brain function identified numerous articles on this topic that were found in non-healthcare journals that are probably not read by healthcare providers. Though being in nature may rarely result in negative outcomes such as infections from rabid animals, mosquitos, or ticks, overwhelming evidence suggest positive brain health benefits of nature exposure. Exposure to nature could affect neurobehavior, cognitive function, multiple sclerosis, Parkinson’s disease, cerebrovascular conditions, Alzheimer’s disease, and various neuropsychiatric disorders. The risks and benefits of solar exposure in addition to it vitamin D-making powers are described. Research evidence supports numerous brain health benefits of being in nature in both blue and green spaces; however, further investigation is needed with stronger designs to identify causal mechanisms, pathways of relationships, and variations by context and populations and with different health outcomes. Longitudinal studies and randomized clinical trials are needed with more valid, reliable measures. This review of the research provides a basis upon which to continue to develop future research and recommendations for public health policy and public dissemination to promote brain health and disease prevention.KeywordsNature and neuroprotectionNature and mental healthNature and depressionPhysical activity in natureGreen spaceBlue spacePollutionClimateSolar exposure
Full-text available
Background: Green spaces have been associated with improved mental health in children; however, available epidemiological evidence on their impact on child behavioral development is scarce. Objectives: We investigated the impact of contact with green spaces and blue spaces (beaches) on indicators of behavioral development and symptoms of attention deficit/hyperactivity disorder (ADHD) in schoolchildren. Methods: This study was based on a sample of 2,111 schoolchildren (7-10 years of age) from 36 schools in Barcelona in 2012. We obtained data on time spent in green spaces and beaches and Strengths and Difficulties Questionnaires (SDQ) from parents, and ADHD/DSM-IV questionnaires from teachers. Surrounding greenness was abstracted as the average Normalized Difference Vegetation Index (NDVI) in buffers of 100 m, 250 m, and 500 m around each home address. Proximity to green spaces was defined as living within 300 m of a major green space (≥ 0.05 km2). We applied quasi-Poisson mixed-effects models (with school random effect) to separately estimate associations between indicators of contact with green spaces and SDQ and ADHD total and subscale scores. Results: We generally estimated beneficial associations between behavioral indicators and longer time spent in green spaces and beaches, and with residential surrounding greenness. Specifically, we found statistically significant inverse associations between green space playing time and SDQ total difficulties, emotional symptoms, and peer relationship problems; between residential surrounding greenness and SDQ total difficulties and hyperactivity/inattention and ADHD/DSM-IV total and inattention scores; and between annual beach attendance and SDQ total difficulties, peer relationship problems, and prosocial behavior. For proximity to major green spaces, the results were not conclusive. Conclusion: Our findings support beneficial impacts of contact with green and blue spaces on behavioral development in schoolchildren.
When making sampling distribution inferences about the parameter of the data, theta, it is appropriate to ignore the process that causes missing data if the missing data are 'missing at random' and the observed data are 'observed at random', but these inferences are generally conditional on the observed pattern of missing data. When making direct likelihood or Bayesian inferences about theta, it is appropriate to ignore the process that causes missing data if the missing data are missing at random and the parameter of the missing data process is 'distinct' from theta. These conditions are the weakest general conditions under which ignoring the process that causes missing data always leads to correct inferences.
Aim: Sedentary behaviour is a risk factor for diabetes, cardiovascular disease etc., independently of level of physical activity. Availability of recreational green space is associated with physical activity, but is unknown in relation to sedentary behaviour. The aim of this study is to examine the association between availability of green space and sedentary leisure time in a Danish population. Methods: The study was based on a random sample of 49,806 adults aged 16 + who answered a questionnaire in 2010, including sedentary leisure time. Objective measures of density green were calculated for each respondent using Geographical Information System (GIS). A multilevel regression analysis, taking neighbourhood and individual factors into account, was performed. Results: 65% of the respondents were sedentary in leisure time for more than 3h/day. We found that poor availability of forest and recreational facilities in the neighbourhood is associated with more sedentary leisure time; OR: 1.11 (95% CL: 1.04-1.19), after adjusting for individual, and neighbourhood, level characteristics. Conclusions: Among adult inhabitants, sedentary leisure time of more than 3h/day was more frequent in neighbourhoods with less green surroundings. Intervention efforts may benefit from emphasising the importance of having recreations options in residential areas to provide alternatives to sedentary activities.
Several multiple imputation techniques are described for simple random samples with ignorable nonresponse on a scalar outcome variable. The methods are compared using both analytic and Monte Carlo results concerning coverages of the resulting intervals for the population mean. Using m = 2 imputations per missing value gives accurate coverages in common cases and is clearly superior to single imputation (m = 1) in all cases. The performances of the methods for various m can be predicted well by linear interpolation in 1/(m — 1) between the results for m = 2 and m = ∞. As a rough guide, to assure coverages of interval estimates within 2% of the nominal level when using the preferred methods, the number of imputations per missing value should increase from 2 to 3 as the nonresponse rate increases from 10% to 60%.
Amelia II is a complete R package for multiple imputation of missing data. The package implements a new expectation-maximization with bootstrapping algorithm that works faster, with larger numbers of variables, and is far easier to use, than various Markov chain Monte Carlo approaches, but gives essentially the same answers. The program also improves imputation models by allowing researchers to put Bayesian priors on individual cell values, thereby including a great deal of potentially valuable and extensive information. It also includes features to accurately impute cross-sectional datasets, individual time series, or sets of time series for different cross-sections. A full set of graphical diagnostics are also available. The program is easy to use, and the simplicity of the algorithm makes it far more robust; both a simple command line and extensive graphical user interface are included.