Habitat International 56 (2016) 84e95

Contents lists available at ScienceDirect

Habitat International
journal homepage: www.elsevier.com/locate/habitatint

Urban green spaces, their spatial pattern, and ecosystem service value:
The case of Beijing
Liyan Xu a, b, Hong You a, Dihua Li a, Kongjian Yu a, *
a
Peking University, China
b
Massachusetts Institute of Technology, USA

a r t i c l e i n f o a b s t r a c t

Article history: Green spaces provide various kinds of ecosystem service functions. Though some of them, such as the
Received 11 January 2016 carbon-sinking and biodiversity preservation functions are of value to everyone, others, especially those
Received in revised form related to aesthetic and recreational functions, only benet people who have direct access to green
16 April 2016
spaces. In urban settings, where ecosystem services in the second category prevail, this means the spatial
Accepted 25 April 2016
dimension of urban green spaces, including their richness, accessibility, shape conguration, and
distributional characteristics, may considerably inuence the realization of their ecosystem service value,
and is therefore subject to scrutiny. In this paper, we study how the spatial pattern of urban green spaces
Keywords:
Urban green spaces
inuence the realization of their ecosystem service value by utilizing the Hedonic Price Modeling (HPM)
Ecosystem service method. Taking Beijing as the case, we use the price and other information in the city's developable land
Spatial pattern transaction records from 2000 to 2004 to construct the HPM, and use Landscape Ecological Metrics
Landscape Ecological Metrics (LEM) (LEM) as proxies of the spatial characteristics of urban green spaces. Four LEMs are used to measure the
Hedonic Price Model (HPM) above mentioned spatial characteristics of urban green spaces. While subject to certain shortcomings in
Beijing data quality and quantitative estimations of the magnitude of the spatial effects cannot be made, results
show that most spatial characteristics of urban green spaces do inuence their ecosystem service value
as embedded in land value, except for the shape conguration characteristic for which the study yields
no result. Further, specically for Beijing, results indicate that in order to effectively realize their
ecosystem service value, green spaces should occupy between 2.20% and 13.40% of the total urban area,
located within a 50e550 m range from other developments, with green space patches so divided that
each patch occupies more than 3.00% but less than 62.50% of the total green space area, and the
ecosystem service value will be at the optimal level when each patch occupies 20.00% of the total green
space area. Lastly, we stress the practical signicance of the ndings, urging an integration of the spatial
patterns aspect of urban green spaces in urban planning practices.
2016 Elsevier Ltd. All rights reserved.

1. Introduction Green spaces, including forest, grassland, farmland, etc., play a

Urban areas across the world have been facing threatens from
environmental degradation (McMichael, 2000), and the situation is
especially severe in fast urbanizing and industrializing developing
countries, such as China (Stern, Common, and Barbier, 1996; Liu &
Diamond, 2005; Economy 2011). Environmental degradation not
only causes physical harms like air and water pollution (Booth &
Jackson, 1997; Shao, Tang, Zhang, & Li, 2006), but also inicts
mental problems to the urban residents (Jiang, Zhang, and Sullivan,
2015), and thus constitutes an urgent issue to address.
* Corresponding author.
crucial role in the global ecosystem, and urban green spaces in
particular are commonly regarded as a remedy to the urban envi-
ronmental problems. They help remove air and water pollution (Jim
& Chen, 2008; De Ridder et al., 2004), preserve biodiversity
(Mortberg & Wallentinus, 2000), and create an amenable atmo-
sphere which benets people's physical and mental health (De
Vries et al. 2003; Hillsdon, Panter, Foster, & Jones, 2006; Jim &
Chen, 2006; Jiang, Larsen, Deal, & Sullivan, 2015).
An interesting question, though, is how much value have urban
green spaces realized in improving the environment. Technically, a
feasible way to answer the question is by evaluating the ecosystem
service value generated by urban green spaces (Go  mez-Baggethun
& Barton, 2013). However, in the absence of a market for urban

http://dx.doi.org/10.1016/j.habitatint.2016.04.005
0197-3975/ 2016 Elsevier Ltd. All rights reserved.
 L. Xu et al. / Habitat International 56 (2016) 84e95
green spaces, one needs non-market evaluating methods to esti-
mate the former's value in monetary terms (McConnell & Walls,
2005). Seminal works of this kind can be attributed to Costanza
et al. (Costanza et al. 1997), who invented a constructed
willingness-to-pay method to estimate the service value of the
global ecosystem. This study established a unit value-based
paradigm for evaluating the value of ecosystem services, which
despite some innate drawbacks (Bingham, Bishop, & Brody, 1995;
Pearce, 1998), has since been widely applied in ecosystems of
various types and scales (Rapport, Gaudet, & Karr, 1998), and green
spaces in particular, with a fairly large geographical coverage
(Costanza, Stern, & Fisher, 2004; Cilliers, Cilliers, Lubbe, & Siebert,
2012; Bateman et al. 2013).
However, the unit value-based method is built implicitly on
the assumption that the ecosystem service value is independent of
the spatial dimension. This may hold true at the global or other very
large scales, as in the case of the above mentioned Costanza et al.s
work. But at smaller scales, the assumption is not as sound from a
landscape ecology perspective, which states though some
ecosystem service functions, such as carbon sinking, are location-
insensitive, others are not. Indeed, as Forman (Forman, 1995)
noted, green space patches with different spatial attributes (rich-
ness, accessibility, shape conguration, and distributional charac-
teristics) may have different ecological functions in a landscape,
and they therefore should convey varied amount of ecosystem
service value. For example, Xie et al. (Xie, Xiao, & Lu, 2006)
demonstrate that the soil and water preservation service of forests
is much less important in plains than in slope terrains, thus is of less
ecosystem service value in the former case. Particularly, in urban
contexts, where the most prominent ecosystem service functions
are regarding the aesthetic and recreational aspects, the spatial
dimension matters even more profoundly. It is therefore necessary
to study the relationship between the spatial pattern of urban
green spaces and their ecosystem service value, a subject the
existing literature sheds little light on.
The lack of research attention on the issue not only constitutes
an academic gap, but has also inicted negative inuences in real-
world practices. The argument above suggests that urban planners
and managers should pay as much attention to the spatial dimen-
sion of urban green spaces as the quantity. However, despite certain
degrees of academic coverage, the notion appears not quite
commonly appreciated in practice, resulting an overemphasis on
the latter and neglect of the former in many occasions (Haase et al.
2014). For example, some large cities in China have adopted an
occupation/compensation balance of urban green spaces policy,1
requiring developers who destruct urban green spaces to create
new ones elsewhere. In practice, however, for quite understandable
reasons, such make-up green spaces usually locate in the exurbia or
even remoter areas. Such practices, letting alone the obvious po-
litical ecological problem they imply (Heynen, Perkins, & Roy,
2006; Wolch, Byrne, & Newell, 2014), cause losses in ecosystem
service value, too, as urban green spaces ecosystem service func-
tions, such as the city beautication and micro-climate control, are
only meaningful where population concentrates, and the loss of
such services in the city center cannot be compensated by as
large, or even larger green spaces in remote areas where their
ecosystem service value hardly realizes.
In this paper, we study the relationship between urban green
spaces spatial pattern and the ecosystem service value they convey.
1
For example, Guidelines for the Basic Ecological Control Line Management for the
city of Wuhan, see http://www.wuhan.gov.cn/frontpage/pubinfo/PubinfoDetail.
action?id1201205242201590014; and Regulations on Urban Green Spaces for the
city of Qingdao, see http://rules.yuanlin.com/Html/Detail/2012-2/1599_2.html.
85
Specic spatial characteristics to be examined include the richness,
accessibility, distribution, and shape conguration of urban green
spaces. Using empirical data from Beijing, China, we construct
quantitative models to examine the spatial effects, and also explore
their practical implications. We hope our work will not only
contribute to the academic literature, but also have practical im-
pacts and could thus help build a better urban environment.
The paper is organized as follows. Section two provides a review
of relevant studies and methods. Section three discusses the
modeling approach, measurement of spatial characteristics, and
data issues. Section four presents the results, with detailed
demonstration of the spatial effects resulting from different rich-
ness, accessibility, distribution, and shape conguration variables.
Section ve offers further analysis on the quantied estimation of
the spatial effects, as well as discussions on the possible reasons for
the non-signicant results. We conclude the paper in section six
with a summary of the study and discussions on the study's prac-
tical implications.
2. Evaluating the ecosystem service value of urban green
spaces and the inuence of their spatial patterns: a literature
review
As there does not exist a market for urban green spaces in most
occasions, people's willingness-to-pay for urban green spaces could
only be measured through indirect approaches (McConnell & Walls,
2005). The price of real estate, for example, is a commonly used
proxy, which is considered to include a green space premium e
the willingness-to-pay for accessibility to urban green spaces so as
to enjoy the ecosystem service they provide. The Hedonic Price
Model (HPM) (Chau, Ma, & Ho, 2001) is a typical method to sepa-
rate out the green space premium from real estate prices.
HPM assumes that the price of a commodity, such as an apart-
ment unit or a land parcel, includes the contributions from its
various innate and environmental characteristics (Lancaster, 1966).
Therefore, one may identify the willingness-to-pay for each feature
involved using analytical techniques such as multivariate regres-
sion. Specically, for urban real estates, their prices are usually
considered to consist of the contributions from three categories of
characteristics: the structural (such as the size of a land parcel or
the unit plan type of an apartment), neighborhood (such as the
transportation accessibility), and environmental (such as nearby
amenity and recreational facilities) variables (Poudyal, Hodges, &
Merrett, 2009). A typical Hedonic Price Model is thus formulated
as follows:
X X X
ln pi b0 bj Sij bk Nik bl Eil i (1)
where ln pi denotes the logarithm of the price of the i-th real estate
object, Sij denotes its j-th structural variable, Nik denotes its k-th
neighborhood variable, and Nik denotes its l-th environmental
variable. b0 , bj , bk , bl , and i are the respective estimates of
regression coefcients and the residual term.
Therefore, the HPM can be used to evaluate the value of urban
green spaces when they serve as the environmental variables in
Equation (1). Loads of works of this sort have been done during the
past half-century (McConnell & Walls, 2005), covering various
green space types such as natural habitats, parks, planted forests,
wetlands, and farmlands. The divergent inuences of different
green space types have also been widely discussed (Bolitzer &
Netusil, 2000; Neumann, Boyle, & Bell, 2009).
Particularly, regarding the urban green spaces, the subject of this
study, detailed studies have been conducted concerning their in-
uence on real estate value (Bolitzer & Netusil, 2000; Poudyal et al.,
86 L. Xu et al. / Habitat International 56 (2016) 84e95
2009). While some studies fail to nd a relationship between the
two (Morancho, 2003), most empirical evidences show that real
estate values are indeed positively correlated to the existence of
nearby urban green spaces. Other inquiries have gone further to
reveal such relationship's dependency on spatial proximity, as
measured by the distance between urban green space locations and
real estate sites (Czembrowski & Kronenberg, 2016; Morancho,
2003; Sander & Polasky, 2009; Tajima, 2003; Tyrvainen &
Miettinen, 2000).
With these progresses, however, the majority of the literature
has not addressed other spatial characteristics of urban green
spaces, e.g. the richness, distribution, and shape conguration
characteristics, and their inuence on nearby real estate values.
Among the few works that do cover this issue, Kong, Yin, and
Nakagoshi (2007) show that real estate values in Jinan, China are
indeed inuenced by the spatial patterns of urban green spaces as
represented by a size-distance index, as well as by the richness of
urban green space. Poudyal, Hodges, Tonn, & Cho, 2009, studying a
real estate transaction sample in the City of Roanoke, Virginia, have
gone further and showed that open space plots with square shape
and smooth, straight edges were preferred to those with more
complex shapes and irregular edges, and that open spaces in few
larger plots are preferred to many smaller scattered pieces. Meth-
odologically, what is worth noting in these works is that they use
Landscape Ecological Metrics (LEM) to quantify the spatial char-
acteristics of urban green spaces, and thus to construct the envi-
ronmental variables in the HPM. We borrow this method in our
paper, too, which we elaborate in the next section. Despite these
works, research on the spatial aspect of urban green space and their
ecosystem service values is quite limited in both types of spatial
characteristics involved and real-world cases covered. All in all, to
remedy the lack of relevant research, a systematic examination on
the spatial effects of urban green spaces on real estate values in a
real-world setting is needed.
3. Methodology
3.1. Data
3.1.1. Urban green spaces and other basic geographical features
We use an ofcial Geographic Information System (GIS) data-
base of Beijing, produced in 2004, as our basic geographical data-
base in this paper. The database consists the location and shape of
all major urban green spaces in the city, as well as other
geographical features such as administrative boundaries, streets,
roads, and public transit lines and stops which are necessary for
computing the neighborhood variables for the HPM. The key
geographical features are shown in Fig. 1, and the location and
shapes of Beijing's urban green spaces are shown in Fig. 2. Ac-
cording to the data, Beijing's urban green spaces occupy about 10%
of the area in the city's Inner Eight Districts,2 which is also the
spatial extent covered in this study (i.e. whenever referred, Bei-
jing means the Inner Eight Districts of the city).
It should be noted, however, that the urban green spaces here do
not include all urban green spaces in the general sense. According
to China's land use classication system, urban land are classied
into certain main categories, including residential, commercial,
industrial, transportation, and green spaces, etc. However, within
the rst-order categories such as residential or commercial land
use, there are second-order land use sub-categories, such as green
2
Including: Dongcheng, Xicheng, Chongwen, Xuanwu, Haidian, Chaoyang,
Fengtai, Shijingshan. Chongwen and Xuanwu had been annexed to Dongcheng and
Xicheng Districts, respectively in 2010.
spaces, which in some cases could occupy as much as 35% of the
total area. These second-order land uses are not designated at the
master planning level. Rather, they are designated ad hoc when it
comes to the detailed planning of specic land parcels, and there-
fore these second-order land uses, such as green spaces, do not
count in the statistics of the respective land use class. As a result,
the database that we use in this study only consist of the land
parcels that are classied into the rst-order green spaces land
use class. This data quality issue may lead to certain estimation
errors, as previous studies have otherwise pointed out (Derkzen,
van Teeffelen, & Verburg, 2015). In the absence of any better
geographical data source of urban green spaces, we use the above
mentioned database anyway. However, conceptually, the omission
of second-order urban green spaces may almost surely result in
estimation errors in the following modeling. We address this
problem whenever encountered later in this paper.
3.1.2. Land transaction records
We use the complete record of developable residential land
transactions in Beijing from 2000 to 2004 as basic dataset for
analysis. As noted above, the HPM requires market price data of
housing or land transactions. In this study, the main reason for
using land rather than housing data is data availability concerns.
With reliable ofcial or third-party database for housing trans-
actions unavailable, there seems to be no alternative but to use the
land data for which the full record is available. Moreover, there are
evidences that the housing market during that period was domi-
nated by speculation and therefore bore a bubble (Hui & Yue,
2006), thus violating the market equilibrium assumption which is
essential for the HPM.
The land transaction data, however, has its own problems. On
the one hand, as noted above, the database consists of all residential
land transaction records during the study period, thus constituting
a valuable source of land market data. On the other hand, however,
whether the land price-formulating mechanism in Beijing during
the study period was completely free-market based is questionable,
which may result in some problems. We explain the problem as
follows.
China's urban land is owned by the state by law, and the urban
land market (strictly speaking, it is the market for the leasehold
rights of urban land) has been evolving since the 1988 real estate
reform. Different land price-formulating mechanism dominates in
different periods, and these mechanisms form a full spectrum from
totally central-planned economy to totally free-market styles. The
traditional way, known as the land acquisition by governmental
allocation approach, is not a market-based mechanism at all.
Indeed, land price under such a mechanism is usually 0. Since early
1990s, however, a new market-based mechanism began to prevail,
which is known as the land transfer by agreement approach.
Under this mechanism, land price is based on the agreement be-
tween the local government and the entities seeking leasehold
rights, and market-based methods, such as the market comparison
method and the replacement costs method, have been developed
to determine the land price. However, though having some market
features, this mechanism is, to say the most, a semi-marketized one
(Mak, Choy, & Ho, 2007), as the procedure of land price determi-
nation is usually not transparent, and there is usually no free entry
to the land market for all market entities. Lastly, rst experimented
in some cities in 2002 and becoming prevail nationwide after Aug
31, 2004, a new, fully marketized mechanism, known as the land
transfer by bidding, auction, and listing mechanism was intro-
duced, and land price under this mechanism should represent a
totally free-market value.
Specically, in Beijing, the bidding, auction, and listing
mechanism was rst introduced in Dec 8, 2003, and became the
 L. Xu et al. / Habitat International 56 (2016) 84e95 87

Fig. 1. Key geographical features of the study area.

only mechanism after Sep 1, 2004. This means that in our land prices to reect constant 2000 Yuan value. The location and price of
transaction database, which covers the land transaction records the land transactions are also shown in Fig. 2.
between 2000 and 2004, only a fraction of the transactions are
based on the bidding, auction, and listing mechanism, and the
3.2. Measurements of the spatial pattern: Landscape Ecological
rest are all based on the agreement mechanism. As noted above,
Metrics
the latter is not a fully marketized one. Problem is, we do not know
how marketized the prices are. Irrational (by a free market stan-
The quantication of the various spatial characteristics of urban
dard) land prices may exist in some cases, rational ones may exist in
green spaces is key to the study, for which Landscape Ecological
others. In the absence of a better data source, and considering other
Metrics are the most widely used tool (Wu & Hobbs, 2002). In this
merits of the database (for example, China's current land regulation
paper, we select four indices to measure the richness, accessibility,
system prohibits any secondary land market, which has repressed
distribution, and shape conguration characteristics of green space
any chance of speculation and thus prevented a bubble to form,
patches, as shown in Table 1.
maintaining a relatively rational market land price), we opt to still
We compute the DIST metric in ESRI ArcGIS, and all others in
use the database. However, the above mentioned problems may
Fragstats 4.2. As the CAR, LDI, and LSI metrics are only meaningful
affect the analysis results, which we address later when
in a given area of landscape, one must rst designate the specic
encountered.
landscape area to calculate these metrics. In this study, we desig-
There are in total 3659 land parcels in the dataset, with infor-
nate a 500  500 m window as the basic landscape area for analysis,
mation on land area size, land use (residential for all cases),
as it is roughly the average distance between green space patches in
approved development intensity (Floor-to-Area Ratio, FAR), and
the study area such that at least one patch is included in each
transaction price for each parcel. We have ination-adjusted all
window. Moreover, to get the metric values within the whole
88 L. Xu et al. / Habitat International 56 (2016) 84e95
Fig. 2. Green space parcels and land transactions in Beijing, 2000e2004.
landscape (study area), we perform a convolution with a sliding
window of the 500  500 m size mentioned above, and calculate
the average landscape metric values within the landscape. The re-
sults are shown in Fig. 3. It should to be noted here that due to the
existence of some No Data areas produced in the convolution
process, the real estate transaction records which happen to locate
in these holes cannot get a value of the metrics, and thus must be
excluded from the respective model. Hence, though the model with
the DIST variable can include all 3659 samples, the ones with the
CAR, LDI and LSI variables only have a sample size of 2922.
3.3. Specication of the Hedonic Price Model
We specify the HPM in this study following Equation (1). The
HPM typically takes a semi-log form, with the dependent variable
logarithmic transformed, for two reasons. First, housing or land
price samples are usually lognormal distributed (Poudyal et al.,
2009). Second, researchers usually focus on the changes rather
than the absolute levels of the dependent variable, the semi-log
equation form is thus more appropriate. In our case, moreover,
preliminary data explorations show that the landscape metrics are
also lognormal distributed. Therefore, we use a modied semi-log
OLS regression to estimate the model, with both the dependent
variable (land price) and the environmental variables (landscape
metrics) logarithmic transformed.
Regarding the variables, we rst include two variables: land
parcel size and approved structure area size in the structural group
Sij . As noted above, all land parcels are of the same designated land
use (residential), so land use is not included as a variable here.
Second, we include ve variables: distances to city center, district
center, nearest highway entrance, nearest urban street, and nearest
bus stop in the neighborhood group Nik . It should be noted that not
all the variables would necessarily enter the nal model, for issues
such as multicollinearity may render some variables statistically
insignicant.
Lastly and most importantly, variables in the environmental
group Eil include the spatial metrics discussed above. However, the
specic way in which the variables enter the model is worth noting
here. A search of relevant literatures reveals two ways of including
environmental variables in the HPM. On one hand, the rst
approach is to directly use the values of landscape metrics. For
example, one can construct a variable indicating a land parcel's
accessibility to urban green spaces using the distance from the land
parcel to the nearest green space patch, as in the case of Kong, Yin,
 L. Xu et al. / Habitat International 56 (2016) 84e95 89

Table 1
Landscape ecological metrics in the study.

Spatial characteristics LEM LEM denition* Notes

P
Richness Class Area Ratio CAR nj1 aij Ratio between the area of a land use class (urban green space in this case, and the same
(CAR) hereinafter) and the whole landscape area (or the area of the convolution sliding window) (m2)
Accessibility Nearest Distance DIST dnearest Distance from a real estate site to the nearest urban green space patch (m)
(DIST)
Distribution Landscape Division  2 A measurement of the fragmentation level of a land use class in a certain area. Ranging from
Pn aij
Index (LDI) LDI 1  j1 A 0 (Not fragmented at all) to 1 (Perfectly fragmented)
ei
Shape Landscape Shape LSI mine A measurement of the shape conguration of patches, LSI equals the total length of edge (or
i
Conguration Index (LSI) perimeter) involving the corresponding class, divided by the minimum length of class edge (or
perimeter) possible for a maximally aggregated class, which is achieved when the class is
maximally clumped into a single, compact patch (McGarigal, Cushman, & Ene, 2012).
LSI 1 implies a square patch; as LSI increases, the patch gets more complex in shape.

*Where aij area (m2) of patch ij; dnearest distance to the nearest patch; ni number of patches in the land use class (type) i; A total landscape area (m2); ei total length of
edge (or perimeter) of class i in terms of number of cell surfaces; min ei minimum total length of edge (or perimeter) of class i in terms of number of cell surfaces; includes all
landscape boundary and background edge segments involving class i. In this paper, there is only one possible value for i e the green spaces. The denitions of the metrics are
given by Fragstats(McGarigal, Cushman, and Ene 2012). For detailed description of the indices, refer to the source above.

and Nakagoshi (2007) and Poudyal et al. (2009). The second
approach, on the other hand, uses dummy variables. Specically, it
rst designate a few sections across the range of a landscape metric,
and then construct dummy variables accordingly, thus convey the
spatial pattern information in an indirect way. Examples of this
approach include Bolitzer and Netusil (2000) and Tyrvainen and
Miettinen (2000). The rst approach would directly reveal the in-
uence of the spatial effects on the dependent variable (land price),
but would also require high-quality data supports. The second
approach, on the contrary, is less effective in revealing the specic
inuence of the spatial effects, but also requires weaker data sup-
port. Therefore, we face a trade-off between data quality and
modeling efciency. As discussed above, both the urban green
space geographical data and the land price data used in this study
could be far from perfect, implying the potential benet of adopting
the second approach. Although this is a compromise we have to
make because of data limitations, we argue that imperfect data,
rather than high-quality data, is the normal in most real-world
settings, especially in developing countries such as China. In this
sense, the second approach has its unique merits.
It should also be noted that by following the second modeling
approach, the main purpose of this study is not to quantify the
monetary value premium brought about by the environmental
amenities (though we try to give some rough estimations later in
the paper). Rather, we focus on the very existence (or not) of the
spatial effects per se. To this end, we need only to do the following
hypothesis test at a given (for example, 5%) signicance level:
H0 : bl s0 and H1 : bl 0
We do the hypothesis tests in a series of models to nd the
effective range of spatial characteristics in which the spatial effects
exist. Taking the DIST variable as an example, we rst set a series of
cutoff values, ranging from 50 m to 100, 150, ..., 1000 m, and then
create a 0/1 dummy variable and assign the value for it based on the
cutoff values. For the cutoff value 50 m, for instance, we create a
dummy variable DIST 50, which is dened as follows:
DIST50 1, if DIST <50 m;
DIST50 0, Otherwise
Then, we use the new dummy variable DIST 50 to t the rst
model, denoting HPM (DIST 50). If the regression coefcient of the
dummy variable is statistically signicant, then we can safely
conclude that the presence of urban green spaces within 50 m of a
land development site does inuence its value. Similarly, we create
a series of dummy variables with different cutoff values and t a
group of models accordingly (Table 2). Then we only need to focus
on the regression coefcients of the dummy variables (DISTCOV),
and identify the range of cutoff values within which the coefcients
of DISTCOV are statistically signicant, thus indicating the effective
range of the spatial effects.
The criteria for setting the cutoff values is worth noting here. On
the one hand, for those landscape metrics with physical units (for
example, DIST, measured in meters), we simply arbitrarily set a
series of cutoff values such that the entire value range is evenly and
reasonably divided, a method similar to other researchers' (Bolitzer
& Netusil, 2000). On the other hand, for the standardized metrics

Fig. 3. The spatial pattern (LEM values) of the CAR, LDI and LSI variables.
90 L. Xu et al. / Habitat International 56 (2016) 84e95
Table 2
The group of HPMs for an environmental variable (DIST).
Model no. Model name
1 HPM (DIST 50)
2 HPM (DIST 100)
n HPM (DIST COVn)
which don't have an explicit physical meaning, we divide their
value ranges such that the sample size within each subinterval is
the same to avoid possible bias induced by radically skewed dis-
tribution of samples among the subintervals. Moreover, in both
occasions we make sure there are not too many subintervals that
unnecessarily complicates the analysis. In sum, the variables, their
descriptive statistics, and expected signs of regression coefcients
are shown in Table 3.
One major drawback of the above described modeling approach,
though, is that it eliminates the possibility of constructing a
complete model including all spatial-related characteristics. For
example, if we designate m dummy variables for spatial charac-
teristic 1 and n dummy variables for spatial characteristic 2, then
we would need m * n equations in a complete model group. As the
number of spatial characteristics examined increase, it would soon
become manifest that it is impractical to include all the spatial
characteristics in the same model. Besides, including all spatial
characteristic variables in the same model may not only be
impractical, but also undesirable, as different landscape metrics
deal with different aspects of spatial characteristics. For example, it
is easy to nd that the four landscape metrics used in this paper are
not correlated with each other at all, which is manifest as revealed
in the denitions of the landscape metrics per se. Thus, interactive
terms of different landscape metrics hardly make any practical
sense. Therefore, instead of tting a general but very complicated
model with all the environmental variables, we construct a series of
HPMs, with all possible variables in Sij and Nik in every model, but
only one spatial metric in each model as the Eil variable, such that
we focus on one spatial dimension at a time. Overall, each model
group would reveal the marginal effective range of the respective
spatial pattern, so that one can directly combine the marginal re-
sults from each model group to reach a comprehensive conclusion.
4. Results
Results show that, for different models with different environ-
mental variables, only three independent structural and neigh-
borhood variables are statistically signicant (at the 5% signicance
level) to enter all models. These are the approved construction area
on the land parcel (built_size), the distance to the city center
(dist_city), and the distance to the district center (dist_district).
Moreover, the regression coefcients for all the three variables
remain quite stable not only within each model group, but also
across different model groups. The coefcients for the environ-
mental dummy variable, however, vary dramatically between
models within each group, with some statistically signicant while
others not (Table 4). The fact means that the way the dummy var-
iables are constructed does inuence their relevance. Putting
another way, the spatial dimension does matter.3 We discuss the
model groups one by one in this section.
3
To avoid excessive details, we only demonstrate the ranges of R2, regression
coefcients (Beta), and T-values in each model group. Detailed information is
available upon request.
DIST cutoff Value Regression equation
P P P
50 ln pi b0 bj Sij bk Nik bl DIST50i i
P P P
100 ln pi b0 bj Sij bk Nik bl DIST100i i
P P P
COVn ln pi b0 bj Sij bk Nik bl DISTCOVn i i
4.1. Model group 1: the richness and ecosystem service value of
urban green spaces
The rst model group, with the land use area ratio (CAR) of
urban green spaces being the environmental variable, examines
whether e and if yes, how the richness of urban green spaces in-
uences their ecosystem service value as reected in land prices. As
shown in Table 5, the answer to the rst question depends on how
much area is occupied by green spaces in the urban landscape. The
dummy variable's regression coefcient is not statistically signi-
cant when the cutoff value is small, but becomes signicant (at a 5%
signicance level) as the cutoff value increases to reach a tipping
point, when green space patches occupy 2.20% (i.e. exp(3.8073))
of the total area, and then basically remains signicant until the
cutoff value reaches a high threshold (between 9.20% and 13.40%
green space ratio).
A notable problem, however, is that the regression coefcients
(standard beta) for the CAR variable are negative in all models, and
show no trend between models. This is against the intuitive ex-
pected sign of the variable, and is also inconsistent with the results
from other researchers (McConnell & Walls, 2005). We elaborate on
this issue in the discussions section. Nevertheless, despite the
negative regression coefcients, the main result of the model
group, i.e. that the existence of urban green spaces would
contribute to nearby land value when the former's size ratio in the
landscape exceeds a certain threshold, still holds.
4.2. Model group 2: the accessibility and ecosystem service value of
urban green spaces
The second model group, with the distance between land
development parcels and the nearest urban green space patch
(DIST) being the environmental variable, examines whether e and
if yes, how the accessibility of urban green spaces inuences their
ecosystem service value as reected in land prices. As shown in
Table 6, when a green space patch is within 50 m of a land parcel, its
inuence on the land price is statistically insignicant. However,
when a green space patch is within the 50e550 m range of distance
from a land development parcel, its inuence on the land price is
statistically signicant in all models. Lastly, when the distance be-
tween a green space patch and a land development parcel is larger
than 600 m, the former's inuence on the land price fades away
again as the respective regression coefcients again become sta-
tistically insignicant.
Overall, these results are in line with common sense, except for
the 50-m lower threshold of the spatial effects. We argue that this
lower threshold is very likely an estimation error, and the error
comes from a special data limitation. As indicated above, we only
include rst-order urban green spaces. These green space patches
usually occupy entire urban blocks, and are separated from other
land parcels by major urban streets, which, in the specic case of
Beijing, are seldom narrower than 50 m. Thus, in principle, the
number of land transaction sites that are within 50 m of nearby
urban green space parcels should be very small. In fact, there are
only 64 such cases in the land transactions database, and even
 L. Xu et al. / Habitat International 56 (2016) 84e95 91

Table 3
Denitions, descriptive statistics, and expected signs of independent variables.a

Variable Denition/Unit Mean Std. deviation Expected sign

Sample 1b Sample 2c Sample1 Sample 2

Common variables in all models

Land_size Area of parcel/m2 16499.97 15476.98 26974.71 25771.64
Built_size Approved construction area on parcel/m2 48474.85 47531.69 64747.60 65426.55
DIST_city Distance to city center/m 9455.55 9229.29 4697.87 4571.87 e
DIST_district Distance to district center/m 5559.41 5145.43 2778.54 2553.09 e
DIST_highway Distance to nearest highway entrance/m 751.21 661.73 811.29 671.98 e
DIST_urbanstreet Distance to nearest urban street/m 768.46 792.48 934.30 979.38 e
DIST_busstop Distance to nearest bus stop/m 402.00 380.18 276.43 253.09 e
Dummy variable groups in different model groups
CARk See Table 1. N/A N/A
DISTl
LDIp - when LDI is too large or too small, otherwise
LSIq Not clear
a
Dependent variable Land price.
b
Observations 3659.
c
Observations 2922.

Table 4
Summary of regression results.

Model group No Observations Environmental Variable R2 Structural Variable(s) entered Neighborhood Variable(s) entered Environmental Variable

Built_size Dist_city Dist_district Std. Beta T Value

Std.Beta T Value Std. Beta T Value Std.Beta T Value

1 2922 CARk High 0.33 0.54 35.19 0.15 9.79 0.06 4.05 0.01 0.54
Low 0.32 0.53 35.04 0.16 10.49 0.07 4.29 0.05 3.26
2 3659 DISTl High 0.35 0.55 40.75 0.17 12.47 0.07 5.01 0.01 0.61
Low 0.34 0.54 40.70 0.17 12.64 0.08 5.47 0.05 3.39
3 2922 LDIp High 0.33 0.54 35.15 0.15 9.72 0.06 4.05 0.04 2.84
Low 0.32 0.53 34.99 0.16 10.38 0.07 4.26 0.00 0.10
4 2922 LSIq High 0.33 0.54 35.18 0.16 10.46 0.07 4.22 0.01 0.83
Low 0.32 0.53 35.10 0.16 10.60 0.07 4.34 0.03 1.83

these 64 cases are possibly produced by mapping errors. Putting
another way, the entire less than 50 m category makes little sense
in the specic study context, so is the insignicant regression
coefcient.
Like in the previous model group, the DIST variables in this
model group show negative regression coefcients, too, still with
no trend between models. Similarly, we discuss possible explana-
tions in the next section.
4.3. Model group 3: the distribution and ecosystem service value of
urban green spaces
The third model group, with the degree of division (LDI) of green
spaces being the environmental variable, examines whether e and
if yes, how the fragmentation status of urban green spaces
inuences their ecosystem service value as reected in land prices.
By denition, given total urban green space area, the larger the LDI
value is, the more fragmented the green space patches are; there-
fore, the LDI is in some sense a comprehensive indicator for the
spatial distribution of urban green space in a landscape. As shown
in Table 7, results show both an upper and a lower threshold in the
ln LDI value range, and only the models with LDI cutoff values in
between yields statistically signicant results in which the positive
regression coefcients indicate a positive inuence on nearby land
values when the degree of division of urban green spaces is within
the range, an instinctive result that is in line with our expectation.
Other models have insignicant environmental variable co-
efcients, thus yielding no meaningful results. We elaborate on the
physical meaning of the ln LDI range in the next section.

Table 5
Regression results for the class area ratio (CAR) model group.

Sub-model no. Environmental dummy Variable name Ln CAR cutoff Std. Beta Sig.

1 CAR_1 5.452 0.008 0.591

2 CAR_2 4.759 0.010 0.506
3 CAR_3 4.232 0.027 0.081
4 CAR_4 3.807 0.032 0.038
5 CAR_5 3.525 0.027 0.082
6 CAR_6 3.258 0.030 0.049
7 CAR_7 2.977 0.042 0.007
8 CAR_8 2.702 0.050 0.001
9 CAR_9 2.385 0.036 0.018
10 CAR_10 2.006 0.023 0.128

The bold, italics gures indicate signicance levels that are below 0.05.
92 L. Xu et al. / Habitat International 56 (2016) 84e95

4.4. Model group 4: the shape conguration and ecosystem service
value of urban green spaces
The last model group, with the LSI of green spaces being the
environmental variable, examines whether e and if yes, how the
shape conguration of urban green spaces inuences their
ecosystem service value as reected in land prices. By denition,
the LSI is an indicator that measures how much a shape is
different from a square, with the value of 1 when the shape is a
square, and larger values when the shape is deviated from a
square. As shown in Table 8, no model in this group yields a sta-
tistically signicant regression coefcient for the environmental
variable (LSI), thus giving no meaningful results. We discuss on
possible explanations for these results in the next section.
5. Analysis and discussions
Among the three model groups yielding statistically signicant
results, on the one hand, the rst and second one (on CAR and DIST,
respectively) give results with explicit physical meanings (in terms
of green space ratios and distances, respectively), though the re-
sults are in some cases against the common sense. We discussion
the possible causes to this problem. The model group for LDI, on the
other hand, does not give results with explicit physical meanings, as
the LDI is a standardized metric that does not have any explicit
physical meaning. However, based on the regression results, we can
interpret the results by quantitatively calculating the range for
urban green spaces degree of fragmentation in which they are
inuential on real estate values, as well as the optimal degree of
fragmentation in which the most inuential effects are realized.
Also, we discuss on possible causes for the non-signicant results in
the Landscape Shape Index (LSI) model group, and give an expla-
nation on it.
5.1. On the counter-intuitive regression coefcients in the richness
and accessibility model groups
In the regression results of the richness and accessibility model
groups, a most notable problem is that the regression coefcients
for the CAR and DIST variables are negative in all models that are
statistically signicant, and show no trend between models. These
results are clearly counter-intuitive, as they imply that increased
richness and accessibility to urban green spaces would decrease
nearby real property values whenever the spatial effects are
effective. We attribute the cause of the problem to data quality
issues. As discussed above, both the land price data and the
geographical data on urban green spaces may be problematic, and
thus constitute sources of errors. On the one hand, with regard to
the geographical urban green space data, the dataset may not be
complete due to administrative reasons, and the omission of certain
second-order green spaces may affect the model estimations. A
rened geographic data on urban green spaces, therefore, may help
eliminate such errors. On the other hand, with regard to the land
price data, it should be noted that Beijing's land market during the
study period was not a totally free-market one, and that the
possibly irrational land prices may result in the strange negative
regression coefcients. The problem, therefore, may disappear in a
real free-market context. We admit that these problems are the
major limitations of this study, and they are subject to future in-
quiries. Nevertheless, despite the unexpected signs of the regres-
sion coefcients, the results concerning the effective range of the
spatial effects are robust. In sum, green spaces should occupy be-
tween 2.20% and 13.40% of the total urban landscape area, and
within a 100e550 m range from nearby real estate development
sites, such that the spatial effects are effectively realized.
5.2. On the degree of fragmentation of urban green spaces: the
effective range and optimal level
The LDI measures the degree of fragmentation of a land use class
(i.e. urban green spaces) in a landscape, and is given in the form of
the ratio between green space patches and the landscape area.
Therefore, to calculate the thresholds and optimal levels of frag-
mentation of urban green spaces as measured by the area ratio
between the patches and the total green space area, one needs
additional information regarding the ratio of area occupied by
green spaces in the urban landscape. In this paper's particular
context of Beijing, as mentioned earlier, the gure is about 10%
within the study area. We base the following calculations on this
gure.
As shown in Table 7, the lower and upper thresholds for statis-
tically signicant regression coefcients for the ln LDI variable

Table 6
Regression results for the Accessibility (DIST) model group.

Sub-model no. Environmental dummy variable name DIST cutoff Std. Beta Sig.

1 DIST_1 50 0.014 0.308

2 DIST_2 100 0.032 0.017
3 DIST_3 150 0.032 0.018
4 DIST_4 200 0.029 0.031
5 DIST_5 250 0.032 0.016
6 DIST_6 300 0.044 0.001
7 DIST_7 350 0.047 0.001
8 DIST_8 400 0.043 0.001
9 DIST_9 450 0.032 0.018
10 DIST_10 500 0.034 0.012
11 DIST_11 550 0.034 0.012
12 DIST_12 600 0.022 0.111
13 DIST_13 650 0.013 0.340
14 DIST_14 700 0.007 0.589
15 DIST_15 750 0.003 0.802
16 DIST_16 800 0.003 0.843
17 DIST_17 850 0.000 0.983
18 DIST_18 900 0.008 0.557
19 DIST_19 950 0.001 0.933
20 DIST_20 1000 0.000 0.992

The bold, italics gures indicate signicance levels that are below 0.05.
 L. Xu et al. / Habitat International 56 (2016) 84e95 93

Table 7
Regression results for the landscape division index (LDI) model group.

Sub-model no. Environmental dummy variable name Ln LDI cutoff Std. Beta Sig.

1 LDI_1 0.0278 0.023 0.129

2 LDI_2 0.0118 0.025 0.102
3 LDI_3 0.0064 0.033 0.034
4 LDI_4 0.0036 0.039 0.013
5 LDI_5 0.0020 0.044 0.005
6 LDI_6 0.0010 0.043 0.006
7 LDI_7 0.0006 0.036 0.018
8 LDI_8 0.0003 0.031 0.045
9 LDI_9 0.0001 0.007 0.654
10 LDI_10 0.00003 0.002 0.917

The bold, italics gures indicate signicance levels that are below 0.05.

Table 8
Regression results for the landscape shape index (LSI) model group.

Sub-model no. Environmental dummy variable name Ln LSI cutoff Std. Beta Sig.

1 LSI_1 0.1335 0.007 0.643

2 LSI_2 0.2231 0.013 0.409
3 LSI_3 0.3064 0.002 0.871
4 LSI_4 0.3795 0.015 0.336
5 LSI_5 0.4418 0.017 0.258
6 LSI_6 0.5108 0.007 0.659
7 LSI_7 0.5718 0.000 0.995
8 LSI_8 0.6286 0.000 0.982
9 LSI_9 0.6931 0.007 0.651
10 LSI_10 0.7481 0.023 0.131
11 LSI_11 0.7885 0.026 0.092
12 LSI_12 0.8313 0.021 0.162
13 LSI_13 0.8928 0.021 0.172
14 LSI_14 0.9372 0.024 0.111
15 LSI_15 0.9860 0.007 0.650
16 LSI_16 1.0427 0.009 0.556
17 LSI_17 1.0935 0.006 0.715
18 LSI_18 1.1477 0.020 0.189
19 LSI_19 1.2303 0.028 0.067

are 0.0064 and 0.0003, respectively, and the value of ln LDI
when the regression coefcient is at the maximum (i.e. realizing
the most ecosystem service value) is 0.0020. Thus:
LDILower exp0:0064 0:9936
LDIUpper exp0:0003 0:9997
LDIOptimal exp0:0020 0:9980
Without loss of generality, assuming all green space patches in
the landscape are evenly divided, and there are n patches in total.
By the denition of LDI:
 2
aj
1  LDI n (2)
A
Let g denote the overall green space ratio in the landscape:
naj
g (3)
A does not mean that a monotonously and continuously changing LSI
Insert Equation (3) into Equation (2), and apply g 10%, we can
calculate that the lower and upper thresholds for n is between 1.60
and 33.00, respectively, and the optimal value is 5.00. Putting in
terms of area ratios, the ratio each green space patch should occupy
in total green space area is 3.00%, 62.50%, and 20.00% at the lower,
upper, and optimal levels, respectively. This is to say, in order to
effectively realize the ecosystem service value embedded in urban
green spaces, the green space patches should be so distributed that
every patch occupies more than 3.00% but less than 62.50% of the
total green space area, and when each patch occupies 20.00% of the
total green space area, the ecosystem service value is optimally
realized.
It should be noted, however, that these gures are only effective
for the Beijing case as they are calculated based on city's land
market data, and also assuming a 10% total green space ratio which
is also specic of Beijing's context. Nevertheless, the analysis and
calculation methods are universally applicable. We thus expect
these methods be tested, applied, and further developed in other
contexts by cohorts in the academic community.
5.3. An explanation for the non-signicant results in the LSI model
group
The models with LSI as the environmental variable have yielded
no statistically signicant regression coefcients for the LSI vari-
able. We attribute these results to the innate ambiguousness in the
very denition of the LSI. As stated earlier, the LSI by denition only
measures how much a shape is deviated from a square, but this
series necessarily implies a continuously changing shape series as
well. For example, it is easy to construct a shape series as shown in
Fig. 4. The series is actually composed of two alternating sub-series,
with the square series becoming more and more convex, and the
half-circle series becoming more and more concave, but the LSI
values for the composed series monotonously increase. In fact, this
example is of specic meaning from a landscape ecological
perspective, which states that the concave and convex patches in a
94 L. Xu et al. / Habitat International 56 (2016) 84e95
Fig. 4. Two alternating shape series with monotonously increasing LSI values.
landscape have different edge effects, and therefore such two so
evolving series may lead to opposite ecological effects, such as
ecosystem service value (Forman, 1995), a result that would not
otherwise have been revealed by the seemingly monotonously
changing LSI series.
Therefore, the LSI alone does not provide enough information
that can reect all necessary shape conguration characteristics of
a green space patch. Therefore, a model solely based on it hardly
makes sense, thus naturally the non-signicant results. In contrast,
indicators such as DIST and LDI convey clear physical meanings
(either explicitly or implicitly), thus the denite results. For a better
measurement of the shape conguration characteristics of green
space patches, one needs to introduce additional indicators, which
is beyond the scope of this paper and we leave it for further studies.
6. Conclusion
The study clearly shows that the spatial pattern of urban green
spaces does matter when it comes to the realization of their
ecosystem service value. Though the direction and magnitude of
the spatial effects are in some cases unclear, or cannot be deter-
mined in other cases as limited by the quality of data, the existence
of the spatial effects per se is manifest. Specically, the distance
from a green space patch to a nearby land development parcel
determines whether the green space patch's ecosystem service
value can contribute to the land parcel's market value, and green
space patches with different sizes and distributional characteristics
may have varying degrees of value-adding effect. Furthermore, an
optimal level of spatial fragmentation of urban green spaces that
maximizes the value-adding effect also exists. To our knowledge,
this is the rst piece of work that systematically examine the
different aspects of the spatial pattern of urban green spaces and
their inuence on urban green spaces' ecosystem service value,
thus these ndings constitute the main contribution of this study to
the academic literature.
Moreover, with the additional information of overall green
space ratio, quantitative estimations on the conditions on which
the spatial effects exist can be made specically for the Beijing case.
It is suggested that green spaces should occupy between 2.20% and
13.40% of the urban landscape area, and within a 100e550 m range
from nearby real estate development sites. Also, green space
patches should be so distributed that every patch occupies more
than 3.00% and less than 62.50% of the total green space area to
ensure effective ecosystem service value-adding effect, and that the
effect is optimized when every green space patch occupies 20.00%
of the total green space area. Again, it should be noted that these
quantitative results are context-dependent and are therefore not to
be simply generalized.
The major limitation of the study is that due to data quality is-
sues, we are not able to nd the specic magnitude of the spatial
effects. The implication of this limitation is twofold. On the one
hand, since the lack of high-quality data is the normal condition in
most cases in a develop country context, the analysis method that
we develop in this study can be applied in such conditions when
only imperfect data are available, and they can still yield results that
are meaningful to some extent. This illustrates the universal
applicability of our modeling and estimation methods. We thus
expect other researchers apply the method in various contexts so as
to yield more empirical results. On the other hand, however, the
specic magnitude of the spatial effects are indeed an important
aspect of the inquiry. Though the original database we use in the
study has little potential of being improved, we do expect that we
analyze the same study area in a different time period, hopefully
with fully marketized price information and rened geographical
data of urban green spaces, such that estimates on the magnitude of
the spatial effects can be made, and that comparative studies can be
done. Also, the analysis on the shape conguration characteristics
of urban green spaces yields no results. As discussed above, this
would require the construction of more effective measurements of
the shape conguration characteristics of urban green spaces. We
also leave this methodological study for future inquires.
Further, in practical terms, a direct practical implication of the
study is that urban green space systems can, and should be spatially
optimized such that they realize the most ecosystem service value.
Such optimization should be carried out in a systematical manner,
which means not only a certain aspects of the spatial characteristics
of urban green spaces are at concern; rather, all relevant aspects of
the spatial characteristics should be included. Items to optimize
thus include the specic way green spaces and other built envi-
ronment elements are spatially arranged, the size of the green
space patches, the way the green space patches are divided, and
also perhaps the shape conguration of the green space patches.
Given the prevailing urban environment problems across the globe,
we suggest such optimization task be explicitly included in urban
planning codes for all cities, and that the results from a number of
empirical research projects may help contribute to the making of a
general quantitative guideline for the spatial planning of green
spaces in the urban environment in a way to optimize their
ecosystem service value.
Acknowledgments
We would like to thank the Beijing Municipal Administration of
Land Resources for permission of use of the land transaction re-
cords data and the Beijing GIS database.
References
Bateman, I. J., Harwood, A. R., Mace, G. M., Watson, R. T., Abson, D. J., Andrews, B.,
et al. (2013). Bringing ecosystem services into economic decision-making: land
use in the United Kingdom. Science, 341(6141), 45e50. http://science.
sciencemag.org/content/341/6141/45.abstract.
Bingham, G., Bishop, R., & Brody, M. (1995). Issues in ecosystem Valuation: Improving
information for decision making.
Bolitzer, B., & Netusil, N. R. (2000). The impact of open spaces on property values in
Portland, Oregon. Journal Of Environmental Management, 59(3), 185e193. http://
gateway.isiknowledge.com/gateway/Gateway.cgi?GWVersion2&SrcAuth
AegeanSoftware&SrcAppNoteExpress&DestLinkTypeFullRecord&DestApp
 L. Xu et al. / Habitat International 56 (2016) 84e95
WOS&KeyUT000089385700002.
Booth, D. B., & Jackson, C. R. (1997). Urbanization of aquatic systems: degradation
thresholds, stormwater detection, and the limits of mitigation. Journal of the
American Water Resources Association, 33(5), 1077e1090.
Chau, K. W., Ma, V. V., & Ho, D. C. W. (2001). The pricing ofLuckinessin the apartment
market.
Cilliers, S., Cilliers, J., Lubbe, R., & Siebert, S. (2012). Ecosystem services of urban
green spaces in african countriese-perspectives and challenges. Urban Ecosys-
tems, 16(4), 681e702. http://dx.doi.org/10.1007/s11252-012-0254-3.
Costanza, R., Arge, R., Rudolf de, G., Farber, S., Grasso, M., Hannon, B., et al. (1997).
The value of the world's ecosystem services and natural capital. Nature,
387(6630), 253e260. http://dx.doi.org/10.1038/387253a0.
Costanza, R., Stern, D., & Fisher, B. (2004). Inuential publications in ecological eco-
nomicas: A citation analysis.
Czembrowski, P., & Kronenberg, J. (2016). Hedonic pricing and different urban green
space types and sizes: insights into the discussion on valuing ecosystem ser-
vices. Landscape and Urban Planning, 146(February), 11e19. http://dx.doi.org/
10.1016/j.landurbplan.2015.10.005.
De Ridder, Kon, Adamec, V., Ban ~ uelos, A., Bruse, M., Brger, M., et al. (2004). An
integrated methodology to assess the benets of urban green space. Science of
the Total Environment, 334, 489e497. Elsevier.
De Vries, Sjerp, Verheij, R. A., Groenewegen, P. P., & Spreeuwenberg, P. (2003).
Natural environments-healthy environments? an exploratory analysis of the
relationship between greenspace and health. Environment and Planning A,
35(10), 1717e1732.
Derkzen, Marthe, L., van Teeffelen, A. J. A., & Verburg, P. H. (2015). Review: quan-
tifying urban ecosystem services based on high-resolution data of urban green
space: an assessment for Rotterdam, the Netherlands. Journal of Applied Ecology,
52(4), 1020e1032. http://dx.doi.org/10.1111/1365-2664.12469.
Economy, & Elizabeth, C. (2011). The River Runs Black: The Environmental Challenge
to China's Future. Cornell University Press.
Forman, R. T. T. (1995). Land Mosaics: The Ecology of Landscapes and Regions. Cam-
bridge university press.
Go mez-Baggethun, E., & Barton, D. N. (2013). Classifying and valuing ecosystem
services for urban planning. Ecological Economics, 86(February), 235e245.
http://dx.doi.org/10.1016/j.ecolecon.2012.08.019.
Haase, D., Larondelle, N., Andersson, E., Artmann, M., Borgstro m, S., Breuste, J., et al.
(2014). A quantitative review of urban ecosystem service assessments: con-
cepts, models, and implementation. AMBIO, 43(4), 413e433. http://dx.doi.org/
10.1007/s13280-014-0504-0.
Heynen, N., Perkins, H. A., & Roy, P. (2006). The political ecology of uneven urban
green space the impact of political economy on race and ethnicity in producing
environmental inequality in Milwaukee. Urban Affairs Review, 42(1), 3e25. Sage
Publications.
Hillsdon, M., Panter, J., Foster, C., & Jones, A. (2006). The relationship between access
and quality of urban green space with population physical activity. Public
Health, 120(12), 1127e1132. Elsevier.
Hui, E. C. M., & Yue, S. (2006). Housing price bubbles in Hong Kong, Beijing and
Shanghai: a comparative study. The Journal of Real Estate Finance and Economics,
33(4), 299e327. Springer.
Jiang, B., Larsen, L., Deal, B., & Sullivan, W. C. (2015). A doseeresponse curve
describing the relationship between tree cover density and landscape prefer-
ence. Landscape and Urban Planning, 139(July), 16e25. http://dx.doi.org/10.1016/
j.landurbplan.2015.02.018.
Jiang, B., Zhang, T., & Sullivan, W. C. (2015). Healthy cities: mechanisms and research
questions regarding the impacts of urban green landscapes on public health
and well-being. Landscape Architecture Frontiers, 3(1), 24e35.
Jim, C. Y., & Chen, W. Y. (2006). Recreation-amenity use and contingent valuation of
urban greenspaces in Guangzhou, China. Landscape and Urban Planning,
75(1e2), 81e96. http://gateway.isiknowledge.com/gateway/Gateway.cgi?
GWVersion2&SrcAuthAegeanSoftware&SrcAppNoteExpress&
DestLinkTypeFullRecord&DestAppWOS&KeyUT000234771600008.
Jim, C. Y., & Chen, W. Y. (2008). Assessing the ecosystem service of air pollutant
removal by urban trees in Guangzhou (China). Journal of Environmental Man-
agement, 88(4), 665e676. Elsevier.
Kong, F. H., Yin, H. W., & Nakagoshi, N. (2007). Using GIS and landscape metrics in
the Hedonic Price Modeling of the amenity value of urban green space: a case
study in jinan city, China. Landscape and Urban Planning, 79(3e4), 240e252.
http://gateway.isiknowledge.com/gateway/Gateway.cgi?GWVersion
2&SrcAuthAegeanSoftware&SrcAppNoteExpress&DestLinkType
FullRecord&DestAppWOS&KeyUT000244584600005.
Lancaster, K. J. (1966). A new approach to consumer theory. The Journal of Political
95
Economy, 74(2), 132e157.
Liu, J., & Diamond, J. (2005). China's environment in a globalizing world. Nature,
435(7046), 1179e1186. Nature publishing group.
Mak, S. W. K., Choy, L. H. T., & Ho, W. K. O. (2007). Privatization, housing conditions
and affordability in the People's Republic of China. Habitat International, 31(2),
177e192. Elsevier.
McConnell, Virginia, & Walls, Margaret A. (2005). The value of open space: Evidence
from studies of nonmarket benets. Washington, DC: Resources for the Future.
McGarigal, K., Cushman, S. A., & Ene, E. (2012). FRAGSTATS v4: spatial pattern
analysis program for categorical and continuous maps. University of Massa-
chusetts, Amherst. In Computer software program produced by the authors at the
Uni-Versity of Massachusetts.
McMichael, A. J. (2000). The urban environment and health in a world of increasing
globalization: issues for developing countries. Bulletin of the World Health Or-
ganization, 78(9), 1117e1126. SciELO Public Health.
Morancho, A. B. (2003). A hedonic valuation of urban green areas. Landscape and
Urban Planning, 66(1), 35e41. http://gateway.isiknowledge.com/gateway/
Gateway.cgi?GWVersion2&SrcAuthAegeanSoftware&SrcApp
NoteExpress&DestLinkTypeFullRecord&DestAppWOS&KeyUT
000187753900004.
Mortberg, U., & Wallentinus, H.-G. (2000). Red-listed forest bird species in an urban
environmentdassessment of green space corridors. Landscape and Urban
Planning, 50(4), 215e226. Elsevier.
Neumann, B. C., J Boyle, K., & R Bell, K. (2009). Property price effects of a national
wildlife refuge: great Meadows National Wildlife Refuge in Massachusetts. Land
Use Policy, 26(4), 1011e1019. http://gateway.isiknowledge.com/gateway/
Gateway.cgi?GWVersion2&SrcAuthAegeanSoftware&SrcApp
NoteExpress&DestLinkTypeFullRecord&DestAppWOS&KeyUT
000266197500015.
Pearce, D. (1998). Cost-benet analysis and environmental policy.
Poudyal, N. C., Hodges, D. G., & Merrett, C. D. (2009). A hedonic analysis of the
demand for and benets of urban recreation parks. Land Use Policy, 26(4),
975e983. http://gateway.isiknowledge.com/gateway/Gateway.cgi?
GWVersion2&SrcAuthAegeanSoftware&SrcAppNoteExpress&
DestLinkTypeFullRecord&DestAppWOS&KeyUT000266197500012.
Poudyal, N. C., Hodges, D. G., Tonn, B., & Cho, S. H. (2009). Valuing diversity and
spatial pattern of open space plots in urban neighborhoods. Forest Policy and
Economics, 11(3), 194e201. http://gateway.isiknowledge.com/gateway/Gateway.
cgi?GWVersion2&SrcAuthAegeanSoftware&SrcAppNoteExpress&
DestLinkTypeFullRecord&DestAppWOS&KeyUT000266155200006.
Rapport, D. J., Gaudet, C., & Karr, J. R. (1998). Evaluating landscape health: Integrating
societal goals and biophysical process.
Sander, H. A., & Polasky, S. (2009). The value of views and open space: estimates
from a hedonic pricing model for Ramsey County, Minnesota, USA. Land Use
Policy, 26(3), 837e845. http://gateway.isiknowledge.com/gateway/Gateway.cgi?
GWVersion2&SrcAuthAegeanSoftware&SrcAppNoteExpress&
DestLinkTypeFullRecord&DestAppWOS&KeyUT000264647500034.
Shao, M., Tang, X., Zhang, Y., & Li, W. (2006). City clusters in China: air and surface
water pollution. Frontiers in Ecology and the Environment, 4(7), 353e361. Eco Soc
America.
Stern, D. I., Common, Michael S., & Barbier, E. B. (1996). Economic growth and
environmental degradation: the environmental Kuznets curve and sustainable
development. World Development, 24(7), 1151e1160. Elsevier.
Tajima, K. (2003). New estimates of the demand for urban green space: implications
for valuing the environmental benets of Boston's Big Dig Project. Journal of
Urban Affairs, 25(5), 641e655. http://gateway.isiknowledge.com/gateway/
Gateway.cgi?GWVersion2&SrcAuthAegeanSoftware&SrcApp
NoteExpress&DestLinkTypeFullRecord&DestAppWOS&KeyUT
000187291200006.
Tyrvainen, L., & Miettinen, A. (2000). Property prices and urban forest amenities.
Journal of Environmental Economics and Management, 39(2), 205e223. http://
gateway.isiknowledge.com/gateway/Gateway.cgi?GWVersion2&SrcAuth
AegeanSoftware&SrcAppNoteExpress&DestLinkTypeFullRecord&
DestAppWOS&KeyUT000086021400005.
Wolch, J. R., Byrne, J., & Newell, J. P. (2014). Urban green space, public health, and
environmental justice: the challenge of making cities just green enough.
Landscape and Urban Planning, 125(May), 234e244. http://dx.doi.org/10.1016/
j.landurbplan.2014.01.017.
Wu, J., & Hobbs, R. (2002). Key issues and research priorities in landscape ecology:
an idiosyncratic synthesis. Landscape Ecology, 17(4), 355e365. Springer.
Xie, G., Xiao, Y., & Lu, C. (2006). Study on ecosystem services: progress, limitation,
and basic paradigm (In Chinese). Journal of Plant Ecology, 30(2), 191e199.