Habitat International 72 (2018) 57e65

Contents lists available at ScienceDirect

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

Modeling urban growth boundary based on the evaluation of the

extension potential: A case study of Wuhan city in China
Qingsong He a, Ronghui Tan b, Yuan Gao a, Mengke Zhang a, Peng Xie a, Yaolin Liu a, c, d, *
a
School of Resource and Environment Science, Wuhan University, 129Luoyu Road, Wuhan, 430079, Hubei Province, PR China
b
College of Management and Economics, Tianjin University, No.92, Weijin Road, Nankai District, Tinjin, 300072, PR China
c
Key Laboratory of Geographic Information System, Ministry of Education, Wuhan University, 129Luoyu Road, Wuhan, 430079, Hubei Province, PR China
d
Collaborative Innovation Center for Geospatial Information Technology, PR China

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

Article history: The urban growth boundary (UGB) concept is useful in the field of urban planning, but models that can
Received 30 May 2016 simulate the change in UGBs remain limited to date. In this paper, we propose a model known as UBEM
Received in revised form that can simulate the future UGB. UBEM combines historical trajectories of UGB development and its
12 September 2016
extension potential in each azimuth to predict the future UGB for one city. UBEM consists of two parts: 1)
Accepted 11 November 2016
Available online 16 November 2016
the radiation method (RM) is used to describe the incremental length between the urban boundaries. In
RM, urban centroids are used as the origin points to generate a set of radial lines from each azimuth, and
we calculate the total and annual urban boundary length increments for each azimuth. 2) the extension
Keywords:
Urban growth boundary
pressure of the urban boundary is evaluated for different azimuths based on the potential value, which is
UBEM generated by selecting a set of variables that are related to urban growth potential. Multiple time series
Radiation method maps were used to calibrate the model to reduce the randomness in future modeling. We compare the
Extension potential calibrated modeling result with those generated by the uncalibrated UBEM and a separate null model,
Wuhan applying two goodness of fit metrics to evaluate model accuracy: percent area match (PAM) quantity and
Urban planning PAM location were used to demonstrate that the calibrated UBEM performed better than the uncali-
brated UBEM and null model when modeling the change in the urban boundary. Wuhan City in central
China is used as a case study to test the viability of UBEM and predict the future UGB in 2020. The
predication result offers helpful guidelines for Wuhan's future urban planning and UGB design.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction
Urbanization promotes social and economic development and
improves people's living standards. However, uncontrolled urban
sprawl generates a series of problems, such as the deterioration of
ecological security and a loss of agricultural land (Habibi & Asadi,
2011; Long, Liu, Hou, Li, & Li, 2014; Su, Jiang, Zhang, & Zhang,
2011). These problems seriously threaten the human environ-
ment as well as economic and social sustainable development. The
science-based and effective control of urban growth has become an
important component of city planning that must be considered in
urban planning and policy design. Urban containment policies, the
* Corresponding author. School of Resource and Environment Science, Wuhan
University, 129Luoyu Road, Wuhan, 430079, Hubei Province, PR China.
E-mail addresses: baihualin2013@163.com (Q. He), rhtan@tju.edu.cn (R. Tan),
gaoyuan920418@163.com (Y. Gao), 424035654@qq.com (M. Zhang), xiepenggis@
163.com (P. Xie), liuyaolin1999@126.com (Y. Liu).
basic goal of which is to increase the density of urban land use and
protect open space, have been widely used (Han, Lai, Dang, Tan, &
Ci-Fang, 2009). This type of policy includes three practices:
greenbelts, urban growth boundaries (UGBs), and urban service
boundaries (USBs). Among these practices, UGBs are the most
widely discussed in academics. UGBs draw lines between the urban
area and the rural area using regional development licenses and
other land use policies to direct the legal urban expansion within
the UGB to control the scale, time sequence, and shape of the urban
area (Calthorpe & Fulton, 2001; Martin, Pendall, & Fulton, 2002).
UGBs, as useful tools for government planning departments to
develop future urban development planning policies, have been
widely used all over the world (Ball, Cigdem, Taylor, & Wood, 2014;
Bengston and Yuon, 2006; Eddo, 2007; Gunn, 2007; Han et al.,
2009; Sinclair-Simth, 2014; Tayyebi, Delavar, Saeedi, & Amini,
2008; Wang, Han, & Lai, 2014; Wan et al., 2014). The UGB concept
originated in the United States in the 1960s and 70s, when the
country was in the middle-late stages of urbanization and many

http://dx.doi.org/10.1016/j.habitatint.2016.11.006
0197-3975/© 2016 Elsevier Ltd. All rights reserved.
58 Q. He et al. / Habitat International 72 (2018) 57e65
cities were facing the problem of disordered urban sprawl. In order
to resolve these issues, many new theories were proposed to
explore and implement UGBs, such as Smart Growth, Growth
Management, Infill Development and New Urbanism (Benfield
et al., 2001; David & Shayne, 2005). After the subsequent use of
UGBs in Great Britain (Gunn, 2007), Australia (Eddo, 2007), Saudi
Arabia (Mubarak, 2004), and Korea (Bengston & Youn, 2006) this
concept was accepted as the most effective way to curb urban
sprawl and resolve corresponding problems. Zhang first introduced
the UGB concept and other urban growth management tools in
China, and they have been applied in some cities throughout the
country (Han et al., 2009; Long, Han, & Mao, 2009, 2013; Wang
et al., 2014). The Ministry of Housing and the Ministry of Land
and Resource selected 14 cities in which to conduct UGB pilot. The
first set of pilot cities in July 2014 included Beijing, Shanghai, and
Guangzhou, among others. Their goals were to delineate the UGBs
when implementing urban planning, limit the expansion scale, and
limit the scope of construction in the city. The next phase will
expand this planning approach to more than 600 cities in China to
develop the UGBs. There is no doubt that UGB as an urban planning
tool plays an important role in promoting sustainable city devel-
opment. However, models that can simulate changes in UGB are
quite limited to date (Tayyebi, Pijanowski, & Pekin, 2011; Tayyebi,
Pijanowski, & Tayyebi, 2011). Therefore in this paper, we propose
a model that can simulate UGB to help urban planners identify and
forecast the future urban boundary extension.
1.1. Literature review on UGB modeling
UGB modeling can be viewed as one form of urban dynamics
simulation. Traditional urban dynamics simulation mainly focused
on modeling the change in individual urban pixels, describing ur-
ban boundaries using what is known as “planar” expansion, while
boundary modeling is referred to as “linear” expansion modeling.
Cellular automata (CA) and CA-based model are the most
frequently used among the “planar” expansion methods. They use
simple and dynamically updated local rules to convert the status of
the cells to simulate the complex process of urban expansion, and
have a wide range of applications in the field of urban dynamic
change modeling around the word (Clarke & Gaydos, 1998; Clarke
et al., 1997; Rienow & Goetzke, 2014; Yang & Xia, 2006; Yeh & Li,
2002; Yeh and Li, 2002). Some scholars use CA in combination
with other models to simulate future urban shape and then
delineate the outline of the built-up area as the UGB (Long et al.,
2009, 2013; Puertas, Henríquez, & Meza, 2014). However, these
efforts have been at very small scales (the pixel level) and they fail
to directly simulate and forecast urban boundaries.
In “linear” urban expansion modeling, Tayyebi and his team
have made significant contributions to urban boundary simulation.
Tayyebi, Pijanowski, and Pekin (2011), Tayyebi, Pijanowski, and
Tayyebi (2011), Tayyebi, Perry, and Tayyebi, (2014) suggest UGB
models that can directly simulate the urban boundary. These
include two rule-based UGB models (TR-UGBM) (Tayyebi,
Pijanowski, & Pekin, 2011), artificial neural networks UGB model
(ANN-UGBM) (Tayyebi, Pijanowski, & Tayyebi, 2011) and spatial
logistic regression UGB model (SLR-UGBM) (Tayyebi et al., 2014)
The TR-UGBM used the two rule-based models Distance Dependent
Method (DDM) and Distance Independent Method (DIM), to
simulate the UGBs, calculating the straight line azimuth and the
distance from the urban center to simulate change in UGBs. The
ANN-UGBM combined artificial neural networks (ANN) with GIS
and RS to predict UGB, using variables that affect the urban
boundary extension as input training parameters for the ANN on
the predictor variables of urban boundary geometry, with satis-
factory results. The SLR-UGBM used spatial logistic regression to
quantify how the driver variables influence the urban boundary
extension at each iteration over time, and forecast the future urban
boundary expansion. Aside from the generation of UGB models,
Tayyebi and others also established quantity and location metrics
that enable us to evaluate the accuracy of urban boundary models:
percent area match (PAM) quantity and location (Tayyebi,
Pijanowski, & Pekin, 2011; Tayyebi, Pijanowski, & Tayyebi, 2011,
2014). However, TR-UGBM simulations of the UGBs only consider
the azimuth and distance and ignore some important driving forces
affecting urban expansion, such as distance to water, distance to
urban centroids, distance to major roads, elevation, and slope
(Müller, Steinmeier, & Küchler, 2010; Poelmans & Rompaey, 2010;
Sunde, He, Zhou, Hubbart, & Spicci, 2014; Vermeiren, Rompaey,
Loopmans, Serwajja, & Mukwaya, 2012). Although ANN-UGBM
and SLR-UGBM take some driving forces of urban expansion into
account, these models only use two temporal periods, It is difficult
to guarantee a reliable conclusion if the result is obtained from
sparse data and a small number of variables. Furthermore, these
UGBMs can be easily applied in cities (in Tehran, Iran and Las Vegas,
US, for example) where natural conditions (such as terrain and
climate) or urban design cause the urban boundary to adopt a
relatively regular morphology without encountering obvious ob-
stacles (for example, big lakes, high mountains); however, such
favorable simulation conditions don't exist in our study area of
Wuhan. Wuhan has a complex urban morphology with lakes and
rivers throughout the territory, so we cannot apply these UGBMs to
simulate the boundary directly. We need to develop a UGBM that
will fit Wuhan's situation.
1.2. Research questions, significance and paper structure
In this paper, we propose a model known as UBEM to simulate
UGBs. UBEM combines the historical development trajectory of the
UGB and its extension potential in each azimuth to predict the
future UGBs. Wuhan City in Central China is selected as a case study
to verify the utility of UBEM. We apply the model to simulate
Wuhan's UGBs for the future and propose some urban planning
guidelines according to the projected results. The research ques-
tions include: (1) How can we combine the radiation method (RM)
with evaluation of the extension potential in each azimuth to
simulate the extension of the urban boundary? (2) How can we use
temporal data to calibrate the model and reduce randomness? (3)
How can the predication results be used to guide urban planners?
UBEM provides a new way to identify trajectories of urban dynamic
change, and this study is significant because the UBEM product
provides a reference for future urban planning, helping city man-
agers design Wuhan's UGBs to promote coordinated urban devel-
opment and environmental protection. This article is organized as
follows: Section 2 gives a brief introduction to the study area and
describes the model construction and calibration of the UBEM in
detail. Section 3 describes the results, including the model cali-
bration result using the temporal data and the predicted urban
boundary for 2020, and Section 4 presents some discussion be-
tween the current study with existing studies in the literature as
well as guidance for the design of Wuhan's UGBs. The final section
provides the conclusions of this research.
2. Materials and methods
2.1. Study area and data
Wuhan, the capital of Hubei Province, China, lies at latitude 29

580 e31
220 N and longitude 113
410 e115
050 (Fig. 1). It is recog-
nized as the political, economic, cultural, financial, educational, and
 Q. He et al. / Habitat International 72 (2018) 57e65
Fig. 1. Location of the study area.
transportation center of Central China. Historically, Wuhan was also
known as the “the crossroads of the 9 provinces” It covers an area of
8467 km2 and has a population of more than 10 million as of 2015.
The historical data used to predict Wuhan's UGBs is comprised
of several different collections of remotely-sensed imagery, spe-
cifically TM/ETM images collected in 1988, 1995, 2000, 2005, 2011,
and 2015. Supervised classification and manual screen digitizations
are combined to extract the urban edge. First, ERDAS software is
used to classify the land cover type using supervised classification
for each of the images. Land cover types consist of urban land,
farmland, forest, water, and bare land. The pixel value is 1 for urban
land; 0 for farmland, forest, and bare land; and 2 for water. Arc-
GIS10.2 software is used to define the boundary of urban land
polygons using by artificial digitization. The largest, and continuous
urban land polygon is chosen to extract the boundary (Tayyebi,
Pijanowski, & Pekin, 2011; Tayyebi, Pijanowski, & Tayyebi, 2011).
2.2. Method
2.2.1. Extension length of the urban boundary in every azimuth
The radiation method (RM) is used to describe the urban
extension for a given azimuth. The basic idea of RM is that urban
centroids are the origins point that generates individual radial lines
from each azimuth. The radial line for a particular azimuth in each
phase of the UGB intersects at one point, and n (n  1) phases of the
boundary result in n points. Thus, the urban centroids should first
be determined. This paper selects the centers of gravity for each of
the five districts (i.e., Jiangan, Qiaokou, Hanyang, Wuchang, and
Qingshan) under the jurisdiction of Wuhan to be the urban cen-
troids. Each centroid, or origin point, generates one radial line from
the origin to the boundary for every degree in a clockwise direction.
The overlapping parts of the radial lines are then removed and the
result is obtained (Fig. 2). Not all boundaries are convex. Some
concave parts exist at the local level and generate more than one
point of intersection where the radial line meets the boundary. We
set rule that the point nearest to the current origin is the only point
of intersection between the radial line and the boundary. Formula
59
(1) is used to calculate the extension length of the urban boundary
in each azimuth.
h i1 2 =
L ¼ ðPXtþ1  PXt Þ2 þ ðPYtþ1  PYt Þ2 (1)
where PXt þ 1, PYt þ 1 and PXt, PYt are the x-coordinate and y-co-
ordinate of the intersection points for the radial line with the
boundary of phases t þ 1 and t, respectively.
Thus, the annual average extension length can be calculated by
Formula (2).
L
Laverage ¼ (2)
n
Where Laverage is the average extension length, n is the time interval,
and L is defined in Formula (1).
2.2.2. Potential for urban growth
The extension potential of the urban boundary in different di-
rections varies due to differences in the resistance. Large rivers,
steep terrain, and a natural conservation area may prohibit the
boundary's outward expansion. The surrounding non-urban areas
that are close to the city center, urban major roads, schools, and
hospitals are transformed into urban areas. Previous studies on
urban driving forces have indicated that topographic driving factors
mainly include slope and elevation, and socioeconomic driving
factors may include the distance to important sites in the city, the
surrounding environment (i.e., neighborhood effect), and land use
policies (Liu, He, Tan, Liu, & Yin, 2016). As the present study is
limited by data acquisition and the convenience of quantitative
expression, eight indicators are chosen (i.e., elevation, slope, dis-
tance to urban centroids, major roads, railways, commercial cen-
ters, and urban build-up) to evaluate the urban growth potential
(Liu et al., 2016; Müller et al., 2010; Poelmans & Rompaey, 2010;
Sunde et al., 2014; Vermeiren et al., 2012). The “Near” analysis
tool found in ArcGIS10.2 software is used to calculate the distance
60 Q. He et al. / Habitat International 72 (2018) 57e65
Fig. 2. Map of radiations.
variables. Given the fact that terrain factors (i.e., elevation and
slope) and distance belong to different dimensions, all results are
standardized to range between 0 and 1 to remove dimensionality. A
number of sample points are selected: 1) non-urban areas in the
previous period that transformed into urban area in the new period
and 2) non-urban area in the previous period and that continued to
be non-urban in the new period. The first transition is denoted by 1,
and the second is denoted by 0. These two situations are treated as
the dependent variables and the urban driving forces are taken as
the independent variables. SPSS software is used to test the
collinearity of the selected variables and uncover the relationship
between urban change situation (or not) and driving forces using
logistic regression. The collinearity test shows that distance to ur-
ban build-up is collinear with distance-to-city commercial centers
and that elevation is similar to slope. The remaining six variables
are selected as the driving forces of urban growth. The possibility of
transforming from non-urban to urban in Wuhan City can be ob-
tained using Formula (3) and finally a land use change potential
map is produced.
1:0
P¼ (3)
1:0 þ eY
The value of P in Formula (3) ranges from (0, 1), and the closer P
is to 1, the higher the probability that a non-urban pixel will
transition into an urban pixel. After all the pixel transformation
potentials are calculated, it is easy to evaluate the extension po-
tential of the urban boundary for each azimuth using Formula (3).
2.2.3. Extension potential of UBGs in different azimuths
Quantitatively describing the expansion potential of the urban
boundary in different azimuth is difficult. Each radial line can be
imagined to be a real-world pipe. The higher the pressure in the
pipe is, the farther the water flows out from the pipe in terms of
distance. Moreover, certain “pressures” promote the extension of
the UGB. These pressures are determined by different driving forces
of urban growth. Based on the land use change potential map
produced in the previous section, the following steps are used to
calculate the “pressure” (extension potential) of UGB in a given
direction.
(1) Taking a radial line as the central axis, a certain distance
expands on both sides and forms a rectangular template as
shown in Fig. 3.
In Fig. 3, the red line denotes the central axis, that is, the radial
line is emitted from the urban center in a different direction. The
zone within the green border is the rectangular template. Each cell
in the template represents one pixel. The variable r is the expansion
distance on both sides of the axis and l is the length from the urban
center to the intersection point of the radiation and boundary.
(2) Based on the land use change potential map, all the pixel
values in the template are calculated using C# programming
language and integrated using the secondary development
interface provided by ArcGIS. The average pixel value, which
is calculated using Formula (4), represents the extension
potential of UBGs in a given direction.
P
pi
PE ¼ (4)
N
P
where PE is the potential value, pi is the sum of the pixel value in
the rectangle, and N is the number of pixels.
2.2.4. UGB predication model
With the historical extension length of the urban boundary and
the evaluation of extension potential in a given direction, the UBEM
Fig. 3. Map of the rectangular template.
 Q. He et al. / Habitat International 72 (2018) 57e65
is created to predict the extension length of the simulated urban
boundary forheg each azimuth.
ELTnþ1 ¼ v*ðTnþ1  T0 Þ*LaverageðT *PðT0 Tn Þ
0 Tn Þ
Where ELTnþ1 is the incremental length of the simulated urban
boundary at time Tn þ 1, compared to the urban boundary in T0,
LaverageðT Tn Þ ; and PðT0 Tn Þ represents the historical annual extension
0
length of urban boundary and the extension potential in one azi-
muth during the period of time T0 to Tn, where T0 is the start year,
and Tn is the year n. In this model, ELTnþ1 and LaverageðT Tn Þ are length,
0
PðT0 Tn Þ is dimensionless potential value, changing from (0, 1), and
(Tn þ 1  T0) is the time interval, another dimensionless value. We
define a model coefficient v for the right side of the equation to
achieve equality on both sides, meaning that this equation may be
tenable if we determine an appropriate value of v.
LaverageðT Tn Þ and PðT0 Tn Þ are easily obtained from the historical ur-
0
ban boundaries, so the key to calculating ELTnþ1 is to determine the
correct value of v. In fact, UBEM needs at least three time steps to
calculate the value of v, and more data time steps will help further
calibrate the value of v and reduce randomness in the future
simulation. The steps for using multi-temporal data to calibrate v
will be presented in Section 2.2.5. At present, we will describe how
to use three time steps to obtain the value of v and simulate the
future urban boundary. For example, if we have the urban bound-
aries for 2005, 2010, and 2015, and we want to predict the urban
boundary for 2020, then the procedure is as follows:
(1) Calculate the value of v for one azimuth
a) The total extension length, the annual average extension
length and the extension potential of the boundary in one
azimuth between periods 2005 and 2011, and 2005 and 2015,
are calculated. The values are represented as EL2005~2011 and
EL2005~2015, Laverageð20052011Þ and Laverageð20052015Þ , P(2005~2011) and
P(2005~2015) respectively.
b) Based on Formula (5), we can deduce that:
EL20052015
v¼
ð2015  2005Þ*Pð20052011Þ Laverageð20052011Þ
EL20052015
¼
10*Pð20052011Þ Laverageð20052011Þ
(2) Simulate the future urban boundary based on v
In Step (1), we get the value of v based on the historical exten-
sion of the urban boundary for one azimuth, and now we can
predict the future extension of the urban boundary for this azimuth
using the equation below:
EL2020 ¼ v*15*Laverageð20052015Þ * Pð20052015Þ
Where v is calculated using the three historical phrase boundaries,
i.e., 2005, 2011 and 2015.
Steps (1)e(2) are repeated for each azimuth, which is described
in Step 2. The prediction points in each direction are connected to
the line and form the final urban boundary.
2.2.5. Calibration of the model coefficient using temporal data
The v coefficient in the UBEM directly affects the precision of the
modeling result. In our introduction of the model details (Section
2.2.4), we shows that the value of v is related to the historical data
that is selected. For example, a value for v can be obtained using
61
data from 2000, 2005 and 2011 to predict 2020, but we can get a
different value for v using the data from 2005, 2011 and 2015 to
predict 2020, which implies a high degree of randomness. If we can
(5) get a relatively stable value of v and make full use of the multi-
temporal historical data, then the prediction result will more reli-
able. Because there is a link between v and the annual extension
length of the urban boundary in one azimuth, a binary linear
regression model is used to detect the relationship between the
different values of v and the average extension length of the UGB in
one azimuth, which is obtained using the historical urban bound-
aries. This is ideal and is a relatively simple way to reduce
randomness. The details of this procedure are as follows.
(1) Equation v ¼ a  Laverage þ b represents the relationship
between the annual extension length L and the v coefficient.
The key is to calculate the coefficients a and b.
(2) The permutation and combination of the n time series data
with three phase data forms one group and C3n unique
combinations are generated. For example, the six time series
data of 1988, 1995, 2000, 2005, 2011, and 2015 used in this
paper can generate C36 combinations (i.e., 1988, 1995, 2000;
1988, 1995, 2005; 1988, 1995, 2011; 1988, 1995, 2015; … and
2005, 2011, 2015). Each combination will result in a value of v
and Laverage to predict the future urban boundary (for
example, the 2020 urban boundary). Then the C# program-
ming language is used to fit the relationship of the two var-
iables (if R2 is less than 0.5, then directly use the average of v
to replace the calibration result). The fitted result of this
procedure for one azimuth is shown in Fig. 4. The vertical and
horizontal axes in the graph are for v and L, respectively.
Thus, the values of a and b are 0.0055 and 1.3093,
respectively.
(3) The calibrated v is used to predict the future urban boundary.
(4) Steps (1)e(3) are repeated for each azimuth.
3. Results
3.1. Evaluation of the extension potential of the urban boundary for
each azimuth
Logistic regression modeling is used to quantify the relationship
between whether an area is urban or not, and the selected driving
forces for urban expansion. The results are shown in Table 1.
Following Formula (3),
Y ¼ 0:908  1:276*Disroad  1:042*Diswater þ 2:333*Disccenter
 0:419*Slope  0:123*Disrailway þ 1:883*Discmcenter
Where Disroad, Diswater, Disccenter, Disrailway, Discmcenter, and Slope are
distance to major roads, water, urban centroids, railways, com-
mercial centers, and slope, respectively.
Fig. 4. Chart of the fitting result.
62 Q. He et al. / Habitat International 72 (2018) 57e65

Table 1 The red line in the figure represents the simulated result of the 2015
Logistic regression model results for the urban land use change. boundary using the uncalibrated v value, the blue line represents
Variables B Standard error t Sig Tolerance the calibrated simulation result, and the black line is the actual
Constant 0.908 0.028 19.124 0
urban boundary in 2015. The red line exhibits a good deal of ir-
Disroad 1.276 0.103 13.122 0 0.534 regularity, many sections of it appearing in a zigzag pattern due to
Diswater 1.042 0.094 10.448 0 0.609 the uncalibrated v value; the blue line, on the other hand, is much
Disccenter 2.333 0.045 22.656 0 0.592 more smooth and similar to the actual boundary thanks to the
Slope 0.419 0.107 8.043 0 0.837
calibration of v. In addition to this, PAM quantity and PAM location
Disrailway 0.123 0.019 2.334 0 0.697
Discmcenter 1.883 0.036 14.743 0 0.605 are 0.9736 and 0.8210, respectively, when v is calibrated, while the
corresponding values are 0.9587 and 0.8016 if v is not calibrated.
2 Log likelihood: 1891.19; Cox and Snell R2: 0.557; Nagelkerke R2: 0.7987.
PCPa: 86.7.
These values suggest that the calibration of v using multi-temporal
a
PCP with cut value 0.5. data could much more reliably simulate the actual urban boundary.

3.3. Model comparison and future urban boundary prediction

3.2. Calibration of v
We use the actual and simulated urban boundaries for 2015 to
validate UBEM before and after calibration. The simulated urban
boundary was calculated in two ways: the first used a value of v that
was derived from three temporal data collections (1988, 2011, 2015)
and the second used a value of v that was derived from five tem-
poral data collections (1988, 1995, 2000, 2005, 2011). Thus, the
incremental length of the simulated urban boundary in 2015 when
compared to the urban boundary from 1988 can be written as:
EL2015 BC ¼ vBC *ð2015  1988Þ*Laverageð19882011Þ *Pð19882011Þ
(6)
EL2015 AC ¼ vAC *ð2015  1988Þ*Laverageð19882011Þ *Pð19882011Þ
(7)
Where EL2015_AC represents the incremental length of the simulated
urban boundary in 2015 compared to the urban boundary of 1988
using the calibrated value of v (i.e., vAC); EL2015_BC is the incremental
length of the simulated urban boundary in 2015 compared to the
urban boundary of 1988 using the uncalibrated value of v (i.e., vBC).
In order to verify the reliability of the modeling result, we use
the percent area match (PAM) quantity and PAM location for ac-
curacy evaluation (Tayyebi, Pijanowski, & Pekin, 2011). The PAM
quantity indicates how close the area enclosed by the simulated
urban boundary is to the actual area, while PAM location represents
a relative match to the boundary without considering the over-
estimate. The closer PAM quantity and PAM location are to 1, the
better the performance of the simulation (Tayyebi, Pijanowski, &
Pekin, 2011; Tayyebi, Pijanowski, & Tayyebi, 2011). Fig. 5 shows
the comparison of the model results before and after calibration.
In Section 3.2, we concluded that the modeling result obtained
using a calibrated v has better performance than one without
calibration; however, this is comparing results within the UBEM, In
order to demonstrate the reliability of the UBEM, we also compare
the result of our UBEM with a null UGBM. The null UGBM simulates
new urban boundaries using different buffer distances around the
base map and finds a suitable buffer size to make the PAM quantity
value as close to 1.0 as possible (Tayyebi, Pijanowski, & Pekin, 2011).
After experimentation, we set the buffer distance to 550 m, at
which distance the PAM quantity is closest to 1.0. We compare our
UBEM results to that of the null UGBM and evaluate their perfor-
mance for urban boundary modeling (Table 2). We find that both of
the models demonstrate good performance based on PAM quantity,
but in terms of the position accuracy (PAM location), our UBEM is
clearly superior to the null UGBM. In urban planning, city managers
not only need to know how much the urban area will increase, but
also where the new urban area may appear. Our UBEM's high
simulation precision in both PAM quantity and location undoubt-
edly provides a meaningful reference for urban planners.
Using the urban boundary from 1988 to 2015, and the model
coefficient v calibrated using six temporal urban boundary phases
(i.e., 1988, 1995, 2000, 2005, 2011, 2015), we predict the future
urban boundary of Wuhan in 2020 based on UBEM (Fig. 6).
Our analysis shows that the urban area in Wuhan covered
240.32, 295.81, 395.22, 557.68, 718.55, and 860.73 km2 in 1988,
1995, 2000, 2005, 2011, and 2015, respectively, and is expected to
increase to 964.87 km2 in 2020, which indicates a rapid growth
trajectory. The gross domestic product increased significantly, from
156.44 billion yuan (about 2.4 billion dollars) in 1988 to 1006.948
billion yuan (about 150 billion dollars) in 2015. These data imply
that the rapid economic growth over a 30-year period influenced
urban spatial expansion in Wuhan.

4. Discussion

4.1. Comparison with existing studies

In this paper we propose an urban growth boundary model,

Fig. 5. Results comparing the model before and after calibration.
known as UBEM, to simulate the urban boundary of Wuhan. UBEM
is based on an evaluation of the extension potential for each azi-
muth from the city center to simulate incremental growth when
compared to the base map. We apply UBEM in Wuhan to predict
the future urban boundary in 2020. Other studies have applied
UGBM outside China; for example, Tayyebi, Pijanowski, and Pekin
(2011), Tayyebi, Pijanowski, and Tayyebi (2011, 2014) applied
UGBMs to simulate the urban boundaries in Tehran, Iran and Las
Vegas, US, with good results. When compared with these studies,
however, ours has the following unique characteristics that make
UBEM more suitable in Wuhan: (1) Contrary to the relatively dry
climate found in Tehran and Las Vegas, Wuhan has rich water
 Q. He et al. / Habitat International 72 (2018) 57e65
Table 2
Comparison of UBEM and null UGBM in simulation accuracy.
Model Predicted change in area(km2) Actual change in area (km2)
2011e2015 2011e2015
UBEM 138.43 142.18
Null UGBM 137.60 142.18
Fig. 6. Simulated boundary.
resources. Wuhan is known as the “city of hundreds of lakes”, while
the world's third longest river, the Yangtze, flows through the city
and divides Wuhan into three parts (see Fig. 1). The large lakes and
rivers in this territory have hindered the natural expansion of
Wuhan, which makes the shape of the city more complex than
Tehran or Las Vegas. Such a complicated and irregularity form lead
to sharp differences in the incremental lengths for different azi-
muths between two time periods. Taking 2005e2015 as an
example, the incremental lengths between the two boundaries for
each azimuth ranged from 8.56 m to 5612.66 m, so a uniform rule
(like the uniform buffer distance in the null UGBM) may not be
suitable for Wuhan. UBEM simulates the urban boundary for each
azimuth separately, thus making the results more reliable. (2) Other
similar models can only use two temporal periods to simulate the
urban boundary, while for UEBM, the number of periods is flexible
and all available temporal data is involved in the calibration of the
model, making the simulation results smoother and with higher
accuracy (see Section 3.2).
4.2. Guidance for the design of Wuhan UGBs
This paper proposes a UBEM to simulate the expansion of an
UGB. The method simulates natural expansion without the
consideration of policy constraints. This paper also provides useful
references for designing future UGBs. Uncontrolled urban sprawl
has brought serious environmental problems, with it, as low-
density urban sprawl has replaced the natural landscape and
agricultural resources. The old city center has seen declines due to
suburbanization, low efficiency municipal facilities, traffic conges-
tion, and the lack of a community atmosphere in suburban areas.
The rapid expansion of the built-up areas in Wuhan has also
resulted in the rapid decline of the number of lakes, from 127 in the
early years of “New China” to only 40 at present; that is, nearly 90
lakes disappeared in the last 60 years. Indeed, the history of
Wuhan's urban sprawl is characterized by “lake-filling”. Therefore,
forming a continuous closed boundary around the city is important
to regulate urban growth and limit the chaotic urban sprawl to
protect the remaining natural open space. Based on the simulated
urban boundary results obtained from our UBEM, the following
recommendations are suggested for the design of Wuhan UGBs.
63
Over-estimated by the simulated and actual area (km2) PAM quantity PAM location
21.17 0.9736 0.8210
51.91 0.9678 0.6027
(1) This study measures the expansion rate of Wuhan in four
directions: north, south, east and west. The expansion rate in
the north is relatively rapid from 2015 to 2020; the north
ranks slower than the west. The northern part contains
Mulan Tianchi National Forest Park. Government planners
should maintain the integrity of the ecosystem in Mulan
Tianchi National Forest Park when designing future UGBs.
The government should adopt a basic strategy of protecting
the ecological environment and its natural resources with
moderate tourism development. The broad and effective
protection of the environment and resources can be achieved
through moderate development. Development and urban
uses that are in conflict with conservation objectives should
be avoided. Stable ecosystems can continue to provide the
public with tourism, scientific research, education, and rec-
reational opportunities, and this provision is an effective and
protective management pattern to responsibly cope with the
tensions between environmental protection and resource
exploitation or utilization
(2) The Wuhan municipal government approved the “Wuhan
green line management approach” on April 12, 2010. The
“green line” can be interpreted as a “green belt” that lies
along the outer highway ring to protect important regional
natural resources and landscapes, as well as to achieve the
orderly urban growth. However, different conservation goals
imply different boundaries for protected areas, such as the
boundary of the forest protected area, the “red line” for basic
agricultural protection, the “blue line” for water resources
protection, the “green line” for greenbelt protection, and the
“purple line” for cultural heritage protection. UGBs should be
considered as an integration of all of these boundary lines,
and in this case the “green line” is only one of those lines.
Therefore, the currently proposed “green line” is an example
of the difficulty in effectively protecting various non-built-up
areas, which are not be designated within urban develop-
ment boundaries. Thus, preventing the occupation of pe-
ripheral ecologically sensitive areas and natural landscapes is
difficult. Fig. 7 shows that the urban border has been largely
extended to basic agricultural protection areas and some lake
areas through uncontrolled growth except for the designa-
tion of urban green space.
(3) Different types of development areas should be established:
ecological baseline areas and areas for ecologically respon-
sible developmental. Ecological baseline areas are top line of
last protection for urban ecological security. They adhere to
the most stringent ecological protection requirements.
Ecological developmental areas allows for low-density, low-
intensity construction when certain conditions are met.
Thus, these two factors should be coordinated when delin-
eating UGBs. The first factor is designed to protect the natural
environment, using geographical information systems and
other relevant technology to analyze and evaluate the
ecological environment capacity and ecological of the urban
area, delineating “rigid” boundaries for urban growth, and
making a clear distinction between urban lands designated
for construction and those that are not. The second factor
64 Q. He et al. / Habitat International 72 (2018) 57e65
Fig. 7. Land occupied by urban expansion.
helps to forecast the scale of city development. The “dynamic
development” line of thinking implies that UGBs are limited
by the concept of “growth” to the point that per capita urban
construction land gradually fails. In the planning process, we
should test different UGB limits and then create city spatial
layouts at different stages of development to guarantee the
sustainability and rationality of urban spatial structure and
morphology.
5. Conclusion
The overall effects of urban development are mainly evidenced
by the movement of the urban boundary, especially outward
extension. Modeling the urban boundary helps to understand the
trajectory of urban development and serves as a reference for urban
planners. UBEM, which combines the historical development of
UGBs and their extension potential in one azimuth, can forecast
future UGBs. Three important conclusions can be drawn:
(1) UBEM has high simulation accuracy in urban boundary
modeling with respect to total length, cover area, and degree
of fit. It achieves satisfactory results when compared with the
actual boundary.
(2) Based on the simulated boundary of Wuhan in 2020,
Wuhan's rapid urbanization will continue in the future
without control. In 2015, the urban area increased to
860.73 km2, and the projected area for 2020 increased to
964.87 km2 with a growth rate of 19.79%. Resolving the
contradiction between land supply restrictions and the de-
mands for urban development is a challenge that is faced by
the government.
(3) The UBEM results also show that natural urban expansion
will occupy some ecological lands, such as lakes and forests,
which have the important role of maintaining the security of
the ecosystem. The government should address this phe-
nomenon and reasonably design UGBs to spatially guide
urban development.
Acknowledgements
This paper was supported by the Special Fund of Ministry of
Land and Resources of China in the Public Interest (ID. 201511001).
References
Ball, M., Cigdem, M., Taylor, E., & Wood, G. (2014). Urban growth boundaries and
their impact on land prices. Environment & Planning A, 46(12), 3010e3026.
Benfield, F. K., Terris, J., & Vorsanger, N. (2001). Solving sprawl: Models of smart
growth in communities across America. Natural Resources Defense Council.
Bengston, D. N., & Youn, Y. C. (2006). Urban containment policies and the protection
of natural areas: The case of Seoul's greenbelt. Ecology & Society, 11(1), 1.
Calthorpe, P. G., & Fulton, W. B. (2001). The regional city: Planning for the end of
sprawl. Books.
Clarke, K. C., & Gaydos, L. J. (1998). Loose-coupling a cellular automaton model and
GIS: Long-term urban growth prediction for San Francisco and Washington/
Baltimore. International Journal of Geographical Information Science, 12(7),
699e714.
Clarke, K. C., Hoppen, S., & Gaydos, L. J. (1997). A self-modifying cellular automaton
model of historical urbanization in the San Francisco bay area. Environment &
Planning B Planning & Design, 24(2), 247e261.
Eddo, Coiacetto (2007). Residential sub-market targeting by developers in Brisbane.
Urban Policy & Research, 25(2), 257e274.
David, Gordon, & Shayne, Vipond (2005). Gross density and new urbanism:
Comparing conventional and new urbanist suburbs in Markham, Ontario.
Journal of the American Planning Association, 71(1), 41e54.
Gunn, S. C. (2007). Green belts: A review of the regions' responses to a changing
housing agenda. Journal of Environmental Planning & Management, 50(5),
595e616.
Habibi, S., & Asadi, N. (2011). Causes, results and methods of controlling urban
sprawl. Procedia Engineering, 21(3), 133e141.
Han, H. Y., Lai, S. K., Dang, A. R., Tan, Z. B., & Ci-Fang, W. U. (2009). Effectiveness of
urban construction boundaries in Beijing: An assessment. Journal of Zhejiang
University Science A, 10(9), 1285e1295.
Liu, Y., He, Q., Tan, R., Liu, Y., & Yin, C. (2016). Modeling different urban growth
patterns based on the evolution of urban form: A case study from Huangpi,
central China. Applied Geography, 66, 109e118.
Long, Y., Han, H., Lai, S. K., & Mao, Q. (2013). Urban growth boundaries of the Beijing
metropolitan area: Comparison of simulation and artwork. Cities, 31(2),
337e348.
Long, Y., Han, H., & Mao, Q. (2009). Establishing urban growth boundaries using
constrained ca. Acta Geographica Sinica, 64(8), 999e1008 (in Chinese).
Long, H., Liu, Y., Hou, X., Li, T., & Li, Y. (2014). Effects of land use transitions due to
rapid urbanization on ecosystem services: Implications for urban planning in
the new developing area of China. Habitat International, 44, 536e544.
Martin, J., Pendall, R., & Fulton, W. (2002). Holding the line: Urban containment in the
United States. Brookings Institution.
 Q. He et al. / Habitat International 72 (2018) 57e65
Mubarak, F. A. (2004). Urban growth boundary policy and residential suburbani-
zation: Riyadh, Saudi Arabia. Habitat International, 28(4), 567e591.
Müller, K., Steinmeier, C., & Küchler, M. (2010). Urban growth along motorways in
Switzerland. Landscape & Urban Planning, 98(1), 3e12.
Poelmans, L., & Rompaey, A. V. (2010). Complexity and performance of urban
expansion models. Computers Environment & Urban Systems, 34(1), 17e27.
Puertas, O. L., Henríquez, C., & Meza, F. J. (2014). Assessing spatial dynamics of urban
growth using an integrated land use model. Application in Santiago metro-
politan area, 2010e2045. Land Use Policy, 38(2), 415e425.
Rienow, A., & Goetzke, R. (2014). Supporting sleutheenhancing a cellular autom-
aton with support vector machines for urban growth modeling. Computers
Environment & Urban Systems, 49, 66e81.
Sinclair-Smith, K. (2014). Methods and considerations for determining urban
growth boundariesdan evaluation of the Cape Town experience. Urban Forum,
25(3), 313e333.
Su, S., Jiang, Z., Zhang, Q., & Zhang, Y. (2011). Transformation of agricultural land-
scapes under rapid urbanization: A treat to sustainability in Hang-Jia-Hu region,
China. Applied Geography, 31(2), 439e449.
Sunde, M. G., He, H. S., Zhou, B., Hubbart, J. A., & Spicci, A. (2014). Imperviousness
change analysis tool (i-cat) for simulating pixel-level urban growth. Landscape
& Urban Planning, 124(2), 104e108.
Tayyebi, A., Delavar, M. R., Saeedi, S., & Amini, J. (June 2008). Monitoring the urban
expansion by multi-temporal GIS Maps. Stockholm, Sweden: Integrating Gener-
ations Fig Working Week.
65
Tayyebi, A., Perry, P. C., & Tayyebi, A. H. (2014). Predicting the expansion of an urban
boundary using spatial logistic regression and hybrid rasterevector routines
with remote sensing and GIS. International Journal of Geographical Information
Science, 28(4), 639e659.
Tayyebi, A., Pijanowski, B. C., & Pekin, B. (2011a). Two rule-based urban growth
boundary models applied to the Tehran metropolitan area, Iran. Applied Geog-
raphy, 31(3), 908e918.
Tayyebi, A., Pijanowski, B. C., & Tayyebi, A. H. (2011b). An urban growth boundary
model using neural networks, GIS and radial parameterization: An application
to Tehran, Iran. Landscape & Urban Planning, 100(s1e2), 35e44.
Vermeiren, K., Rompaey, A. V., Loopmans, M., Serwajja, E., & Mukwaya, P. (2012).
Urban growth of Kampala, Uganda: Pattern analysis and scenario development.
Landscape & Urban Planning, 106(2), 199e206.
Wang, L. G., Han, H., & Lai, S. K. (2014). Do plans contain urban sprawl? A com-
parison of Beijing and Taipei. Habitat International, 42(2), 121e130.
Wan, L., Ye, X., Lee, J., Lu, X., Zheng, L., & Wu, K. (2014). Effects of urbanization on
ecosystem service values in a mineral resource-based city. Habitat International,
46, 54e63.
Yang, Q. S., & Xia, L. I. (2006). Cellular automata for simulating land use changes
based on support vector machine. Computers & Geosciences, 34(6), 592e602.
Yeh, G. O., & Li, X. (2002). A cellular automata model to simulate development
density for urban planning. Environment & Planning B Planning & Design, 29(3),
431e450.