Urban Rail Transit (2021) 7(3):240–255

https://doi.org/10.1007/s40864-021-00153-8 http://www.urt.cn/

ORIGINAL RESEARCH PAPERS

TOD Typology Based on Urban Renewal: A Classification

of Metro Stations for Ningbo City
Liu Yang1 • Xiaoyu Song1

Received: 7 April 2021 / Revised: 9 June 2021 / Accepted: 18 June 2021 / Published online: 12 August 2021
 The Author(s) 2021

Abstract In recent decades, the transit-oriented develop- willingness of residents on the basis of the traditional node-
ment (TOD) concept has been widely used all over the place model as one of the judgment indicators for evalu-
world, especially in China, for the massive construction of ating whether existing stations and surrounding areas are
urban public transportation systems with rail transit as the suitable for TOD improvement. At the same time, tradi-
backbone. However, it is not easy to make significant tional statistical analysis is combined with GIS and
changes in a city while building a transportation system, machine learning technology. Using this method, we pro-
and the transit-guided urban development expected by the pose the TOD improvement-oriented station area classifi-
TOD concept has not been completely realized. The cation and identification method based on TOD typology
transformation of nearby areas with the guidance of transit theory. The results show that Ningbo’s subway stations can
is also becoming the choice of many Chinese cities, be divided into four categories according to the suitability
especially for cities that have only had subways for a few for TOD improvement, and we selected seven stations
years. Unlike other international metropolitan cities, with suitable for TOD improvement according to the charac-
metro systems of considerable scale, the modernization teristics of the node-place model. The practice in Ningbo
process of most of the small and medium-sized cities in has proved that this method is effective for identifying sites
China is being carried out simultaneously with metro-based suitable for TOD improvement, especially for cities that
public transportation guidance. For cities which are still in have recently built subways.
their initial stage of the backbone public transportation
system, there is not enough previous experience and evi- Keywords Transit-oriented development  Node-place
dence to support the suitability of TOD typological anal- model  Machine learning  TOD Improvement  Metro 
ysis based on the node-place model. More research based Ningbo
on the node-place model has also shown practical appli-
cations of the TOD in developed cities. However, there are
very few studies that analyse cities in which rail transit and 1 Introduction
urban development are both in a period of rapid growth.
The goal of this research is to identify which metro stations Previously, activities that guided urban renewal and
in these cities are suitable for TOD improvement and improvement, such as transit-oriented development (TOD),
optimization. This article attempts to expand the were referred to as joint development [1–3]. Although the
activities of transforming cities in this way have a long
& Liu Yang
history [4–7], the concept of TOD was formally proposed
[email protected] by Calthorpe and implied mixed development and a larger
population living near transit lines [8]. TOD is widely
1
China Metro Engineering Consulting Corporation, Building considered to be an efficient approach for integrating
9#, No.31 Xishiku Dajie, Xicheng Dist., Beijing 100034,
China
transportation and land use.
China is a country with high population density. Urban
Communicated by Baoming Han. extension and the increase in the number of private

123
Urban Rail Transit (2021) 7(3):240–255 241

motorized vehicles have caused traffic congestion, social categories of the calculation results, and the recommended
and spatial integration, and negative environmental, social stations for improvement. Chapter 6 analyses the situation
and economic effects [9]. TOD, as a method for urban of the Ningbo metro station based on the model. Finally,
planning, has been practised in the populous Asian regions Chapter 7 discusses the limitations of the model and the
[10–15]. Unlike the urbanized cities in Japan and South space to be optimized, and expansions for future research
Korea, the rapid development of metro systems in main- direction.
land China rarely matches the urban land use [16]. Fur-
thermore, the development of stock land will become the
mainstream of urban construction in China [17]. With the 2 Literature Review
TOD theoretical input and practical experience gained
from other Asian cities, the decision-makers have realized The quantitative study of TOD began with Bertolini’s
the importance of integrating land development with the research [24] on the development prospect of railway sta-
transit system, including guiding the development of new tions and their surrounding areas. Based on the relationship
towns and constructing commerce around stations [18]. For between traffic accessibility and land use, the node-place
instance, Ningbo, as a representative city, has formed the model was established and successfully applied to the
idea that ‘building metro is building a city’ to solve the analysis of railway stations in Amsterdam, Utrecht, and
problem of the disconnect between urban development and Arnhem Nijmegen in the Netherlands [22, 24, 25].
transportation [19]. This conceptual model has been quantified and expan-
More research based on the node-place model has also ded by many researchers to analyse TOD development
shown practical applications of TOD in developed cities status assessment [26, 27], traffic behaviour choices
[14, 20, 21]. However, there are very few studies that [28, 29], land use [30, 31] and many other specific appli-
analyse cities in which rail transit and urban development cations. These studies are often attributed to the TOD
are both in a period of rapid growth. The interactive rela- typology. Although the indicators selected by the model are
tionship between the establishment of the TOD develop- not the same, the TOD analysis framework based on the
ment model and the implementation of rail transit is often node-place model has been widely used. The TOD
the focus of such cities. This is because these cities have typologies enhance related planning and design. For
enough opportunities and time to integrate the relationship instance, the similarities type allows the policymakers and
between urban development and backbone public trans- urban planners to develop more targeted sets of strategies
portation systems and the land-use patterns guided by to promote TOD [32].
them. Higgins and Kanaroglou [33] asserted that there are two
This paper attempts to expand the node-place model related approaches to conceptualizing and estimating
[22] to assist the urban renewal around the stations. We transit station typologies. The first is to classify according
take the willingness of the residents as an important indi- to a certain attribute of TOD, e.g., location of the stations,
cator of pre-planning assessment [20, 23]. In the research, level of commercial activities, and the function of the
besides using traditional statistical analysis methods, we transportation (similar to that mentioned by Huang et al.
combine them with GIS and machine learning technology. [34]). Calthorpe presupposed that there are three types of
The advantage of our developed method enables the data TOD in China: commercial centres, urban centres, and
model to be less affected by human factors. The proposed town centres [35]. However, it has been discovered through
model has high repeatability and portability while main- practice that the attributes of TOD areas are diverse, and
taining the same simple data structure as the traditional the Calthorpe approach [35] is unable to describe the
model. multiple attributes of TOD. Dittmar and Poticha [36] also
The structure of this paper is as follows. Chapter 2 recognized that there can be no ‘one-size-fits-all’ approach
focuses on reviewing the previous node-place model lit- to the TOD. The second approach is justified based on the
erature and our thoughts on the scalability of the node- TOD characteristics, which is more positive. Reusser et al.
place model. Chapter 3 describes the node-place model [37] expanded the node-place model and classified over
extension method based on TOD improvement, focusing 1600 railway stations in Switzerland into five types char-
on the main concepts of the method and the construction acterized by 11 indicators. Chorus and Bertolini [14] used
method of the index system. Chapter 4 describes the the node-place model to study the traffic types and land-use
overall calculation process of the Ningbo case, which factors that affect the spatial development around Tokyo’s
includes the classification method of TOD improvement 99 station areas. Kamruzzaman et al. [32] developed a
based on entropy TOPSIS, machine learning, and other typology by clustering the stations into TOD-based groups.
technologies to find the appropriate stations for TOD The analysis in Brisbane showed that the stations might be
improvement. Chapter 5 analyses the characteristics and classified regardless of their balance. Lyu et al. [26]

123
242 Urban Rail Transit (2021) 7(3):240–255

expanded node-place with the ‘oriented’ dimension to In a specific city, however, there may not be necessarily
quantify the orientation of transit and development com- all five types of stations. Moreover, this conceptual model
ponents towards each other. The authors determined the does not clarify the function expression of each curve and it
accessibility and walking potential by measuring the spatial is only a conceptual model that represents the relative
specifications and walkway density parameters. Because of position of the station. In the related research, there are a
the particular research objects and the requirements of variety of quantitative methods. Some of these methods use
practical projects, the current study adopts the second standard values [26, 39], and some perform z-transforma-
typology. tion [14, 37]. In addition, the indicators used in various
studies are different according to the specific local condi-
tions, including the selection of indicators and the range of
3 Methodology specific indicators (such as the difference in the suit-
able walking ranges). The common feature, however, is
3.1 Theoretical Background: The Node-Place Model that they all fall into two attributes (node, place) or three
attributes (T-attribute/node, D-attribute/place and
The node-place model proposed by Bertolini [22] is an O-attribute).
important tool for TOD analysis for public transportation
stations (see Fig. 1). The basic idea of the model is to treat 3.2 Node-Place Model Based on TOD Improvement
the stations as a combination of nodes in the transportation
network and places of urban functions [38]. Analysis of the According to the basic node-place model theory, the four
influencing factors attributed to the transport functions categories including stress (S), unsustained node (U-N),
(node characteristics) and urban functions (place charac- unsustained place (U-P), and dependency (D) are not ideal
teristics) shows that the coordinate positions of the two station types. Therefore, when selecting the TOD
attributes reflect the role and status of the station and its improvement and optimization object stations, it is con-
surrounding areas throughout the city. sidered to select the stations close to the critical state of the
According to the coordinate position on the map, all four sides of the accessibility (Ac) area, achieving the
theoretically existing stations are divided into the five accessibility (Ac) state as the improvement goal. Note that
following categories: the specific implementation method also depends on the
actual situation of the station. Although some stations have
1. Stress (S area)
certain TOD development bases, such as population and
2. Unsustained node (U-N area)
job density, they have not reached the ideal TOD devel-
3. Accessibility (Ac area)
opment level and status. In some cases, areas with dense
4. Unsustained place (U-P area)
residential and commercial facilities have not become the
5. Dependency (D area)
regional centre, while in others the commercial core area of
the old city has gradually lost its vitality with a geo-
graphical centre position (see Fig. 2). Compared with
Beijing, Shanghai, Hong Kong and other international
cities with more developed and stable rail transits, some
ordinary Chinese cities such as Ningbo, with most of the
stations having not been developed at all (numerous
dependency-type stations tending to the lower-left corner),
are selected as the objects of improvement. In such cases,
although the station itself may be greatly improved, such
stations that have just had metro services are not yet in
urbanization or on the edge of urbanization. Therefore,
there is no basis for the TOD improvement in the short
term. So it is often better to select some stations near the
four arcs near the edge of area A to upgrade these stations
mainly in area D to area A. This will make it easier for the
transformation to take effect, and the cost is lower.

Fig. 1 Node-place typology model [22]

123
Urban Rail Transit (2021) 7(3):240–255 243

3.4 Principles of TOD Improvement Indicator

Selection

There have been many studies discussing the selection of

TOD indicators based on the node-place model
[14, 26, 27, 29, 32, 39, 44]. Based on the above results of
systematic research, there are six common principles as
follows.
• Follow the TOD principle; the most famous is the 3D
(density, diversity, and design) principle proposed by
Cervero [45]. Cervero et al. then extended the theory to
5D [46] (with the extra dimensions including Destina-
tion accessibility and Distance to transit), and later 7D
[47] (with the extra dimensions including demand
management and demographics). In any case, the 3D
principle as the core foundation of TOD is the general
foundation.
• Reflect the characteristics of their respective cities.
Fig. 2 Node-place model based on the TOD improvement (four
Chinese cities are undergoing rapid development and
centripetal arrows show the ideal direction of improvement)
changes in recent years [48], especially the rapid
3.3 TOD Improvement Site Identification Method growth of the backbone public transportation systems,
for Entropy TOPSIS Based on the Node-Place e.g., rail transit and bus rapid transit (BRT). Therefore,
Model TOD in China is related more to public transportation
and urban planning [16, 49, 50].
Measuring station areas worthy of TOD improvement is • Focus on the middle and micro features around the
similar to TOD development level evaluation [32, 37, 40] station. In collating the related studies, it can be found
and station classification [26, 27, 29, 39]. It is necessary to that the social constituent factors and economic service
carry out a quantitative analysis for the influencing factors characteristics are the fundamental aspects that reflect
of multiple dimensions and further calculate according to the development level around the stations.
the method of multi-criteria decision analysis (MCDA) • As a task with TOD improvement and its value
[18, 41, 42]. Considering that the human factors often orientation, it is necessary to survey the public regard-
interfere with the objectivity of the calculation results, we ing their preferences, although such surveys are often
use the entropy TOPSIS (E-TOPSIS) [43] for the tradi- affected by the actual social conditions in various
tional MCDA method. The scores that each station receives regions [20, 23].
by E-TOPSIS analysis reflect the current TOD develop- • The index system has universal applicability. The broad
ment level and the basic value of TOD improvement attributes of the index system should allow direct
typology. Therefore, besides the description of general comparison with those of other cities and related
transport, density, urban design, etc., the indicator system research. The weighting of the indicators should
should also include metro passengers’ attitudes (i.e., their therefore minimize subjective judgments.
revealed preference) and the wishes of potential metro • The calculated result of the index reflects the direction
passengers (i.e., their stated preference). to be transformed for the reference of the implementing
At the same time, these factors should be associated entity. Further, the specific indicators can be appropri-
with one of the T/O/D attributes, so that the entire index ately modified according to the needs of the project, if
system can also obtain the calculation results of each sta- the value orientation of the project is reflected.
tion in the T/O/D three-dimensional coordinate system
based on the node-place model. Therefore, the node-place 3.5 Identification of the TOD Indicators
model combined with the E-TOPSIS calculation results can
be used to screen out the stations that have a certain TOD Based on the above principles and regarding the related
development basis and improvement value. research, we selected 39 indicators within eight categories
covering the three dimensions of T/O/D, as shown in
Table 1. According to the 3D principle [45] and collected
data from Ningbo local planning institute, the population

123
244 Urban Rail Transit (2021) 7(3):240–255

Table 1 Criteria and Indicators for measuring TOD index

Attributes C. Criteria I. no. Indicator
no.

Transit 1 Rail transit (T1) 1 Inbound passenger flow

2 Outbound passenger flow
3 Number of station entrances and exits
4 Connecting passenger flow (0–500 m)
5 Connecting passenger flow between (500–1500 m)
6 Connecting passenger flow (1500 m )
2 Traffic connection (T2) 7 The convenience of connection (CCI)
8 Number of intersections
9 Number of bus stations
10 Number of bus routes
11 The density of the road network suitable for walking
12 The density of the road network suitable for bicycles
13 If there are parking lots for non-motor vehicles
14 If there are P R parking lots
Oriented 3 Residents willingness (O1) 15 Residents’ daily living and travel locations
16 Commercial consumption areas
17 Metro passengers’ willingness to transform
18 Non-metro passengers’ willingness to take the metro
4 Development potential around the station 19 Commercial development potential
(O2) 20 Public space connectivity
21 If there is unbuilt land
Development 5 Density (D1) 22 Population density (0–250 m)
23 Population density (250–500 m)
24 Population density (500–1000 m)
25 Job density (0–250 m)
26 Job density (250–500 m)
27 Job density (500–1000 m)
6 Diversity (land use) (D2) 28 Floor area ratio (0–250 m)
29 Floor area ratio (250–500 m)
30 Floor area ratio (500–1000 m)
31 The entropy of the land properties
32 Rationality index of circle development (Standard Gradient
Index)
7 Social service(D3) 33 Density of restaurants
34 Density of convenience stores
35 Density of cultural and sports facilities
36 Density of medical facilities
37 Density of residential quarters
8 Economic development(D4) 38 Residential prices
39 Commercial rents

density, the job density and the floor area ratio as three land properties. We further introduce the rationality index
types of data for the nine indicators (all of them including of the circle development to describe the different radii
the data of the three layers with radii of 250 m, 500 m, and around the station change trend of floor area ratio. Among
1000 m) are more suitable as indicators for the selected the factors of urban design combined with the needs for
items. According to the mix, and the principle of land use TOD improvement, we select the indicators related to the
[30, 44, 51], we use entropy to calculate the diversity of development potential and transportation connection.

123
Urban Rail Transit (2021) 7(3):240–255 245

The development potential criteria consist of three (oriented) attribute. Density, land use, social services and
indicators: commercial development potential, public space economic development are also classified as D (develop-
connectivity, and presence of unbuilt land. Based on rela- ment) attributes. Based on the TOD typology studies, the
ted research on traffic connection criteria [47, 50, 52, 53], Ningbo TOD improvement project evaluation framework
three indicators are selected to describe the features of this (see Table 1) is constructed as previously described.
aspect: the number of intersections, the density of the road Details are presented later in Chapter 3.6.
network suitable for walking, and the density of the road
network suitable for bicycles. The convenience of the 3.6 Three New Variables
connection is then described using the current average
connection distance of the passenger connected to the The three indicators introduced for the construction of the
subway. The design of the transportation connection index system are as follows:
facilities criteria is based on four indicators: the number of
(a) Convenience of connection index (CCI, indicator 7)
bus stations, the number of bus routes, whether there are
parking lots for non-motor vehicles and P R parking lots. CCI describes the convenience of connection for actual
The selection of traffic indicators is site-specific. The metro passengers at each station and describes the weigh-
Ningbo metro has only been opened to traffic for 6 years, ted average connection distance from the inverse sequence.
the network is not complicated, the vehicle type selection is The closer the connection distance, the higher the CCI and
relatively simple, and the difference in line capacity is the better the connection convenience, and vice versa. CCI
insignificant. Therefore, six indicators are selected to is defined as
describe the traffic criteria including the inbound passenger 3
¼1
flow, the outbound passenger flow, the number of station ¼ 2000  3
ð1Þ
entrances and exits, and the connecting passenger flow of ¼1
the three circles below 500 m, 500–1000 m and beyond where i is the index of the station number between 1 and
1000 m. 63, k is the connecting distance circle layer number (within
Regarding the social services and economic develop- 500 m: k 1, 500–1000 m: k 2, outside 1500 m:
ment criteria, we refer to the related research [18, 44, 54] k 3), Nik in the questionnaire is the number of passen-
combined with the actual situation in Ningbo. The selected gers who took the metro at the i-th station with the con-
indicators are the density of restaurants, convenience necting distance at the k-th circle, Sk is the theoretical
stores, cultural and sports facilities, medical facilities, and average distance of the k-th circle layer (within 500 m:
residential quarters. In addition, residential prices and 250, within 500–1000 m: 750, outside 1500 m: taken as
commercial rents are also selected as indicators of the 2000).
economic status of the station area. Similar studies [18, 44]
also mentioned that the setting of indicators should have a (b) (b) Land-use diversity (the entropy of land proper-
certain level of flexibility to be able to incorporate the ties, indicator 31)
differences between cities. This term gives the land entropy value according to the
The TOD improvement (renewal) project around the general formula while considering that some land-use
metro station differs from the general TOD evaluation. properties in urban TOD development may reduce transit
Smooth progress of the project needs to incorporate the use. It treats the land of these properties as a negative
opinions of the residents. Similar views have also been value.
expressed in related studies [20, 23, 25]. Therefore, it is The entropy of land mix degree of the i-th station is
necessary to conduct a survey of metro and non-metro
passengers within the metro coverage area to get their
opinions on the TOD improvement. The following four
¼  ln ln ð2Þ
indicators are the main factors for the public opinion: ¼1
residents’ daily living and travel locations, commercial
consumption areas, metro passengers’ willingness to where Qij is the percentage of the land of type j at the i-th
transform, and non-metro passengers’ willingness to take station in the total land surrounding the station, and N is the
the metro. total number of land types around all stations.
The above indicators are classified according to the T/O/ (c) The rationality index of the circle development
D attributes, where two types of rail transit indexes and (Standard Gradient Index [SGI], indicator 32)
their connection are classified as the T (transit) attribute,
and the two types of indicators of residents’ willingness This term describes the step-change relationship of the
and the development potential around the station as the O floor area ratio of the three circles with a radius of 250 m,

123
246 Urban Rail Transit (2021) 7(3):240–255

500 m, and 1000 m. A standardized unit circle area range is 4.1 Dataset for Classification
used to describe the difference in floor area ratio. The
development of the three circles is approximately close to The research data in this paper were obtained from the
the step distribution. The higher the SGI, the more rea- following sources:
sonable the development of the circle, and vice versa.
• The inbound and outbound passenger flows were
¼ 1 ð1 þ expð ÞÞ ð3Þ collected from the card data of the operating company.
where • Data related to the traffic conditions around metro
stations and connected passengers were provided by the
¼ ð1 Þ  ð 500  250 Þ 5002  2502 Ningbo Transportation Planning Institute.
ð 1000  500 Þ 10002  5002   106 • Data related to the residents’ willingness were collected
using surveys on mobile phone applications.
250 500 1000 250 m 500 m 1000 m
• Data related to on-site facilities were also collected
range Floor Area Ratio ðFARÞ using field surveys.
• The location of surrounding social service facilities was
taken from GIS data and includes the density of
4 Case Study convenience stores and density of medical and health
facilities.
The goal of this research is to identify which Ningbo metro • The economic data were collected using the survey
stations are suitable for TOD improvement and optimiza- conducted by a professional agency.
tion. Therefore, the defined research area includes all 63 Residents’ preferences were included as part of the
stations covered by the Ningbo metro (see Fig. 3, as of Oct. project goal to renovate the periphery of the metro stations
2019). based on TOD [20, 23]. The passenger flow characteristics
Ningbo is a city on the east coast of China, which is in of the relatively new Ningbo metro system differ from
the south wing of the Yangtze River Delta, with Shanghai those of cities with mature metro systems, e.g., Beijing and
as the Yangtze River Delta area centre. The current pop- Tokyo. The inbound and outbound passenger flows from
ulation of Ningbo city is six million, including three mil- the station do not conform to the power-law distribution
lion residing in the urban areas and three million residing in nor the log-normal distribution [56–58], and the number of
the suburban areas [55]. inbound and outbound passengers is also not related to
employment around the station. It is necessary, therefore,

Fig. 3 Ningbo metro network map (Oct. 2019)

123
Urban Rail Transit (2021) 7(3):240–255 247

to supplement the questionnaires of residents (including 4.2 Data Processing and Classification of Metro
metro passengers and non-metro passengers) and establish Stations
a connection between the number of questionnaires and the
traffic, geographic, economic and other factors affecting If the Ningbo metro had entered a mature development
metro ridership. This establishes a connection between the stage, and the passenger flow and surrounding land were
station volumes so that all the relevant data of the three relatively stable, then the TOD improvement methods of
attributes of T/O/D are integrated, which can be used as the existing stations could be leveraged. Otherwise, the TOD
basis of TOD improvement typology. development situations and improvement methods of sta-
We carried out this research survey for the entire Ningbo tions in better-developed and well-known cities at home
city during January 2020, just before the Chinese New and abroad should be targeted instead. According to the
Year and COVID-19 outbreak. Sixty-three stations covered studies on Beijing, Shanghai, Seoul, etc. [36, 59, 60], in
by the metro were surveyed, with 3087 valid samples East Asian cities with more mature metro development, the
obtained from 57 stations representing 1% of the passenger passenger flow in and out of each station shows a power-
volume. law or log-normal distribution. Passenger flow in and out of
the station is correlated with the development of sur-
rounding land, the number of surrounding populations, and
the number of surrounding jobs. For Ningbo, however, the
inbound and outbound passenger flows do not follow a
Table 2 Inbound and outbound passenger flow distribution test power-law or log-normal distribution (see Table 2, Fig. 4).
Equation Inbound R-square Outbound R-square Furthermore, the passenger flow in and out of the station is
not strictly related to the population and jobs distribution in
Logarithmic .864 .880 radii of 250 m, 500 m, and 1000 m. This suggests that the
Quadratic .680 .711 development level of Ningbo metro TOD is still in its
Cubic .826 .851 infancy (Table 3).
Compound .859 .863 Through public surveys, it is found that the effective
Power .769 .772 sample size of each station and the inbound and outbound
S .400 .389 passenger flow, the sample size and the planned population
Exponential .859 .863 and jobs show different degrees of correlation (Spearman
correlation [59]). This shows that the obtained sample is

Fig. 4 Normality test of passenger flow distribution in and out of the station

123
248 Urban Rail Transit (2021) 7(3):240–255

Table 3 Inbound and outbound passenger flow correlation test

Spearman’s rho Population250 Population500 Population1000 Job250 Job500 Job1000 FAR250 FAR500 FAR1000

(Indicator no.) 22 23 24 25 26 27 28 29 30

Correlation coefficient .180 .185 .226 .348** .468** .565** .157 .449** .470**
Sig. (2-tailed) .157 .147 .075 .005 .000 .000 .219 .000 .000

Correlation coefficient .194 .190 .234 .360** .473** .572** .160 .445** .475**
Sig. (two-tailed) .127 .136 .064 .004 .000 .000 .210 .000 .000
** Correlation is significant at the 0.01 level (2-tailed)

Table 4 Effective sample size correlation test

Spearman’s rho Population250 Population500 Population1000 Job250 Job500 Job1000 FAR250 FAR500 FAR1000

(Indicator no.) 22 23 24 25 26 27 28 29 30
Correlation coefficient .460** .517** .530** .334** .409** .450** .228 .503** .531**
Sig. (two-tailed) .000 .000 .000 .007 .001 .000 .073 .000 .000
** Correlation is significant at the 0.01 level (2-tailed)

representative of the current passenger flow and the dis-

tribution of the surrounding population post. Survey data
can be also used for TOD model data analysis (see Table 4)
The collected data are tested to determine whether the
Cronbach- coefficient is greater than 0.6, which is the
basic condition for data reliability and variable selection.
After testing, the Cronbach- coefficient of all 39 variables
is found to be greater than 0.6, and therefore the data are
considered reliable. The data validity test is based on the
Kaiser–Meyer–Olkin (KMO) value and Bartlett sphericity
test. The KMO value of the test data is 0.737 0.6, and the
-value of the Bartlett sphericity test is 0.01. These also
show that the selected 39 indicators can explain the eval-
Fig. 5 Eight criteria of influencing factors weight (by entropy
uation of TOD improvement at stations. The entropy cal- calculation)
culation weights are shown in Fig. 5. The evaluation of the
TOD development level based on E-TOPSIS is also shown better than SOM and SVM. After applying z to transform
in Fig. 6. the result, the visual output according to the T/O/D three-
It is shown that the use of machine learning methods [K- dimensional coordinates is presented in Figs. 7, 8, and 9.
means [39, 40], self-organizing map (SOM) [29, 60],
support vector machine (SVM), etc.) to classify TOD types
helps reduce the interference of subjective factors in 5 Results
making a judgment. According to the T/O/D attribute
characteristics of different indicators, they are included in 5.1 Four Types of Clusters
the three types of attribute indicators. Based on the cal-
culation of entropy weights, the T/O/D attribute score of Based on the results of K-means clustering, the TOD
each station is obtained, and then the result was calculated improvement of Ningbo metro stations is divided into four
using the K-means cluster analysis. In the case of Ningbo, clusters (see Fig. 10):
we found that the interpretability of K-means results is

123
Urban Rail Transit (2021) 7(3):240–255 249

Fig. 6 The TOD development score ranking of the station in Ningbo metro, 2019 (take the golden ratio of all the stations to mark, as it reflects a
certain level of development, and is suitable as a candidate set for selecting TOD improvement)

Cluster 3: ITOD (improvable TOD). The basic condi-

tions for TOD development such as population and job
density are reasonable, but there are shortcomings in at
least one aspect of transportation supply, just like trans-
portation connection, or economic development, which has
not shown the TOD effect (rail transit has a large passenger
flow and a high degree of urban mixing). Such stations are
often in the construction region and have a large room for
renovation and improvement, and are the key type rec-
ommended for renovation.
Cluster 4: PTOD (potential TOD), population and job
density and other conditions are also good. There is no
obvious shortcoming in the scoring performance. However,
compared with station C1, it is not a city or district centre.
The distribution of the four types of stations in the
Ningbo metro network is illustrated in Fig. 10.

5.2 Target Station Recommendations Based

Fig. 7 A 3D node-place model based on TOD improvement,
including scatter plots of all 63 stations and four clusters on the Clustering Results

Cluster 1: NTOD (no TOD), has not yet shown a station According to the node-place model based on the TOD
with TOD significance, but a station with a better foun- improvement, the recommended station as the target of
dation may be transformed into a regional centre of the TOD improvement is the station near the four arcs
city. enclosing the accessibility area. As the Ningbo metro is
Cluster 2: ATOD (already TOD)/city activities zone still in the initial construction stage, there is no station with
(CAZ), has developed a certain TOD and has become the a strong T-attribute (corresponding to the node attribute).
core station of the city. Therefore, no station enters the U-N area, and the CAZ of
Ningbo belongs to the CBD before the construction of the

123