Integrating Network Centrality and Node-Place Model to Evaluate and Classify Station Areas in Shanghai

Abstract

Transit-oriented development (TOD) is generally understood as an effective urban design model for encouraging the use of public transportation. Inspired by TOD, the node-place (NP) model was developed to investigate the relationship between transport stations and land use. However, existing studies construct the NP model based on the statistical attributes, while the importance of travel characteristics is ignored, which arguably cannot capture the complete picture of the stations. In this study, we aim to integrate the NP model and travel characteristics with systematic insights derived from network theory to classify stations. A node-place-network (NPN) model is developed by considering three aspects: land use, transportation, and travel network. Moreover, the carrying pressure is proposed to quantify the transport service pressure of the station. Taking Shanghai as a case study, our results show that the travel network affects the station classification and highlights the imbalance between the built environment and travel characteristics.

1. Introduction

Rapid urbanization and population growth worldwide have caused many urban problems, such as air pollution, traffic congestion, and excessive reliance on fossil fuels. An essential reason for this phenomenon is the absence of clear and effective integrated development plans for land use and transportation. Transit-oriented development (TOD) has been adopted as a commonly followed urban planning strategy for achieving such integration, promoting efficient and mixed land use near public transportation hubs and stations [1,2]. At the metropolitan scale, metro stations provide greater accessibility to their catchment area and play a crucial role in the public transport network. Meanwhile, stations offer sufficient opportunities to develop diversity and intensity land use, aiming at creating walkable, activity-friendly communities. The implementation of TOD at station level has been a major focus of attention in scholarly work.

Various researchers have attempted to establish comparable TOD typologies for better strategic planning, investment guidance, and station quantification [3,4]. The best-known approach for evaluating TOD typology is the node-place (NP) model developed by [5], which is essentially a conceptual framework for assessing the transport supply and land use characteristics of a station simultaneously. Based on the NP model, various modifications to the original model have been introduced. The most notable modification is the extension of design dimension, which represents a considerable improvement regarding walkability and pedestrian comfort [6]. Specifically, three aspects need to be investigated to achieve effective integration of transportation and land use: (1) multimodal accessibility of the station, (2) land use diversity and intensity, and (3) pedestrian-oriented design [7]. In a similar vein, Ref. [8] extended the model by adding the experience dimension with indicators, which reflect the transit quality from a traveler’s perspective.

Recently, traveler-related attributes, such as ridership, as critical indicators have been studied. For example, Ref. [9] elaborated on the correlation between the travel volume and NP index. Ref. [10] investigated the interaction between node value, place value, and ridership. However, ridership has limitations to reflect the travel network characteristics (e.g., transfer behavior). There is scant research on the synergies between the actual rail transit travel characteristic and the functional configuration of its physical context. To the best of our knowledge, Ref. [11] made the first exploratory study on the relationships between travel network features and the classification results of the node-place-design model.

Inspired by these efforts, we attempt to extend the node-place model by adding a new dimension of network for metro station classification, which contributes to illuminating the complex links between the built environment, land use, and travel network characteristics of stations. The main contributions of this study are as follows:

• Combined official data and open geographic big data, an extended node-place-network (NPN) model was proposed to measure the synergy between the built environment and travel characteristics around stations. Moreover, the carrying pressure indicator was utilized for the quantitative evaluation;
• Employed the proposed analytical methodology to complement empirical knowledge concerning TOD of the station’s area in Shanghai, which verified the effectiveness of the proposed NPN model.

The paper is organized as follows. The next section is a literature review of the NP model and relevant theory. Then, we elaborate on the methodology used and a case study of Shanghai. Afterwards, the results are discussed in Section 5, followed by conclusions in Section 6.

2. Literature Review

2.1. The Node-Place Model

In the past, many researchers examined the gaps between public transport supply and demand. Various approaches were employed to access the public transport performance, for example, network supply model [12], Gini coefficients [13,14], and node-place model [4], etc.

Among these models, the node-place (NP) model, as the best-known approach leading to a TOD typology, was developed by [4]. It defined the double nature of the station: nodes of networks and place in the city. The funding principle of the NP model is balancing transportation with land use to achieve a state of cooperative development between them. An analysis framework, based on a dual axis of node and place indexes, is used to represent the conceptual model (Figure 1). The node dimension measures the offer of transport services (e.g., number of directions served, daily frequency), reflecting the accessibility of the station area by several transportation modes. The place dimension measures the land use features (e.g., density and diversity of activities) found around a station area, reflecting the potential demand for transport services. By identifying five types of station areas (Figure 1), the conceptual framework provides an integrated quantitative indicator for measuring the integration level of transportation and land use as well as assessing the evolution of a system over time [15,16].

During the last two decades, the dimension of NP has been extended in many diverse aspects [18,19]. One major modification is its expansion to the urban design dimension. [6] first investigated the usefulness of adding the walking environment evaluation to the NP model as an urban planning tool. Based on that evaluation, Ref. [20] emphasized the functional and morphological interrelation between transportation and urban conditions, and extended a third, ‘oriented’ dimension to quantify the accessibility of station areas. Likewise, [7] considered different access modes to stations and introduced the design index to measure the influence of the pedestrian accessibility of the station areas. Other feeder modes, such as bicycles, cars were considered in the work of [21], which focused on geographical contexts characterized by medium or low population densities. In a similar vein, several extra dimensions, such as ‘proximity’ [22], ‘user-friendliness’ [23,24], and the spatial configuration of the street network design [25,26] were also investigated in previous studies.

In addition to the extension of the design dimension, some scholars conducted research from a user-based perspective. For instance, the experience value [8] is added to reflect the traveler’s experience at the station in terms of comfort (sheltered waiting, toilets), ambient elements (architecture and recent renovation), and social elements (presence of personnel). The people dimension proposed by [27] includes motivation (education, work, and other), ridership (using the station as an origin station, destination), and effort (walking, biking, and travelling a farther distance).

Furthermore, the high salience of transit-related attributes, especially human mobility in the studies, are identified. Two approaches have been suggested to include travel features in the NP model. The first approach highlighted the importance of the passenger frequencies [20] and adds passenger volume to calculate the node index [26,28]. Given unique characteristics and importance of travel flow data, the second approach made a comparative analysis between the travel characteristics and the results of station classification. According to this point, Ref. [29] explored the intersection of two created classifications for New York City and unveiled the consistency between context and use of the station. Similarly, Ref. [11] added the centrality analysis of the movement network to investigate the performance of the Greater London metro station areas at both local and system levels. They found that there were substantial discrepancies in the criticality of station areas within the same cluster with similar node-place-design, suggesting the various roles they played at multiple levels.

2.2. TOD and the Travel Network

Although improvements of the NP model varied to some extent, they have the common objective to increase transit ridership and create a more sustainable transit system. Compared with other NP indicators, travel flow has its own unique characteristics. First, it is the objective data generated by individual travel, reflecting the actual demand of people. Second, travel distribution reflects the characteristics of the entire travel system rather than those of a local station. Based on that, it is reasonable to think that taking travel flow as an indicator of the node index within the classification of TOD has limitations.

As for the specific index calculation method of the passenger travel dimension, most previous studies have used passenger frequency as an indicator [3,30,31]. One exception is [11], who introduced a strategic network (criticality) component to enhance the value of the NP model from a perspective of network analysis. As pointed by [32], ‘characterizing the network anatomy is important because structure always affects function and vice versa’. The study of networks pervades all of science. In particular, this new wave of spatial networks and flows in cities using moving data provides new insights into understanding traffic network systems and urban spatial structure. For example, Refs. [33,34] used three-year smart card data to reveal a polycentric urban development phenomenon in Singapore. Similarly, Ref. [35] constructed a multi-layer network from three sources (i.e., bus GPS observations, smart card data, and roadside Blue tooth detector records) to identify the network structure at each layer and reveal comparable features of the network organization for different spatial layers. In addition, some scholars also used other moving flows, such as the bicycle sharing flow and social media check-in stream to reveal the hierarchical structure of urban space [36,37].

Numerous empirical studies have shown that the importance of nodes and the type of metro stations can be further identified using the metrics of complex network theory [38]. For example, Ref. [39] developed a new weighted composite index to position hub stations in a subway network. Ref. [40] used several improved network indexes to distinguish the key stations and sections of the metro network in Beijing, China. Ref. [41] used human mobility patterns and improved the PageRank algorithm to evaluate station importance.

Consequently, we develop a new model, which is called the node-place-network (NPN) model with an extended network dimension, as a complement to the NP model. Below we discuss how we developed and applied this extended model in a case study of Shanghai.

3. Methodology

In this section, we first introduce the node-place-network model including the index selection and station classification with the K-means++ method. Second, we calculate the carrying pressure to verify the meaning of travel network dimension. Taken Shanghai as case study, our methodology has been depicted in Figure 2.

3.1. Node-Place-Network Model

As discussed in Section 2, the actual travel characteristics of passengers and the importance of the station in the network system were ignored in previous node-place model. Hence, an extended node-place-network model, as shown in Figure 3, is proposed. The node and place dimension represent the traffic services and land use around the stations, respectively. The network dimension indicates the importance of the station in the whole subway network system.

3.1.1. Index Selection

(1) Network indicators

Based on the smart card data (SCD), we first construct an OD (origin-destination) matrix of travel between all metro stations, and then convert it to an undirected weighted network form, viewed as a passenger movement network:

G = (V, E, W), E = W * W

(1)

where V is a list of vertices (i.e., stations) of network G. E is a set of edges of network G, and W, which includes the weights w_ij denoting the volume of travel between stations i and j.

Degree centrality and centrality betweenness are the two most representative indicators in complex network studies [42,43].

• Degree centrality

The degree centrality refers to the number of links connected to a node in the network, which is the number of travel records of a metro station in our study.

$$S_i={\displaystyle \sum }_{j\in N_i}{w}_{ij}{N}_i$$

(2)

where N_i is the set of adjacent stations of node i, and w_ij is the weight between nodes i and j. In our study, it indicates the number of travel flow between stations i and j.

• Betweenness centrality

Betweenness centrality is an index that measures the level of node connectivity, which is a key indicator for identifying metro network hubs. A node has a higher centrality C_betweenness(k), the greater the number of shortest paths that traverse it and is defined as follows:

$$C_{betweenness}\left(k\right)={\displaystyle \sum }_{ij}{\delta }_{ij}\left(k\right)/{\delta }_{ij}$$

(3)

where δ_ij(k) is the number of shortest paths between any two stations i and j that pass through the station k and δ_ij is the total number of such paths between i and j.

One point to note here is that both centrality indicators are calculated based on the passengers’ movement network instead of the features of transit infrastructure network.

(2) Node indicators

Node indicators can be divided into two groups depending on the transit mode: the transit quality of rail and feeder transport (

Table 1. Indicators of the node dimension.

Node Index	Description
Rail
Number of directions served by metro	n1 = number of metro services offered at station
Daily frequency of metro services	n2 = number of metro departing from station on working day
Number of stations within 20 min of travel	n3 = number of stations reachable within 20 min by metro
Feeder transport
Number of directions served by other public transport	n4 = number of public transport services offered at station
Number of car parking places	n5 = number of car parking within 600 m.
Distance from the closest motorway access	n6 = distance to closest motorway access

).

(3) Place indicators

In terms of place indicators, three aspects of the station’s area are measured following the ‘three-Ds’ principle, namely density, diversity, and design [2]. Based on previous studies [20,27,44], the indicators of place are summarized in

Table 2. Indicators of the node dimension.

Place Index	Description
Density
Population density	p1 = density of population within 600 m
Number of residences	p2 = number of establishments in residence within 600 m
Number of workers in business	p3 = number of establishments in business within 600 m
Number of workers in green space	p4 = number of establishments in green space within 600 m
Number of workers in transportation	p5 = number of establishments in transportation within 600 m
Number of workers in public service	p6 = number of establishments in public service within 600 m
Number of workers in industry	p7 = number of establishments in industry within 600 m
Number of POIs	p8 = number of points of interest (POIs) within 600 m
Diversity
Land use mix	p9 = $-\frac{{{\displaystyle \sum }}_{i=2}^7\left(p_{i1}·ln{p}_{1i}\right)}{6},$ where $p_{1i}$ is the normalization of $p_i$
Design
Intersection density	p10 = density of intersections per hectare
Accessible network length	p11 = length of the accessible network (meters)

.

(4) Data normalization

The first step in descriptive statistical analysis is data transformation. With formula (4) and (5), all indicators were box-cox transformed [45] and rescaled to reduce the skewness of their univariate distribution and improve their comparability.

$$x_i^{\prime }=\left\{\begin{array}{cc}\frac{x_i^{\lambda }-1}{\lambda },& if \lambda \ne 0;\\ log{x}_i ,& if \lambda =0.\end{array}\right.$$

(4)

where $x_i^{\prime }$ is the transformed variable; λ is the transformation parameter, ranging from −5 to 5. The optimum value of λ is obtained using the maximum likelihood estimate.

$$x^{*}=\frac{\max \left(x_i^{\prime }\right)-x_i^{\prime }}{\max \left(x_i^{\prime }\right)-\min \left(x_i^{\prime }\right)}$$

(5)

Then, the two-sided Spearman correlation analysis was also conducted to understand the direction and strength of the relations between the indicators. Furthermore, the relative importance of independent variables will be explored with the entropy method [46].

3.1.2. Station Classification

After data preprocessing, the similarity of metro stations was explored by applying the K-means++ method [47]. K-means++ can choose the initial cluster centers as far apart from each other as possible. The silhouette coefficient, ranging of [−1, 1], is used to evaluate the clustering performance [48].

3.2. Carrying Pressure

To verify to what extent the travel network dimension adds meaning to the findings of a traditional NP type of classification, we conduct a comparative analysis based on two classification strategies. The first strategy primarily focuses on the original NP model, which describes the built environment properties around the station. The second strategy focus on the NPN model, including extended passenger travel characteristics, which describes the real travel flow distribution of rail transit externally.

Furthermore, a new index called the carrying pressure of the node was proposed to quantitatively describe the development synergy between the travel flow and the supporting physical environment of the station. The carrying pressure value of the node can be determined using the following formula:

$$cp\left(i\right)=\frac{T\left(NPN\right)}{T\left(NP\right)}=\frac{\left[Index\left(Node\right)+Index\left(Place\right)+Index\left(Network\right)\right]/3}{\left[Index\left(Node\right)+Index\left(Place\right)\right]/2}$$

(6)

where T, the score of model, is the average of integrated indices, representing the comprehensive measure of TOD development status of the station. Index(.) represents the score of one dimension. For example, Index (Node) represents the composite score after weighting node indicators.

4. Study Area and Data

4.1. Study Area

China has been experiencing rapid urbanization and a massive expansion of infrastructure construction during the past decades. Shanghai, as China’s largest economic center, has a 548 km network, which is officially the largest network by route length in the world [49]. In 2018, the daily ridership of Shanghai Metro exceeded 10 million on weekdays and 7 million on weekends. The total number of subway passengers reached approximately 3.7 billion, which increased 1.62 billion over that of 2017.

According to the Master Plan, officials intend to have over 1000 km of network by track length, and over 600 metro stations, forming an extensive network covering the entire city. In particular, the conception of TOD has been introduced and considered a promising solution by local governors in land use planning. In this regard, it is necessary to have a universal approach to identifying the TOD variations and typology among the metro station areas and providing general knowledge on understanding the complex relationship between metro trip demands, transportation, and land use. Based on our data source, all 286 metro stations, covering over 600 km that are part of the analysis, are shown on the map (Figure 4).

4.2. Data

This study used three sources of open data, including 642,724 POIs extracted from Gaode Map, LandScan, OpenStreetMap (OSM), government website data, and a month of smart card data (SCD) from Shanghai Metro during March 2018.

The recorded SCD contains detailed information on each trip, including the card ID, time, fare, and the station name. From the fare field, the state of travel can be detected (i.e., whether boarding or alighting). The government website data offers the basic information of the metro network, such as the metro daily frequency, number of directions served by metro, etc. In addition, road-related information, like intersection density and accessible network length, can be obtained from OSM data. The population distribution data was from the LandScan (http://web.ornl.gov/sci/landscan, accessed on 1 September 2020). The description of data can be seen in Appendix A.

5. Results and Discussion

In this section, we first carried on the variable selection and the comprehensive score calculation of node, place, and network dimensions. Then, the indexes of NP and NPN were clustered respectively to obtain different station categories. Finally, the difference of station categories under NP and NPN were investigated and the guidance for individual stations are discussed.

5.1. Node-Place-Network Model

5.1.1. Variables

The direction and strength of correlations between the indicators were examined (Pearson’s), and the results show a high and significant correlation between the variables (see Appendix B). Nevertheless, the high correlation between some variables was not considered problematic, as the indexes of each dimension are then given different weights and finally integrated into one index.

Table 3. Indicators of the node dimension.

Dimensions	Independent Variables	Rank	Relative Importance (%)
Node	n1: Number of directions served by metro	1	23
	n2: Daily frequency of metro services	3	16
	n3: Number of stations within 20 min of travel	2	27
	n4: Number of directions served by other public transport	5	14
	n5: Number of car parking places	4	15
Place	p1: Population density	3	10
	p2: Number of residences	1	15
	p3: Number of workers in business	3	10
	p4: Number of workers in green space	6	9
	p5: Number of workers in transportation	2	11
	p6: Number of workers in public service	3	10
	p7: Number of workers in industry	6	9
	p8: Number of POIs	8	8
	p9: Land use mix	9	7
	p10: Intersection density	11	4
	p11: Accessible network length	9	7
Network	c1: Degree	1	50
	c2: Betweenness	1	50

shows the relative importance of independent variables with the entropy method. For the node index, the number of directions served by the metro is the most important variable, which emphasizes the importance of a transfer station. For the place index, the density and diversity of land use have greater weight, followed by the mixing degree of land use, and finally, the density of intersections. This result is consistent with the previous study [50] that the population density and employment density are strongly positively associated with ridership. For the network index, the degree and betweenness indicator have identical weights, which indicates that these two variables are equally important in reflecting the importance of stations.

5.1.2. Indexes

A descriptive statistical analysis of the three dimensions is shown in

Table 4. Descriptive statistics of the node, place, and network indexes.

Index	Minimum	Maximum	Mean	Std. Deviation
Node	0.1480	0.8674	0.4908	0.0204
Place	0.1135	0.8167	0.5351	0.0231
Network	0.1556	1	0.5235	0.0457

. The ranges of data indicate that the distribution of real passenger flow distribution between subway stations has a greater variation (which is 0.046) than the node dimension (0.020). The result indicates that the average network accessibility is not being fully incorporated into transportation development.

The geographic distribution of the node and place indexes follows a broadly concentric circle-shape, generally radiating away from the central area of Shanghai (Figure 5). The node and place with better development status are basically located in the central region, while the relatively underdeveloped regions occur on the outskirts of the city, which also reflects the strong monocentric urban structure of Shanghai. Notably, the spatial distribution pattern of the network dimension shows a substantial difference between node and place indexes. A higher value of network index is no longer located in the central area and appears in the periphery of city. It is evident that the periphery/suburbs of the city have a high travel demand, whereas the supply of these places is in relatively short supply with an undeveloped physical environment (including traffic facilities and land use allocation).

In addition to the spatial analysis of all station areas, the relationships among these indexes were also analyzed to reveal the complex links between stations. As shown in Figure 6, Shanghai metro station areas exhibit substantial differences among these indexes: the results of a positive relationship between node and place (r2 = 0.69) indicate that the public transport accessibility has a good synergy associated with land use patterns. The relationship between network index and the other two dimensions is relatively low (0.36 and 0.48), which is consistent with [11] and that there is a significant difference between network centrality and the attribute of local station areas.

5.2. Cluster Analysis

5.2.1. Typologies of the NP Model

Through K-means++ cluster, five clusters are chosen with the best silhouette value 0.38. The F-statistics of clusters of the place index are 541, higher than node index, which are 428. The lower descriptive power of the node index indicates that the transportation service level does not differ greatly. The reason for this situation may be that we mainly analyzed the urban area of Shanghai, which has developed public transportation.

Using the geographic map and NP scores, the node index and place index in Shanghai are generally in a state of equilibrium (Figure 7 and

Table 5. Summary of the five clusters.

ID	Num	Example of Stations	Node (Avg.)	Place (Avg.)	All (Avg.)
1	62	People’s Square, Jing’an Temple	0.67	0.72	0.69
2	80	Hongqiao railway station, Xinzhuang	0.57	0.59	0.58
3	70	Jiuting, Huinan	0.44	0.49	0.47
4	48	An’ting, Nanxiang	0.32	0.41	0.36
5	26	Xinchang, Zhoupu east	0.27	0.22	0.25

). The geographic distribution follows a broadly concentric circle-shape radiating from the central area of Shanghai. For instance, cluster 1 and cluster 2, two better performing clusters, are primarily located in central urban areas with good traffic accessibility and diverse land use. Followed by cluster 3, which is mostly located from the edge of the central city area and has a moderate level of development in terms of urban traffic and land use. The last two clusters, cluster 4 and cluster 5, are located at the edge of the city or at the end of the rail transit station.

5.2.2. Typologies of the Node-Place-Network Model

The same cluster analysis was again applied to the NPN model and five clusters are identified (the best silhouette value is 0.32). The F-statistics (network index = 345, place index = 204, and node index = 205) shows that higher descriptive power of the network. Further, in the case of similar built environment, the actual travel flow distribution among different stations is quite different. This result once again proves the necessity of travel flow for site classification.

Each of the five categories has its own distinctive characteristics (

Table 6. Description summary of the five clusters.

ID	No.	Stations	Node (Avg.)	Place (Avg.)	Network (Avg.)	All (Avg.)
1	57	People’s Square, Jing’an Temple	0.65	0.70	0.79	0.71
2	84	Baoshan Road, Longqiao Road	0.60	0.62	0.56	0.59
3	49	Jiuting, Zhuanqiao	0.39	0.46	0.61	0.48
4	61	An’ting, Longhua	0.41	0.47	0.36	0.41
5	35	Longyao Road, Luoshan Road	0.28	0.28	0.16	0.23

and Figure 8). Cluster 1 (the number of stations is 57) is characterized by the three highest indexes, predominantly located in the central area of the city. Two exceptions are Hongqiao Railway Station (0.69, 0.66, 0.87) and Xinzhuang (0.59, 0.65, 0.89), which are clearly positioned as sub-centers of urban development according to the Master Plan of Shanghai (2017–2035).

Cluster 2 (the number of stations is 84) has most stations, which has a balanced development status. Cluster 3 has a similar score for network with cluster 2, with relatively high scores for the network but a low score for node and place. This is possibly because around these stations, the land use is usually dominated by residential types, and more commuters need to travel between workspace in the city center and residential areas in the suburbs. One typical case is Sijing station (0.32, 0.42, 0.78), a typical living place, located in an area with a dense residential population, large travel flow during the peaks and insufficient facilities.

Cluster 4 (the number of stations is 61) and cluster 5 (the number of stations is 35), are generally located at the end of a metro line. Both of them have a relatively low score of network, corresponding to lower demand for transportation. Furthermore, the cluster 5 needs to be improved comprehensively in terms of travel demand, transport services, and land supply.

5.3. Comparative Analysis

Figure 9 provides an alluvial diagram that depicts how the clusters in different strategies interact with one another. Between the two categories, there are thick edges that connect the clusters. This situation indicates that the set of nodes that form a particular cluster within the NP model constitutes a substantial portion of another cluster within the NPN model, reflecting the similar development consistencies under different dimensions. Cluster 1 to 5 is arranged in order of the average score from large to small. A large number of thin edges change to the neighbor cluster IDs, which indicates that some stations perform slight differences between the two categories. However, several thin edges change to the non-neighbor cluster IDs, reflecting the significant mismatches between environmental attributes and travel demand. For example, Jinke station (network = 0.82, node = 0.55, place = 0.49), which belonging to cluster 3 under NP classification, is transferred to cluster 1 under NPN classification.

To quantitatively describe the development synergy between the travel flow and the physical environment of the station, we compare the carrying pressure within different categories across these two classifications. As shown in Figure 10, the score of NPN shows more variability and comparability, while there is no clear distinction in the NP model. The results from the NPN model also clearly show that the nodes in cluster 3 have a higher T(NPN) score, while cluster 5 shows a lower performance in the public transport system. The results of the comparison study show the validity of the extended NPN method.

5.4. Guidance for Individual Stations

The carrying pressure analysis on different clusters provides us not only a general impression of station groups, but also specific details for 286 stations in Shanghai (see

Table 7. Top 10 stations of carrying pressure.

Station	Clusters (NPN)	Clusters (NP)	Score (NPN)	Score (NP)	CP
Hesha Hangcheng	3	5	0.37	0.26	1.46
Xinchang	3	5	0.33	0.23	1.40
Zhoupu East	3	5	0.36	0.26	1.39
Sijing	3	4	0.51	0.38	1.36
Shenshe Road	3	4	0.51	0.38	1.33
Nanxiang	3	4	0.53	0.41	1.30
Jiuting	3	3	0.59	0.47	1.27
Guanglan	3	3	0.53	0.42	1.26
Jiading North	3	4	0.45	0.36	1.26
Jiading Xincheng	3	4	0.40	0.32	1.23

). The stations with a high carrying pressure can be characterized as follows: (1) transportation hubs, such as Hongqiao Railway Station and Pudong International Airport, which are reasonable for having large travel demand because of the dominant position of transportation; (2) large residential districts, such as Jiuting and Guanglan Road stations, which are generally located on the edge of a main urban area or outside the city. These areas serve as sub-centers for future urban development. (3) The terminal stations, such as Shenshe Road station, which have a low degree of the physical environment but a large travel demand. Moreover, our findings are consistent with the Shanghai Master Plan 2017–2035. For instance, four integrated and promoted town clusters are found in Figure 11 (Areas 1 to 4), in planning that includes Nanxiang-Jiangqiao, Jiuting-Sijing-Dongjing-Xinqiao, Pujiang-Zhoupu-Kangqiao-Hangtou, Tangzhen-Caolu-Heqing. There are also two prominent clusters, which correspond to the planned central towns, Jiading and Nanhui.

6. Conclusions

In this paper, we extended the existing node-place (NP) model by adding a third dimension—the network dimension, which results in a new model, the node-place-network (NPN) model. The NPN model can be used to assess the extent to which the cooperation between the land use and public transport as well as travel demand. An empirical case study of Shanghai demonstrated the effectiveness of our method. Three key findings are obtained: First, the results show that the TOD guidelines have already been implemented in Shanghai, although there is also a large mismatch between the travel demands and the surrounding environment of stations. This result is largely due to the complex relationship between transport systems and travel patterns. Second, a comparative analysis of the discrepancies between two classifications, the stations that need possible improvements are identified. Third, through the index of the carrying pressure, the matching degree of travel characteristics and environmental development can be quantitatively evaluated, which provides a guideline for the development of multi-center cities.

Our results also provide recommendations for policy making. The stations with a relatively high carrying pressure have the potential for encouraging densification and diversification to increase the land use efficiency. We recommend planners to apply our methodology to a selection of potential places as sub-centers. Moreover, the reason for the occurrence of stations with a high carrying pressure is worth monitoring. Heavy commuting burdens pose a major challenge to improving the quality of public transportation. In particular, the rapid growth of urban populations has placed enormous pressure on transportation services and land use allocation [51].

This study can be extended in several directions for future investigations. First, we only used the entire travel records during a day, while more fine-grained event resolution, such as 1-h temporal interval can be considered. In other contexts, Ref. [29] used smart card data to conduct station classification, which brings finer resolution for boarding and alighting information. Second, we chose a buffer zone of 600 m in this study, which is the service scope planned by the Shanghai Municipal Government. Due to the presence of the modifiable areal unit problem (MAUP) [52,53], the influence of different research scopes and research scales can be considered in the future. Third, the different methods of index weighting can be adopted, e.g., the fuzzy approach and ANN [54,55]. Furthermore, as Shanghai is a monocentric city, further research will be carried out in cities with polycentric development to verify the performance of the proposed method.