Assessing polycentric urban development in Shanghai,China,with detailed passive mobile phone data

Abstract

Despite growing studies on the distinction between morphological and functional polycentricity, the present methods for identifying polycentricity often focus on the morphological dimension due to a lack of information about intra-urban functional flows, and are limited by the multifarious nature of people’s spatiotemporal interactions. This study proposes a new approach, examining the degree of polycentricity in Shanghai at the intra-urban level using passive mobile phone data. A series of polycentricity indicators are used and are benchmarked against previous studies. Notably, we found that people’s daily movements within a subcenter indicate that morphological polycentricity is also at play in Shanghai. We conclude that morphological and functional polycentricity may coexist at the intra-urban level, and that a mobile phone data approach can offer an alternative method to elucidate both the morphological and functional features of subcenters.

Keywords

Functional polycentricity human activity mobile phone data Shanghai

Introduction

Urban spatial structure, in general, refers to the spatial organization of population and employment in metropolitan areas. Many recent studies have suggested a shift from monocentric cities toward more polycentric urban configurations in many parts of the world (De Goei et al., 2010). For example, the rise of polycentric urban patterns may be due to suburbanization, as people may opt to reduce their commuting costs by living close to employment subcenters (Cervero and Wu, 1997; Giuliano and Small, 1991). Empirical studies have subsequently assessed how such a shift toward polycentricity may be associated with economic performance (Li and Liu, 2018; Suarez-Villa and Walrod, 1997), commuting time (Aguilera, 2005; Schwanen et al., 2004), environmental sustainability (Vasanen, 2012; Veneri and Burgalassi, 2012), and the social well-being of cities (Horton and Reynolds, 1971). Polycentricity is also seen as a stretched concept (Van Meeteren et al., 2016): one that can hold different meanings and can exist at different scales (Davoudi, 2003). The extant literature on polycentricity can be fitted to two broad axes: scale and concept of function versus morphology (Burger and Meijers, 2012; Green, 2007; Kloosterman and Musterd, 2001).

Despite these contributions toward overcoming some of the analytical and empirical ambiguities of polycentricity, the present measurement techniques for the degree of polycentricity at the intra-urban scale are challenged by the highly diverse and intertwined ways in which people actually make use of urban space (Roth et al., 2011; Zhong, 2014; Zhong et al., 2014). Since polycentricity can refer to the multi-nodal development of complicated human activity in a self-sufficient urban entity (Kloosterman and Musterd, 2001), Burger et al. (2013) and Zhong et al. (2017) called for further investigations into the multifarious spatial interactions within an urban system. The traditional travel survey is not able to delineate human activities for all citizens (Zhong et al., 2017). With the emergence of new technologies, individual-level and activity travel data from sources such as mobile phones may provide new opportunities to move beyond travel-survey-based functional connection (Zhang et al., 2016a). In fact, limited studies have already used mobile phones to study polycentric structures at the regional scale (Lee et al., 2018; Louail et al., 2015).

This study attempts to contribute in this area, examining Shanghai’s degree of polycentricity at the intra-urban level by utilizing passive mobile phone data. We extend the present framework of functional intra-polycentricity—we believe that mobile phone data combine many aspects of polycentricity since it contains a vast amount of spatiotemporal information on the density, frequency, and direction of people’s actual usage of space. Therefore, clustering the density of multi-purpose trips and activities gives rise to a more comprehensive definition of polycentricity at the intra-urban scale. This research explores these ideas through the case study of Shanghai, selected because the evidence so far from China on the functional aspects remains relatively scarce at the intra-urban level (Yue et al., 2019a; Zhang et al., 2019; Zhou et al., 2018).

Relevant literature reviews

Polycentricity and its measures

The traditional monocentric model assumes that all jobs are located in the central business district (CBD), but there has since been much evidence to show that the metropolitan area has become increasingly decentralized and deconcentrated, especially for post-industrial cities, as suburban areas increasingly evolve into local centers, developing their own economic activities and demand for jobs (Hu et al., 2019). Kloosterman and Musterd (2001) and Hall and Pain (2006) interpreted polycentric development to be a spatial process, resulting from the outward diffusion of (often higher-order) urban functions from major centers to nearby smaller centers. Although polycentricity is a very popular concept in academia, it is still, however, a fuzzy concept (Markusen, 2003)—polycentricity can be explained not only analytically but also through normative approaches at different spatial scales (Kloosterman and Musterd, 2001; Waterhout et al., 2005).

According to the extant literature, there are two analytically distinct approaches for understanding and measuring polycentricity: the morphological approach and the functional approach. Van Meeteren et al. (2016) found that the functional approach appeared to be more influential among European megaregionalists. Specifically, the morphological approach refers to the nodal balance features of the spatial distribution of urban (sub)centers (Burger and Meijers, 2012), using methods such as location quotients, rank–size relation, employment density, and employment-to-work ratio (Batty, 2001). For example, Giuliano and Small (1991) defined a subcenter as having a minimum employment density of 10 employees per acre and a minimum of 10,000 employees. These specifications seem subjective and lead to the identified centers being highly sensitive to the spatial scale of the analysis (Anas et al., 1998). McMillen (2001) used a two-stage nonparametric procedure to identify how employment density declines with distance to the CBD. Given that, Redfearn (2007) pointed out that this identification of subcenters is heavily reliant on the parametric form and local knowledge, among other factors.

However, many scholars point out that the morphological approach ignores the actual structure of the urban system (Burger and Meijers, 2012; Sohn, 2005; Vasanen, 2012). In particular, the geographical proximity of each subcenter does not necessarily imply a functional connection between these subcenters (Lambooy, 1998). Taylor et al. (2010) stated that the central place theory might be invalid for the specific case of the economic trade center. This implies that the spatial logic behind contemporary urban regions may be that of increased spatial interaction; the central flow theory is, therefore, a clear complement to central place theory. Urban regions may be physically separated but functionally linked by, for instance, the flows of commuting trips (Hall and Pain, 2006; Hall and Tewdwr-Jones, 2010). In this sense, the functional approach emphasizes a multi-directional set of functional linkages between (sub)centers (Hall and Pain, 2006). Indeed, functional polycentric urban structure was first observed in the field of commuting patterns (Cervero and Wu, 1998; Van der Laan, 1998), where it was found that the spatial pattern of commuter flows no longer followed the traditional monocentric model, and seemed increasingly complex in suburban areas (Van der Laan, 1998). As Parr (2004) remarks, the polycentric urban region is accompanied by complex internal commuting patterns. Recently, Vasanen (2012) proposed a new approach for identifying the degree of functional polycentricity through the connectivity density of individual centers to the whole urban system. Zhong et al. (2014, 2017) proposed a trip-based centrality index for understanding spatial organization within the metropolitan area, studying the numbers of people attracted to different locations, but also the diversity of activities that the people were engaged in. Yue et al. (2019b) determined that the degree of polycentricity depends on the degree of connectivity between locations, measured using trip inflow/outflow data. In this sense, they established three flow- and network-based centrality metrics for measuring polycentric structure using travel data.

These studies concerned the distinguishing features and concepts between morphological and functional polycentricity and showed that these are rapidly evolving. Many scholars agree on moving away from the morphological idea as it can lead to the erroneous specification of polycentricity (Louail et al., 2015), since some spatial politics may be tied in with their morphology; for example, it may be the capital city of an autonomous region, or a special economic zone. Nevertheless, Kloosterman and Musterd (2001) pointed out that these two trains of thought are somehow coherent. Hoyler et al. (2008) called for the combining of morphological characteristics with functional relations into a single approach, in a systematic way, in order to address the multilayered nature of polycentricity. Green (2007) stated that functional relations within the morphologically polycentric system may not change synchronously. Both the morphology and function can be delineated by social network analysis techniques. Burger and Meijers (2012) developed their analysis to combine the morphological and functional aspects of polycentricity in a coherent manner. While these contributions have resolved important research problems, polycentricity has still not been rigorously defined with clear demarcation (analytical and prescriptive or normative approaches) (Burger and Meijers, 2012; Hoyler et al., 2008; Liu and Wang, 2016; Sarkar et al., 2018; Vasanen, 2012; Zhong et al., 2017).

The previous trip-based centrality index (Zhong et al., 2014, 2017) failed to provide a full picture of people’s daily spatial and temporal activity patterns with multi-purpose and scale, and the multifarious nature of spatial interactions that combine trips for shopping or leisure has rarely been systematically studied (Burger et al., 2014; Zhong et al., 2017). Louail et al. (2015) proposed a generic method to identify urban spatial structure using mobile phone data at a regional scale, and their results showed that cities differ by their proportion of two types of flows: integrated and random.

Polycentricity identification in China

Due to rapid urbanization, China’s megacities are struggling to cope with the pressures of population concentration. Wu and Phelps (2011) pointed out that unlike the suburbanization of North America, Chinese suburban development has also experienced three stages, moving from separated self-contained places to industrial relocation and housing programs and finally to the recent subcenter of polycentric development—also known as a post-suburb development. Liu and Wang (2016) investigated the polycentric degree of 318 cities in China, and they found that over 90% of Chinese cities have four or fewer intra-city centers, and that the pattern of GDP and population-based centers are associated with the city master plan. Lv et al. (2017) confirmed that the Beijing Metropolitan Area still has strong monocentric characteristics, with 70% of its employment distribution explained by the monocentric density function. Moreover, Wu (1998) stated that land-use change can be used as a prompt and reliable indicator of polycentric urban development in China, but that polycentric urban regions could be understood as being geographically separated but functionally integrated. Liu and Wang (2016), Liu (2019) and Liu et al. (2016) stated that the measurement of functional polycentricity is challenging, mainly due to the lack of suitable data in China. Zhang et al. (2016b, 2018) employed social media to understand the degree of regionalization in the Yangtze River Delta, focusing on the perspective of inter-city daily mobility through recorded Weibo position data. Li and Phelps (2016) utilized co-publication information to examine the knowledge links for the Yangtze River Delta region; they found that functional polycentricity decreases with increasing geographical scale.

In China, identification of the type and degree of polycentric development at intra-city level is the focus of concern for both academics and local municipal government, since the goal is to arrange residential employment and commuting patterns through urban structure optimization in order to balance economic development and urban livability. However, existing studies of urban polycentricity in China often focus on the morphological dimension due to a lack of information about intra-urban functional flows. For example, Zhang et al. (2019) examined employment centralization in Shanghai between 2008 and 2013, but they found that at the intra-city level, Shanghai is dispersed at the larger scale and polycentric at the smaller scale. Yue et al. (2019a) used remote sensing data of nighttime light and surface temperature to explore the dynamic of polycentricity urban development and the thermal environment in Hangzhou from 2000 to 2010.

Polycentricity can refer to the multi-nodal development of complicated human activity (Kloosterman and Musterd, 2001). According to Roth et al. (2011) and Zhong et al. (2017), the present measurement approaches for the degree of polycentricity are challenged by the highly diverse and intertwined ways in which people actually make use of urban space. Following this logic, Yue et al. (2019b) compared land development and functional linkage to better understand polycentric development in Shanghai. They extracted one day’s records of mobile phone calls to measure the linkage across spatial units, and found that morphological and functional polycentricity in Shanghai is mismatched. Wei et al. (2020) employed pick-up and drop-off information extracted from taxi GPS data to examine the polycentric structure of Shanghai. They confirmed that an obvious polycentric structure exists in Shanghai but that it is sensitive to scale effects.

On these premises, this paper proposes a new approach to achieve adequate measurement of the polycentric spatial structure of Shanghai at the intra-urban scale by using passive mobile phone data. We hypothesize that mobile phone data have natural advantages for the identification of the nodal features and actual usage of space with multiple centers, combining the morphological and functional aspects of polycentricity found in the extant literature. This data approach contains multiple attributes of urban structure; specifically, population and employment density can be identified from the user’s home and work location, and the diversity of their activities including the actual movements and recreational, shopping, and touristic behaviors can be distinguished spatially from people’s daily activity patterns and movements through the consideration of particular specifications in mobile positioning data (Raun et al., 2016; Shoval and Ahas, 2016; Xiao et al., 2019). More importantly, it provides a systematic way of telling the direction of flow by determining if a location is the destination or origin (Louail et al., 2015; Sarkar et al., 2018). Consequently, it is intuitively believed that the clustering of these activities with the actual usage of urban space may be key to gaining a comprehensive understanding of polycentricity.

Data and methodology

Study area

Shanghai is China’s financial center and one of the most internationally influential cities in China. By the end of 2018, Shanghai had a total permanent population of 24.24 million and a GDP of 3267.99 billion yuan, ranking first in China. Shanghai has 16 districts with 230 subdistricts—“Jiedao”—which are the smallest administrative units. It is noted that Wei et al. (2020) pointed out that both the size and shape of the analysis unit would affect the results of polycentric identification, with higher resolution bringing more accurate identification. This study takes the Jiedao (subdistrict) as the spatial analysis unit for policy implementation (Figure 1).

Figure 1.

Study area and Jiedao units.

Over the past decade, Shanghai has responded to the sprawl of its urban center by building nine new towns to accommodate industrial development and population growth; these are mainly distributed within 50 km of the central urban area of Shanghai. During the 11th Five Year Plan (2006–2010), the Shanghai government proposed an urban structure, named “1966,” which represented 1 central city, 9 satellite cities, 60 new towns, and 600 main residential areas. Among them, Songjiang, Jiading, Lingang, and other satellite cities have been identified as key locations in the planned transfer of industry and population from Shanghai’s center. The master plan of Shanghai (1990–2020) proved to be a watershed of polycentric development, with the “One City Nine Towns” program suggested to strengthen the “anti-magnetic force” of planned suburban new towns. Since then, polycentricity has been prioritized in Shanghai’s optimization agenda for spatial structures and new towns have focused more on an integrated development of housing estates and industrial zones (Deng and Liao, 2013). In the recent master plan (2017–2035), the government further proposed a multi-center urban system comprising “main center, subcenter, regional center, community center.” In addition to the planned new cities, Fengxian and Haibin are also included within the scope of subcenters. Due to preferential policies and the development of industrial areas, the suburbs have become the main areas of population growth. After 2000, the proportion of the population in the central urban area dropped sharply—in 2010 it dropped to 30.34%, and the proportion of population in suburban areas increased to 41.49%.

Methodology and mobile phone data

The traditional studies on human movement and activities have mainly relied on two major sources: a survey from respondents in the form of travel diaries (Jones and Pebley, 2014; Wong and Shaw, 2011; Zhong et al., 2017), and GPS tracking data (Shen et al., 2013). With the increasing pervasiveness of new technologies, individual-level and activity travel data from sources such as location-based social media (Facebook, twitter etc.) and smart card data have become available and may provide new opportunities to move beyond traditional human activity measures. For example, Huang and Wong (2016) introduced an approach to determine users’ home and work locations via twitter data in Washington, DC. Roth et al. (2011) used the tube travel card to reveal the spatial organization of the city, and they found that London has a polycentric structure characterized by large flows organized around a limited number of activity centers.

Recently, numerous studies have noted that passive mobile phone data contain rich geo-coded trajectories of population movement over space and time, which have a high degree of temporal and spatial regularity; it is, thereby, possible to identify people’s daily mobility and activities (Gonzalez et al., 2008; Xiao et al., 2019). Grinberger and Shoval (2015) adopt this logic, identifying multiple and dynamic centers of consumption via mobile phone records. Many urban applications of mobile phone data (especially within the Chinese context) rely on aggregated counts (Roth et al., 2011; Zhong, 2014; Zhong et al., 2014) and do not fully realize the potential of mobile phone data for characterizing urban dynamics at fine spatial and temporal scales. Yang et al. (2018) explored the spatial structure characteristics generated by human commutes to work by using one workday’s mobile phone data in Shenzhen. Their results found some inconsistencies between the detected subcenters and the areas proposed by urban planning.

The mobile phone data used in this study have been collected by China Mobile, in Shanghai. The time period for acquiring the mobile phone data in this study is from 15 March 2014 to 28 March 2014 (10 workdays and 3 weekend days), with an average of 6–8 million mobile phone signal records per day. Once the user’s phone is powered on, their position is traced. There were 13,154,173 unique residences identified from mobile phone data in Shanghai, from 8,311,791 commuters. For most big data research, data representation is a principal issue and so we need to test the validation of our data set. We checked the identified results with the census information, and the correlation result is above 0.75 (Figure 2), which indicates that our mobile phone data are valid for representing the population pattern of Shanghai. It is noted that the mobile phone data record the signal generated by the contact between the mobile phone and the mobile communication base station, and determine the time–space position through the base station location. There are 36,000 mobile base stations in Shanghai, which are the means of sending and receiving mobile phone signal.

Figure 2.

Correlation test between population from census and population from mobile phone identification.

Analysis framework

In general, the approach taken involved three main steps. The first step was to identify daily commuting activities via the mobile phone data. We then constructed nine indicators to reflect how people actually use the space from three dimensions: density, travel cost, and function. The third step was to employ principal component analysis (PCA) and K-means clustering to examine the degree of polycentricity.

Identifying mobile phone users’ trajectories

We followed the approach of Ahas and Mark (2005) and Xiao et al. (2019) to identify the mobile phone users’ daily behaviors. Since people’s daily movements and activities are rhythmic—such as sleeping and working—these activities could be identified by a threshold time. For example, we accumulated the duration of the “stay location” (a place where a user visits) from the user’s continuous trajectory during the night (19:00–08:00). The user’s “night destination” is the stay location with a maximum duration exceeding 120 min. We determined that the user’s “residence” should be the same daily night destination for more than nine days in two weeks. We used the same logic to identify people’s “workplace” by amending the time period (08:00–18:00), and other daily activities.

Indicators for polycentricity measures

We took account of three patterns of commuting trips: inflow, outflow, and local trips, since people’s movement trajectory may be associated with the multi-scalar nature of polycentricity. We intuitively believe that this information could indicate the balance of workplace and housing in a certain area, and that those Jiedao with a high proportion of workplaces would be associated with a high degree of local commuting.

Furthermore, we segmented three different types of commuting activities with three dimensions: density, distance, and proportion (Table 1). It is noted that the density reflects the absolute and relative concentration of jobs, the distance may reflect the socioeconomic status of commuters, and the proportion ratio may reflect the level of land-use mix.

Table 1.

Descriptive statistics of nine indicators of commuting pattern.

Indicator	Equation	Mean	Std Dev	Min	Max
Density (person/m²)
Density of inflow commuters	$D_{I n - C} = \frac{{NUM}_{I n - C}}{AR}$	2,180.73	3,012.76	1.16	13,893.22
Density of outflow commuters	$D_{Out - C} = \frac{{NUM}_{Out - C}}{AR}$	1,798.24	1,985.30	1.26	7,967.49
Density of local commuters	$D_{Loc - C} = \frac{{NUM}_{Loc - C}}{AR}$	617.94	533.82	0.00	2,969.33
Distance (km)
Average distance of inflow commuters	${DIS}_{I n - C} = \frac{\sum_{i = 1}^{\infty} {DIS}_{I n - i}}{{NUM}_{I n - C}}$	7.29	2.41	3.10	18.20
Average distance of outflow commuters	${DIS}_{Out - C} = \frac{\sum_{i = 1}^{\infty} {DIS}_{Out - i}}{{NUM}_{Out - C}}$	6.76	1.86	2.96	15.58
Average distance of local commuters	${DIS}_{Loc - C} = \frac{\sum_{i = 1}^{\infty} {DIS}_{Loc - i}}{{NUM}_{Loc - C}}$	1.31	0.66	0.00	4.08
Land-use mix
Percentage of inflow commuters	$P_{I n - H} = \frac{{NUM}_{I n - C}}{{NUM}_{H}}$	0.31	0.38	0.01	3.28
Percentage of outflow commuters	$P_{Out - H} = \frac{{NUM}_{Out - C}}{{NUM}_{H}}$	0.22	0.07	0.01	0.52
Percentage of local commuters	$P_{Loc - H} = \frac{{NUM}_{Loc - C}}{{NUM}_{H}}$	0.15	0.10	0.00	0.41

Empirical results

Commuting pattern of Shanghai

Figure 3 shows a map of Shanghai’s commuting network patterns, where red and green indicate the strength of the commuting connection, and the arrow shows the direction of the flow. There are several interesting results in this map; for instance, all hotspots for commuting inflows are outside the central urban area, in line with the obvious characteristics of a subcenter for employment in the suburbs. When we further investigate these hotspots, it is found that all hot spots for commuting are in the location of Shanghai new towns and that, among the nine new towns, the attraction effects of Minhang new town, Songjiang new town, and Nanqiao new town are the strongest. Of interest is that, for the mobile phone trajectories shown for each new town, its workers are mainly coming from the surrounding areas rather than the central city. At the same time, there is also a very complicated commuting network structure among the new towns, indicating that there are also certain functional connections between the new towns. Indeed, compared with the new towns, the commuting network within the urban central area presents a more complex and random network structure, with no obvious inflow hotspots, whereas with suburban commuting behavior, commuting distances in the central urban area are relatively short. Shanghai experienced a prominent and strategic change in urban spatial structure through its “One City Nine Towns” development plan, launched in 2001, which addressed the structural problem of overcrowding in the old central city. Based on the results of the commuting network, Shanghai’s polycentric spatial development policy seems to have worked well.

Figure 3.

Commuting patterns in Shanghai.

As mentioned above, the traditional polycentric identifications based on the commuter network are constrained by the data itself, which cannot completely capture the direction of the flow and the multiple purposes of the travel (Zhong et al., 2017). Therefore, we mapped nine cases with different commuting patterns in three dimensions (Figure 4). Regarding the density of the three commuting patterns, we were surprised to find that there were a large number of externally employed people in the central area of the city. At the same time, the outflow areas and inflow areas were highly consistent, indicating that there was an employment and residence mismatch in Shanghai. The density distribution of local commuters is relatively balanced, with no obvious high-value area compared with inflow commuters and outflow commuters.

Figure 4.

Spatial distribution of nine indicators.

The mapping of average distance patterns of the three commuter types are provided in Figure 4. As expected, many people living in the suburbs commute across districts, with longer commuting distances, whereas people living in the central city have shorter commuting distances. People face a trade-off between higher housing prices and commute times (Holzer, 1991). Surprisingly, the longest average distance between the inflow areas was not in the central city. The results show that the employment locations of residents in the suburbs are not actually in the city center, but rather that suburban residents commute between suburban areas to their employment. This may be a result of the processes of gentrification in Shanghai (He and Wu, 2007), whereby the low-income population who used to live in the urban central area spread into urban suburb areas. Such gentrification has disrupted the original urban spatial structures and exacerbated the imbalance between employment and housing, especially for vulnerable groups.

The proportions of inflow, outflow, and local commuting may reflect the level of land-use mix. It is assumed that the proportion of outflow will be higher in areas dominated by residential land use, while the proportion of inflow will be higher in areas dominated by business center/workplace land use. The results (Figure 4) show that the central urban area and the suburbs have opposing workplace–housing relationships. Specifically, the areas with a high proportion of inflows and outflows within the city’s central urban area are more obviously separated. However, outside the outer ring, and especially in the new town regions, it is found that the areas with a higher proportion of inflows that are always associated with a high proportion of outflows is high, which may indicate an issue with poor self-containment in Shanghai’s new town. Furthermore, it seems very interesting that we found that the proportion of employment in suburban areas, especially for non-new town areas, is generally higher—because most of the areas are rural areas, and most of the residents are employed locally. Table 1 provides statistic descriptions of nine indexes at Jiedao level. It is found that the average commuting distance from the inflow commuting patterns and outflow commuting patterns are much higher than those of local commuting. The average commuting distance in Shanghai is 8 km, far lower than that of London, Paris, and Tokyo (Frost et al., 1998; Kawabata and Shen, 2006), which indicates that most commuters in Shanghai choose employment close to their residence.

PCA analysis

Table 2 shows the PCA results. Among all nine components, the eigenvalue of three factors are greater than one, accounting for 79.62% of the total combined variance. In this vein, we keep these three factors, and rename them Component I, Component II, and Component III, respectively (Table 3).

Table 2.

The results of PCA of commuting pattern index.

	Component I	Component II	Component III
Eigenvalue	4.3402	1.7022	1.1236
Difference	2.6380	0.5786	0.4329
Proportion	0.4822	0.1891	0.1248
Cumulative	0.4822	0.6714	0.7962
Density of inflow commuters	0.8092	–0.0276	0.4787
Density of outflow commuters	0.8943	–0.3079	–0.0307
Density of local commuters	0.8826	–0.0911	–0.1769
Average distance of inflow commuters	–0.2794	0.8091	0.2092
Average distance of outflow commuters	–0.0871	0.4406	–0.6245
Average distance of local commuters	–0.7525	0.4152	–0.0711
Percentage of inflow commuters	–0.0088	–0.0393	0.9231
Percentage of outflow commuters	0.0823	–0.8194	0.3601
Percentage of local commuters	–0.5184	0.6929	–0.2111

LR test: independent vs. saturated: chi²(36) = 1367.50 Prob>chi² = 0.0000.

Table 3.

The results of K-means clustering.

Cluster	Subcenter I	Subcenter II	Suburban residences
Workplace–housing balanced	1.12(1.23)	0.16(0.89)	–0.63(0.43)
Residence	0.25(0.55)	–0.71(0.60)	0.82(0.86)
Workplace	1.67(1.46)	–0.26(0.55)	–0.29(0.51)

It is seen that Component I is associated with a high density of inflow commuters (0.81), high density of outflow commuters (0.89), high density of local commuters (0.88), and a short average distance of local commuters (–0.75). Component I meets most commuters’ needs for place of residence and workplace, since they have strong attraction. In this sense, we named Component I a “mixed region” type (Jiedao), which has both functions of work and residence. Some studies have found that mixed land-uses are related to commuting patterns (Van Acker and Witlox, 2011). Such mixed regions account for the highest proportion of the total, 48%.

The second level is Component II, which accounts for 19% of regions. These areas are associated with a small percentage of outflow commuters (–0.82), large percentage of local commuters (0.69), and long average distance of inflow commuters (0.81). This information indicates that the accessibility of these Jiedaos is poor and the location of the Jiedaos is relatively remote—there is a relatively small number of outflow commuters, with local commuters more prevalent in these Jiedaos. In this sense, we may refer to this type of region as a “residence region.”

Finally, we found that there were fewer Component III Jiedaos, but that they were distinctive, with a large percentage of inflow commuters (0.92) and small average distance of outflow commuters (–0.62). Inflow dominates outflow for commuting and so we named this type of region an “employment region.”

Clustering analysis

The final step for identifying polycentric structure is to use k-means cluster estimation. The clustering results are provided in Table 3 and Figure 5. In general, there are three clusters: the first cluster has a total of 32 Jiedaos, accounting for 14.16% of the total number of Jiedaos, and is based on Component III in the PCA analysis. We named this cluster “employment center,” since these areas mainly contain employment function, and are distributed in the old urban areas as well as the bounded areas shown in Figure 5 and export processing zones in the Pudong new area. The old urban area is the traditional commercial employment center of Shanghai. The bonded zone and export processing zone of the Pudong new area are the new employment centers, which have developed in recent years. It is unexpected that the Lujiazui Business Center would be excluded from this cluster; this may be due to the modifiable area unit problem (MUAP), which is encountered when studying aggregated spatial units. Lujiazui Business Center has the highest value of inflow commuting, with the other Jiedao all being residential developments with fewer employment opportunities. In this sense, when we group them, the inter-group variance does not account for the extra variance observed. The identified “employment centers” also include many industrial areas in the suburbs, such as the industrial zone of petrochemicals and fine chemicals manufacturing in the Jinshan district and the industrial zone of shipbuilding in Changxing island. Although they are not traditional business centers, they are identified by our mobile phone data, as they provide a large number of manufacturing jobs.

Figure 5.

Spatial pattern of cluster analysis.

The second cluster is based on Component I, which has a total of 109 Jiedaos, accounting for 48.23% of the total number of Jiedaos and 21.62% of the total urban area of Shanghai. As discussed above, this type of area contains both residential and employment functions, but does not imply a perfect balance of workplace and housing. Mixed regions are mainly distributed in the central urban areas and new towns. We consider the mixed areas of the suburbs as the subcenters, since people may live close to employment subcenters (Cervero and Wu, 1997; Giuliano and Small, 1991). The rest of the regions belong to the third cluster, which is mainly residential areas, accounting for 37.61% of the total number of Jiedaos and 76.17% of the total urban area of Shanghai. The residential regions are dominated by residential functions and are mainly distributed outside of the central urban area.

Figure 4 shows that regardless of whether it is an employment center or a mixed subcenter, Shanghai displays a polycentric spatial structure. Wu (1998) stated that land-use change can be used as a quick and reliable indicator of polycentric urban development in China, and Yue et al. (2019a) built on this point, using land-use patterns to identify the degree of morphological polycentricity in Shanghai. We were very surprised to find that the polycentric results based on land and activity were almost identical; an example of this was seen in the subcenter of petrochemicals and fine chemicals manufacturing in Jinshan, and the subcenter of equipment manufacturing in Linggang. This strongly suggests that the flow activity derived from mobile phone data can be complexly associated with the nodal features of each subcenter.

Discussion and conclusion

This study builds upon recent interest in the identification of polycentric urban structure, attempting to examine Shanghai’s degree of polycentricity at the intra-urban level using passive mobile phone data. Our approach adopted Sarkar et al. (2018)’s method, accounting for both inflow and outflow of each region to establish absolute and relative concentration of jobs, and constructed nine indicators to capture the multi-nodal connectivity features of people’s actual usage of space from the dimensions of employment density, commuting costs, and land-use mix. In general, the results of clustering estimation show a clearly apparent polycentric structure in Shanghai, especially in urban suburb areas. These results meaningfully complement the existing literature in three respects.

Firstly, this research provides a new analytical framework for using mobile phone data that can reflect the multi-scalar nature of polycentricity. We identified two types of subcenter: the employment subcenter and the self-sufficient subcenter. Ducruet and Beauguitte (2014) and Poorthuis and Van Meeteren (2019) emphasized that each node may take up a different position in each layer or level of the network. Human daily activities occurred at multi-scalar; therefore, it would help to understand how multi-level urban systems are neatly nested. Thus, the proposed framework could also be extended to any urban–regional level, such as a regional scale (Lee et al., 2018; Louail et al., 2015), or 300 × 300 urban grid (Wei et al., 2020).

Secondly, it is surprising to find that the morphological and functional features of polycentricity in Shanghai are perfectly matched. Our results align with Zhang et al. (2019), who followed a traditional framework using employment density data to identify the degree of morphological polycentricity in Shanghai. All subcenters they identified are included in our results. This provides evidence to support the idea that morphological and functional polycentricity may coexist (Burger and Meijers, 2012), which indicates that the intensity and flow of people’s movement could capture—entirely or in part—the features of both the nodality and centrality of subcenters. Our results imply that morphological and functional polycentricity could be developed synchronously with an urban plan. It also noted that our finding is inconsistent with the work from Yue et al. (2019b), who found that morphological and functional polycentricity in Shanghai is mismatched. One of the potential reasons for this is different measurement of regional centrality linkages.

Thirdly, it is interesting that the majority of subcenters in the emerging suburban areas are adherent to Shanghai’s new town developments launched in 2001 (One City Nine Towns Development Plan). The form and dynamics of China’s suburban development is by no means the same as that of the classic post-war suburbanization in the West. Many scholars pointed out that new town development of most East Asian cities can be treated as planned suburbanization with a logic of mixing together market and state authority, with the municipal government guiding the polycentric development through drafting and revising master plans and establishing development zones (Wu and Phelps, 2008; Yue et al., 2010). As stated by Wu and Phelps (2011), China’s institutional advantage lies in state entrepreneurialism, which can effectively allocate industries and promote the balance of employment and housing in the new towns. Liu and Liu (2018) found that only the planned subcenters in Beijing have higher employment growth and a stronger influence on employment and population distribution than those in Shanghai. In this sense, our results may imply the success of “planned polycentricity” in Shanghai, suggesting that new towns planned with industrial parks can provide more employment opportunities and have higher self-containment for commuting, which could be a policy tool for the polycentric development of another place or city.

The growing availability of new data could shed some light on the understanding of the polycentricity concept at multiple scales (Zhong et al., 2017). The urban system has become more complex than ever, and the mobile phone approach opened a research agenda for identification of polycentricity, since it could produce a full spatiotemporal picture of people’s daily lives (Xiao et al., 2019); however, the application of mobile phone data to this subject is still in its development phase, since the MUAP problem is one of limit for this research. Wei et al. (2020) pointed out that both the size and shape of the analysis unit would affect the results of polycentric identifications, with higher resolutions bringing more accurate identification. China’s mobile phone service is provided by China Mobile Communications Group, China Unicom and China Telecom. This paper uses the mobile phone signaling data of China Mobile. Moreover, it is intuitively believed that major users with multiple mobile phones will not use the same mobile phone operator, because there are differences in the package services of the three operators: one person owning more than one phone wants to enjoy different services. Furthermore, our approach failed to identify those people whose lifestyle pattern involves a reversal of day and night activities, and does not identify those people with more than one mobile phone.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by The National Social Science Fund of China [No. 19BSH035], and The Natural Science Foundation of China (51878457, 52008281).

ORCID iD

Yang Xiao

Yang Xiao, Associate Professor, College of Architecture and Urban Planning, Tongji University, 1239 Siping Road, Shanghai 200092, China. Email: yxiao@tongji.edu.cn

Yao Wang, Research Fellow, College of Architecture and Urban Planning, Suzhou University of Science and Technology, 1701, Binghe Road, Suzhou, Jiangsu 215009, China. Email: 245540789@qq.com

Siyu Miao, PhD Student, College of Architecture and Urban Planning, Tongji University, 1239 Siping Road, Shanghai 200092, China. Email: mcr@tongji.edu.cn

Xinyi Niu, Professor, College of Architecture and Urban Planning, Key Laboratory of Ecology and Energy‑Saving Study of Dense Habitat, Ministry of Education, Tongji University, Shanghai 200092, China. Email: niuxinyi@tongji.edu.cn

References

Aguilera

(2005) Growth in commuting distances in French polycentric metropolitan areas: Paris, Lyon and Marseille. Urban Studies 42: 1537–1547.

Ahas

Mark

(2005) Location based services—New challenges for planning and public administration? Futures 37: 547–561.

Anas

Arnott

Small

(1998) Urban spatial structure. Journal of Economic Literature 36: 1426–1464.

Batty

(2001) Exploring isovist fields: Space and shape in architectural and urban morphology. Environment and Planning B: Planning and Design 28: 123–150.

Burger

Meijers

(2012) Form follows function? Linking morphological and functional polycentricity. Urban Studies 49: 1127–1149.

Burger

Van Der Knaap

Wall

(2014) Polycentricity and the multiplexity of urban networks. European Planning Studies 22: 816–840.

Burger

Van Oort

Wall

, et al. (2013) Analysing the competitive advantage of cities in the Dutch Randstad by urban market overlap. In: Klaesson J, Johansson B and Karlsson C (eds) Metropolitan Regions. Berlin: Springer, pp.375–391.

Cervero

K-L

(1997) Polycentrism, commuting, and residential location in the San Francisco Bay area. Environment & Planning A 29: 865–886.

Cervero

K-L

(1998) Sub-centring and commuting: Evidence from the San Francisco Bay area, 1980-90. Urban Studies 35: 1059–1076.

10.

Davoudi

(2003) European briefing: Polycentricity in European spatial planning: From an analytical tool to a normative agenda. European Planning Studies 11: 979–999.

11.

De Goei

Burger

Van

Oort FG

, et al. (2010) Functional polycentrism and urban network development in the Greater South East, United Kingdom: Evidence from commuting patterns, 1981–2001. Regional Studies 44: 1149–1170.

12.

Deng

Liao

(2013) Transformation of City Space: From Single-Center to Multi-Center. Shanghai: Shanghai People's Publishing House (in Chinese).

13.

Ducruet

Beauguitte

(2014) Spatial science and network science: Review and outcomes of a complex relationship. Networks and Spatial Economics 14: 297–316.

14.

Frost

Linneker

Spence

(1998) Excess or wasteful commuting in a selection of British cities. Transportation Research Part A: Policy and Practice 32: 529–538.

15.

Giuliano

Small

(1991) Subcenters in the Los Angeles region. Regional Science and Urban Economics 21: 163–182.

16.

Gonzalez

Hidalgo

Barabasi

A-L

(2008) Understanding individual human mobility patterns. Nature 453: 779.

17.

Green

(2007) Functional polycentricity: A formal definition in terms of social network analysis. Urban Studies 44: 2077–2103.

18.

Grinberger

Shoval

(2015) A temporal-contextual analysis of urban dynamics using location-based data. International Journal of Geographical Information Science 29: 1969–1987.

19.

Hall

Tewdwr-Jones

(2010) Urban and Regional Planning. Abingdon: Routledge.

20.

Hall

Pain

(2006) The Polycentric Metropolis: Learning from Mega-City Regions in Europe. Abingdon: Routledge.

21.

(2007) Socio-spatial impacts of property-led redevelopment on China’s urban neighbourhoods. Cities 24: 194–208.

22.

Holzer

(1991) The spatial mismatch hypothesis: What has the evidence shown? Urban Studies 28: 105–122.

23.

Horton

Reynolds

(1971) Effects of urban spatial structure on individual behavior. Economic Geography 47: 36–48.

24.

Hoyler

Kloosterman

Sokol

(2008) Polycentric puzzles–Emerging mega-city regions seen through the lens of advanced producer services. Regional Studies 42: 1055–1064.

25.

Yang

, et al. (2019) Urban spatial structure and travel in China. Journal of Planning Literature. Epub ahead of print 5 June 2019. DOI: 10.1177/0885412219853259.

26.

Huang

Wong

(2016) Activity patterns, socioeconomic status and urban spatial structure: What can social media data tell us? International Journal of Geographical Information Science 30: 1873–1898.

27.

Jones

Pebley

(2014) Redefining neighborhoods using common destinations: Social characteristics of activity spaces and home census tracts compared. Demography 51: 727–752.

28.

Kawabata

Shen

(2006) Job accessibility as an indicator of auto-oriented urban structure: A comparison of Boston and Los Angeles with Tokyo. Environment and Planning B: Planning and Design 33: 115–130.

29.

Kloosterman

Musterd

(2001) The polycentric urban region: Towards a research agenda. Urban Studies 38: 623–633.

30.

Lambooy

(1998) Polynucleation and economic development: The Randstad. European Planning Studies 6: 457–466.

31.

Lee

K-S

You

Eom

, et al. (2018) Urban spatiotemporal analysis using mobile phone data: Case study of medium-and large-sized Korean cities. Habitat International 73: 6–15.

32.

Liu

(2018) How did urban polycentricity and dispersion affect economic productivity? A case study of 306 Chinese cities. Landscape and Urban Planning 173: 51–59.

33.

Phelps

(2016) Megalopolis unbound: Knowledge collaboration and functional polycentricity within and beyond the Yangtze River Delta Region in China, 2014. Urban Studies 55: 443–460.

34.

Liu

(2019) Characterizing broken links on national and local expressways in Chinese city-regions. Regional Studies 53(8): 1137–1148.

35.

Liu

Derudder

(2016) Measuring polycentric urban development in China: An intercity transportation network perspective. Regional Studies 50(8): 1302–1315.

36.

Liu

Wang

(2016) How polycentric is urban China and why? A case study of 318 cities. Landscape and Urban Planning 151: 10–20.

37.

Liu Z Liu S (2018) Polycentric development and the role of urban polycentric planning in China’s mega cities: An examination of Beijing’s metropolitan area. Sustainability 10(5): 1588.

38.

Louail

Lenormand

Picornell

, et al. (2015) Uncovering the spatial structure of mobility networks. Nature Communications 6: 6007.

39.

Zheng

Zhou

, et al. (2017) Decentralization and polycentricity: Spatial changes of employment in Beijing metropolitan area. Sustainability 9(10): 1880.

40.

Markusen

(2003) Fuzzy concepts, scanty evidence, policy distance: The case for rigour and policy relevance in critical regional studies. Regional Studies 37: 701–717.

41.

Mcmillen

(2001) Nonparametric employment subcenter identification. Journal of Urban Economics 50: 448–473.

42.

Parr

(2004) The polycentric urban region: A closer inspection. Regional Studies 38: 231–240.

43.

Poorthuis

Van Meeteren

(2019) Containment and connectivity in Dutch urban systems: A network-analytical operationalization of the three-systems model. Journal of Economic and Social Geography. Epub ahead of print 19 November 2019. DOI: 10.1111/tesg.12391.

44.

Raun

Ahas

Tiru

(2016) Measuring tourism destinations using mobile tracking data. Tourism Management 57: 202–212.

45.

Redfearn

(2007) The topography of metropolitan employment: Identifying centers of employment in a polycentric urban area. Journal of Urban Economics 61: 519–541.

46.

Roth

Kang

Batty

, et al. (2011) Structure of urban movements: Polycentric activity and entangled hierarchical flows. PloS One 6: e15923.

47.

Sarkar

Levinson

(2018) Measuring polycentricity via network flows, spatial interaction and percolation. Urban Studies. Epub ahead of print 3 April 2019. DOI: 10.1177/0042098019832517.

48.

Schwanen

Dieleman

Dijst

(2004) The impact of metropolitan structure on commute behavior in The Netherlands: A multilevel approach. Growth and Change 35: 304–333.

49.

Shen

Kwan

M-P

Chai

(2013) Investigating commuting flexibility with GPS data and 3D geovisualization: A case study of Beijing. Journal of Transport Geography 32: 1–11.

50.

Shoval

Ahas

(2016) The use of tracking technologies in tourism research: The first decade. Tourism Geographies 18: 587–606.

51.

Sohn

(2005) Are commuting patterns a good indicator of urban spatial structure? Journal of Transport Geography 13: 306–317.

52.

Suarez-Villa

Walrod

(1997) Operational strategy, R&D and intra-metropolitan clustering in a polycentric structure: The advanced electronics industries of the Los Angeles basin. Urban Studies 34: 1343–1380.

53.

Taylor

Hoyler

Verbruggen

(2010) External urban relational process: Introducing Central flow theory to complement central place theory. Urban Studies 47: 2803–2818.

54.

Van Acker

Witlox

(2011) Commuting trips within tours: How is commuting related to land use? Transportation 38(3): 465–486.

55.

Van der Laan

(1998) Changing urban systems: An empirical analysis at two spatial levels. Regional Studies 32: 235–247.

56.

Van Meeteren

Poorthuis

Derudder

, et al. (2016) Pacifying Babel’s Tower: A scientometric analysis of polycentricity in urban research. Urban Studies 53: 1278–1298.

57.

Vasanen

(2012) Functional polycentricity: Examining metropolitan spatial structure through the connectivity of urban sub-centres. Urban Studies 49: 3627–3644.

58.

Veneri

Burgalassi

(2012) Questioning polycentric development and its effects. Issues of definition and measurement for the Italian NUTS-2 regions. European Planning Studies 20: 1017–1037.

59.

Waterhout

Zonneveld

Meijers

(2005) Polycentric development policies in Europe: Overview and debate. Built Environment 31(2): 163–173.

60.

Wei

Luo

Wang

, et al. (2020) Multiscale identification of urban functional polycentricity for planning implications: An integrated approach using geo-big transport data and complex network modeling. Habitat International 97: 102134.

61.

Wong

Shaw

S-L

(2011) Measuring segregation: An activity space approach. Journal of Geographical Systems 13: 127–145.

62.

(1998) Polycentric urban development and land-use change in a transitional economy: The case of Guangzhou. Environment and Planning A: Economy and Space 30: 1077–1100.

63.

Phelps

(2008) From suburbia to post-suburbia in China? Aspects of the transformation of the Beijing and Shanghai global city regions. Built Environment 34: 464–481.

64.

Phelps

(2011) (Post) suburban development and state entrepreneurialism in Beijing's outer suburbs. Environment and Planning A: Economy and Space 43: 410–430.

65.

Xiao

Wang

Fang

(2019) Exploring the disparities in park access through mobile phone data: Evidence from Shanghai, China. Landscape and Urban Planning 181: 80–91.

66.

Yang

Fang

Yin

, et al. (2018) Understanding the spatial structure of urban commuting using mobile phone location data: A case study of Shenzhen. Sustainability 10: 1435.

67.

Yue

Liu

Fan

(2010) Polycentric urban development: The case of Hangzhou. Environment and Planning A: Economy and Space 42: 563–577.

68.

Yue

Qiu

, et al. (2019a) Polycentric urban development and urban thermal environment: A case of Hangzhou. Landscape and Urban Planning 189: 58–70.

69.

Yue

Wang

Liu

, et al. (2019b) Mismatch of morphological and functional polycentricity in Chinese cities: An evidence from land development and functional linkage. Land Use Policy 88: 104176.

70.

Zhang

Qiu

, et al. (2018) Intensification of hydrological drought due to human activity in the Middle reaches of the Yangtze River, China. Science of the Total Environment 637: 1432–1442.

71.

Zhang

Sun

Li W , et al. (2019) Polycentricity or dispersal? The spatial transformation of metropolitan Shanghai. Cities 95: 102352.

72.

Zhang

Derudder

Wang

, et al. (2016a) Using location-based social media to chart the patterns of people moving between cities: The case of Weibo-users in the Yangtze River Delta. Journal of Urban Technology 23: 91–111.

73.

Zhang

Jin

Stuecker

, et al. (2016b) Unraveling El Niño's impact on the East Asian monsoon and Yangtze River summer flooding. Geophysical Research Letters 43:11375–11382.

74.

Zhong

(2014) Urban Transformation towards Polycentricity: Detecting Functional Urban Changes in Singapore from Transportation Data. Zurich: ETH Zurich.

75.

Zhong

Arisona

Huang

, et al. (2014) Detecting the dynamics of urban structure through spatial network analysis. International Journal of Geographical Information Science 28: 2178–2199.

76.

Zhong

Schläpfer

Müller Arisona

, et al. (2017) Revealing centrality in the spatial structure of cities from human activity patterns. Urban Studies 54: 437–455.

77.

Zhou

, Yeh AGO, Li W, et al. (2018) A Commuting Spectrum Analysis of the Jobs–housing Balance and Self-Containment of Employment with Mobile Phone Location Big Data. Environment and Planning B: Urban Analytics and City Science 45: 434–451.