The spatial distribution pattern of enterprises in Beijing and its influencing factors analysis based on POI data

Abstract

Different urban elements may exhibit various aggregation patterns. It is of great significance to quantitatively investigate the disparity and connection among various aggregation patterns of urban elements for understanding the mechanism of urban development and supporting urban planning. In this paper, the point of interest (POI) of Beijing is taken as an example, and the distribution pattern and the level of agglomeration of POI in different industries are analyzed by kernel density estimation (KDE). The study found that the distribution density of POI in various industries in Beijing showed a trend of “higher in the eastern part and lower in the western part” and gradually decreased from the center to the outer. The aggregation of other industries’ POI, which is centered on enterprise POI, is analyzed by k-nearest method. The results show that the retail industry, bus station, and catering service industry are in a relatively concentrated distribution around the enterprise POI, and other urban elements are rarely distributing. In addition, this paper analyzes the kernel density chart by the vector analysis theory on landscape pattern, and how the spatial distribution pattern of enterprises is revealed by using the perspective of classical mechanics. It can be concluded that the formation of this kind of distribution comes from the centripetal force of the various industries and the axial traction of the northwest–southeast to the traffic trunk. Overall, the results of enterprise distribution analysis based on the POI data can explain part of the difference in business activities and economy distribution within urban areas. The study results of enterprise activities are also conducive to the strategy-making process of both governments and enterprises.

Keywords

Beijing enterprise POI kernel density estimation k-nearest method POI vector analysis theory on landscape pattern

Introduction

The spatial agglomeration characteristic is regarded as the objective trend of modern enterprise development (Higano, 2011; Fang et al., 2017). While industrialization tends to be the unification of total expansion and spatial aggregation in the developed countries, remarkable dispersion and layout characteristics have appeared in Chinese enterprises since the Reform and Opening-up until now. However, the initial state of distributed layout can be influenced along with the expansion of the enterprise scale and the change of external system, economy, resource and environment, and other factors, thus promoting the evolution of the enterprises to space agglomeration (Song Wei, 2010). Business activities of an enterprise are carried out in the market, which is also subject to the political, economic, technological and social culture of the country. Thus the analysis of the production and operation activities of enterprises should take these environmental factors into account. There is a close relationship between an enterprise and its environment: the environment lays the foundation of enterprise development on one hand; on the other hand, instead of being simply passively dominated by the environment, the enterprise (a kind of dynamic social organization) is adaptable to, and has inversely impact on, the environment (Gibb, 1997). This would promote the social progress and economic prosperity. Characteristics of enterprise distribution have an impact on the enterprises’ trading behavior, products, technical implementation, talent introduction, and other related important links.

The spatial analysis, which provides support to quantitatively study space patterns, is considered as an important means of spatial cognition and evaluation. It enables us to comprehensively understand the importance of spatial location and spatial interaction (Anselin, 1999). Much research has been conducted on the distribution characteristics of various industries (Jiaming, 2016), including office space (Zhang & Chen, 2011), production services (Han & Qin, 2009), automobile services (Feng et al., 2012), urban logistics enterprises (Fang & Chen, 2005) and retail industry (Cao, 2011). For example, the location characteristics of large commercial outlets in Changchun were explored by means of point pattern analysis, ordered polychotomous logistic regression and so on (Wang & Li, 2015). Taking Guangzhou city as an example, the hotspots of retail activities and inner aggregating distribution of retail formats were analyzed by using the method of KDE (Kernel Density Estimation) and Getis-Ord G* (Chen, W. et al., 2015). With five types of urban elements such as impervious surface, Point of Interest (POI) and so on, the city centers of Wuhan city, Central China, were identified by using the KDE method and Concentration Degree Index (CDI) to reflect the aggregation degree of urban elements (Chen Weishan et al., 2015).

In general, the analysis on spatial distribution and agglomeration characteristics of various industries, based on big data, has become a trend of quantitative research in various industries. However, the empirical research on the influence of external forces on the distribution of enterprises is less involved, in which the multi-industry co-distribution in the geographic space is considered. At present, there is an increasing public demand for spatial information. To enhance the level of urban geographic information service, the geographical information collection and public service in China are gradually transitioned from the government to the enterprise market. POI, as an important source of geographic information, records the location information and attribute information of urban public infrastructure (Zheng et al., 2014). At the same time, it brings about abundant information for recognizing urban development patterns, social-economic structure, social behavior mentality, environmental monitoring and indoor and outdoor navigation (McKenzie, G., 2014; Min et al., 2015). In terms of expressing interest points, the kernel density estimation (KDE) is better than other density expressions (e.g. quadrant density, Voronoi diagram density and so on), because of the location influence of Tobler’s First Law (Wang & Lee, 2015; Wenhao et al., 2015).

Therefore, by taking Beijing (a densely enterprise POI distributed city) as an example, the regional distribution characteristics of enterprises were studied in this paper. In the study, the KDE method was used to analyze the aggregation of various industries, and the congenerous distribution between enterprise POI and those of other urban services (e.g. retails, government offices) was analyzed by the k-nearest analysis. Moreover, the vector analysis theory on landscape pattern (VATLP) (Zhang, S., 2006) was first introduced to analyze the external force condition of the enterprises in the city, so as to better reveal the formation mechanism of the spatial distribution of the urban elements. This study can help to understand the clustering rules of urban enterprises and provide decision support for urban planning.

Methods

Kernel density estimation (KDE)

The kernel density estimation (KDE) method is used to analyze the spatial clustering characteristics of different industry elements. KDE method is widely used in spatial cluster analysis based on point position data. As a nonparametric method, KDE reflects the relative concentration of points’ distribution, which is calculated as follows:

f (x) = \frac{1}{n h} \sum_{i = 1}^{n} K (\frac{x - x_{i}}{h})

Here, $f (x)$ is the kernel function, $K (\cdot)$ is the kernel – a non-negative function that integrates to one, h is a smoothing parameter called the bandwidth. A kernel with subscript h is called the scaled kernel and defined as $K_{h} (x)$ = $1 / h \cdot k (x / h)$ . In this paper, the POI density is analyzed by the kernel density analysis module of Seaborn Library which is based on matplotlib in the Python development environment.

K-nearest neighbor relationship

The k-nearest neighbor relationship is a non-parametric method used for classification and regression (Altman, 1992; Bian, 2009), which is calculated by using the distance value measurement to find the frequent collection of adjacent spatial features.

K-nearest neighbor definition:

Let $R = {r_{1}, r_{2}, \dots r_{m}}$ be a set of m reference points in a d dimensional space, and let $Q = {q_{1}, q_{2}, \dots q_{m}}$ be a set of n query points in the same space. The k-nearest neighbor relationship consists in searching the k-nearest neighbors of each query point $q_{i} \in Q$ in the reference set R given a specific distance (Garcia et al., 2008).

In this paper, the k-nearest neighbor relationship is used to analyze the cooperative distribution relationship between the enterprise distribution and those of other industries. The proportion of other POIs around the enterprise POI is first measured, and the relationships of mutual interaction and mutual dependence between enterprises and other industries are then discussed (see Figure 1).

Figure 1.

The theory of kernel-nearest neighbor (KNN) in measuring the nearest points.

The vector analysis theory on landscape pattern (VALTP)

The vector analysis theory on landscape pattern (VATLP) is based on the quantitative description of plane graphics properties in classical mechanics theory, in which both the patch geometric characteristic and patch orientation are considered simultaneously. VATLP is briefly introduced as follows:

Figure 2 illustrates a patch (S), whose centroid is C. In the patch, there are two axes, namely the major principal axis and the minor principal axis, pass through the centroid. The major principal axis and the minor principal axis have the minimum moment of inertia (I_yy) and the maximum moment of inertia (I_xx), respectively. The angle (θ) between the major principal axis and the horizontal axis can be calculated as follows:

tan 2 θ = \frac{2 I_{x y}}{I_{x x} - I_{y y}}

where I_xy refers to the product of the inertia. The detailed calculation method of I_xx, I_yy and I_xy can be found in Zhang et al. (2006). The orientation of the major principal axis (i.e. θ) is then assigned as the patch orientation (PO). The landscape index of patch orientation (PO) can be used for quantifying landscape anisotropy and revealing the corresponding driving forces.

Figure 2.

A patch (S) with centroid C ( $\bar{x}, \bar{y}$ ), A is the area, dxdy is the differential area of point (x, y), xoy is the geographic coordinate system.

The interaction between internal clusters and external environmental factors (such as government macroeconomic regulation and control, the abundance of available resources, etc.) often results in directive distributions of urban POIs. In this paper, VATLP is thus utilized for analyzing the directionality of urban POI distribution.

Experiment results and analysis

Data source

This study is exemplified by using Beijing city, the capital and the political, economic and cultural center of China. The POI data of the city originated from the Baidu Map is used here, as illustrated by Figure 3. The POI data consists of 10 broad categories (BC POIs) such as government agencies, financial services, commercial buildings, hotels, entertainment, medical services, scientific research institutions, catering services, residential area, and enterprises, and 15 types of functional sites (denoted as FS POIs), such as bus station and railway station: a total of 15 types. Table 1 demonstrates the corresponding attribute name, feature ID and number of features in each type of BC or FS POIs. The data of POIs is also divided into POIs of enterprises or companies (denoted as enterprise POIs) and those for 24 other types (denoted as non-enterprise POIs).

Figure 3.

The urban POI distribution in Beijing.

Table 1.

The Beijing POI data set description.

Type name	Type identification	Feature number
Governmental agencies	‘7085’	10245
Airport	‘8100’	18
Railway station	‘8085’	1946
Car station	‘8083’	131
Bus station	‘8087’	40630
Gas station	‘4082’	1062
Parking lots	‘4102’	7277
Motorway service areas	‘8301’	1061
Toll-gate	‘8401’	367
Financial services	‘1199’	10721
Commercial buildings	‘1600’	3820
Retail industry	‘2003’	53697
Hotel	‘5380’	6440
Recreation and entertainment	‘6081’	20626
Medical service	‘7280’	11500
Research and educational
institutions	‘1701’	14158
Companies	‘1980’	43189
Park and square	‘7300’	950
Residential quarters	‘1900’	8932
Comprehensive service	‘7180’	24832
Food and beverages	‘1380’	33548
Automobile service	‘4500’	7305
Famous scenery	‘9080’	4398
Telecommunication services	‘1202’	1393
Comfort station	‘7880’	8688

Results and analysis

The spatial distribution characteristics of enterprises in Beijing

Figure 4(a) and Figure 4(b) illustrate the POI distribution density and the corresponding results of kernel density estimation (KDE). From the figures, it can be seen that the distribution density of enterprises gradually decreases from the inner to the outer. At the same time, the distribution density of the enterprises in Beijing tends to “higher in the eastern part and lower in the western part”. Two peaks, varied significantly in values, appear in the density curve like the saddle. This coincides with the previous research (Ma & Pei, 2010) that, with an obvious wide range, the employment of the eastern region is higher than that of the west. In the western region, except Zhongguancun (ZOL) and the Financial Street high-end functional area, the density level is lower overall.

Figure 4.

The kernel density estimation of enterprise POI.

Co-distribution between enterprise POIs and non-enterprise POIs

Figure 5 demonstrates the distribution characteristics of the nearest 24 non-enterprise POIs adjacent to an enterprise POI. From the figure it can be seen that the retail industry, bus stations, and catering services are distributed with a relative high proportion, up to 14.4%, 14.1%, and 11.7%, respectively. The high-density distribution of the retail industry may result from the formation of the direct supply chain formed by producers and the sale of goods to the residents (the living consumption) or social groups (the public consumption). The dense distribution of bus stations guarantees the circulation of business and the stable output of production. At the same time, the dense distribution of enterprises enables urban populations to work in a relatively concentrated place. In addition, the large population movement can also promote the development of the catering industry, which gives rise to the intensively distributed catering industry around the enterprise.

Figure 5.

Co-distribution between enterprise POIs and non-enterprise POIs.

According to Tobler’s First Law, there are certain interactions between the neighboring individuals or groups, and the influence of various space activities and the intergroup restraint tends to decrease along with the spatial distance. For enterprises, there is an inherent demand to save working costs in various economic activities, thus causing the social resources and elements to operate on the principle of proximity. Moreover, due to the limitation of spatial perception, the surrounding or nearby areas usually take priority when achieving the related decision-making information. The results and analysis of this paper are, therefore, consistent with the above-mentioned theories.

Force analysis of enterprise distribution in Beijing based on the VALTP

To better understand the distribution of enterprises and the corresponding driving forces, the spatial distribution of other broad categories of POIs (BC POIs) was analyzed. As illustrated by Figure 6, most of the other nine BC POIs (e.g. retail industry, hotels) demonstrate obvious centrality, that is, these POIs are mostly distributed in the central region. The agglomeration of each of these industries generally leads to an inverted “U” structure of its spatial distribution, as shown by the up and right curves in the figure. This means that the agglomeration effect of these industries is greater than the traction of external factors. The enterprises in the central region are therefore squeezed by other industries, leading to the “hollowing phenomenon” of the enterprise distribution.

Figure 6.

The results of kernel density estimation (KDE) about various industries.

The distribution of enterprises is influenced not only by driving forces interior to the city, but by those outside of the city, in particular the traffic trunk lines instead of the small traffic lines interior to Beijing. Therefore, the KDE chart of the enterprise POIs is overlapped with the expressway as well as railway map, and further analyzed on the basis of VATLP. As shown by Figure 7, the agglomeration of the enterprises is significantly affected by the axial traction of the Beijing-Tianjin Expressway (Figure 7(a)), the Beijing-Shanghai Expressway (Figure 7(a)), and the Beijing-Baotou Railway (Figure 7(b)). In fact, the influences of these three trunk lines on enterprise distribution are not identical. Due to the large tractive effect from the former two lines compared with the last one, a higher density of enterprise distribution along the southeast direction (i.e. directionality) could be found, as illustrated by Figure 3b. The combination of the “hollowing phenomenon” and the directionality gives rise to the final saddle-shape distribution of the enterprises, as shown by upper and right curves in Figure 3(b).

Figure 7.

The force analysis based on the VATLP.

Conclusions

In this paper, we analyzed the spatial distribution characteristics of enterprise POIs in Beijing and their influencing factors by using KDE, KNN and vector analysis theory of landscape pattern (VATLP). It can be found that the characteristics of enterprise distribution demonstrate both the “hollowing phenomenon” in the central region and the directionality along the southeast direction. The corresponding driving forces were further analyzed. We conclude that the enterprise distribution is influenced by the driving forces interior (i.e. the impact from other industries) to the city and those exterior (i.e. the major traffic lines) to the city. This study will allow us to better understand the development mechanism of enterprise spatial patterns.

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 research was jointly supported by the National Key Research and Development Program [Grant numbers: 2017YFB0503600].

ORCID iD

Peng Wu

References

Altman

N. S.

(1992). An introduction to kernel and nearest-neighbor nonparametric regression. American Statistician, 46(3), 175–185.

Anselin

(1999). The future of spatial analysis in the social sciences. Geographic Information Sciences, 5(2), 67–76.

Bian

(2009). A novel spatial co-location pattern mining algorithm based on k-nearest feature relationship. Geomatics & Information Science of Wuhan University, 34(3), 331–334.

Cao

W. D.

(2011). Spatial pattern and location evolution of urban logistics enterprises: Taking Suzhou as an example. Geographical Research, 30(11), 1997–2007.

Fang

X. Y.

Chen

Z. N.

(2005). Agglomerate types of retail business around urban subway station – taking Guangzhou as an example. Economic Geography, 25(4), 525–527.

Fang

Zhang

Wang

Zhang

(2017). Spatial agglomeration of exhibition enterprises on a regional scale in China. Chinese Geographical Science, 27(3), 497–506.

Feng

Yang

Y. U.

Xia

Lin

(2012). The spatial agglomeration characteristics of automotive service industry: A case study of Nanjing. Scientia Geographica Sinica, 32(10), 1200–1208.

Gao

Janowicz

Montello

D. R.

Yang

J. A.

Mckenzie

(2017). A data-synthesis-driven method for detecting and extracting vague cognitive regions. International Journal of Geographical Information Science, 31(6), 1245–1271.

Gao

Janowicz

Zhang

(2014). Constructing gazetteers from volunteered big geo-data based on Hadoop. Computers Environment & Urban Systems, 61, 172–186.

10.

Garcia

Debreuve

Barlaud

(2008). Fast k-nearest neighbor search using GPU. Computer Vision and Pattern Recognition Workshops, 2008. Piscataway, NJ: IEEE.

11.

Gibb

A. A.

(1997). Small firms’ training and competitiveness. Building upon the small business as a learning organization. International Small Business Journal, 15(3), 13–29.

12.

Han

S. S.

Qin

(2009). The spatial distribution of producer services in Shanghai. Urban Studies, 46(4), 877–896.

13.

Higano

(2011). The spatial economy: Cities, regions, and international trade. MIT Press Books, 86(1), 283–285.

14.

Jiaming

L. I.

Zhang

Yejin

L. I.

Yang

Jianhui

Y. U.

(2016). The characteristics of industrial agglomeration based on micro-geographic data. Geographical Research, 35(1), 95–107.

15.

Limin

Jiao

(2017). The characteristics and patterns of spatially aggregated elements in urban areas of Wuhan. Acta Geographica Sinica, 72(8): 1432–1443.

16.

Pei

(2010). Exploratory spatial data analysis of regional economic disparities in Beijing during 2001–2007. Available at: http://www.isprs.org/proceedings/XXXVIII/part2/Papers/52_Paper.pdf (accessed 3 January 2018).

17.

McKenzie

Janowicz

Gao

Yang

J. A.

(2015). POI pulse: A multi-granular, semantic signature-based information observatory for the interactive visualization of big geosocial data. Cartographica the International Journal for Geographic Information & Geovisualization, 50(2), 71–85.

18.

Min

Tinghua

A. I.

Wei

L. U.

Cheng

(2015). A real-time generalization and multi-scale visualization method for POI data in volunteered geographic information. Acta Geodaetica et Cartographica Sinica, 44(2), 228–234.

19.

Song

Wei

. (2010). Spatial Evolution of Enterprises. Social Sciences Academic Press.

20.

Wang

(2015). Analysis and visualization of POI distribution density based on urban network space. Urban Geotechnical Investigation & Surveying, (1), 21–25.

21.

Wang

Hao

Jiang

(2015). Locations and their determinants of large-scale commercial sites in Changchun, China. Acta Geographica Sinica, 70(6), 893–905.

22.

Chen

Liu

Liang

(2016). Retail center recognition and spatial aggregating feature analysis of retail formats in guangzhou based on poi data. Geographical Research, 35(4), 703–716.

23.

Wenhao

Y. U.

Tinghua

A. I.

Liu

Yakun

H. E.

(2015). Network kernel density estimation for the analysis of facility POI hotspots. Acta Geodaetica et Cartographica Sinica, 44(12), 1378–1383, 1400.

24.

Zhang

Chen

(2011). Industrial distribution and clusters of urban office space in Beijing. Acta Geographica Sinica, 66(10), 1299–1308.

25.

Zhang

Cropp

(2006). Vector analysis theory on landscape pattern (VATLP). Ecological Modelling, 193(3–4), 492–502.

26.

Zheng

Capra

Wolfson

Yang

(2014). Urban computing: Concepts, methodologies, and applications. ACM Transactions on Intelligent Systems & Technology, 5(3), 1–55.