Quantitative Analysis of Spatial Heterogeneity and Driving Forces of the Urban Spatial Structure’s Development Level Based on Multi-Source Big Data: A Case Study of Beijing, China

Abstract

Many developed cities in the world put forward a spatial development strategy in their construction planning. Analyzing the development level of the urban spatial structure and the influence of driving factors has become a hot topic. Based on 709,232 points of interest data in Beijing, China, this paper integrates nighttime light data and population density data to select eight key driving factors from three perspectives: urban function configuration, economic activity intensity and population spatial distribution. Geodetector is used to optimize data discreteness and highlight the spatial heterogeneity of the development level. We use the technique for order of preference by similarity to ideal solution (TOPSIS) model improved with the entropy weight method to grade the spatial differentiation characteristics of the comprehensive development level. The driving factors and their effects on space are further discussed using Geodetector. The results are as follows: (1) The quartile method can achieve the optimal dispersion of all urban functions. The standard deviation can achieve the optimal dispersion of economic activity intensity and population spatial distribution; (2) A comparison with the “Beijing Urban Master Plan (2016–2035)”, finds that the optimized evaluation system can effectively reflect the spatial heterogeneity of urban spatial structure development. It verifies the rationality of the evaluation methods and factors; (3) The driving force of the population under single-factor driving is 0.8428. The dual-factor driving force with population participation ranges from 0.8992 to 0.9550. The results of the study are significant and reflect the prominent role of population on the development level of spatial structure in Beijing. This paper aims to provide a new idea for the study of the interior space planning of large inland cities.

1. Introduction

The urban spatial structure, also known as the urban internal spatial structure [1], is the spatial projection of the urban economic and social structures [2]. As the spatial form of the existence and development of the urban social economy, it is generally manifested in urban density [3], urban morphology [4] and other forms. Since its reform and opening, China has accelerated the development process of urbanization. Rapid population growth has significantly changed the inner urban space and functions of many large cities [5]. However, the continuous expansion of urban regional land has aggravated the imbalance the level of urban development. It leads to the phenomenon of “big city disease” in many places [6] and has a serious impact on the city’s sustainable development. In recent years, with the transformation of urban development from a single-center structure to a multi-center structure, social problems such as pseudo-urbanization have gradually been alleviated. Studies on the development of the urban spatial structure have also attracted much attention and favor from scholars [7]. Therefore, studying the spatial structure and its internal driving factors in large cities helps urban planners understand urban characteristics and make informed urban management decisions. Currently, government decision makers in most large cities must consider long-term needs. These needs are related to multi-center development when they create urban development planning policies. For example, the “Beijing Urban Master Plan (2016–2035)” clearly emphasizes building a world-class urban agglomeration with the capital as the core. It aims to improve the urban system and establish an urban spatial structure of “one core, one main site and one subdivision, two axes, multiple points and one district” in Beijing. The plan intends to change the development mode of single-center agglomeration and construct a new urban development pattern for Beijing [8]. However, there is a lack of sufficient objective analysis methods to judge urban development. This makes it difficult to determine if development aligns with the urban planning intentions. To reduce the social problems caused by urban land expansion, this paper evaluates the development level of Beijing’s spatial structure. We use an objective method and analyze the driving factors that may affect spatial heterogeneity. The goal is to provide decision-making support for sustainable development planning in large urban inner spaces.

Traditional urban researchers have limited technical skills and data acquisition abilities. They typically rely on research data from statistical yearbooks or field surveys [9]. However, data obtained through these methods are mostly static with long update cycles and weak continuous characteristics. With the popularity of the Internet, the wide application of geospatial big data (e.g., point of interest (POI) data, mobile phone signaling data [10]) and remote sensing big data [11]) have been accessible. They have contributed greatly to urban research in terms of the urban spatial structure, human social behavior and other aspects [12,13,14]. Compared to traditional statistical data, big data have strong timeliness and high spatial resolution [15]. Geospatial big data help to fully utilize the advantages of remote sensing (RS) and geographic information systems (GIS) in dynamic monitoring and spatial analysis. It can identify urban morphology more effectively. In turn, the data can provide more specific and detailed decision making for government management [16]. For example, Cao et al. used smart card transaction big data to study the spatial differentiation law of public housing prices in Singapore [17]. He et al. used spatio-temporal big data, such as location-based services (LBS), POI and PM2.5 site detection, to assess the exposure to PM2.5 and health risks in cities during COVID-19 [18]. Chen Z, et al. used nighttime light (NTL) data to make an analogy between the characteristics of the urban structure and earth topography to identify the urban center of Shanghai [19]. Ma T, et al. used regression models to quantify the long-term relationship between weighted NTL area and urbanization variables, such as GDP, built-up area, power consumption, etc. [20]. Zhang Q and Seto K C monitored urban change at the regional and global scales using multi-period NTL data [21]. POI, NTL and population density data have a clear spatial correlation. They can significantly improve the effectiveness and accuracy of the study of the urban spatial structure [22]. At present, there are few studies that integrate the data above for the study of the urban spatial structure. Therefore, considering the similarity of spatial distribution rules, the factors above will more accurately reflect the characteristics of the urban spatial structure.

Previous studies have used various methods to study the influencing mechanisms of spatial heterogeneity on the urban spatial structure. These methods include multiple linear regression, geographic weighted regression (GWR) and other models [23,24]. The focus is mainly on exploring the influence of a single factor on spatial heterogeneity. However, less attention has been paid to the interaction between two factors on the explanatory power of a single factor [25]. Most geographical phenomena have a complex driving mechanism. The internal interaction of socio-economic and population factors mainly affects the distribution trend of the urban spatial structure. This complex interaction plays a key role in driving the generation and development of societies and cities [26]. Some researchers have attempted to address this issue using Geodetector [27,28,29,30]. They used the q statistic to measure the determining power of variables, with promising results. However, they overlook the importance of Geodetector in optimizing the expression of spatial heterogeneity. This leaves room for improvement in subsequent studies. Depending on the purpose of the research and data availability, there are different factors for studying urban structure. These factors include public services, economy, population, etc. Some scholars also measure the urban spatial structure from the perspective of green space [31]. Moreover, the research scale changes with the purpose of the research. In existing research, the research objects of urban spatial structure are generally divided into regional, city and street level [32]. On this basis, the research objects and data can be transformed into raster data grids. This helps to unify the spatial scale of the research and understand the distribution characteristics of the spatial structure as a whole. It also helps accelerate the transformation from monocentric to polycentric structures and promote the coordinated development of regions [33,34].

The research framework of this paper (Figure 1) is as follows: First, the kernel density estimation of POI data was performed. Under the grid-level research scale, we used Geodetector to perform the optimal discretization of factors and presented the optimized evaluation system. Then, we used the improved TOPSIS model to determine factor weights and score the extracted research units. The inverse distance weight (IDW) was used for interpolation to obtain the differentiation pattern of the urban spatial structure’s development level, which was compared with the “Beijing Urban Master Plan (2016–2035)”. Finally, this paper again used Geodetector to detect the single and interactive influences of the factors. Our innovation lies in the selection of the optimal discrete method and the number of intervals with Geodetector from a statistical perspective. This can fully demonstrate the heterogeneity of Beijing’s spatial development level. Interaction detection excavates and quantifies multi-factor intrinsic interactions. A more realistic representation of the urban spatial structure’s formation and development mechanisms is another advantage. Our research provides a feasible quantitative analysis scheme of the level of urban development. This is of great significance for optimizing the mode of spatial development and promoting the sustainable development of Beijing and other developed cities.

2. Materials and Methods

2.1. Study Area

Beijing, the capital city of China, where the urbanization process is rapidly developing, is selected as the study area (Figure 2). Beijing (115°25′ E–117°30′ E, 39°26′ N–41°03′ N) is located in the northern part of the North China Plain and has jurisdiction over 16 districts with a total area of 16,410.54 km² [35]. Since the reform and opening, the urban population of Beijing has been increasing. As a leader in regional development, Beijing has attracted a large number of resources and investment to the city. This has brought opportunities for international exchange and cooperation to Beijing and promoted the development of surrounding areas. Beijing plays an important role as the engine of regional development in the urban spatial structure and plays an important role in radiating and driving the surrounding areas. Its urban spatial structure showcases the latest ideas and practices of urban planning and construction. Therefore, the selection of Beijing as the research area plays a demonstrative role and reference for the development of other cities.

2.2. Data Sources

In recent studies, POI data [36], the Baidu heat map [37], mobile phone signaling data [38], NTL data [39] and other big data have been widely applied to the analysis of urban spatial structure. Based on extensive reading of the literature, we made full use of the advantages of emerging big data. We constructed an evaluation system for the urban spatial structure from the social aspects of regional functional division, industrial structure, economic development level, population agglomeration degree and distribution characteristics.

2.2.1. POI Data

As the basic type of geospatial big data, POI data can directly reflect the regional richness of various urban functions. It is often used in the research of urban functional area identification [40], urban vitality [41], urban land use change [42], etc. Compared with traditional non-point-based statistics, POI data correspond to the actual physical space of the city. It can accurately reflect the functional distribution and activities of different areas. Higher spatial precision facilitates comparison with the social typologies of the districts or with the main features of the urban structure to draw more accurate and reliable conclusions. We used Python to obtain POI data in 2020 for 15 categories from the Autonavi Open Platform Interface (https://lbs.amap.com, accessed on 1 December 2020) according to the platform’s introduction to data acquisition. The data include business residence, science and education, culture, entertainment, etc. The data information includes the name, category, longitude and latitude of the POI. Next, we used Python to batch convert POI data to the WGS84 coordinate system, eliminate POI data with missing information and divide them into six categories of urban functions (as shown in

Table 1. POI data classification information.

City functions	Primary classification	Secondary classification	Number of POI
Residential life	Business residence	Villa areas, commercial and residential buildings, community centers, dormitories, residential areas, etc.	32,258
Public service	Science and education culture	Museums, science and technology museums, scientific research units, higher education, secondary schools, art exhibitions, planetariums, libraries, etc.	44,815
	Life service	Telecommunications office, photography and printing, laundry, information and consultation center, post office, etc.	100,984
	Healthcare	First aid centers, disease prevention, clinics, general hospitals, etc.	25,540
	Automobile related	Charging station, gas station, car repairing, car sales, car maintenance, car washing, etc.	23,455
Commercial finance	Company enterprise	Factories, companies, etc.	77,047
	Financial institutions	ATM, insurance, investment finance, banking, etc.	12,551
	Hotel accommodation	Economic chain hotels, youth hostels, three-star hotels, four-star hotels, five-star hotels, etc.	19,211
Recreation	Shopping spending	Convenience stores, supermarkets, home building materials, shopping streets, etc.	155,921
	Dining and gourmet	Cake and dessert store, foreign food, snack fast food, Chinese food, etc.	106,813
	Sports and fitness	Fitness center, equestrian and horse racing, water sports, taekwondo, gymnasium complex, basketball, soccer, table tennis, etc.	13,585
	Entertainment	KTV, cinema, bar, theater, farmhouse, chess room, Internet cafe, playground, etc.	14,489
Tourist attractions	Park green space	Zoos, botanical gardens, aquariums, forest parks, squares, etc.	10,541
	Places of interest	Red tourism, memorials, world heritage, etc.
Transportation	Transportation facilities	Subways, bus stations, toll booths, parking lots, etc.	72,022

). Finally, there are 709,232 effective POI for six types of urban functions.

2.2.2. NTL Data

The intensity of nighttime light reflects the distribution of nighttime brightness in a city, which is often used to identify the urban spatial structure and its center. It has statistical significance in measuring the intensity of economic activity in a city. The data come from the National Geographic Data Center of the National Atmospheric and Oceanic Administration. Compared with the previous DMSP/OLS data, the NPP/VIIRS data of 2020 used in this paper has a higher spatial resolution (500 m). It reduces the saturation of nighttime light data [43]. The obtained data were extracted by mask according to the administrative boundary of Beijing to prepare for subsequent analysis.

2.2.3. Population Density Data

Population density data can be used to reflect the degree of population concentration in different regions and study the population’s spatial characteristics. Compared with traditional regional demographic data, WorldPop population data integrate multiple open data sets, such as roads, vegetation and land cover. It generates population spatial data with higher spatial resolution (100 m) based on the regression mapping method of the stochastic forest model. In a more detailed and real way, urban spatial structure features can be identified from the perspective of the population’s spatial characteristics [44]. The data selected in this paper is from 2020, which was extracted by mask based on the boundary of the Beijing administrative region.

2.3. Research Methods

2.3.1. Analysis of Spatial Distribution Characteristics of Urban Functions

The first law of geography states that everything is related to everything else, but closer things are more related than distant things [45]. Based on this law, the weight of the neighborhood of a POI decreases with the increase in the distance of the kernel density estimation. It considers that the aggregation degree of the POI reflects the distribution of human activities. Considering the distribution pattern of points, this method is often used to identify the hot spot distribution of point elements in the study of the urban spatial structure [46,47].

For any point s in the space region, the calculation formula of its kernel density estimation is shown in Equation (1) [48].

$$f\left(s\right)=\frac{1}{n{h}^2}{{\displaystyle \sum }}_{i=1}^{n}k\left(\frac{s-s_i}{h}\right),$$

(1)

where $f\left(s\right)$ is the kernel density estimation, n is the number of entities in the distance threshold range, h is the distance decay threshold between two points and $k\left(\frac{s-s_i}{h}\right)$ represents the weight function.

2.3.2. Comprehensive Development Level Evaluation Method

In this paper, the technique for order of preference by similarity to ideal solution (TOPSIS) model improved with the entropy weight method is adopted to objectively reflect the contribution of each evaluation factor to the development level of the urban spatial structure. It helps determine the pros and cons of the development of different regions. Then, we complete the comprehensive evaluation of different development levels of the urban spatial structure. The method effectively avoids the subjectivity of the TOPSIS model’s evaluation [49]. It reflects the differentiation of the urban spatial structure’s development level in a more comprehensive and objective way. We make full use of the advantage of the entropy weight method where the larger the information, the more accurate the weight.

ArcGIS 10.8 was used to conduct raster a turn point operation on the results of the kernel density estimation. The geographical processing was carried out on the obtained research units. Finally, we obtained 31,276 effective research units. Next, 31,276 research units and 8 established evaluation indicators were entered into the model of this paper. The specific implementation process is as follows:

(1)

Standardization of decision matrix

The indicators selected in this paper are all positive indicators. No positive processing will be carried out. Considering the differences in the evaluation index units, the matrix is standardized to eliminate the impact of dimension. The standardized matrix is

$$A=\left[\begin{array}{cccc}{a}_{11}& a_{12}& \cdots & a_{1m}\\ a_{21}& a_{ij}& \cdots & a_{2m}\\ ⋮& ⋮& \ddots & ⋮\\ a_{n1}& a_{n2}& \cdots & a_{nm}\end{array}\right],$$

(2)

where $A$represents the standardized matrix;$i$ represents the evaluation object, $i=1,2,\cdots ,n$; $j$represents the evaluation index, $j=1,2,\cdots ,m$; $a_{ij}$represents the value of the$i$-th evaluation object on the $j$-th evaluation index after standardization.

(2)

Weight calculation

Based on information entropy theory, the sum entropy weight of the$j$-th evaluation index can be determined according to Equations (3) and (4):

$$H_j=-k{{\displaystyle \sum }}_{i=1}^n{p}_{ij}ln{p}_{ij},$$

(3)

$$W_j=\frac{1-H_j}{{{\displaystyle \sum }}_{j=1}^m\left(1-H_j\right)} , W_j\in \left[0, 1\right],$$

(4)

where$H_j$is the information entropy,$W_j$is the entropy weight and $p_{ij}$represents the proportion of the $i$-th evaluation object in the $j$-th evaluation index.

(3)

Comprehensive evaluation

Determine the positive and negative solutions of the vector according to Equations (5) and (6):

$$A^{+}=\left(a_{i1}^{+},a_{i2}^{+},\cdots ,a_{im}^{+}\right),$$

(5)

$$A^{-}=\left(a_{i1}^{-},a_{i2}^{-},\cdots ,a_{im}^{-}\right),$$

(6)

where $A^{+}$ is the positive solution and$A^{-}$is the negative solution.

The Euclidean distance between each evaluation object and positive ideal solution and negative ideal solution can be calculated as

$$D_i^{+}=\sqrt{{{\displaystyle \sum }}_{j=1}^m{\omega }_j{\left(a_{ij}^{+}-a_{ij}\right)}^2}$$

(7)

$$D_i^{-}=\sqrt{{{\displaystyle \sum }}_{j=1}^m{\omega }_j{\left(a_{ij}^{-}-a_{ij}\right)}^2}$$

(8)

where$D_i^{+}$is the distance between$a_{ij }$and the positive ideal solution; $D_i^{-}$is the distance between$a_{ij}$and the negative ideal solution.

Calculate the proximity, namely overall score index, according to Equation (9):

$$S_i=\frac{D_i^{-}}{D_i^{+}+D_i^{-}},$$

(9)

where$S_i\in \left[0, 1\right]$. The higher the value of $S_i$, the closer the evaluation object is to the positive level, indicating the higher the development level of the urban spatial structure.

2.3.3. Raster Data Gridding

Meshing is an important method to transform spatial point elements into surface elements by using spatial topological relations. Commonly used mesh shapes include square, regular hexagon, regular triangle and so on. Meshing can reduce not only spatial data redundancy, but also process raster data with different spatial resolutions into uniform-sized grid cells. In “central place theory”, Christian Taylor pointed out that the urban spatial service structure has significant hexagonal morphological characteristics. The hexagonal grid is closer to the real urban structure than other shaped grids [50]. Therefore, in this paper, we create a regular hexagonal honeycomb map with Beijing as the boundary. Grid processing is convenient for subsequent research on the spatial heterogeneity of Beijing’s spatial structure at the same scale. It highlights the spatial differentiation pattern of the urban spatial structure in Beijing.

2.3.4. Analysis of Driving Factors

Geodetector is a new statistical method capable of detecting spatial heterogeneity. It can reflect the degree of influence of driving factors on spatial heterogeneity through factor detection, risk detection, ecological detection and interaction detection [51]. In this paper, we use factor detection, ecological detection and interaction detection to analyze the influence of driving factors. The calculation formula of factor detection is

$$q=1-\left(\frac{{{\displaystyle \sum }}_{h=1}^L{N}_h{\sigma }_h^2}{N{\sigma }^2}\right),$$

(10)

where$q$is the influence of the driving factor on the spatial heterogeneity of the urban spatial structure’s development level, and the value range is $\left[0,1\right]$; h is the number of subregions of the driving factor; L is the stratification of variables; $N$ and $N_h$ are the total number of samples and the number of samples in region h, respectively$; {\sigma }^{2 }$and${\sigma }_h^{2 }$are the variance of the total region and the variance of region h, respectively. The larger the value of$q$, the greater the influence of the factor on the spatial differentiation pattern.

Ecological detection is used to explore whether there are significant differences in different driving factors in influencing the spatial heterogeneity of the development level, usually using the F-test metric:

$$F=\frac{N_{X1}\left(N_{X2}-1\right)SS{W}_{X1}}{N_{X2}\left(N_{X1}-1\right)SS{W}_{X2}},$$

(11)

where N_X1 and N_X2, respectively, represent the sample size of two factors X₁ and X₂; SSW_X1 and SSW_X2 represent the sum of in-layer variances formed by X₁ and X₂, respectively. The null hypothesis is SSW_X1 = SSW_X2. The rejection of the null hypothesis at the significance level of α indicates that there are significant differences in the effects of the two factors, X₁ and X_2, on the spatial distribution of attribute Y.

Interaction detection is the biggest advantage of Geodetector when compared with other statistical methods. Whether the influence of different driving factors on spatial heterogeneity is independent can be detected by the statistic. In addition, we can determine whether any combination of driving factors enhances or weakens the influence. The expression analysis results are shown in

Table 2. Dual-factor interaction expression analysis.

Expressions	Interaction
$p\left(x\cap y\right)<min\left(p\left(x\right),p\left(y\right)\right)$	Non-linear weakening
$min\left(p\left(x\right),p\left(y\right)\right)<p\left(x\cap y\right)<max\left(p\left(x\right),p\left(y\right)\right)$	Single-factor linear weakening
$p\left(x\cap y\right)>max\left(p\left(x\right),p\left(y\right)\right)$	Dual-factor linear enhancement
$p\left(x\cap y\right)=p\left(x\right)+p\left(y\right)$	Mutual independence
$p\left(x\cap y\right)>p\left(x\right)+p\left(y\right)$	Non-linear reinforcement

.

3. Results

3.1. Optimal Discrete Results of the Evaluation System

In this paper, Geodetector was used to perform optimal discretization of the kernel density estimation results of six types of POI, NTL and population density data. Different discretization methods and the number of discretization intervals have an impact on the statistical q value. The level of the statistical q value affects the precision of the spatial heterogeneity expression. The higher the statistical q value, the more detailed the presentation of spatial heterogeneity. Figure 3 shows that among the commonly used discrete interval numbers three, four and five, the optimal number of discrete intervals for all factors is five. The spatial heterogeneity of the development level was divided according to five levels: relatively high, high, medium, low and relatively low (

Table 3. Evaluation system of the urban spatial structure’s development level.

Evaluation Factor		Development Level/Grade					Weight
		Relatively low/1	Low/2	Medium/3	High/4	Relatively High/5
Urban function configuration	Residential life	0	0–0.7	0.7−2.6	2.6−8.8	>8.8	15.29%
	Public services	0	0−1.8	1.8−7.3	7.3−29.2	>29.2	7.77%
	Economical finance	0	0−1.4	1.4−4.1	4.1−13.8	>13.8	7.05%
	Recreation	0	0−3.5	3.5−13.9	13.9−52.1	>52.1	8.91%
	Tourist attraction	0	0−0.2	0.2−0.6	0.6−1.5	>1.5	6.73%
	Transportation	0	0−0.7	0.7−2	2−8.7	>8.7	6.62%
Economic activity intensity	NTL	0.4	0.4−2.7	2.7−7.7	7.7−20.1	>20.1	29.41%
Population spatial distribution	Population density	0.1	0.1−4.8	4.8−14.4	14.4−48.1	>48.1	18.22%

). For the six types of urban functions, the optimal discrete method is the quartile method. For the intensity of economic activity and the population’s spatial distribution, the optimal discrete method is the standard deviation method. We used the TOPSIS model improved with the entropy weight method to assign weights to the evaluation factors. The overall score indexes calculated were used to rank each evaluation unit. Finally, we obtained the weights of the evaluation factors (Table 3) and the ranking of each evaluation unit (

Table 4. Ranking of evaluation unit scores.

Unit	Positive Ideal Solution Distance (D+)	Negative Ideal Solution Distance (D−)	Overall Score Index	Ranking
Unit_1	0.0157	0.0000	0.0000	4022
Unit_2	0.0157	0.0000	0.0000	4022
Unit_3	0.0154	0.0010	0.0633	4016
--	--	--	--	--
Unit_11477	0.0008	0.0156	0.9540	3
Unit_11478	0.0029	0.0137	0.8264	64
Unit_11479	0.0034	0.0129	0.7923	96
--	--	--	--	--
Unit_31273	0.0150	0.0018	0.1077	3968
Unit_31274	0.0151	0.0016	0.0983	3986
Unit_31275	0.0156	0.0008	0.0460	4019
Unit_31276	0.0157	0.0000	0.0000	4022

).

3.2. Spatial Differentiation Pattern of the Urban Spatial Structure’s Development Level

Figure 4a,b show that the development level of Beijing’s spatial structure follows a “low-high-low” pattern. The analysis of Figure 4a reveals that the largest relatively low development level area accounts for 38.05% of Beijing’s total area. This area lies in the city’s northwest direction and edges; it is characterized by sparse urban functions, a weak economy and low population density. Conversely, the relatively high-level spatial structure area comprises 5.46% of Beijing’s total area; it is concentrated mainly in the Dongcheng and Xicheng districts. The urban spatial structure radiates from Dongcheng district, with its development level decreasing towards the periphery. The high development level area in the urban fringe is sparse and small-scale. Overall, Beijing exhibits a core–satellite city structure with the spatial structure’s development level characterized as “low in the northwest, high in the southeast, low in the periphery and high in the center”.

The analysis results align with the “Beijing Urban Master Plan (2016–2035)” [8], indicating the validity of the evaluation method and factor selection. Future efforts could focus on improving the region’s urban functional configuration. This could enhance local economic activities, relieve population pressure in central areas and optimize the urban spatial structure.

3.3. Single-Factor Analysis

In order to study the influence of each factor on spatial heterogeneity, this paper analyzes eight factors with factor detection. According to the factor detection results in

Table 5. Single-factor ranking of urban spatial structure development.

Independent Variable		q Statistic	p Value	Whether It Is Significant	Ranking
Urban function configuration	Residential life	0.8400 ***	0.000	Y 1	2
	Public service	0.8299 ***	0.000	Y 1	3
	Business finance	0.7898 ***	0.000	Y 1	7
	Recreation	0.7968 ***	0.000	Y 1	6
	Tourist attraction	0.4990 ***	0.000	Y 1	8
	Transportation	0.8083 ***	0.000	Y 1	5
Economic activity intensity		0.8207 ***	0.000	Y 1	4
Population spatial distribution		0.8428 ***	0.000	Y 1	1

¹ Y for yes, and *** means the confidence level is 99%.

, the values of all factors are shown as 0.000. It indicates that the results are significant, and the driving force levels of all factors are of practical significance. Among them, the strongest driving force on the development level of Beijing’s spatial structure is the population spatial distribution (q = 0.8428), followed by the residential life function (q = 0.8400). The difference between them is not evident. The weakest driving force is the tourist attraction function, with q = 0.4990. It is significantly lower than the driving force of other factors on the spatial structure of Beijing. The driving force of each factor is sorted as follows: population spatial distribution (q = 0.8428), residential life function (q = 0.8400), public service function (q = 0.8299), economic activity intensity (q = 0.8207), transportation function (q = 0.8083), recreation function (q = 0.7968), commercial finance function (q = 0.7898) and the tourist attraction function (q = 0.4990).

Next, this paper conducts a spatial analysis of the development level of Beijing’s urban spatial structure. The analysis focuses on three perspectives: urban function configuration, economic activity intensity and population spatial distribution.

3.3.1. Urban Function Configuration

From the perspective of urban function (Figure 5), Xicheng, Dongcheng, the adjacent areas of Haidian, Chaoyang, Fengtai and Shijingshan districts have the functions of life, finance, recreation, tourism, etc. These areas have convenient transportation facilities and the most complete urban functions. The development of the urban spatial structure is high in the center and decreases with distance. Yanqing, Huairou, Miyun and Pinggu districts on the edge of Beijing have small-scale, highly developed residential life centers. Mentougou District has the weakest residential life function because it has the smallest proportion of residential areas. The development level of public service, commercial finance, recreation and transportation functions is higher in the southeast direction than in the northwest direction. It indicates that these four functions are easier to expand in the southeast direction, resulting in better spatial structure. Beijing’s topography, with high northwest and low southeast, affects the spatial structure of these functions. Tourist attractions in Beijing are uniformly distributed with multi-center grouping. It indicates that the distribution is less impacted from terrain based on the distribution of historical and cultural heritage and tourism resources.

3.3.2. Economic Activity Intensity

From the perspective of economic activities (Figure 6a), Xicheng, Dongcheng, Haidian, Chaoyang, Fengtai and Shijingshan districts in Beijing have high-intensity and uniformly distributed nighttime lights. This indicates these areas have high-intensity and active development of economic activities. Changping, Daxing, Tongzhou and Shunyi districts have relatively weak nighttime light brightness. This indicates that economic activity intensity in these areas is at a medium level. Yanqing, Huairou, Miyun and Pinggu districts on the edge of the city have areas with high nighttime light brightness. This indicates that these areas are the center of local economic activities, with a high potential for economic development. These areas also play a radiating role in promoting regional economic development, reflecting the high level of urban spatial development. Mentougou and Fangshan districts have the lowest nighttime light brightness. This indicates that the level of local economic development is relatively backward. The economic activity intensity in Beijing follows a distribution trend of “low in the northwest and high in the southeast”. This distribution belongs to the core and satellite city structure.

3.3.3. Population Spatial Distribution

From the perspective of the population’s spatial distribution (Figure 6b), high-value areas of population density in Beijing are concentrated in Xicheng, Dongcheng, Chaoyang and Fengtai districts and adjacent areas. Population concentration decreases with distance from the core, with slower decrease in southeast direction. This reflects an unbalanced distribution of the population’s spatial structure in Beijing. There are scattered and small high-value areas of population density in central zones of urban fringe areas, such as Pinggu and Miyun districts. Population in Tongzhou and eastern Fangshan districts is evenly distributed, with a medium level of population concentration. Overall, the population spatial distribution of Beijing follows a core and satellite city structure.

Comparing the spatial structure of economic activity intensity and the population’s spatial distribution, the central geosphere of the former is more obvious than that of the latter. The range of the population’s spatial distribution is weaker than that of the intensity of economic activity. In addition, the paper compares the spatial distribution characteristics of the two with the road network planning of Beijing. Results show that the scope and direction of the population’s spatial distribution are consistent with the road network’s distribution. The population’s spatial distribution and spatial accessibility are closely related.

3.4. Dual-Factor Driving Analysis

According to the ecological detection function of Geodetector (Figure 7), residential life function and six other factors make no significant difference in the spatial heterogeneity of the urban spatial structure’s development level, except for the population’s spatial distribution. This was completed to further study the influence of various factors on the urban spatial structure’s development level. Similarly, public services and five other factors, except for the population’s spatial distribution, make no significant difference in the spatial heterogeneity of the urban spatial structure’s development level. However, the population’s spatial distribution and other driving factors show significant differences in the spatial heterogeneity of the development level. The results of ecological detection effectively reflect the significant difference among driving factors on the spatial differentiation of Beijing’s urban structure. On this basis, the heat map is drawn according to the results of the interaction detection function (Figure 8). The result shows that the q value of any two factors are greater than that of the single factor. It indicates the interaction relationship is enhanced by two factors. There is no weakening or independent relationship. This reflects that the interaction of each of the two factors can effectively enhance the driving force for the urban spatial structure’s development level.

In the interaction detection of all factors, the q value of the interaction between the population’s spatial distribution and the commercial finance function is the highest, reaching 0.9550. It indicates that the two have the strongest interactive influence on the urban spatial structure’s development level. Figure 8 shows that under the interaction between the population’s spatial distribution and other driving factors, the value of the other dual-factor driving forces, except q (population spatial distribution ∩ tourist attraction) = 0.8992, is about 0.95. This reflects the key role of the population’s spatial distribution in promoting the development of Beijing’s urban spatial structure. This is because the high population density is often accompanied by the characteristics of high consumption demand and strong consumption power. The urban function configuration, such as public services and recreation, tends to be distributed in such areas, which effectively stimulates local economic consumption. This not only helps to obtain higher economic benefits, but also greatly promotes the formation and development of the urban spatial structure.

4. Discussion

4.1. Optimal Dispersion Method of Driving Factors

It is crucial to choose an appropriate discrete method and number of intervals for grading the evaluation of the urban spatial structure. Most previous studies have utilized the natural break point classification method to divide urban space into several levels and evaluate the urban spatial structure [52,53,54,55]. Although this method can be discretized according to the data distribution, it has its shortcomings. To some extent, the results are subjective and ignore the influence of spatial factors, resulting in the inability to show spatial heterogeneity effectively. For instance, Saravana Ganesh Manoharan et al. divided the population density of cities into three levels: low, medium and high. Then, they evaluated the cities based on the average value when studying urban social vulnerability in India [54]. Similarly, when examining the distribution pattern of air pollution in Beijing, Zhang et al. also utilized the natural break point classification method to divide pollution levels into three categories: low, medium and high [55]. However, this method failed to reflect the spatial heterogeneity of urban air pollution comprehensively, leading to the loss of many details. A common issue in previous studies is that some areas may be evaluated at the same level despite having different urban spatial structures. Therefore, these studies failed to fully explore the distributional characteristics of spatial data and construct an evaluation system that is generally rough. Consequently, the spatial heterogeneity and comprehensive development level of urban space cannot be fully reflected.

In contrast, our study utilizes Geodetector in R language and integrates statistical analysis advantages with geographic information technology. We optimally discretize the kernel density estimation results of six types of urban functions, NTL data and population density data. We selected a suitable discrete method and number of intervals for the spatial heterogeneity of Beijing’s spatial structure’s development level. This approach considers regional development level differences caused by Beijing’s terrain differences and retains the spatial characteristics of the low data distribution frequency area. Thus, the research results more accurately reflect the complexity and heterogeneity of Beijing’s urban spatial structure. It provides a more scientific reference for urban planning and management.

To sum up, in constructing the evaluation system, we fully consider the regional characteristics of driving factors and the data distribution differences caused by them. Our method meets actual needs better and significantly improves the evaluation effect of Beijing’s urban spatial structure’s development level. Our research method is more advanced and accurate, providing a new perspective for studying the spatial heterogeneity of the urban spatial structure.

4.2. Exploring the Guiding Power of “People-Oriented” Policy to the Development of Urban Spatial Structure

Previous studies have suggested that the economy is the primary driving force behind Beijing’s development pattern [56,57]. However, our research finds that the spatial distribution of the population has a greater impact on urban development patterns than the density of economic activity. This suggests that as urbanization advances, population factors become more influential than economic factors. The “Beijing Urban Master Plan (2016–2035)” proposes various concepts, such as population centers and public service centers, to optimize and adjust the urban spatial structure [8]. This highlights the importance of considering population factors in urban planning to meet the needs and expectations of people and promote sustainable development. The guidance of policy validates our conclusion and emphasizes the role of policy in the development of urban spatial patterns. Urban planning plays a critical role in determining a city’s scale and direction of development [58]. We should focus on the construction of northwestern marginal areas, aiming to improve the development efficiency and level of urban function configuration [59]. This will help attract the inflow of enterprises and talents and promote the overall development of Beijing. Currently, core areas such as Dongcheng, Xicheng and Chaoyang District have high development levels. In contrast, the development level of Yangyang, Huairou and other marginal areas is low, and the urban function configuration is not perfect. By evacuating the functions of core urban areas, such as Dongcheng and Xicheng districts, to share regional infrastructure, the remote areas of Beijing can enjoy greater benefits. This will give full force to the synergistic effect between driving factors and promote the optimization of the urban spatial structure. In addition, it can avoid the efficiency loss caused by excessive economic agglomeration under the single-center structure. In this way, the healthy growth of the talent scale and the promotion of urban competitiveness can be achieved.

The urban spatial structure should meet people’s various needs. Population density, a driving factor with people as the main body, significantly impacts Beijing’s urban spatial structure. Areas with high population density tend to form the urban center with more urban functions and services, while areas with low population density tend to be the mountainous areas on the northwestern edge of the city. This is due to Beijing’s topography and geomorphology, consistent with previous research results [60]. However, this study found that population density also affects the urban spatial structure’s internal distribution and structure type. Areas with high population density tend to have an intensive structure, while areas with low population density tend to have a dispersed structure. The findings are significant for future urban planning and design, emphasizing the need to pay more attention to the impact of population density on the urban spatial structure. In the future, Beijing should promote the connection and coordination between the core and the surrounding areas through talent introduction to drive the construction of urban transportation. This is conducive to the formation of an urban spatial structure with talents as the core driving force for development.

This study calls on urban decision makers to pay attention to the driving forces of population density in planning and construction and clarify the meaning of “people-oriented”. The development of cities should give priority to meeting people’s needs and improving people’s quality of life. Measures such as adjusting the population spatial distribution and improving supporting infrastructure will help Beijing build a “green, low-carbon, smart and livable” city with sustainable development. This will foster Beijing’s leading role in the coordinated development of the Beijing–Tianjin–Hebei region. According to the characteristics of different types of central urban areas, urban planners formulate corresponding construction schemes. It helps realize the rational layout, optimization and upgrading of the urban spatial structure.

4.3. Limitations and Future Work

The concept of “people-oriented” development is currently reflected only by population density data, due to the limited selection of basic research data. The leading factors of the urban spatial development level are greatly influenced by policies. To gain a more comprehensive understanding of the evolutionary process of Beijing’s spatial structure, future studies could analyze the differences and changes in core driving factors before and after the release of the “Beijing Urban Master Plan (2016–2035)” from a multi-temporal perspective. To further explore and study the formation and future development of Beijing’s spatial structure, future studies could incorporate natural factors in addition to social factors. This would allow for a more complete understanding of the heterogeneity of the urban spatial structure’s development level.

5. Conclusions

Based on the development status of the spatial structure in Beijing, this paper puts forward an evaluation system of the urban spatial structure from three perspectives. These perspectives include the urban function configuration, economic activity intensity and population spatial distribution. By using the TOPSIS model improved with the entropy weight method, the research units are weighted and ranked. Then, we evaluate the comprehensive development level of Beijing’s urban spatial structure. Furthermore, the spatial heterogeneity and influence of Beijing’s spatial structure are analyzed from single-factor and dual-factor perspectives. The following conclusions can be drawn:

(1)

The evaluation system proposed in this paper integrates economic activity intensity, population spatial distribution and other factors. It improves the traditional evaluation index system based on POI data. The Geodetector of the R language version is used to optimize data discretization. After comparing the statistical results of different classification methods, the optimal discrete interval number for all factors is determined to be five. The quartile method and standard deviation method are found to better reflect the spatial heterogeneity of Beijing’s spatial structure in the three aspects. The comprehensive results of this analysis are compared with the “Beijing Urban Master Plan (2016–2035)” and are shown to be reasonable, effective and conforming to the future master plan of Beijing;

(2)

On the whole, the urban spatial structure of Beijing is the core and satellite city structure. The high-value areas of development are mainly concentrated in the central areas represented by Xicheng District and Dongcheng District. These areas account for the lowest proportion among all levels of development. The low-value area is surrounded by the western and northern edge of the main urban area of Beijing. It occupies the highest proportion among all levels of development. The development level of the urban spatial structure presents a spatial heterogeneity of “low in the northwest, high in the southeast, low in the periphery and high in the center”;

(3)

The single factor results show that the population’s spatial distribution has the strongest explanatory power for the development level’s spatial heterogeneity. The tourist attraction function, on the other hand, has the weakest explanatory power. The residential life function, when compared with the spatial distribution characteristics of other factors, is the most concentrated. It is mainly distributed in the core areas of Dongcheng District and Xicheng District. In the fringe area of Beijing, the tourist attraction function is the most uniform. This is in contrast to the central areas where the function is more concentrated. Additionally, compared with the intensity of economic activity, the central geospheric characteristics of the population’s spatial distribution are not as apparent. However, they are more closely related to spatial accessibility, indicating a relationship between the population’s spatial distribution and transportation infrastructure;

(4)

Further analysis of the interaction detection results shows that the interaction between the population’s spatial distribution and other factors is the strongest driving force. It indicates that the population’s spatial distribution is crucial to the development of Beijing’s spatial structure. The results are significant. This further demonstrates from a quantitative perspective that the population’s spatial distribution is the key factor affecting the spatial heterogeneity of the development level in Beijing. Areas in Beijing with a low development level can promote improvement of economic activity intensity and urban function configuration by attracting population flows. This can alleviate the effect of high-density population pressure in the main urban areas. It makes contributions to the optimization of urban spatial structure and the improvement of the overall development level, which has positive practical significance.