Refined Identification of Urban Functional Zones Integrating Multisource Data Features: A Case Study of Lanzhou, China

Abstract

Identifying urban functional zones is one of the important foundational activities for urban renewal and the development of high-quality urban areas. Efficient and accurate identification methods for urban functional zones are significant for smart city planning and industrial layout optimization. However, existing studies have not adequately considered the impact of the interactions between human activities and geographical space provision on the delineation of urban functional zones. Therefore, from the perspective of integrating the spatiotemporal characteristics of human activities with the distribution of urban functional facilities, by incorporating mobile signaling, POI (point of interest), and building outline data, we propose a multifactorial weighted kernel density model that integrates ‘human activity–land feature area–public awareness’ to delineate urban functional zones quantitatively. The results show that the urban functional zones in the central city area of Lanzhou are primarily characterized by dominant single functional zones nested within mixed functional zones, forming a spatial pattern of ‘single–mixed’ synergistic development. Mixed function zones are widely distributed in the center of Lanzhou City. However, the area accounted for a relatively small proportion, the overall degree of functional mixing is not high, and the inter-district differences are obvious. The confusion matrix showed 85% accuracy and a Kappa coefficient of 0.83.

1. Introduction

Urban functional zones, serving as spatial carriers for the division of labor among city functions, are critical mediators for the exchange of various elements and resources and fundamental units for urban planning, management, and resource allocation [1,2,3]. Refined urban functional zoning is crucial for enhancing urban management capacity and advancing the scientific nature of urban planning [4]. With the increasing demand for diversified urban functions [5], the contradiction between the current insufficiency of public service resources, the unbalanced spatial distribution within functional zones, and the public desire for a high-quality and improved life have intensified. The above issues highlight the importance of accurately identifying and optimizing the urban functional zones from a human-centered perspective [6].

Identification methods for urban functional zones are the research hotspot of state-of-the-art urban science and the fundamental theme of urban complexity research [7,8,9]. In the traditional data context, urban functional area identification methods rely on remote sensing images, land use data, and other sources [10,11,12,13]. These methods have achieved high levels of accuracy in functional zone identification. However, these methods, based on remote sensing images, still have some shortcomings, such as high cost and the absence of socioeconomic features because only natural features can be obtained. As data collection capabilities continue to improve, point of interest (POI) data have been applied in the study of urban functional zones [14]. POI data represent data points within a geographic coordinate system that are used to identify specific locations, such as business establishments, attractions, or facilities. In the early applications of POI data, owing to the limited comprehensiveness of the dataset, studies on urban functional zones based on POI data primarily focused on individual urban functions. For instance, some studies investigated the patterns of distribution and clustering of retail trade within the city [15]. As POI data has continued to improve, the vast majority of public hotspot areas have been included. POI data have become an important tool for identifying urban functional zones due to their extensive coverage and easy accessibility. Related research has expanded to employ the quantitative distinctions among various categories of POI data for classifying urban functional zones, thereby enabling the differentiation of diverse types of urban functional zones [16]. For instance, Li Yuanfu et al. [17] employed a random forest model to calculate the weights for each POI and classified the primary urban area of Chongqing into 10 distinct functional zones. Huang Chong et al. [18] used a combination of density index and ANNI methods to calculate POI data and classify urban functions into 26 categories in Futian District, Shenzhen. Luo Shaohua et al. [19] classified the urban functions of Xicheng District, Beijing, into 36 categories based on the frequency density method and category ratio. However, since POI data are point-based and lack area information, despite the large volume and wide variety of function categories, this limits their further use in detailed studies of urban functional zone identification. Therefore, current researchers often refer to existing business classifications and urban public service facility planning standards to uniformly assign weights and scores to POI types corresponding to different functional land uses [20,21,22].

POI data identify urban functional zones solely from the perspective of the geographical spatial distribution of urban facilities, i.e., without considering the intensity of human activity demands for different POI types. The emergence of novel big data aptly fills this gap. Driven by novel big data, large-scale data sources (including mobile signaling data [23], social media data [24], GPS data [25], and traffic and travel data) based on human activity, which are characterized by high spatial and temporal resolution, timeliness, and sensitivity to social functional attributes, provide a new perspective for the functional zone identification methods. For example, using taxi trajectory data, Liu Xudong et al. [26] analyzed hourly traffic trips on both weekdays and weekends. They classified the functional zones within the Chengdu Ring Road into six categories. Zhi Ye et al. [27] classified Shanghai’s urban functional zones into four categories using social media check-in data at different times of the day with the consideration of the activity patterns of the interviewed residents. Pei Tao et al. [28] classified land use types in Singapore into five categories based on the different land use types characterized by vectors of calls per hour and total calls. Mitchell et al. [29], utilizing commuting data from residents and employing the Intramax clustering method, categorized New South Wales into 24 functional zones based on the residents’ commuting behaviors and economic activities. Obaco et al. [30] used travel time, population density, and OSM road network data to identify and connect urban cores, categorizing Ecuador into 28 functional urban zones. Compared to fragmented activity data, which often reflect minority groups through sources such as social media and transportation records, mobile signaling data stands out due to the continuous dynamic recording and high user engagement rate. This approach makes the data type particularly advantageous for identifying urban functional zones based on population distribution and activity trajectories. For example, Novak et al. [31] utilized mobile signaling data and the O-D matrix and Intramax method to categorize Estonia into several functional areas based on daily activity patterns and population dynamics. Komaki et al. [32] analyzed commuting and consumer behavior patterns using mobile signaling data and census data, identifying the city’s core areas and multiple regions with distinct functional characteristics. Toole et al. [33] utilized mobile signaling data and employed the random forest algorithm to analyze human activity temporal sequences and spatial distribution, classifying the Greater Boston area into five types of urban functional zones. While mobile signaling data can capture high-frequency human activity information, it is constrained by the limitations in user numbers and the accuracy of user footprint localization, leading to the loss of some urban functional attributes and, to some extent, creating errors in the recognition results. However, POI data can compensate for the lack of geographical spatial entity information in novel big data that include human activity information, as well as for the loss of some urban functional attributes caused by limitations in user numbers and the accuracy of user footprint localization. Combining POI data with novel types of big data containing information on human activity facilitates the capture of diverse socioeconomic factors and resident travel patterns [34], enabling a more comprehensive interpretation of the spatial data and a more detailed depiction of urban functionalities.

Currently, some scholars are overcoming the deficiencies of single data sources by utilizing multisource data fusion. However, there are still some deficiencies in the research on urban functional zone identification that involves the fusion of remote sensing data, POI data, and novel big data. First, the current research only assigns artificial weights to POI data lacking area information. For example, different POI categories are weighted and scored based on the existing business classification standards and urban public service facility planning norms according to their area sizes [35,36]. This method can only assess the area of POIs of the same type without evaluating and allocating the area coverage of urban functional facilities for each piece of POI data. Moreover, the integration of building outline data, which includes information on building coverage area and number of floors, allows for the estimation of the geographic area associated with each POI. This approach adequately considers the impact of size variations in the physical entities represented by each POI on the delineation of functional zones. In addition, the varying definitions and collection methods of multi-source geographic data determine that their applications in the study of urban functional zones exhibit distinct characteristics. POI data can identify the urban functional zones of a city from the perspective of geospatial provisioning by reflecting the clustering of economic activities in these zones. Comparatively speaking, mobile signaling data, social media data, GPS data, and traffic and travel data can analyze the spatial differences of urban functional zones from the perspective of distribution and mobility of human activities. Complex internal mechanisms exist between the spatial information reflected in different human activities and the geographical spatial provisioning elements [31]. Therefore, it is crucial to fully consider the significant influence of human–land interaction in identifying urban functional zones to understand the current state of urban functions and promote sustainable spatial regulation. However, these studies have not yet fully considered the impact of the interactions between human activities and geographical spatial supply on the delineation of urban functional zones. By considering the integration of diurnal variations in human activities from mobile signaling data with the distribution of urban functional facilities represented by POI data, precise identification and optimization of urban functional zones can be achieved from a human-centric perspective. Meanwhile, the smallest research unit commonly used in previous studies identifying urban functional zones is typically a regular grid or a road network scale [37]. Regular grids embody the advantages of a simple structure, easy comparison, and flexible spatial scales in identifying urban functional zones. However, they cannot adequately capture the structural–functional boundaries and internal heterogeneity of cities [38]. The process of selecting urban road networks using only road data can lead to errors in the selection results, due to topological issues such as a large number of redundant intersecting roads and the lack of comprehensive consideration of highly heterogeneous and complex scenario units [39]. Combining detailed control planning data, OSM road network data, and GF-2 images, which collectively consider factors such as the transportation network and land use, finely delineating study units can assist policymakers in implementing precise spatial control measures and achieving sustainable urban development.

This paper begins from the perspective of integrating the spatiotemporal characteristics of human activities with the distribution of urban functional facilities. It integrates mobile signaling data containing information on human activities with POI data and building outline data, reflecting urban functional facilities’ distribution and scale. A method for identifying urban functional zones is proposed, based on a multifactorial weighted kernel density model that integrates ‘human activity–land feature area–public awareness’. The contributions of this study are as follows: (1) By incorporating the spatiotemporal characteristics of human activities and the distribution of urban functional facilities into the delineation of urban functional zones, this study offers new methods and perspectives for the refined identification of urban functional zones. (2) Consideration is given to the impact of size variations in the physical entities represented by each POI on the delineation of functional zones. The land feature area corresponding to each POI is estimated using building outline data. (3) Research units are segmented based on detailed control planning data that integrates transportation networks, land uses, community service facilities, the OSM road network, and GF-2 images, enhancing their rationality. (4) We have categorized urban function types objectively and comprehensively by integrating the day–night variations in the utilization intensity of urban functional facilities, along with classified POI data and urban construction land types. This contribution enhances the practical relevance of our research findings in urban planning. Using Lanzhou City, the central city of inland Northwest China, as a case study, this research aims to provide a scientific basis for accurately understanding the current state of urban functional development, optimizing the internal spatial structure of the city, and formulating a reasonable urban planning scheme.

2. Methods

2.1. Study Area

Lanzhou, as the capital of Gansu Province, is an important industrial base in Northwest China. It is positioned along the central route of the ‘One Belt, One Road’ initiative, benefiting from its locational advantages with a ‘central hub with six connections’, radiating across Northwest China. The city comprises five districts and three counties, covering a total land area of 13,100 square kilometers, with a resident population of 4,425,100 at the end of 2022. The urban design marked by ‘two mountains on either side of a river, with the Yellow River running through the city’, has given rise to Lanzhou’s unique linear and clustered layout. Under its influence, urban functions are scattered and complex, displaying good typicality and representativeness. Currently, urban functional area identification research has predominantly focused on southeastern coastal cities with well-established urban functions and strong economic power, while giving limited consideration to central cities located at Northwestern inland China, with Lanzhou as a typical example. Therefore, the central urban area of Lanzhou City has been chosen as the study area in this paper (Figure 1). This central urban area is situated in the southern part of Lanzhou and comprises four districts: Chengguan, Qilihe, Anning, and Xigu. It encompasses 50 streets, includes a high-tech development zone management committee, and represents the concentrated area for businesses, population, and service industries in Lanzhou.

2.2. Data Source and Process

This study utilizes multi-source data, including mobile signaling data, POI data, Building outline data, OSM road network data, GF-2 images, and the detailed regulatory planning of central areas within Lanzhou city, as depicted in

Table 1. Study data description.

Data	Year	Application
Mobile signaling data	April 2023	Calculates day-time and night-time human activity for different functional types of POIs within the smallest parcel unit, reflecting the intensity of urban function demand.
POI data	April 2023	Provides a detailed classification of urban functions by integrating day–night variations in functional zone utilization intensity and urban land use types.
Building outline data	April 2023	Estimates the area information of urban functional facilities corresponding to each POI
GF-2 image	April 2023	Utilized for delineating research units in articles.
Detailed regulatory planning for Lanzhou city center	April 2018	Utilized for delineating research units in articles and testing functional area identification results.
OSM road network	April 2023	Highways, railroads, township arterials, urban arterials, and trails were chosen to delineate the research units in the article.
Baidu’s online map	April 2023	Utilized for testing functional area identification results.
Field investigation	April 2023	The field surveys were conducted in urban functional zones with ambiguous delineations to validate the identification results.

. The mobile signaling data were collected from gridded data provided by China Mobile in Lanzhou City on a weekday, 26 April 2023. The data have a time scale of hours and a spatial scale of 200 m × 200 m. They were divided into two time periods, 6:00–18:00 (day-time) and 18:00–6:00 (night-time). The number of people in each study unit was counted separately for day-time and night-time, and weights for the day-time and night-time human activity were calculated to assess the intensity of resident utilization of various urban functional zones and their temporal differences. The building outline data were obtained through the open API interface provided by Baidu Maps, containing building contour lines and height information, and a total of 41,911 building outline data were obtained for August 2023 within the study area after data cleaning and clipping.

This study obtained Lanzhou City’s POI data for April 2023, comprising a total of 91,787 items, through the open API interface provided by Gaode Map. Due to problems such as type redundancy, cross repetition, and inconsistent coordinates in the original POI data, this paper combines POI categories, day–night intensity variations of functional zone utilization, and the ‘Urban Land Use Classification and Planning and Construction Land Use Standards’ to subdivide POI data into 21 types meticulously. Based on the characteristics of the main urban functional zones, 21 categories of POI data were progressively categorized, resulting in six types of urban functional zones. These are the transportation zone, residential zone, industrial zone, green space zone, public services zone, and commercial zone, as shown in

Table 2. Classification for urban functional area categories.

Functional Zone Classification	Subcategories of Functions within the Functional Area	Classification of POI Primitives	Urban Land Use Subcategory	Day–Night Activity	General Urban Land Use Categories
Transportation zone	-	Transportation facilities Services	Transportation hub site	Day and night	Street and transportation land use
		Road ancillary facilities	Transportation yard site
		Access facilities	Urban road land
Residential zone	-	Business/residential-related	Class II residential land use	Night	Residential land use
		Residential areas	Class I, II, and III residential land use
Industrial zone	-	Business office	Class I industrial land/business facilities land	Day and night	Industrial land use
		Company	Class I industrial land/business facilities land
		Industrial park	Class I industrial land
Green space zone	-	Scenic spots	Parkland	Day	Green space and city square
		Public square	Plaza
Public services zone	Scientific, educational, and cultural functions	Science, Education, and culture	Educational and scientific research land/Land for cultural facilities	Day and night	Administrative and public service land use
		Sports and leisure	Sports ground
	Healthcare function	Health care services	Healthcare land	Day and night
	Service functions of government agencies	Government agencies and social organizations	Administrative office space	Day
Commercial zone	Life service function	Shopping services	Commercial facilities land	Day and night	Commercial and business land use
		Automotive services	Commercial facilities land/Other service facilities Land
		Motorcycle services	Commercial facilities land/Other service facilities land
		Life services	Business facilities land/Recreation and leisure facilities site
		Accommodation services	Commercial facilities land
	Catering service functions	Catering services	Commercial facilities land	Day and night
	Financial service function	Financial and insurance services	Business facilities land	Day

. Due to the presence of various sub-functional categories closely associated with human activities in the public services and commercial zones, although they coexist within the same functional zone, they are significantly different regarding the spatiotemporal characteristics of the corresponding human activities and the POI distribution. Therefore, we also explored their functional subcategories to better delineate the public services and commercial zones. These include the scientific, educational, cultural and healthcare functions, and service functions of government agencies in the public services zone and life service, catering service, and financial service functions in the commercial zone.

The data for the detailed regulatory planning of the central urban area in Lanzhou were sourced from the Lanzhou Municipal Bureau of Natural Resources. The GF-2 image data were acquired from the Gansu Data and Application Center of the High-Resolution Earth Observation System (HREDOS), with a multi-spectral resolution of 1 m. The OSM road network data were sourced from the OpenStreetMap geospatial platform. OSM road networks are frequently used for the delineation of study units for urban functional zoning [40]. Topological errors, including the large number of disconnected and redundantly intersected roads in the OSM roadway network, prevent it from forming closed study units. In this study, preprocessing was performed to address the interrupted paths. For independent roads, those with a distance of less than 10 m between their ends and neighboring roads were extended and connected, while the others were deleted. Nevertheless, segmenting the study unit using only road network data still engendered problems such as fragmented geographic scenarios. Therefore, this study integrates the detailed regulatory planning of Lanzhou city center, which is used to control road locations, with reference to the GF-2 images, to appropriately incorporate and exclude road information, including highways, railways, township main roads, urban main roads, and minor roads within the study area. Eventually, the central area of Lanzhou City was subdivided into 1011 parcel units (Figure 2).

2.3. Research Methods

Early methods for identifying and delineating urban functional zones were typically qualitative, such as the breakpoint theory and the field strength method [41]. Since 1960, as quantitative geography has evolved and matured, methods for identifying urban functional zones have increasingly shifted towards quantification [42]. Generally speaking, while economic statistics-based identification of functional zones is straightforward and effective, data scientists often find it challenging to obtain the necessary data for this method. Although remote sensing imagery can be effectively used to identify urban functional zones, its collection is costly and primarily captures natural land features, lacking social and economic characteristics [12,13]. With the gradual development and maturation of information technology, the widespread use of new types of geographic big data, represented by POI data, mobile signaling data, and social media data, has provided new data sources for the identification of urban functional zones. Many studies determine urban functional zones by constructing frequency densities and category ratios [43,44]. However, this method only considers the impact of the number of various POIs on the identification of functional categories. The kernel density estimation method can quantify the distance decay effect of different POI types in spatial distribution [41], making it superior to the frequency density method. However, POI data are point-based and lack area information. Current studies only assign weights to different POI categories based on existing business classification standards and the size of areas without assessing and allocating the area range of urban functional facilities corresponding to each POI [35,36]. Moreover, POI data identify urban functional zones solely from the perspective of the geographic spatial distribution of urban facilities, without considering the intensity of human activity demand for different POI types. Combining POI data with data that include the spatiotemporal characteristics of human activities can capture rich socioeconomic elements and resident travel information.

By integrating mobile signaling data, points of interest (POI) data, building outline data, and other supplementary datasets, this study proposed a novel approach for identifying urban functional zones, as illustrated in Figure 3. First, the urban road network was processed in accordance with road topology principles, and additional auxiliary data were used to delineate the smallest parcel units within the urban area. Secondly, the kernel density of each POI category was estimated at the smallest parcel units. These kernel densities were then weighted and superimposed based on three assessment factors: day-time and night-time human activity, land feature area, and public awareness. Again, based on the delineation rules, the functional categories of the smallest parcel units were determined. We then applied the local hotspot analysis method to identify hotspot areas. Finally, accuracy verification was accomplished by confusion matrix method.

2.3.1. Multi-Factor Weighted Kernel Density Calculation Model

Compared with the traditional estimation method based on frequency density, the kernel density estimation method accounts for the locational influence of the first law of geography, thoroughly considers the spatial autocorrelation of POI data, and weakens the discretization phenomenon of POI data. Because each type of POI exhibits varying levels of demand, scale, and scope of influence, this study proposed a multi-factor weighted kernel density calculation model that integrates ‘human activity–land feature area–public awareness’, as shown in Figure 4.

(1)

Kernel Density Estimation Methods

The kernel density estimation method was utilized to quantify the distance attenuation effect of different POI types in the spatial distribution [45]. The formula is as follows:

$$f(x,y)={\displaystyle \frac{3}{n{h}^2\pi }}{{\displaystyle \sum _{i=1}^n\left[1-\frac{{(x-x_i)}^2+{(y-y_i)}^2}{h^2}\right]}}^2.$$

(1)

In this formula, $f(x,y)$ denotes the kernel density at spatial location $(x,y)$, and $h$ denotes the search radius or bandwidth. In this research article, the bandwidth interval is determined as [700 m,800 m] based on the research area and the research target situation; $x_i$ and $y_i$ are the coordinates of the sampling point $i$; $n$ is the number of sampling points that are less than or equal to $h$ from the position $(x,y)$; $x$ and y are the coordinates of the grid center point; and ${(x-x_i)}^2+{(y-y_i)}^2$ denotes the grid center point with the sampling point $i$.

(2)

Calculation of Day-time and Night-time Human Activity Weights

The weights of day-time and night-time human activity are utilized to reflect the resident demand for various types of functional zones during different time periods. The functional zones are categorized into three types: day-time active (D), night-time active (N), and both daytime and night-time active (B). The formula is as follows:

$$H_{ij}=\left\{\begin{array}{c}{\displaystyle \frac{P_{i,day}}{M_{i,day}}},j\in D\\ {\displaystyle \frac{P_{i,night}}{M_{i,night}}},j\in N\\ {\displaystyle \frac{P_{i,day}}{M_{i,day}}}+{\displaystyle \frac{P_{i,night}}{M_{i,night}}},j\in B\end{array}\right..$$

(2)

In this formula, $P_{i,day}$ denotes the number of day-time active people in region $i$; $P_{i,night}$ denotes the number of night-time active people in region $i$; $M_{i,day}$ denotes the total number of day-time active POIs in region $i$; and $M_{i,night}$ denotes the total number of night-time active POIs in region $i$.

(3)

Calculation of land feature area Weights

The area information of urban functional facilities corresponding to POI data is important for accurate identification of urban functional zones. In this study, we estimated the land feature area of each POI using building outline data. First, we calculated the total building area based on the coverage area and the number of floors in the building outline data. The formula is as follows:

$$T_{\mathrm{i}}=S_i\times f_i.$$

(3)

In this formula, $T_i$ represents the total area of the $i$-th building. $s_i$ and $f_i$ represent the coverage area and the number of floors of the $i$-th building, respectively. Assuming that all entities within the same building are similar, meaning all POIs within the same building share a total building area, the land feature area for each POI is estimated as follows:

$$B_{\mathrm{i}}={\displaystyle \frac{T_i}{NumPOIs}},$$

(4)

where $B_i$ represents the area of the $i$-th POI, and $NumPOIs$ denotes the number of POIs within the $i$-th building.

(4)

Calculation of Public Awareness Weights

The public awareness weights were utilized to represent the general public’s perception of the salience of each POI category. Various types of POIs exhibit distinct impact levels. Some are less spatially dispersed but enjoy higher public awareness such as train stations and airports. In contrast, other POI types, like retail stores located near train stations, exhibit concentrated spatial distribution but lower public awareness compared to train stations. Accordingly, based on the research by Zhao Weifeng [46] and Wang et al. [22], this paper incorporates public awareness as an evaluation factor, as depicted in

Table 3. Public awareness scores of POI subcategories.

Subcategory	Score	Subcategory	Score	Subcategory	Score
Transportation facilities services	1	Accommodation services	0.5562	Company	0.3057
Road ancillary facilities	0.01	Scenic spots	0.8245	Shopping services	0.8146
Business/residential-related	0.3057	Public square	0.6548	Automotive services	0.8146
Residential areas	0.01	Science education and culture	0.6706	Government agencies and social organizations	0.355
Business office	0.01	Sports and leisure	0.501	Health care services	0.5069
Industrial park	0.3057	Life services	0.8146	Catering services	0.5562
Financial and insurance services	0.3057	Access facilities	0.01	Motorcycle services	0.8146

.

(5)

Multi-Factor Weighted Kernel Density

In order to address the quantitative disparities among day-time and night-time human activity weights, land feature area weights, and public awareness weights, we have selected the min–max scaling method for normalization in this study to ensure their comparability. The formula is as follows [17]:

$$x_i^{*}={\displaystyle \frac{x_i-x_{\min }}{x_{\max }-x_{\min }}}.$$

(5)

In this formula, $x_i^{*}$ is the normalized value of each weight; $x_i$ is the original weight value; $x_{\min }$ is the minimum value of each weight; and $x_{\max }$ is the maximum value of each weight.

The multi-factor weighted kernel density calculation method developed in this research is as follows:

$${\widehat{f}}_k(x,y)=x_1^{*}{f}_k(x,y)+x_2^{*}{f}_k(x,y)+x_3^{*}{f}_k(x,y).$$

(6)

In this formula, ${\widehat{f}}_k(x,y)$ denotes the multi-factor weighted kernel density value for the $k$-th POI at location $(x,y)$; $f_k(x,y)$ is the kernel density for the $k$-th POI at location $(x,y)$; $x_1^{*}$ is the standardized value of the day-time and night-time human activity weights for the $k$-th POI; $x_2^{*}$ is the standardized value of the land feature area weights for the $k$-th POI; and $x_3^{*}$ denotes the standardized value of the public awareness weights for the $k$-th POI standardized value, $k$ = 1, 2, …, 21.

2.3.2. The Judgment of the Functional Zone Type

For urban functional zone units composed of irregular grids, the functional type is determined by computation of the category ratio (CR). The formula is as follows [22]:

$$D_k={\displaystyle \frac{d_k}{S}}\ast 100\%,$$

(7)

$$d_k={\widehat{f}}_k(x,y),$$

(8)

$$S={\displaystyle \sum _{k=1}^n{\widehat{f}}_k(x,y)}.$$

(9)

In this formula, $D_k$ denotes the multi-factor weighted kernel density value for the $k$-th POI as a percentage of the sum of the multi-factor weighted kernel density values for all POIs in the study cell, $k$ = 1, 2, …, 21; $d_k$ denotes the multi-factor weighted kernel density value for the $k$-th POI; and $S$ denotes the sum of multi-factor weighted kernel density values for all POIs in each research unit. According to the formula, if $D_k$ ≥ 50%, the unit is judged to be a single functional zone. If $D_k$ < 50%, the unit is judged to be a mixed functional zone, with the mixed functional category depending on the first two functional types, and if there are no kernel density values, it is judged to be an area with no data.

2.3.3. Hot Spot Analysis

Getis-Ord G* is a local spatial autocorrelation indicator based on a distance weighting matrix that detects clusters of high and low values in functional zones. The formula is as follows [47]:

$$G^{*}={\displaystyle \frac{{\displaystyle \sum _{j=1}^n{W}_{ij}d(x_j)}}{{\displaystyle \sum _{j=1}^n{X}_j}}}.$$

(10)

In this formula, $x_j$ is the attribute value of the spatial region unit $j$, and $W_{ij}$ is the spatial weight matrix. The element in the spatial weight matrix is 1 if the distance between unit $i$ and unit $j$ is at the predetermined distance $d$, and 0 otherwise.

3. Results and Discussion

3.1. Day-Time and Night-Time Human Activity and Functional Zoning Results

3.1.1. Daytime and Night-Time Human Activity Recognition

To capture comprehensive resident travel data, mobile signaling data in this study are divided into two time periods: 6:00–18:00 (day-time) and 18:00–6:00 (night-time) at a spatial scale of 200 m × 200 m. By spatially connecting the smallest parcel units in the study area, information on the day-time and night-time human activity of each parcel unit was obtained for different time periods. In addition, we counted the POI categories and the percentage of POIs in areas characterized by ‘higher day-time than night-time’ and ‘higher night-time than day-time’ population densities to analyze urban functions with various daytime and night-time activity.

Figure 5 displays the outcomes of day-time and night-time human activity identification, revealing that the number of night dominant regions (553) is 1.2 times more than the number of day dominant regions (458) in the central city of Lanzhou. As demonstrated in

Table 4. Table of daytime and nighttime active area percentage for POIs.

Functional Zone TYPES	Transportation Zone	Green Space Zone	Residential Zone	Industrial Zone	Public Services Zone	Commercial Zone
Day dominant region	3.11%	0.50%	2.85%	8.24%	16.74%	68.56%
Night dominant region	2.66%	0.24%	3.22%	7.93%	14.19%	71.76%

, the daytime activity in the green space zone, which is open only during the day, is far greater than at night, whereas the night-time activity in the commercial and residential zones is far greater than during the day. Variations in day-time and night-time for different functional utilization intensities are influenced by resident travel characteristics and the operating hours of functional service facilities. Hence, it is justified to finely categorize the day-time and night-time activity of distinct functions.

3.1.2. Results of Functional Partitioning

By calculating the multi-factor weighted kernel density values that integrate ‘human activity-land feature area-public awareness’ for various functional zones in different block units, we obtained the box plot as shown in Figure 6. The graph indicates that the multi-factor weighted kernel density indices for industrial and public service zones are generally higher, whereas those for green spaces are relatively lower. Simultaneously, based on the category ratio (CR) method, we obtained the identification results for urban functional zones (Figure 7). A total of 14 functional area categories were identified, comprising six single functional zones and eight mixed functional zones. In general, Lanzhou central city is predominantly composed of single functional zones (constituting 67% of the study area), nested with mixed functional zones, forming a spatial pattern of ‘single–mixed’ synergistic functional development.

3.2. Spatial Distribution Characteristics of Single Functional Zones

Single functional zones are primarily characterized by the industrial zone and public services zone (Figure 8). Among these, the industrial zone exhibits a layout characterized by ‘traditional industries in the western region and modern industries in the eastern region’, Moreover, industrial zones are clustered in large contiguous areas in Xigu District, which is the core industrial zone of Lanzhou. In other areas, they are dispersed and independent. The area of the public services zone is the second largest, exhibiting a pattern of multi-point distribution across the four main city districts.

The residential zones show a spatial pattern of ‘high density in the east and low density in the west’. These zones are primarily located in Chengguan District, the most populous district, and adjacent Qilihe District. They often mix with commercial, public services, and industrial functions. The commercial zone exhibits ‘multi-aggregation center characteristics’ and is widely distributed in the areas of ‘Xiguan Shizi Road-Nanguan Shizi Road’, Dongfanghong Square, and other commercial districts within Lanzhou City. The transportation zone is mainly distributed along the east–west traffic arteries and is predominantly situated at major transportation hubs, including Lanzhou Station and Lanzhou West Station. The green space zone has the smallest area, exhibiting distribution characteristics of ‘overall dispersion, local concentration’. It is primarily situated on both sides of the Yellow River and the northern and southern mountain regions.

3.3. Spatial Distribution Characteristics of Mixed Functional Zones

The mixed functional zones cover a relatively small area, comprising only half of the single functional zones, which indicates the relatively low level of functional mixing in the urban area. Thus, the overall functional structure still needs to be upgraded. The mixed functional zones are primarily a combination of public services and commercial zones, and residential and public service zones, which account for 31.95% and 24.39% of the total mixed functional zone (Figure 9). Among them, this combination of public service and commercial zones are mainly distributed in the Chengguan core cluster and the Anning-Qilihe area. In this region, public facilities, such as schools, organizations, and hospitals, are widely distributed, and commercial activities are more developed. The combination of residential and public service zones is mostly located in Chengguan and Qilihe Districts. Public facilities are constructed in accordance with the population size and are frequently mixed with residential functions.

The combination of industrial and public service zones is mostly located in Chengguan District. Owing to topographical constraints and the limited land resources in the city center, the newly developed industrial areas in recent years on the eastern side have displayed a notable mixed layout pattern, incorporating public services, commercial activities, and other functions. The combination of residential and commercial zones is widely distributed in the four districts of the main city, forming a number of residential and commercial complexes. The layout of mixed functional zones, encompassing combinations of commercial and industrial activities, transportation and public services, residential and industrial, and transportation and industrial zones, exhibits a dispersed pattern with an overall multi-point distribution layout.

3.4. Identification of Urban Functional Hotspots

According to the identification results of urban functional zones in Lanzhou City, both the public services zone and the commercial zone, in the form of a single or mixed function, occupy a large proportion of the study area. Therefore, this study utilizes the local hotspot analysis method to identify hotspot areas for multiple secondary subcategory functions within the two types of functional areas, which are characterized by significant clustering differences and are close relevance to residents’ daily lives, as illustrated in Figure 10 and Figure 11.

In terms of the hotspot distribution in the public services zone, the scientific, educational, and cultural functions exhibit a double-core spatial distribution, mainly concentrated in the densely populated Chengguan District and Anning District. While this region contains ample higher education resources, it lacks supporting sports facilities and large-scale professional venues. The distribution of the healthcare function is characterized by multiple cores, mainly concentrated in Chengguan District, Qilihe District, and Xigu District. The service function of government organizations exhibits a single-core spatial distribution, predominantly concentrated in Chengguan District known as the old city and its surrounding areas. Among them, provincial and municipal administrative organizations are primarily located on both sides of Zhangye Road–Qin’an Road, often mixed with commercial and residential functions.

In terms of the hotspot distribution in the commercial zone, the catering service function exhibits a spatial distribution with multiple cores, and its functional layout is relatively well-designed. The hotspots are primarily concentrated on Zhangyelu Street in Chengguan District, Xihu Street in Qilihe District, and Xigucheng Street in Xigu District. The financial service function exhibits the spatial distribution of a single core, with its hotspot concentrated in Chengguan District, which has the highest specialization degree in the financial service function. The life service function exhibits a multi-core aggregation feature of ‘face + point’, and the hotspots are primarily distributed along the intersections of the main and secondary roads within the city.

In the future, it is imperative to follow the planning concept of ‘balanced development of sub-districts’ to enhance the commercial service capacity of Anning and Xigu Districts and their public service system.

3.5. Recognition Result Verification and Comparison with Traditional Methods

3.5.1. Recognition Result Verification

To assess the identification accuracy of the urban functional zone, this study adopts the confusion matrix, a widely used evaluation method for image classification accuracy. The specific accuracy evaluation metrics include user accuracy, mapping/production accuracy, overall accuracy, and kappa coefficient, as follows [48]:

$$User accuracy {=\mathrm{X}}_{ii}/X_{i+}\ast 100,$$

(11)

$$Mapping/Production accuracy=X_{ii}/X_{+i}\ast 100\%,$$

(12)

$$Overall accuracy={\displaystyle {\sum }_{i=1}^r{X}_{ii}/M\ast 100\%},$$

(13)

$$Kappa={\displaystyle \frac{M{\displaystyle {\sum }_{i=1}^r{X}_{ii}-{\displaystyle {\sum }_{i=1}^r{X}_{i+}}{X}_{+i}}}{M^2-{\displaystyle {\sum }_{i=1}^r{X}_{i+}}{X}_{+i}}}.$$

(14)

In this formula, $X_{ii}$ represents the element at the intersection of row $i$ and column $i$ in the confusion matrix; $X_{i+}$ denotes the sum of row $i$; $X_{+i}$ signifies the sum of column $i$; $M$ stands for the total number of samples; and $r$ represents the number of rows and columns. The kappa coefficient value falls within the range of [−1, 1], where the high kappa value indicates a high classification accuracy.

Based on the detailed regulatory planning of Lanzhou central urban area and Baidu Maps, the ground truth of urban functional zones is acquired through manual visual interpretation, with field investigations carried out in areas with uncertain boundaries of the urban functional zones. From the various functional types, excluding mixed functional zones, 20 sample plots were randomly and uniformly selected for each type. In total, 120 sample plots were selected to evaluate the identification accuracy of the urban functional zone. As demonstrated in

Table 5. Confusion matrix for urban functional zone classification.

Functional Zone	Number of Identified Zone Categories						Total Number of Parcels	Mapping/ Production Accuracy
	Residential Services	Industrial	Green Space	Commercial	Transportation	Public Services
Residential	16	0	0	1	0	3	20	0.8
Industrial	3	15	0	1	0	1	20	0.75
Green space	0	0	19	0	1	0	20	0.95
Commercial	3	1	0	15	0	1	20	0.75
Transportation	0	0	1	0	18	1	20	0.9
Public services	0	0	0	0	1	19	20	0.95
Total Number of Parcels	22	16	20	17	20	25	120	/
User Accuracy	0.73	0.94	0.95	0.88	0.9	0.76	/	/

Note: Overall accuracy is 85%; Kappa coefficient is 0.83.

and Figure 12, the overall identification accuracy of the urban functional area in this paper is 85%, with a kappa value of 0.83. In general, high identification accuracy was achieved, confirming the effectiveness of the proposed method.

The green space zone and the transportation zone exhibit higher user accuracy and production accuracy between them. However, some industrial zones and commercial zones have been erroneously categorized as residential zones due to geographical constraints and limited land resources within the central urban area. Particularly, in recent years, the intermixing phenomena have been increasing significantly in new industrial and residential land on the eastern side. Moreover, a multitude of commercial service functions, including catering services and life services, are predominantly concentrated around residential areas, leading to an inaccurate categorization of industrial and commercial zones.

Additionally, the identification results were further validated by comparing the findings of this study with GF-2 images and photos from field investigations. The GF-2 image data were acquired from the Gansu Data and Application Center of the High-Resolution Earth Observation System, with a multi-spectral resolution of 1 m. We randomly selected six single-functional and one mixed-functional parcel unit as samples and visually interpreted the GF-2 images of all the samples, as shown in Figure 13. Among them, the parcel units identified as the Green Space Zone are shown as the Renshou Mountain Cultural Tourism Area. The parcel units identified as the industrial zone are shown as Lanzhou Petrochemical Hongda Company. The parcel units identified as the commercial zone are shown as Lanzhou Wanda Plaza and Gansu Dongli Outlets Mall. The parcel units identified as the public services zone are shown as Lanzhou University. The parcel units identified as the residential zone are shown as Baye Residential Community and Gansu Electronics Company Residential Quarters. The parcel units identified as the transportation zone are shown as Lanzhou Station. The parcel units identified as the public services-commercial mixed-function xone, verified through field surveys and GF-2 images, are shown as Lanzhou Olympic Sports Center and Lanzhou Wanda Mall, corresponding with the identification results. These results demonstrate the effectiveness of the methods used in this study.

3.5.2. Comparison with Traditional Methods

In traditional methods, many studies typically identify urban functional zones by constructing a frequency density approach combined with the category ratio [44,45]. Traditional methods only consider the impact of POI counts on the identification of urban functional zones. Figure 14 depicts the distribution of urban functional zones identified using traditional methods. When evaluating the identification results of the traditional method, 120 sample parcels tested by the proposed method were selected to assess the accuracy of urban functional zone identification. The display of relevant sample parcels is shown in Figure 14. As indicated in

Table 6. Confusion matrix for urban functional zone classification under traditional methods.

Functional Zone	Number of Identified Zone Categories						Total Number of Parcels	Mapping/ Production Accuracy
	Residential Services	Industrial	Green Space	Commercial	Transportation	Public Services
Residential	14	0	1	2	1	2	20	0.7
Industrial	5	12	0	1	0	2	20	0.6
Green space	0	0	18	0	2	0	20	0.9
Commercial	5	2	0	10	0	3	20	0.5
Transportation	0	4	2	0	14	0	20	0.7
Public services	0	0	1	2	4	13	20	0.65
Total number of parcels	24	18	22	15	21	20	120	/
User accuracy	0.58	0.67	0.81	0.67	0.67	0.65	/	/

Note: Overall accuracy is 68%; Kappa coefficient is 0.65.

and Figure 15, the overall accuracy of the urban functional area identification using traditional methods is 68%, with a Kappa coefficient of 0.65. Compared to the identification results in this study, the accuracy of traditional methods is lower. This result underscores the effectiveness of the proposed multifactorial weighted kernel density model integrating ‘human activity–land feature area–public awareness’.

3.6. Discussion

Multi-source big data have significant potential in urban functional area identification [49,50]. The new geographic big data expand the conventional land use perspective, offering a novel interpretation of urban functional zone identification from the perspective of human activities [51,52]. The analysis presented in this study demonstrates that the multi-factor weighted kernel density computation model of ‘human activity–land feature area–public awareness’, which integrates mobile signaling data, POI data, and building outline data, achieves satisfactory accuracy and effectiveness in urban functional zone identification studies. Compared to previous studies on urban functional zone identification that relied solely on a single data source [26,27,28], our study integrates POI data, building outline data, and mobile signaling data, including spatiotemporal human activity characteristics. This approach enables us to capture detailed information on the spatiotemporal dynamics of socioeconomic elements and human activities, offering a novel methodology and perspective for refined urban functional zone identification.

Currently, China’s urban planning approach has shifted from ‘incremental land expansion’ to ‘tapping into the potential of existing land stocks’, and optimizing the city’s functional structure has become a focal point in urban planning [53,54,55]. As the spatial carrier for the division of various urban functions, the urban functional zone serves as the fundamental unit of urban planning and management. It is crucial to consider fully residents’ needs for high-quality urban life at the scale of the city’s functional zones and to address the incompatibility between planned and existing land uses. We have found that identifying urban functional zones by integrating the spatiotemporal characteristics of human activities with the distribution of urban functional facilities helps to bridge the gap between urban planning and actual usage. The method for urban functional zone identification proposed in this study can assist urban planners in precisely recognizing the current land use conditions in cities, facilitating the scientific allocation of urban infrastructure. Furthermore, the calculation of ‘day–night human activity’ based on mobile signaling data can more intuitively reflect the spatial needs of urban residents. Based on the land feature area estimation for each POI using building outline data, we can more accurately reflect the impact of size differences among the entities represented by POIs on functional zone identification, thereby enhancing the precision and efficiency of functional zone recognition. We recommend that urban planners and policymakers fully utilize spatiotemporal big data to identify urban functional zones, facilitating effective interaction between urban functionalities and population needs. Given the current urban traffic congestion, road widening and structural improvements should be prioritized for urban functional zones with high human activity identified by mobile signaling data, such as industrial and commercial service zones. This requirement can be fulfilled by increasing the number of lanes, designing dedicated bus lanes, and creating a circular traffic system, to reduce traffic bottlenecks and enhance road capacity. Concurrently, the layout of mixed functional zones should be optimized to minimize residents’ commuting needs. For instance, through rational land planning, integrating industrial, residential, and commercial functions within a specific area can significantly reduce long-distance commuting, thereby alleviating traffic congestion.

Although this study is confined to Lanzhou, the central city of inland Northwest China, the data sources used in our urban functional zone identification model, such as mobile signaling data, POI data, OSM road network data, and GF-2 images, are equally obtainable in other cities. Therefore, the methodology of this study is theoretically applicable to various other cities. Nevertheless, this study has certain limitations. First, while mobile signaling data can capture high-frequency human activity information, the data are constrained by relatively low cell phone ownership among children and elderly residents, which, to some extent, causes errors in the recognition results. In our multifactorial weighted kernel density model, the calculation of day-time and night-time human activity weights primarily depends on the accuracy of mobile signaling data. However, the absence of data for two demographic groups—children and elderly residents—introduces errors in calculating certain diurnal human activity weights, potentially reducing the accuracy of identifying specific urban functional zones. For example, the accuracy of identifying urban functional zones, such as the public services zone, which includes educational functions for children and healthcare functions for the aged population, will be somewhat affected. In future studies, we should incorporate social media data to supplement information on activities of demographic groups that may be overlooked by mobile signaling data. Second, due to the limitations of POI data and classification labels, we classified the POI data into 21 types. We organized them based on the primary characteristics of urban functions, dividing the urban functional zones into six categories. This constraint hindered further refinement of urban functional division. Third, the recognition capability of POI data is comparatively limited in areas with low building density and at the urban periphery. Owing to the spatial distribution characteristics of the data, this study only focuses on the central urban area with a dense transportation network and high population density. Therefore, we did not discuss the characteristics of functional zones across the entire Lanzhou City. In the future, combining multi-temporal data to analyze the spatiotemporal evolution characteristics and driving mechanisms of urban functional zones is necessary.

4. Conclusions

This study uses Lanzhou City, the central city of inland Northwest China, as a case study. Starting from integrating the spatiotemporal characteristics of human activities with the distribution of urban functional facilities and coupling mobile signaling with POI and other multisource data, we propose a multifactorial weighted kernel density calculation model based on ‘human activity–land feature area–public awareness’ for identifying urban functional zones. The study results indicate the following:

(1) The detailed classification of urban functions, integrating day–night variations in functional zone utilization intensity, POI categories, and urban land use types, is reasonable. Since POI data lack information such as residential demand level, scale, and influence scope, the employment of a multi-factor weighted kernel density calculation model based on ‘human activity–land feature area–public awareness’ can moderately improve the identification accuracy. The identification results for various types of functional areas were satisfactory, achieving an overall accuracy of 85%, confirming the effectiveness of the method. The urban functional zone identification method proposed in this study can assist urban planners in refining the identification of discrepancies between the current and actual urban land uses, thereby providing a foundation for the sustainable use and development of urban land resources.

(2) In the central urban area of Lanzhou City, there are a total of 14 functional types, including 6 types of single functional zones and 8 types of mixed functional zones. Overall, a spatial pattern of ‘single–mixed’ synergistic functional development is formed, predominantly composed of single functional zones nested within mixed functional zones. Single functional zones are primarily industrial zones and public services zones; the mixed functional zones are primarily combinations of public services and commercial zones, as well as residential and public service zones; and mixed functional zones comprise only half the area of single functional zones, indicating a relatively low degree of urban functional mixing. City managers should implement spatial planning adjustments to enhance the mix of urban functions by integrating additional public and commercial services near residential zones, to promote the sustainable development of urban resources.

(3) Within the public services zone, science, education, culture, and sports functions are mainly distributed in Chengguan District and Anning District, where higher education institutions are concentrated; healthcare functions are mainly concentrated in Chengguan District, Qilihe District, and Xigu District; and the service functions of government agencies are mainly located in Chengguan District, also known as the old city. Within the commercial zone, catering service functions are mainly concentrated on Zhangyelu Street in Chengguan District, Xihu Street in Qilihe District, and Xigucheng Street in Xigu District; financial service functions are mainly concentrated in Chengguan District; and life service functions are mainly distributed along the intersections of the city’s primary and secondary roads. The disparities in public and commercial service functions between districts in central Lanzhou are significant. The public service system in Lanzhou needs further enhancement to promote sustainable socioeconomic development across all districts.