POI Data–Driven Identification and Representation of Production–Living–Ecological Spaces at the Urban and Peri–Urban Scale: A Case Study of the Hohhot–Baotou–Ordos–Yulin Urban Agglomeration

Abstract

The identification of the multifunctional combination of production–living–ecological spaces (PLES) in urban agglomerations, particularly in urban cores and peri–urban areas, is a critical issue in the urbanization process. This study, using the Hohhot–Baotou–Ordos–Yulin (HBOY) urban agglomeration, a key node in China’s “Two Horizontals and Three Verticals” urbanization strategy, proposes a hexagonal grid–based PLES quantification framework using POI data. A three–level POI classification system was developed, with functional element weights determined via the Analytic Hierarchy Process and public perception surveys. The framework quantifies PLES within hexagonal grids and analyzes its patterns and functional coupling mechanisms using spatial overlay, Average Nearest Neighbor Index (ANNI), kernel density analysis, and spatial autocorrelation analysis. The following results were obtained. (1) PLES classification accuracy reached 90.83%, confirming the reliability of the method. (2) The HBOY urban agglomeration exhibits a dominant production space (40.84%), balanced living and ecological spaces (29.37% and 29.36%, respectively), and a severe shortage of mixed spaces (0.43%). (3) Production and living spaces show significant clustering ($ANNI$ ≤ 0.581), mixed spaces follow ($ANNI$ = 0.660), and ecological spaces are relatively evenly distributed ($ANNI$ = 0.870). (4) The spatial distribution patterns show that production and living spaces exhibit “core concentration with peripheral dispersion”, ecological spaces show “block concentration with point–like distribution”, and mixed spaces show “point–like dispersion”. (5) Production and living spaces exhibit strong spatial autocorrelation ($Moran’s I$ > 0.7) and the highest spatial correlation ($Bivariate Moran’s I$ = 0.692), while the spatial correlation with ecological spaces is weakest ($Bivariate Moran’s I$ = 0.150). The proposed PLES identification framework, with its efficiency and dynamic updating potential, provides an innovative approach to urban spatial governance under the global Sustainable Development Goals. The findings offer integrated decision–making support for spatial diagnosis and functional regulation in the ecologically vulnerable areas of northwest China’s new urbanization.

1. Introduction

With the rapid development of industrialization and the acceleration of urbanization, the complex reorganization of urban spaces and the increasing distribution of innovative resources at the urban periphery have become key areas of focus [1]. Moreover, suburbanization has emerged as a new phenomenon in the 21st century, especially in suburban agricultural areas, where imbalances between urbanization, industrialization, agricultural resources, and the ecological environment have become more prominent [2]. To address these challenges, scientists worldwide have conducted research from environmental [3], economic [4], and social [5] perspectives to promote multifunctional and sustainable land use, aiming to achieve coordinated urban development. The United Nations’ 2030 Agenda for Sustainable Development proposed new objectives, emphasizing the assessment of land sustainability within the global sustainable development framework, with a focus on three key dimensions: intensive production, harmonious living, and ecological balance [6]. In recent years, the Chinese government has also released a series of policies advocating the gradual shift from production–dominated spaces to the coordinated development of production–living–ecological spaces (PLES) [7,8,9]. In response to these policy demands, the rapid, precise, and widespread identification of PLES structures and components, along with their quantitative analysis, is crucial for optimizing urban spatial structures and ensuring sustainable land use.

The formation mechanism of production–living–ecological spaces (PLES) is shown in Figure 1, originating from the synergistic coupling of social, economic, and environmental systems, and forming a multifunctional composite through land resources as the carrier [10]. International research on land functions has shifted from a single–dimensional approach to a multifunctional synergy framework. For example, Gebhard [11], De Groot [12], and Chen et al. [13] proposed a triadic division of production, ecological, and social functions. Chinese scholars such as Zhang et al. [14] classified land use into production, living, and ecological spaces from a land use function perspective, corresponding to industrial and agricultural products, social security, and ecological services, respectively. Zhu et al. [15], from a human activity perspective, defined production space as areas for production and business activities, living space as areas for residence and consumption, and ecological space as areas for providing ecological products. As shown in Figure 2, PLES functions at the micro–scale (such as counties [16], townships [17], and administrative villages [18]) are relatively simple, while at macro and meso–scales (such as national [19], provincial [20], urban agglomeration [21], and city [22] levels), spatial interactions create complex multifunctional combinations [13]. In summary, the complexity of the multifunctional combination of PLES at the urban agglomeration scale presents a significant challenge to existing PLES identification methods.

Currently, there are two main methods for PLES identification: the aggregation classification method and the quantitative measurement method. The aggregation classification method categorizes land use based on existing conditions. Although this method facilitates the connection between land use functions and urban land classification standards, its accuracy is limited, and it often exhibits a certain degree of delay. Additionally, significant differences in the accuracy of land use data and classification systems make it difficult to compare results across different studies [23,24]. On the other hand, the quantitative measurement method identifies different functions by constructing an evaluation system, offering higher accuracy. However, this method involves a relatively complex data standardization process and requires further improvement in multi–scale integration [25,26].

POI data, represented by geographic entity attributes such as facility type and spatial location, characterizes human activities and functional distribution, becoming a core tool for urban spatial analysis [27,28,29]. Scholars have applied POI data in urban heat island analysis [30], tracking small–scale emission sources [31], and industrial land renewal [32]. With its high precision, real–time nature, and extensive coverage, POI data offers a new path for PLES identification. However, in the application of POI data for PLES identification, the primary identification units are global equal–area grids and blocks [33], leading to limitations due to the inherent flaws of traditional spatial units. Equal–area grids suffer from “hot grid” and “cold grid” phenomena caused by the Earth’s curvature, leading to data bias at higher latitudes [34], while the block method, which generates land parcels defined by road networks, is complex to implement and its accuracy is affected by road network density, limiting the precision of the results.

To address the limitations of existing methods, this study innovatively employs a hexagonal grid unit with a 1000 m interval [35], whose equal–area characteristics and neighborhood consistency overcome the limitations of traditional units, significantly improving the ability to capture local spatial variation [36]. This method not only addresses the high–latitude data bias issue but also avoids the road network density limitations of the block unit, providing a new technological path for identifying PLES in the urban cores and peri–urban areas of urban agglomerations. This study constructs a PLES classification system based on POI data, collects expert and stakeholder scoring matrices, and assigns weights through the Analytic Hierarchy Process (AHP). Additionally, comprehensive weights are obtained by combining a public perception survey on the size of geographic entities. The production, living, and ecological spaces are quantified, followed by a systematic analysis using the ArcGIS spatial overlay module and $Moran’s I$ autocorrelation model.

We use the Hohhot–Baotou–Ordos–Yulin (HBOY) urban agglomeration in northwest China as a case study. Over the past two decades, this region has experienced rapid socio–economic development, rapid urban expansion, and diverse industrial, agricultural, and pastoral production modes, yet there has been little research on PLES. The main objective of this study is to construct a POI–based PLES classification system for high–precision identification, explore the spatial distribution and differentiation patterns of PLES in the HBOY urban agglomeration under a 1000 m hexagonal grid, and systematically analyze the spatial coupling relationships between production, living, and ecological functions. The framework developed in this study not only provides a quantitative example for high–precision grid–based PLES identification at the national urban agglomeration level but also fills the research gap in multifunctional land use theory in northwest China. Our findings provide foundational support for sustainable development and spatial planning optimization in urban agglomerations and offer important reference insights for optimizing new urbanization construction.

2. Materials and Methods

2.1. Study Area

2.1.1. Overview of the Study Area

The Hohhot–Baotou–Ordos–Yulin (HBOY) urban agglomeration (106°28′ E—112°18′ E, 36°49′ N—42°44′ N) is located in the northwest of China, encompassing the cities of Hohhot, Baotou, and Ordos in Inner Mongolia, as well as Yulin in Shaanxi Province (Figure 3). Covering an area of 175,000 square kilometers, the agglomeration administers 14 districts and 25 counties. As of 2023, the permanent population is 12.19 million, with a regional GDP of RMB 2.1 trillion, accounting for 1.67% of the national total. According to the “HBOY Urban Agglomeration Development Plan”, the urban agglomeration aims to promote the coordinated development of the economy, ecology, and industry, optimize production efficiency, improve living quality, and enhance ecological protection, laying the foundation for open cooperation and sustainable development in the central and western regions of China [37]. Therefore, this study method focuses on high–frequency human activity areas within the HBOY urban agglomeration (urban core and peri–urban areas) to promote the balanced development of PLES in these regions, thereby driving the sustainable development of the HBOY urban agglomeration.

2.1.2. Data Sources and Preprocessing

The vector boundary data are sourced from the Tianditu Center (https://www.tianditu.gov.cn, accessed on 25 February 2025) in GeoJSON format, with map review number GS(2024)0650. The data have only undergone re–projection and format conversion, with no modifications to the map. The digital elevation model is derived from the China 1 km resolution digital elevation model dataset provided by the National Cryosphere Desert Data Center (https://www.ncdc.ac.cn/portal/, accessed on 25 February 2025) [38]. The land cover dataset is obtained from the “30 m annual land cover datasets and its dynamics in China from 1985 to 2023” (https://zenodo.org, accessed on 25 February 2025) [39].

The Jilin–1 satellite imagery and POI data are both sourced from the BiGEMAP GIS Office (Chengdu, China). A total of 230,738 POI records were obtained, including five attributes: name, address, phone number, category, and coordinates. According to the classification system of BiGEMAP GIS Office, the POI data comprises 19 classification levels. After screening, redundant and duplicate entries were identified, and after preprocessing, a final set of 195,635 valid records was retained.

This study uses a 1000 m spaced hexagonal grid as the unit for PLES identification, totaling 341,183 study units. Through spatial connection, 12,947 valid grids containing POI data were selected, with the spatial extent defined as follows.

• Urban Core Areas

The spatial range includes the entire administrative boundaries of the 14 districts in the HBOY urban agglomeration, specifically including:

Baotou City (6 districts)—Baiyun Obo Mining, Donghe, Jiuyuan, Kundulun, Shiguai, Qingshan; Ordos City (2 districts)—Dongsheng, Kangbashi; Hohhot City (4 districts)—Huimin, Saihan, Yuquan, Xincheng; Yulin City (2 districts)—Hengshan, Yuyang.

Functional determination: Hexagonal grids containing POI data within the administrative boundaries of the aforementioned districts.

2.

Peri–Urban Areas

The spatial range includes the entire administrative boundaries of the 25 counties within the HBOY urban agglomeration, specifically including:

Baotou City (3 counties)—Darhan Muminggan United, Guyang, Tumed Right; Ordos City (7 counties)—Dalad, Otog, Otog Front, Hanggin, Uxin, Ejin Horo, Jungar; Hohhot City (5 counties)—Helingeer, Qingshuihe, Tumed Left, Tuoketuo, Wuchuan; Yulin City (10 counties)—Dingbian, Fugu, Jia, Jingbian, Mizhi, Qingjian, Shenmu, Suide, Wubu, Zizhou.

Functional determination: Hexagonal grids containing POI data within the administrative boundaries of the above–mentioned counties.

3.

Non–Study Areas

Areas without POI coverage: Hexagonal grids without POI data within the administrative boundaries of the 14 urban districts and 25 counties.

2.2. Methods

This study constructs a framework for PLES identification and analysis, integrating POI data and GIS spatial analysis, which includes four stages: data preparation, ArcGIS spatial integration, result validation, and PLES feature analysis. The specific process is outlined as follows (Figure 4).

2.2.1. PLES Identification Based on POI Data

• Data Preparation and Classification System Construction

Based on the formation mechanism of PLES (Figure 1) and the corresponding relationship between production, living, and ecological functions (Figure 2), and in conjunction with studies such as the Urban Land Classification and Planning Construction Land Standards (GB 50137–2011) [40] and Land Use Status Classification (GB/T 21010–2017) [10,41,42,43], a three–level POI classification system was constructed (

Table 1. POI classification system and relevance value (α), influence value (β), and comprehensive weight (W) data.

Objective Layer	Criterion Layer	Element Layer	Industry Classification	Relevance Value (α)	Impact Value (β)	Comprehensive Weight (W)
Production Space	Commercial Space	Company Enterprises	Business buildings, advertising, decoration, construction companies, agricultural and forestry bases, etc.	0.0918	39	3.6077
		Financial Insurance	Banks, ATMs, investment management, credit unions, pawnshops, etc.	0.1113	36	4.0196
	Industrial Space	Factories	Factories, mines, workshops, etc.	0.0883	33	2.9233
		Warehousing and Logistics	Warehouses, logistics, railway stations, etc.	0.0939	29	2.7339
		Automotive Services	Car repair, car sales, car beauty services, car parts, car rentals, car testing centers, etc.	0.0936	27	2.5586
	Management Space	Government Agencies	Government agencies, etc.	0.2605	35	9.0916
	Transportation Space	Transportation	Subway stations, bus stations, parking lots, etc.	0.2605	40	10.4463
Living Space	Residential Space	Housing	Residential areas, commercial residential areas, communities, villas, etc.	0.1192	53	6.2820
	Life Service Space	Retail Stores	Retail stores, shops, specialty stores, convenience stores, etc.	0.0997	40	4.0104
		Supermarkets and Shopping	Shopping centers, department stores, home improvement stores, electronics stores, supermarkets, etc.	0.0589	31	1.8489
		Dining Services	Chinese restaurants, international restaurants, fast–food outlets, bakeries, teahouses, bars, cafes, etc.	0.0792	38	2.9865
		Accommodation Services	Budget hotels, apartment hotels, star hotels, homestays, etc.	0.0530	39	2.0513
		Life Services	Beauty salons, photo studios, photography shops, logistics companies (delivery stations), telecom business halls, etc.	0.0959	35	3.3761
		Medical and Healthcare	General hospitals, specialist hospitals, veterinary clinics, pharmacies, clinics, disease control centers, etc.	0.1411	33	4.6902
		Science and Cultural Spaces	Schools, museums, libraries, science centers, cultural centers, exhibition halls, etc.	0.1142	40	4.5255
		Sports and Recreation	Sports facilities, massage parlors, internet cafes, KTV, rural accommodations, movie theaters, resorts, etc.	0.0832	31	2.6170
		Public Facilities	Public restrooms, newspaper kiosks, etc.	0.0832	38	3.1680
		Public Squares	Leisure squares, etc.	0.0722	31	2.2629
Ecological Space	Ecological Spaces	Parks and Green Spaces	Parks, zoos, botanical gardens, aquariums, etc.	0.3542	29	10.3948
		Scenic Spots	Natural sites, historical sites, religious venues, national scenic spots, beach areas, etc.	0.6458	37	24.1735

). The target level includes three main functional spaces: production space, living space, and ecological space. The standard level consists of 7 subcategories of functional spaces; the element level lists 20 specific industry classification elements.

2.

Comprehensive Weight Calculation

(1) Calculation of Correlation Value ($\alpha$)

The correlation value ($\alpha$) is determined using the Analytic Hierarchy Process (AHP), with the following steps. First, expert scoring is conducted: eight experts (two from urban planning, land management, ecology, and stakeholders each) are invited to use a nine–point scale method (1 = equally important, 9 = extremely important) to perform pairwise comparisons between the standard and element levels, constructing a judgment matrix. Next, the consistency ratio (CR) for each matrix is calculated, and invalid data with CR ≥ 0.1 is excluded. Then, the arithmetic mean is calculated for the valid matrices to obtain the individual expert weights (Equation (1)). Finally, weight aggregation is performed by calculating the arithmetic mean of the weights ($w_{ik}$) of the eight experts to determine the final $\alpha$ value (Equation (2)).

$$w_{ik}={\displaystyle \frac{1}{n}}\sum _{j=1}^n{\alpha }_{ij}$$

(1)

$${\alpha }_i={\displaystyle \frac{1}{8}}\sum _{k=1}^8{w}_{ik}$$

(2)

where $w_{ik}$ is the individual expert weight, ${\alpha }_{ij}$ is the element of the judgment matrix, and $n$ is the order of the matrix.

(2) Calculation of Impact Value ($\beta$)

Due to the area differences in various geographic entities, their impact on PLES also varies [44,45,46]. Therefore, a survey questionnaire was designed (Element Layer in Table 1) to assess the public’s perception of the area of geographic entities, quantifying the subjective perception scores (ranging from 1 to 100) for different POI data entities. After min–max normalization, the average value is taken as $\beta$.

(3) Calculation of Comprehensive Weight ($W$)

That is, the product of the correlation value and the impact value, as shown in the following formula:

$$W_i={\alpha }_i\times {\beta }_i$$

(3)

3.

Spatial Integration and Function Identification

First, a 1000 m hexagonal grid is created using QGIS3.36.1 to cover the study area. Then, ArcGIS Pro 3.0.1 is used to spatially integrate the pre–processed and reclassified POI data with the hexagonal grid of the study area. Next, the number of POIs within each grid cell is counted. Finally, the number of PLES functions within each grid cell is calculated using Formula (4):

$$Z_i=S_i\times W_i,i=1,2,3,\dots \dots ,n$$

(4)

where $Z_i$ represents the number of PLES functions for the $i$-th element in the grid cell, $S_i$ is the number of POIs for the $i$-th element in the grid cell, and $W_i$ is the comprehensive weight of the $i$-th element.

By summing the function counts of each element, the total number of ecological, production, and living functions in the grid cell can be obtained. The formula is as follows:

$$Y=\sum _i^n{S}_i\times W_i,i=1,2,3,\dots \dots n$$

(5)

where $Y$ is the total sum of the function counts of all elements in the target layer within the grid cell.

The identification results of production functions (Figure 5), living functions (Figure 6), and ecological functions (Figure 7) are shown below.

By performing spatial overlay on the calculated results of ecological, production, and living functions, production space, living space, ecological space, and mixed space are identified using spatial analysis and the volumetric ratio method. The formula for the volumetric ratio method is as follows:

$$Q_j={\displaystyle \frac{Y_j}{{\sum }_j^n{Y}_j}},j=1,2,3$$

(6)

where $Q_j$ represents the proportion of the $j$th functional unit, and $Y_j$ represents the number of the $j$th functional unit in the grid. If $Q_j$ ≥ 50%, the function is considered dominant, and the grid cell is classified as a single–function space. If $Q_j$ < 50%, it is classified as a mixed–function area [10]. If there is no POI data within the grid (i.e., $Q_j$ is empty), it is classified as a no–data area [44].

In mixed–function areas, three types are formed based on the priority of the “Three Lines”: ecological–production space, production–living space, and ecological–living space. Ecological–production space prioritizes the integration of ecological protection red lines and farmland protection lines to ensure ecological and food security; production–living space combines farmland protection and urban development boundaries to balance agriculture and urban development; ecological–living space emphasizes the coordination between ecological protection and living needs [47,48,49].

4.

Result Validation

To validate the accuracy and reliability of the identification results, this study conducted a random sampling test and ensured consistency by cross–referencing with Jilin–1 satellite imagery (September 2024) with a resolution of 0.597 m.

2.2.2. Spatial Pattern Analysis

• Distribution Pattern Analysis

The Average Nearest Neighbor Index (ANNI) is used to determine spatial clustering, which can reveal characteristics such as clustering, randomness, or uniformity [50]. Smaller values indicate stronger clustering, values close to 1 represent uniform distribution, and larger values indicate stronger dispersion. In this study, ANNI is used to analyze the distribution patterns of production, living, ecological, and mixed spaces. The formula is as follows:

$$ANN{I}_i={\displaystyle \frac{{\sum }_{j=1}^{N_i}{d}_{ij}}{N_i\cdot d_e}}$$

(7)

where $i \in \{1, 2, 3, 4\}$ represents the Average Nearest Neighbor Index for production space ($i=1$), living space ($i=2$), ecological space ($i=3$), and mixed space ($i=4$); $N_i$ is the number of points in the $i$-th space; $d_{ij}$ is the distance from the $j$-th point in the $i$-th space to its nearest neighbor; and $de$ is the expected nearest neighbor distance, calculated as follows:

$$d_e={\displaystyle \frac{0.5}{\sqrt{\frac{N_i}{A}}}}$$

(8)

where $A$ is the area of the study region. By calculating the $ANN{I}_i$ for each space type, the distribution patterns of different space types can be compared.

2.

Kernel Density Analysis

Kernel density estimation is used to estimate the density distribution of spatial points, thereby identifying hotspot areas and revealing the distribution patterns of production space, living space, ecological space, and mixed space [50]. The formula is as follows:

$$f\left(x\right)={\displaystyle \frac{1}{n{h}^d}}\sum _{i=1}^{n}K\left({\displaystyle \frac{x-x_i}{h}}\right)$$

(9)

where $f\left(x\right)$ represents the density estimate at location $x$; $n$ is the number of points; $h$ is the kernel bandwidth, defining the influence range of the kernel function; $d$ is the dimensionality of the area; $x_i$ is the coordinate of a point; and $K$ is the quartic kernel function, which has its maximum value at the point and decreases with increasing distance, becoming zero outside the specified radius.

3.

Spatial Autocorrelation Analysis

Spatial autocorrelation analysis is used to describe the spatial correlation of variables by calculating $Moran’s I$, which measures the degree of spatial clustering or dispersion. The value of $Moran’s I$ ranges from [−1, 1]. When 0 < $Moran’s I$ < 1, it indicates spatial clustering; when $Moran’s I$ = 0, it indicates random distribution; and when −1 < $Moran’s I$ < 0, it indicates spatial dispersion [51,52].

Based on the Python 3.11 environment with the geopandas, esda, and libpysal packages, this study calculates global $Moran’s I$ and bivariate $Moran’s I$ to analyze the spatial clustering degree and interactions of production, living, and ecological spaces.

Global $Moran’s I$ is used to evaluate the clustering degree of each spatial element, and the formula is as follows:

$$I={\displaystyle \frac{N}{{\sum }_{i=1}^N{\sum }_{j=1}^N{w}_{ij}}}\cdot {\displaystyle \frac{{\sum }_{i=1}^N{\sum }_{j=1}^N{w}_{ij}\left(x_i-\overline{x}\right)\left(x_j-\overline{x}\right)}{{\sum }_{i=1}^N(x_i-\overline{x}{)}^2}}$$

(10)

where $N$ is the total number of regions, $x_i$ and $x_j$ are the observation values of region $i$ and region $j$, $\overline{x}$ is the mean value, and $w_{ij}$ is the spatial weight.

Bivariate $Moran’s I$ is used to quantify the interaction between different spatial elements, and the formula is as follows:

$$I_{xy}={\displaystyle \frac{{\sum }_{i=1}^N{\sum }_{j=1}^N{w}_{ij}\left(x_i-\overline{x}\right)\left(y_j-\overline{y}\right)}{\sqrt{{\sum }_{i=1}^N(x_i-\overline{x}{)}^2\cdot {\sum }_{j=1}^N(y_j-\overline{y}{)}^2}}}$$

(11)

where $x_i$ and $y_j$ are the observation values of variables $x$ and $y$ in regions $i$ and $j$, respectively, $\overline{x}$ and $\overline{y}$ are the mean values of variables $x$ and $y$, and $w_{ij}$ is the spatial weight.

3. Results

3.1. Identification Results and Validation

3.1.1. Identification Results of Production–Living–Ecological Spaces

The identification results of PLES, based on POI data, are presented in Figure 8. The total area of production space is 4579.54 km², comprising 5288 units; the total area of living space is 3292.63 km², comprising 3802 units; the total area of ecological space is 3291.76 km², comprising 3801 units; and the total area of mixed space is 48.50 km², comprising 56 units. Within the mixed spaces, the total area of ecological–production space is 18.19 km², comprising 21 units; the total area of production–living space is 19.92 km², comprising 23 units; and the total area of ecological–living space is 10.39 km², comprising 12 units. The results suggest that production space is predominant in the HBOY urban agglomeration, while living space and ecological space cover similar areas, each constituting a significant proportion.

3.1.2. Validation of PLES

In this study, 120 hexagonal grid cells (numbered from 0 to 119), each covering an area of 0.87 square kilometers, were randomly selected as validation areas. The PLES identification results were then compared with satellite imagery. The validation outcomes revealed that 109 areas (72 fully consistent, 37 partially consistent) aligned with the satellite imagery, representing 90.83% of the total. However, 11 areas did not match the satellite imagery (

Table 2. Details of inconsistencies in production–living–ecological space identification results.

Validation Areas	Space Type	Assessment of Satellite Imagery Conditions	Validation Results
61	Living Space	Mountain peak, farmland	Inconsistent
62	Living Space	Natural landscape, farmhouse	Inconsistent
63	Living Space	Farmland, health clinic	Inconsistent
68	Living Space	Small residential area, farmland, mountain peak, health clinic	Inconsistent
74	Living Space	Factory, farmland	Inconsistent
78	Living Space	Orchard, farmland, homestead	Inconsistent
81	Living Space	Abandoned school, farmland, provincial road	Inconsistent
95	Production Space	Resort, lake	Inconsistent
111	Production Space	Natural landscape, few companies	Inconsistent
112	Production Space	Resort, natural landscape	Inconsistent
117	Production Space	Mountain peak, hotel	Inconsistent

).

This inconsistency is mainly attributed to the lack of POI data for farmland, mountains, and forests. For example, validation area 61 encompasses both peaks and farmland. The validation results indicate that the method of identifying PLES in urban agglomerations using POI data exhibits high accuracy and operational feasibility, as shown in Figure 9.

3.2. Spatial Distribution Analysis of PLES

3.2.1. Spatial Distribution Types of PLES

The PLES identification results were converted into point features and analyzed using the Average Nearest Neighbor Index (ANNI) method. The results show that the ANNI for production space is 0.581, significantly below 1, indicating a clustered distribution within the region. The ANNI for living space is 0.577, also suggesting clustering, reflecting the tendency for residential and living service spaces to aggregate. The ANNI for ecological space is 0.870, close to 1, indicating some degree of clustering but a relatively uniform distribution. The ANNI for mixed space is 0.660, suggesting moderate clustering, which lies between the clustering tendencies of production and living spaces and the uniformity of ecological space.

3.2.2. Spatial Distribution Patterns of PLES

Figure 10 illustrates the distribution characteristics of production, living, ecological, and mixed spaces within the study area.

Production spaces (Figure 10a) exhibit a distribution pattern defined by “core concentration and peripheral dispersion”. Core cities, including Hohhot (Saihan District, Xincheng District, Huimin District, Yuquan District), Baotou (Donghe District, Kundulun District, Qingshan District, Jiuyuan District), Ordos (Dongsheng District, Kangbashi District), and Yulin (Yuyang District, Hengshan District), show high production space density. Secondary production clusters have emerged in regions such as Tumed Left Banner in Hohhot, Shiguai District in Baotou, Ejin Horo Banner, Dalad Banner, and Jungar Banner in Ordos, as well as Jingbian County and Dingbian County in Yulin. In contrast, production spaces in the southern and western peripheral regions are more scattered.

Living spaces (Figure 10b) exhibit a “core concentration and peripheral dispersion” pattern. The density of living spaces is primarily concentrated in core cities, including Hohhot (Saihan District, Xincheng District, Huimin District, Yuquan District), Baotou (Donghe District, Kundulun District, Qingshan District, Jiuyuan District), Ordos (Dongsheng District, Kangbashi District), and Yulin (Yuyang District, Hengshan District). Secondary clusters are located in areas such as Tumed Left Banner in Hohhot, Shiguai District in Baotou, Dalad Banner in Ordos, and Shenmu City and Suide County in Yulin. In contrast, living spaces in the southern and western peripheral regions are more scattered.

Ecological spaces (Figure 10c) follow a “block concentration with point–like distribution” pattern. High–density ecological spaces are concentrated in Hohhot (Tumed Left Banner, Wuchuan County), Baotou (Shiguai District), and the southern part of Yulin (Hengshan District, Zizhou County), forming core areas for ecological functions. Secondary ecological clusters are found in Ordos (Dongsheng District, Kangbashi District, Jungar Banner) and Yulin (Shenmu City, Dingbian County). In contrast, ecological spaces in the western peripheral regions are more scattered.

Mixed spaces (Figure 10d) follow an overall “point–like dispersion” pattern. These spaces are mainly concentrated in core cities, including Hohhot (Saihan District, Xincheng District, Huimin District, Yuquan District), Baotou (Donghe District, Kundulun District, Qingshan District, Jiuyuan District), Ordos (Dongsheng District, Kangbashi District), and Yulin (Yuyang District, Hengshan District). Secondary mixed clusters are found in Tumed Left Banner in Hohhot and Dalad Banner in Ordos. Mixed spaces in other regions are relatively sparse.

3.3. Correlation of the Distribution Characteristics of PLES

A univariate global spatial autocorrelation analysis was performed for production spaces, living spaces, and ecological spaces (Figure 11). The results passed the 0.01 significance test, confirming significant spatial clustering characteristics. The $Moran’s I$ values for production spaces, living spaces, and ecological spaces were 0.735, 0.733, and 0.130, respectively, indicating significant spatial autocorrelation for all three. Among these, production spaces exhibited the strongest clustering.

To further investigate the relationships among clustered spaces, a bivariate global spatial autocorrelation analysis was performed for production spaces, living spaces, and ecological spaces. The results passed the 0.001 significance level test (Figure 11). The analysis revealed that the correlation between living spaces and production spaces was the highest, with an $Moran’s I$ value of 0.692. The correlation between ecological spaces and living spaces had an $Moran’s I$ value of 0.159, while the correlation between ecological spaces and production spaces was the lowest, with an $Moran’s I$ value of 0.150. This suggests that urban production and living spaces are mutually dependent and foster each other’s development. In contrast, the weaker clustering relationship between ecological spaces and production spaces may reflect the impact of extensive production methods on the ecological environment, leading to weaker spatial connections between the two.

A univariate and bivariate spatial autocorrelation analysis was performed for production space elements (Figure 12). The results passed the 0.01 significance level test, confirming significant clustering characteristics. In the univariate analysis, the $Moran’s I$ values (from high to low) were as follows: transportation (0.719) > companies and enterprises (0.564) > finance and insurance (0.532) > government agencies (0.457) > warehousing and logistics (0.312) > automotive services (0.300) > factories (0.201). Among these, transportation, companies and enterprises, and finance and insurance exhibited the highest clustering levels, indicating significant spatial concentration.

In the bivariate analysis, a $Moran’s I$ value greater than 0.5 indicates a strong spatial association between elements. Based on this criterion, the following conclusions were drawn. Elements closely related to transportation include companies and enterprises (0.542) and finance and insurance (0.553), suggesting that transportation facilities are primarily located in areas with a concentration of commercial and service facilities that support production activities. Elements closely related to finance and insurance include transportation (0.553) and government agencies (0.449), indicating that financial and insurance facilities show strong clustering effects around areas with convenient transportation and policy support. Elements closely related to companies and enterprises include transportation (0.542) and finance and insurance (0.500), demonstrating the significant attraction of transportation and financial services to companies and enterprises.

Overall, the univariate analysis shows that transportation, companies and enterprises, and finance and insurance exhibit high levels of clustering. The bivariate analysis reveals significant spatial associations among transportation, finance and insurance, and companies and enterprises, emphasizing their critical roles in the spatial layout of production spaces and providing a supportive spatial framework for production activities.

A univariate and bivariate spatial autocorrelation analysis was performed for living space elements (Figure 13). The results passed the 0.01 significance level test, confirming significant clustering characteristics. In the univariate analysis, the $Moran’s I$ values (from high to low) were as follows: residential areas (0.699) > living services (0.669) > catering services (0.632) > science, education, and culture (0.630) > healthcare (0.606) > supermarkets and shopping (0.452) > retail stores (0.436) > accommodation services (0.476) > sports and recreation (0.439) > public squares (0.141) > public facilities (0.124). Among these, residential areas, living services, and catering services exhibited the highest levels of clustering, indicating significant spatial concentration.

In the bivariate analysis, a Moran’s I value greater than 0.5 indicates a strong spatial association between elements. Based on this criterion, the following conclusions were drawn. Elements closely related to residential areas include living services (0.601), science, education, and culture (0.598), and catering services (0.578), indicating that living, educational, and catering facilities significantly attract residential areas. Elements closely related to living services include catering services (0.645) and science, education, and culture (0.609), demonstrating that living service facilities exhibit strong clustering effects around catering and educational facilities. Elements closely related to catering services include living services (0.645) and residential areas (0.578), suggesting that catering facilities are often located near living services and residential areas, providing convenient support for residents’ daily lives.

Overall, the univariate analysis shows that residential areas and living services exhibit significant spatial clustering. The bivariate analysis further reveals notable spatial associations among living space elements, emphasizing the important roles of living services, science, education, culture, and catering facilities in shaping the spatial layout of residential areas.

A univariate and bivariate spatial autocorrelation analysis was performed for ecological space elements. The results show that the spatial autocorrelation of each element is relatively low, with no significant clustering characteristics. In the univariate analysis, the $Moran’s I$ value for parks and green spaces is 0.052, and for scenic spots, it is 0.128, both indicating dispersed distributions without clear clustering. In the bivariate analysis, the $Moran’s I$ value between parks and green spaces and scenic spots is 0.037, indicating a low spatial association and a lack of interconnectivity between the two.

Overall, ecological space elements in the region display a dispersed layout, with no significant spatial clustering or association characteristics.

4. Discussion

4.1. Analysis of PLES Identification Results Based on POI Data and Application Advantages

The HBOY urban agglomeration is located at the northern end of the “Baotou to Kunming” corridor in China’s “Two Horizontals and Three Verticals” urbanization strategy. It plays an important role in promoting the new pattern of Western China’s development, advancing new urbanization, and improving the layout of open development along the border. The “Two Horizontals and Three Verticals” urbanization strategy is a spatial development framework that integrates the overall development of eastern, central, and western China, as well as the north and south. By combining transportation axes with urban agglomerations, it facilitates the efficient flow of population, environment, and economic factors, providing strategic support for building a new development pattern [53]. However, unlike China’s macro–coordinated “Two Horizontals and Three Verticals” urban development strategy, which is unique to the country, urbanization in Western countries is primarily focused on urban districts, with an emphasis on local conditions. This approach prioritizes market–driven solutions for human concerns, industrial development, and multi–functional coordination with environmental resources [54]. Therefore, in advancing China’s urbanization strategy, we must not only consider the macro “Two Horizontals and Three Verticals” framework but also learn from Western countries in accounting for the spatial heterogeneity of each city in China, thus addressing the relationship between urbanization and ecological environments. This remains a critical challenge in both academic circles and government decision–making departments, and has become a global strategic issue, though it currently remains an underexplored and weak aspect of the issue [55].

This study, by constructing a POI–based PLES classification system, quantitatively analyzes the spatial distribution patterns of the economic, social, and ecological subsystems, revealing that the PLES distribution in the HBOY urban agglomeration presents a “production–dominated, ecological–living balance” feature (production space 40.84%, with living and ecological spaces close at 29.37% and 29.36%, respectively). The result reflects the high level of economic activity and industrialization in the HBOY urban agglomeration, with productive land dominating the spatial layout. Furthermore, these findings are consistent with those of Kong et al. [56], who pointed out that the concentration of productive land use in economically developed areas contributes to regional economic growth. Secondly, production space exhibits a “core concentrated, peripheral dispersed” distribution pattern, which is highly consistent with the “core–periphery” structure of industrial production space in the Pearl River Delta urban agglomeration [57], indicating the cross–regional similarity of production space aggregation in economically oriented regions. However, this also highlights the importance of balancing the development of ecological and living spaces. Notably, the implementation of national projects such as the National Forest Protection Project, Beijing–Tianjin Sandstorm Source Control Project, the 3–North Shelter Forest Program, the Grain for Green Program, and Grassland Restoration Program may have positively influenced the balanced distribution of living and ecological spaces in the HBOY urban agglomeration [58]. The coordinated development of living and ecological spaces is crucial for regional sustainability [59], as it not only helps improve ecological quality but also raises the regional standard of living, ensuring sustainable development. The proportion of mixed space in the HBOY urban agglomeration is only 0.43%, indicating clear functional zoning of PLES and distinct land use divisions. The low proportion of mixed space may be due to stricter land use control policies in the study area, and compared to more developed regions, such as the Yangtze River Delta urban agglomeration, which has a more integrated industrial mixed layout, there are still considerable gaps in areas such as labor, technology, and industrial coordination [60,61]. Overall, the spatial layout of this urban agglomeration provides a solid foundation for economic development, but more attention needs to be given to the protection and coordinated layout of living and ecological spaces to promote regional sustainable development.

The ANNI for mixed space is 0.660, indicating moderate clustering, which lies between the stronger clustering of production space (0.581) and living space (0.577) and the relatively uniform distribution of ecological space (0.870). This is closely related to the differences in regional economic development stages. In highly urbanized areas (e.g., Guangdong–Hong Kong–Macao [62]), mixed land use forms dense clusters through functional overlap, whereas in areas with stronger ecological constraints (e.g., Ningxia [63]), mixed space shows a dispersed distribution due to the single nature of the industrial types. Production space and living space exhibit strong clustering because they bear the core economic and social functions, typically concentrated in the high–density areas of core cities. This finding aligns with the spatial differentiation pattern of the “core–periphery” theory [64], which states that production factors gather in core cities to take advantage of economies of scale. In contrast, ecological space focuses on environmental functions, presenting a “block concentration, point distribution” pattern, which has a minimal effect on the distribution of mixed space. The distribution characteristics of mixed space are consistent with the view of Hao et al. [65], who argue that the formation of multifunctional areas typically relies on the concentration of economic and service functions in core areas. Since mixed space integrates production, living, and ecological functions, its spatial demand is more diverse, typically concentrated in industrial parks, commercial districts, and integrated development zones. The distribution of these areas is more influenced by the aggregation of these functions. This pattern is similarly validated in studies of mixed space distribution in the Chengdu–Chongqing urban agglomeration [55], but in those areas, the centers of aggregation often extend along transportation axes, while in HBOY, it presents a “island–like” distribution around resource–based industrial parks, highlighting the path dependence of mixed space development in energy–based urban agglomerations.

The spatial autocorrelation analysis of PLES shows that the synergy between living space and production space is the strongest ($Moran’s I$ = 0.692), while the synergy between ecological space and living space is weaker ($Moran’s I$ = 0.159), and the synergy between ecological space and production space is the weakest ($Moran’s I$ = 0.150). The distribution of production and living spaces exhibits strong interdependence, particularly in high–density aggregation areas such as Hohhot (Saihan District, Xincheng District, Huimin District, Yuquan District) and Baotou (Donghe District, Kundulun District, Qingshan District, Jiuyuan District). The overlap of production space and living space in these areas reflects the polarization effect of resource concentration and functional distribution in core cities. This phenomenon is consistent with the “integration of production and living” characteristic in core cities. Existing studies suggest that the high concentration of economic activities and living functions in core cities strengthens functional coordination and spatial interdependence within the region [66]. In contrast, the high–density distribution of ecological space is more dispersed, primarily concentrated in areas such as Hohhot (e.g., Tumd Left Banner, Wuchuan County), Baotou (e.g., Shiguai District), and southern Yulin (e.g., Hengshan County, Zizhou County), with minimal overlap with high–density production and living spaces. This reflects the potential land use conflicts between ecological functions and production space. Similar spatial conflicts have also been observed in studies of other regions, indicating that balancing ecological protection and economic development is an urgent issue to be addressed in current urban agglomeration planning [67].

The validation of PLES identification results based on Jilin–1 satellite imagery shows an accuracy of 90.83%, demonstrating that the POI–based identification method has high precision and feasibility. Compared to other methods, POI data, with its large capacity, wide coverage, and real–time updates, makes PLES identification in urban agglomerations simpler, more flexible, and accurate. Sun et al. [68] conducted a systematic review of the POI matching method, and Fu et al. [43] elaborated on the application of POI data in urban planning in Wuhan. Both studies emphasize the advantages of POI data in understanding human–environment interactions and supporting urban management, providing strong technical support for regional resource management and policy formulation.

The findings of this study provide several insights for optimizing the spatial planning of the HBOY urban agglomeration. First, the layout of production space should be optimized to address the current overrepresentation of production space. By achieving a balance between production, living, and ecological functions, it will be possible to promote industrial upgrading, high–quality development, and improved land use efficiency. Second, the livability of living spaces should be improved by enhancing residential environments and optimizing the layout of public service facilities. Reducing the high concentration of living spaces will help alleviate the resource and environmental pressures in core cities and promote more sustainable urban living conditions. Third, the protection of ecological spaces should be strengthened to prevent the expansion of production space at the expense of ecological areas. Ecological corridors and green networks should be scientifically planned to enhance the connectivity of “block concentrated” ecological areas, thereby improving their ecological service functions and ensuring long–term environmental sustainability. Finally, the multifunctionality of mixed spaces should be fully utilized. Industrial parks and commercial districts should be prioritized, promoting functional integration and reducing the disorderly expansion of land in urban planning. This approach will not only improve land use efficiency but also support the coordinated development of urban agglomerations.

4.2. Limitations and Future Prospects

However, this study also has certain limitations. The method shows stronger applicability in highly urbanized and developed urban agglomerations [69], as the abundance of POI data and functional intensity exhibit spatial coupling. In ecologically fragile areas, however, due to the volunteer geographic information nature of POI data, coverage bias may occur [70], potentially leading to an underestimation of the true functional distribution in low–frequency human–environment interaction areas.

Despite achieving a validation accuracy of 90.83% with the proposed POI–hexagonal recognition framework for the HBOY urban agglomeration, limitations in POI data were found during the identification of urban core and peri–urban areas. In the core areas, it was primarily found that the production–living space had high quantity and identification accuracy, due to the high density of POI data for production and living aspects. However, in the peri–urban areas, limitations in the representation of ecological–mixed space were noted; ecological space depends on POI categories such as “parkland” and “scenic areas”, which typically have larger physical areas but are represented as a single data point in POI data. This difference in granularity between the core and peri–urban areas reflects the lower sensitivity of POI data in identifying peri–urban areas and functional overlap regions. Therefore, to extend this method to regional scales (such as including more ecologically vulnerable areas), additional data sources should be incorporated, or a quadtree method can be used for merging in edge regions to alleviate recognition differences caused by varying spatial granularity across urban agglomeration regions.

In summary, future research can continue to explore the following directions. First, optimize multi–source data fusion by integrating remote sensing vegetation indices, mobile signaling, and other spatiotemporal big data to construct a “heaven–earth–space” collaborative observation system, addressing data gaps in remote areas. Second, incorporate policy simulation and risk warning studies by introducing system dynamics models to simulate spatial game processes under policy interventions such as ecological protection red lines and energy industry transformation, providing decision support for flexible land spatial planning. Finally, organically integrate the “core–periphery” theory with PLES theory to analyze how transportation corridors (such as the “Two Horizontal and Three Vertical” axis) influence population movement and ecological barriers, revealing the unique evolutionary patterns of human–environment relationships in western border urban agglomerations.

5. Conclusions

This study innovatively integrates POI data with hexagonal spatial grids to construct a PLES classification system, using the HBOY urban agglomeration as a case study. It uncovers the differentiation patterns of PLES in the core and peri–urban areas of resource–based urban agglomerations, providing essential scientific support for optimizing human–environment systems in arid and semi–arid regions. The key contributions are as follows.

(1) Method reliability validation: The constructed POI–hexagonal identification framework achieved a verification accuracy of 90.83% in the HBOY urban agglomeration, confirming the method’s significant advantage in characterizing the complex spatial functions of ecologically vulnerable urban agglomerations. It provides an efficient tool for dynamic monitoring of land use evolution in resource–based regions.

(2) Theoretical breakthrough in spatial patterns: The pattern, with production space dominating (40.84%), balanced ecological and living spaces (29.36% and 29.37%, respectively), and the lowest proportion of mixed space (0.43%), reveals the typical feature of resource–based urban agglomerations being “production–heavy and coordination–light”. The high aggregation of production and living spaces ($ANNI$ ≤ 0.581) and strong spatial synergy ($Moran’s I$ = 0.692) validate the applicability of the “core–periphery” theory in energy base cities. Meanwhile, the uniform distribution of ecological space ($ANNI$ = 0.870) and its weak synergy with production functions ($Moran’s I$ = 0.150) expose the deep contradictions between ecological security and industrial expansion in arid and semi–arid regions, filling the gap in existing theories related to human–environment relationships in resource–based urban agglomerations.

(3) Practical implications for planning decisions: The identified distribution patterns of production and living spaces show “core concentration, peripheral dispersion”, while ecological space exhibits “block concentration, point distribution”, and mixed space shows “point dispersion”. These patterns provide targeted guidance for optimizing the functional layout along the “resource corridor” of the HBOY urban agglomeration. For example: The core area of Hohhot–Baotou should control excessive industrial agglomeration, the ecologically fragile areas in southern Yulin should strictly limit the expansion of mining land, and the “point dispersion” feature of mixed space indicates the need to strengthen multi–functional development in industrial parks.

This study confirms that the integration of spatial computing technology and geographic big data can accurately capture the development challenges of resource–based urban agglomerations, providing a scientific paradigm for coordinating ecological civilization construction and new urbanization strategies.