Research on the Spatial Evolution and Planning Strategies of Green Belts in Metropolises

Abstract

Green belts in metropolises face a significant contradiction between ecological protection constraints and urban sprawl, necessitating effective planning and management. Existing studies have primarily focused on a single dimension, while the factors influencing the spatial evolution of green belts are complex and diverse. This study establishes a multi-objective quantitative analysis framework, utilizing quantitative analysis methods such as average nearest neighbor analysis, landscape ecological index analysis, land–use transition matrix, kernel density estimation, and spatial autocorrelation models. Taking the green belt area of Shijiazhuang as a case study, this research systematically analyzes the spatial evolution characteristics of the region from 2015 to 2024. The findings reveal spatial patterns such as the small-scale and dispersed expansion of industrial land, increasing fragmentation of ecological spaces, ongoing encroachment on agricultural land, differentiated growth of service industry spaces, and the uncontrolled sprawl of residential areas in villages and towns during rapid urbanization. These patterns lead to increased ecological risks, imbalanced urban–rural development, and lagging infrastructure. To address these challenges, this study proposes a planning strategy of “adjusting the primary industry, restricting the secondary industry, and promoting the tertiary industry,” aiming to resolve the conflict between ecological protection and urban expansion in metropolitan green belts, ensuring their orderly development. This research provides insights for the sustainable development of green belts in Metropolises of developing countries during the rapid urbanization process.

1. Introduction

The unchecked, disorderly sprawl of large cities continues to encroach upon surrounding green spaces, leading to the erosion of agricultural land, fragmentation of ecological spaces, and inefficiency in land-use. Brabe [1] notes that urban sprawl often results in the fragmentation of agricultural landscapes, weakening the integrity of ecosystems and thereby exacerbating environmental vulnerability. Beckers et al. [2] found that the expansion of urban built-up areas not only consumes high-quality farmland but also disrupts the continuity of ecological networks. Salvati et al. [3], through a comparative study of Europe and North America, emphasizes that such processes significantly increase regional ecological risks, threatening the long-term sustainability of urban environments.

To mitigate the negative effects of urban sprawl, many countries have adopted green belts as key spatial governance tools, aimed at curbing urban expansion and protecting green spaces. Han and Go [4] argues that countries such as the United Kingdom, South Korea, and Canada have, to some extent, curbed excessive urban expansion and preserved the integrity of ecological spaces through green belts policies. London serves as a representative case, with Amati and Yokohari [5] pointing out that since the 1950s, the city’s green belt has consistently worked to limit urban sprawl, safeguard open spaces, and strike a balance between development and conservation. However, recent studies have shown that the effectiveness of green belts is not fixed. Koster [6] indicates that while green belts have prevented the chaotic expansion of central urban areas, they have also suppressed housing supply, pushing development activities outward beyond the green belt boundaries. Bengston and Youn [7] highlight that the actual success of green belts heavily depends on the capacity for spatial governance, transportation integration, and housing market dynamics. As a spatial governance tool controlling urban expansion, the green belts must strike a balance between ecological preservation and socio-economic development.

From the international experience in metropolitan green belts planning, it is clear that green belts are not static; they evolve dynamically under the influence of institutional environments, land markets, and societal demands. A multi-objective balance must be achieved between ecological protection, urban expansion, and social equity. The rapid urbanization in China has intensified the loss of green belt farmland resources, continuously increasing environmental pressure, posing severe challenges to the ecological security and sustainable development of metropolises. Many cities have progressively initiated green belt construction. For example, Beijing began planning the first and second phases of its green belts in 1993 to create an ecological barrier for the city’s core area; Shanghai planned its outer-ring green belts to demarcate the boundaries of urban development; and Chengdu launched the “198” green ecological belts in 2003, integrating ecological restoration with spatial planning.

At the same time, discussions surrounding green belts have deepened, with the focus of research gradually shifting from spatial form planning to spatial policy design. Zhan et al. [8] applied a multi-index system to examine the contraction–expansion dynamics of two green belts in Beijing, revealing the complexity and dynamic nature of policy interventions in spatial morphology. Pourtaherian et al. [9] conducted a quantitative analysis and found that cities with established green belts experienced a significantly greater reduction in their average sprawl index between 2006 and 2015 than cities without such belts, with the inhibitory effect particularly evident in metropolises with populations exceeding one million. However, as implementation has advanced, rigid control models have increasingly revealed multiple dilemmas, prompting scholars to reassess policy adaptability. Smith [10] argued that insufficient housing supply within green belts has displaced development outward, increasing commuting-related carbon emissions by 23%. Walton [11] observed that although public participation can restrict residential construction, it may also exacerbate the shortage of affordable housing. Dockerill and Sturzaker [12] showed that freezing land supply contributed to a deficit of social housing and the proliferation of high-rise apartments, generating social tensions such as household dissatisfaction. Eswar [13] highlighted that Bangalore’s green belt neglected farmers’ livelihood transitions, resulting in the invisible conversion of non-agricultural land. Choi et al. [14] provided empirical evidence that South Korea’s “green belt + new town” policy was shaped by key political actors, producing irreversible path dependence.

To counteract the limitations of qualitative research, spatial policy analysis gradually shifts to quantitative methods. Jun [15] uses machine learning simulation and finds that lifting the green belts would have absorbed 6% of the metropolitan population and 13% of employment, but would have weakened the connectivity of ecological corridors by 20%. Therefore, quantitative policy evaluation is necessary. Lee and Yoon [16] conduct an economic analysis and find that lifting the green belts would increase the land values within 500 m by 11.2% and would depreciate peripheral lands due to the decrease in ecological services. Ma and Jin [17] disclose the cost of regulation of green belts and find that the strict implementation of regulation would have decreased the consumption by 2010 of 202 million USD, and reflects the rigid implementation and market forces conflicts. However, most quantitative studies are focused on one dimension and lack of systematic indicators.

Overall, qualitative analysis of green belts in metropolises is used to determine their functions, offering the advantage of providing in-depth insights into spatial cognition and governance logic. However, it struggles to address the complex realities of green belt space evolution. Quantitative analysis, applied in land-use studies of green belts, accurately identifies the characteristics of spatial evolution through spatial statistics and econometric models, but it often overlooks economic and social factors influencing spatial changes. Balagou et al. [18] noted that although green belts have been extensively studied in metropolitan areas of developed countries, they remain underexplored in developing regions, particularly in Africa, where empirical research is still scarce and fragmented. Moreover, Monstadt and Meijer [19] argued that existing studies often focus on single objectives—such as containment, housing, or ecology—without establishing an integrated, multi-criteria quantitative framework capable of addressing the complex governance interactions inherent in metropolitan green-belt systems. To bridge these gaps, this study combines both qualitative and quantitative analyses, using the green belts area of Shijiazhuang as a case study. A multi-objective quantitative analysis framework is developed to explore the spatial evolution characteristics and mechanisms between 2015 and 2024, and spatial optimization strategies are proposed. This research provides valuable insights for the sustainable development of green belts in metropolises of developing countries during rapid urbanization, enriching the global body of research on the planning of green belts in metropolises.

2. Methodology

2.1. Research Area

Shijiazhuang, the capital of Hebei Province, is situated in the south-central region of the province and serves as a core city within the world-class Beijing–Tianjin–Hebei urban agglomeration. By the end of 2024, the city’s permanent population reached 11.25 million, including 8.17 million urban residents, resulting in an urbanization rate of 72.66%. Between 2015 and 2024, the central urban area expanded from approximately 278 km² (China City Statistical Yearbook, 2016 [20]) to 660 km² (China City Statistical Yearbook, 2021 [21]). However, this rapid urban expansion has introduced multiple ecological risks, including encroachment on surrounding green spaces and water systems, occupation of agricultural land, and intensification of the urban heat island effect. To curb this disorderly sprawl, Shijiazhuang designated a 438.0 km² green belt within the urban–rural transition zone between the central urban area and the adjacent counties of Luquan, Luancheng, Gaocheng, and Zhengding (Figure 1). This study takes green belts of Shijiazhuang as the research subject, combining both qualitative and quantitative analyses to determine the evolution characteristics of industrial, ecological, agricultural, service, and residential spaces. The study reveals the mechanisms of evolution and proposes planning strategies.

2.2. Methods

To address the limitations of single-dimensional quantitative approaches, this study developed a multi-dimensional, comprehensive quantitative analysis method. Based on the evolutionary characteristics of different spatial types, five quantitative methods were employed: average nearest neighbor analysis, landscape pattern indices, land-use transfer matrix, kernel density estimation, and spatial autocorrelation. These methods were integrated through cross-validation and complementary enhancement to reliably characterize green belts spatial evolution and to inform spatial-optimization strategies.

2.2.1. Average Nearest Neighbor Analysis (ANN)

Average nearest neighbor analysis determines the amount of spatial clustering or dispersion of point features by comparing the observed mean distance to the expected mean distance if the points were randomly distributed. The formula is expressed as follows:

$$\begin{array}{c}R={\displaystyle \frac{d_0}{d_e}}={\displaystyle \frac{{{\displaystyle \sum }}_{i=1}^n{d}_i/n}{0.5/\sqrt{n/A}}}\end{array}$$

where $n$ is the number of points in the dataset, $A$ is the area in which the points are distributed, $d_0$ is the average observed distance, $d_0={\displaystyle \sum }_{i=1}^n{d}_i/n$; $d_e$ is the expected average distance, $d_e=0.5/\sqrt{n/A}$.

Most of the previous studies used density-based methods to analyze the spatial distribution of industrial,, but these methods could not quantitatively measure the degree of industrial agglomeration or dispersion. Therefore, this study used ANN tool in ArcMap 10.8. The significance tests based on the nearest neighbor ratio ($R$ value) and $Z$ score were used to determine whether the spatial distribution of industrial land was clustered, random or dispersed.

2.2.2. Landscape Ecological Index Analysis

• Mean Patch Area (AREA_MN)

The mean patch area represents the average size of all patches of a specific type within the study region (hm²), reflecting the typical scale of patches. A smaller value indicates smaller average patch sizes, suggesting a more fragmented landscape dominated by small patches. The calculation formula is:

$$AREA\_MN=\frac{1}{n}\sum _{i=1}^n{a}_i$$

where $a_i$ is the area of the $i$-th patch, and $n$ is the total number of patches in this type of landscape.

2.

Standard Deviation of Patch Area (AREA_SD)

This metric measures the standard deviation (hm² or km²) of the patch areas of a particular type within the region, indicating the variability in patch size distribution. Larger values reflect higher variability in patch size, whereas smaller values indicate sizes closer to the mean. The calculation formula is:

$$AREA\_SD=\sqrt{{\displaystyle \frac{1}{n}}{\displaystyle \sum }_{i=1}^n{(a_i-AREA\_MN)}^2}$$

where $a_i$ is the area of the $i$-th patch, and $AREA\_MN$ is the average patch area.

3.

Patch Density Index (PD)

The patch density index refers to the number of patches per unit area. It is calculated in two forms: the ratio of ecological patches to the total regional area (PD₁, patches/km²) and the ratio of ecological patches to the ecological land area (PD₂, patches/km²). Higher values suggest greater fragmentation. The calculation formula is:

$$PD={\displaystyle \frac{n}{S}}$$

where $\begin{array}{c}n\end{array}$ is the total number of patches; $S$ is the total area of the landscape.

4.

Largest Patch Index (LPI)

The largest patch index measures the proportion (%) of the largest patch relative to the total landscape area, reflecting the dominance of core patches in the overall landscape pattern. Larger values indicate the presence of dominant, continuous patches, while smaller values suggest fragmentation of core patches. The calculation formula is:

$$\begin{array}{c}LPI={\displaystyle \frac{\underset{1\le i\le n}{max}\left(a_i\right)}{A}}\times 100\%\end{array}$$

where $\max (a_i)$ is the maximum patch area, and $A$ is the total landscape area.

Previous studies on ecological spaces have concentrated on area changes, which hinders an understanding of the internal spatial patterns and structural characteristics. This study uses a correlated set of landscape metrics—mean patch size, standard deviation of patch size, patch density and largest patch index—to characterize the fragmentation of ecological space in green belts of Shijiazhuang from the perspective of scale, variation, degree of fragmentation and dominance of core patches.

2.2.3. Land-Use Transfer Matrix Model

The land-use transfer matrix model quantitatively describes the magnitude, direction, and structural characteristics of the mutual transformations among different land-use types within the study area over a specific time interval. The model constructs a two-dimensional matrix that records the area transferred from the $i$-th land-use type to the $j$-th land-use type during different time periods. The dimension of the matrix is determined by the total number of land-use categories n and is expressed as follows:

$$S_{ij}=\left[\begin{array}{ccc}{S}_{11}& \dots & S_{1n}\\ ⋮& \ddots & ⋮\\ S_{n1}& \dots & S_{nn}\end{array}\right]$$

where $S_{ij}$ represents the area converted from the $i$-th land-use type at the beginning of the period to the $j$-th land-use type at the end of the period, and $n$ represents the total number of land-use types divided in the study area.

Existing studies have primarily focused on net changes in land-use categories, which makes it difficult to reveal the dynamics of conversion between land-use types. By contrast, a land-use transfer matrix captures the direction, pathway, and magnitude of these conversions. In this study, a transfer matrix was constructed for three time periods—2015, 2020, and 2024—to quantify not only the extent of agricultural land encroached upon by built-up land but also the intensity of such conversions at different stages.

2.2.4. Kernel Density Estimation

Kernel density estimation (KDE) treats each observed point as the center of a localized probability mass or influence range. It then calculates the sum of the kernel function values of all point features at any given spatial location x and normalizes this sum to obtain the estimated density at that position.

$$f_n\left(x\right)={\displaystyle \frac{1}{nh}}{\displaystyle \sum }_{i=1}^{n}k\left({\displaystyle \frac{x-x_i}{h}}\right)$$

where: $f_n\left(x\right)$ is the kernel density estimate at point $x$; $n$ is the total number of points; $h$ is the distance decay threshold; $k\left(?\right)$ is the kernel function; $x_i$ is the position of the $i$-th observation point.

Previous studies investigating the spatial differentiation of the service industry often used point-of-interest (POI) distribution maps, which are unable to reflect gradual changes in spatial density. This study applies KDE using the raster calculator in ArcMap 10.8 to calculate the difference in kernel density between two categories of service industry. By doing so, we are able to illustrate the differentiation process and gain further insights into the spatial evolution of the service industry.

2.2.5. Spatial Autocorrelation Model

• Global Spatial Autocorrelation

This study employs the global Moran’s I index as a core indicator to quantify the degree of spatial clustering or dispersion of POI points within village and town residential spaces. The formula for calculating Moran’s I is as follows:

$$Moran\prime s I={\displaystyle \frac{n{{\displaystyle \sum }}_{i=1}^n{{\displaystyle \sum }}_{j=1}^m{W}_{ij}\left(x_i-\stackrel{-}{x}\right)\left(x_j-\stackrel{-}{x}\right)}{{{\displaystyle \sum }}_{i=1}^n{{\displaystyle \sum }}_{j=1}^m{W}_{ij}{{\displaystyle \sum }}_{i=1}^n{(x_i-\stackrel{-}{x})}^2}}$$

where $n$ is the total number of spatial cells. $x_i$ and $x_j$ are the attribute values of spatial cell $i$ and its neighboring cell $j$, respectively. $\stackrel{-}{\mathrm{x}}$ represents the average of all attribute values of spatial cells. $w_{ij}$ represents the element of the spatial weight matrix, which defines the spatial adjacency relationship between cells $\mathrm{i}$ and $j$. If cells $\mathrm{i}$ and $j$ are adjacent, $w_{ij}$ = 1; otherwise, $w_{ij}$ = 0.

The value of Moran’s I typically ranges from −1 to 1, and its magnitude reflects both the nature and strength of spatial autocorrelation. Specifically, Moran’s I > 0 indicates positive spatial autocorrelation, meaning that spatial units with similar attribute values tend to cluster together. Moran’s I < 0 suggests negative spatial autocorrelation, indicating that dissimilar units are more likely to be adjacent. Moran’s I ≈ 0 implies no spatial autocorrelation, suggesting that the spatial pattern is nearly random. A larger absolute value corresponds to stronger spatial dependence, reflecting more pronounced clustering or dispersion.

2.

Local Spatial Autocorrelation

Local spatial autocorrelation analysis focuses on the relationships between attribute values of the central spatial unit and attribute values of neighboring spatial units at a local scale, and finds some local areas of clustering or spatial outliers. Local Moran’s I significance maps can be categorized into the following four types of spatial association. High–High and Low–Low, which represent local spatial clustering, i.e., high-value or low-value areas are clustering in space. High–Low and Low–High, which represent local spatial dispersion, i.e., high–value and low–value areas are dispersed in space. The formula to calculate local Moran’s I index is as follows:

$$I_i{ = (x}_i-\stackrel{-}{x}){\displaystyle \sum }_{j=1}^n{\omega }_{ij}{(x}_j-\stackrel{-}{x})$$

where $x_i, x_j$ represent the index or coupling coordination degree of the region $i$ and region $j$, which is mainly analyzed through the Local Indicators of Spatial Association cluster diagram.

Related work on the expansion of residential space has mainly focused on statistical analysis of expansion areas or overlay analysis of spatial extent. In contrast, this article proposes using global Moran’s I and local spatial autocorrelation analysis to quantitatively analyze not only the degree of clustering of residential space sprawl but also its temporal characteristics. Based on the LISA cluster maps, we can accurately locate hotspots, lagging areas, and spatial abnormalities in the sprawl process and provide a scientific basis for constructing differential residential space management policies.

2.3. Data Sources

2.3.1. Remote Sensing Image Data and Preprocessing

This study focuses on three key timeframes in Shijiazhuang’s green belts: 2015, 2020, and 2024. Landsat 8–9 OLI/TIRS Collection 2 Level-2 remote sensing imagery, provided by the National Aeronautics and Space Administration (NASA) and the United States Geological Survey (USGS), was acquired from the Geospatial Data Cloud Platform of the Chinese Academy of Sciences (http://www.gscloud.cn/, accessed on 14 July 2025), with a spatial resolution of 30 m × 30 m. Preprocessing was conducted primarily using ArcMap 10.8 and ENVI 5.3, including image fusion, mosaicking, cropping, and supervised classification.

2.3.2. POI Data and Preprocessing

Point-of-interest (POI) data for Shijiazhuang’s green belts was obtained from the AutoNavi Open Platform (https://lbs.amap.com/, accessed on 25 July 2025). Following the POI data classification standards, five categories were selected for analysis: (1) companies and enterprises, (2) science, education, and cultural services, (3) leisure and entertainment services, (4) lifestyle services, and (5) commercial and residential properties. During preprocessing, the raw POI dataset was cleaned and projected into the WGS84 coordinate system. Duplicate entries and erroneous coordinates were removed, while spatial matching and registration were applied to ensure temporal and spatial consistency across different years. The processed dataset was subsequently reclassified using ArcMap 10.8 (

Table 1. Reclassification of POI Data.

Spatial Category	Primary POI Classification	Secondary POI Classification
Industrial Space	Companies and Enterprises	Factories
Service Industry Space-Cultural Tourism and Leisure Oriented Services	Science, Education, Culture, Leisure, and Entertainment	Museums, science and technology museums, archives, libraries, cultural centers, radio and television stations, cinemas, karaoke (KTV) venues, retirement and holiday centers, bars, agritourism farmhouses, chess and card rooms, internet cafés, amusement parks, etc.
Service Industry Space-Life Supporting Services	Lifestyle Services	Public utilities, post offices, agencies and intermediaries, lottery outlets, logistics, photography and printing services, beauty and hairdressing salons, information and consultation centers, etc.
Living Space	Commercial and Residential Properties	Village and town residential areas

).

3. Results

3.1. Small Scale and Dispersed Distribution of Industrial Land

Quantitative analysis using landscape ecological indices shows that the mean patch area of industrial land was 11.9646 hm² in 2015, declining to 11.6481 hm² in 2020 and further to 10.2953 hm² in 2024 (Figure 2). This indicates a continuous reduction in the size of individual industrial land patches, reflecting the absence of large, contiguous industrial clusters and the predominance of small-scale development. Moreover, the standard deviation of patch area dropped sharply from 248.1834 hm² in 2015 to 121.7657 hm² in 2024 (Figure 2), suggesting a substantial decrease in size variation among patches and a more uniform overall distribution. Collectively, these findings demonstrate that industrial land during 2015–2024 was characterized by fragmentation and small-scale dispersion.

Results of the ANN were also supportive of these trends (

Table 2. Results of Nearest Neighbor Analysis of Industrial Land in Shijiazhuang’s Green Belts (2015–2024).

Index	2015	2020	2024
Mean observed distance (m)	598.751	374.671	90.596
Expected random distance (m)	1448.645	1066.842	1128.391
Nearest neighbor ratio	0.413	0.351	0.080
z-score	−11.335	−19.309	−33.569
p-value	0.000	0.000	0.000

). The calculated mean distance between industrial sites was 598.75 m in 2015 and dropped dramatically to 90.60 m in 2024, which suggested that patches were becoming closer to each other. In addition, the nearest neighbor ratio decreased from 0.413 to 0.080 and the z-score decreased from −11.335 to −33.569 during the same period. Because the ratios were below 1 and the absolute values of the z-scores were larger than 2.58, the two ratios were significantly lower than the corresponding random values, meaning that the patch clustering did not result from a random distribution. These results collectively revealed that although industrial land presented localized clusters, its overall distribution was dispersed, and therefore presented a fragmented pattern.

In total, industrial land in Shijiazhuang’s green belts expanded steadily from 2015 to 2024, and its spatial configuration was characterized by small-scale and dispersed distribution. This may be due to two reasons. First, although large industrial parks were forbidden in green belts, as local villages and towns attempted to increase fiscal revenue, they had tacitly tolerated the development of small and micro-scale industries. Second, this scattered distribution enabled industries to obtain land at relatively low cost in ecologically marginal areas and to avoid paying the full cost of infrastructure development. However, the uncontrolled expansion of industrial land also brought about great ecological risks, such as breaking ecological corridors, increasing land surface temperature, enhancing urban heat islands, and causing bare land.

3.2. Fragmentation of Ecological Space

The alterations in the Maximum Patch Index (LPI) demonstrate a notable reduction in the scale of core ecological patches (Figure 3). The LPI for agricultural terrain saw a notable reduction from 19.52% in 2015 to 7.23% in 2020, and further to 6.82% in 2024, indicating a substantial weakening of its role as a continuous, large-scale ecological patch. Grassland showed a fluctuating pattern, with its LPI briefly increasing to 3.03% in 2020 before declining to 1.42% in 2024, suggesting that its capacity to maintain large, stable core patches has also diminished. The findings demonstrate a fragmentation and reduction in previously continuous ecological patches, significantly compromising the connectivity of the ecological space.

As shown in Figure 4, when it comes to the process of leapfrog urbanization, the change trend of patch density (PD) shows the tendency of ecological fragmentation in whole. The PD of cultivated land decreased from 2.069 patches/km² in 2015 to 1.573 patches/km² in 2024. It is not only the fragmentation of large patch but also the breaking away of small patch or merging of small patch into big patch leading to the great decrease in total area. As for forestland, its PD increased greatly from 0.087 patches/km² in 2015 to 1.957 patches/km² in 2020, and the PD decreased slightly in 2024 although, which also reflected the great fragmentation degree. More smaller patches and more fragmented patches were appeared. The PD of grassland decreased firstly and then increased. Combined with the decrease in LPI, it also reflected the fragmentation of remaining or newly added patches, which were small patches and scattered patches.

In summary, the core ecological patches of green belts in Shijiazhuang experienced the process of constant fragmentation in the period from 2015 to 2024. That is, the patches became more and more scattered, more and more small and highly fragmented ecological environment. It was caused by the process of leapfrog urbanization. The built-up land occupied more and more ecological areas around ecological zone. To seek for less cost and better location, the enterprise tended to set the scattered sites around ecological area, which would lead to the phenomenon of ecological fragmentation and degradation.

3.3. Encroachment on Agricultural Space

Quantitative results from the land–use transition matrix (

Table 3. Land-Use Transfer Matrix of Green Belts in Shijiazhuang (2015–2024).

Land-Use Type	Cropland	Woodland	Grassland	Water	Built-Up Land	Unused Land	Total	Area Change
Cropland	103.89	0.00	4.14	0.00	6.04	3.70	117.77	55.95
Woodland	1.31	0.66	2.69	0.05	1.25	2.07	8.04	−7.16
Grassland	31.84	0.07	24.01	0.08	17.26	16.67	89.93	−30.23
Water	0.24	0.01	1.30	6.15	3.82	9.93	21.45	−14.49
Built-up Land	24.97	0.08	18.12	0.65	106.01	15.93	165.75	−23.97
Unused Land	11.48	0.07	9.43	0.03	7.40	16.77	45.17	19.91
Total	173.72	0.88	59.69	6.96	141.78	65.08	448.12
Area Change	−55.95	7.16	30.23	14.49	23.97	−19.91

Note: Vertical data represent land-use types in 2015, and horizontal data represent land-use types in 2024. The area unit is hectares (ha).

) indicate a net decrease of 55.95 ha of cultivated land between 2015 and 2024, with the largest conversion being 24.97 ha transferred to built-up land. The second major transition involved 18.12 ha converted from grassland to built-up land. This process is also evident in the land-use maps: the 2015 map (Figure 5a) shows that cultivated land patches were still large and contiguous, forming the regional agricultural base. Over time, however, built-up land gradually infiltrated and spread across this base (Figure 5b,c). These transformations reveal increasing fragmentation of farmland patches, which—when combined with the scattered distribution of rural settlements—further reduces land-use efficiency.

Stage-specific analysis (

Table 4. Land-Use Transfer Matrix of Green Belts in Shijiazhuang (2015–2020).

Land-Use Type	Cropland	Woodland	Grassland	Water	Built-Up Land	Unused Land	Total	Area Change
Cropland	93.92	0.00	1.99	0.00	2.78	1.74	100.43	73.29
Woodland	1.78	0.70	4.40	0.13	2.69	2.37	12.06	−11.18
Grassland	49.66	0.06	27.40	0.01	19.08	20.51	116.73	−57.04
Water	0.19	0.01	0.58	6.33	2.70	4.2	14.09	−7.12
Built-up Land	17.51	0.05	14.20	0.44	105.71	11.43	149.33	−7.55
Unused Land	10.66	0.07	11.13	0.05	8.81	24.75	55.48	9.60
Total	173.72	0.88	59.69	6.96	141.78	65.08	448.12
Area Change	−73.29	11.18	57.04	7.12	7.55	−9.60

Note: Vertical data represent land-use types in 2015, and horizontal data represent land-use types in 2020. The area unit is hectares (ha).

and

Table 5. Land-Use Transfer Matrix of Green Belts in Shijiazhuang (2020–2024).

Land-Use Type	Cropland	Woodland	Grassland	Water	Built-Up Land	Unused Land	Total	Area Change
Cropland	85.31	0.66	24.26	0.01	4.25	3.28	117.77	−17.34
Woodland	0.08	2.25	2.96	0.10	0.57	2.07	8.04	4.02
Grassland	6.64	3.70	49.69	0.27	12.38	17.24	89.93	26.80
Water	0.01	0.56	1.32	12.13	3.20	4.23	21.45	−7.37
Built-up Land	5.18	1.81	18.47	1.53	125.15	13.63	165.75	−16.43
Unused Land	3.21	3.08	20.03	0.05	3.78	15.02	45.17	10.31
Total	100.43	12.06	116.73	14.09	149.33	55.48	448.12
Area Change	17.34	−4.02	−26.80	7.37	16.43	−10.31

Note: Vertical data represent land-use types in 2020, and horizontal data represent land-use types in 2024. The area unit is hectares (ha).

) shows that 2015–2020 period was the year range during which most cultivated land was converted for other uses (net decrease of 73.29 ha). During this stage, 17.51 ha of land were converted to built-up land, which accounted for 23.9% of the decrease in cultivated land. The 2015–2020 land-use map reveals that there were large clusters of built-up land encroaching on the edges of cultivated land. This phenomenon clearly demonstrates the pressure on cultivated land caused by the rapid development of industry.

In contrast, the rate of decrease in cultivated land during 2020–2024 slowed slightly, with a net decrease of 17.34 ha (5.18 ha of land converted to construction, accounting for 29.9%). However, the 2020–2024 land-use map still displays scattered expansion of patches of built-up land, and there are even isolated new developments in the hinterlands of cultivated land that are far away from the built-up area. Therefore, we found that agricultural space was still being eroded.

Overall, the continuous erosion of agricultural space within the green belts of Shijiazhuang from 2015 to 2024 indicates a structural problem in the returns from land-use. The significantly higher returns per unit of land from industrial land compared with that from agriculture have eroded the internal incentives for local stakeholders to protect farmland. This vicious circle of the fragmentation of farmland and decreasing returns has weakened the agricultural production capacity of the green belts and has become a bottleneck to regional sustainable development. This trend aligns with the land–use transition theory, indicating that under the pressures of urbanization, the conversion of agricultural land into non-agricultural built-up land is an irreversible process. It reveals the characteristics of spatial structural adjustments driven by economic growth, reflecting the dual pressures of agricultural function degradation and ecological risks within green belts.

3.4. Spatial Differentiation of Service Industry Growth

Quantitative analysis based on kernel density estimation reveals pronounced spatial differentiation in the development of cultural–tourism–leisure services versus lifestyle-related services within Shijiazhuang’s green belts between 2015 and 2024. In 2015, high-density clusters of lifestyle-related services were primarily concentrated in the western region, whereas cultural–tourism–leisure services were more dispersed in the northeast. The spatial distribution and intensity of these two sectors differed significantly, suggesting that lifestyle-related services, initially dependent on existing settlements, had a clear first-mover advantage in service coverage. By contrast, cultural–tourism–leisure services, constrained by longer development cycles and limited resource utilization, had not yet established widespread dominance.

The spatial distribution of kernel density values exhibited substantial alterations between 2015 and 2020 (Figure 6a,b). The value range expanded appreciably to −318,272.438 to 62,637.084. The southward progression of lifestyle-related service high-density zones along key transportation corridors exemplifies their persistent intrusion into a range of peri-urban residential spaces. Simultaneously, the cultural–tourism–leisure sectors commenced clustering in the northwest; however, their expansion was limited by disparate resource distribution and a tardy pace of development. The overlap of these clusters with the dispersed expansion of lifestyle-related services further intensified the functional differentiation of regional service industries.

From 2020 to 2024, the spatial differentiation of kernel density values intensified (Figure 6c), with the delineation between negative-value and positive-value dominated regions exhibiting greater complexity. An escalation in the cultural–tourism–leisure services’ peak kernel density was noted, reaching 93,955.628–104,395.141, converging into a contiguous mass in the northwest. The rise in high-end consumer demand and consistent policy endorsements for cultural tourism have positioned these services as the central element in regional industrial advancement. Concurrently, lifestyle-oriented services exhibited exceptionally elevated kernel density levels (124,248.489–310,621.219). In the burgeoning eastern communities, a decentralized distribution is concurrently sustained across newly established rural settlements, which have emerged alongside urban expansion, to fulfill livelihood requirements.

In summary, the service industry within Shijiazhuang’s green belts evolved from differentiation to interaction between 2015 and 2024. The transformation was driven by resource allocation, policy direction, and profit orientation, reflecting the adaptive adjustment of service industries to diversified demand amid rapid urban expansion. This pattern aligns with the theories of “functional re-differentiation” and “core–periphery diffusion” in urban spatial structure, as tourism and leisure-oriented services tend to cluster along ecological boundaries forming new growth poles, whereas life-supporting services spread outward to accommodate residential needs, together reflecting the dual pressures of economic growth and ecological protection during the urbanization process.

3.5. Spread and Expansion of Residential Space in Villages and Towns

Results of the spatial autocorrelation model revealed the trend of disorderly sprawl and expansion of rural and urban residential space in green belts of Shijiazhuang during 2015–2024. At the macro-level, the Global Moran’s I index indicated the overall sprawl of residential space. As shown in Figure 7, the Global Moran’s I index was 0.953 in 2015, 0.958 in 2020 and 0.961 in 2024. It indicated that the degree of spatial agglomeration in high-value areas gradually increased. Therefore, the residential distribution pattern changed from “point-like” to “belt-like” and even “surface-like” pattern gradually. Combined with the time-series characteristics of the Moran’s I scatter plots, it can be observed that: in 2015, scatter points were primarily concentrated in the high–high quadrant, reflecting a strong basis of positive spatial correlation; in 2020, scatter points tended to cluster in high-value neighborhoods, indicating that residential space had begun to penetrate surrounding low-value areas; in 2024, scatter points shifted further into the high-value quadrant, highlighting the trend of disorderly spillover and erosion of residential space.

Micro-level examination of the 2015 cluster data (Figure 8a) shows a scattered, multi-core arrangement of high–high clusters, predominantly located in the west and north, establishing independent residential areas and beginning to encroach upon adjacent low-value areas, thus paving the way for subsequent disorganized growth. In the eastern and southern regions, low–low clusters were prevalent, corresponding to non-constructed land. The 2020 cluster analysis (Figure 8b) demonstrates a substantial expansion of high–high clusters, which have been merging and encroaching upon low-low clusters. Clusters expanded along transportation corridors and development axes, evolving from a “point-like” to a “belt-like” configuration, encroaching upon agricultural and ecological spaces incrementally. The transformation was predominantly propelled by the concentration of employment and the escalating residential requirements resulting from industrial expansion. The 2024 cluster analysis (Figure 8c) indicates that the high–high clusters in the west and north further merged on a large scale, forming broader and more continuous agglomerations. The process persistently depleted low-density clusters, prompting their southeastward migration and subsequent fragmentation. Observations revealed a leapfrog pattern of expansion into environmentally sensitive regions, highlighting spatial imbalances between central residential densities and peripheral sprawl. In addition, leapfrogging into ecologically sensitive areas was observed, revealing a spatial imbalance between dense residential cores and disorganized peripheral regions.

In summary, the steady rise in the Global Moran’s I index confirms the trend of residential space expansion and sprawl. Cluster analysis results clearly demonstrate the transformation of residential space from “island-like dispersion” to “ribbon-like extension” and ultimately to “continuous encroachment.” This change aligns with the patterns revealed by urban sprawl and edge city theories, where under the combined pressures of ecological constraints and residential demand, peripheral areas develop a “multi-core, low-density” sprawl pattern. This reflects the dual role of green belts policy: it curbs disorderly expansion of the urban core yet can inadvertently trigger peripheral aggregation and functional reorganization.

Overall, this study employs a range of quantitative analysis methods that mutually validate each other. The average nearest neighbor analysis reveals the clustering and dispersal characteristics of industrial land; the landscape ecological index system assesses the degree of fragmentation in ecological spaces; the land–use transition matrix dynamically reflects the conversion pathways and structural features between agricultural and built-up land, shedding light on the spatial transformation mechanisms of the green belts; kernel density estimation illustrates the differentiated distribution patterns of service industry spaces; and the spatial autocorrelation model quantifies the intensity and sprawl trends of residential space expansion. Together, these methods form a comprehensive analytical framework, providing quantitative tools for understanding the spatial evolution of green belts in metropolises.

4. Discussions

4.1. The Challenge of Small-Scale and Dispersed Distribution of Industrial Land

Between 2015 and 2024, industrial land within Shijiazhuang’s green belts exhibited a small-scale and dispersed distribution pattern, creating major challenges for infrastructure efficiency, economies of scale, and ecological protection. Smith [10] found that because green belt policies strictly restrict the expansion of contiguous built-up areas, industrial development was forced to adopt a leapfrog expansion model, bypassing the belts for decentralized, low-density development in peripheral zones. This led to negative outcomes such as longer commuting distances, increased automobile dependence, and higher carbon emissions. Hu and Wang [22] used the PLUS model to simulate the impacts of green belt scenarios on ecosystem service values, showing that irrational land-use changes can significantly diminish ecosystem functions. At the same time, the small-scale and dispersed distribution pattern of industrial land also echoes Nelson’s [23] “spatial paradox” in urban sprawl control theory, where rigid boundaries set to limit sprawl may inadvertently trigger decentralized development on the periphery.

Overall, excessive dispersion of industrial land results in duplicate infrastructure construction, rising commuting costs, and ecological fragmentation. Therefore, while maintaining ecological protection as the foundation, policies should consider local development stages and resource endowments to guide moderate concentration and cluster-based development of industrial land. This would help achieve a dynamic balance between ecological security and industrial growth.

4.2. Threats of Ecological Space Fragmentation

From 2015 to 2024, the ecological space in the green belts of Shijiazhuang became progressively fragmented. The changes in maximum patch index and patch density reflect a decrease in connectivity and an increase in fragmentation, and further induce ecological risks, including temperature increasing, urban heat island enhancing, vegetation cover decrease, and high ecological value area diminishing. Do et al. [24] believed that in addition to the direct occupation of a large amount of green space, the rapid urbanization also leads to the degradation of the ecosystem, which seriously affects the sustainable development of the city. Kardani-Yazd et al. [25] believed that the direct fragmentation reduces biodiversity and ecological services. While Zhou et al. [26], in their study of the Beijing green belt, found that although the green belt policy aims to curb urban sprawl, poor management or design flaws have led to increased fragmentation of the landscape within the green belt. This differs from the fragmentation pattern driven by urbanization observed in the present study. Additionally, Costanza et al. [27], from an ecological economics perspective, highlight that this fragmentation signifies the loss of natural capital and ecosystem services.

This fragmentation poses a severe threat to regional ecological security and sustainable development, manifesting in reduced biodiversity, worsened heat island effects, and weakened ecological barriers. Therefore, strengthening ecological connectivity should be prioritized. Constructing wedge-shaped green corridors and integrated blue–green networks would repair fragmented patches, enhance ecological integration, and restore key ecosystem functions.

4.3. Threats of Agricultural Space Encroachment

Between 2015 and 2024, eroded agricultural space located in Shijiazhuang’s green belts was continuously decreased. Land-use transfer matrices revealed that the reduction in cultivated land and increasing patch splitting are mainly induced by land return imbalance caused by the rapid development of industry. Han et al. [28] found that the liberalization of green belt policies leads to accelerated conversions of green belts into urban land, causing a large decrease in cultivated land, a serious threat to farmers’ lives and incomes, a challenge to regional food security, and a decrease in the sustainable supply of ecosystem services. Akimowicz et al. [29] point out that similar land conversion pressures are also observed in the green belts of the United Kingdom and Canada, where urban expansion has weakened the resilience of agricultural systems. This suggests that in land–use transitions, economic incentives often play a more decisive role than regulatory mechanisms. Zhou and Wang [30] argue that rapid urbanization has led to significant encroachment and fragmentation of agricultural land, while green belt policies have, to some extent, facilitated the recovery of grasslands, highlighting the impact of policy interventions on green belts. In addition, Porter et al. [31] warned that farmland protection policies that offer economic incentives may lead to “fake protection” of farmland. That is, the land on the books is protected but poorly managed, offering little ecological or productive service.

These findings suggest that agricultural protection policies must fully account for the imbalance in land-use benefits. Promoting multifunctional agricultural models is essential to achieving coordinated ecological conservation and agricultural productivity. At the same time, safeguards must be established against “false protection,” ensuring that preserved farmland retains both ecological and production value.

4.4. Challenges of Spatially Differentiated Service Industry Growth

Between 2015 and 2024, the spatial distribution of service industries in the green belt area of Shijiazhuang exhibited a distinct pattern of differentiated growth, with tourism and leisure-oriented services and life-supporting services both displaying characteristics of spatial clustering and dispersion. This differentiated spatial pattern echoes Hall’s [32] concept of “functional re-differentiation” in urban sprawl control theory, where under the constraints of green belt space, the service industry adapts through relocation and differentiation. Studies by Amati & Yokohari [5] on London and Choi et al. [14] on Seoul have found that tourism and leisure services tend to cluster along the edges of green belts, forming an urban fringe economic zone. This suggests that the spatial differentiation of services in Shijiazhuang’s green belts is not an isolated phenomenon, but rather a trend reflecting the ongoing post-industrial transformation. Conversely, based on the London case, Smith [10] argued that concentrated service industry layouts, though market-driven and economically efficient, often occur at the expense of ecological land.

This evidence underscores that spatially differentiated growth of the service industry has profound implications for balancing functional coordination and ecological protection. Policy frameworks must therefore reconcile ecological protection with dynamic livelihood needs, and market efficiency with regulatory guidance, to achieve multi-objective coordination among ecological conservation, livelihood security, and economic benefits.

4.5. Challenges of Residential Space Sprawl and Expansion

Between 2015 and 2024, residential space in Shijiazhuang’s villages and towns within the green belts exhibited continuous sprawl, as reflected in the steadily rising global Moran’s I. Cluster analysis revealed that “high–high” clusters evolved from multi-core discreteness to large-scale contiguous agglomerations, intensifying challenges to ecological connectivity, infrastructure capacity, and social equity. Wei et al. [33], in a study of suburban Wuhan, proposed a comprehensive evaluation framework showing that low-density, sprawling residential expansion leads to socioeconomic inefficiency and ecological service loss, a finding consistent with this study. However, Ma and Jin [17] emphasized that while strict green belt policies may restrain sprawl, they can also excessively compress residential development into limited corridors, producing uncontrolled density and infrastructure overload in transport, water, and energy systems. Studies by Choi et al. [14] and Eswar [13] have found similar phenomena in Seoul and Bangalore, where rigid green belt policies have restricted infill development and increased carbon emissions associated with housing. These cases suggest that if control measures are too stringent, the constraints imposed by green belt policies may inadvertently exacerbate spatial inequality and environmental inefficiency, highlighting the necessity for establishing adaptive governance mechanisms.

Thus, although sprawling residential expansion may temporarily satisfy housing demand, it undermines ecological connectivity, encroaches on farmland, and aggravates infrastructure deficiencies. To address these issues, residential development should be guided toward moderate intensification, avoiding disorderly sprawl in villages and towns. Comprehensive spatial strategies must balance ecological constraints, infrastructure capacity, and residents’ livelihood needs to achieve sustainable residential land-use.

5. Policy Recommendations

To address the contradictions between ecological protection and urban expansion within Shijiazhuang’s green belts, this study proposes a structural policy mechanism designed to coordinate industrial, ecological, agricultural, service, and residential spaces. The mechanism is summarized as “adjust the primary industry, restrict the secondary industry, and promote the tertiary industry.” Specifically, “adjust” refers to optimizing the spatial structure of agricultural land and transforming it from a single-function production system into a multifunctional structure that integrates both ecological and economic values. “Restrict” emphasizes dual constraints on disorderly industrial expansion and inefficient residential sprawl. “Promote” focuses on advancing the high-quality development of the service industry, enhancing the integration of ecological spaces, and encouraging compact and intensive use of residential land. Collectively, these three components constitute a coordinated structural policy mechanism that balances ecological protection and urban development, providing a generalizable theoretical framework for metropolitan green belts governance and ensuring the long-term, orderly, and sustainable evolution of green belts. The following section presents detailed policy recommendations for the green belts of Shijiazhuang City.

5.1. Enhancing the Value of Agricultural Space

Guided by the “one adjustment” principle, agricultural spatial layout should be optimized to transform agriculture from a single-function system into a multifunctional model, shifting farmland from passive erosion to proactive value creation. Promoting high-value-added, green, and sustainable agricultural practices can increase farmers” incomes, thereby reducing economic pressures that drive the conversion of farmland to non-agricultural uses. Improving the allocation and efficiency of agricultural resources is essential, particularly through the development of modern agricultural industrial parks. By consolidating fragmented farmland into contiguous cultivation zones and integrating production, processing, and leisure services, agricultural space can achieve higher land-use efficiency and greater ecological and economic benefits.

5.2. Aggregating Industrial Space

Following the “two limitations” strategy, the disorderly proliferation of dispersed industries must be curbed by redirecting industrial development toward spatial concentration. Establishing clear boundaries for industrial agglomeration and enforcing a “negative list” for industrial land within green belts are crucial steps to gradually phase out highly polluting, low-efficiency small-scale industries. Such measures address the root causes of leapfrog industrial expansion that threaten ecological land. At the same time, promoting the clustering of high-value-added and R&D-intensive industries within designated industrial parks can strengthen economies of scale, improve land-use efficiency, and enhance the ecological compatibility of industrial activities.

5.3. Coordinating Service Industry Development

In line with the “three developments” approach, service industry space should be diversified and spatially balanced to reconcile livelihood needs with ecological protection. Ensuring equitable distribution of livelihood-supporting services can prevent overconcentration in core areas while extending services to underserved regions. Leveraging the ecological assets of green belts, low-intrusion and ecologically adaptive cultural tourism and leisure industries should be promoted. By adopting “park-like” renewal concepts, cultural and tourism facilities can be embedded within ecological networks, fostering the organic integration of cultural, ecological, and agricultural spaces. This strategy not only enhances the service functions of green belts but also supports their ecological and social sustainability.

5.4. Integrating Ecological Space

A systematic approach to ecological space integration is essential for strengthening the structural and functional stability of ecosystems. Connectivity among ecological corridors should be enhanced to establish resilient and adaptive ecological networks. Rivers and woodlands can serve as the backbone of such networks, forming ecological corridors and connectivity systems that safeguard biodiversity and landscape resilience. By applying “wedge-shaped green corridors” and “blue–green networks,” fragmented ecological patches can be reconnected, scattered ecological lands integrated, and the integrity and continuity of the regional ecological landscape significantly improved.

5.5. Intensifying Residential Space

Residential development should follow a model of concentrated distribution to reduce inefficient land consumption and align housing expansion with industrial layout and ecological carrying capacity. Efforts should focus on concentrating rural and town residential areas within core township centers or along transportation corridors, where public service infrastructure can be efficiently provided. Expansion into ecologically sensitive zones and contiguous farmland must be strictly prohibited to prevent the degradation of ecological and agricultural functions. At the same time, leapfrog residential development in marginal areas should be curbed to enhance land-use efficiency and ensure that residential growth is both spatially rational and environmentally sustainable.

6. Research Prospects

This study, using green belts in Shijiazhuang as a case, has revealed the spatial evolution characteristics and mechanisms of green belts during rapid urbanization through various quantitative analysis methods. However, the study still has limitations, as outlined below:

At the data level, the study relies on remote sensing images and POI data from the years 2015, 2020, and 2024 to uncover the spatial evolution characteristics of Shijiazhuang’s green belts. However, the use of non-continuous, interannual data presents challenges in revealing the underlying evolutionary mechanisms. The relatively long observation intervals make it difficult to capture short-term land-use changes, potentially introducing bias. Future studies could collect continuous time-series data and employ more frequent observations to improve the accuracy of spatial-governance models.

At the analytical level, due to limitations in interdisciplinary expertise and data availability, a supporting economic analysis framework could not be established, which limited our ability to assess fiscal costs and benefits, market responses and social welfare effects of spatial policy implementation. Therefore, future studies should adopt an interdisciplinary approach involving land economics, urban planning and public policy evaluation to establish a coupled “policy–space–economy” analytical model to simulate different policy scenarios, which would improve the robustness of policy design and provide better economic justification for spatial governance policies.

7. Conclusions

This study takes green belts of Shijiazhuang as a case study and integrates various analytical methods, including average nearest neighbor analysis, landscape ecological index, land–use transition matrix, kernel density estimation, and spatial autocorrelation models, to systematically analyze the spatial evolution characteristics of the Shijiazhuang green belts from 2015 to 2024. The results show that industrial land has expanded in a small-scale and dispersed manner, the fragmentation of ecological spaces has intensified, agricultural land continues to be encroached upon, service industry spaces have grown with increasing differentiation, and residential spaces in villages and towns have spread uncontrollably. These findings reveal the structural contradiction between the rigid constraints of ecological protection and the real demands for spatial development during urbanization, leading to increased ecological risks, imbalanced urban–rural development, and lagging infrastructure. In response, this study proposes a planning strategy of “adjusting the primary industry, restricting the secondary industry, and promoting the tertiary industry,” aiming to resolve the conflict between ecological protection and urban expansion in metropolitan green belts.

The various quantitative methods used in this study provide a solid foundation for understanding the spatial evolution mechanisms of green belts in rapidly urbanizing areas. Among them, the land–use transition matrix and spatial autocorrelation model play a central role. The former effectively reveals the dynamic conversion relationships between different land types, while the latter accurately identifies the temporal and spatial patterns of aggregation and diffusion. Meanwhile, methods such as landscape ecological indices and kernel density estimation demonstrate unique value in revealing spatial structural evolution and functional differentiation, proving to be effective quantitative tools for specific case studies. Future research could further incorporate supplementary analytical tools, such as spatial coupling coordination models, geographical detectors, and geographically weighted regression (GWR), to deepen the analysis of the mechanisms behind green belts spatial evolution in metropolises. By integrating socio-economic, ecological, and policy data, these approaches can provide more forward-looking decision support for spatial planning strategies in metropolitan green belts.