Trade-Offs, Synergies, and Driving Factors of Ecosystem Services in the Urban–Rural Fringe of Beijing at Multiple Scales

Abstract

Urban–rural fringe areas are critical transition zones where ecological functions and human activities interact intensely, often leading to complex spatial patterns and trade-offs among ecosystem services (ESs). Understanding these patterns and their socio-ecological drivers across multiple spatial scales is essential for sustainable land-use planning and ecosystem management. This study, using the urban–rural fringe (URF) of Beijing as an example, quantified eight representative ecosystem services at the 1 km grid, 3 km grid, and township scales. It employed hotspot analysis, Moran’s Index, and the Spearman correlation to analyze trade-offs and synergies (TOSs) among ESs. The study also applied a self-organizing map and the NbClust function to identify and determine the optimal number of ecosystem service bundles (ESBs) for ecological functional zoning. Redundancy analysis was used to explore the impacts of six socio-ecological drivers on the spatial distribution of ESs. The results revealed the following: (1) The spatial distribution of ESs in Beijing’s URF exhibits clustering and cross-scale variations, with spatial clustering intensifying as the scale expands. (2) TOSs among ESs vary in strength and direction across the three spatial scales. (3) The primary drivers of TOSs at all three scales are the normalized vegetation index and annual precipitation. (4) Based on the supply intensity of various ESs, the study area was classified into four types of ESBs across the three scales: ecologically restricted areas, food production areas, ecologically balanced areas, and high-quality ecological areas. The township scale is more conducive to planning and management, while the 1 km and 3 km grid scales are more helpful for understanding the relationship between land use and ESs.

1. Introduction

Ecosystem services (ESs) refer to the benefits provided by nature, which directly or indirectly contribute to sustainable human well-being [1]. However, global land use change and the overexploitation of natural resources are driving the continuous degradation of ecosystem functions. According to the Millennium Ecosystem Assessment, approximately 60% of ecosystem services worldwide are currently in decline, with the impacts particularly evident in regions undergoing rapid urban expansion [2]. A deeper understanding of the interactions among different ecosystem services is crucial for restoring ecosystem functions and enhancing their service capacity in areas experiencing fast-paced urbanization [3,4]. The interactions among ecosystem services are primarily manifested in two forms: trade-offs and synergies. Trade-offs refer to situations where the enhancement of one service often comes at the expense of another, while synergies occur when multiple services increase or decrease simultaneously [5,6]. Under specific geographic and ecological contexts, multiple ecosystem services often co-occur in space and time, forming stable spatial groupings known as ecosystem service bundles (ESBs) [7]. ESBs represent the spatial integration of trade-offs and synergies, reflecting the co-evolutionary mechanisms among multiple services within ecosystems. As such, they serve as a critical unit for revealing ecosystem multifunctionality and for informing the spatial optimization of ecosystem services.

In recent years, researchers have increasingly adopted the framework of ecosystem service bundles (ESBs) to characterize the multifunctional patterns of ecosystem services across different regions [8,9]. This analytical approach has proven effective in identifying potential trade-offs and synergies among services and provides valuable support for developing targeted regional management strategies [10,11]. For instance, Liu et al. employed the Spearman correlation coefficient analysis in combination with principal component analysis (PCA) and K-means clustering to explore the trade-offs and synergies of ecosystem services in a semi-arid valley basin. They identified ESBs under multiple future scenarios, offering theoretical and practical guidance for sustainable land use, ecological spatial optimization, and adaptive ecosystem management [12]. Similarly, Qi Liu et al. applied the Pearson correlation method to examine the interactions among various ecosystem services in the Yellow River Basin and used a self-organizing map (SOM) model to identify eight distinct ESBs. Their study highlighted the importance of understanding service interactions at the bundle level to inform regional ecosystem management [13]. Liao et al. integrated Moran’s I, PCA, the NbClust function, and K-means clustering to investigate ESBs in southern Jiangxi Province, China. Their results revealed that trade-offs were most likely to occur between provisioning services (e.g., food production) and regulating or cultural services. This study demonstrated that ESBs not only reflect the nature of service interrelationships but can also be used to delineate ecological functional zones, thereby providing a spatial foundation for green space planning strategies [14].

The spatial patterns of ecosystem service bundles (ESBs) exhibit high heterogeneity [15], driven by a range of influencing factors. These include anthropogenic drivers such as urban expansion, industrial restructuring, changes in population density, and vegetation restoration programs, as well as natural factors like climatic conditions, topography, and soil characteristics [16,17,18]. Existing studies have shown that identifying the spatial distribution of ecosystem service bundles (ESBs) and understanding their driving mechanisms are key directions for advancing ecological restoration and enhancing regional ecosystem service capacity [19].

Although research on ecosystem service bundles (ESBs) and the trade-offs and synergies among ecosystem services has been increasing, it has largely focused on broad spatial scales such as national, watershed, or large regional levels [20,21,22]. Most of these studies have been conducted at a single spatial scale, overlooking the significant scale dependency between ecosystem services and their driving factors [23,24]. Neglecting scale effects may obscure the true interactions among key services and lead to misinterpretations in policy-making and spatial planning processes [25]. Given that China’s current territorial spatial planning system comprises multiple levels of decision-making, the scientific management of ecosystem services urgently requires research from a multi-scale perspective to support coordinated governance across spatial levels [26].

The urban–rural fringe (URF) lies in the transitional zone between urban and rural areas [27]. As a region with characteristics of both urban and rural land use, the URF plays a unique role in urban construction and development during rapid urbanization, providing large amounts of land and space. It strengthens connections between urban centers and surrounding hinterlands, with increasingly intertwined relationships and blurred boundaries [28,29]. However, it is also a frontier of urban–rural gradient expansion, where ecological land is most severely encroached upon by urban construction land, leading to highly fragmented landscapes [30]. Despite the coexistence of ecological degradation and regeneration, the urban–rural fringe (URF) occupies a distinct and strategic position within the urban ecological structure. On the one hand, compared to urban cores, URF areas often retain a higher proportion of green spaces and remaining natural patches, offering relatively strong ecological functions. These include supporting regional biodiversity, regulating microclimates, and conserving water resources. On the other hand, due to the relatively flexible and transitional nature of land use in these areas, they hold significant potential for ecological restoration and spatial reorganization. Although the ecological importance of URFs has gained increasing attention, systematic investigations into the trade-offs and synergies among ecosystem services and their driving mechanisms in these regions remain limited. Most existing studies focus on either urban centers or rural hinterlands, while the URF is often overlooked due to its blurred spatial boundaries and high socioeconomic heterogeneity [31]. However, the condition of ecosystem services in the urban–rural fringe has a profound influence on the overall ecological sustainability of cities. Therefore, advancing research on ecosystem services in these transitional zones has become increasingly urgent to support urban ecological protection and sustainable development.

Beijing is one of the fastest-changing cities in the world and, after Shanghai, has the highest urbanization rate in China [32]. With population growth and accelerated urbanization, human interference with the natural ecosystems of Beijing’s URF has increased in both scale and intensity. The “Beijing Urban Master Plan (2016–2035)” emphasizes the need to “strictly control urban growth, adhere to the resource and environmental carrying capacity as rigid constraints, and implement reductions in scale, burden, and density”. The URF is a key area for constructing ecological security patterns in the plains, mitigating potential hazards to the capital, and preventing uncontrolled urban sprawl. It is also a focal area for population control, the relocation of non-capital functions, industrial transformation, and environmental pollution control across the city [33].

Based on this context, this study takes the urban–rural fringe (URF) of Beijing as a case area and addresses pressing issues under rapid urbanization, including the degradation of ecosystem service (ES) quality, increasing complexity in the relationships among ESs, and unclear management scales. Specifically, it aims to answer the following research questions: (1) Do significant trade-offs and synergies exist among ecosystem services in Beijing’s URF? (2) Do these relationships exhibit spatial heterogeneity across different scales? (3) What are the key driving factors behind these patterns? To address these questions, the study employs a combination of methods—including the Spearman correlation analysis, Global Moran’s I, hotspot analysis, self-organizing maps (SOM), and redundancy analysis (RDA)—to reveal the spatial distribution patterns of ecosystem services and their trade-offs and synergies across multiple scales, as well as the primary socio-ecological drivers in the URF of Beijing in 2022. Based on the quantified relationships between ESs, their trade-offs, synergies, and the socio-ecological drivers of ESs, this study develops spatial planning and management strategies tailored to different scales, aiming to provide scientific evidence and references for regional management.

2. Study Area and Data Sources

2.1. Study Area

In China, the State Council’s official document Notice on Strengthening the Supervision and Administration of Urban and Rural Planning clearly defined the criteria for delineating urban–rural fringe (URF) areas. According to the document, URF areas have the following characteristics: (1) they are adjacent to urban centers, sharing both urban and rural features, and are administratively governed by suburban townships; (2) non-agricultural industries are well developed; and (3) population density falls between that of urban and rural areas and continues to change with urban expansion [34]. Beijing is located at the northwestern edge of the North China Plain, between 115°20′–117°30′ E and 39°28′–41°05′ N. According to the Beijing Urban Master Plan (2016–2035), the spatial scope of the URF is defined in terms of two boundaries: the policy zone boundary and the implementation unit boundary. The policy zone boundary includes areas outside the planned centralized construction zone between the Fourth and Sixth Ring Roads. The implementation unit encompasses all township-level administrative districts involved in the policy unit boundary. Considering research validity and data availability, this study selected the implementation unit of the URF as its research scope, located between 115°58′–116°47′ E and 39°37′–40°15′ N. It covers a total of 109 complete township-level administrative districts across 10 municipal districts, with a total area of approximately 2780.97 km² (Figure 1).

The URF has a terrain that slopes from high in the west to low in the east, with an average elevation of 61.42 m. The climate is characterized as a warm temperate, semi-humid, and semi-arid monsoon climate. Summers are hot and rainy, winters are cold and dry, and spring and autumn are brief. The annual precipitation ranges from 500 to 700 mm, and the annual average temperature is between 10 °C and 12 °C [35]. The main rivers flowing through the area include the Yongding River, Wenyu River, Qing River, and North Canal. The primary land use types are township and village land, as well as cropland, accounting for 73.09% and 14.47% of the total area, respectively. This region is critical for promoting high-quality, modern development, supporting the relocation of non-capital functions, building new development frameworks, implementing rural revitalization strategies and environmental improvement, fostering deep urban–rural integration, and meeting the growing demand for a better quality of life among residents [36].

2.2. Data Sources

Based on local administrative divisions and geographic and hydrological distribution characteristics, the study area was divided into 3131 grids of 1 km × 1 km, 434 grids of 3 km × 3 km, and 109 township-scale administrative units. Multi-source datasets were used to evaluate the spatial distribution of ESs, their TOSs, and the identification and quantification of their driving factors. The datasets include basic geographic data, satellite remote sensing data, soil- and vegetation-related data, meteorological data, socio-economic data, and statistical data. For specific information such as dataset names, data sources, data resolutions, and data formats, please refer to Supplementary Material Table S1. All raster data were resampled to a spatial resolution of 30 m × 30 m.

3. Research Methods

3.1. Quantification of ESs

The classification of ecosystem services in the urban–rural fringe of Beijing was determined based on the following considerations. First, the Millennium Ecosystem Assessment (MA), one of the most authoritative and influential global initiatives in the field of ecosystem services, classifies ecosystem services into four categories: provisioning, supporting, regulating, and cultural services [4]. This study follows the MA framework. Second, several official planning documents—including the Beijing Urban Master Plan (2016–2035) [33], the Beijing Landscape and Greening Special Plan (2018–2035) [37], the Development Plan for Green Isolation Areas in Beijing during the 14th Five Year Plan Period [38], and the Three-Year Action Plan for Reducing Development in Urban–Rural Areas of Beijing (2021–2023) [39]—have emphasized the need to improve the ecological environment in urban–rural fringe areas. These plans highlight measures such as enhancing food and freshwater supply capacities, improving soil and water conservation, strengthening air and agricultural pollution control, and promoting ecotourism development. Third, based on nearly 20 years of Beijing Ecological Environment Status Bulletins [40] and relevant academic studies [41], urban–rural fringe areas in Beijing face multiple environmental pressures, including ecosystem degradation, water scarcity, and non-point source agricultural pollution. In addition, food security and national carbon neutrality goals further emphasize the ecological importance of these areas. Accordingly, this study selected eight ecosystem service indicators. These include one supporting service: habitat quality (HQ); two provisioning services: food production (FP) and water yield (WY); four regulating services: soil conservation (SC), carbon sequestration (CS), nitrogen purification (NP), and phosphorus purification (PP); and one cultural service: recreational opportunities (RO).

In this study, habitat quality, annual water yield, soil conservation, nitrogen purification, phosphorus purification, and carbon storage in the urban–rural fringe of Beijing in 2022 were estimated using corresponding modules of the InVEST model, specifically: the Habitat Quality, Annual Water Yield, Sediment Delivery Ratio, Nutrient Delivery Ratio, and Carbon Storage and Sequestration modules. Previous studies have demonstrated a significant linear correlation between agricultural production and NDVI [42]. Accordingly, we used agricultural yield data for major crops in various districts of Beijing’s urban–rural fringe in 2022, as reported in the 2023 Beijing Regional Statistical Yearbook [43], and spatialized these statistics based on NDVI raster data and land use data to generate a spatial distribution map of major crop production. Following the approach of Komossa et al. and Zhao et al., NDVI, slope, and DEM were selected as key indicators for evaluating recreational opportunities [44,45]. Detailed calculation methods for the eight ecosystem service indicators used in this study are provided in Supplementary Table S2 [46,47,48,49,50,51,52,53,54,55,56,57].

All calculations were performed using the InVEST model and ArcGIS 10.6. The resulting outputs were processed using the Zonal Statistics tool in ArcGIS 10.6 to compute the mean values, and ecosystem service values were subsequently aggregated to 1 km × 1 km grids, 3 km × 3 km grids, and the township level. The 1 km × 1 km grid scale was selected to ensure data accuracy, while the township scale corresponds to the basic unit commonly used for landscape planning and administrative decision-making. The 3 km × 3 km grid scale represents an intermediate resolution, approximately reflecting the average area between the 1 km grids and township units.

3.2. Exploratory Spatial Data Analysis

Global spatial autocorrelation analysis is a fundamental tool in spatial statistics, used to determine whether a given variable exhibits a clustered or dispersed distribution across the entire study area. It provides essential insights into the spatial structure and distribution patterns of ecosystem services [58]. Accordingly, this study employed exploratory spatial data analysis to examine the spatial correlation and heterogeneity of eight ecosystem services across the urban–rural fringe of Beijing at three spatial scales: 1 km × 1 km grids, 3 km × 3 km grids, and the township level. The equation is as follows:

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

(1)

In Equation (1), $n$ represents the number of spatial units for the specified ESs, such as the total number of 1 km grids, 3 km grids, and townships; $x_i$ and $x_j$ are the attribute values of the ESs for units $i$ and $j$; $\overline{x}$ is the mean attribute value of the ESs; and $w_{ij}$ is the spatial weight matrix, which adopts an inverse distance weighting standard in this study. Moran’s I ranges from [−1, 1], where values greater than 0 indicate spatial clustering—the larger the value, the stronger the spatial clustering; values less than 0 indicate spatial dispersal—the greater the absolute value, the stronger the spatial dispersion; and a value of 0 indicates a random spatial distribution.

The Getis-Ord Gi* statistic in ArcGIS 10.6 was employed to identify the spatial distribution of statistically significant high-value clusters for the eight ecosystem services at the 1 km, 3 km, and township scales [59]. These significant high-value areas represent regions with a stable and consistently high capacity for providing ecosystem services over time [60]. Each ES defined its top 10% as hotspots, with hotspot and non-hotspot values assigned as 1 and 0, respectively [61]. The hotspots of all ESs were overlaid to create a hotspot distribution map of ESs in the URF of Beijing.

3.3. Interactions Among ESs

Simultaneous increases or decreases in ES indicators indicate the existence of trade-offs or synergies among them. To analyze the relationships among various ESs, the Kolmogorov–Smirnov test was conducted, revealing that not all ES data followed a normal distribution [62]. Therefore, Spearman’s non-parametric correlation analysis was performed using the “corrplot” package in R 4.1.2 to measure the interactions between paired ESs. The equation is as follows:

$$r={\displaystyle \frac{\sum i(x_{ij}-\overline{x})(y_{ij}-\overline{y})}{\sqrt{\sum i{(x_{ij}-\overline{x})}^2\sum i{(y_{ij}-\overline{y})}^2}}}$$

(2)

In Equation (2), $r$ represents the correlation coefficient, ranging from [−1, 1]. When r > 0, it indicates a synergistic relationship between the two services—that is, the capacities of the two ecosystem services increase or decrease together. When r < 0, it implies a trade-off relationship, meaning that an increase in one ecosystem service corresponds to a decrease in the other [63]. The absolute value of $r$ reflects the strength of the TOSs, categorized into 3 levels: $\left|r\right|\ge 0.5$ indicates strong correlation, $0.3\le \left|r\right|<0.5$ indicates moderate correlation, and $0.1\le \left|r\right|<0.3$ indicates weak correlation [64]. $p<0.05$ denotes a statistically significant correlation. $x_{ij}$ and $y_{ij}$ represent the data values for different types of ESs. Before performing the calculations, each ES was normalized using Min–Max scaling in R 4.1.2. The equation is as follows:

$${ES}_{std}={\displaystyle \frac{{ES}_i-{ES}_{min}}{{ES}_{max}-{ES}_{min}}}$$

(3)

In Equation (3), ${ES}_{std}$ is the normalized value of any ES, ranging from [0, 1]. ${ES}_i$ is the original evaluation value [65]. To account for sensitivity to minimum and maximum values and avoid errors caused by outliers, ${ES}_{max}$ and ${ES}_{min}$ were set as the 95th and 5th percentiles of each ES, respectively [66].

3.4. Identification of ESBs

Given the complex interrelationships among ecosystem services (ESs) and their pronounced spatial heterogeneity, traditional clustering methods often face limitations when processing high-dimensional and nonlinear ecological data. The self-organizing map (SOM), an unsupervised neural network algorithm, offers robust capabilities in nonlinear dimensionality reduction and spatial pattern recognition. It effectively captures the spatial co-occurrence patterns of ESs and demonstrates strong robustness and interpretability [67]. Meanwhile, the NbClust function integrates multiple cluster validity indices to avoid biases associated with single-index evaluation, thereby improving the stability and scientific reliability of clustering results [68]. Therefore, this study adopts a combined approach using SOM and NbClust to identify ecosystem service bundles (ESBs) at three spatial scales within the urban–rural fringe of Beijing: 1 km × 1 km grid, 3 km × 3 km grid, and township level.

The SOM algorithm, originally proposed by Kohonen [69], employs a competitive learning mechanism to map high-dimensional input data onto a two-dimensional neural lattice, enabling the self-organized classification of spatial patterns [70,71]. In this study, the SOM network was constructed using the “kohonen” package in R to identify ESBs. The optimal number of clusters was then determined using the NbClust function, which synthesizes various clustering validation metrics, in combination with the elbow method. The elbow method identifies the “elbow point”—where the rate of decrease in the within-cluster sum of squares (WSS) significantly slows down—thus indicating the optimal number of clusters. This approach helps to avoid overfitting or underfitting caused by inappropriate selection of cluster numbers. All analyses were conducted using R version 4.1.2.

3.5. Analysis of Socio-Ecological Drivers of ESBs

Based on references [41,72], this study selected six representative potential driving factors from two dimensions—natural and socio-economic aspects—including digital elevation model (DEM), mean annual precipitation (PRE), normalized difference vegetation index (NDVI), mean annual evapotranspiration (ET), gross domestic product density (GDP), and population density (POP). The NDVI data were selected from August 2022, as the period from June to August represents the peak growing season for vegetation in Beijing. The data sources, spatial resolutions, and data formats mentioned above can be found in Supplementary Material Table S1. Data preprocessing was performed. First, using ArcGIS 10.6 and the “Zonal Statistics” tool, the raw data for the six potential socio-ecological drivers were aggregated to 1 km × 1 km and 3 km × 3 km grids, as well as the township level. The average value of each factor was calculated within each spatial unit, achieving spatial matching and standardization of driving factors across multiple scales. Next, a 95% winsorization procedure was applied to minimize the influence of extreme values, ensuring smoother and more stable data distributions. Min–Max normalization was then employed to eliminate dimensional differences among the socio-ecological drivers, allowing for comparisons on a consistent scale [73]. To reduce indicator redundancy, multicollinearity among the driving factors was examined using variance inflation factors (VIFs) in the R 4.1.2 environment with the “car” package. Factors with VIF > 7.5 were excluded from the analysis [74]. Finally, redundancy analysis (RDA) was conducted to explore the relationships between ecosystem services and socio-ecological drivers at different spatial scales. This analysis was performed in R 4.1.2 using the “vegan” and “rdacca.hp” packages.

4. Results

4.1. Spatial Patterns of ESs at Different Scales

In 2022, all eight ESs in the URF of Beijing exhibited clear spatial patterns across various scales (Figure 2). As the research scale increases from 1 km × 1 km grids to 3 km × 3 km grids and the township level, the spatial differences in individual ES provisioning weakened, and the spatial distribution became more concentrated. Spatially, the western part of the study area consists of shallow mountainous areas, while the central and eastern parts are relatively flat. Due to Beijing’s concentric urban spatial structure, the degree of urban development increases closer to the core of Beijing and gradually decreases outward. Except for provisioning services (i.e., food production and annual water yield), most other services reach high values in the western shallow mountainous areas and along the Yongding River and Wenyu River basins (Figure 2).

Provisioning services displayed similar spatial patterns across scales. High FP values were mainly distributed in the plain areas on the outer edges of the study area. The southern and eastern plains had the highest agricultural yields. At larger scales, such as the township level, some high-value areas were downgraded to medium-high-value areas. The annual water yield (WY) exhibits a consistent spatial pattern across the 1 km × 1 km grid, 3 km × 3 km grid, and township scales, with higher values in the southern part of the study area compared to the north. By comparing the water yield map in Figure 2 with the land use classification map of the urban–rural fringe in Beijing shown in Figure 1c, it is evident that areas with higher water yield were mainly concentrated in the ring-shaped zones near the city center and in the southern part of the study area, which are predominantly characterized by urban and village construction land. The high annual water yield in these areas may be attributed to the precipitation distribution in 2022, as well as the high proportion of impervious surfaces in urban areas and the resulting low rainfall infiltration rate [75]. In contrast, the northern, northeastern, and northwestern parts of the study area showed lower water yield. At the township scale, the low-value zones of annual water yield along the Xiaoqing River, Yongding River, and Wenyu River disappeared. Supporting services, such as HQ, and regulating services, such as NP, PP, and CS, exhibited similar spatial patterns. At the 1 km and 3 km grid scales, their high values are distributed as planar patterns in the western shallow mountainous areas, as linear patterns along the Xiaoqing, Yongding, and Wenyu Rivers, and as point patterns in other areas with strong habitat quality, nitrogen and phosphorus purification, and carbon storage capacity. Low-value areas were primarily concentrated in the central ring-shaped township and village land near the core urban area. As the research scale increased, the clustering of high-value areas became more apparent, and low-value areas decreased. SC, another regulating service, reached its maximum in the forest-dense western shallow mountainous areas. As the research scale increased from the 1 km × 1 km grid and 3 km × 3 km grid to the township level, the high-value areas for SC in the eastern and northern parts of the study area disappeared. Cultural services, such as RO, showed increasing clustering of high values in the western shallow mountainous areas and outer ring areas as the research scale increased, while low-value areas gradually decreased.

By comparing the annual nitrogen and phosphorus purification maps in Figure 2 with the land use classification map of Beijing’s urban–rural fringe in Figure 1c, it is observed that areas with strong nitrogen and phosphorus purification capacity are primarily distributed near forested areas and rivers. In contrast, areas with weaker purification capacity are mainly concentrated in flatter regions. This may be due to low vegetation cover and limited ability to intercept pollutants, increased nitrogen and phosphorus outputs from agricultural areas due to the use of pesticides and fertilizers, and intensified urban construction and human activities that lead to greater pollutant emissions, ultimately resulting in weaker purification capacity in these regions [76]. When the research scale expanded from the 1 km grid scale to the township scale, areas with high nitrogen and phosphorus purification capacity became significantly clustered in the western shallow mountainous areas, as well as near the Yongding River and Wenyu River basins.

4.2. Spatial Autocorrelation Analysis of ES Patterns at Different Scales

To study the impact of scale variation on the spatial distribution of ESs in the URF of Beijing, this study measured the spatial autocorrelation coefficients of ES patterns at different scales using exploratory spatial data analysis methods (

Table 1. Spatial autocorrelation coefficients of ESs in the study area at the three scales of analysis (1 km grid, 3 km grid, and township scales).

Scale	HQ	FP	WY	SC	NP	PP	CS	RO
1 km Grid	0.755	0.759	0.787	0.739	0.513	0.494	0.757	0.662
3 km Grid	0.677	0.720	0.727	0.766	0.506	0.491	0.713	0.626
Township	0.630	0.592	0.800	0.709	0.489	0.481	0.676	0.594
p	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001

Note: HQ: habitat quality; FP: food production; WY: water yield; SC: soil conservation; NP: nitrogen purification; PP: phosphorus purification; CS: carbon sequestration; RO: recreational opportunities.

). The results show that the Moran’s I values for all ESs at the 1 km grid, 3 km grid, and township scales were greater than 0 (p < 0.001), indicating significant geographic clustering of ESs at all three scales. Except for WY, the spatial correlation of ESs weakened as the scale increased. At the 1 km grid scale, the Moran’s I values were the highest, indicating the strongest spatial correlation of ESs at this scale.

The top 10% high-value areas of each ES were defined as hotspot regions, and their spatial overlay was used to generate ES hotspot maps for Beijing’s URF at the three scales (Figure 3). The results show that no single unit had a hotspot value of 7–8, indicating that the maximum number of ESs provided by a single unit in the area was 6. A comparison of hotspot spatial distribution characteristics across different scales reveals similar patterns with some localized differences.

Under the scenario involving the eight ecosystem service indicators selected in this study, areas providing 5–6 ESs were mainly concentrated in the western shallow mountainous areas of the URF region, where land types are primarily forests and garden plots. Areas providing 3–4 ESs were predominantly distributed along the Xiaoqing River, Yongding River, Liangshui River, Wenyu River, Chaobai River, and North Canal, with land types mainly consisting of forests, garden plots, grasslands, croplands, wetlands, and inland waters. Areas providing 1–2 ESs were primarily located in the southern and northeastern edge zones of the study area, where the main land types are townships, villages, and croplands. Areas with no hotspots were mainly distributed in the northern part of the study area, closer to the urban core, with township and village land as the dominant land type.

At the 1 km grid scale, hotspot distribution was more scattered and smaller in area. At the 3 km grid scale, hotspot distribution was more concentrated compared to the 1 km scale. At the township scale, the hotspot distribution pattern was slightly different from the 1 km and 3 km grid scales, with a coarser and more aggregated distribution.

Overall, the hotspots for the selected ES of Beijing’s URF at all three scales were concentrated primarily in the western shallow mountainous areas (

Table 2. Number and area of hotspots in the study area at the three scales of analysis (1 km grid, 3 km grid, and township scales).

Scale	1 km Grid		3 km Grid		Township
Hotspot Value	Area (km2)	Share (%)	Area (km2)	Share (%)	Area (km2)	Share (%)
0	848.923	30.526	827.669	29.762	819.519	29.469
1	1106.290	39.781	1199.892	43.147	876.537	31.519
2	300.851	10.818	356.054	12.803	401.016	14.420
3	257.835	9.271	155.381	5.587	318.822	11.464
4	94.367	3.393	56.967	2.048	20.238	0.728
5	31.321	1.126	23.966	0.862	78.868	2.836
6	141.378	5.084	161.041	5.791	265.967	9.564
7	0	0	0	0	0	0
8	0	0	0	0	0	0

). Approximately 8.7–13.1% of the area provided hotspots for 4–6 types of ESs. Although about 30% of the area had no hotspots, 57.4–59.8% of the area provided hotspots for 1–3 types of ESs, with areas having one hotspot accounting for the largest proportion of the study area across all three scales (31.5–43.1%). The proportion of areas without hotspots decreased as the scale increased.

4.3. Interactions Among ESs at Different Scales

The TOSs of ESs vary across different scales. By calculating the Spearman correlation coefficients, 28 potential TOS relationships were identified in 2022 for the URF of Beijing at three scales (Figure 4,

Table 3. TOSs and significance statistics in the study area at the three scales of analysis (1 km, 3 km, and township scale).

Scale	Significance (p < 0.05)	Non-Significance (p > 0.01)	Trade-Offs	Synergies
1 km Grid	28	0	12	16
3 km Grid	28	1	11	16
Township	28	1	9	18

). At the 1 km × 1 km grid scale, 12 ecosystem service (ES) pairs exhibited trade-off relationships. These included water yield with habitat quality, food production, soil conservation, nitrogen purification, phosphorus purification, carbon sequestration, and recreational opportunities; as well as food production with soil conservation, nitrogen purification, phosphorus purification, carbon sequestration, and recreational opportunities. In contrast, 16 ES pairs demonstrated synergistic relationships. Specifically, habitat quality showed synergies with food production, soil conservation, nitrogen purification, phosphorus purification, carbon sequestration, and recreational opportunities; soil conservation was synergistically linked with nitrogen purification, phosphorus purification, carbon sequestration, and recreational opportunities; nitrogen purification showed synergies with phosphorus purification, carbon sequestration, and recreational opportunities; phosphorus purification was synergistic with carbon sequestration and recreational opportunities; and carbon sequestration and recreational opportunities also exhibited a synergistic relationship. At the 3 km × 3 km grid scale, the number of trade-off and synergy pairs was 11 and 16, respectively. Notably, the relationship between habitat quality and food production shifted from synergy to trade-off, while the relationship between water yield and soil conservation became statistically non-significant. At the township scale, 9 ES pairs were identified as trade-offs and 18 as synergies. At this level, habitat quality and food production returned to a synergistic relationship, and food production became synergistic with nitrogen purification and phosphorus purification. The relationship between water yield and soil conservation remained non-significant. Overall, synergies dominated the interactions among ESs at all three scales. The direction of TOSs remained relatively stable across scales, while the strength of these interactions varied significantly. Supporting services, regulating services, and cultural services exhibited significant moderate-to-high synergies (0.3 ≤ r ≤ 1, p < 0.01), with correlation coefficients increasing as the scale expanded, indicating stronger synergies at larger scales. Provisioning services generally exhibited trade-offs with other types of services across all three scales. FP showed significant moderate trade-offs with SC, CS, and RO at all three scales (−0.5 < r ≤ −0.3, p < 0.01). However, the relationships between FP and HQ, WY, NP, and PP were inconsistent across scales, with FP showing weak correlations with HQ and WY. WY exhibited significant moderate-to-high trade-offs with HQ, NP, PP, CS, and RO across all three scales (−1 ≤ r ≤ −0.3, p < 0.01). The relationship between WY and SC was a strong trade-off at the 1 km grid scale but became insignificant at the 3 km grid and township scales.

4.4. Spatial Patterns of ESBs at Different Scales

Clustering analysis was conducted on eight ecosystem services in the urban–rural fringe of Beijing in 2022 using the NbClust function in R, at three spatial scales: 1 km grids, 3 km grids, and township units. The elbow method was applied to determine the optimal number of clusters by analyzing the relationship between the number of clusters (K) and the within-cluster sum of squares (WSS). Distinct “elbows” were observed near K = 4 across all three spatial scales, indicating a clear inflection point where increasing the number of clusters yields diminishing returns (Figure 5). Therefore, four clusters were identified as the optimal solution at each scale, demonstrating both stability and representativeness. Figure 6 presents the ecosystem service bundles (ESBs) derived from the self-organizing map (SOM) model when K = 4, across the three spatial scales. As the spatial scale increased, fine-grained spatial patterns gradually diminished, and the spatial aggregation of ecosystem services became more pronounced, i.e., the bundled ESs were prone to spatial aggregation when the scale was coarser, similar findings were reported by Gao et al. [6]. Based on the dominant ecosystem services and their spatial distributions within each cluster, the ESBs were categorized and named as follows: (1) an “ecologically restricted cluster”, located near the urban core, dominated by water yield but with low levels of all other services; (2) a “high-quality ecological cluster”, situated in the western part of the study area, characterized by high levels of habitat quality, soil conservation, nitrogen and phosphorus purification, carbon storage, and recreational opportunities; (3) an “ecologically balanced cluster”, found in the northern region, where ecosystem services were relatively balanced; and (4) a “food production cluster”, located in the southern area, dominated by food production and water yield. According to the richness of ecosystem services provided by each ESB, the study area was classified into four ecological functional areas, ranked from low to high service richness: ecologically restricted area, food production area, ecologically balanced area, and high-quality ecological area.

Ecologically restricted areas showed similar distributions across the three spatial scales, corresponding to ESB1, ESBI, and ESBa at the 1 km grid, 3 km grid, and township scales, respectively. These areas were in the central part of the study region near Beijing’s core area, dominated by township and village land use in densely populated plain regions. ESs were limited, primarily focused on WY within provisioning services. As the study scale increased, the proportion of ecologically restricted areas gradually decreased (Figure 7).

Food production areas also exhibited similar distributions across the three spatial scales, corresponding to ESB2, ESBII, and ESBb, at the 1 km grid, 3 km grid, and township scales, respectively. These areas were in the southern plains of the study region, far from the core area, dominated by townships, villages, and croplands, serving as a hotspot area for FP and exhibiting a certain level of WY services. As the research scale increased from the 1 km × 1 km grid and 3 km × 3 km grid to the township level, the proportion of FP areas gradually expanded.

Ecologically balanced areas showed slight variations across the three spatial scales, corresponding to ESB3, ESBIII, and ESBc, at the 1 km grid, 3 km grid, and township scales, respectively. At the 1 km grid scale, these areas were distributed more finely, primarily in the northern, eastern, and southern plains of the study region, far from the core area, overlapping with croplands, garden plots, forests, grasslands, and inland waters. As the study scale increased, these areas became coarser and more spatially aggregated. At the township scale, ecologically balanced areas disappeared in the southeastern part of the study region but formed large clusters in the southwestern part. These areas had higher proportions of garden plots, forests, grasslands, and inland waters than ecologically restricted areas and food production areas, providing diverse regulating, provisioning, supporting, and cultural services with a relatively balanced supply of ESs.

High-quality ecological areas showed similar distributions across the three spatial scales, corresponding to ESB4, ESBIV, and ESBd, at the 1 km grid, 3 km grid, and township scales, respectively. These areas were mainly clustered in the western shallow mountainous areas, dominated by forests with minimal human activity, exhibiting extremely high levels of supporting, regulating, and cultural services. Across all scales, the proportion of high-quality ecological areas within the total study region remained relatively stable.

Based on hotspot analysis, ESB4, ESBIV, and ESBd were in regions with hotspot values of 5–6; ESB3, ESBIII, and ESBc in regions with hotspot values of 3–4; ESB2, ESBII, and ESBb in regions with hotspot values of 1–2; and ESB1, ESBI, and ESBa in regions with hotspot values of 0 (Figure 3 and Figure 6). This further highlighted the distinctions among ecologically restricted areas, food production areas, ecologically balanced areas, and high-quality ecological areas.

4.5. Driving Factors of ESs at Different Scales

RDA (

Table 4. Explanation, variance contribution, and significance of socio-ecological drivers’ impact on ESs at the three scales of analysis (1 km grid, 3 km grid, and township scales).

First-Level Indicator	Second-Level Indicator	1 km Grid			3 km Grid			Township Scale
		EXP (%)	IC (%)	p	EXP (%)	IC (%)	p	EXP (%)	IC (%)	p
Biophysical indicator	NDVI	37.5	53.8	0.001 ***	38.3	52.1	0.001 ***	41.2	55.4	0.001 ***
	PRE	22.8	32.7	0.001 ***	23.1	31.1	0.001 ***	23.5	25.2	0.001 ***
	DEM	6.3	9	0.001 ***	7.5	10.1	0.001 ***	4.9	5.3	0.001 ***
	ET	0.2	0.3	0.001 ***	0.9	1.2	0.001 ***	3.4	3.6	0.001 ***
Socio-economic indicator	POP	0.4	0.5	0.001 ***	0.5	0.7	0.016 *	0.4	0.4	0.018 *
	GDP	2.6	3.7	0.001 ***	3.5	4.8	0.001 ***	9.8	10.1	0.019 *

* p < 0.05; *** p < 0.001. EXP: explained variance; IC: individual contribution of variance.

) revealed significant correlations between ESs and driving factors in the URF of Beijing in 2022 at the 1 km grid, 3 km grid, and township scales (p < 0.05). The influence strength of socio-ecological drivers on the distribution of ES varied, and the cumulative explanatory power of axes 1 and 2 in RDA increased as the scale expanded. At the 1 km grid, 3 km grid, and township scales, the analysis explained 69.6%, 73.8%, and 83.3% of the variance in ESs distribution (adjusted R²), respectively. At all three scales, the NDVI and PRE were the two most explanatory factors, with total explanatory power of 60.3%, 61.4%, and 64.7% at the 1 km grid, 3 km grid, and township scales, respectively, increasing as the scale expanded. DEM and GDP followed, with total explanatory power of 8.9%, 11.0%, and 14.7% at the three scales, ranking second only to NDVI and PRE (Table 4). The driving direction of different socio-ecological factors varied for different ESs. As shown in Figure 8, NDVI and DEM were positively correlated with SC, CS, HQ, NP, PP, and RO. This indicates that areas with dense vegetation and high altitudes in the study region provided higher levels of regulating, cultural, and supporting services, which is consistent with the findings of Zhou et al. [57]. Conversely, socio-economic factors such as POP and GDP were negatively correlated with SC, CS, HQ, and RO. This suggests that regions with high population density and economic activity exhibited lower levels of regulating, cultural, and supporting services. PRE was the main positive determinant of the spatial distribution of WY, indicating that precipitation directly determined WY levels in the study area. Additionally, FP was primarily influenced by a negative correlation with DEM at all three scales. FP was negatively influenced by POP and GDP at the 1 km grid and township scales, but positively influenced by these factors at the 3 km grid scale.

5. Discussion

The URF of Beijing, located in the transitional zone between the central urban area and the outer suburbs, represents an important component of the ecological landscape. This study provides a multi-scale approach to analyzing the spatial patterns of ESs, their trade-offs, and synergies at the 1 km gird, 3 km gird, and township scales. It explores the relationships between ESs and socio-ecological drivers at different spatial scales, offering a comprehensive understanding of the multi-scale differences in ESs. This study provides new insights into the effects of spatial scale on ESBs, enhances understanding of how socio-ecological drivers shape ESs across scales, and incorporates the scale dependence of ESs into hierarchical governance, contributing to sustainable regional ecological management.

5.1. Scaling Characteristics of ESs Spatial Patterns and Relationships

The spatial distribution of ESs in the URF of Beijing exhibits significant spatial heterogeneity, but the spatial distribution of the same ES shows similarities across different scales. The western shallow mountainous areas, serving as a transition zone between mountains and plains, act as a critical ecological source and barrier for the capital. This region, dominated by ecological control zones, limits urban and agricultural expansion and population growth to some extent. Efforts such as Beijing’s two rounds of million-acre afforestation and greening projects [77] and the Protection Plan for the Shallow Mountainous Areas of Beijing (2017–2035) [78] have created a connected and systematic green space network, improving ecological quality and stability. These factors support HQ, NP, PP, SC, CS, and RO. In addition, with the continuous development of green spaces in green isolation areas and the advancement of water ecological protection and restoration efforts in Beijing, as of 2022, the “first green isolation belt” between the Fourth and Fifth Ring Roads has been completed, and more than 40 parks have been opened in the “second green isolation belt” near the Sixth Ring Road. The water and soil conservation functions of Beijing’s five major river basins have been continuously enhanced. Consequently, large green spaces such as the Xiaoqing River, Yongding River, and Wenyu River banks, as well as major forest parks, suburban parks, and wetland parks, provide high-quality HQ, NP, PP, CS, and RO. Conversely, the central ring of the study area near the urban core is dominated by township and village land use. Rapid urban expansion, high population density, and encroachment on green spaces have destabilized ecosystems, reducing nitrogen and phosphorus purification capacity, increasing carbon loss, and weakening recreational functions. The spatial distribution of WY is primarily determined by annual precipitation, largely unaffected by other socio-ecological factors [79]. Consequently, low WY is observed in areas with low precipitation, while high WY occurs in rainfall-rich regions, contrasting significantly with other ESs.

Since the biophysical interactions between ESs may depend on scale, changing the research scale could strengthen, weaken, or reverse relationships among ESs [80]. Additionally, the magnitude of driving effects may vary across scales, changing the relationship between ESs [81]. In this study, most ES relationships were consistent across scales, with a few reversing due to scale changes, indicating that ES TOSs in Beijing’s URF are relatively stable and hold significant potential for adaptive environmental governance and spatial management.

Correlation analysis revealed significant moderate-to-high synergies among supporting, regulating, and cultural services, with stronger synergies at larger scales. This suggests that multiple types of services may be better realized at the township scale. High CS areas typically have high vegetation cover [82] and showed strong synergies with HQ, SC, and RO (0.5 ≤ r ≤ 1, p < 0.01) across all three scales, reflecting the influence of forests, garden plots, and grasslands. Provisioning services, however, generally exhibited trade-offs with other services. FP had significant moderate trade-offs with SC, CS, and RO across all three scales due to the negative impacts of food production on soil conservation, carbon sequestration [83], and recreational opportunities. FP also had moderate trade-offs with NP and PP at the 1 km and 3 km grid scales, as food production requires substantial nitrogen and phosphorus input. Agricultural residues often accumulate nitrogen and phosphorus in the soil, reducing purification capacity and highlighting agriculture as a major source of non-point pollution [84]. However, at the township scale, FP showed weak synergies with NP and PP, reversing its relationship at smaller scales. This suggests that agriculture is not the sole source of nitrogen and phosphorus pollution, consistent with findings by Jian Zhang et al. [82]. WY showed significant moderate-to-high trade-offs with NP, PP, CS, and RE across all three scales, likely due to the large proportion of township and village land generating runoff. Township and village land also exhibited weak supporting, regulating, and cultural services. At the 3 km and township scales, the relationship between WY and SC was not significant. Across the three scales, the relationships between WY, HQ, and FP exhibited opposite patterns at different scales. WY showed a significantly strong trade-off relationship with RO, but as the scale increased, the correlation first decreased and then increased. This is because the spatial distribution of WY depends on annual average precipitation, which has distinct spatial characteristics across the three scales. Moreover, the trend of spatial clustering intensity for WY differs from that of other ESs. As the scale expanded, the Moran’s I value of WY decreased from 0.787 to 0.727 and then increased to 0.800, whereas the spatial clustering intensity of other ESs decreased with increasing scale. This discrepancy also caused the TOS relationships involving WY to exhibit cross-scale variations different from other ES pairs, which is consistent with findings from studies by Yuli Deng, Qiang Liao, and others [14,79].

5.2. Scale Effects of ESBs, ESs, and Their Socio-Ecological Drivers

This study analyzed the spatial distributions of ESs and ESBs at the 1 km grid, 3 km grid, and township scales. Specifically, when the evaluation scale was smaller, the spatial information of ESs and ESBs became more detailed, consistent with the findings of Wanxu Chen, Muyi Huang, and Yongxiu Sun [85,86,87]. Differences in the spatial locations and extents of ESs and ESBs at different scales suggest that selecting an appropriate research scale based on study objectives is critical when formulating ES management strategies to better meet stakeholder needs. Management strategies should also consider national spatial planning and land use controls. Using grids as evaluation units helps overcome differences in geomorphology, elevation, and administrative size among URF districts, facilitating the design of finer-scale ecological management and compensation measures [88]. Furthermore, research at the township scale enables policymakers to plan and manage ecosystem spaces within basic administrative units [89,90].

At the three research scales, the overall spatial distribution of ESBs was similar across scales, with approximately 40% of the land classified as ecologically balanced areas, making this the largest category. About 29% of the land was classified as ecologically restricted areas, 24% as food production areas, and 7% as high-quality ecological areas (Figure 6 and Figure 7). Differences in ESB distributions emerged at different scales; coarser scales resulted in more aggregated and generalized ESB patterns, better capturing the effects of rapid land use changes on ES variations. The land use composition revealed that areas with high ESB supply intensity had a higher proportion of forests and a lower proportion of croplands, townships, and villages. Overall, the URF of Beijing showed balanced ecosystem services supply across certain regions, but attention is needed for areas with low ecosystem services supply (ecologically restricted areas and food production areas), where targeted governance policies should be implemented.

The direction and intensity of the potential impacts of social and natural factors on ESs vary with scale [23]. Axes 1 and 2 of RDA explained 69.6%, 73.8%, and 83.3% of ES spatial distribution at the 1 km, 3 km, and township scales, respectively (p < 0.05), demonstrating that the six selected socio-ecological drivers were the primary factors influencing the Ess’ spatial patterns in the study area. The spatial distribution of SC, CS, HQ, and RO are mainly positively driven by the NDVI and DEM, while negatively influenced by POP, GDP, and ET. This is primarily because areas with dense vegetation and high altitudes experience lower human disturbance, resulting in better-preserved ecosystems. These conditions help reduce soil erosion, enhance carbon absorption and storage, and provide higher-quality habitat and recreational services. Such regions also tend to be more attractive for leisure and recreational activities. Conversely, high population density and economic activities can lead to land use changes and vegetation destruction, increasing deforestation and soil erosion and reducing carbon storage capacity. This results in habitat fragmentation or environmental pollution, lowering habitat quality and diminishing the appeal of tourism resources. Areas with good vegetation typically exhibit high annual evapotranspiration, as forests and grasslands can effectively mitigate soil erosion and water loss, explaining the negative correlation between SC and ET. NP and PP are positively correlated with NDVI but negatively correlated with POP and GDP. This indicates that vegetation enhances nitrogen purification through root absorption and interaction with soil, while vegetation cover slows water flow, promoting the deposition and adsorption of phosphorus. In contrast, regions with high population density and intense economic activity often experience discharges of agricultural nitrogen, phosphorus, and potassium fertilizers, exceeding the ecosystem’s purification capacity and increasing its burden. At all three scales, the spatial distribution of WY is mainly positively driven by average annual precipitation, with other factors having a smaller impact. FP is primarily negatively influenced by DEM, as low-altitude plains are more favorable for agricultural activities, with most cropland located in these areas. Additionally, at different study scales, the effects of POP and GDP on FP are inconsistent. This suggests that, at certain scales, socioeconomic factors can boost food production, but they may also exert pressure on it, such as through rapid economic development and population growth, leading to reduced farmland or soil and water resource degradation. Kaiping Wang et al. noted that natural factors are the primary drivers of service clusters in the Beijing area, which is consistent with findings from this study in the URF of Beijing [41].

5.3. Implications for Regional Scale Management and Policy Formulation

The study area is in the URF, which, like other URFs domestically and internationally [91,92], represents a transitional region between urban and rural areas. This zone exhibits transitional characteristics in ecological, economic, and social aspects, forming a unique ES supply model. To enhance the resilience of ESs and balance the supply of different services, it is crucial to comprehensively and deeply assess how socio-ecological factors in the URF influence the spatial distribution of ESs and their cross-scale relationships [93,94]. This study emphasizes the need to formulate a comprehensive spatially sustainable management policy package based on the characteristics of ESs and ESBs at different spatial scales, along with the impact of social driving factors on ESs across spatial scales. These policies should include two levels: first, common strategies applicable to all three spatial scales; and second, customized strategies tailored to the characteristics of the grid scale and township scale.

Firstly, for high-quality ecological areas, ESs exhibit high levels of supporting, regulating, and cultural services across all three spatial scales. These areas can leverage their resource and environmental carrying capacities to maintain or further develop ecological and cultural tourism. For instance, based on current land use, forest parks, suburban parks, and wetland parks can be established to promote local fiscal revenues while providing economic incentives for maintaining these ESs [39].

Secondly, for food production areas, strict adherence to national farmland protection regulations is essential. Basic farmland and arable land protection lines must be upheld, and the relationship between urban expansion and farmland protection should be managed appropriately. Improving the quality of farmland, enhancing agricultural mechanization and intelligence, and boosting agricultural productivity are critical to ensuring food security and meeting public demand. Furthermore, ecological principles should be upheld by promoting modern agricultural technologies, advancing agricultural innovation, implementing precision fertilization, and optimizing nitrogen and phosphorus ratios in fertilizers to reduce pollutant emissions. Strengthening the agricultural non-point source pollution monitoring network, building an environmental supervision information platform for the URF, and improving pollution control assessment are also vital.

Thirdly, for ecologically balanced areas, the primary strategy should focus on preserving and maintaining the health and integrity of ecosystems while balancing the demands of agricultural production and ecological protection. In regions unsuitable for agriculture, reforestation and afforestation should be implemented to increase vegetation coverage and gradually reduce human activity intensity. In areas suitable for agriculture, effective measures should be taken to enhance the integrated allocation of agricultural and forestry resources under the premise of ecological protection. Utilizing the advantages of forestry and grass industries in soil and water conservation, carbon storage, habitat quality improvement, and nitrogen and phosphorus purification, diverse agroforestry management models should be explored to enhance land productivity and vegetation’s environmental improvement capabilities. These efforts can alleviate conflicts in land resource use while improving environmental and economic benefits.

Lastly, for ecologically restricted areas, mainly distributed in the ring-shaped region near the core area of Beijing, where township and village land use are highly concentrated, all supporting, regulating, cultural, and provisioning services, including food production, are relatively low. Based on the main adjustable driving factors at all three scales (NDVI, ET, and POP), the following strategies are recommended: (1) Establish ecological red lines to control urban expansion and reduce human disturbances to prevent further biodiversity loss. (2) Promote ecological restoration projects, integrate governance of mountains, rivers, forests, fields, lakes, and grasslands, and focus on the integrated development of forestry and water resources, emphasizing river, tidal flat, and wetland restoration. (3) Construct ecological barriers by continuing the afforestation and greening policies of the “million-acre forestation” project, advancing greening efforts in the URF, and building park belts to address the shortfall in ecological resources, optimize ecosystem patterns, and enrich residents’ ecological well-being. These efforts will expand green ecological spaces, perfect the ecological framework defined by the city’s overall plan, and enhance resilience to climate and natural disasters. For ecologically restricted areas, in addition to the above measures, real-time monitoring and assessment of environmental indicators in these areas should be conducted. Ecological red lines should be dynamically adjusted based on the assessment results.

The above strategies apply to all three scales. At the township scale, regional resource planning and layout should be carried out with administrative divisions as units, setting short- and long-term implementation goals. At the 1 km gird and 3 km grid scales, the spatial distribution of ESs and ESBs should guide site selection for tourism and green resources, farmland quality improvement, key nitrogen and phosphorus emission reductions, agricultural non-point source pollution monitoring, ecological red line establishment, and ecological restoration projects. Ecosystem service capacities should be adjusted and enhanced across scales by integrating ecosystem service drivers.

5.4. Limitations and Future Research

This study focuses on the URF of Beijing and uses parameter simulations to model the spatial distribution of ESs and ESBs in 2022. However, the research has potential limitations. First, the study evaluates 8 ESs that are currently at risk or policy-driven, which may overlook relationships with other important potential ESs. Second, the methods used in this study, which involve the InVEST model, NDVI, and other parameters to estimate FP and RO, are widely used but simplify ecological processes, which may result in discrepancies between estimated results and actual conditions. Lastly, although the methods used in this study effectively reflect the spatial distribution of ESs and ESBs, the TOSs between ESs, and the driving relationships between ESs and socio-ecological factors, they only provide a static analysis of ESs for a single year. As ESs are continuously changing, future studies should include a broader range of important ES categories, explore more precise ES assessment methods, and utilize long-term datasets to further clarify the temporal and spatial dynamics of ESBs.

6. Conclusions

This study focused on the urban–rural fringe area of Beijing and conducted a multi-scale assessment of ecosystem services (ESs) across 1 km grids, 3 km grids, and township scales. It systematically analyzed the spatial distribution of ESs, their trade-offs and synergies, and the socio-ecological driving mechanisms behind them. The spatial patterns of ecosystem service bundles (ESBs) across scales were also explored, offering practical insights for regional ecological management. The main conclusions are as follows:

(1)

The spatial heterogeneity of ecosystem services exhibits significant scale effects.

Driven by socio-ecological factors, the eight representative ESs show distinct spatial differences. Regulating, supporting, and cultural services are mainly concentrated in the western hilly areas, while food production is largely distributed across the peripheral plains. In contrast, water yield is primarily influenced by precipitation patterns. Spatial aggregation of ecosystem services becomes more pronounced at coarser scales.

(2)

The direction of trade-offs and synergies among ecosystem services remains stable across scales, while their strength varies significantly.

Across all three scales, synergies dominate the interactions among ESs, indicating cross-scale consistency. However, the intensity of these relationships is sensitive to spatial resolution, suggesting that scale must be carefully considered in ecological management and planning.

(3)

Ecosystem services and their trade-off and synergy patterns are jointly driven by multiple socio-ecological factors.

Across all scales, NDVI and mean annual precipitation consistently emerge as the primary drivers of the spatial distribution of ESs. However, the influence of socio-economic variables (e.g., population density and GDP) on specific services such as food production differs in direction across scales, underscoring the importance of integrating multi-scale information into decision-making processes.

(4)

Four distinct ecosystem service bundles were identified, providing a basis for ecological zoning.

Across all three spatial scales, four ESBs were delineated: ecologically optimized zones, ecologically balanced Zones, food production zones, and ecologically sensitive zones. Township-level divisions, due to their administrative operability, offer practical advantages for policy formulation and spatial management. In contrast, the 1 km and 3 km grid scales reveal clearer spatial associations between land use and ecosystem services, enhancing the understanding of land–ecosystem service interactions.

In conclusion, this study establishes a comprehensive framework integrating multi-scale spatial analysis, clustering methods, and driver identification for ecosystem service assessment. It provides scientific support for sustainable ecosystem management and spatial optimization in urban–rural fringe areas and offers a replicable pathway for ecological governance in other peri-urban regions.