Multi-scale spatiotemporal trade-offs and synergies of ecosystem services: implications for zoned ecological governance in the Shanxi section of the Yellow River Basin

Abstract

Clarifying the spatial distribution and evolutionary characteristics of ecosystem services (ESs) across multiple scales is crucial for maintaining ecological stability and sustainability. This study assessed the spatiotemporal changes of ESs in the Shanxi section of the Yellow River Basin (SYRB) from 2005 to 2020 using the InVEST model. We analyzed the multi-ecosystem service landscape index and their driving factors, explored the trade-offs and synergy at the county and 1-km scale, and formulated differentiated ecological management strategies based on local ecological characteristics. The findings demonstrated that (1) land use change was the dominant driver influencing ESs (2) carbon storage (CS) and habitat quality (HQ) showed a stronger synergistic relationship at both the county and 1-km scales; and (3) ecosystem service bundles were divided into four types: CS–HQ synergistic bundle, low synergistic bundle, key synergistic bundle, and integrated ecological bundle at the 1-km scale, while CS–HQ synergistic bundle, key synergistic bundle, water purification–water yield–soil conservation synergistic bundle, and integrated ecological bundle revealed at the county scale. From 2005 to 2020, an increasing number of regions transitioned from single ES to synergistic ESs. This study reveals the spatial mechanisms of trade-offs and synergies among ESs in SYRB, and constructs an effective partitioning scheme and targeted ecological management strategy for SYRB, which are of great significance for regional ecological conservation and sustainable development.

1 Introduction

With rapid economic development and accelerated urban construction, human interference in nature has become increasingly pronounced. Humans gain ecological benefits by modifying the structure and function of ecosystems, however, this process has simultaneously triggered a range of environmental issues, such as ecological space fragmentation and ecological function reduction (Newton et al., 2024). Large-scale land development, utilization, and functional conversion profoundly impact the operational mechanisms of ecosystems. In this context, amid rapid economic growth, exploring the balance between ecological environmental protection and socioeconomic development, and constructing a sustainable development model characterized by harmonious coexistence of human and nature, have become a critical contemporary challenge (Doncaster and Bullock, 2024).

Ecosystem services (ESs) refer to the various benefits that natural ecosystems provide to humans, including provisioning, regulating, supporting, and cultural services (Baskent, 2020). The scientific understanding, assessment, and management of ESs are not only relevant to biodiversity conservation but also serve as the foundation for achieving regional sustainable development and ecological security (Feng et al., 2023; Tao et al., 2022). The generation, delivery, and value realization of ESs are distinctly scale-dependent (Cao et al., 2025). The interaction between natural geographical processes and socioeconomic activities across different scales exerts a profound impact on spatial patterns and dynamic changes of ESs (Lindborg et al., 2017; Martin-Lopez et al., 2019). Therefore, an in-depth understanding of the multi-scale characteristics of ESs and their cross-scale associations is essential to improve the scientific and management precision. Within this context, multi-scale comprehensive evaluation frameworks and scale-based zoning management strategies have emerged as frontiers and research hotspots (Li S. et al., 2025; Zhang et al., 2024).

Given the spatial heterogeneity of ESs, the concept of ecosystem service bundles (ESBs) has been proven to be a powerful analytical framework for identifying spatially co-occurring sets of ESs with similar characteristics, thereby facilitating the systematic characterization of multifunctional landscapes (Shen et al., 2023). From a methodological perspective, recent research has progressively shifted from valuation of individual ecosystem services to the identification of integrated bundles and corresponding functional zoning. For example, spatially explicit clustering approaches was applied to characterize the internal structures and spatial disparities of ESBs in fast-urbanising areas including the Beijing–Tianjin–Hebei region (Li et al., 2024). Complementarily, integrated assessment frameworks combined with trade-off modelling facilitated ecological function zones at the county administrative scale in the Nansi Lake Basin (Liu et al., 2023). Alongside such methodological advances, research perspectives have expanded beyond static spatial mapping to incorporate dynamic supply–demand coupling analyses. A typical illustration derived from the Sichuan–Yunnan ecological buffer zone, where scholars tracked 15-year the spatiotemporal evolution of ESBs and quantitatively disentangled their dominant socioecological driving forces (Yang B. et al., 2024). Collectively, these advances underscore that quantitatively revealing the evolution, trade-offs, and synergistic relationships of ESs is significant for formulating scientific ecological protection policies and promoting the harmonious coexistence between humans and nature (Gao et al., 2022).

Current research on ESs has formed a mature paradigm consisting of model simulation,trade-off and synergy identification, driving factor detection and ESB classification, which is instrumental in characterizing the spatial patterns and internal relationships of ESs (Li and Li, 2025). A rich body of literature has systematically explored scale effects, scale mismatches and multiscale ES variations across diverse geographies (Chen et al., 2025), laying solid theoretical and methodological foundations for ES multiscale research. Nevertheless, fewer studies have conducted side-by-side comparative analyses between fine grids and administrative units specifically within the Loess Plateau. This research gap is particularly prominent in this geomorphic region, which is characterized by highly fragmented loess landforms and rigid, fixed administrative demarcations. Current regional ES assessments mostly adopted a single scale and fall into two separate strands: high-resolution grids evaluations capturing fine ecological heterogeneity, and county scale analyses oriented toward administrative governance. Each scale analysis has notable limitations, which underscores the necessity and urgency of multi-scale comparative research. While grid-scale analyses yield scientifically rigorous results, these findings are often difficult to translate directly into management policies at the administrative division level (Gao et al., 2025). This creates a disconnect between research outcomes and practical management needs that impedes their effective implementation. Conversely, simplistic county-scale analyses tend to obscure subtle changes in ecological functions driven by within-county differences in topography and land use, leading to an “average trap” (Chang et al., 2025) that precludes the accurate identification of key ecological issues and vulnerable areas. To address this dilemma, building on previous research findings (Wu Q. et al., 2025; Xu T. et al., 2025), we introduced a multi-scale comparative analysis framework and selected 1-km grid scale and county scale for targeted comparison aligned with the ecological and management characteristics of the Shanxi section of the Yellow River Basin (SYRB). The core rationale for this design is that the 1-km grid scale can effectively capture subtle fluctuations in ESs amid the fragmented terrain of the Loess Hilly and Gully Region, while balancing computational efficiency and assessment accuracy. As the most basic unit of administrative management and policy implementation in China, county-scale results are directly linked to local government performance evaluations and the design of ecological compensation mechanism (Tian et al., 2025), which accompanies closely with the practical management needs of the SYRB. By comparing results across these two scales, we can precisely diagnose the spatial mismatch between natural ecological boundaries and artificial administrative boundaries.

In September 2019, ecological protection and high-quality development of the Yellow River Basin (YRB) was elevated to a major national strategy in China. As a core region in the middle reaches of the YRB, the SYRB occupies an exceptionally pivotal strategic position. Located on the eastern edge of the Loess Plateau, the SYRB is one of the regions with the highest risk of soil erosion and most fragile ecological environment (Zhu et al., 2019). The SYRB is also an important ecological barrier that prevents the eastward invasion of sand and wind from moving southward, and has an irreplaceable role in guaranteeing the ecological security (Jiang et al., 2019). Accordingly, ESs research in the SYRB delivers regionally representative empirical evidence while directly responding to core demands of national ecological conservation strategies. Spatially, relevant assessments have evolved from coarse basin-scale evaluations to targeted analyses of key sub-basins and segments, such as the Henan (Niu et al., 2022) and Inner Mongolian (Yang J. et al., 2024). Distinct trade-off and synergy relationships among ESs were jointly shaped by local climatic conditions and human activities. Thematically, research scopes have expanded from separate quantification of individual ESs to integrated analytical frameworks that link ecological spatial patterns, environmental quality, and ecosystem service functions (Yue et al., 2025). Notwithstanding these advances, a prominent research gap remains. Most existing studies relied on a single fixed spatial scale and implicitly assumed that analytical scales aligned seamlessly with optimal ecological management scales. This assumption overlooks the pervasive scale mismatches between natural biogeophysical processes and human administrative boundaries. This problem is particularly acute in the SYRB, where highly fragmented loess landforms intersect with rigid administrative demarcations. We hypothesized that such scale mismatches impaired ecological governance efficiency in two primary ways: county-level analyses tended to mask fine-scale ecological spatial heterogeneity, and high-resolution grid-based analyses generated scientifically robust outputs difficult to translate into implementable administrative policies. Against this backdrop, this study established a multiscale comparative analytical framework that systematically contrasts results derived from 1-km raster grids and county administrative units. We aimed to quantify risks of governance inefficiency stemming from scale mismatches and identify feasible pathways for coordinated cross-scale ecological management.

In this context, this study explored ES interactions in the SYRB with three core objectives: (1) quantify spatiotemporal patterns of five dominant ESs and their driving factors from 2005 to 2020; (2) compare ES trade-off and synergy characteristics across 1 km grid and county scales, and unpack scale-dependent mechanisms; and (3) identify multiscale ES bundles and establish a zoned cross-scale governance framework to resolve natural-administrative scale mismatches. This study proposes differentiated ecological restoration strategies for the YRB, and provides empirical support for regional sustainable ecological development.

2 Data and methods

2.1 Study area

The SYRB (110°14′–114°33′E, 34°34′–40°31′N) is located in the middle reaches of the Yellow River (Figure 1). It encompasses the main stem of the YRB and its tributaries that traverse the interior of Shanxi Province of China (Ma J. et al., 2025). The Yellow River enters Shanxi Province at Laoniuwan, Pianguan County, and exits the province at Nianpangou, Yuanqu County, with an in-province reach of approximately 965 km, accounting for nearly one-fifth of the Yellow River’s total length. The SYRB covers 73.13% of Shanxi Province’s total land area, with elevation across the region ranging from 190 m to 3061 m. It contains key mountain ecosystems, including those of the Lvliang and Taihang Mountains. While approximately 20% of its total area comprises valley plains and basins (Wang et al., 2024b). However, the SYRB faces severe challenges including water scarcity, ecosystem degradation, and intensifying human–land conflict (Zuo et al., 2024). The region has a temperate continental monsoon climate, with a mean annual temperature of 10.0 °C and mean annual precipitation ranging from 350 mm to 680 mm. Long-term, intensive resource exploitation in the region has exerted substantial pressure on the local ecosystem, and the efficacy of ecological restoration and conservation measures is crucial to the overall high-quality development of the YRB.

FIGURE 1

2.2 Data sources and processing

This study selected 2005, 2010, 2015, and 2020 as the research time nodes, mainly for the following reasons: it is highly consistent with the cycles of China’s 11th to 13th Five-Year Plans, facilitating the analysis of the phased impacts of major national policies; in addition, these years have high data availability, ensuring the robustness and comparability of long-term analysis. The data sources are shown in Table 1. All datasets were uniformly resampled to a 30 m spatial resolution for ES analysis using ArcGIS software (Esri, Inc., Redlands, CA, USA), and projected to the new coordinate system (WGS 1984 UTM Zone 48N).

All continuous raster outputs generated by the InVEST model (WY, CS, SC, HQ) at 30 m resolution were aggregated to a uniform 1 km grid using block mean aggregation via the ArcGIS Aggregate tool. Given that a 1 km cell contains approximately 1,111 30 m subpixels, the arithmetic mean value of all valid subpixel ecosystem service magnitudes within each 1 km grid was assigned as the output cell value. A 1 km grid resolution was adopted in this study to balance computational efficiency and regional ecological assessment accuracy, consistent with the multi-scale matching framework proposed by Wang et al. (2026). Finer 30 m datasets capture micro-topographic details but lead to unaffordable computational burdens when coupled with county-scale socioeconomic data, while the 1 km scale facilitates the integration of biophysical simulation results and administrative governance units at the regional level, albeit with inevitable loss of micro-landscape heterogeneity.

2.3 Research framework

In this study, we performed a comprehensive spatiotemporal assessment of ESs in the SYRB over the period 2005–2020. Given the region’s well-documented challenges of severe water scarcity and widespread soil erosion, we focused on five core ESs critical to regional ecological security: water yield (WY), carbon storage (CS), soil conservation (SC), habitat quality (HQ), and water purification (WP). Our analytical workflow was structured as follows: (1) We quantified the five target ESs using the Integrated Valuation of Ecosystem Services and Trade-offs (InVEST) model, and derived the Multi-Ecosystem Services Landscape Index (MESLI) through standardized normalization of the ES results. (2) We employed the Geographical Detector (GD) model to disentangle the key driving factors underlying the spatiotemporal variations of individual ESs. (3) We explored trade-offs and synergies among ESs across multiple spatial scales via Spearman’s rank correlation analysis and geographically weighted regression (GWR). Finally, (4) We identified ESBs at different scales using the Self-Organizing Map (SOM) model, to determine the dominant service functions of each regional unit, and to develop scale-adaptive, targeted management strategies for regional ecosystem conservation and restoration.

2.4 Methods

2.4.1 Quantification of ecosystem services

The calculation methods are listed as follows (Table 2).

Localized parameter calibration was implemented for each ES module within the InVEST model. For the WY module, the Z parameter was assigned a value of 1.5, consistent with recent empirical investigations focused on the Loess Plateau (Song et al., 2025; Donohue et al., 2012). Modeled WY outputs were validated against multi-source observational datasets, including the Shanxi Provincial Statistical Yearbook and Shanxi Water Resources Bulletin, alongside existing InVEST-based research targeting Shanxi and the broader Loess Plateau (Xu Y. et al., 2025). For carbon storage, carbon density values were derived from field measurements in the SYRB (Wang et al., 2024b). To validate the rationality of the selected carbon density parameters, the simulated results were compared with the sampling survey data from previous studies conducted in the region. For SC assessment, the rainfall erosivity (R) factor and soil erodibility (K) factor were separately derived from meteorological observations and the Harmonized World Soil Database (HWSD). Indirect validation was performed by comparing spatial patterns of simulated potential soil loss against documented regional erosion distributions from published work (Zhang et al., 2022). With respect to HQ and nutrient retention modules, core parameter configurations followed well-established parameterization schemes from previous SYRB-focused research (Tang et al., 2024; Wu S. et al., 2025), and the reliability of the results was verified through comparison with published spatial patterns and trend analysis.

2.4.2 Calculation of MESLI

The MESLI is a comprehensive environmental indicator that can effectively quantify the capacity of a given region to supply multiple ESs (Rodríguez-Loinaz et al., 2015). This index is calculated as the sum of the standardized ES indicators.

Min–max standardization was implemented for positive ecosystem service indicators (i.e., WY, CS, SC, HQ), wherein larger values correspond to higher service provision. By contrast, reverse normalization was adopted for WP_N and WP_P—two metrics quantifying pollutant export loads instead of ecosystem purification capacity—to guarantee that elevated MESLI scores uniformly reflect superior comprehensive ecosystem service supply capacity. The calculation formula is expressed as follows:where i denotes the type of ES, n denotes the number of ESs, is the data for the ith ES, and and represent the maximum and minimum values of the ith ES over the entire study period (2005–2020), respectively.

2.4.3 Analysis of geographical detector

The GD is a spatial statistical method used to analyze the relationship between within-strata variance and total variance of the dependent variable, and to identify the driving force of each factor on the dependent variable by leveraging spatial stratified heterogeneity (Wang and Hu, 2012). The core analytical principle of the GD is rooted in the spatial stratified heterogeneity of independent variables. This model only accepts categorical independent variables as input; therefore, all continuous independent variables must undergo discretization (grading/classification) prior to inclusion in model calculations (Cao et al., 2013). In this study, we discretized all continuous variables using the natural breaks (Jenks) method and classified each variable into 5 categories. To evaluate the robustness of the adopted discretization scheme, we compared the variations in q-statistics across three class levels (3, 5, and 7 categories), along with results derived from the quantile and equal interval discretization methods. The results indicated that although the absolute values of the q-statistics exhibited slight fluctuations, the ranking of the relative importance of the main influencing factors remained stable, suggesting that the findings of this study were insensitive to discretization parameters and had strong robustness. The calculation formula for the q-statistic is expressed as follows:where h is the stratification of variable Y or factor X, and N are the number of units in layer h and the whole region, respectively, and are the variance of Y values in layer h and the whole region, respectively, SSW and SST are the intra-layer variance and total variance of the whole region, respectively, and q is the influence of the driving factor, which takes a value in the range of 0–1. The larger the q value, the stronger the explanatory power of the factor.

The spatial heterogeneity of ESs arises from the interaction of multiple influencing factors. In according with the actual situations of the study area and previous relevant studies (Li W. et al., 2025; Xu Y. et al., 2025), we initially selected potential driving factors from two broad categories: natural factors and socio-economic factors. We first performed multicollinearity diagnostics on raw continuous predictors (GDP, population density, NDVI, annual precipitation, DEM, mean annual temperature) following Williams et al. (2018). Any continuous factor with a variance inflation factor greater than 10 or strong pairwise linear correlation was removed to eliminate linear redundancy. This collinearity test only covered continuous variables and excluded the categorical land use dataset, as VIF algorithms are incompatible with categorical inputs. In addition, the root-restriction layer depth exhibited high collinearity with the Digital Elevation Model (DEM), and there was a spatial mismatch between their research scales. Therefore, this indicator was excluded. All reserved continuous driving factors were discretized into multiple grades via the natural breakpoint method to satisfy the categorical input requirement of GD. Land use type was retained directly without discretization due to its vital ecological significance for regional ESs; the GD model natively accepts categorical variables, enabling joint analysis with discretized continuous drivers. Ultimately, seven driving factors were retained for model operation:GDP (X1), population density (X2), land use type (X3), NDVI (X4), annual precipitation (X5), DEM (X6), and mean annual temperature (X7).

2.4.4 Quantification of trade-offs and synergies between ecosystem services

Following the identification of key driving factors for ES spatial heterogeneity using the Geographical Detector model, we next explored pairwise relationships among the five target ESs to systematically quantify trade-offs and synergies across the study area. For this purpose, we employed Spearman’s rank correlation coefficient analysis—a nonparametric statistical method that does not require normality assumptions for input data, making it particularly suitable for ES spatial data with skewed distributions (Xiao et al., 2024). We implemented all correlation analyses in R version 4.5.0, covering the entire 2005–2020 study period in the SYRB. To examine scale-dependent patterns in ES relationships, we performed the analysis at two distinct, ecologically and administratively relevant scales: (1) a fine 1-km grid scale, and (2) a coarse administrative county scale. Spearman’s rank correlation coefficients were interpreted at the conventional p < 0.05 and p < 0.01 significance levels as follows: positive values indicated a synergistic relationship between paired ESs, negative values indicated a trade-off, and values not significantly different from zero indicated no linear monotonic relationship.

GWR reveals the spatial differentiation characteristics of the research object at a specific scale by constructing local regression equations for each point across the spatial dimension. With ES variables as independent variables, the application of the GWR model to detect spatial interactions can reflect the spatial heterogeneity of trade-off and synergistic relationships in different regions (Xia et al., 2023). During model execution, we employed a fixed kernel function and optimized the bandwidth using the AICc minimization criterion to ensure the model could accurately capture the spatial heterogeneity of the variables. The modeling equation of GWR is as follows:where is the dependent variable, is the spatial location of point i, and are the intercept and regression coefficient of point i, respectively, is the value of the kth independent variable at the ith point, n denotes the number of independent variables, and denotes the random error. Positive regression coefficients calculated by GWR indicate a spatial synergy between the two variables, whereas negative regression coefficients suggest a spatial trade-off relationship.

Spearman’s rank correlation and GWR serve complementary roles in our analysis of ES trade-offs and synergies. Spearman’s correlation provides a global, scale-explicit assessment, indicating whether two ESs exhibit overall synergy or trade-offs across the study area. However, it cannot reveal where these relationships vary spatially. GWR fills this gap by capturing spatial non-stationarity, thereby revealing local variations in the direction and magnitude of ES relationships across sub-regions. In essence, while Spearman’s correlation defines the regional baseline, GWR pinpoints where this relationship intensifies, weakens, or reverses. This dual framework is widely adopted in ecosystem service interaction studies to separate broad-scale patterns from local-scale heterogeneities.

Fisher’s r-to-z transformation (Fisher, 1915) was adopted to test whether Spearman correlation coefficients of ecosystem service pairs exhibited statistically significant differences between the 1 km grid scale and county scale in 2005, 2010, 2015 and 2020, so as to quantitatively identify scale-dependent divergence of ecological service trade-offs and synergies. The calculation formula is expressed as follows:where Z₁ and Z₂ are the Fisher-transformed values of the correlation coefficients at the 1-km and county scales, respectively, and n₁ and n₂ are the corresponding sample sizes.

2.4.5 Identification of ecosystem service bundles

Following the multi-scale analysis of ES trade-offs and synergies, we identified ESBs to spatial group units with highly similar ES supply profiles. The SOM is an unsupervised neural network algorithm that preserves the topological structure of input data, making it well-suited for identifying non-linear, spatially explicit ES clusters (Qu et al., 2021). We implemented SOM analysis using the kohonen package in R version 4.5.0 to reveal the potential non-linear relationships among ESs and delineate ESBs across the SYRB. Within the WP service, WP_N and WP_P were highly correlated, and WP_N had been identified as the primary pollutant affecting ecosystem health in the broader YRB (Cui et al., 2024). Therefore, we selected WP_N as the representative indicator for WP to simplify subsequent clustering analysis. The SOM analysis workflow was structured as follows: First, we input the normalized, dimensionless values of the five target ESs into R version 4.5.0. Second, we determined the optimal number of ESBs by examining the trend of the total within-sum-of-squares values, a widely used criterion for unsupervised clustering (Xu et al., 2024). Third, we identified the final ESBs for the study area using the trained SOM model. Finally, we spatially visualized and mapped the ESB distribution using ArcGIS software (Esri, Inc., Redlands, CA, USA).

3 Results

3.1 Spatiotemporal distribution of ecosystem services

Evident spatial heterogeneity was observed in ESs across the SYRB over 2005-2020 (Figure 2). WY, HQ, CS, and SC showed a consistent spatial pattern: high values were concentrated in the mountainous areas on the eastern and western sides of the basin, while low values in the high urbanized central basin areas. Low HQ values were extensively observed across a large proportion of the study area, whereas regions with high HQ values were limited; this spatial pattern is consistent with the findings of Wang et al. (2023) and Yang et al. (2022). In contrast, WP_N and WP_P exhibited the opposite spatial distribution to the aforementioned four ESs, with high-value areas mainly concentrated in the central belt and the cultivated land on both sides of the basin. Increased WP_N and WP_P indicated a decline in water purification service capacity. This spatial pattern was spatially associated with the distribution of anthropogenic production activities and agricultural land uses, suggesting a potential linkage to non-point source pollution and phosphorus losses (Tu et al., 2025; Xu Y. et al., 2025).

FIGURE 2

In terms of changes in the spatial distribution of ESs from 2005 to 2020 (Figure 2), WY decreased in the southern part and increased in the northern region. HQ showed minimal overall change, and the regions with HQ declines were spatially consistent with the areas of expansion in construction land and cultivated land expansion. This spatial coincidence suggested a potential negative association between intensive land use conversion and habitat quality maintenance, consistent with findings reported in previous studies (Wang et al., 2023). CS remained relatively stable between 2005 and 2020. Areas with CS increases were spatially coincident with regions where cultivated land and construction land were converted to forest land, while CS declines occurred in areas where forest land was converted into other land use types. This spatial correspondence is consistent with the expectation that land use transitions involving forest establishment or removal exert measurable effects on carbon storage capacity. SC showed an overall increasing trend. The timing of this increase overlapped with the implementation of a series of ecological protection and restoration initiatives in Shanxi Province since 2006, suggesting a potential temporal correspondence that merits further investigation through causal attribution methods (Ma Y. et al., 2025). However, the SC capacity in basin areas and regions with rapid urbanization remained comparatively low, indicating a key priority for subsequent ecological governance. Although WP_N and WP_P remained stable from 2005 to 2020, areas surrounding construction land showed relatively higher nitrogen and phosphorus output values, potentially reflecting the combined effects of construction land expansion and agricultural fertilizer application. However, direct attribution to specific anthropogenic sources would require more detailed agricultural input and land management data.

During 2005 to 2020, WY, SC, and WP_P showed fluctuating upward trends, whereas WP_N, CS, and HQ showed decreasing trends (Figure 3). The observed trends of ESs in the SYRB from 2005 to 2020 were temporally coincident with multiple ongoing changes in both climatic conditions and anthropogenic activities. These co-occurring increases may be related to changes in regional precipitation patterns and large-scale vegetation restoration efforts such as the Grain for Green Program (returning farmland to forest). Correlatively, growing precipitation could potentially boost regional water yield, and vegetation rehabilitation was likely to strengthen soil conservation capacity. Meanwhile, elevated surface runoff linked to increased WY may facilitate the transport of phosphorus accumulated from farmland production, which could partly explain the upward trend of WP_P. The downward trend of WP_N overlapped chronologically with local agricultural non-point source pollution control policies, which may have contributed to nitrogen retention reduction. By comparison, CS and HQ declined continuously across the study timeframe. Spatially, these losses were concentrated in zones undergoing urban expansion, industrial development and mineral exploitation, implying that habitat occupation and landscape fragmentation are plausible contributing drivers.

FIGURE 3

3.2 Spatiotemporal distribution of MESLI

The MESLI values for 2005-2020 indicated that the ecosystems in the SYRB could provide multiple ESs simultaneously (Figure 4). MESLI showed a consistent spatial pattern, with high values concentrated in the mountainous areas on the eastern and western sides of basin, and low values distributed in the central basin belt and intensively urbanized regions. Notably, high MESLI values were more highly aggregated in the southeastern part of the study area, whereas areas with concentrated urban construction land consistently had low values, which was consistent with the findings of previous related studies (Wang B. et al., 2024; Wang et al., 2024a). From 2005 to 2020, MESLI showed an overall upward trend, with the average value increasing from 2.057 in 2005 to 2.168 in 2010, followed by a slight decrease to 2.163 in 2015, and a further increase to 2.247 in 2020. In terms of spatial changes, the north-central region of the SYRB showed the most evident improvement over the study period, while the southeastern region exhibited a slight decrease. This spatiotemporal trend was temporally consistent with the implementation of large-scale ecological restoration projects across the region, though the extent to which these projects directly caused the observed MESLI changes requires further attribution analysis. The MESLI serves as a synthetic indicator of overall ecosystem multifunctionality and allowed us to validate our ESB classification.

FIGURE 4

3.3 Factors influencing ecosystem services

Single-factor detection results from the GD revealed the independent explanatory power of each driver for the spatial patterns of ESs across four time points (Figure 5). WY was primarily governed by land use type (multi-year mean q = 0.535) and annual precipitation (multi-year mean q = 0.437). The q-statistic of land use type for WY spatial heterogeneity rose continuously from 2005 to 2020, whereas the explanatory power of annual precipitation declined gradually. Moderate explanatory effects on WY were also detected for DEM, NDVI and mean annual temperature, with their individual q-values increasing year by year.

FIGURE 5

CS was dominated by land use type, whose explanatory power consistently exceeded 0.980 across the entire study period. NDVI (mean q = 0.364), DEM (mean q = 0.246) and annual temperature (mean q = 0.215) also exerted substantial influences on the spatial distribution of CS, and the explanatory magnitude of each factor remained relatively stable over time.

SC were driven by NDVI (mean q = 0.172), DEM (mean q = 0.234), and annual average temperature (mean q = 0.194).

HQ spatial heterogeneity was jointly controlled by DEM, annual mean temperature, NDVI and land use type. The multi-year average q values ranked as DEM (0.113) > annual temperature (0.116) > NDVI (0.082) > land use type (0.075). GDP and population density exerted negligible impacts on HQ, with q-values consistently below 0.03 across all years.

Land use type acted as the leading driver for both WP_N and WP_P. The q-value of land use type for WP_N stabilized at roughly 0.785 during 2005–2020, while its explanatory power for WP_P averaged approximately 0.713. NDVI, DEM and mean annual temperature constituted secondary influencing factors for WP_P.

Across all four ES metrics, GDP and population density exhibited consistently low q-statistic values. This indicated that natural drivers (climate, terrain, vegetation) explained more spatial variation in ESs than the socioeconomic variables selected in this framework within the study watershed. Nevertheless, this result could not be interpreted as evidence that socioeconomic factors exerted weaker causal impacts on ESs. Socioeconomic forces may shape ESs indirectly via mediating variables included in our model, or their full effects may only be detectable at spatial scales finer or coarser than the two scales adopted in this study.

3.4 Trade-offs and synergies of multi-scale ecosystem services

In this study, a correlation analysis of the six indicators (WY, CS, SC, HQ, WP_N, and WP_P) was performed at the 1-km and county scales across 2005, 2010, 2015, and 2020. At the 1-km grid scale, most ESs showed a synergistic relationship, with the HQ-CS pair showing the highest synergy (Figure 6). This result confirmed the internal consistency of these two services as structural ESs: dense vegetation can simultaneously improve carbon sequestration capacity and biodiversity maintenance function by forming complex habitat structures, resulting in a typical synergistic gain effect (Yang J. et al., 2025). In contrast, WP_P exhibited consistent trade-off relationships with most other ESs, with the most pronounced trade-off observed between WP_P and CS. This finding indicated that at the 1-km grid scale, areas with high CS were often accompanied by a higher risk of phosphorus output, reflecting the resource competition between vegetation cover and hydrological processes (Li et al., 2025a). Specifically, although WP_P showed weak trade-off relationships with SC and HQ, the result indicated the presence of a certain degree of compatibility between hydrological regulation functions and terrestrial ecosystem services. Notably, the correlation between WP_N and SC exhibited distinct interannual volatility during 2005–2020, showing significant correlation in 2005, non-significant correlation in 2010, extremely significant correlation in 2015, and significant correlation again in 2020. Different from the consistently stable and highly significant relationships of the other 14 ES pairs, this unique temporal fluctuation was steadily reproduced through repeated data screening, outlier inspection, and recalculation, excluding potential methodological artifacts, outlier interference, or computational errors. All datasets adopted in this study were official standardized open-source products with unified quality control. Due to the lack of publicly available high-precision ground-truth data, further quantitative verification and calibration of individual dataset accuracy is currently unavailable. Accordingly, this study conservatively interpreted the WP_N–SC fluctuating relationship as the inherent interannual variability of regional ecosystem service interactions. From the perspective of long-term evolutionary trends, most trade-off relationships showed an optimizing trend from 2005 to 2020, with the intensity of trade-offs weakening or shifting to synergy relationships. However, the synergy between WP_P and other ESs generally weakened, and the corresponding trade-off relationships continued to strengthen, suggesting that the water purification function faced increasingly prominent trade-off pressure in regional ecosystem management.

FIGURE 6

Compared with those at the 1-km scale, most paired ESs exhibited more synergistic relationships at the administrative county scale (Figure 7). The exceptions were the WP_N–SC and WP_P–CS pairs, which exhibited consistent trade-off relationships; notably the WP_P–SC pair shifted from synergistic relationships to a trade-off over the study period. The paired relationships with the strongest synergy and strongest trade-off at the county scale were consistent with those at the 1-km scale. Two core paired relationships remained stable and unchanged in their direction (synergistic or trade-off) throughout 2005–2020. Five pairs of synergistic relationships were strengthened or weakened by trade-offs, whereas eight pairs of synergistic relationships were weakened or strengthened by trade-offs. In terms of correlation changes, the relationships of SC–WY, HQ–CS, WP_N–CS, WP_N–HQ, and WP_N–WP_P showed highly significant correlations. WP_P–CS was significantly correlated, whereas WP_N–WY and WP_P–WY weakened over time until they had no correlation.

FIGURE 7

To statistically compare the correlation coefficients between the two scales, we applied Fisher’s r-to-Z transformation across the four study years. The results revealed that most ES pairs (14 out of 15) did not show statistically significant differences in their correlation coefficients between the 1-km and county scales (|Z| < 1.96, p > 0.05), indicating that the overall direction of ES relationships was generally consistent across scales for the majority of service pairs. However, several notable exceptions were identified. The CS–SC pair showed significantly stronger synergy at the county scale than at the 1 -km scale in 2010 (Z = 2.31, p < 0.05), 2015 (Z = 2.40, p < 0.01), and 2020 (Z = 2.03, p < 0.05), although no significant scale difference was detected in 2005 (Z = 1.94, p > 0.05). Additionally, the HQ–WP_N pair exhibited a significantly stronger synergy at the county scale compared to the 1 -km scale in 2020 (Z = 2.14, p < 0.05). These findings confirmed that scale effects were particularly evident for pairs involving CS, SC, and WP services, especially in later study years, while other ES relationships showed greater cross-scale consistency.

The GWR analysis results showed significant spatial heterogeneity in the trade-offs and synergies among ESs across the SYRB. At the 1-km scale, strong trade-offs were mainly reflected in the local spatial relationships of four ES pair: WY–CS, CS–HQ, CS–WP_P, and SC–WP_P. These strong trade-off zones were concentrated in the mountainous forest land areas on the eastern and western sides of the study area (Figure 8). This spatial pattern can be explained by land use types and topographic features of these regions. Forest land is typically a hotspot for the supply of multiple ESs, it is often difficult for different services to reach their optimal state simultaneously within the same spatial unit, especially in mountainous areas with complex terrain.

FIGURE 8

Regions with strong synergistic relationship for the WY–CS and WY–HQ pairs were concentrated in the central basin belt basin of the study area. Regions with strong spatial synergies for the CS–SC, CS–HQ, WY–WP_N, CS–WP_N, SC–WP_N, and HQ–WP_N pairs were scattered across the study area. With the expansion of the spatial scale from 1-km grid to the county level, the trade-off or synergistic relationships between the same ES pairs became more obvious, and most service relationships exhibited the strongest synergistic effects at the county scale. The spatial distribution of ES trade-offs and synergies at the 1-km grid scale showed high temporal stability from 2005 to 2020. In terms of the overall spatial distribution of ES trade-offs and synergies, the proportion of grid units with synergistic relationships was generally higher than that with the trade-off relationships across the study area (Figure 9). Among the 15 paired ES combinations, only the CS–WP_P pair had a higher areal proportion of trade-offs than synergies. For all remaining 14 pairs, the areal proportion of synergies exceeded that of trade-offs; of these, the WY–HQ, WY–WP_N, WY–WP_P, CS–SC, CS–HQ, CS–WP_N, SC–WP_P, HQ–WP_N, HQ–WP_P, and WP_N–WP_P pairs all showed a substantially higher areal proportion of spatial synergies than trade-offs. From 2005 to 2020, the spatial proportion of synergies for 13 of the 15 ESs pairs increased gradually, while only the proportions for the WY–WP_P and HQ–WP_P pairs decreased over time.

FIGURE 9

At the administrative county scale, all 15 paired ESs relationships showed more extensive spatial synergies compared with the 1-km grid scale, and spatial trade-off zones were more dispersedly distributed (Figure 10). This indicated that the county scale held greater synergistic potential for the coordinated management of multiple ESs. However, during 2005–2020, although the overall areal proportion of trade-off areas remained relatively small, it exhibited a continuous expanding trend over time (Figure 11). This suggested that regional ESs face growing developmental conflicts, which highlights the necessity for targeted optimization pathways to improve the comprehensive supply capacity and long-term stability of regional ecosystems. Notably, the SC–HQ pair showed a synergistic relationship across the study area between 2005 and 2010, but shifted to a dominant trade-off relationship across the study area during 2015–2020, reflecting a growing imbalance and prominent functional contradiction between SC and HQ maintenance.

FIGURE 10

FIGURE 11

The results of the correlation coefficients between the two scales revealed that the synergy between HQ and CS was significantly stronger at the county scale than at the 1-km scale (Z = 3.56, p < 0.01). Similarly, the trade-off between CS and WP_P was significantly weaker at the county scale (Z = −2.89, p < 0.01), consistent with the observed directional reversal for this pair. For most other ES pairs, although the correlation coefficients differed in magnitude between scales, the differences were not statistically significant (p > 0.05). These results confirmed that scale effects are particularly pronounced for pairs involving CS and WP_P, while other relationships exhibited greater cross-scale consistency in their overall direction.

3.5 Spatiotemporal distribution and characterization of ecosystem service bundles

To determine the optimal number of ESBs, we applied the elbow method (Kodinariya and Makwana, 2013). The total within-sum-of-squares (WSS) was computed for k = 2 through k = 6 at both scales. The elbow point where the marginal decrease in WSS began to flatten was clearly identified at k = 4 for both scales (Figures 12a, 13a), indicating that four clusters optimally balance model fit and parsimony. Accordingly, four ESBs were identified at both scales across the SYRB. At the 1-km grid scale, the four ESBs were classified as the CS–HQ synergistic bundle, low synergistic bundle, key synergistic bundle, and integrated ecological bundle (Figure 12). From 2005 to 2020, the overall spatial layout of the ESBs showed minimal change, with no significant transitions between the different ESB types. The CS–HQ synergistic bundle was dominated by the core ecological functions of HQ and CS, and was mainly distributed in areas with extensive concentrated cultivated land and grassland. This bundle accounted for the largest areal proportion of in the study area, with its total area decreasing from 73235 km² in 2005–68635 km² in 2020 (Figure 14). The low synergy bundle was characterized by consistently low supply levels across all ESs, was mainly distributed in areas with concentrated construction land and unused land, and showed an increasing trend in areal extent from 2005 to 2020. Notably, significant differences in the mean MESLI values were detected across the four identified ESB types (ANOVA, F = 37.61, p < 0.001). At the 1 km grid scale, the integrated ecosystem service bundle exhibited the highest average MESLI (3.95 ± 0.28), followed by the key synergy bundle (2.57 ± 0.27), the CS-HQ synergy bundle (2.08 ± 0.21), and the low-synergy bundle (1.43 ± 0.17). This gradient demonstrated that the ESB classification established in this study effectively captures overall disparities in ecosystem service supply capacity. Regions categorized as low-synergy bundles required targeted ecological management interventions.

FIGURE 12

FIGURE 13

FIGURE 14

The key synergistic bundles were dominated by WY, HQ, and CS, and was mainly distributed in forest land areas. These bundles effectively delivered core ecological regulatory functions, but their SC supply capacity was insufficient. The total area of this bundle decreased from 33046 km² in 2005–32611 km² in 2020 (Figure 14). The integrated ecological bundles accounted for the smallest areal proportion across the study area, and was mainly distributed in core forestland regions with high HQ and relatively complete ES supply capacity, holding high ecological functional potential. The total area of this bundle decreased slightly from 818 km² in 2005 to 765 km² in 2020 over the study period. The continuous expansion of the low synergistic bundle and the shrinkage of the integrated ecological bundle indicated that the overall supply capacity and stability of ESs across the study area were facing severe challenges.

At the administrative county scale, three of the four ESB categories, CS–HQ synergistic bundles, integrated ecological bundles, and key synergistic bundles, were functionally consistent with the classification defined at the 1-km grid scale, whereas a fourth, scale-specific WP–WY–SC synergistic bundles emerged, reflecting significant scale effects in ES clustering patterns (Figure 13). This functional consistency validated the robustness of our classification, while the county-specific WP-WY-SC bundle captured a genuine scale-averaging effect. This indicated that regional ecosystems exhibited different ES aggregation characteristics at different spatial scales. The WP–WY–SC synergistic bundle was dominated by WP, WY, and SC functions, and showed no significant spatial transitions with other ESBs types from 2005 to 2020. Its total areal extent remained stable at 1610.7 km², accounting for only 1.38% of the total study area (Figure 15). Consistent with the 1 km scale results, the CS–HQ synergistic bundles accounted for the largest proportion of the study area at the county scale. Temporally, its total area increased from 69002.9 km² in 2005–82611.5 km² in 2015, and then dropped back to 69606.7 km² in 2020. The area diagram showed that from 2005 to 2020, a substantial number of county units shifted from the CS–HQ synergy bundle (dominated by two core ES functions) to the key synergy bundle and integrated ecological bundle (characterized by multi-ES dominance). This trend indicated that the study area was shifting toward diversified, multi-functional ES supply at the county scale.

FIGURE 15

4 Discussion

4.1 Impacts of drivers on ecosystem services

WY reflects ecosystem water retention capacity and is generally considered primarily controlled by precipitation in semi-arid regions (Li et al., 2021). However, our GD results revealed that land use type, rather than precipitation, dominated the spatial variation of WY in the SYRB, presenting a counterintuitive finding. This paradox can be reasonably explained as follows: first, the InVEST WY module calculated actual evapotranspiration based on land-use biophysical parameters. Land-use differences directly altered evapotranspiration and water loss, thereby strongly regulated spatial WY patterns and even weakened precipitation-driven gradients. Second, continuous precipitation data must be discretized in GD analysis, which may underestimate its explanatory power relative to the inherently categorical land-use variable. Third, the fragmented loess terrain of the SYRB led to highly heterogeneous rainfall interception and infiltration, which were better captured by land-use patterns than by precipitation alone (Yang et al., 2019). Spatially, the southeastern forested areas showed low evapotranspiration and high WY, whereas other land-use types exhibited greater water loss. Therefore, land-use-mediated evapotranspiration variation overrided precipitation effects and became the dominant driver of WY spatial distribution in the study area.

The high-value areas of CS were mainly distributed in forest lands with dense vegetation cover, which exhibited favorable vegetation conditions and strong carbon sequestration capacity, especially the accumulation of soil organic carbon (Fu et al., 2024). By contrast, low-value CS areas were concentrated in cultivation and construction lands, where vegetation has been reduced and carbon density is relatively low owing to urbanization expansion (Zhai et al., 2021). Land use type and NDVI played important roles in the distribution of CS. For SC, previous work in the middle and upper reaches of the YRB identified slope and relative humidity as primary controls (Li et al., 2025b). However, in this study, land use type, DEM, temperature, and NDVI jointly influenced the SC capacity of the SYRB. Owing to the different natural conditions, the influencing factors of the different sub-basins in the YRB were different. The HQ was generally low across most of the study area, and a few areas with high values were distributed in the core areas of forest land on the east and west sides. Consistent with our GD results, DEM, annual temperature, NDVI, and land use types were endogenous drivers that determined the spatial distribution of HQ in the SYRB (Yan et al., 2018). The main influencing factor for WP was also land use type.

Moreover, the GD method is intrinsically more sensitive to categorical predictors than to continuous ones that require discretization—a methodological feature that may inflate the q-statistic for land use relative to climate gradients. On the other hand, this statistical dominance is not merely a model artefact; it also captures genuine ecological differences among land use categories. In the SYRB, functional divergence between forests, grasslands, croplands, and construction lands was substantially greater than intra-category variability driven by climatic factors alone. More importantly, rather than treating land use as a purely independent causal agent, we interpreted its high q-statistic as an integrative proxy that encapsulated the cumulative effects of multiple intertwined drivers. It reflected topographic constraints (through land suitability), vegetation conditions (via NDVI), and anthropogenic interventions (including ecological restoration policies and agricultural practices). Thus, the finding does not constitute a circular tautology—provided we recognize land use as a synthetic indicator of coupled human–natural system dynamics, rather than as an isolated biophysical variable.

Admittedly, multiple InVEST modules simulated ESs based on land use and cover data as core inputs, which may produce inherent statistical correlation between land use types and modelled ES outputs and introduce analytical artifacts. However, cross-regional comparative Geographical Detector evidence demonstrated that land use does not invariably serve as the dominant driving factor for ES spatial patterns. Li S. et al. (2025) and Xu Y. et al. (2025) confirmed that topography and climate showed stronger explanatory power in other watersheds even when land use was included as a candidate factor. The extremely high q-statistic of land use within the Shanxi Yellow River reach arose jointly from two aspects: the inherent input structure of the InVEST model, and the dramatic landscape heterogeneity shaped by ecological restoration, as well as urban expansion and mining activities in this region. Distinct land-use types corresponded to vastly different vegetation, soil and disturbance levels, and generated genuine gaps in ecosystem service supply. Therefore, the dominant role of land use revealed by GD analysis reflects both model characteristics and the unique local ecological background, rather than a meaningless circular inference derived solely from model parameter settings.

The dominant role of land use as an integrative proxy is further substantiated by the region’s policy and terrain context. From 2005 to 2020, it overlapped with the consolidation stage of China’s Grain for Green project. In the SYRB, a region typified by steep slopes and highly fragmented terrain, this policy has accelerated the transformation of sloping cultivated (>25°) to forestland and grass land. Such a top-down policy intervention, combined with the region’s inherent topographic constraints, has produced a synergistic, superimposed effect on ecosystem structure and function. Research on the Loess Plateau confirms that over the past 2 decades, implementation of the Grain for Green Program has become the main driver of enhanced carbon sequestration, vegetation restoration, and reduced soil erosion (Ma Y. et al., 2025). Within this terrain-limited scope, policies provide the economic and social impetus to promote and guide land use changes. This concept aligns with the findings from global semi-arid regions, where terrain complexity regulates the relationship between land management practices and ecosystem outcomes (Wang et al., 2025).

Therefore, rather than treating land use as a singular causal agent, we recognized it as a synthetic indicator of coupled human-natural system dynamics. This nuanced interpretation, we contend, more faithfully represents the socio-ecological reality of the SYRB—a region characterized by complex loess topography and intensive policy-driven landscape transitions.

4.2 Impacts of multi-scale management on trade-offs and synergies

Spatial trade-off and synergy patterns among ESs are not a fixed intrinsic properties, but substantially reshaped with shifting analytical scale, highlighting the necessity of multi-perspective ES interaction analysis to formulate targeted ecological management schemes. In this study, both the magnitude and spatial distribution of ES trade-offs and synergies showed distinct disparities between the 1-km and county scales. For instance, CS–HQ showed stronger synergies at the county scale, whereas HQ–WP_P displayed weak trade-offs at the 1-km scale yet mild synergies at the county scale. Overall, synergistic effects of ESs intensified and the trade-off effect weakened at coarser administrative scales, consistent with Liu et al. (2024) and Zhang et al. (2025). Temporal trends from 2005 to 2020 further revealed a gradual decline in synergy intensity at the county scale. This finding indicates that temporal evolutionary characteristics of ES interactions should be jointly considered alongside scale-dependent laws governing ES trade-offs and synergies.

This study identified significant scale dependence in the CS–WP_P linkage. At the 1-km grid scale, the pair exhibited a trade-off relationship, while at the county scale, the relationship reversed to synergy. This phenomenon is closely linked to the modifiable areal unit problem, a well-documented source of uncertainty in spatial statistical analysis. As Zen et al. (2019) reported for the Eastern Alps, aggregating high-resolution ES datasets to administrative jurisdictions inevitably erodes fine spatial heterogeneity, restructuring spatial clustering patterns. At the fine 1-km grid scale, raw data retain detailed topographic and land-use heterogeneity, capturing localized ecological resource competition. On steep slopes, high forest coverage boosts CS while accelerating phosphorus loss via surface runoff, generating a statistically significant negative correlation (trade-off) between CS and WP_P. County-level spatial averaging eliminates these micro-scale extreme differences and overturns the statistical correlation. Finer grid units amplify intra-unit ecological variability and highlight local divergences, while coarser aggregation creates an averaging effect that masks micro-terrain and land-use disparities (Zeng et al., 2024).

Such scale-driven directional reversal can also be explained by shifting dominant driving forces across spatial scales. At the fine 1-km grid scale, the WP_P–CS interactions are mainly controlled by local biophysical attributes including slope, soil type, and vegetation coverage. By contrast, county scale ES relationships are dominated by broad-scale environmental gradients, such as climatic and geomorphic zones (Liu et al., 2024). Micro-scale patterns are primarily shaped by natural biophysical conditions, while macro county-scale dynamics are increasingly modulated by socioeconomic factors. Most ES pairs feature stronger interaction magnitudes at coarser scales (Niu et al., 2022). Counties with favorable hydrothermal conditions deliver elevated overall ES supply: abundant precipitation and moderate temperature simultaneously promote vegetation carbon sequestration and runoff-related phosphorus output, forming synergistic correlations. Larger analytical extents also sustain higher biodiversity and resource utilization efficiency, further reshaping ES interaction patterns (Su et al., 2020).

Beyond biophysical drivers, the cross-scale relationship reversal may also reflect institutional administrative partitioning, corresponding to the natural-administrative scale mismatch hypothesis proposed in the Introduction. As China’s basic unit for policy enforcement, countiesimplement unified territorial spatial planning, major functional zone delineation and ecological compensation policies. Although grid-scale patches face resource competition between ecological and cultivated land functions, integrated county-level governance coordinates multiple ecological targets and yields holistic ES synergy. A single county concurrently manages water conservation zones (reducing WP_P loss to safeguard water purification) and permanent basic farmland protection areas. These two governance priorities operate in parallel rather than competing, shaping county-wide synergistic ES supply at the county scale (Zeng et al., 2024). This finding validates our hypothesis that scale mismatch impairs governance efficiency: county-level aggregated policies overlook localized trade-offs on steep upstream terrain and hinder optimal resource allocation.

Existing basin-scale literature yields inconsistent ES interaction conclusions. Yang et al. (2021) concluded that a trade-off exists between WY and CS in the YRB, while Xu T. et al. (2025) found significant trade-offs between HQ, CS, and WP_N. Shao et al. (2023) found that the relationship between ESs in the YRB was dominated by weak trade-offs and synergies at a temporal scale, with trade-offs strengthened in the Upper YRB and the Middle YRB and synergies strengthened in the Lower YRB. Such discrepancies stem from divergent analytical scales and heterogeneous regional environmental backgrounds, emphasizing the value of region-specific multi-scale assessments like this study instead of universal basin-scale generalizations.

Notable volatility was observed in the WP_N-SC linkage. Interannual agricultural activities and climate fluctuations are potential drivers, yet limited time-series data on fertilizer application and land management intensity prevent robust validation. We therefore frame these interpretations as tentative hypotheses requiring further exploration with high-frequency datasets and process-based models. To improve the identification of local features and explanatory power of ES relationships, future research should combine multi-scale data, introduce high-resolution land use information, and develop cross-scale ecosystem management strategies that balance fineness and wholeness.

It is important to distinguish between two types of stability to resolve apparent logical ambiguity. Within the constraints of specific ecological contexts and functional distributions, the relationships between these ESs are relatively stable (Zhang et al., 2022). Cross-scale comparison demonstrates that ES pair linkages lack scale invariance: switching from fine grids to county units can completely reverse the direction of trade-offs and synergies, as evidenced by the CS–WP_P pair. No universal fixed ES interaction pattern exists across all spatial scales. Recognizing this cross-scale lability constitutes the core merit of our multi-scale design and provides empirical evidence to avoid scale mismatch biases in territorial ecological policymaking.

4.3 Management strategies for multi-scale ecosystem service bundles

The CS-HQ synergistic bundle, the most widely distributed bundle at both 1-km and county scales, delivers comprehensive ESs and ranks as the core conservation priority in the study area. At the 1-km scale, this bundle concentrates in the biodiversity reserves including the Taihang Mountain and Zhongtiao Mountain, where fragmented grassland, cropland and built-up land create persistent anthropogenic disturbance. Corresponding management must coordinate dual spatial scales: county governments formulate unified territorial spatial plans, while 1-km fine zoning guides localized refined control. In line with the Three Zones and Three Lines policy, these zones should be demarcated as coordinated ecological function protection areas to curb cropland and construction land expansion—the primary land-use conversion threat identified in our spatial analysis. Local policies shall shift from single-target ecological restoration to multi-objective synergy optimization. Pilot ecological product value realization projects are proposed to convert carbon sinks, water conservation, and biodiversity maintenance into tradable ecological assets (Spake et al., 2017). Supported by our cross-scale disparity findings, a basin-wide ecological compensation mechanism should be implemented: downstream beneficiary counties transfer fiscal payments to upstream counties that restrict development to safeguard CS–HQ synergies. Moreover, afforestation shall follow near-natural restoration rather than single-species high-water-consumption plantations, which prior local studies found induce soil desiccation and reduced water yield (Zhao et al., 2025).

The integrated ecological bundle experienced notable areal shrinkage from 818 km² (2005) to 765 km² (2020) in our long-term grid-scale assessment, representing a continuous decline in high-level comprehensive ES supply. This bundle covers zones with balanced provision of carbon storage, soil conservation, water purification and habitat quality, and requires tiered cross-scale protection arrangements. At the 1-km fine scale, shrinkage hotspots shall be marked as first-level ecological redlines to limit scattered industrial and residential encroachment. At the county administrative scale, these integrated zones shall be designated key ecological protection units within county master plans, with annual ecological performance assessments tied to local government appraisal. Given their balanced multi-service capacity, integrated bundle areas shall serve as core demonstration zones for coordinated ecological governance, linking grid-scale degradation monitoring to county-level ecological restoration funding allocation to reverse the observed area contraction trend.

The Low synergistic bundle, identified at the 1-km scale, clustered in densely populated central basin urban agglomerations where urbanization drives severe ES trade-offs. Uniform restrictive development policies are inappropriate here; instead, ecological constraints must be embedded within county economic development frameworks, with fine grid zoning delivering site-specific regulatory standards. Since this bundle is only discernible at the fine 1-km scale, it underscores the need for targeted, localized interventions within broader county-level plans. Counties shall enforce total construction land quota control, linking new land approval to equivalent ecological land compensation within identical watershed units to preserve natural ecological integrity. An annual habitat quality monitoring system based on 1-km ESB maps shall be launched: if habitat quality degrades for two consecutive years within low-synergy zones, county authorities automatically tighten urban development boundaries and cut subsequent construction land quotas.

The WP–WY–SC synergistic bundle, identified at the county scale, aggregation feature covering two complete administrative units. Driven primarily by land-use change and NDVI variations per our geographic detector results, integrated watershed governance is required. First, differentiated pollutant discharge permits should be allocated: counties dominated by this bundle adopt stricter total emission caps than provincial benchmarks, with emission trading confined within single watersheds to avoid cross-basin pollution transfer. Second, following semi-arid riparian management standards validated by Yang S. et al. (2025), hierarchical riparian restoration buffers are deployed along river networks; primary tributaries adopt 100 m ecological buffer zones and secondary tributaries adopt 50 m buffers to mitigate soil loss and non-point source pollution. Third, water-saving irrigation and rainwater harvesting should be promoted to secure river ecological base flow and wetland recharge, coordinating water yield and habitat quality improvements across the watershed.

Crucially, this study’s core concern regarding scale mismatch between natural ecological units and administrative jurisdictions is addressed via a two-tier cross-scale governance framework. The 1-km grid acts as a fine ecological early-warning tool, detecting localized ES trade-offs masked by county-level spatial averaging to deliver targeted micro-restoration schemes. The county scale serves as the operational unit for policy implementation, fiscal compensation and inter-departmental coordination. We propose a clear hierarchical linkage mechanism: county governments incorporate 1-km ESB spatial outputs into territorial planning to prioritize protection of fine-scale low-synergy shrinkage hotspots and integrated bundle degradation zones. Meanwhile, county-level aggregated ESB classification underpins macro regional policies including basin ecological compensation and watershed pollutant trading. This framework integrates the high diagnostic resolution of grid data with the institutional operability of administrative units, resolving the scale mismatch dilemma and translating our multi-scale empirical findings into actionable governance tools.

Ecosystem service bundle zoning provides targeted, data-driven support for differentiated ecological management rather than generic policy recommendations. All strategies above are tailored to the scale-dependent spatial patterns, dominant human–natural threats, long-term area change trends and driving mechanisms identified in our dual-scale ESB assessment. This multi-tiered cross-scale governance model can be replicated for ecological zoning in fragmented loess hilly regions, supporting coordinated improvement of regional ecosystem functions and long-term sustainable development of the Yellow River Basin.

4.4 Limitations and prospects

This study explored multi-scale associations between ESs and their driving factors and identified dominant ES bundles to provide targeted ecological management implications. Nevertheless, several limitations of this study should be acknowledged. First, the temporal observations were sampled at 5-year intervals from 2005 to 2020 rather than using continuous time-series data, and such discrete temporal resolution may bias the detected synergies and trade-offs among ESs. Second, inherent uncertainties exist in the InVEST modelling framework. The model simplifies real-world ecological processes, and its outputs are highly sensitive to parameter configurations and input data quality. Key sensitive parameters vary across the four modules adopted in this study: the Z parameter regulating precipitation–evapotranspiration interactions dominates WY simulations; land-use-specific carbon densities bring substantial uncertainties to CS estimation; the R and K erodibility factors primarily govern SC results; and distance-decay coefficients of threat factors largely shape HQ outputs. Notably, only the water yield module was calibrated against official statistical yearbook and water resources bulletin data, while all other InVEST outputs lack field observational validation, which may introduce additional uncertainties into the overall ES assessment.

Notably, another scale-related limitation arises from the 30 m-to-1 km spatial aggregation implemented in this study. Fine-resolution 30 m model outputs were aggregated to 1 km grids using block mean statistics to balance computational efficiency and regional-scale analytical applicability. Inevitably, inevitably, this scale generalization smooths fine-scale topographic heterogeneity and micro-landscape variations across the fragmented Loess Plateau terrain, which may partially obscure small-scale ecological variability and weaken localized spatial differences in ES distributions. This scale trade-off represents a necessary compromise for multi-scale analysis yet introduces inherent analytical limitations.

In addition, spatial resolution mismatch exists between datasets: fine-resolution biophysical data were integrated with coarse administrative-level socioeconomic statistics, which may hinder accurate characterization of fine-scale spatial interactions between ecological supply and human demand. Furthermore, detailed soil properties were excluded from driving factor analysis due to data availability constraints. Nonetheless, slope, NDVI, and precipitation—the predominant controls governing soil conservation and carbon sequestration across the Loess Plateau—were fully incorporated into our analysis. Although soil attributes were not directly quantified, GD can partially capture edaphic effects through vegetation and topographic interactions. Still, unmeasured spatially heterogeneous soil properties may induce unquantified biases, which can be better addressed in future studies with high-resolution soil datasets.

Beyond data-level improvements, we highlight two key research prospects. First, while the GD effectively quantifies the global explanatory capacity of individual drivers, it fails to capture spatially nonstationary driver–ES relationships across subregions. Systematic factor interaction detection was not performed in this study, as our core objective was to identify and rank dominant independent drivers to support targeted management. Ecological theories imply considerable interactive effects between land use and precipitation for WY, as well as between topography and vegetation for SC and CS on the Loess Plateau, though quantitative assessment of such couplings remains insufficient. Emerging geographically weighted machine learning approaches, such as Geographical Random Forest, can effectively disentangle localized driving mechanisms and interactive effects at grid and subregional scales, enabling more refined ecosystem regulation (Xu et al., 2026). Second, the scale mismatch issue identified in this work highlights the need for robust cross-scale linkage frameworks that can translate fine-scale ecological insights into administrative-scale policies without losing localized spatial information. Addressing these multi-scale challenges is critical to advancing scale-adaptive and spatially precise ecosystem governance.

5 Conclusion

References

• 1

  BaskentE. Z. (2020). A framework for characterizing and regulating ecosystem services in a management planning context. Forests11, 102. 10.3390/f11010102

  • CrossRef
  • Google Scholar
  • View reference in article

• 2

  CaoF.GeY.WangJ. (2013). Optimal discretization for geographical detectors-based risk assessment. GISci Remote Sens.50, 78–92. 10.1080/15481603.2013.778562

  • CrossRef
  • Google Scholar
  • View reference in article

• 3

  CaoY.SuD.WangJ.LiG.FangX.WuQ.et al (2025). Uncovering scale effects on spatial patterns and interactions of multiple cropland ecosystem services. Environ. Dev. Sustain.27, 6781–6810. 10.1007/s10668-023-04167-z

  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  ChangY.WangY.YuanL.YaoX.ZhangR.LiuX.et al (2025). Spatial scale effects on the trade-offs and synergies of ecosystem services in China’s Huaihe river basin. Sci. Rep.15, 22301. 10.1038/s41598-025-06753-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 5

  ChenY.MengR.LiK. (2025). Exploring the spatial heterogeneity of trade-offs and synergies across ecosystem service bundles based on supply and demand relationships in the Wuhan Metropolitan area. Integr. Environ. Assess.21, 1025–1038. 10.1093/inteam/vjaf066

  • CrossRef
  • Google Scholar
  • View reference in article

• 6

  CuiY.LiR.ChenX. (2024). Analysis of nitrogen flow in the yellow river basin over a long time series. Environ. Monit. Assess.197, 28. 10.1007/s10661-024-13505-1

  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  DoncasterC. P.BullockJ. M. (2024). Living in harmony with nature is achievable only as a non-ideal vision. Environ. Sci. Policy152, 103658. 10.1016/j.envsci.2023.103658

  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  DonohueR. J.RoderickM. L.McVicarT. R.FarquharG. D. (2012). Impact of CO2 fertilization on maximum foliage cover across the globe's warm, arid environ-ments. Geophys. Res. Lett.39, L17403–3035. 10.1002/grl.50563

  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  FengD.ZhouP.WangD.ShiP. (2023). Reconstructing the ecological restoration pattern from the perspective of ecosystem health assessment in a typical black soil region of Northeast China. Front. Environ. Sci.11, 1184517. 10.3389/fenvs.2023.1184517

  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  FisherR. A. (1915). Frequency distribution of the values of the correlation coefficient in samples from an indefinitely large population. Biometrika10, 507–521. 10.2307/2331838

  • CrossRef
  • Google Scholar
  • View reference in article

• 11

  FuS.ZhenZ.ZhouH.WangB.QiaoQ. (2024). Spatio-temporal evolution and prediction of carbon storage at the source of the Fen river and Sanggan river based on a PLUS-InVEST model. Front. Environ. Sci.12, 1449576. 10.3389/fenvs.2024.1449576

  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  GaoS.YangL.JiaoH. (2022). Changes in and patterns of the tradeoffs and synergies of production-living-ecological space: a case study of Longli county, Guizhou province, China. Sustainability14, 8910. 10.3390/su14148910

  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  GaoY.FuD.HuangH.JiangJ.LiuQ.ZhuL.et al (2025). Spatio-temporal evolution and scenario-based optimization of urban ecosystem services supply and demand: a block-scale study in Xiamen, China. Ecol. Indi.172, 113289. 10.1016/j.ecolind.2025.113289

  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  JiangC.ZhangH.ZhangZ.WangD. (2019). Model-based assessment soil loss by wind and water erosion in China’s Loess Plateau: dynamic change, conservation effectiveness, and strategies for sustainable restoration. Glob. Planet. Change172, 396–413. 10.1016/j.gloplacha.2018.11.002

  • CrossRef
  • Google Scholar
  • View reference in article

• 15

  KodinariyaT. M.MakwanaP. R. (2013). Review on determining number of cluster in K-Means clustering. Int. J. Adv. Res. Comput. Sci. Manag. Stud.1, 90–95.

  • Google Scholar
  • View reference in article

• 16

  LiP.LiT. (2025). Spatiotemporal analysis of ecosystem services and the impact of new-type urbanization: a case study of Chengdu, China. Ecol. Indi.171, 113191. 10.1016/j.ecolind.2025.113191

  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  LiM.LiangD.XiaJ.SongJ.ChengD.WuJ.et al (2021). Evaluation of water conservation function of Danjiang river basin in Qinling Mountains, China based on InVEST model. J. Environ. Manage.286, 112212. 10.1016/j.jenvman.2021.112212

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  LiQ.LiD.WangJ.WangS.WangR.FuG.et al (2024). Spatial heterogeneity of ecosystem service bundles and the driving factors in the Beijing-Tianjin-Hebei region. J. Clean. Prod.479, 144006. 10.1016/j.jclepro.2024.144006

  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  LiS.LiR.WangL.FuM.TianY.ZhangJ. (2025). Study on the multi-scale characteristics of ecosystem service trade-offs, synergies and ecosystem service bundles in Beijing. Environ. Earth Sci.84, 359. 10.1007/s12665-025-12363-5

  • CrossRef
  • Google Scholar
  • View reference in article

• 20

  LiW.ChenX.ZhengJ.ZhangF.YanY.HaiW.et al (2025). Effectiveness and driving mechanisms of ecological conservation and restoration in Sichuan Province, China. Ecol. Indi.172, 113238. 10.1016/j.ecolind.2025.113238

  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  LiX.LiX.LiG.YaoQ.WangS.DongZ.et al (2025a). Multi-scale divergences of ecosystem services: bridging grid-level biophysical and town socio-economic drivers for hilly ecological management. J. Clean. Prod.534, 147070. 10.1016/j.jclepro.2025.147070

  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  LiX.YuK.XuG.LiP.LiZ.ShiP.et al (2025b). Exploring ecosystem service dynamics and drivers in the upper and middle yellow river basin under large-scale ecological restoration. Ecol. Eng.217, 107643. 10.1016/j.ecoleng.2025.107643

  • CrossRef
  • Google Scholar
  • View reference in article

• 23

  LindborgR.GordonL. J.MalingaR.BengtssonJ.PetersonG.BommarcoR.et al (2017). How spatial scale shapes the generation and management of multiple ecosystem services. Ecosphere8, e01741. 10.1002/ecs2.1741

  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  LiuY.JingY.HanS. (2023). Ecological function zoning of Nansi Lake basin in China based on ecosystem service bundles. Environ. Sci. Pollut. Res.30, 77343–77357. 10.1007/s11356-023-27723-6

  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  LiuQ.QiaoJ.LiM.HuangM. (2024). Spatiotemporal heterogeneity of ecosystem service interactions and their drivers at different spatial scales in the yellow river basin. Sci. Total Environ.908, 168486. 10.1016/j.scitotenv.2023.168486

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  MaJ.HaoZ.ShenY.ZhenZ. (2025). Spatial-temporal evolution of carbon storage and its driving factors in the Shanxi section of the yellow river basin, China. Ecol. Modell.502, 111039. 10.1016/j.ecolmodel.2025.111039

  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  MaY.LiJ.CaoW.HuangL. (2025). Grain for green program to grassland might lead to carbon sink leakage in the Loess Plateau. Earth's Future.13, e2024EF005261. 10.1029/2024EF005261

  • CrossRef
  • Google Scholar
  • View reference in article

• 28

  Martin-LopezB.Felipe-LuciaM. R.BennettE. M.NorstromA.PetersonG.PlieningerT.et al (2019). A novel telecoupling framework to assess social relations across spatial scales for ecosystem services research. J. Environ. Manage.241, 251–263. 10.1016/j.jenvman.2019.04.029

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  NewtonA. C.CowardF.ElliottS.JenkinsE.LindenM. V.RirisP.et al (2024). Understanding long-term human ecodynamics through the lens of ecosystem collapse. Holocene34, 1439–1453. 10.1177/09596836241259785

  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  NiuH.LiuM.XiaoD.ZhaoX.AnR.FanL. (2022). Spatio-temporal characteristics of trade-offs and synergies in ecosystem services at watershed and landscape scales: a case analysis of the yellow river basin (Henan section). Int. J. Environ. Res. Public Health19, 15772. 10.3390/ijerph192315772

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 31

  QuS.ShiZ.LiangX.WangG.HanJ. (2021). Multiple factors control groundwater chemistry and quality of multi-layer groundwater system in Northwest China coalfield-using self-organizing maps (SOM). J. Geochem. Explor.227, 106795. 10.1016/j.gexplo.2021.106795

  • CrossRef
  • Google Scholar
  • View reference in article

• 32

  Rodríguez-LoinazG.AldayJ. G.OnaindiaM. (2015). Multiple ecosystem services landscape index: a tool for multifunctional landscapes conservation. J. Environ. Manage.147, 152–163. 10.1016/j.jenvman.2014.09.001

  • Pubmed Abstract
  • CrossRef
  • Google Scholar
  • View reference in article

• 33

  ShaoY.LiuY.LiY.YuanX. (2023). Regional ecosystem services relationships and their potential driving factors in the yellow river basin, China. J. Geogr. Sci.33, 863–884. 10.1007/s11442-023-2110-1

  • CrossRef
  • Google Scholar
  • View reference in article

• 34

  ShenJ.LiS.WangH.WuS.LiangZ.ZhangY.et al (2023). Understanding the spatial relationships and drivers of ecosystem service supply-demand mismatches towards spatially-targeted management of social-ecological system. J. Clean. Prod.406, 136882. 10.1016/j.jclepro.2023.136882

  • CrossRef
  • Google Scholar
  • View reference in article

• 35

  SongG.WeiT.ZhuQ.BiH.QiuJ.ZhangJ. (2025). Spatiotemporal dynamics and trade-off analysis of ecosystem services in the Caijiachuan watershed of the Loess Plateau. Agronomy15, 1707. 10.3390/agronomy15071707

  • CrossRef
  • Google Scholar
  • View reference in article

• 36

  SpakeR.LasseurR.CrouzatE.BullockJ.LavorelS.ParksK.et al (2017). Unpacking ecosystem service bundles: towards predictive mapping of synergies and trade-offs between ecosystem services. Glob. Environ. Chang.47, 37–50. 10.1016/j.gloenvcha.2017.08.004

  • CrossRef
  • Google Scholar
  • View reference in article

• 37

  SuC.DongM.FuB.LiuG. (2020). Scale effects of sediment retention, water yield, and net primary production: a case-study of the Chinese Loess Plateau. Land Degrad. Dev.31, 1408–1421. 10.1002/ldr.3536

  • CrossRef
  • Google Scholar
  • View reference in article

• 38

  TangB.BaiX.ZhangP.JiaX.YaoY. (2024). Soil water retention and release in a small catchment affected by erosion and deposition on China's Loess Plateau. CATENA249, 108683. 10.1016/j.catena.2024.108683

  • CrossRef
  • Google Scholar
  • View reference in article

• 39

  TaoQ.GaoG.XiH.WangF.ChengX.OuW.et al (2022). An integrated evaluation framework for multiscale ecological protection and restoration based on multi-scenario trade-offs of ecosystem services: case study of Nanjing city, China. Ecol. Indi.140, 108962. 10.1016/j.ecolind.2022.108962

  • CrossRef
  • Google Scholar
  • View reference in article

• 40

  TianF.LiX.WuX.GongL. (2025). Identifying supply–demand mismatches of ecosystem services and social-ecological drivers at different scales to support land use planning. Ecol. Indi.174, 113462. 10.1016/j.ecolind.2025.113462

  • CrossRef
  • Google Scholar
  • View reference in article

• 41

  TuG.LuQ.ZhangF.XiaY.YanB. (2025). The spatio-temporal interactions between rapid urbanization and multiple ecosystem services at the county scale in the Poyang Lake basin. Geomatics Nat. Hazards Risk16, 2480252. 10.1080/19475705.2025.2480252

  • CrossRef
  • Google Scholar
  • View reference in article

• 42

  WangJ.HuY. (2012). Environmental health risk detection with GeogDetector. Environ. Modell. Softw.33, 114–115. 10.1016/j.envsoft.2012.01.015

  • CrossRef
  • Google Scholar

• 43

  WangJ.LiY.WangS.LiQ.LiL.LiuX. (2023). Assessment of multiple ecosystem services and ecological security pattern in Shanxi province, China. Int. J. env. Res. Pub. He.20, 4819. 10.3390/ijerph20064819

  • CrossRef
  • Google Scholar

• 44

  WangB.FuS.HaoZ.ZhenZ. (2024). Ecological security pattern based on remote sensing ecological index and circuit theory in the Shanxi section of the yellow basin. Ecol. Indi.166, 112382. 10.1016/j.ecolind.2024.112382

  • CrossRef
  • Google Scholar

• 45

  WangJ.LvZ.CaoY.WangS.ZhenZ. (2024a). Spatial-temporal evolution and influencing factors of ecological resilience in urban agglomerations: a case study of Shanxi section of the yellow river basin. Front. Environ. Sci.12, 1385604. 10.3389/fenvs.2024.1385604

  • CrossRef
  • Google Scholar

• 46

  WangJ.LvZ.ZhenZ.WuQ. (2024b). Evolution of the ecological security pattern of the yellow river basin based on ecosystem services: a case study of the Shanxi section, China. Front. Environ. Sci.12, 1477843. 10.3389/fenvs.2024.1477843

  • CrossRef
  • Google Scholar

• 47

  WangX.HeH.ZhangM.DengJ.RenX.LvY.et al (2025). Grain for green project dominates greening in afforested areas rather than that in grass revegetation areas of the Loess Plateau, china-using deep crossing LSTM age network. Envir. Res. Lett.20, 084068. 10.1088/1748-9326/adec02

  • CrossRef
  • Google Scholar

• 48

  WangS.LiuY.OuyangZ. (2026). Mapping the “Supply–Demand–Flow” of ecosystem services for ecosystem management in China. Adv. Sustain. Syst.10 (7), 2522070. 10.1002/advs.202522070

  • CrossRef
  • Google Scholar

• 49

  WilliamsS. T.MareeN.TaylorP.BelmainS. R.KeithM.SwanepoelL. H. (2018). Predation by small mammalian carnivores in rural agro-ecosystems: an undervalued ecosystem service?Ecosyst. Serv.30, 362–371. 10.1016/j.ecoser.2017.12.006

  • CrossRef
  • Google Scholar

• 50

  WuQ.CaoY.SuD.CaoY. (2025). A multi-scale framework for understanding spatial scale effects on ecosystem service heterogeneity, interactions, drivers and their socio-ecological impact pathways for adaptive management. J. Clean. Prod.516, 145757. 10.1016/j.jclepro.2025.145757

  • CrossRef
  • Google Scholar

• 51

  WuS.LiR.YanY.DuY.WuZ.MaK. (2025). Scale-dependent drivers of ecosystem service supply-flow-demand couplings in the Shanxi yellow river basin, China. ISci.28, 114002. 10.1016/j.isci.2025.114002

  • CrossRef
  • Google Scholar

• 52

  XiaH.YuanS.PrishchepovA. V. (2023). Spatial-temporal heterogeneity of ecosystem service interactions and their social-ecological drivers: implications for spatial planning and management. Resour. Conserv. Recycl.189, 106767. 10.1016/j.resconrec.2022.106767

  • CrossRef
  • Google Scholar

• 53

  XiaoS.XiaH.ZhaiJ.JinD.GaoH. (2024). Trade-Off and synergy relationships and driving factor analysis of ecosystem services in the Hexi region. Remote Sens.16 (17), 3147. 10.3390/rs16173147

  • CrossRef
  • Google Scholar

• 54

  XuL.HeY.ZhangL.BaoG.XuH. (2024). Spatial variation in ecosystem service relationships in alpine ecosystems: a case study of the Daxing'anling forest area, Inner Mongolia. Ecol. Indi.166, 112351. 10.1016/j.ecolind.2024.112351

  • CrossRef
  • Google Scholar

• 55

  XuT.TianG.LinT.LiW.LiuC. (2025). Multiscale analysis of ecosystem service interactions and driving factors in the Loess Plateau: implications for ecological management. J. Clean. Prod.495, 145074. 10.1016/j.jclepro.2025.145074

  • CrossRef
  • Google Scholar

• 56

  XuY.TangQ.ZhangH.DuanD.ChenP.LuoY.et al (2025). Trade-offs and synergies of ecosystem services in the hilly purple soil area of southwestern China: implications for sustainable land management. J. Mt. Sci.22, 2145–2158. 10.1007/s11629-024-9399-9

  • CrossRef
  • Google Scholar

• 57

  XuY.WangM.TangW.LiS.AiW.XuW.et al (2026). Territorial spatial carbon balance governance from a spatially refined perspective: incorporating sub-regional differences in spatial driving effects into zoning regulation. Environ. Impact Assess.117, 108222. 10.1016/j.eiar.2025.108222

  • CrossRef
  • Google Scholar

• 58

  YanS.WangX.CaiY.LiC.YanR.CuiG.et al (2018). An integrated investigation of spatiotemporal habitat quality dynamics and driving forces in the upper basin of Miyun Reservoir, North China. Sustainability10, 4625. 10.3390/su10124625

  • CrossRef
  • Google Scholar

• 59

  YangD.LiuW.TangL.ChenL.LiX.XuX. (2019). Estimation of water provision service for monsoon catchments of South China: applicability of the InVEST model. Landsc. Urban Plan.182, 133–143. 10.1016/j.landurbplan.2018.10.011

  • CrossRef
  • Google Scholar

• 60

  YangJ.XieB.TaoW.ZhangD. (2021). Ecosystem services assessment, trade-off, and bundles in the yellow river basin, China. Diversity13, 308. 10.3390/d13070308

  • CrossRef
  • Google Scholar

• 61

  YangZ.ZhanJ.WangC.Twumasi-AnkrahM. (2022). Coupling coordination analysis and spatiotemporal heterogeneity between sustainable development and ecosystem services in Shanxi province, China. Sci. Total Environ.836, 155625. 10.1016/j.scitotenv.2022.155625

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 62

  YangB.XueD.MiaoP. (2024). Spatiotemporal evolution and driving factors of ecosystem supply and demand bundles: a case study in the Sichuan-Yunnan ecological buffer area, China. Sustainability16, 4977. 10.3390/su16124977

  • CrossRef
  • Google Scholar

• 63

  YangJ.JiaL.HaoJ.LuoQ.ChiW.WangY.et al (2024). Temporal and spatial variation characteristics of the ecosystem in the Inner Mongolia section of the yellow river basin. Atmosphere15 (7), 827. 10.3390/atmos15070827

  • CrossRef
  • Google Scholar

• 64

  YangJ.ChenZ.ZhangW.MengG.CaoM.LiJ.et al (2025). Relationships among multiple ecosystem services in mountainous regions: a case study of the Gaoligong Mountains. J. Environ. Manage.387, 125765. 10.1016/j.jenvman.2025.125765

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 65

  YangS.LiP.ShiP.LiuX.MengY.CaoL.et al (2025). Spatial and temporal distribution of soil nitrogen and phosphorus and ecological risk assessment in Yanhe river riparian zone on the Loess Plateau of China. Environ. Earth Sci.84, 194. 10.1007/s12665-024-12020-3

  • CrossRef
  • Google Scholar

• 66

  YueH.WangZ.LiuY. (2025). Comprehensive assessment of the ecosystem in yellow river basin based on pattern-quality-service model. Environ. Model. Assess.30, 337–348. 10.1007/s10666-024-10004-y

  • CrossRef
  • Google Scholar

• 67

  ZenM.CandiagoS.SchirpkeU.ViglL.GiupponiC. (2019). Upscaling ecosystem service maps to administrative levels: beyond scale mismatches. Sci. Total Environ.660, 1565–1575. 10.1016/j.scitotenv.2019.01.087

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 68

  ZengS.JiangC.BaiY.WangH.LiuE.GuoL.et al (2024). Understanding scale effects and differentiation mechanisms of ecosystem services tradeoffs and synergies relationship: a case study of the Lishui river basin, China. Ecol. Indic.167, 112648. 10.1016/j.ecolind.2024.112648

  • CrossRef
  • Google Scholar

• 69

  ZhaiT.ZhangD.ZhaoC. (2021). How to optimize ecological compensation to alleviate environmental injustice in different cities in the yellow river basin? A case of integrating ecosystem service supply, demand and flow. Sustain. Cities Soc.75, 103341. 10.1016/j.scs.2021.103341

  • CrossRef
  • Google Scholar

• 70

  ZhangX.YangY.ZhaoM.HanR.YangS.WangX.et al (2022). Trade-off analyses of multiple ecosystem services and their drivers in the Shandong yellow river basin. Int. J. Environ. Res. Public Health19 (23), 15681. 10.3390/ijerph192315681

  • CrossRef
  • Google Scholar

• 71

  ZhangX.HanR.YangS.YangY.TangX.QuW. (2024). Identification of bundles and driving factors of ecosystem services at multiple scales in the eastern China region. Ecol. Indic.158, 111378. 10.1016/j.ecolind.2023.111378

  • CrossRef
  • Google Scholar

• 72

  ZhangL.ChenG.ZhaoJ.LinY.YangH.HeJ. (2025). Spatiotemporal characteristics and scale effects of ecosystem service bundles in the Xijiang river basin: implications for territorial spatial planning and sustainable land management. Sustainability17, 1967. 10.3390/su17051967

  • CrossRef
  • Google Scholar

• 73

  ZhaoY.LuoP.ZhangJ.YanP.RijalM.YangZ.et al (2025). Coordinating socio-economic development and ecological protection in the yellow river basin: pathways to new quality productivity. Environ. Sustain. Indic.28, 100987. 10.1016/j.indic.2025.100987

  • CrossRef
  • Google Scholar

• 74

  ZhuM.HeW.ZhangQ.XiongY.TanS.HeH. (2019). Spatial and temporal characteristics of soil conservation service in the area of the upper and middle of the yellow river, China. Heliyon5, e02985. 10.1016/j.heliyon.2019.e02985

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 75

  ZuoQ.DingX.CuiG.ZhangW. (2024). Yellow river basin management under pressure: present state, restoration and protection II: lessons from a special issue. Water16, 999. 10.3390/w16070999

  • CrossRef
  • Google Scholar