Scenario-driven optimization of land-use/land-cover conversion pathways to improve ecosystem service performance in the Yellow River Basin

Abstract

Future land use/land cover (LULC) conversions provide a critical pathway for improving ecosystem service (ES) outcomes in large river basins, yet their contributions to comprehensive ES performance and functional synergies remain insufficiently quantified. This study projected the future dynamics of five representative ES metrics: carbon storage (CS), water yield (WY), nutrient export (NE), soil conservation (SC), and habitat quality (HQ) across multiple spatiotemporal scales in the Yellow River Basin, China. An Integrated Ecosystem Services Index (IESI) was constructed by aggregating the five normalized ES metrics to assess overall ES performance, identify priority areas for optimizing future LULC projections, compare alternative LULC conversion pathways, and evaluate socioeconomic associations with IESI, thereby providing insights for ES management. The results showed that before LULC optimization, basin-wide WY exhibited a declining trend from the historical to future periods, whereas the other ES metrics generally increased. Because NE is a negative attribute, its increase indicates weakened water-purification performance and may reflect intensified land-use pressure or agricultural activity. More diverse LULC configurations were generally accompanied by more synergies among ESs, while trade-offs mainly occurred between HQ and WY, particularly in downstream regions. IESI broadly increased, except in “spatially clustered low-performance” or “spatial mismatch” areas characterized by low-low clusters or low-high outliers of IESI and often located in upstream and midstream regions with cropland expansion. A moderate conversion of cropland and undeveloped land in LULC-optimization priority areas into a combination of grassland (12.2%–15.7% of the priority-area extent) and forest (4.3%–6.8%) was identified as the preferred conversion pathway. This pathway increased mean IESI within the priority areas by 6.2%–6.4% and basin-wide mean IESI by approximately 0.9%–1.1%. Under this LULC conversion pathway, the modelled socioeconomic-IESI relationships changed differently, with the negative association between population and IESI weakening and gross domestic product (GDP) shifting from a negative to a positive association with IESI, while significant spatial spillover effects indicated interdependence among neighboring cities. These findings warrant greater consideration of targeted LULC optimization and cross-city interactions in future basin-scale ES management.

1 Introduction

The term “Ecosystem Service” (ES) refers to the environmental conditions generated and maintained by ecosystems, as well as the contributions from these ecosystems to human survival and wellbeing (Costanza et al., 1997). The United Nations has announced the Sustainable Development Goals to promote the harmony between human societal progress and ESs. The meaning of ESs can be broadly recognized as being closely linked to human welfare: provisioning services, including water supply for humans (Bai et al., 2018); regulating services, contributing to water purification and carbon reduction in terrestrial ecosystems (Xu et al., 2023); supporting services, affecting soil retention and habitat biodiversity (Bhattarai and Dutta, 2007; Chuenchum et al., 2020); and cultural services, providing natural landscapes of rivers/lakes and resources for ecotourism (Kosanic and Petzold, 2020; Yang et al., 2021). Research on ESs to support resource management and policy decisions has become a focal point for policymakers, scientists, and multi-sector stakeholders (Laterra et al., 2012; Schulp et al., 2014; Wan et al., 2015; Li et al., 2022). It serves as an essential component in the sustainable development research for large rivers, cross-regional areas, and trans-watersheds (Sherrouse et al., 2011; Li et al., 2024).

Land use/land cover (LULC), as an explicit manifestation of human impacts on Earth’s surface, reflects changes in urban socio-economic systems and profoundly shapes dynamic patterns of ESs (Jiang et al., 2016; Meraj et al., 2022). Recent research increasingly extends beyond single aspects of LULC, underscoring the necessity of integrating LULC into ES assessments (Yang et al., 2021; Ji et al., 2023). Significant alterations in LULC have led to conflicts within multiple ESs, resulting in several forms of ecological degradation, including reduced carbon storage, unstable biodiversity maintenance, habitat loss, declines in water resource supply and quality, and increased soil erosion (Cortinovis and Geneletti, 2019). Therefore, investigating trade-offs (i.e., the compromise of certain ESs to enhance others) and synergies (i.e., simultaneous increases or decreases across multiple ESs) from the perspective of LULC drivers provides essential insights into the intricate interrelationships among various ESs (Schulp et al., 2014).

Future projections of socioeconomic development and ecosystem evolution in river basin systems are increasingly drawing attention (Jiang et al., 2023). The Coupled Model Intercomparison Project Phase 6 (CMIP6) employs a combination of Shared Socioeconomic Pathways (SSP) and Representative Concentration Pathways (RCP) to provide predictions and scenario settings for future LULC and socioeconomic changes under various SSP-RCP scenarios. This approach has garnered significant attention from researchers (Chen et al., 2020; Chen et al., 2022; Ji et al., 2023). In existing studies on future ES predictions based on LULC scenarios, integrating spatiotemporal dynamics of LULC with ES is more comprehensive and convincing than focusing solely on either temporal or spatial scales (Bai et al., 2018). Various modeling methods have been developed, including the Soil and Water Assessment Tool (SWAT) model (Gassman et al., 2007), Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) (Sharp et al., 2016), Artificial Intelligence for Ecosystem Services (ARIES) (Bagstad et al., 2014), Social Values for Ecosystem Services (SolVES) (Sherrouse et al., 2011), and Ecological Asset Inventory and Management (EcoAIM) (Meraj et al., 2022). InVEST has become popular and widely applied worldwide (Xu et al., 2023) due to its advantages, such as fewer model parameters, relatively accessible data, high global applicability, and adjustable spatiotemporal resolution to reduce uncertainty (Gassman et al., 2007; Sharp et al., 2016; Chen et al., 2022). In studies concerning China, the use of administrative boundaries, such as provinces, cities, or counties, is prevalent (Ji et al., 2023; Xu et al., 2023). Some research adopting watershed, cross-regional, or multiscale perspectives still relies predominantly on administrative units along rivers, which do not fully represent river basin systems and urban agglomeration. Most studies have examined only a limited number of ecosystem services, lacking a comprehensive assessment of future LULC impacts. Moreover, the limited number of CMIP6 SSP-RCP scenarios considered in current studies may constrain the representation of potential future spatiotemporal changes in ESs.

The Yellow River Basin is of considerable strategic importance, serving as a vital ecological barrier, a key economic development area, and a central region for achieving China’s “carbon neutrality” targets (Li et al., 2024). It is also one of the most impacted regions globally by human activities, undergoing substantial LULC transformations and rapid socio-economic growth (Ji et al., 2023; Xu et al., 2023). The basin faces high pressure on its finite resources and a vulnerable ecological system, which constrains its potential for coordinated development (Jiang et al., 2023). The Chinese government places significant emphasis on ecological security and has formulated policies such as the “Ecological Red Line”, “Ecological Civilization Construction and Function Zoning”, and the “Yellow River Basin Ecological Protection and High-Quality Development” initiatives. Nonetheless, information on ESs available for watershed management remains insufficient.

To address the aforementioned knowledge gaps, this study selects the Yellow River Basin as a typical research area and aims to achieve the following objectives: (1) project the spatiotemporal patterns of five ESs from 2030 to 2100 at the pixel, prefecture-level, provincial, and watershed scales based on LULC predictions under five SSP-RCP scenarios; (2) construct an integrated ES index to evaluate the performance of multiple ESs and apply it in multi-scale assessments to compare alternative LULC conversion pathways; (3) examine the spatial associations between socioeconomic factors and IESI across different periods; and (4) identify a beneficial LULC conversion pathway to improve ES performance. This study is expected to support water and soil resource management and the development of ES strategies for sustainable development in large river basins.

2 Study area and data

2.1 Overview of study area

The Yellow River is the second-longest river in China, stretching from 95° E to 120° E and 32° N to 42° N (Figure 1), with a total length of 5,464 km and a drainage area of approximately 795,000 km². Originating from the Qinghai-Tibet Plateau, it traverses nine provincial-level regions, with Qinghai, Sichuan, Gansu, Ningxia, and Inner Mongolia mainly in the upper reaches, Shanxi and Shaanxi in the middle reaches, and Henan and Shandong in the lower reaches, ultimately discharging into the Bohai Sea (Li et al., 2024). The Yellow River water-receiving area encompasses 67 prefecture-level cities. The Yellow River Basin spans the Qinghai-Tibet Plateau, Loess Plateau, and North China Plain. It is predominantly distinguished by semi-arid to arid climate types, with highly uneven precipitation ranging from 240 to 600 mm annually. Despite significant ecological improvements resulting from governmental strategic initiatives aimed at protecting the Yellow River environment, the basin remains vulnerable due to recent social development and resource exploitation (Fang et al., 2021). Persistent challenges such as land desertification, declining water conservation capacity, agricultural non-point source pollution, and soil erosion require ongoing attention and effective solutions (Ji et al., 2023; Xu et al., 2023).

FIGURE 1

2.2 Data and scenarios

All basic data used are listed in Table 1. All gridded inputs were processed to a common spatial resolution of 1 km in the WGS_1984_Cylindrical_Equal_Area projection coordinate system. For the future projections, five coupled SSP-RCP scenarios developed within the CMIP6 framework (Eyring et al., 2016) were used to represent alternative socioeconomic development and radiative-forcing pathways underlying future LULC change. In this notation, “SSPx” denotes the underlying socioeconomic development pathway, whereas “RCPy” represents the corresponding radiative forcing level by 2100. The scenarios included SSP1-RCP26 (Sustainability; low greenhouse gas emissions), SSP2-RCP45 (Middle of the Road; intermediate greenhouse gas emissions), SSP3-RCP70 (Regional Rivalry; relatively high greenhouse gas emissions), SSP4-RCP60 (Inequality; a radiative forcing level of approximately 6.0 W m^-2 by 2100), and SSP5-RCP85 (Fossil-fueled Development; very high greenhouse gas emissions). The future LULC dataset developed by cellular automata and future land use simulation (FLUS) models under CMIP6 SSP-RCP scenarios in Chen et al. (2022) was adopted in this study, as it shows the fine resolution and the latest socio-climatic scenarios. All LULC data were reclassified into six categories: water body, forest, grassland, cropland, built-up land, and undeveloped land (Supplementary Figure SA1). In the future ES simulations, because this study focused on isolating the effects of LULC change, only the scenario-specific LULC maps varied during 2030–2100, whereas precipitation and reference evapotranspiration were fixed at their 1950–2022 climatological mean values and the other non-LULC inputs were held constant. Accordingly, the projected differences in ESs primarily represent LULC-mediated responses under alternative SSP-RCP pathways rather than the direct effects of future climate change. In the analysis of socioeconomic associations with future ES performance, population and GDP projections under SSP1-SSP5 were obtained from Chen et al. (2020) and Wang and Sun (2022), respectively, and future urbanization rates were obtained from projections of the Yellow River Conservancy Commission.

3 Methods

An overview of the technical route of this research is shown in Figure 2. Spatial scales included pixel, prefecture-level cities, provincial boundaries, and the entire river basin. The temporal scope from 1980 to 2100 was divided into three periods: historical (1980–2020), near future (2030–2050), and distant future (2055–2100). The two future periods incorporated the aforementioned five SSP-RCP scenarios for LULC.

FIGURE 2

3.1 ES metrics based on LULC

Five ES metrics (Table 2) were selected based on three criteria: the academically recognized importance (Fang et al., 2021; Ji et al., 2023), the availability and integrity of recent data (Yan et al., 2020; Yang et al., 2021; Xu et al., 2023), and the alignment with basin policy goals and public demand (Bai et al., 2018; Li et al., 2024).

3.1.1 Carbon storage

Carbon storage (CS) is estimated in the InVEST model using four principal carbon pools: aboveground biomass (, comprising all living plant material above the ground), belowground biomass (, encompassing the active root systems of vegetation), soil (, containing both organic and mineral components), and dead organic matter (, including litter, fallen branches, and dead trees). This model simplifies the descriptions of carbon cycling processes by associating these carbon pools with pixel-scaled LULC maps, though it does not incorporate biophysical complexities or dynamic factors such as tree growth, changes in soil chemistry, or meteorological variations. It estimates the net carbon retained in land units across periods by calculating the average carbon density for a certain LULC type. The carbon density values for these types (Supplementary Table SA1) were downloaded from the National Ecosystem Science Data Center (http://www.nesdc.org.cn/) and adjusted based on the modifications outlined in Xu et al. (2023). These values were used to estimate the total carbon stock for area of interest, namely, .

3.1.2 Water yield

Water yield (WY) module in the InVEST model calculates the annual WY () using the annual total precipitation () and actual annual total evapotranspiration () on each pixel , as shown in Equation 1.

InVEST distinguishes vegetated and non-vegetated LULC classes through the parameter LULC_veg (Supplementary Table SA2). For classes with LULC_veg = 0, was calculated as min(). For classes with LULC_veg = 1, was estimated using the Budyko formulation in Equations 2, 3:where , , and are annual WY (mm), annual total precipitation (mm), and annual total evapotranspiration (mm) on a pixel , respectively. is the potential evapotranspiration that can be estimated from the reference evapotranspiration and the evaporation factor for each LULC and is a non-physical parameter that reflects the inherent climatic and soil characteristics. is the plant available water content (mm) computed by the product of two parts: (1) the plant available water capacity and (2) the minimum of root restricting layer depth and vegetation root depth . measures the gap between field capacity and wilting point and can be computed according to the available information on soil texture, following the estimation method proposed by Zhou et al. (2005), which adapts to the physicochemical properties of soil in China. values are approximated by the depth to bedrock provided by Yan et al. (2020). is an adjustable constant reflecting the precipitation regime and other related hydrogeological properties of the study area. was set to 3.7 because the deviation between the simulated WY and the natural runoff during the multi-year observed period (1990–2020) was the smallest when (Supplementary Figure SA2). The biophysical table for each LULC type summarizes the parameter values of and root depth (Supplementary Table SA2), which were obtained from other research (Yang et al., 2021), FAO crop reference coefficients (Droogers and Allen, 2002), and parameters suggested by the InVEST designer (Sharp et al., 2016).

3.1.3 Nutrient export

Nutrient export (NE) is estimated using the Nutrient Delivery Ratio module of the InVEST model, which calculates nutrient loads across the landscape based on loading rates for different LULC types and further adjusts them according to the local runoff potential index using Equation 4.where is the modified nutrient loads on a pixel . , is the log-transformed value of the cumulative water yield of all pixels along the flow path upstream of pixel , and is the average over all pixels of the whole river basin. The NE variable at pixel is obtained by multiplying and the nutrient delivery ratio using Equation 5:where is estimated from nutrient retention efficiencies (eff_n) and maximum retention distances (crit_len_n). This research concentrated on nitrogen as a representative nutrient in the Yellow River Basin, assuming that all nutrients enter the stream through surface flow (conceptually, including shallow subsurface flow). The parameter values of the nitrogen loadings (load_n), retention efficiencies (eff_n), and maximum retention distances (crit_len_n) for each LULC type in the study area were referred to expert consultation (Supplementary Table SA3).

3.1.4 Soil conservation

Soil conservation (SC) for different LULC types is measured using Equation 6, based on the revised universal soil loss equation (RUSLE) (Renard et al., 1997), which has been demonstrated as reliable in terms of assessing water-eroded soil loss (Bhattarai and Dutta, 2007; Chuenchum et al., 2020).where is the amount of soil conservation on pixel (t·hm^-2·yr^-1), is the potential soil erosion amount in the absence of cover-management vegetation (t·hm^-2·yr^-1), is the actual soil erosion (t·hm^-2·yr^-1), is the rainfall erosivity (MJmm·hm^-2·h^-1·yr^-1) and can be derived from monthly precipitation, is the soil erodibility factor (thm²·h·MJ^-1·mm^-1·hm^-2), is the combined slope-length () and slope-steepness () factor (unitless), is the cover-management factor (unitless) to measure the inhibitory impacts of vegetation cover and management on erosion control, and is the support practice factor (unitless) to reflect water and soil conservation measures. For application, were derived from gridded monthly precipitation following the method in Wischmeier and Smith (1978), were computed by the erosion/productivity impact calculator (EPIC) model with the information of the soil texture (Williams and Arnold, 1997), and were calculated from the multiplication of slope length and slope steepness using the previously developed formulas (McCool et al., 1989; Liu et al., 1994). and were determined (Supplementary Table SA4) according to the previous documents (Renard et al., 1997; Fang et al., 2021). , , , and were set as time-constant values given their relative stability and difficulties in time series predictions.

3.1.5 Habitat quality

Habitat quality (HQ) represents an ecosystem’s capacity to furnish favorable conditions for maintaining individual and population viability, which can be regarded as a proxy indicator of biodiversity status. The formulation in the InVEST model is shown by Equations 7–9:where is a continuous variable indicating the habitat quality on a pixel with LULC type and is the habitat suitability of LULC type . indicates a higher habitat quality and if . is the half-saturation constant defined as 50% of the highest degradation value on pixel and is the fixed parameter equal to 2.5 by default. represents the total threat extent on pixel with LULC type . is the number of pixels occupied by threat . is the weight of threat ; indicates the threat on pixel (equal to 1 for existence, otherwise 0), with the total number of ; denotes the consequence of threat from pixel on the habitat/LULC type at pixel ; represents the relative accessibility of habitat pixels to the modeled threats, ranging from 0 (inaccessible) to 1 (fully accessible). Although some key ecological function zones have been designated in the Yellow River Basin, these zones are inherently ecologically sensitive and therefore remain exposed to persistent pressures and environmental threats. Given the lack of consistent information on the effectiveness of protection measures in reducing the modeled threats, and following previous studies (Fang et al., 2021), was set to 1 for all pixels as a precautionary and conservative assumption to avoid overestimating their threat-mitigation effects. represents the sensitivity of habitat on LULC type to the threat . is the linear distance between pixel and pixel . is the maximum effective distance of threat across space. The operating parameters of all threats (Supplementary Table SA5) and the sensitivity ratings of each habitat to each threat (Supplementary Table SA6) for the Yellow River Basin were collected from the related studies (Fang et al., 2021; Ji et al., 2023).

3.2 Integrated ecosystem services index (IESI)

In addition to individual ES metrics (Schulp et al., 2014; Wan et al., 2015), an integrated ecosystem services index (IESI) was developed through a weighted sum of all individual ES metrics to reflect the overall ecosystem service performance. Five weighting schemes were examined: equal weighting, entropy weighting (Shannon, 1948), CRiteria Importance Through Intercriteria Correlation (CRITIC) (Diakoulaki et al., 1995) weighting, period-specific composite weighting, and fixed composite weighting. Under the equal-weighting scheme, each of the five ES metrics was assigned a weight of 0.2. The entropy method determines weights according to the variation and information content of each ES metric, whereas the CRITIC method considers both contrast intensity and correlations among metrics. The composite scheme integrates the complementary information captured by entropy and CRITIC, thereby leveraging both the variability and inter-metric relationships of different ESs (Chang and Zhu, 2020), and was adopted for the main analysis to reduce reliance on a single weighting criterion. The entropy, CRITIC, and composite weights were calculated separately for the historical, near-future, and distant-future periods and were held constant for all observations within the corresponding period. To examine whether period-specific weighting affected IESI comparisons, the fixed composite weights were derived from the pooled dataset containing all historical observations and future scenarios and applied consistently across the three periods.

For each weighting scheme, the IESI for pixel in year was calculated by Equation 10:where is the total value of IESI on a pixel for the study area of interest for year ; is the number of ESs considered (i.e., ). is the composite weight attributed to the ES, with its normalized value as for each year under the corresponding weighting scheme and study period, to ensure comparability and eliminate the influence of differences in units and magnitudes among the ES metrics.

The composite weight was calculated by Equation 11:where and are the weights obtained using the entropy and CRITIC methods, respectively. In the absence of evidence supporting the prioritization of either method, the two objective weighting methods were treated as equally important, and was set to 0.5.

The value is obtained from the information content from the information content Ci using Equations 15, 16.where is the sample standard deviation of each normalized ES metric and is the correlation coefficient between the and normalized ES metrics.

3.3 Spatial dependence and clustering analysis

Trade-offs, synergies, and neutral relationships between ES pairs were identified using Pearson correlation analysis (Yang et al., 2021) at both the basin and pixel scales. It should be noted that NE was reverse-normalized using Equation 12, such that higher values represented lower nutrient export and better ecosystem-service performance. At the basin scale, the spatially averaged value of each ES across the entire Yellow River Basin was calculated for each selected year and scenario. Pearson correlation coefficients were then calculated for all possible pairs of ESs within each spatial unit and study period. A positive Pearson correlation coefficient that was statistically significant (p < 0.05) was classified as a synergistic relationship, indicating that the two ESs tended to change in the same direction. A negative Pearson correlation coefficient that was statistically significant (p < 0.05) was classified as a trade-off relationship, indicating that an increase in one ES was generally accompanied by a decrease in the other. Non-significant correlations were classified as neutral relationships because there was insufficient statistical evidence to classify them as either synergistic or trade-off. These Pearson-based classifications are interpreted as exploratory associations. The analysis was conducted separately for three periods: the historical period (1980–2020), the near-future period (2030–2050), and the distant-future period (2055–2100). For the historical period, the analysis was based on the areal mean values of the eight selected historical years, resulting in a sample size of 8. For the near-future period, five selected years were evaluated under each of five SSP-RCP scenarios. The scenario-specific annual values were pooled to represent the overall range of possible near-future conditions, resulting in a sample size of 25 (5 years × 5 scenarios). Similarly, the distant-future analysis included 10 selected years under each of the five scenarios, resulting in a sample size of 50 (10 years × 5 scenarios). At the pixel scale, Pearson correlations between each ES pair were calculated for individual grid cells based on their temporal series and classified as trade-offs, synergies, or neutral relationships using the same criteria described above. The area percentages of the three relationship categories were then calculated for each ES pair during the historical, near-future, and distant-future periods, with the five future SSP-RCP scenarios analyzed independently. The spatial contributions of the upper, middle, and lower Yellow River Basin were further quantified separately. Specifically, for each basin section and relationship category, the area percentage was calculated as the area of pixels assigned to that category within the section divided by the total valid pixel area of the entire Yellow River Basin and multiplied by 100%.

Moran’s indices in Equation 17 were utilized to assess the overall spatial association of different ESs across the entire geographical adjacent units of the study area using ArcGIS.where is Moran’s index. denotes a spatial aggregation of similar clusters; indicates a spatial dispersion, where opposite values fluctuate between high and low, likely alternating in distribution and thus showing the presence of negative spatial autocorrelation; indicates a random data distribution with no significant spatial autocorrelation. is the total number of pixels; and are the ES values of regions and , respectively, having the average value ; represents the spatial proximity between regions and ; is the variance of observations.

The local indicators of spatial association (LISA) map is a visualization tool for detecting and displaying local spatial autocorrelation at each location in geographic units, with the local Moran’s index defined in Equation 18.

To highlight how each location’s ES value relates to the values of its neighboring locations, the LISA map uncovers four spatial patterns of aggregations or anomalies, including high-high (hotspot, viewed as clusters of high values), low-low (cold spot, viewed as clusters of low values), high-low (outliers with high values encircled by low values), and low-high (outliers with low values encircled by high values). A positive value signifies the existence of high-high and low-low clusters, whereas a negative value signifies the existence of high-low and low-high clusters.

3.4 Spatial durbin models (SDM)

To assess the associations between socioeconomic development and IESI, this study employed a spatial Durbin model (SDM; LeSage and Pace, 2009), as expressed in Equation 19.where is the response variable; and are the explanatory variables, representing population and gross domestic product, respectively; is the spatial autoregressive coefficient, reflecting the contemporaneous spatial dependence of IESI. denotes control variables included to account for other factors associated with . In this study, was represented by urbanization rate, measured as the proportion of urban population in the total population, while denotes its spatially lagged term. is the intercept term; and are the coefficients of local and , while and are the coefficients of their spatially lagged terms and , respectively. Similarly, and are the coefficient vectors of the local control variables and their spatially lagged terms; is the random error term; and denote individual (regional) fixed effects and time-fixed effects, respectively, with and indicating prefecture-level cities and time. (when ; otherwise ) is the spatial weight matrix, constructed based on the geographical centroid distance between cities and , i.e., spatial correlation weakens as geographical distance increases. was row-standardized for all model estimations and spatial-effect decompositions. To reduce numerical differences and enhance robustness, all response, explanatory, and control variables were log-transformed. The spatial effects derived from the SDM can be separated into the direct (local) and indirect (spatial spillover) effects of each explanatory variable on IESI according to the partial differential decomposition method (LeSage and Pace, 2009). The total effect is the sum of the direct and indirect effects.

The SDM was estimated using panel data for 67 prefecture-level cities, with each sample representing a (city × year) pair. City-level IESI values were obtained by averaging all pixels within each city. Based on the data availability summarized in Table 1, the historical, near-future, and distant-future panels comprised 6 years (1995, 2000, 2005, 2010, 2015, and 2020), 3 years (2030, 2040, and 2050), and 5 years (2060, 2070, 2080, 2090, and 2100), respectively. Variables were temporally subsampled to these common observation years, and no temporal interpolation was applied. Accordingly, the three panels contained 402, 201, and 335 (city × year) samples, respectively. For the future periods, scenario averaging was conducted separately for each (city × year) sample according to the projection framework of each variable. IESI was averaged across the five SSP-RCP scenarios, whereas POP and GDP were averaged across the five SSP-based projections (SSP1-SSP5). To ensure valid model specification, the LM-lag, LM-error, robust LM-lag, robust LM-error, LR-lag, LR-error, Wald-lag, and Wald-error tests were conducted to examine whether the SDM could be simplified to a spatial autoregressive model or a spatial error model. The Hausman test was subsequently used to determine whether fixed or random effects were appropriate.

3.5 LULC optimization procedure

A rule-based, model-conditioned LULC scenario comparison was conducted to evaluate alternative LULC conversion pathways. Priority areas were identified based on the LISA results, including spatially clustered low-performance areas represented by significant low-low clusters and spatial mismatch areas represented by high-low and low-high outliers. The projected LULC maps were overlaid with these priority areas, and only cropland and undeveloped-land pixels within them were considered eligible for conversion. The availability of cropland and undeveloped land within the priority areas was first quantified through spatial overlay analysis. Based on land availability and cropland-protection and food-security considerations, the maximum conversion proportions of undeveloped land and cropland were then determined. All conversion proportions were defined relative to the total priority-area extent. Candidate pixels were ranked according to their pre-optimization IESI values, with lower-IESI pixels selected first until the specified conversion area was reached. During the scenario simulations, undeveloped land was progressively reduced in increments of 1 percentage point and cropland in increments of 0.1 percentage points up to their respective maximum conversion proportions. The areas removed from cropland and undeveloped land were allocated to grassland, forest, or a combination of both, with the allocated area adjusted in increments of 0.1 percentage points.

Three conversion pathways were evaluated by converting the selected pixels into grassland, forest, or a combination of both. For each pathway and conversion proportion, modified LULC maps were generated for the near-future and distant-future periods and used to recalculate the corresponding ESs and IESI values. To ensure direct comparability between the pre-optimization and optimized cases, the metric-specific normalization bounds and corresponding period-specific ES weights adopted in the pre-optimization analysis were retained when recalculating the IESI under all conversion pathways, while all non-LULC inputs remained unchanged. The robustness of the optimization results was further assessed using a single fixed composite weight vector across periods. The preferred pathway and conversion proportions were determined according to improvements in mean IESI and the responses of the individual ESs. To assess the influence of the pixel-selection rule, eligible pixels were randomly selected from the same candidate areas under the preferred conversion proportions. Random allocation was repeated 100 times for each pathway and future period, and the resulting changes in mean IESI were summarized using their medians and 95% empirical intervals, with comparisons between the low-IESI-priority allocation and random allocation. The procedure identifies the best-performing pathway among the tested scenarios rather than a universally optimal land-management solution.

4 Results

4.1 Spatiotemporal evolutions of ESs

The spatiotemporal dynamics in ES metrics are illustrated for three periods: 1980–2020, 2030–2050, and 2055–2100 (Figure 3). CS was highest in the midstream regions, exhibiting an increase in multi-year average values, which ranged from 44.97 Mg C/hm² to 47.98 Mg C/hm² and further increased to 48.13 Mg C/hm² by 2055–2100, with considerable contributions from the middle and lower regions. WY displays higher values concentrated in the upstream southwest regions, with a gradual reduction observed from 62.94 mm (1980–2020) to 62.45 mm (2030–2050), and further to 61.81 mm (2055–2100) in the future. NE increased slightly from 205.39 kg/km² to 257.28 kg/km², showing higher values in the central regions, particularly in the southern regions. This increase indicates weakened water-purification performance and intensified land use pressures or agricultural activity. SC evolved from 76.03 t hm^-2·yr^-1 to 162.74 t hm^-2·yr^-1, being the highest in the midstream regions, and showing an overall increase in soil conservation capacity in areas prone to erosion. HQ improved on average from 0.16 to 0.23, and further to 0.26, being highest in the central or southern regions, and tending to improve in the middle and lower sections of the basin.

FIGURE 3

4.2 IESI estimation and distribution characteristics

The weights assigned to the five ES metrics under the equal, entropy, CRITIC, and composite weighting schemes are presented in Supplementary Table SA7 and were used to calculate the corresponding IESI values. Despite differences in the weights assigned to individual ES metrics, the alternative weighting schemes produced only minor differences in mean IESI, while the temporal trends and spatial rankings remained highly consistent with those obtained using the composite weights (Supplementary Figure SA3). For clarity and conciseness, only the IESI results based on the period-specific composite weights (A, B, and C under the composite scheme in Supplementary Table SA7) are presented in the main text. Figure 4 shows that the mean IESI values were 0.4310, 0.6379, and 0.6911 during the historical (1980–2020), near-future (2030–2050), and distant-future (2055–2100) periods, respectively, indicating a marked improvement in integrated ES performance. During 1980–2020, relatively low IESI values were mainly concentrated in the western and northern parts of the basin. During 2030–2050, IESI increased noticeably in the middle and lower reaches. During 2055–2100, IESI values exceeded 0.60 across most of the basin, with the highest values concentrated in the central and southern parts. The fixed composite weight sensitivity analysis similarly produced the increasing temporal pattern, with basin-average IESI increasing from 0.4408 in the historical period to 0.6225 and 0.6857 in the near- and distant-future periods, respectively (Supplementary Figure SA3). Although the magnitude of change differed slightly from that obtained using the period-specific composite weights, the increasing temporal trend in IESI was robust to temporal variations in the weighting scheme.

FIGURE 4

Figure 5 compares the spatiotemporal heterogeneity of IESI levels in the past and future. Historical IESI values were significantly lower than those under future scenarios. The upper and middle regions exhibit higher IESI values compared to the lower regions. The IESI values were generally lower under the SSP5-RCP85 scenario than under the SSP3-RCP70 scenario.

FIGURE 5

4.3 Analysis of trade-offs/synergies between ES pairs

The trade-offs, synergies, and neutral relationships between ES pairs were analyzed using Pearson correlations (Figure 6). CS and WY consistently showed a synergistic relationship across all periods. CS and SC showed significant synergies in the historical and near-future periods, but their relationship became neutral in the distant-future period. The relationship between WY and SC showed a synergistic trend during 1980–2020 and 2055–2100. NE and SC evolved from a trade-off relationship to a synergistic one after 2050. WY and NE exhibited a significant synergistic relationship in future periods after 2030, while the CS-NE and SC-HQ relationships showed significant synergies only during 2030–2050. The NE-HQ relationship shifted from synergistic during 2030–2050 to a trade-off after 2050. WY-HQ exhibited trade-offs after 2050.

FIGURE 6

The bar plots in Figure 7 depict the area percentages of trade-offs, synergies, and neutral relationships between ES pairs across the three study periods and five future scenarios. Different colors represent the upper, middle, and lower sections of the Yellow River Basin, and each section is further divided into trade-offs, synergies, and neutral relationships. Neutral relationships generally accounted for relatively small proportions and did not alter the dominant patterns of trade-offs and synergies. Figure 7 shows that trade-off relationships were more prevalent across most areas under SSP5-RCP85 scenario than under SSP1-RCP26 scenario. Low-emission scenarios exhibited higher LULC diversity than high-emission scenarios (Supplementary Figure SA4) and were also associated with a greater proportion of synergistic relationships among ESs. For CS-SC and CS-WY, synergies (positive correlations) dominated overall, with upstream regions contributing a higher proportion. For CS-HQ, trade-offs and synergies were broadly comparable, accompanied by a relatively small proportion of neutral relationships. For HQ-SC and HQ-NE, synergies (positive correlations) prevailed in upstream and midstream regions, while downstream regions shifted from historical synergies to trade-offs under future high-emission scenarios, suggesting potential for improving synergies. For CS-NE, most areas exhibited synergies during the historical period but transitioned to trade-offs under future high-emission scenarios. For WY-HQ, upstream and midstream regions shifted from a mixture of trade-offs and synergies during the historical period to predominantly synergy relationships under most future scenarios, although trade-offs dominated under SSP3-RCP70 during the distant-future period (2055–2100) and remained more prominent in the downstream region. For NE-SC, trade-offs dominated during the historical period but transitioned to synergies in the future. For WY-SC and WY-NE, synergies dominated, with the highest proportions in upstream regions and increasing proportions observed in the middle section.

FIGURE 7

4.4 Mapping of spatiotemporal clusters/outliers in IESIs

The multi-year average Moran’s indices during 1980–2020, 2030–2050, and 2055–2100 (Supplementary Table SA8) indicate significant positive spatial autocorrelations of IESI values, with Moran’s indices spanning from 0.2438 to 0.4199 (p < 0.0001). Figure 8 presents the LISA maps of average IESI values for three time periods: 1980–2020, 2030–2050, and 2055–2100. The detailed spatial clusters or outliers of IESI values under the five scenarios are illustrated in Supplementary Figure SA5. The results show that Qinghai, Gansu, Shaanxi, Shanxi, and Henan exhibited significantly high-high clusters with geographically concentrated distributions during 1980–2020. In future periods, the spatial pattern of high-high clusters underwent notable shifts, with localized regions featuring relatively rare high-low and low-high outliers, which were identified as “spatial mismatch” areas. Additionally, areas corresponding to statistically significant low-low clusters were identified as “spatially clustered low-performance” areas, indicating spatial concentrations of relatively low IESI values. Inner Mongolia, along with localized areas in northern Ningxia and northwest Shaanxi, displayed significant low-low clusters during 1980–2020, though these became less pronounced after 2030. The combined “spatially clustered low-performance” and “spatial mismatch” areas, hereafter referred to as the LULC-optimization priority areas, decreased from 29.3% (approximately 232,935 km²) during 1980–2020 to 18.0% (approximately 143,100 km²) during 2030–2050, and then increased to 20.7% (approximately 164,565 km²) during 2055–2100. In the two future periods, undeveloped land accounted for 1.8% and 2.1% of the basin area, equivalent to approximately 10% of the corresponding priority-area extent, while cropland accounted for 11.6% and 13.9% of the basin area, equivalent to 64.4% and 67.1% of the priority areas, respectively. Accordingly, the maximum conversion of undeveloped land was set at 10% of the priority-area extent, approximately corresponding to its available area. Although cropland was more abundant, its conversion was also capped at 10% to limit cropland loss and account for cropland-protection and food-security requirements. These limits were adopted in the subsequent LULC optimization.

FIGURE 8

4.5 Optimization of future LULC projections

Figures 9, 10 compare the mean IESI and corresponding ES responses before and after optimization within the LULC-optimization priority areas. All reported LULC conversion percentages are expressed relative to the total priority-area extent, and basin-wide IESI is reported separately. The IESI improvements under the low-IESI-priority allocation fell within the 95% empirical intervals of the 100 random allocations, while the ranking of the three pathways remained stable in both future periods (Supplementary Figure SA6). The optimized mean IESI values varied slightly across weighting schemes, but pathway C consistently achieved the highest mean IESI among the tested LULC conversion pathways and was therefore identified as the preferred pathway. This consistency suggests that the pathway ranking was generally stable across alternative weighting schemes (Supplementary Figure SA7).

FIGURE 9

FIGURE 10

For 2030–2050 (Figure 9), the mean pre-optimization IESI within the priority areas was 0.5358. The three pathways were ranked as pathway C (0.5690; +6.2%) > pathway A (0.5615; +4.8%) > pathway B (0.5529; +3.2%). Pathway A converted 20.0% of the priority-area extent (10.0% cropland and 10.0% undeveloped land; approximately 28,620 km²) to grassland. Pathway B converted 15.2% (5.2% cropland and 10.0% undeveloped land; approximately 21,751 km²) to forest. Pathway C converted 19.0% (9.0% cropland and 10.0% undeveloped land; approximately 27,189 km²) into 12.2% grassland and 6.8% forest. Under pathway C, CS, WY, SC, and HQ increased by 10.8%, 16.5%, 13.1%, and 15.7%, respectively, whereas NE decreased by 14.9%. At the basin scale, mean IESI increased from 0.6379 to 0.6425, 0.6410, and 0.6439 under pathways A, B, and C, respectively.

For 2055–2100 (Figure 10), the mean pre-optimization IESI within the priority areas was 0.5906. The three pathways were again ranked as pathway C (0.6284; +6.4%) > pathway A (0.6213; +5.2%) > pathway B (0.6065; +2.7%). Pathway A converted 20.0% of the priority-area extent (10.0% cropland and 10.0% undeveloped land; approximately 32,913 km²) to grassland. Pathway B converted 13.2% (3.2% cropland and 10.0% undeveloped land; approximately 21,723 km²) to forest. Pathway C converted 20.0% (10.0% cropland and 10.0% undeveloped land; approximately 32,913 km²) into 15.7% grassland and 4.3% forest. Under pathway C, CS, WY, SC, and HQ increased by 11.5%, 14.1%, 10.7%, and 8.6%, respectively, whereas NE decreased by 16.3%. At the basin scale, mean IESI increased from 0.6911 to 0.6974, 0.6944, and 0.6989 under pathways A, B, and C, respectively.

4.6 Urban socioeconomic associations with IESI

Table 3 presents the estimation results of the fixed-effects SDMs, with the corresponding diagnostic test results for all five SDM specifications reported in Supplementary Table SA9. The LM tests indicated significant spatial dependence, while the LR and Wald tests rejected simplification of the SDM to either the spatial autoregressive model or the spatial error model. The Hausman tests further supported the use of fixed effects. Accordingly, fixed-effects SDMs were adopted for all five specifications. The spatial autoregressive coefficients were positive and statistically significant at the 1% level across all periods, indicating significant positive spatial dependence in IESI after controlling for socioeconomic variables, other covariates, and city and time fixed effects. For the future periods, the SDM estimates represent scenario-averaged associations based on the five scenario pathways. For the historical period (1995–2020), POP and GDP both showed significant negative associations with IESI. The effect decomposition results in Table 4 further show that both variables had significant negative direct, indirect, and total effects during this period. For the future periods, the association between POP and IESI weakened and became statistically insignificant. In contrast, GDP shifted from a significant negative association in the historical period to significant positive associations in the future. Its direct, indirect, and total effects were also positive and significant, and became stronger after LULC optimization in both future periods. These future estimates are conditional on the internally consistent SSP-based socioeconomic and LULC projections and should therefore be interpreted as scenario-conditioned socioeconomic-IESI associations rather than independent causal effects.

5 Discussion

5.1 Management implications

The Chinese government has implemented a range of ES management strategies to address water scarcity and ecological challenges, including soil and water conservation, the “Three-North Shelterbelt Project”, “Grain for Green” policy, “Ecological Function Zoning”, and water-security initiatives (Jiang et al., 2016; Bai et al., 2018; Li et al., 2024). Despite these efforts, ES relationships varied between synergies and trade-offs across spatiotemporal scales (Figures 6, 7). Low-emission scenarios generally exhibited a greater prevalence of ES synergies together with higher LULC diversity (Supplementary Figure SA4). Similar synergistic relationships have been reported in the Yellow River Basin (Geng et al., 2022; Shen et al., 2024; Li G. et al., 2021; 2025). However, multiple ESs respond to common drivers such as precipitation and vegetation conditions, which may themselves vary across environmental contexts. Filoso et al. (2017) showed that forest restoration may reduce water yield through increased evapotranspiration, whereas Li G. et al. (2021) reported higher mean water yield for grassland and positive LULC effects in parts of the Yellow River Basin. Consistent with this context dependence, Supplementary Table SA10 shows that, under pathway C, modeled mean WY within the priority areas increased by 24.65 mm (16.5%) during 2030–2050 and by 23.89 mm (14.1%) during 2055–2100, with undeveloped-land-to-grassland and cropland-to-grassland conversions making the largest contributions. The positive WY response obtained here therefore represents a modelled, context-specific outcome conditional on the adopted InVEST parameterization, including the different formulations applied to vegetated and non-vegetated land, as well as the specific LULC transitions considered.

The identified “spatially clustered low-performance” and “spatial mismatch” areas, marked by cropland expansion (Supplementary Figure SA1), are particularly vulnerable to transitions or degradation in ESs and therefore warrant prioritization with targeted LULC adjustment. Among the tested pathways, the combination of grassland and forest showed superior IESI performance. However, this superior performance is a model-based result and should be interpreted with caution, as practical implementation may require further assessment of restoration costs, hydrological carrying capacity, food-security requirements, and other socioeconomic constraints. Accordingly, this preferred pathway represents the best-performing modelled configuration within the tested scenario space and could thus be regarded as a potential adjustment direction rather than a universally applicable land-management prescription.

Previous studies suggest that ecological management from river to basin systems is a multidimensional engineering undertaking (Jiang et al., 2023), requiring a holistic approach that integrates institutional governance, cross-administrative collaboration, and social participation to enhance ES synergies and reduce adverse cross-regional effects (Li et al., 2024). These insights are also relevant to differentiated management across the Yellow River Basin. The source region is highly sensitive to precipitation variability and plays an important role in downstream water–sediment regulation, emphasizing water-resource conservation and ecological stability (Yang et al., 2021; Li et al., 2024). In the middle reaches, vegetation restoration involves potential trade-offs between water use and soil-conservation benefits (Fang et al., 2021; Xu et al., 2023). In the highly urbanized and intensively cultivated downstream region, ecological restoration needs to be balanced with agricultural production and land-development pressures, with particular attention to land-use regulation and riparian restoration (Li et al., 2022). The SDM-derived spatial spillover effects indicate cross-city interdependence in IESI under the scenario-averaged future projections, underscoring the relevance of cross-city coordination.

5.2 Limitations and prospects

This study provides a practical and efficient assessment framework for projecting the spatiotemporal patterns of ESs and optimizing LULC projections. The assessment framework considers a broad yet limited set of ESs and involves certain model simplifications (Sharp et al., 2016). For instance, the CS module does not simulate seasonal variations in carbon density or carbon loss due to natural succession and human interventions in LULC. The WY module neglects extreme precipitation and hydropower effects and may not accurately capture water balance processes under complex geological conditions. The NE module may underestimate water purification efficiency because some chemical and biological pollutant-removal processes are not fully represented. The SC module primarily relies on the RUSLE model, which excludes sediment sources from gullies, streambanks, landslides, etc. Resampling the DEM from 30m to 1 km may smooth fine-scale topographic variability and introduce resolution-related uncertainty into and SC estimates, particularly in steep terrain. However, DEM resolution has been reported to affect the and factors in opposite directions, with coarser DEMs potentially overestimating the factor while underestimating the factor, resulting in partial error cancellation in the factor (Li A. et al., 2021). The HQ module does not explicitly describe the transformations between threats and habitats under a specific planning context. The uniform setting of represents a conservative assumption and may not sufficiently capture spatial differences in the actual effectiveness of ecological protection measures. Large-scale field validation of all parameter configurations remains difficult; therefore, model parameters mainly follow values recommended or adjusted in previous studies. Uncertainties arising from LULC diversity (Xu et al., 2023) and spatial resolution resampling (Li et al., 2022) may affect simulated ES values, although their influence on basin-scale patterns may be comparatively limited. The broad consistency of the estimated ES values with previous studies provides additional support for the reliability of the basin-scale results (Jiang et al., 2016; Fang et al., 2021; Xi et al., 2023; Zhang et al., 2024). For example, the estimated basin-wide carbon stock of 3.58 Pg C fell within the range of 3.15–4.91 Pg C reported by comparable InVEST-based studies (Xi et al., 2023; Zhang et al., 2024). Additionally, the IESI is a data-driven composite indicator and does not directly represent absolute ecological importance or policy preference. Potential temporal dependence was not explicitly incorporated into the Pearson analysis. Future basin-level correlations combine variation across scenarios and years, while pixel-level analyses, especially for 2030–2050, are based on relatively limited temporal observations. The identified trade-off and synergy patterns are therefore interpreted as exploratory Pearson associations, and statistical significance is treated cautiously instead of being taken as strong evidence of functional interactions among ESs. Future ES simulations and IESI optimization results mainly reflect scenario-dependent LULC changes under conservative considerations regarding cropland-retirement constraints.

To reduce the influence of uncertainty, several measures were adopted. (1) WY estimates were validated against in situ observations (Supplementary Figure SA2), whereas CS, NE, SC, and HQ were parameterized using multi-source data from remote sensing products, statistical records, and published studies. (2) Analyses were conducted at the pixel, prefecture-level, provincial, and watershed scales to improve spatial representativeness and capture scale-dependent patterns. (3) Five weighting schemes were compared to assess the robustness and consistency of IESI estimates and optimization results. (4) Three alternative LULC conversion pathways were evaluated, together with robustness tests for pixel allocation and weighting schemes. Future research could further incorporate hydrological, socioeconomic, and other practical constraints to improve the representation of ES interactions and the feasibility of LULC management.

6 Conclusion

References

• 1

  BagstadK. J.VillaF.BatkerD.Harrison-CoxJ.VoigtB.JohnsonG. W. (2014). From theoretical to actual ecosystem services: mapping beneficiaries and spatial flows in ecosystem service assessments. Ecol. Soc.19, 64. 10.5751/ES-06523-190264

  • CrossRef
  • Google Scholar

• 2

  BaiY.WongC. P.JiangB.HughesA. C.WangM.WangQ. (2018). Developing China’s ecological redline policy using ecosystem services assessments for land use planning. Nat. Commun.9, 3034. 10.1038/s41467-018-05306-1

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

• 3

  BhattaraiR.DuttaD. (2007). Estimation of soil erosion and sediment yield using GIS at catchment scale. Water Resour. Manag.21, 1635–1647. 10.1007/s11269-006-9118-z

  • CrossRef
  • Google Scholar
  • View reference in article

• 4

  ChangY. J.ZhuD. (2020). Urban water security of China’s municipalities: Comparison, features and challenges. J. Hydrol.587, 125023. 10.1016/j.jhydrol.2020.125023

  • CrossRef
  • Google Scholar
  • View reference in article

• 5

  ChenY.GuoF.WangJ.CaiW.WangC.WangK. (2020). Provincial and gridded population projection for China under shared socioeconomic pathways from 2010 to 2100. Sci. Data7, 83. 10.1038/s41597-020-0421-y

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

• 6

  ChenG.LiX.LiuX. (2022). Global land projection based on plant functional types with a 1-km resolution under socio-climatic scenarios. Sci. Data9, 125. 10.1038/s41597-022-01208-6

  • CrossRef
  • Google Scholar
  • View reference in article

• 7

  ChuenchumP.XuM.TangW. (2020). Predicted trends of soil erosion and sediment yield from future land use and climate change scenarios in the lancang-mekong river by using the modified RUSLE model. Int. Soil Water Conse.8, 213–227. 10.1016/j.iswcr.2020.06.006

  • CrossRef
  • Google Scholar
  • View reference in article

• 8

  CortinovisC.GenelettiD. (2019). A framework to explore the effects of urban planning decisions on regulating ecosystem services in cities. Ecosyst. Serv.38, 100946. 10.1016/j.ecoser.2019.100946

  • CrossRef
  • Google Scholar
  • View reference in article

• 9

  CostanzaR.d’ArgeR.De GrootR.FarberS.GrassoM.HannonB.et al (1997). The value of the world’s ecosystem services and natural capital. Nature387, 253–260. 10.1038/387253a0

  • CrossRef
  • Google Scholar
  • View reference in article

• 10

  DiakoulakiD.MavrotasG.PapayannakisL. (1995). Determining objective weights in multiple criteria problems: the critic method. Comput. Oper. Res.22, 763–770. 10.1016/0305-0548(94)00059-H

  • CrossRef
  • Google Scholar
  • View reference in article

• 11

  DroogersP.AllenR. G. (2002). Estimating reference evapotranspiration under inaccurate data conditions. Irrig. Drain. Syst.16, 33–45. 10.1023/A:1015508322413

  • CrossRef
  • Google Scholar
  • View reference in article

• 12

  EyringV.BonyS.MeehlG. A.SeniorC. A.StevensB.StoufferR. J.et al (2016). Overview of the coupled model intercomparison project phase 6 (CMIP6) experimental design and organization. Geosci. Model. Dev.9, 1937–1958. 10.5194/gmd-9-1937-2016

  • CrossRef
  • Google Scholar
  • View reference in article

• 13

  FangL.WangL.ChenW.SunJ.CaoQ.WangS.et al (2021). Identifying the impacts of natural and human factors on ecosystem service in the Yangtze and Yellow River Basins. J. Clean. Prod.314, 127995. 10.1016/j.jclepro.2021.127995

  • CrossRef
  • Google Scholar
  • View reference in article

• 14

  FilosoS.BezerraM. O.WeissK. C.PalmerM. A. (2017). Impacts of forest restoration on water yield: a systematic review. PLoS ONE12, e0183210. 10.1371/journal.pone.0183210

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

• 15

  GassmanP. W.ReyesM. R.GreenC. H.ArnoldJ. G. (2007). The soil and water assessment tool: historical development, applications, and future research directions. T. a. s. a. b. E.50, 1211–1250. 10.13031/2013.23637

  • CrossRef
  • Google Scholar
  • View reference in article

• 16

  GengW.LiY.ZhangP.YangD.JingW.RongT. (2022). Analyzing spatio-temporal changes and trade-offs/synergies among ecosystem services in the Yellow River Basin, China. Ecol. Indic.138, 108825. 10.1016/j.ecolind.2022.108825

  • CrossRef
  • Google Scholar
  • View reference in article

• 17

  JiX.SunY.GuoW.ZhaoC.LiK. (2023). Land use and habitat quality change in the Yellow River Basin: a perspective with different CMIP6-based scenarios and multiple scales. J. Environ. Manage.345, 118729. 10.1016/j.jenvman.2023.118729

  • CrossRef
  • Google Scholar
  • View reference in article

• 18

  JiangC.WangF.ZhangH.DongX. (2016). Quantifying changes in multiple ecosystem services during 2000-2012 on the loess Plateau, China, as a result of climate variability and ecological restoration. Ecol. Eng.97, 258–271. 10.1016/j.ecoleng.2016.10.030

  • CrossRef
  • Google Scholar
  • View reference in article

• 19

  JiangE.QuB.JiaJ.LiuC.WangY.LiD.et al (2023). Study on the coordination of flow and sediment transport, ecological environment, and crop production in the lower yellow river since 2000. River2, 326–335. 10.1002/rvr2.61

  • CrossRef
  • Google Scholar
  • View reference in article

• 20

  KosanicA.PetzoldJ. (2020). A systematic review of cultural ecosystem services and human wellbeing. Ecosyst. Serv.45, 101168. 10.1016/j.ecoser.2020.101168

  • CrossRef
  • Google Scholar
  • View reference in article

• 21

  LaterraP.OrúeM. E.BoomanG. C. (2012). Spatial complexity and ecosystem services in rural landscapes. Agr. Ecosyst. Environ.154, 56–67. 10.1016/j.agee.2011.05.013

  • CrossRef
  • Google Scholar
  • View reference in article

• 22

  LeSageJ.PaceR. K. (2009). Introduction to Spatial Econometrics. Chapman and Hall/CRC.

  • Google Scholar
  • View reference in article

• 23

  LiA.ZhangX. C.LiuB. (2021a). Effects of DEM resolutions on soil erosion prediction using Chinese soil loss equation. Geomorphology384, 107706. 10.1016/j.geomorph.2021.107706

  • CrossRef
  • Google Scholar
  • View reference in article

• 24

  LiG.JiangC.ZhangY.JiangG. (2021b). Whether land greening in different geomorphic units are beneficial to water yield in the Yellow River Basin?Ecol. Indic.120, 106926. 10.1016/j.ecolind.2020.106926

  • CrossRef
  • Google Scholar
  • View reference in article

• 25

  LiL.JiangE.YinH.WuK.DongG. (2022). Ultrashort-term responses of riparian vegetation restoration to adjacent cycles of ecological water conveyance scheduling in a hyperarid endorheic river basin. J. Environ. Manage.320, 115803. 10.1016/j.jenvman.2022.115803

  • CrossRef
  • Google Scholar
  • View reference in article

• 26

  LiL.JiangE.LiuC.XuC. Y. (2024). How does top-down water unified allocation and regulation decelerate water utilization? Insights from the Yellow River, China. J. Clean. Prod.447, 141402. 10.1016/j.jclepro.2024.141402

  • CrossRef
  • Google Scholar
  • View reference in article

• 27

  LiX.YuK.XuG.LiP.LiZ.ShiP.et al (2025). 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

• 28

  LiuB. Y.NearingM. A.RisseL. M. (1994). Slope gradient effects on soil loss for steep slopes. 37, 1835–1840. 10.13031/2013.28273

  • CrossRef
  • Google Scholar
  • View reference in article

• 29

  McCoolD. K.FosterG. R.MutchlerC. K.MeyerL. D. (1989). Revised slope length factor for the universal soil loss equation. Trans. ASAE. 32. 1571–1576. 10.13031/2013.31192

  • CrossRef
  • Google Scholar
  • View reference in article

• 30

  MerajG.SinghS. K.KangaS.IslamM. N. (2022). Modeling on comparison of ecosystem services concepts, tools, methods and their ecological-economic implications: a review. Model. Earth Syst. Env.8, 15–34. 10.1007/s40808-021-01131-6

  • CrossRef
  • Google Scholar
  • View reference in article

• 31

  RenardK.FosterG.WeesiesG.McCoolD.YoderD. (1997). “Predicting soil erosion by water: a guide to conservation planning with the revised universal soil loss equation (RUSLE),”, 703. U.S. Department of Agriculture, Agriculture Handbook, 404.

  • Google Scholar
  • View reference in article

• 32

  SchulpC. J.BurkhardB.MaesJ.Van VlietJ.VerburgP. H. (2014). Uncertainties in ecosystem service maps: a comparison on the European scale. PLoS ONE9, e109643. 10.1371/journal.pone.0109643

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

• 33

  ShannonC. E. (1948). A mathematical theory of communication. Bell Syst. Tech. J.27, 379–423. 10.1002/j.1538-7305.1948.tb01338.x

  • CrossRef
  • Google Scholar
  • View reference in article

• 34

  SharpR.TallisH.RickettsT.GuerryA.WoodS.Chaplin-KramerR.et al (2016). “InVEST user’s guide,” in The Natural Capital Project. Stanford University, University of Minnesota, the Nature Conservancy, and World Wildlife Fund.

  • Google Scholar
  • View reference in article

• 35

  ShenJ.ZhaoM.TanZ.ZhuL.GuoY.LiY.et al (2024). Ecosystem service trade-offs and synergies relationships and their driving factor analysis based on the Bayesian belief network: a case study of the Yellow River Basin. Ecol. Indic.163, 112070. 10.1016/j.ecolind.2024.112070

  • CrossRef
  • Google Scholar
  • View reference in article

• 36

  SherrouseB. C.ClementJ. M.SemmensD. J. (2011). A GIS application for assessing, mapping, and quantifying the social values of ecosystem services. Appl. Geogr.31, 748–760. 10.1016/j.apgeog.2010.08.002

  • CrossRef
  • Google Scholar

• 37

  WanL.YeX.LeeJ.LuX.ZhengL.WuK. (2015). Effects of urbanization on ecosystem service values in a mineral resource-based city. Habitat Int.46, 54–63. 10.1016/j.habitatint.2014.10.020

  • CrossRef
  • Google Scholar
  • View reference in article

• 38

  WangT.SunF. (2022). Global gridded GDP data set consistent with the shared socioeconomic pathways. Sci. Data9, 221. 10.1038/s41597-022-01300-x

  • CrossRef
  • Google Scholar
  • View reference in article

• 39

  WilliamsJ. R.ArnoldJ. G. (1997). A system of erosion-sediment yield models. Soil Technol.11, 43–55. 10.1016/S0933-3630(96)00114-6

  • CrossRef
  • Google Scholar

• 40

  WischmeierW. H.SmithD. D. (1978). “Predicting rainfall erosion losses - a guide to conservation planning,” in Agriculture Handbook No. 537. USDA/science and Education Administration US Govt. Washington, DC: Printing Office.

  • Google Scholar

• 41

  XiF.LinG.ZhaoY.LiX.ChenZ.CaoC. (2023). Land use optimization and carbon storage estimation in the Yellow River Basin, China. Sustainability15, 11278. 10.3390/su151411278

  • CrossRef
  • Google Scholar

• 42

  XuC.ZhangQ.YuQ.WangJ.WangF.QiuS.et al (2023). Effects of land use/cover change on carbon storage between 2000 and 2040 in the Yellow River Basin, China. Ecol. Indic.151, 110345. 10.1016/j.ecolind.2023.110345

  • CrossRef
  • Google Scholar

• 43

  YanF.ShangguanW.ZhangJ.HuB. (2020). Depth-to-bedrock map of China at a spatial resolution of 100 meters. Sci. Data7, 2. 10.1038/s41597-019-0345-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 44

  YangJ.XieB.ZhangD.TaoW. (2021). Climate and land use change impacts on water yield ecosystem service in the Yellow River Basin, China. Environ. Earth Sci.80, 72. 10.1007/s12665-020-09277-9

  • CrossRef
  • Google Scholar

• 45

  ZhangJ.YangJ.LiuP.LiuY.ZhengY.ShenX.et al (2024). Effects of land use/cover change on terrestrial carbon stocks in the Yellow River Basin of China from 2000 to 2030. Remote Sens.16, 1810. 10.3390/rs16101810

  • CrossRef
  • Google Scholar

• 46

  ZhouW.LiuG.PanJ.FengX. (2005). Distribution of available soil water capacity in China. J. Geogr. Sci.15, 3–12. 10.1007/BF02873101

  • CrossRef
  • Google Scholar