Trade-offs and synergies of ecosystem services in the yellow river basin: spatial dynamics and driving factors Authors: Weisi Ma, Zhiwei Zhao, Jinshuo Zhang, Mengyao Xun Published: 2026 ABSTRACT Understanding the interactions among ecosystem services and their relationships with dominant drivers are crucial for effective ecosystem management. Previous studies have mainly focused on trade-offs and synergies between ESs, overlooking no-effect interactions. The effect of bidirectional land cover transitions on ESs remained insufficiently quantified. Additionally, driving force analyses typically focused on the overall impacts, neglecting spatial heterogeneity. In the Yellow River Basin, we used the InVEST model and spatial allocation methods to assess the spatiotemporal distributions of carbon storage (CS), soil retention (SR), water yield (WY), food production (FP), habitat quality (HQ), and landscape aesthetics (LA) from 2003 to 2023. ES changes due to land cover transitions were evaluated using a Cross-Tabulation Matrix. An improved pairwise method analyzed the spatial diversity of interactions among ES pairs, while key drivers were identified using the optimized parameter-based geographical detector (OPGD) model. Additionally, bivariate local spatial autocorrelation was used to analyze the spatial relationships between ESs and their key drivers. The results indicated that FP and WY were higher in the south and east, SR and CS were more concentrated in the center and south, while HQ and LA were predominantly distributed in the west. Seven ES pairs exhibited predominant synergies, including HQ-SR, HQ-CS, HQ-LA, SR-FP, SR-WY, CS-LA, and FP-WY, whereas five involving HQ-FP, HQ-WY, WY-LA, FP-LA, and SR-LA showed trade-offs. For CS-FP, CS-WY, and SR-CS pairs, synergy, trade-off, and no-effect relationships occupied comparable spatial area proportions. CS and HQ are primarily shaped by landscape configuration, SR by slope, vegetation, and precipitation, WY by climate and topography, and FP by socioeconomic intensity and elevation. Spatially, HH areas linking CS, SR, WY, HQ, and LA to their main drivers cluster in the upstream Loess Plateau gully region, the Sichuan Tibet alpine canyon region, and the southern Tibetan Plateau, whereas the northwestern desert region and the middle and lower reaches are dominated by LL, HL, LH, or not significant patterns. In contrast, HH hotspots for FP and its main drivers are concentrated in the middle and lower reaches, especially in the agropastoral transition zone and irrigated plains. These findings suggest that regulating key drivers, balancing vegetation restoration with regional water capacity, and protecting wetlands and other sensitive areas can help manage tradeoffs, enhance synergies, and support sustainable regional development.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-026-43436-5. Introduction Ecosystem services (ESs) are the benefits and conditions that natural environments provide to support human survival, including supporting, provisioning, regulating, and cultural services1. These services play a crucial role in sustaining ecosystem functionality, promoting sustainable development, and ensuring societal progress and human well-being2. However, urban and industrial expansion, coupled with poorly regulated land development, have contributed to deforestation, wetland degradation, and biodiversity loss. Studies show that 15 out of 24 key ecosystem services have been degraded over the past half-century, severely impairing ecosystems’ self-regulating capacities and their ability to deliver essential services3–6. Faced with these increasingly escalating challenges, Goal 15 outlined in the 2030 UN Agenda on Sustainable Development emphasizes the urgency of protecting and restoring terrestrial ecosystems7–9. Assessing ESs is a fundamental prerequisite that supports sustainable management of ecosystems while contributing to the effective promotion of human welfare10. The diverse nature of ES categories, the spatial heterogeneity of ES distributions, and the selective utilization by humans, have given rise to complex interactions among various services, manifesting as trade-offs, where the gain of one service comes at the expense of another, and synergies, where multiple services benefit simultaneously11–14. A meta-analysis by Howe et al. (2014) of 1,324 cases revealed trade-offs to be three times more common than synergies15. Similarly, Turkelboom et al. (2018) analyzing 24 case studies, highlighted the significant complexity of ES trade-offs, especially between provisioning and cultural services, which exceeds what has been previously recognized in the literature16. This complexity suggests that modifying ecosystems to boost one service inevitably impacts the delivery of others8. Hence, accurately capturing the spatio-temporal trajectories of ESs and detecting the patterns that underlie trade-offs and synergies is critical for effective ecosystem management17,18. To this end, various analytical methods have been employed to explore the interactions among ES. While correlation analysis remains a commonly used approach to quantify overall trade-offs and synergies4,19, it lacks the capacity to reveal their spatial characteristics. To address this, geospatial approaches such as bivariate spatial autocorrelation analysis18, constraint line analysis3, and spatial overlay approaches20 have been utilized to analyze the spatial configurations of ES interactions. Nevertheless, these methods frequently simplify the relationships into binary categories of “trade-off” and “synergy,” neglecting the possibility of “no-effect” interactions21,22, where the change in one service does not result in any significant impact on the other22. Therefore, we have used an improved method to explore potentially independent relationships22. The identification of the impacts of different driving forces on ESs is crucial for informing sustainable development strategies. Since the Anthropocene began, rapid urban expansion has increasingly encroached upon natural resources and ecological spaces, heightening human-environment conflicts23,24. Changes in land use, have significantly reshaped human- nature interactions, profoundly altering ecosystem structure and function25. However, existing research on the effects of land use change on ESs have predominantly focused on assessing the contributions of land use type transitions to service supply26. For instance, afforestation policies have promoted the expansion of forests and water bodies, enhancing the provision of regulating services19,27. Additionally, statistical methods have been used to evaluate how the conversion of one or two land use types at a single time point interacts with ES changes19. This approach, however, fails to capture the dynamic effects of bidirectional land-cover transitions and stable land-cover areas on ESs. Beyond land use change, key drivers of ESs include biophysical factors like topography and climate, as well as socioeconomic conditions such as population and economic development28,29. Natural factors, as inherent ecosystem attributes, affect service supply by altering ecosystem structures and functions, thereby influencing services such as soil retention, water regulation, and wind erosion control30. Socio-economic factors modify land use and landscape patterns, affect animal population dynamics, and consequently lead to alterations in ecosystem services11. However, many current studies on driving factors focus primarily on their overall impact on ecosystem services, with limited consideration of spatial heterogeneity. Therefore, this study first employs the optimized parameter-based geographical detector (OPGD) to pinpoint the primary drivers influencing each ES. OPGD represents an advanced approach that improves spatial heterogeneity analysis by automatically optimizing spatial data discretization, strata breakpoints, and scale parameters19,31. Unlike traditional Geographical Detector methods, which rely on subjective parameter choices, OPGD enhances the extraction of geographical features from explanatory variables, leading to more accurate and robust identification of dominant drivers across different spatial data types32,33. Subsequently, we examine the spatially heterogeneous associations between ESs and their dominant drivers using bivariate local spatial autocorrelation, which identifies local co-location clusters, spatial mismatches, and non-significant associations. The Yellow River Basin (YRB) serves as a critical ecological security barrier and a core region for economic development in China, bearing dual responsibilities for ecological conservation and high-quality development at the national strategic level34. Located in an arid and semi-arid climate zone, the YRB exhibits pronounced ecological vulnerability, facing long-term pressures from climate change and human activities13. While restoration measures such as microtopography modification, the Grain for Green Program, and ecological reserve construction have partially improved the basin’s ecological functions35,36, severe ecological challenges persist across the upper, middle, and lower reaches, and the underlying fragility of the ecological foundation remains unresolved23. Aligned with the YRB’s Ecological Protection and High-quality Development Plan, this study quantified and mapped six key ESs in the YRB, namely water yield, carbon storage, habitat quality, soil retention, food production, and landscape aesthetics. Fifteen potential influencing factors across four domains: human activities, climate, topography, and landscape patterns were used to construct an integrated indicator system under the framework of natural, social, economic, and landscape conditions21,22. The primary objectives of this research are: (1) to analyze the spatiotemporal characteristics of six ESs in the YRB for the years 2003, 2013, and 2023 using the InVEST 3.13.0 and ArcGIS 10.8.2; (2) to assess the responses of these ESs to bidirectional land use and land cover transitions from 2003 to 2023; (3) to elucidate the spatiotemporal dynamics of trade-offs, no-effects, and synergies among ESs; (4) to reveal the dominant drivers and their spatially heterogeneous associations with ESs by integrating OPGD and bivariate local spatial autocorrelation. Accordingly, we address three questions: how have ES patterns changed over the past two decades? How do bidirectional LULC transitions reshape ES interactions, including trade-off, synergy, and no-effect relationships? Which drivers dominate each ES, and where do local co-location clusters and spatial mismatches between ESs and their drivers emerge? This provides a spatially explicit foundation for differentiated ecological management and watershed-scale planning in the YRB. Data sources and methods Research framework This study constructs a comprehensive framework to explore the trade-offs and synergies among ecosystem services (ESs) and their driving mechanisms in the Yellow River Basin (Fig. 1). The research consists of five main components. The first part reviews domestic and international studies on ecosystem service dynamics, identifies research gaps, and clarifies the study’s objectives, including quantifying multiple ESs, revealing their interrelationships, and determining the dominant driving factors and their spatial heterogeneity. The second part integrates multi-source environmental data, such as land use, vegetation, climate, topography, and soil, to quantify six key ESs: carbon storage (CS), soil retention (SR), water yield (WY), food production (FP), habitat quality (HQ), and landscape Aesthetics (LA). The third part employs the Optimal Parameter Geodetector (OPGD) model to identify major biophysical, climatic, landscape, and anthropogenic drivers influencing ecosystem service changes. The fourth part analyzes the complex spatial and temporal relationships between driving factors and ecosystem services. The fifth part synthesizes the findings to reveal the principal spatial patterns of trade-offs and synergies and identifies key regions requiring focused management, providing theoretical and methodological support for optimizing land-use patterns and enhancing ecosystem resilience. Fig. 1Research framework diagram. The map was created using ArcGIS Desktop V. 10.8 and the URL link is http://www.esri.com/sofware/arcgis. Research framework diagram. The map was created using ArcGIS Desktop V. 10.8 and the URL link is http://www.esri.com/sofware/arcgis. Study area The Yellow River, ranking as China’s second-longest river, extends for 5,464 km and drains an area of approximately 795,000 km². It traverses northern China through three principal geomorphic terraces, namely the eastern, central, and western terraces, and passing through nine provinces, including Qinghai, Sichuan, Gansu, Ningxia, Inner Mongolia, Shanxi, Shaanxi, Henan, and Shandong (Fig. 2). The basin exhibits pronounced topographic variation, with elevation gradually declining from west to east. Grasslands and barren lands are the dominant land use types. Annual precipitation ranges from 116 mm to 1,038 mm, and the mean annual temperature varies from − 13 °C to 15 °C. The basin supports more than 20 provincial-level and 29 national-level nature reserves, which provide a wide range of ecosystem services. Urbanization and industrialization have driven significant socio-economic transformations over the past 20 years. The population increased by 14%, rising from 280 million to 320 million, while GDP grew nearly tenfold, from 2.01 trillion RMB to 20.83 trillion RMB. The built-up area expanded from 41,407.5 km² to 61,855.7 km²13. These trends highlight the urgent need to establish sustainable development strategies that integrate ecological conservation, economic growth, and the support of livelihoods and human well-being in the YRB. Fig. 2Study area. (a) Location of the study area. (b) Elevation and watershed boundaries (A. Qinghai-Tibet Plateau; B. Qaidam Basin; C. Qilian Mountains; D. Alxa Plateau; E. Inner Mongolia Plateau; F. Yin Mountains; G. Hetao Plain; H. Hobq Desert; I. Ordos Plateau; J. Mu Us Sandy Land; K. Loess Plateau; L. Qinling Mountains; M. Guanzhong Plain; N. Fenhe Valley; O. Taihang Mountains; P: Huang-Huai-Hai Plain; Q. Sichuan-Tibet alpine canyon region. (c) land use/land cover in 2003. (d) land use/land cover in 2013. (e) land use/land cover in 2023. The map was created using ArcGIS Desktop V. 10.8 and the URL link is http://www.esri.com/sofware/arcgis. Study area. (a) Location of the study area. (b) Elevation and watershed boundaries (A. Qinghai-Tibet Plateau; B. Qaidam Basin; C. Qilian Mountains; D. Alxa Plateau; E. Inner Mongolia Plateau; F. Yin Mountains; G. Hetao Plain; H. Hobq Desert; I. Ordos Plateau; J. Mu Us Sandy Land; K. Loess Plateau; L. Qinling Mountains; M. Guanzhong Plain; N. Fenhe Valley; O. Taihang Mountains; P: Huang-Huai-Hai Plain; Q. Sichuan-Tibet alpine canyon region. (c) land use/land cover in 2003. (d) land use/land cover in 2013. (e) land use/land cover in 2023. The map was created using ArcGIS Desktop V. 10.8 and the URL link is http://www.esri.com/sofware/arcgis. Data sources and preparation Multisource datasets were used to spatially evaluate ESs and spatial determinants (Table 1). To ensure consistency in data resolution and projection, all raster data were resampled to a 1 km×1 km resolution in ArcGIS using nearest neighbor method and projected to the KGCS_Krasovsky_1940 coordinate system. Table 1Summary of the primary data.Data typeApplicationSpatial resolutionData sourceLand use/land coverHQ, CS, SR, FP, WY, LARasterZenodo Repository (https://zenodo.org/record/12779975)PrecipitationSR, WY, SEDRasterNational Tibetan Plateau Science Data Center (https://data.tpdc.ac.cn/)ElevationSR, SED,RasterGeospatial data cloud (http://www.gscloud.cn/)WatershedSR, WY,polygonResource and Environment Science and Data Center (www.resdc.cn)Root depth, soil texture, and organic carbon contentSRRasterThe dataset is provided by National Cryosphere Desert Data Center. (www.ncdc.ac.cn)Normalized difference vegetation index (NDVI)FP, SED,RasterNasa Earth Data (https://www.earthdata.nasa.gov)Potential evapotranspirationWY, SEDRasterNational Tibetan Plateau Science Data Center (https://data.tpdc.ac.cn/)GDPSEDRasterResource and Environment Science and Data Center (www.resdc.cn)Population density (POP)SEDRasterResource and Environment Science and Data Center (www.resdc.cn)Nighttime lighting index (NLI)SEDRasterResource and Environment Science and Data Center (www.resdc.cn)Land surface temperature (LST)SEDRasterNational Tibetan Plateau Science Data Center (https://data.tpdc.ac.cn/)Food yieldFPSpreadsheetStatistical Yearbook (www.stats.gov.cn)Distance from the settlement (SD)SEDRasterAfter extracting the settlement from LULC, the Euclidean distance is calculated using ArcGISCONTAG, LSI, LPI, AREAMN, GYRSEDRasterInput LULC data into Fragstats (Version 4.2) software for measurementNote: HQ indicates habitat quality; CS indicates carbon storage; SR indicates soil retention; FP indicates food production; WY indicates water yield; LA indicates landscape aesthetics; SED indicates social-ecological drivers; CONTAG indicates contagion index, LSI indicates landscape shape index, LPI indicates largest patch index, AREAMN indicates mean patch area, GYR indicates radius of gyration. Table 1Summary of the primary data.Data typeApplicationSpatial resolutionData sourceLand use/land coverHQ, CS, SR, FP, WY, LARasterZenodo Repository (https://zenodo.org/record/12779975)PrecipitationSR, WY, SEDRasterNational Tibetan Plateau Science Data Center (https://data.tpdc.ac.cn/)ElevationSR, SED,RasterGeospatial data cloud (http://www.gscloud.cn/)WatershedSR, WY,polygonResource and Environment Science and Data Center (www.resdc.cn)Root depth, soil texture, and organic carbon contentSRRasterThe dataset is provided by National Cryosphere Desert Data Center. (www.ncdc.ac.cn)Normalized difference vegetation index (NDVI)FP, SED,RasterNasa Earth Data (https://www.earthdata.nasa.gov)Potential evapotranspirationWY, SEDRasterNational Tibetan Plateau Science Data Center (https://data.tpdc.ac.cn/)GDPSEDRasterResource and Environment Science and Data Center (www.resdc.cn)Population density (POP)SEDRasterResource and Environment Science and Data Center (www.resdc.cn)Nighttime lighting index (NLI)SEDRasterResource and Environment Science and Data Center (www.resdc.cn)Land surface temperature (LST)SEDRasterNational Tibetan Plateau Science Data Center (https://data.tpdc.ac.cn/)Food yieldFPSpreadsheetStatistical Yearbook (www.stats.gov.cn)Distance from the settlement (SD)SEDRasterAfter extracting the settlement from LULC, the Euclidean distance is calculated using ArcGISCONTAG, LSI, LPI, AREAMN, GYRSEDRasterInput LULC data into Fragstats (Version 4.2) software for measurementNote: HQ indicates habitat quality; CS indicates carbon storage; SR indicates soil retention; FP indicates food production; WY indicates water yield; LA indicates landscape aesthetics; SED indicates social-ecological drivers; CONTAG indicates contagion index, LSI indicates landscape shape index, LPI indicates largest patch index, AREAMN indicates mean patch area, GYR indicates radius of gyration. Summary of the primary data. Note: HQ indicates habitat quality; CS indicates carbon storage; SR indicates soil retention; FP indicates food production; WY indicates water yield; LA indicates landscape aesthetics; SED indicates social-ecological drivers; CONTAG indicates contagion index, LSI indicates landscape shape index, LPI indicates largest patch index, AREAMN indicates mean patch area, GYR indicates radius of gyration. Ecosystem services assessment Ecosystem services were classified according to the Millennium Ecosystem Assessment framework into four types: provisioning, regulating, supporting, and cultural1, this study identified and assessed six representative services in the YRB (Table 2). The selection of these services was guided by the availability of data, the feasibility of quantitative analysis, and their suitability for spatial assessment, while also considering the ecological features of the region and local needs. Food production (FP), a key provisioning service, supports human life and social activities. Given the scarcity and uneven distribution of water resources, water yield (WY) was included as a critical indicator. Carbon storage (CS) and soil retention (SR) were considered as major regulating services, essential for maintaining ecosystem stability by reducing carbon emissions and preventing soil erosion. Habitat quality (HQ), a supporting service, plays a vital role in biodiversity protection and spatial planning. Landscape aesthetics (LA), representing cultural service, reflects the quality of human life and contributes to sustainable land-use planning, which justified its inclusion in the analysis (Supplementary Figure S1-S6). Table 2Overview of ecosystem services assessed.CategoryEcosystem serviceAbbreviationMethodsDescriptionSupportingserviceHabitat qualityHQInVEST modelThe ability of ecosystems to provide conditions suitable for the persistence of individualsand populationsRegulatingserviceCarbon storageCSInVEST modelThe amount of carbon stored by terrestrialecosystemsSoil retentionSRInVEST modelThe soil conservation capability of ecosystemsProvisioningserviceFood productionFPNormalized differencevegetation index (NDVl)The yield of staple food cropsWater yieldWYInVEST modelThe yield of annual waterCulturalserviceLandscape aestheticsLALandscape metrics-based assessmentAn index compounded by naturalness,landscape diversity, and landscape uniquenessNote: HQ, dimensionless index (0–1); CS, t·ha⁻1; SR t·ha⁻1; FP, t·ha⁻1; WY, mm; LA, dimensionless index (0–1). Table 2Overview of ecosystem services assessed.CategoryEcosystem serviceAbbreviationMethodsDescriptionSupportingserviceHabitat qualityHQInVEST modelThe ability of ecosystems to provide conditions suitable for the persistence of individualsand populationsRegulatingserviceCarbon storageCSInVEST modelThe amount of carbon stored by terrestrialecosystemsSoil retentionSRInVEST modelThe soil conservation capability of ecosystemsProvisioningserviceFood productionFPNormalized differencevegetation index (NDVl)The yield of staple food cropsWater yieldWYInVEST modelThe yield of annual waterCulturalserviceLandscape aestheticsLALandscape metrics-based assessmentAn index compounded by naturalness,landscape diversity, and landscape uniquenessNote: HQ, dimensionless index (0–1); CS, t·ha⁻1; SR t·ha⁻1; FP, t·ha⁻1; WY, mm; LA, dimensionless index (0–1). Overview of ecosystem services assessed. Supporting service The ability of ecosystems to provide conditions suitable for the persistence of individuals and populations Regulating service The amount of carbon stored by terrestrial ecosystems Provisioning service Normalized difference vegetation index (NDVl) Cultural service An index compounded by naturalness, landscape diversity, and landscape uniqueness Note: HQ, dimensionless index (0–1); CS, t·ha⁻1; SR t·ha⁻1; FP, t·ha⁻1; WY, mm; LA, dimensionless index (0–1). The response of ESs to LULC transition To quantify the ESs changes per inward-outward LULC transitions, a Cross-Tabulation Matrix was constructed in this study37, which was calculated by ArcGIS 10.7 as follows:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${ES}_{ij}=\left(\begin{array}{ccc}{ES}_{ii}&\cdots&{ES}_{ij}\\\vdots&\ddots&\vdots\\{ES}_{ji}&\cdots&{ES}_{jj}\end{array}\right)$$\end{document} \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${ES}_{ij}$$\end{document} is the value of ESs on pixel where LULC type i in 2003 was transferred to type j in 2023. The diagonal elements (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${ES}_{ii}$$\end{document} to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${ES}_{jj}$$\end{document}) represent the values of ESs on pixel with persistence LULC (no change between 2003 and 2023). \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${ES}_{ij}$$\end{document} is the value of ESs on pixel where LULC type i in 2003 was transferred to type j in 2023. The diagonal elements (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${ES}_{ii}$$\end{document} to \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${ES}_{jj}$$\end{document}) represent the values of ESs on pixel with persistence LULC (no change between 2003 and 2023). Synergies and trade-offs analysis The spatial non-stationarity inherent in the relationships among ES pairs and their driving factors is essential for the governance of ecosystem services, a topic that has become a central focus in recent research5,8,18,22. Studies have examined the variations in ESs pairs, classified as synergies (≥ 0) and tradeoff (< 0)28. This approach failed to incorporate the consideration of independent (no effect) relationship. Hu et al. (2023) advanced this methodology by categorizing the result into three categories: trade-offs (< 0), independence (= 0), and synergy (> 0). Within this framework, independent relationships encompass no-effect relationships; in addition, overlapping stable pixels in each pair should be regarded as part of the synergy21. Therefore, the detailed pairwise interactions among ESs are presented as follows:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left[\varDelta{ES}_{i}\mathrm{*}\varDelta{ES}_{j}<0\right]\to\:Tradeoff$$\end{document}3\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left[\frac{\varDelta{ES}_{i}=0and\varDelta{ES}_{j}\ne0}{\varDelta{ES}_{i}\ne0and\varDelta{ES}_{j}=0}\right]\to\:No-effect$$\end{document}4\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\left[\left(\varDelta{ES}_{i}*\varDelta{ES}_{j}>0\right)+overlapingpersistencepixels\right]\to\:Synergy$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varDelta{ES}_{i}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\varDelta{ES}_{j}$$\end{document}, represent the changes in \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${ES}_{i}$$\end{document} and j from 2003 to 2023, respectively. Analytical methods for optimal parameters-based geographical detector model Optimal Parameter Determination The geodetector serves to identify heterogeneity in spatial stratification and reveal potential drivers. Detecting spatial stratification entails creating a consistent framework of discretized spatial data layers38. One advantage of this method is that it can account for both linear and nonlinear correlations32. The validity of the analysis is assessed using the q-value, with a higher q-value indicating better partitioning39. We applied multiple classification methods (natural breaks, quantile, equal, geometric, and standard deviation) from the R-based GD package across a range of 5–12 levels. The parameter combination (classification method and number of classes) with the highest q-value is selected for discretization. The driver stratification scheme defined by the highest q-value possesses the greatest explanatory power32. Geodetector method Factor detection in the geographical detector identified the factors driving spatial heterogeneity in ESs across the YRB, based on optimal parameter selection. Factor detection analyzes the explanatory power of individual factors on ESs, while interaction detection reveals how interactions between two variables affect ESs. These interactions may include non-linear weakening, single-factor weakening, two-factor enhancement, independence, and non-linear enhancement39. This can be expressed as follows:5\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$q=1-\frac{\sum_{h=1}^{L}{N}_{h}{\sigma}_{h}^{2}}{N{\sigma}^{2}}=1-\frac{SSW}{SST}$$\end{document} where q represents the factor’s explanatory power for ES spatial heterogeneity, ranging from 0 to 1, the closer to 1, with higher values indicating stronger explanatory power. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$N$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${N}_{h}$$\end{document} refer to the total number of cells in the entire region and in stratum \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h$$\end{document}, respectively, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$L$$\end{document} is the stratification of the factor, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h$$\end{document} denotes the strata of explanatory variables. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma}^{2}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma}_{h}^{2}$$\end{document} represent the variance of Y-values for the whole region and stratum \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$h$$\end{document}, respectively. SSW and SST denote the within-stratum and total variances, respectively. Selection and treatment of drivers ES spatial heterogeneity results from the convergence of natural drivers, encompassing climatic variables, biophysical attributes, and landscape configuration, as well as human endeavors. Consequently, by integrating previous studies with the specific conditions of the YRB, we selected 15 driving factors from four categories: climate, biophysical attributes, landscape configuration, and human activities (Table 3). The variance inflation factor (VIF) method was used to test for multicollinearity among the influencing factors. The results showed that all variables had a VIF less than 5 (Supplementary Table S6), indicating their suitability for subsequent analysis. All factors were rasterized in ArcGIS using the projected coordinate system KGCS_Krasovsky_1940 at a 1 km × 1 km resolution. Table 3Driver category selection and data processing.CategoryDriving factorsAbbreviationNatural factorsClimate factorsAnnual precipitationPrePotential evapotranspirationPetLand surface temperatureLstBiophysical factorsElevation/SlopeSlopeNormalized difference vegetation indexNdviLandscape configurationContagion indexContagLandscape shape indexLsiLargest patch indexLpiMean patch areaAreamnRadius of gyrationGyrSocio-economicfactorsAnthropogenic factorsGross domestic productGdpPopulation densityPopDistance from the settlementSdNighttime lighting intensityNli Table 3Driver category selection and data processing.CategoryDriving factorsAbbreviationNatural factorsClimate factorsAnnual precipitationPrePotential evapotranspirationPetLand surface temperatureLstBiophysical factorsElevation/SlopeSlopeNormalized difference vegetation indexNdviLandscape configurationContagion indexContagLandscape shape indexLsiLargest patch indexLpiMean patch areaAreamnRadius of gyrationGyrSocio-economicfactorsAnthropogenic factorsGross domestic productGdpPopulation densityPopDistance from the settlementSdNighttime lighting intensityNli Driver category selection and data processing. Socio-economic factors Bivariate local indicator of spatial association (LISA) To reveal the spatial heterogeneity in the relationships between ESs and driving factors, we employed the bivariate local Moran’s I (bivariate LISA) to quantify local spatial associations between each ES and its driving factors. Specifically, this approach assesses the local association between the standardized value of one variable and the average of the neighboring values for another variable40, thereby identifying local co-location patterns and spatial mismatches. A first-order queen contiguity matrix was used to construct the spatial weights, which were row-standardized. Statistical significance was assessed using 999 random permutations, with a significance level of p < 0.05. All analyses were conducted in R using the sf, spdep, and dplyr packages for computation and data processing.6\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${I}_{i}^{ES,D}={z}_{ES}\left(i\right){\sum}_{j=1}^{N}{w}_{ij}{z}_{D}\left(j\right)$$\end{document} where \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${z}_{ES}\left(i\right)=\frac{E{S}_{i}-\overline{E}S}{{\sigma}_{ES}}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${z}_{D}\left(j\right)=\frac{{D}_{j}-\overline{D}}{{\sigma}_{D}}$$\end{document}, \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$E{S}_{i}$$\end{document} represents the value of the ecosystem service at spatial unit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i$$\end{document}, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${D}_{j}$$\end{document} represents the value of the driving factor at neighboring unit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j$$\end{document}. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{E}S$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overline{D}$$\end{document} are the mean values of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ES$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$D$$\end{document} across the study area, and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma}_{ES}$$\end{document} and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma}_{D}$$\end{document} are their standard deviations. \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${w}_{ij}$$\end{document} is an element of the spatial weight matrix \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$W$$\end{document}, which defines the spatial relationship between unit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$i$$\end{document}and its neighboring unit \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$j$$\end{document}. Furthermore, based on the sign combination of \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${z}_{ES}\left(i\right)$$\end{document} and the spatial lag of the driving factor, four types of local spatial association are identified: HH, LL, HL, and LH. Specifically, HH indicates high values of both \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ES$$\end{document}and its neighboring driving factors (High–High), while LL indicates low values of both \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ES$$\end{document}and its neighboring driving factors. HL indicates a high \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ES$$\end{document} but low neighboring driving factors, and LH indicates a low \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ES$$\end{document} but high neighboring driving factors. For example, when a unit is classified as HL, it means the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ES$$\end{document}in that unit is higher than the average level of the study area, while the neighboring driving factors are lower than the average, reflecting a spatial mismatch between \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ES$$\end{document}and its driving factors. Results and analyses Ecosystem services dynamics and their response to land cover transitions This study quantified grid-scale spatiotemporal metrics of six ESs within the YRB (Fig. 3). HQ was relatively high in the baseline year of 2003 but showed a declining trend thereafter, with a spatial gradient that decreased from the southeast to the northwest. High values clustered in areas with limited human disturbance, such as forests and grasslands, while low values were observed in urbanized and desert regions, particularly in the Guanzhong Plain, Huang-Huai-Hai Plain, and Alxa Plateau (Fig. 3). FP demonstrated a consistent increasing trend, with spatially higher levels distributed in the southern and eastern regions of the basin. Notably, regions upstream of Lanzhou and the Henan-Shandong section both saw a phased shift from increase to decline over the two decades. CS and SR showed highly similar spatial distributions, with their high values areas primarily located in densely vegetated regions dominated by forests and grasslands, including the Qilian Mountains, Bayan Har Mountains, and Qinling Mountains. Regions that experienced a decline in CS and SR from 2003 to 2013 generally saw recovery during 2013–2023, and vice versa, reflecting dynamic temporal fluctuations. WY exhibited a gradient distribution pattern, characterized by a gradual decrease from southern mountainous regions and eastern coastal areas toward the arid inland deserts of the northwest. However, a substantial reduction in WY occurred in the middle and lower reaches over the study period, primarily attributed to decreased precipitation and increased water demand. In contrast, some upstream areas that declined in the first decade exhibited recovery in the second, while previously stable regions showed signs of degradation. LA were relatively high in 2003, particularly in western regions with complex terrain and well-preserved natural landscapes. Nonetheless, LA values showed a consistent decline, especially in the middle and eastern parts of the basin, driven by urban expansion and simplified agricultural landscapes, indicating a gradual reduction in landscape heterogeneity. Fig. 3Spatial variation of ecosystem services and temporal dynamics. CS: carbon storage; SR: soil retention; WY: water yield; FP: food production; HQ: habitat quality; LA: landscape aesthetics. The map was created using ArcGIS Desktop V. 10.8 and the URL link is http://www.esri.com/sofware/arcgis. Spatial variation of ecosystem services and temporal dynamics. CS: carbon storage; SR: soil retention; WY: water yield; FP: food production; HQ: habitat quality; LA: landscape aesthetics. The map was created using ArcGIS Desktop V. 10.8 and the URL link is http://www.esri.com/sofware/arcgis. Land cover changes had significantly heterogeneous effects on ESs (Fig. 4). SR increased most in areas transitioning from shrubland to forest, while it declined in regions where grassland converted to bare land. Transformations of cropland, forest, grassland, and shrubland into settlements or bare land generally enhanced WY, with the greatest increase observed in forest-to-bare land conversions. In contrast, WY declined most in areas where cropland and grassland were converted to wetlands. CS increased in areas where forest and shrubland expanded, while it decreased where cropland, forest, grassland, and shrubland were replaced by water bodies, settlements, or bare land. HQ improved most in areas where cropland transitioned to grassland and water bodies, but declined in grassland and shrubland areas converted to settlements. LA decreased in regions where forest, grassland, or shrubland were converted to settlements or cropland, but improved in areas where cropland transitioned to forest, water bodies, or bare land. Although stable land cover areas did not undergo type transitions, ESs in these regions still experienced changes, albeit less pronounced. For example, HQ, SR, and WY declined in regions continuously covered by forest, shrubland, or grassland, while CS continued to increase. FP showed an upward trend in stable cropland areas, and LA improved in areas dominated by water bodies, barren land, and forest due to reduced human disturbance, thereby enhancing landscape aesthetic value. Fig. 4Mean change in ESs per transitions of major land cover types (2003–2023). The x-axis represents discrete categories of land-use transitions, and the connecting lines are used only to indicate general trends. Mean change in ESs per transitions of major land cover types (2003–2023). The x-axis represents discrete categories of land-use transitions, and the connecting lines are used only to indicate general trends. Spatial Interrelationships between ecosystem services The interactions between pairs of ESs from 2003 to 2023 showed spatial variability (Fig. 5). In areas such as the Ulan Buh Desert, Tengger Desert, northern Yin Mountains, western Qilian Mountains, and high-altitude zones of the Qinghai-Tibet Plateau, most ES pairs exhibited no-effect relationships due to the cold alpine climate and arid conditions. These harsh environmental factors, including limited water availability, low temperatures, and sparse vegetation, reduce the potential for interdependencies between different ESs. The persistent land cover types in these regions further limit the interaction between ESs. Synergistic relationships were more widespread than trade-offs, particularly for HQ-SR, HQ-CS, HQ-LA, SR-FP, SR-WY, CS-LA, and FP-WY. Notably, FP-WY, SR-FP, and SR-WY exhibited the highest synergistic areas, covering 73.74%, 71.13%, and 65.74% of the basin, respectively. These synergies were predominantly observed in the central and western subregions of the YRB, where ecological restoration efforts such as terracing, reforestation, and grassland restoration have enhanced water conservation, soil quality, and agricultural productivity. Synergies in HQ-SR, HQ-CS, HQ-LA, and CS-LA were mainly driven by terrain variability, precipitation, and vegetation coverage. Trade-offs were prominent in the HQ-FP pair, resulting from conversions between cropland and forest, grassland, or shrubland. Declining trends in WY were observed in the west, while increasing trends occurred in the south, central, and east, reflecting spatiotemporal precipitation patterns. The expansion of built-up and cropland areas contributed to habitat degradation in newly developed zones, leading to opposing trends between HQ and WY, especially in southern, eastern, and western regions of the basin, causing dominant trade-offs between WY and HQ. In the western upstream, water scarcity and complex topography led to trade-offs in WY-LA, FP-LA, and SR-LA, with pronounced conflicts in WY-LA and SR-LA. In the central region, high agricultural intensity resulted in significant FP-LA and SR-LA trade-offs. In the downstream region, characterized by intense human activity, WY-LA and FP-LA trade-offs were dominant. For CS-FP, CS-WY, and SR-CS pairs, synergies and trade-offs occupied equal spatial proportions. Urban expansion in the central and lower reaches caused vegetation loss, weakening carbon sequestration. However, in the upstream, grassland-based animal husbandry fostered CS-FP synergies. Divergent trends between CS and WY in the central-eastern region led to CS-WY trade-offs, while stronger synergies were observed in the Loess Plateau Gully Region and the Shandong Peninsula Urban Agglomeration. Ecological restoration projects in the upstream enhanced vegetation cover, reinforcing CS-SR synergies, while unregulated expansion in the middle and lower reaches led to CS-SR trade-offs. Fig. 5Spatial patterns and area proportions of pairwise relationship between the change in ecosystem services. CS: carbon storage; SR: soil retention; WY: water yield; FP: food production; HQ: habitat quality; LA: landscape aesthetics. The map was created using ArcGIS Desktop V. 10.8 and the URL link is http://www.esri.com/sofware/arcgis. Spatial patterns and area proportions of pairwise relationship between the change in ecosystem services. CS: carbon storage; SR: soil retention; WY: water yield; FP: food production; HQ: habitat quality; LA: landscape aesthetics. The map was created using ArcGIS Desktop V. 10.8 and the URL link is http://www.esri.com/sofware/arcgis. Comprehensive analysis of potential driving factors for ESs Key drivers of ES spatial dynamics All 15 drivers significantly affected ES spatiotemporal dynamics in the YRB (all p < 0.05), as determined by optimal parameter-based geographical detector analysis (Fig. 6). To clarify the determinants of spatial heterogeneity in ESs, we divided them into two categories based on their leading drivers: those dominated by natural factors and those influenced by a mix of natural and socioeconomic factors. The first occurs when both main drivers are natural, while the second arises when one is natural and the other socioeconomic. CS, WY, HQ, and LA were more sensitive to natural factors. For CS, the pivotal drivers from 2003 to 2023 included Lsi, Lpi, and Gyr, with their explanatory power reflected in q-values of 0.29, 0.25, and 0.16, respectively. WY was primarily influenced by Pre, elevation, and Pet during the study period, with q-values of 0.47, 0.10, and 0.10, respectively. HQ relied on Lsi, Areamn, and Lpi, with q-values of 0.31, 0.30, and 0.22, respectively. For LA, Elevation, Pre, and Lpi had relatively strong influences. SR and FP, as natural-socioeconomic dominated ESs, were shaped by the interplay of natural and socioeconomic factors. For SR, the dominant explanatory variables from 2003 to 2023 were Slope, Pre, and Gdp, reflected in q-values of 0.36, 0.14, and 0.06, respectively. For FP, pivotal drivers during the study period included Gdp, Pop, and Elevation, with q-values of 0.63, 0.50, and 0.44, respectively. In general, Spatial heterogeneity of ESs in the YRB from 2003 to 2023 was primarily governed by topographic, meteorological, landscape, and anthropogenic drivers. Notably, Nli exerted a certain degree of influence on all ESs, highlighting urbanization’s intensifying influence, especially on CS, FP, and LA. Fig. 6Explanatory effects of dynamic drivers on ESs change (2003–2023). CS: carbon storage; SR: soil retention; WY: water yield; FP: food production; HQ: habitat quality; LA: landscape aesthetics. Pre: annual precipitation; Pet: potential evapotranspiration; Lst: land surface temperature; Ndvi: normalized difference vegetation index; Pop: population density; Sd: distance from the settlement; Gdp: gross domestic product; Nli: nighttime lighting intensity; Slope: slope; Contag: contagion index; Gyr: radius of gyration; Areamn: mean patch area; Lsi: landscape shape index; Lpi: largest patch index. Explanatory effects of dynamic drivers on ESs change (2003–2023). CS: carbon storage; SR: soil retention; WY: water yield; FP: food production; HQ: habitat quality; LA: landscape aesthetics. Pre: annual precipitation; Pet: potential evapotranspiration; Lst: land surface temperature; Ndvi: normalized difference vegetation index; Pop: population density; Sd: distance from the settlement; Gdp: gross domestic product; Nli: nighttime lighting intensity; Slope: slope; Contag: contagion index; Gyr: radius of gyration; Areamn: mean patch area; Lsi: landscape shape index; Lpi: largest patch index. Interactions for driving forces of ESs To further investigate potential ES enhancement resulting from the intersection of these two driving factors, we examined two-factor interactions leveraging all 15 drivers (Fig. 7). Results revealed interdependent driving effects on ESs in the YRB, displaying a nonlinear pattern of dual-factor synergy. This nonlinear interaction effectively accounted for regional differences in the spatial variability of ESs, as illustrated in Fig. 7, where the interplay of the two factors demonstrated a clear “1 + 1>2” effect. In ESs governed by natural factors, CS interactions demonstrated both two-factor and nonlinear enhancement patterns. Significant interaction pairs include Lpi∩Elevation, Lsi∩Elevation, Lpi∩Gdp, and Lsi∩Gdp, with impact coefficients of 0.41, highlighting the salient role of elevation and Gdp with Lsi and Lpi. Notably, nonlinear enhancement also occurred when Lsi interacted with Nli, Pop, and Slope, significantly affecting CS spatial heterogeneity. For SR, interactions displayed both two-factor and nonlinear enhancement types. Primary interaction factors included Slope∩Pre (0.64), Slope∩Gdp (0.44), and Slope∩Pet (0.42), highlighting the synergistic effects between climate, population, and slope. The Slope∩Pre interaction was most significant. Despite its low q-value, Pre still predominantly exhibited nonlinear enhancement in interactions, amplifying its potential influence on SR’s driving role. Additionally, Lsi and Areamn, while not key individually, synergized with Slope to significantly boost SR. WY interactions manifested through both dual and nonlinear enhancement patterns. Mainly driven by Pre∩Elevation (0.73), Pre∩Pet (0.66), and Pre∩Ndvi (0.61), emphasizing the significant roles of climate, topography, and vegetation. Remarkably, Gdp, Sd, and Pop were also pivotal drivers, since nearby urbanized zones and farmland generally showed reduced evapotranspiration capacities, thereby increasing WY. Conversely, regions with higher vegetation coverage exhibited stronger water retention capacities, leading to lower WY. Furthermore, the interaction between precipitation and landscape structure exhibited two-factor enhancement effect. In natural-socioeconomic factor-dominated ES types, FP interactions were predominantly enhanced by two-factor interactions. Interactions linked with both anthropogenic (Gdp and Pop) and biophysical factors (Elevation and Ndvi) demonstrated significant strength for FP spatial heterogeneity. Chief among these interactions was Ndvi∩Gdp (0.71), exerting a paramount influence, indicating that vegetation cover and population jointly influenced FP. For HQ, the interactions involved dual-factor synergy and nonlinear enhancements. The dominant interaction drivers encompassed Lsi∩Gdp (0.39) and Elevation∩Lsi (0.38), consistent with factor analysis results and revealing the joint influence of human disturbance and topographic conditions. Notably, the Lsi∩Areamn (0.37) interaction drove HQ two-factor enhancement, as large and complex patches typically exhibit higher habitat heterogeneity, enhancing HQ, and vice versa. For LA, nonlinear enhancement served as the primary mode of interaction. The dominant interaction combinations were Pre∩Contag (0.17), Pre∩Elevation (0.13), and Elevation∩Sd (0.11), reflecting that topography, precipitation, landscape configuration, and distance to settlements collectively shaped LA spatial heterogeneity. Fig. 7Interactions of dynamic drivers on ESs change (2003–2023). CS: carbon storage; SR: soil retention; WY: water yield; FP: food production; HQ: habitat quality; LA: landscape aesthetics. Lst: land surface temperature; Ndvi: normalized difference vegetation index; Pop: population density; Sd: distance from the settlement; Gdp: gross domestic product; Nli: nighttime lighting intensity; Slope: slope; Contag: contagion index; Gyr: radius of gyration; Areamn: mean patch area; Lsi: landscape shape index; Lpi: largest patch index. Interactions of dynamic drivers on ESs change (2003–2023). CS: carbon storage; SR: soil retention; WY: water yield; FP: food production; HQ: habitat quality; LA: landscape aesthetics. Lst: land surface temperature; Ndvi: normalized difference vegetation index; Pop: population density; Sd: distance from the settlement; Gdp: gross domestic product; Nli: nighttime lighting intensity; Slope: slope; Contag: contagion index; Gyr: radius of gyration; Areamn: mean patch area; Lsi: landscape shape index; Lpi: largest patch index. Spatial Interactions between ESs and major driving factors However, statistical methods such as OPGD offer a broad statistical overview. These methods do not capture the spatiotemporal nonstationary relationships between ESs and key drivers, a topic explored in Subsection 3.4. Figure 8 depicts the spatial typologies of ES changes and their driving factors. For CS, CS-Lsi is dominated by LH, forming relatively continuous patches in the Alxa Plateau and the Huang-Huai-Hai Plain, whereas HH mainly clusters in the Loess Plateau and the Qinling Mountains; HL is concentrated in the southern Tibetan Plateau. CS-Lpi and CS-Gyr show broadly similar patterns, with HH primarily aggregated in the southern Tibetan Plateau meadow zone and only sporadic occurrences in the Loess Plateau and the Huang-Huai-Hai Plain. In addition, CS-Lpi exhibits LL clustering in the Alxa Plateau, and CS-Gyr shows pronounced LH clustering there. For SR, HH clusters of SR-Slope and SR-Ndvi are primarily distributed in the Loess Plateau hilly-gully region and the Sichuan-Tibet alpine canyon region. For SR-Pre, HH clusters occur in the Loess Plateau gully region, the Sichuan-Tibet alpine canyon region, and the Qilian Mountains, whereas LH clusters are concentrated in high-precipitation areas of the middle and lower reaches. Moreover, SR exhibits predominantly LL or nonsignificant associations with these three factors in the northwestern desert areas, low-relief terrain, and regions with high development intensity. For WY, WY-Pre shows HH concentrated in the central-southern-eastern basin, HL and LH mainly in the high-altitude Tibetan and Loess Plateau regions, and largely nonsignificant patterns in sandy desert areas and the Qaidam Basin. WY- Elevation exhibits HH in the upper reaches, LL clustered in the Alxa Plateau, and widespread HL across the Loess Plateau and the Huang-Huai-Hai Plain. WY-Pet shows LH in the downstream and desert regions, LL prevailing in the middle reaches, scattered HH, and HL mainly in the southern areas. For FP, FP-Gdp and FP-Pop show similar patterns, with HH concentrated in the middle-lower reaches in the agro-pastoral transition zone and irrigated plains, HL widespread across the Loess Plateau, and LL or not significant prevailing in the upper reaches and the northwestern arid regions. FP-Elevation exhibits strong topographic differentiation, with LH and LL or not significant mainly in the high-elevation, rugged upper reaches, while HL dominates across the flatter middle-lower reaches. For HQ, HQ-Lsi shows HH predominating in the central part of the YRB and the northwestern region, whereas a mosaic of HL, LL, and LH is more common in the Tibetan Plateau meadow zone and the North China Plain. HQ-Areamn and HQ-Lpi display broadly similar local association patterns, with relatively concentrated HH clusters in the Tibetan Plateau meadow zone. In contrast, LL clusters are concentrated in Shanxi Province, Henan Province, and the Qaidam Basin, while HL forms contiguous patches across the Alxa Plateau. For LA, LA-Elevation shows HH clusters mainly concentrated in the upstream high-elevation mountain regions, whereas the mid-lower reaches with low-relief plains are dominated by LL and a mosaic of HL and LH. LA-Pre shows HH forming contiguous clusters in the central-eastern basin, while HL exhibits a pronounced belt-like distribution in the Qinling Mountains and the northwestern region. LA-Gyr shows a clear HH core in the central-eastern basin, whereas HL and LH are more prevalent along the upstream margins and in the more intensively developed downstream areas, occurring as scattered and patchy mosaics. Fig. 8LISA cluster map between ES changes and the top three drivers of ESs (2003–2023). CS: carbon storage; SR: soil retention; WY: water yield; FP: food production; HQ: habitat quality; LA: landscape aesthetics; Gyr: radius of gyration; Lsi: landscape shape index; Lpi: largest patch index; Areamn: mean patch area; Pre: annual precipitation; Pet: potential evapotranspiration; Pop: population density; Gdp: gross domestic product; Slope: slope; Ndvi: normalized difference vegetation index. The map was created using ArcGIS Desktop V. 10.8 and the URL link is http://www.esri.com/sofware/arcgis. LISA cluster map between ES changes and the top three drivers of ESs (2003–2023). CS: carbon storage; SR: soil retention; WY: water yield; FP: food production; HQ: habitat quality; LA: landscape aesthetics; Gyr: radius of gyration; Lsi: landscape shape index; Lpi: largest patch index; Areamn: mean patch area; Pre: annual precipitation; Pet: potential evapotranspiration; Pop: population density; Gdp: gross domestic product; Slope: slope; Ndvi: normalized difference vegetation index. The map was created using ArcGIS Desktop V. 10.8 and the URL link is http://www.esri.com/sofware/arcgis. Discussion Model Verification Model validation requires systematic comparison of simulated results with measured data. In this study, outputs from the InVEST model for WY, CS, HQ, and SR were validated at a 100 km² grid scale using reference data from the study area. Since FP was calculated directly from statistical yearbook data on grain yields, the yields of food crops can be directly aggregated. Moreover, the LA results are dimensionless composite values ranging from 0 to 1. Accordingly, it is considered reasonable to omit validation for the FP and LA results, given their derivation and characteristics. In the WY module, the Zhang coefficient, an essential seasonal parameter reflecting precipitation characteristics, is necessary for estimating WY. Given the vast expanse of the YRB, direct calibration against measured runoff data is challenging. This study estimated the annual water yield across the basin. According to the Comprehensive Planning of the YRB (2012–2030) and other studies, the multi-year average natural river runoff in the basin is approximately 534.80 billion m³. After multiple model runs, setting the Zhang coefficient to 3.2 resulted in a simulated average WY of 537.42 billion m³ for the years 2003, 2013, and 2023. The small deviation from official statistics suggests that the model reasonably reflects the actual water resource conditions of the basin. HQ simulation results from the InVEST model were validated using annual human footprint data from Mu et al. (2022)41. Previous studies have demonstrated that human activities act as a major factor driving habitat degradation in the region, implying an inverse relationship between human activity and habitat quality11. By analyzing the spatial relationship between human footprint and HQ at the grid scale, we confirmed the reliability of the simulation results. As shown in Supplementary Figure S7, there was strong spatial consistency between the simulated HQ and the human footprint data, with fit values ranging from 0.74 to 0.83, demonstrating the model’s effectiveness in representing the natural habitat conditions essential for species survival. To validate the simulated CS, this study utilized the woody vegetation carbon density (CD) data from the global terrestrial aboveground biomass dataset published by Xu et al. (2021)42. Given the 10 km spatial resolution of this dataset, the analysis was conducted at a corresponding grid scale. As illustrated in Supplementary Figure S8, the simulated CS and CD exhibit strong spatial concordance, with an average R² of 0.70 over three years, indicating the model’s precision in depicting regional CS distribution. SR is a crucial ecosystem regulatory service in the YRB, playing a key role in preventing land degradation43. Total soil retention in the basin rose from 9.44 billion tons in 2003 to 16.53 billion tons in 2023, This significant increase is consistent with findings by Guo. (2022)44, further supporting the reliability of the model results (Supplementary Table S7). Variations in ESs and impacts of bidirectional land cover transitions The YRB ecosystem has experienced notable spatiotemporal changes. The overall annual mean of CS remained relatively stable but exhibited an initial increase and a subsequent decline, likely due to ecological policies that promoted increased vegetation cover expansion and enhanced carbon sequestration capacity45. However, in the later period, vegetation degradation caused by the expansion of agriculture, transportation, residential areas, and water infrastructure in the middle and lower reaches led to a marked decline in CS. HQ exhibited a sustained declining trend throughout the study period. In some regions, habitat degradation has been somewhat mitigated, reflecting a delayed recovery from ecological protection measures. Nonetheless, issues like wetland shrinkage and biodiversity loss persist, especially in ecologically sensitive areas such as estuaries and deltas, indicating the ecosystem’s limited resilience to withstand intensive development pressures46,47. High SR areas are concentrated in the upper Yellow River, including the Qilian Mountains, Bayan Har Mountains, southern Qinling slopes, and parts of Shanxi Province. These areas feature composite ecosystems of forests, shrublands, and grasslands, which bolster SR by stabilizing soil structure and reducing wind and water erosion13,20,48. WY distribution trends are closely aligned with precipitation gradients, as shown by Shen et al. (2024)49. Key contributors to increased WY include the southern slopes of the Bayin Har and Qinling Mountains. Wetland restoration in the YRB has notably enhanced downstream WY. FP is concentrated in the southeastern plains, characterized by high soil fertility and contiguous farmland, making it highly productive. Conversely, the arid northwest suffers from limited productivity due to water scarcity and fragmented plots50. LA encompasses regional biodiversity, uniqueness, aesthetic value, and recreational potential51. The Yellow River source region, southern Qinling Mountains, and central hilly areas boast diverse ecosystems and high natural integrity, leading to elevated LA values. While the northwest is largely deserts and barren plateau, certain primitive and unique areas retain significant aesthetic value52. There exist supportive or conflicting relationships between LULC types and ESs, indicating complex interactions between LULC transitions and ESs, which is a critical aspect requiring comprehensive understanding22,30. In recent decades, China’s ecological restoration measures, exemplified by the “Three-North Shelterbelt” and “Grain for Green,” have transformed bare land and cropland into grassland, forest, and shrubland, thereby improving ecological functions. Forests and grasslands, through root-based soil stabilization and dense vegetation, mitigate rainfall impact on surface soil, stabilize soil structure, reduce erosion, and significantly decrease soil loss in the YRB, thereby enhancing HQ, LA, and CS. However, this comes at the expense of reducing WY and FP. Wang et al. (2024) and Wu et al. (2025) indicate that vegetation restoration in the YRB has reached the region’s water carrying capacity8,36. Extensive afforestation leads to excessive water consumption by plants, altering the relationship between rainfall and runoff and reducing streamflow under consistent precipitation conditions. Conversely, converting other land cover types to farmland and settlements positively impacts WY and FP, consistent with recent studies3,11,19. Although expanding built-up areas may temporarily boost WY13, it leads to vegetation loss, exacerbates the urban heat island effect, and diminishes other ecosystem services53. Cropland significantly improves FP but offers lower supporting, regulating, and cultural services compared to other land types11,28. Interestingly, some grassland areas with dense vegetation have low soil retention, especially in the Loess Plateau and Qinghai-Tibet Plateau edges, where steep terrain, concentrated rainfall, and high erosion sensitivity are prevalent. Although grasslands have certain soil retention potential, their shallow roots, loose surface structure, and lack of artificial soil and water conservation measures make them less effective in resisting runoff from intense rainfall, thereby limiting SR54. In contrast, some agricultural lands in central mountainous regions enhance SR through terracing and check dams, even in high erosion risk zones, further supporting the perspective proposed by Shifaw et al. (2024)22. Interactions between ESs and driving factors Our study area features diverse environmental conditions and varied human activity patterns34, contributing to the spatial heterogeneity of ESs. Xu et al. (2017) noted that trade-offs and synergies among ESs are shaped by land use conflicts and consistencies55. Specifically, land use consistency refers to land use types that can simultaneously support multiple ESs, fostering synergies among services22. In the Middle Reaches of the Yellow River and the southeastern Tibetan Plateau, areas with high vegetation coverage, dominated by forests, grasslands, and shrublands, not only serve as the main land types supporting CS and SR, but also significantly boost biodiversity. This finding aligns with studies by Pei et al. (2024) and Xu et al. (2025)4,30. Soil and water conservation efforts in the YRB over recent decades have significantly ameliorated soil fertility and crop yields, while providing long-term soil retention benefits56, fostering synergies between SR and FP. Agricultural land use practices, such as terracing and check dams, are designed to synchronize crop water requirements with precipitation cycles. When combined with spatial optimization strategies, like cultivating grains on plateaus and storing water in channels, these practices establish an efficient “production-water retention” system, enhancing the interplay between FP and WY. Furthermore, a pronounced synergy between SR and WY was evident in regions with low natural rainfall and minimal surface runoff, effectively reducing soil erosion risks4. The introduction of ecological restoration initiatives in the YRB has improved the LULC composition by transforming marginal lands into ecologically valuable areas such as forests, grasslands, and wetlands. This transformation fosters synergies among HQ, CS, and LA. These investigations underscore the role of consistent land use in bolstering synergies between ES pairs. Conversely, land use conflict arises when a particular land use type cannot support multiple ES simultaneously, leading to trade-offs between ESs57. FP presented trade-offs with HQ and CS. Croplands and artificial grasslands are the main land uses supporting FP, but they offer weaker ecological functions than natural land types like forests and wetlands, leading to conflicts between FP and HQ and CS30. Vegetation enhances HQ-CS synergies by absorbing carbon dioxide and providing habitats. However, its evapotranspiration increases ecological water demand, reducing WY supply5. While cropland expansion and soil conservation projects have boosted SR and FP, they have also disrupted natural vegetation and topography51,58, thereby decreasing LA, as observed by Xia et al. (2023) and Liu et al. (2024)11,13. In regions with stable LULC structures, especially in the northwest, ES pairs mainly show no-effect relationships due to homogeneous land cover and simplified ecosystems, making trade-offs and synergies between ES pairs unlikely12. Recent studies have identified landscape structure, population, topography, and climate as critical drivers of ESs18,22,59. This study quantitatively assessed and spatially mapped the relationships between ESs and their primary drivers, with results consistent with recent research11,13. CS and HQ were significantly influenced by landscape structure. In the southern Tibetan Plateau meadow area, CS and HQ co-located with landscape factors, suggesting that large patches and high connectivity support carbon accumulation and habitat maintenance under low disturbance. In contrast, agricultural intensification and urban expansion in the Huang-Huai-Hai Plain have increased fragmentation, while in the Alxa Plateau, desert-gobi conditions, water constraints, and sparse vegetation hinder ecological function, leading to a mismatch between ‘landscape gain’ and ‘ecological gain22. In erosion-sensitive areas such as the Loess Plateau and the Sichuan-Tibet al.pine canyon, slope and vegetation cover amplify erosion processes. Soil and water conservation measures, such as terracing and reforestation, have enhanced surface roughness and vegetation cover, stabilizing the co-location between SR and Slope, Ndvi, and Pre60. In contrast, in the middle and lower reaches of the basin, high precipitation tends to result in spatial mismatches between Pre and SR, suggesting that increased precipitation in areas with strong disturbance and low retention conditions may lead to erosive runoff, diminishing conservation benefits. In the northwestern desert and areas with high development intensity and low relief, the dominant pattern is low or non-significant, indicating that water constraints and surface disturbance jointly limit the effectiveness of SR3. Among the influencing variables, Pre, Elevation, and Pet have the greatest impact on WY. In low-altitude, high-precipitation areas with sparse vegetation, limited plant water consumption enhances water yield. In the North China Plain, urban expansion and the urban heat island effect decrease water yield, highlighting the trade-off with Pet3. In high-altitude areas, steep slopes facilitate runoff more easily than flat terrains, promoting the synergy between elevation and WY3. The spatial pattern of FP reveals the dual influence of human input and topographic constraints. In the middle and lower reaches, FP is closely linked to Gdp and population, driven by irrigation and economic activities. In contrast, in the Loess Plateau, agricultural production relies more on soil and management practices, with high output not necessarily correlating with population or economic intensity. In high-altitude areas, topographic variation and low temperatures limit agricultural mechanization and contiguous land use61. LA are shaped by a topographic foundation together with water–thermal conditions and land-use configuration. High-value clusters in the upstream mountainous areas reflect terrain differentiation and stable landscape features like canyons and alpine meadows. In contrast, plains in the middle and lower reaches are more prone to low values due to agricultural homogenization and urban expansion. The co-location of Pre and LA in the central-eastern basin suggests that water-thermal conditions enhance aesthetic perception through improved vegetation and water body landscapes, while Gyr in areas with high development intensity tends to signal fragmentation and reduce aesthetic quality22,59. Implications for sustainable regional management and ES governance Integrating ESs assessment and management into spatial planning is essential for prioritizing conservation and restoration efforts to foster sustainable development and harmonious human-environment interactions4,62. This study elucidates the complex interactions among regional ESs and shows that large-scale biophysical land management initiatives, including the Three-North Shelterbelt Program, the Natural Forest Protection Program, the Grain for Green Program, and sloping farmland improvement, have substantially improved ecological conditions, as evidenced by reduced soil erosion and desertification, increased CS, enhanced HQ, and maintained FP. However, afforestation and terracing can significantly reduce soil water storage and surface runoff, resulting in a decline in WY, which is the most prioritized service in the ecological management of the basin. These findings demonstrate that the effectiveness of ecological restoration projects does not linearly increase with intervention intensity, and excessive vegetation cover could reduce water yield and soil retention capacity36. Therefore, vegetation configuration should prioritize evapotranspiration constraints. In the desert–Gobi transition zone and wind-eroded areas, drought-tolerant shrubs and grasses with low evapotranspiration such as hippophae rhamnoides and haloxylon ammodendron should be preferred to reduce canopy water consumption and enhance sand fixation. In the Loess Plateau gully region, restoration should focus on native shrub–grass communities and low-water-demand broadleaf shrubs. In relatively humid mountainous and valley areas, mixed broadleaf or conifer–broadleaf stands can be deployed, integrating mixed species, uneven-aged structure, and forest–grass mosaics to reduce the high water-demand risk of single-species plantations and improve community stability5. In built-up development zones, management should prioritize controlling development intensity and minimizing ecological disturbance by directing new construction land primarily to infill within existing built-up areas. This supports a land-use strategy that prioritizes redevelopment of existing land while treating new land expansion as supplementary, and it restricts ribbon sprawl along rivers and transport corridors as well as extensive, low-density “pancake-style” urban expansion31. In the upper reaches, where agricultural infrastructure and farming conditions are limited, eco-friendly agriculture can be promoted, including organic vegetable, orchard, and medicinal crop production. In alpine pastoral areas, organic rangeland certification and eco-livestock systems can be advanced alongside grassland restoration and rotational grazing to balance conservation with income generation. In the middle and lower reaches, where agriculture is intensive and water constraints are stronger, the priority should be efficiency gains rather than increased irrigation, by promoting drip or micro-sprinkler irrigation, fertigation, and field practices such as mulching for moisture conservation and conservation tillage to improve soil water and nutrient retention. We evaluated the effects of class-to-class LULC conversion on ESs. By quantifying the changes in ESs across both transformed and stable LULC areas in the YRB, policymakers can develop sustainable land management strategies. For instance, afforestation on degraded grasslands, croplands, and bare lands can enhance SR, CS, and HQ. Meanwhile, regional climatic hydrological, and soil conditions must be considered when formulating strategies. Targeted vegetation conservation and restoration, adjustments to the forest-grassland ratio, and engineering measures can reduce WY consumption. Additionally, the scattered and poorly planned expansion of residential areas has significantly degraded HQ, CS, and LA. This necessitates enhancing the efficiency of built-up land use through science-based spatial planning, stringent regulatory enforcement, urban regeneration, and eco-architecture adoption. Wetlands and water bodies are vital for maintaining HQ and LA, necessitating the protection of regions experiencing vegetation degradation and wetland shrinkage, such as the Yellow River source, estuarine areas, and delta region46,47. Simultaneously, establishing ecological monitoring networks for forests and grasslands can enable routine evaluation of environmental conditions4,31. Moreover, diverse and visually appealing landscapes, including mountains, snow-capped peaks, rivers, wetlands, urban parks, and well-managed agricultural areas, not only provide cultural and recreational services but also support local economies through ecotourism and the enhancement of regional green capital. In regions with both high natural landscape value and ecological vulnerability, such as the Yellow River source area and the Inner Mongolia Plateau, comprehensive ecological risk assessments are essential. Integrating scientifically informed ecological restoration and conservation measures with rational spatial planning can ensure that ecosystem management and urban development simultaneously maintain regulating and provisioning ecosystem services while also preserving aesthetic and cultural values. Policymakers should also recognize the environmental impacts of agriculture, including pesticides and fertilizer pollution, and inefficient water use25. Strategies like ecological compensation, restoration, crop rotation, fallowing, and region-specific eco-friendly agriculture should be promoted to prevent farmland degradation. A penalty-reward system for the ecological value of farmland is also recommended63. Notably, HQ, SR, and WY have decreased in stable forest, shrubland, and grassland areas from 2003 to 2023, indicating that factors beyond LULC change need to be incorporated into sustainable land management. Understanding the changes in ESs, their interactions, and key driving factors are critical inputs for minimizing trade-offs and enhancing synergies8,30,49. This spatial information can enable policymakers to more effectively plan regional governance frameworks and science-driven environmental policies. Limitations and future research directions This study explored the spatiotemporal evolution of synergies and trade-offs among six ESs in the YRB and their relationships with driving factors. However, several limitations must be acknowledged. (1) Although six ESs were selected and results aligned with other studies3,13,34,49, they did not fully capture the overall ecological conditions of the basin. Future research could include additional ESs, such as water conservation, wind prevention, sand fixation, and wind energy production. (2) While multiple driving factors were considered, human demand for ESs received less attention. Future work should focus more on the supply-demand balance and resource allocation47,48. Moreover, spatial scale remains a critical focus in ES research11,30, as varying scales could shape both ES relationships and their associations with key drivers. (3) This study used the InVEST model suite to quantify multiple ecosystem services. Although widely applied, InVEST relies on simplified, largely static relationships and empirical parameters, and thus cannot fully represent process feedbacks, nonlinear thresholds, or management-driven biophysical dynamics. Such simplifications may bias absolute service estimates and introduce uncertainty in subregional comparisons. Here, we emphasize robust spatial patterns, relative differences, and co-location typologies rather than exact magnitudes. To reduce uncertainty, we applied consistent inputs and parameter settings across years, harmonized all datasets to a common grid and projection, and conducted independent validation where external data were available. Future work could improve robustness by coupling InVEST with process-based or data-driven models (e.g., SWAT for hydrology and water yield, and MaxEnt or connectivity-based frameworks for habitat-related services) and by conducting formal sensitivity and uncertainty analyses. (4) Socio-cultural dimensions, including recreational preferences, cultural heritage values, and public perceptions of ESs, were not addressed, which may limit the comprehensiveness of the assessment. Future research should therefore expand the scope of ESs, incorporate socio-cultural indicators, integrate human demand, and adopt multi-model approaches to address scale effects and model uncertainties. These improvements would advance a more holistic and robust framework for ES assessment and regional ecological governance. Conclusions This study systematically assessed the distribution dynamics and spatial interactions of critical ecosystem services across the YRB, along with their underlying drivers. The results demonstrated that: (1) CS and SR were higher in the southeastern Qinghai-Tibet Plateau and Shanxi Province, while WY and FP were concentrated in the southern and eastern regions. LA and HQ were primarily found in areas with high vegetation coverage in the southern, western, and central regions. Over time, HQ and LA showed consistent declines, while WY and FP steadily increased. CS and SR exhibited upward trends in the upper reaches, but CS declined in the middle and lower reaches, with SR showing a reverse trend in the middle reaches. (2) In areas with persistent LULC, 15 ES pairs and their relationships with key driving factors primarily showed a no-effect relationship. In areas with LULC changes, synergies were more prevalent between SR and other ESs, as well as in HQ-CS, HQ-LA, CS-LA, and FP-WY pairs, while trade-offs were more pronounced in HQ-FP, HQ-WY, FP-LA, WY-LA, CS-FP, and SR-LA pairs. (3) Bivariate LISA indicates pronounced zonal heterogeneity in ES–driver linkages across the YRB. CS and HQ are primarily shaped by landscape configuration, SR by slope, vegetation, and precipitation controls, WY by climate and topographic conditions, and FP by socio-economic intensity and topographic accessibility. Strong co-location tends to occur in upstream ecological source areas, while arid, low-relief, and highly disturbed regions exhibit weaker associations and more frequent spatial mismatches. Based on these findings, we recommend promoting afforestation, shrub expansion, and grassland restoration to enhance ESs, while managing trade-offs through spatial planning and urban renewal. Priority should be given to soil and water conservation in areas sensitive to changes in WY and SR, to optimize ecosystem service synergies and foster ecological resilience across the YRB. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1