Interactive Effects of Climate and Land-Use Changes on the Spatiotemporal Evolution of Water Ecosystem Services in the Yellow River Basin, China

Abstract

The Yellow River Basin (YRB) faces escalating pressures from climate variability and intensive land-use change; however, how these drivers jointly shape the water ecosystem services (WES) remains insufficiently quantified. This study assessed water yield, soil conservation, and water purification capacity across the YRB from 2000 to 2020 using the InVEST model. The effects of driving forces, including climate and land-use changes, on WES were examined using scenario simulation and the geographical detector method. During 2000–2020, water yield and soil conservation displayed fluctuating upward trends, increasing by 26.24% and 30.19%, respectively, whereas nitrogen and phosphorus exports declined slightly by 4.82% and 3.08%, indicating a modest improvement in water purification. All indicators exhibited elevated values in the southeast and lower values in the northwest, with clear distinctions among the three topographic zones. The contribution rates of climate change to variations in water yield, soil conservation, nitrogen export, and phosphorus export were 97.4–99.3, 94.5–98.3, 87.2–96.0, and 85.7–95.2%, respectively, indicating the central regulatory role of climatic factors in watershed-scale water-related ecological processes. However, the contribution of land-use factors increased over time and had a significantly greater impact on water purification capacity. Distinct spatial heterogeneity was observed among the WES. Interactions among climatic, topographic, and other factors significantly enhanced the spatial variability of water yield and soil conservation services, while land-use interactions with other drivers had greater impacts on the spatial variability of water purification capacity. Thus, regionally differentiated policy strategies are essential. Our findings provide a quantitative basis for differentiating climate versus land-use interventions in water resource management and designing spatially targeted ecological restoration policies in the YRB and similar regions worldwide.

1. Introduction

Water ecosystem services (WESs) denote the direct and ancillary benefits that humans derive from water ecosystems through their structure, functions, and processes and represent a specific form of ecosystem services in the field of water ecology [1]. As the core functions of water ecosystems [2], water yield, soil conservation, and water purification services are central to the systematic research aimed at maintaining regional water resource security, strengthening ecological barriers, and ensuring environmental stability. Currently, the rising frequency of extreme climate events, coupled with rapid urbanization across China, is exacerbating environmental challenges, including water scarcity and deterioration of water ecosystems, placing significant stress on their functional performance. Therefore, investigating the influencing factors and driving mechanisms of changes in WESs is urgently required.

The assessment of WESs has evolved from single-indicator estimation to multi-indicator comprehensive evaluation. Early studies primarily focused on quantifying individual services, such as the water yield and soil erosion. With the refinement of the conceptual framework of ecosystem services, the water yield, soil conservation, and water purification have been identified as the three core functions of water ecosystems. Accordingly, the methods for comprehensive assessment have advanced from statistical-based empirical models to process-based physical models [2]. Recently, with the widespread use of computational techniques, models such as InVEST and SWAT have been widely applied due to their ability to spatially represent the formation processes of ecosystem services [3]. For example, studies have examined the functional characteristics of WES across the East African highlands [4], the Nitra area in Slovakia [5], and the Tungabhadra Basin in India [6] based on the InVEST model. Chinese scholars have focused on watershed and regional scales, primarily assessing water yield, soil conservation, and water purification services in basins such as Poyang Lake [7], Dongting Lake [8], Taihu Lake [9], Luan River [10], Wujiang River [11], Beipan River [12], and Longxi River in Chongqing [13], as well as in regions such as the Yangtze River Economic Zone [14], southwestern China [15], and the Beijing–Tianjin–Hebei region [16]. However, the localization of model parameters and quantification of trade-offs and synergies among multiple services are major challenges in current assessment methods. Therefore, for the ecologically vulnerable Yellow River Basin (YRB), the application of a systematic method and its validation for accurately characterizing the spatiotemporal heterogeneity of core WESs are lacking.

Climate change and land-use change are the core drivers of changes in WESs [17]. Climate change affects water ecosystems by influencing vegetation distribution, water cycles, and nutrient cycles [18]. In contrast, land-use change affects ecosystem service functions by altering water and nutrient migration processes, soil structural characteristics, and the formation processes of non-point source pollution [19]. Notably, land-use change is the most dynamic factor affecting ecosystem service supply over short timescales. Different land-use types exhibit fundamental differences in their capacity to supply ecosystem services, with secondary land-use classes further amplifying these differences. Meanwhile, the intensification of land use produces a “hidden” effect, where different management practices on the same land-use type can lead to significant variations in service output [20]. The spatial distribution pattern of land use influences regional ecological processes by controlling the flow of materials and energy within ecosystems. Previous studies have shown that the dominant driving factors vary significantly across regions: in arid and semi-arid areas, climate change contributes more than 60% to the water yield service [21,22,23]; conversely, in regions with rapid urbanization, land-use change is the primary cause of water quality degradation [21,24]. Notably, climate change and land-use change do not act independently but form complex nonlinear interactive effects through vegetation–soil–hydrology feedback processes [17]. However, current research has primarily identified driving factors based on statistical correlation analyses and remains insufficient to quantitatively disentangle the interactions between climate and land use, as well as to identify threshold effects.

The Yellow River Basin (YRB) has a relatively fragile natural ecological base, characterized by low resistance to disturbance and high sensitivity. Currently, amid the combined pressures of climatic shifts and anthropogenic disturbances, ecological and environmental issues in water systems, such as sharp reductions in runoff, severe soil erosion, and deterioration in water quality, have arisen. Therefore, scientifically assessing temporal and spatial variations as well as the underlying forces affecting WES within the YRB is urgently needed. Current research in this basin has mainly focused on ecosystem service assessments covering various service functions [25,26,27], ecological security pattern construction [28], trade-offs and synergies [29]. For instance, Fan et al. [30] systematically analyzed three general services (carbon storage, water yield, and soil conservation) in the YRB of Henan Province, concluding that the three services share similar spatial distribution patterns. These studies have greatly deepened our understanding of ecosystem service functions in specific areas of the YRB. However, certain deficiencies remain. Considering research themes, most studies focus on general services or static evaluations, with few systematic dynamic assessments of the three core water ecosystem services (water yield, soil conservation, and water purification) at the basin scale. Considering mechanism analysis, most studies qualitatively describe the driving factors and lack quantitative separation of the individual contributions and interactive effects of climate and land-use changes. These deficiencies greatly limit the systematic understanding of the spatiotemporal evolution patterns and causal mechanisms of WES in the YRB.

Building on these insights, we aim to address two key scientific questions: (1) What are the spatiotemporal evolution characteristics of the core WES in the YRB during 2000–2020? (2) What are the individual contributions and interaction mechanisms of climate change and land-use change to these variations? To this end, this study comprehensively evaluated the temporal and spatial evolution patterns of the core WES indicators (water yield service, soil conservation function, and water purification capacity) in the YRB from 2000 to 2020 using the InVEST model. Furthermore, it employs scenario simulation and the geographical detector method to quantitatively analyze the degree of influence and interactions among driving factors, such as climate change and land-use change, on variations in WES. This study aims to provide scientific support for ecological protection and high-quality development in the YRB.

2. Study Area and Research Methods

2.1. Study Area

The Yellow River originates from the Bayan Har Mountains in the Qinghai–Tibet region. With Hekou Town in Tuoketuo County, Inner Mongolia, together with Taohuayu in Zhengzhou, Henan Province, functioning as dividing points between its upper, middle, and lower sections, it traverses nine provinces from west to east—Qinghai, Sichuan, Gansu, Ningxia, Inner Mongolia, Shanxi, Shaanxi, Henan, and Shandong. It ultimately discharges into the Bohai Sea via Kenli County, Shandong Province, spanning a total length of 5464 km, making it the river with the highest sediment concentration in the world [31] (Figure 1).

The YRB (32° N–42° N, 96° E–119° E) covers an area of 795,000 km². As water ecological functions are primarily constrained by topographic factors (elevation and topographic relief), this study adopted a regional division that differs from the traditional upper-, middle-, and lower-reach classification. Based on topographic characteristics and prefecture-level administrative boundaries, the basin is categorized into three zones. The headwater area upstream of Lanzhou (Zone I) features numerous mountains with elevations exceeding 4000 m and annual precipitation exceeding 400 mm, as well as extensive glaciers and permanent snow cover concentrated in the Anyemaqen and Bayan Har Mountains, which serve as critical water sources for the Yellow River. The prevailing land-cover categories are grassland, forestland, and unused land (primarily alpine deserts and bare rock in the Chinese land classification system). The zone covering the upper reaches downstream of Lanzhou and the middle reaches (Zone II) is dominated by the Loess Plateau, with altitudes ranging from 1000 to 2000 m and annual precipitation ranging from 200 to 600 mm. Croplands, grasslands, and forestlands constitute the main land-use patterns. The lower reach region (Zone III) has a general elevation below 500 m, experiences a temperate monsoon climate with annual rainfall exceeding 600 mm, and is characterized by land use dominated by croplands, construction land (including urban land, rural settlements, and infrastructure), and water bodies.

2.2. Data Sources

In this study, we primarily used meteorological and land-use datasets for five periods (2000, 2005, 2010, 2015, and 2020), along with soil, topographic, vegetation, and socioeconomic datasets. The specific information of these datasets is listed in

Table 1. Introduction to data sources.

Type	Description	Resolution	Sources
Meteorological data	Precipitation Potential Evapotranspiration Mean annual temperature	Raster, 1 km	National Earth System Science Data Center (https://www.geodata.cn/): Annual Precipitation Dataset for China (1901–2023) [32] Monthly Potential Evapotranspiration Dataset for China (1901–2024) [33] Mean Annual Temperature Dataset for China (1901–2023) [32]
Land use data	Land use types	Raster, 1 km	Resource and Environmental Science Data Platform, Chinese Academy of Sciences (https://www.resdc.cn/): Annual China Multi-period Land Use Remote Sensing Monitoring Dataset (CNLUCC)
Soil data	Plant available water content Soil erodibility factor	Raster, 1 km	National Glacier, Permafrost and Desert Science Data Center (http://www.ncdc.ac.cn/): China Soil Map based Harmonized World Soil Database (HWSD) (v1.1)
	Rooting depth	Raster, 1 km	Depth-to-bedrock Map of China (https://doi.org/10.1038/s41597-019-0345-6) [34]
Topographic data	DEM	Raster, 30 m	Geospatial Data Cloud (https://www.gscloud.cn/)
	Topographic relief	Raster, 1 km	Relief Degree of Land Surface Dataset of China (https://doi.org/10.3974/geodp.2018.02.04) [35]
Vegetation data	NDVI	Raster, 30 m	National Ecological Science Data Center (https://nesdc.org.cn/): China 30 m Annual Maximum NDVI Dataset (2000–2022) [36]
Socioeconomic data	Population density GDP	Raster, 1 km	Resource and Environmental Sciences Data Platform, Chinese Academy of Sciences (https://www.resdc.cn/): Annual China 1 km Gridded Population Dataset Annual China 1 km Gridded GDP Dataset

. Notably, high-altitude areas may experience uncertainty in research results owing to the low reliability of meteorological data. To achieve data standardization and facilitate integrated analysis, all raster data were resampled to a 1 km × 1 km grid, and the projected coordinate system was standardized to Krasovsky_1940_Albers.

2.3. Research Methods

2.3.1. Methods for Assessing Water Ecosystem Services

The Integrated Valuation of Ecosystem Services and Trade-offs (InVEST) model (version 3.14.1 [37]) is an open-source assessment system jointly developed by Stanford University (Stanford, CA, USA), The Nature Conservancy (TNC) (Arlington, VA, USA), and World Wildlife Fund (WWF) (Washington, DC, USA). Its core function is to simulate the dynamic evolution of ecosystem services. The model has distinct advantages, including easy accessibility of driving data, high accuracy in quantitative assessment, and clear spatial representation of the assessment processes and results [37]. In recent years, the InVEST model has achieved a leading position in both academic recognition and application popularity in the field of ecosystem service research [38]. Therefore, this study is mainly based on the InVEST model to calculate and analyze the core functions of the water ecosystem in the YRB. The model was run at 5-year intervals for the years 2000, 2005, 2010, 2015, and 2020.

(1)

Water yield

The water yield service represents the ecosystem’s water-supply capacity, defined as the water resource supply per unit area per unit time within a specific zone. In the InVEST model, the annual water yield assessment module quantitatively estimates water supply capacity using raster grids based on the Budyko water–heat coupling equilibrium principle. The calculation formulae are as follows [14]:

$$Y_x=\left(1-{\displaystyle \frac{AE{T}_x}{P_x}}\right)\times P_x$$

(1)

$${\displaystyle \frac{AE{T}_x}{P_x}}={\displaystyle \frac{1+{\omega }_x{R}_x}{1+{\omega }_x{R}_x+\frac{1}{R_x}}}$$

(2)

$${\omega }_x=Z\times {\displaystyle \frac{PAW{C}_x}{P_x}}$$

(3)

$$R_x={\displaystyle \frac{K_x\times ET{o}_x}{P_x}}$$

(4)

where Y_x represents mean annual water yield (mm) at grid cell x; AET_x corresponds to annual actual evapotranspiration (mm) occurring at pixel x; P_x indicates mean yearly precipitation (mm) received by pixel x; ω_x corresponds to the quotient of modified plant available water capacity relative to mean annual precipitation; R_x indicates the Budyko drying coefficient; Z is the Zhang coefficient; PAWC_x signifies plant-accessible water storage (mm) with pixel x; K_x represents the vegetation evapotranspiration factor; and ETo_x is the annual average reference evapotranspiration (mm).

The annual water yield assessment module in the InVEST model requires the input of the root depth and K_x for each land-use type (

Table 2. Parameter sheet for the calculation of water yield.

Land-Use Type	Root Depth (mm)	Kx
Cropland	2100	0.65
Forest land	5300	1
Grassland	2400	0.65
Water area	100	1
Construction land	100	0.3
Unused land	100	0.5

), which were obtained from previous studies [39] and the parameters recommended in the InVEST model manual [37]. Additionally, the Zhang coefficient is an important input parameter for the model, and it was set to 3.6 based on a previous study [39].

(2)

Soil conservation function

The soil conservation function refers to an ecosystem’s capacity to regulate and prevent soil loss from wind, runoff, and other erosional processes. According to the Universal Soil Loss Equation (USLE), the sediment transport module in the InVEST model quantifies soil conservation by measuring the difference between potential and actual soil erosion. The calculation formulae are as follows [40]:

$$RKLS=R\times K\times LS$$

(5)

$$USLE=R\times K\times LS\times C\times P$$

(6)

$$SD=RKLS-USLE$$

(7)

where SD corresponds to soil conservation capacity (t·hm⁻²); RKLS signifies potential soil erosion (t·hm⁻²); USLE indicates actual soil erosion (t·hm⁻²); R means the rainfall erosion factor; K means the soil erodibility factor; L and S constitute topographic factors (slope length factor and slope gradient factor, respectively); C means the plant cover factor; and P means the soil conservation measure factor.

In the model, C and P are two fixed parameters used to adjust the actual deviation in the estimation of soil conservation, and their specific values are determined based on the relevant literature [41] (

Table 3. Parameter sheet for the calculation of soil conservation.

Land-Use Type	Cropland	Forest Land	Grassland	Water Area	Construction Land	Unused Land
C	0.23	0.08	0.24	0	0	1
P	0.3	1	1	0	0	1

).

(3)

Water purification capacity

Water purification capacity refers to the natural self-purification capacity of water systems achieved through the interception and adsorption of nutrients (e.g., nitrogen and phosphorus) in runoff by vegetation, microorganisms, and aquatic organisms. In the InVEST model, the nutrient transport module calculates nitrogen and phosphorus outputs to water bodies, reflecting watershed water quality. Lower nitrogen and phosphorus outputs indicate a stronger capacity for water purification. Nutrients are transported via both surface runoff and groundwater flow, but this study focuses on surface runoff due to lack of pathway-specific data. The calculation equations are as follows [42]:

$$X_{exp,i}=loa{d}_{surf,i}\times ND{R}_{surf,i}$$

(8)

$$ND{R}_{surf,i}=ND{R}_{0,i}{\left[1+exp\left({\displaystyle \frac{I{C}_i-I{C}_0}{k}}\right)\right]}^{-1}$$

(9)

where X_exp,i denotes the nutrient discharge (nitrogen and phosphorus) per cell i within the watershed (kg·hm⁻²). load_surf,i represents surface nutrient loading (N and P) at pixel i (kg·hm⁻²), derived by calibrating the average nutrient loads corresponding to distinct land cover classes with the runoff potential index (represented by precipitation). NDR_surf,i is the surface nutrient transport rate for grid i, derived from the maximal nutrient retention efficiency (NDR_0,i), topographic index (IC_i, IC₀), and calibration parameter k, which governs the relationship between pixel i and the stream.

Among the model’s input parameters, the nitrogen and phosphorus nutrient load coefficients (load_n and load_p) are used to calculate the surface nutrient loading load_surf,i. Moreover, the vegetation interception efficiencies (eff_n and eff_p) correspond to the maximal nutrient retention efficiency (NDR_0,i); all the aforementioned parameters are determined based on the InVEST model manual [37] (

Table 4. Parameter sheet for the calculation of the water purification capacity.

Land-Use Type	load_p	eff_p	load_n	eff_n
Cropland	3.57	0.48	27	0.25
Forest land	1.36	0.67	1.4	0.4
Grassland	0.93	0.6	4	0.35
Water area	0	0.4	0	0.02
Construction land	2.1	0.26	6.3	0.05
Unused land	0.79	0.26	6.3	0.05

).

2.3.2. Scenario Simulation Method

Using scenario simulations [39,43] with 5-year intervals as the research period, the contributions and differences in climatic and land-use changes to variations in WES were assessed in the YRB. To avoid cumulative bias from a single fixed baseline, we adopted a rolling baseline approach, wherein the climate and land-use type of the starting year of each 5-year interval is used as the reference for that interval. The scenario settings for each interval are presented in

Table 5. Land-use and climate change scenario settings.

Input Parameter	Climate Change Parameters	Land-Use Parameters
Scenario 1	2005	2000
Scenario 2	2000	2005
Reference scenario	2000	2000
Scenario 1	2010	2005
Scenario 2	2005	2010
Reference scenario	2005	2005
Scenario 1	2015	2010
Scenario 2	2010	2015
Reference scenario	2010	2010
Scenario 1	2020	2015
Scenario 2	2015	2020
Reference scenario	2015	2015

. Considering the period from 2000 to 2005 as an example, scenario 1 assumed no changes in land-use types with only climate change, whereas scenario 2 assumed no climate change with only changes in land-use type. Subsequently, a comparative analysis of the results under each scenario setting was conducted using the InVEST model based on the control variable concept. Thus, the scenario simulations encompass the entire study period from 2000 to 2020 through a series of 5-year rolling comparisons. The contribution rate is calculated as follows [23]:

$$\Delta E{S}_{cli}=E{S}_{cli}-E{S}_1$$

(10)

$$\Delta E{S}_{lu}=E{S}_{lu}-E{S}_1$$

(11)

$$R_{cli}={\displaystyle \frac{|\Delta E{S}_{cli}|}{|\Delta E{S}_{cli}|+|\Delta E{S}_{lu}|}}$$

(12)

$$R_{lu}={\displaystyle \frac{|\Delta E{S}_{lu}|}{|\Delta E{S}_{cli}|+|\Delta E{S}_{lu}|}}$$

(13)

where ES₁ denotes the WES under the reference scenario and ES_cli and ES_lu represent the simulated values for Scenarios 1 and 2, respectively. The ΔES_cli and ΔES_lu are variations induced by climatic shifts and land-use modifications, respectively. R_cli and R_lu indicate the proportional contributions of climatic and land-use drivers to these variations.

2.3.3. Geodetector Method

The aforementioned scenario analysis is primarily used to quantify the temporal contribution of climate versus land use to changes in WES over time, whereas Geodetector (accessed on 30 October 2024 from http://www.geodetector.cn) is used to identify the spatial explanatory power of individual and interacting factors on WES patterns. These two approaches address different scientific questions and are therefore complementary.

To identify core drivers of spatial heterogeneity in WES across the YRB, this study employs a combination of “primitive factors + newly added factors” for factor selection. Primitive factors serve as key input variables in service assessment processes, primarily precipitation and land use types. Factor selection in geographic detection is not about “more is better.” Introducing excessive factors with weak correlations to the dependent variable’s spatial heterogeneity would introduce “noise” that interferes with the identification of critical drivers. This research aims to reveal the key determinants rather than indiscriminately increasing variable count. Therefore, based on the combination of natural and human activity factors, we selected representative influential factors: precipitation and annual average temperature as climatic factors; digital elevation model (DEM) and topographic relief as topographic factors; normalized vegetation index (NDVI) as a vegetation factor; and land-use type, regional GDP, and population density [23,40,44] as human activity factors. The Jenks natural breaks method was used to classify the above continuous factor variables. Significance tests indicated that most factors passed the 0.05 significance level, whereas only GDP and population density failed the test in certain years. Constrained by administrative boundaries, the GDP and population density exhibited a stepwise urban–rural distribution, making it difficult for the natural breaks method to identify valid spatial stratification. This resulted in excessively large within-stratum variance and an insignificant q-statistic. Through factor detection and interaction analysis, we examined how these factors accounted for the spatial variability of WES and the intensity of their interactions. Factor detection is expressed as follows [45]:

$$q=1-{\displaystyle \frac{{\displaystyle \sum _{h=1}^L{N}_h}{\sigma }_h^2}{N{\sigma }^2}}$$

(14)

where q ranges from 0 to 1, with higher values signifying an enhanced explanatory capacity for the factor. h denotes the hierarchical level of factors, L represents the aggregate number of stratifications; N_h and N denote cell counts within stratum h and across the entire region, respectively; and ${\sigma }_h^2$ and ${\sigma }^2$ represent WES variance within stratum h and across the entire region, respectively.

Building on this framework, the interaction effect was assessed to determine whether the combined effects of any paired factors amplified or reduced the explanatory capacity regarding the spatial heterogeneity of WES. Interaction effects were classified into five distinct categories (

Table 6. Classification of factor interactions.

Criterion	Interaction
q(x1 ∩ x2) < Min[q (x1), q(x2)]	nonlinear weakening
Min(q(x1), q(x2)) < q(x1 ∩ x2) < Max[q(x1), q(x2)]	Single-factor nonlinear weakening
q(x1 ∩ x2) > Max[q(x1), q(x2)]	dual-factor amplification
q(x1 ∩ x2) = q(x1) + q(x2)	independence
q(x1 ∩ x2) > q(x1) + q(x2)	nonlinear amplification

).

3. Results

3.1. Spatial-Temporal Patterns of Water Ecosystem Services

3.1.1. Water Yield

Between 2000 and 2020, the total water yield in the YRB showed a fluctuating increase, rising from 301.56 mm in 2000 to 380.70 mm in 2020, representing a 26.24% increase (Figure 2a). This indicated a marked improvement in the basin’s water-yield service capacity. Spatially, the water yield service function was ranked as follows: Zone III > Zone I > Zone II, with multiyear average values of 458.87, 434.50, and 244.42 mm, respectively.

The water yield reflects the water resource supply within a region, and its generation capacity is strongly correlated with factors such as precipitation and evapotranspiration. As shown in Figure 3a, Zone I, located in a high-altitude area, receives abundant rainfall and has low evapotranspiration, indicating a strong overall water-yield capacity ranging from 393.45 to 484.25 mm. This capacity showed elevated values in the southern regions and lower values in the northern areas, with high-altitude zones in the south exhibiting water yields exceeding 800 mm. Zone II occupies a transitional position between semi-humid temperate monsoon and semi-arid temperate continental climate zones. Featuring inadequate rainfall and elevated evapotranspiration, coupled with severe soil erosion across the Loess Plateau that hinders water yield, this region maintains an overall water yield between 204.36 and 323.20 mm, exhibiting a stepwise reduction from southeast to northwest. Zone III, situated in a plain area, saw natural vegetation replaced by farmland and urban development due to population density and accelerated industrialization, which somewhat suppressed the water yield capacity. However, as this region lies in eastern China, where precipitation is relatively high, its overall water yield remains strong and exhibits minimal spatial variation, ranging from 395.97 to 511.54 mm. Notably, this result is partly a function of the InVEST water yield module, which calculates the water yield by subtracting the actual evapotranspiration from precipitation. In areas with high impervious surfaces (e.g., construction land in Zone III), reduced evapotranspiration leads to a higher water yield, even if the hydrological connectivity and baseflow are compromised. Therefore, the simulated high-water yield does not necessarily indicate water resource abundance.

The temporal variation characteristics of water yield are shown in Figure 4a. From 2000 to 2005, the water yield in Zone I increased over a large area, especially in the central part (mainly in central Qinghai Province), with a significant increase of 100–200 mm. In contrast, Zones II and III generally exhibited a decreasing trend in water yield, with reductions ranging from −100 to 0 mm. From 2005 to 2010, the water yield in Zones I and III generally decreased, particularly in the southwestern part of Zone I (mainly in southwestern Qinghai Province), where the reduction exceeded 100 mm. Meanwhile, Zone II showed an increasing trend in water yield, with an increase of 100–200 mm in the northeastern part (mainly in the central-eastern Inner Mongolia Autonomous Region). From 2010 to 2015, the water yield decreased across the entire YRB, especially in the south-central part of Zone II (mainly in Shaanxi Province), where the reduction exceeded 100 mm. From 2015 to 2020, the water yield increased across the entire basin, particularly in the southern part of Zone I (mainly in Sichuan Province) and the central-eastern part of Zone II (mainly in Shanxi Province), with increases exceeding 200 mm.

3.1.2. Soil Conservation

Between 2000 and 2020, the soil conservation capacity across the YRB rose from 88.18 t·hm⁻² in 2000 to 114.80 t·hm⁻² in 2020, representing a 30.19% increase (Figure 2b). This demonstrated a significant enhancement in the basin’s overall soil conservation capacity. Given that vegetation growth is closely tied to precipitation patterns and directly impacts soil stabilization, the fluctuation characteristics in Zone I became particularly pronounced under the combined effects of ecological policies (e.g., the Sanjiangyuan Ecological Protection and Development Initiative and the Gansu Province Soil and Water Conservation Regulations) and nearly two decades of upstream precipitation variations. Spatially, the soil conservation service function was ranked as follows: Zone I > Zone III > Zone II, with multiyear average values of 171.90, 118.91, and 61.64 t·hm⁻², respectively.

The soil conservation capacity reflects a region’s ability to resist erosion, which is closely related to factors such as topographic slope, soil structure, water flow velocity, vegetation coverage, and human activities. As shown in Figure 3b, extensive forest and grassland areas in Zone I significantly reduced the surface runoff from precipitation combined with limited human activity, resulting in a strong soil conservation capacity ranging from 146.20 to 198.14 t·hm⁻². In contrast, Zone II generally exhibits low soil conservation levels, with notable spatial variation. For instance, the northwestern part of Zone II features loose soil, sparse vegetation, and strong winds. Resulting from the synergistic influences of climatic conditions, soil, and vegetative cover, this area has become a “lowland zone” for soil conservation in the YRB, with soil conservation capacity typically below 25 t·hm⁻². The central Loess Plateau region in southeastern Zone II, influenced by regulations such as the “Regulations on Returning Farmland to Forest” and the “Shaanxi Province Soil and Water Conservation Regulations,” has seen widespread construction of soil conservation projects, including terraces and silt dams, significantly enhancing regional soil conservation capabilities. Notably, the soil conservation capacity along the Fen River and Wei River reaches 150 t·hm⁻². The population of the lower reaches (Zone III) was relatively dense. Although long-term agricultural development and the expansion of construction land have weakened the soil’s resistance to erosion, the low, flat terrain has led to substantial sediment deposition, thereby maintaining the local soil conservation amount between 107.87 and 124.10 t·hm⁻².

Figure 4b shows the temporal variation characteristics of soil conservation. From 2000 to 2005, soil conservation in Zone I increased over a large area, especially in the central region (mainly in central Qinghai Province and southwestern Gansu Province), with an increase of more than 100 t·hm⁻². In contrast, Zones II and III showed little difference in the area of increase and decrease in soil conservation, with changes ranging from −50 to 50 t·hm⁻². From 2005 to 2010, soil conservation in Zones I and III decreased, with reductions across both zones mostly ranging from −50 to 0 t·hm⁻². Meanwhile, Zone II exhibited an increasing trend in soil conservation, with increases across the zone ranging from 0 to 50 t·hm⁻². From 2010 to 2015, soil conservation decreased across the entire YRB, especially in the central part of Zone I (mainly in Gansu Province), with a reduction of more than 50 t·hm⁻². From 2015 to 2020, soil conservation increased across the entire basin, particularly in the south-central part of Zone I (mainly in Gansu Province and Qinghai Province) and the eastern part of Zone II (mainly in Shanxi Province), with increases exceeding 100 t·hm⁻².

3.1.3. Water Purification

From 2000 to 2020, the nitrogen and phosphorus outputs in the YRB declined slightly, by 4.82% and 3.08%, respectively, indicating some progress in water pollution control (Figure 2c,d). Spatially, both nitrogen and phosphorus outputs were ranked as Zone III > Zone II > Zone I, with multiyear average nitrogen output values of 6.31, 2.72, and 1.98 kg·hm⁻², respectively, and multiyear average phosphorus output values of 0.65, 0.27, and 0.24 kg·hm⁻², respectively.

Nitrogen and phosphorus outputs serve as fundamental indicators of water purification capacity and are shaped by both natural background conditions and human activities. Spatially, both nitrogen and phosphorus displayed a clear spatial characteristic of elevated discharge in the southeast and diminished levels in the northwest (Figure 3c,d). Zone III, a traditional agricultural production area and region of rapid socioeconomic development, experiences excessive fertilizer use, land expansion, and increased impermeable surfaces, resulting in significantly higher nitrogen and phosphorus outputs than Zones I and II, ranging from 5.612 to 7.138 kg·hm⁻² and 0.591 to 0.729 kg·hm⁻², respectively. Similarly, the northeastern outskirts of Lanzhou (Zone I) and southeastern Zone II, which are densely populated urbanized areas, have experienced substantial land expansion driven by rapid industrialization and urbanization, becoming high-emission zones for nitrogen and phosphorus across the basin. Low-emission regions predominantly occurred northwest of Zone II (with scarce rainfall and poor soil) and southwest of Zone I (where extensive forests and grasslands prevailed). These regions have weaker agricultural foundations and lower levels of socioeconomic development, resulting in reduced fertilizer application intensity, moderate urbanization, and lower rates of surface hardening. These factors help maintain the ecosystem’s pollutant-retention capacity and effectively reduce nitrogen and phosphorus outputs.

As shown in Figure 4c,d, the ranges of variation in nitrogen and phosphorus outputs during the study period were between −5 and 5 kg·hm⁻² and between −0.1 and 0.1 kg·hm⁻², respectively. From 2000 to 2005, nitrogen and phosphorus outputs increased in Zone I, whereas they both decreased in Zones II and III. From 2005 to 2010, nitrogen and phosphorus outputs decreased in Zones I and III but increased in Zone II. From 2010 to 2015, nitrogen and phosphorus outputs in the YRB decreased in south-central Zone I and central Zone II, whereas they increased in Zone III and other parts of Zones I and II. From 2015 to 2020, nitrogen and phosphorus outputs increased in south-central Zone I and northeastern Zone II, whereas they decreased in Zone III and other parts of Zones I and II.

3.2. Influence of Driving Factors on Water Ecosystem Services

3.2.1. Influence of Climate

The impact of climate on the different WESs in the YRB varied significantly between 2000 and 2020 (Figure 5). The simulation results revealed that climatic factors accounted for 97.4–99.3% of the changes in water yield during this period. Notably, Scenario 1 recorded a 38.2% increase in water yield compared with the reference scenario from 2015 to 2020, indicating a substantial enhancement in water production capacity due to climate change. The climatic contribution rate to soil conservation decreased by 3.26%. This decline occurred because climate characteristics in 2005–2010 and 2010–2015 weakened soil-stabilization capacity, whereas climate fluctuations in 2000–2005 and 2015–2020 significantly improved soil-conservation services. Climate contributions to nitrogen and phosphorus outputs continued to decrease by 9.17% and 9.98%, respectively. A comparison of nitrogen and phosphorus output simulations between Scenario 1 and the reference scenario across different periods revealed that climate change has consistently enhanced water-quality purification over the past two decades. In summary, climate change dominated the changes in all WES. While climatic factors have a greater influence on water yield and soil conservation than on nitrogen and phosphorus output, their impact on these services gradually diminishes over time.

3.2.2. Influence of Land Use

With China’s heightened focus on ecological conservation, initiatives such as converting farmland to forests and grazing land to grasslands have reduced cultivated and unused land areas within the YRB from 2000 to 2020, while forest and grassland coverage have expanded. Concurrently, rapid urban development has led to a significant expansion of construction land. These changes intensified the impact of land-use transformation on WES within the basin, with pronounced spatial variations observed (Figure 6). Regarding water yield changes, land-use contributions generally decreased from east to west. Notably, downstream regions, such as Henan and Shandong, showed higher land-use contribution rates than the mid- and upper reaches due to expansion of construction land. Regarding soil conservation services, Ningxia and Inner Mongolia’s expanded forest and grassland areas have substantially enhanced local resistance to soil erosion, resulting in land-use contributions to changes in soil-conservation functions that far exceed those of the other provinces. For water quality purification, land-use factors significantly influenced variations in nitrogen and phosphorus outputs across provinces. In contrast, their impact on water yield and soil conservation remained relatively minor compared with fertilizer application and construction land expansion.

The rate of land-use change relative to changes in WES generally showed an upward trend; however, from 2005 to 2010, provinces such as Qinghai, Sichuan, Shanxi, Henan, and Shandong experienced an abnormal surge (

Table 7. Land use changes by province from 2000 to 2020/10⁴ hm².

	Land-Use Type	Qinghai	Sichuan	Gansu	Ningxia	Neimeng	Shanxi	Shaanxi	Henan	Shandong
2000–2005	Cropland	−0.52	0	−8.34	−9.38	0.09	−4.39	−17.17	−4.39	2.74
	Forest land	−0.04	0.11	5.05	2.48	6.63	−0.13	10.98	−0.2	0.02
	Grassland	−7.07	−0.1	2.05	2.53	−19.79	2.31	2.85	−0.36	−3.96
	Water area	0.3	−0.01	0.13	0.08	−0.69	0.52	0.54	2.35	0.49
	Construction land	0.28	0	1.44	1.51	2.9	1.66	3	2.85	1.33
	Unused land	7.05	0	−0.33	2.78	10.86	0.03	−0.2	−0.25	−0.62
2005–2010	Cropland	2.93	0.1	−14.68	1.2	25.3	−14.48	−19.89	−1.46	−2
	Forest land	0.03	3.53	3.42	1.53	−6.19	1.71	2.52	−0.84	−0.98
	Grassland	72	−10.7	6.07	−3.98	−12.07	−2.91	12.68	−4.76	−5.67
	Water area	1.97	0.8	0.3	0.01	−0.8	−2.06	−0.89	−0.03	1.49
	Construction land	0.49	0.23	3.46	4.96	0.82	17.75	7.68	7.31	8.77
	Unused land	−77.42	6.04	1.43	−3.72	−7.06	−0.01	−2.1	−0.22	−1.61
2010–2015	Cropland	−0.21	−0.02	−4.2	1.43	−1.61	−2.44	−0.45	−3.16	−1.89
	Forest land	0.08	−0.25	−0.2	0.03	0.08	−1.06	−0.48	0.17	0.12
	Grassland	−0.83	0.01	1.1	−3.28	−0.11	0.64	−2.24	−0.16	0.1
	Water area	0.2	−0.15	0.29	0.26	0.36	0.09	0.21	0.78	0.23
	Construction land	0.77	0	2.48	3.32	3.72	2.78	2.29	2.42	1.45
	Unused land	−0.03	0.42	0.43	−1.73	−2.42	−0.02	0.67	−0.01	0.02
2015–2020	Cropland	0.03	0	−5.31	−4.23	−26.37	0.46	−9.92	−4.57	−0.65
	Forest land	−0.33	0.07	−0.14	−0.12	8.53	−0.45	−1.65	0.81	−0.16
	Grassland	−1.4	−0.35	2.35	1.64	12.06	−3.08	3.61	−0.59	0.06
	Water area	0.68	0.36	0.21	0.3	1.42	0.32	1.21	0.03	2.08
	Construction land	1.34	0.07	4.65	3.39	11.69	2.74	7.32	4.27	−1.27
	Unused land	−0.26	−0.17	−1.7	−0.99	−7.35	0	−0.53	0.01	−0.12

). This period coincided with the implementation phase of China’s “Eleventh Five-Year Plan” (2006–2010), during which national strategies, including the Western Development, Grain for Green, and the Rise of Central China, overlapped, triggering drastic land-use changes that subsequently affected WES.

Regarding water yield, the significant increase in contribution rates in the Shanxi, Henan, and Shandong provinces was primarily attributed to large-scale urbanization. The construction land in these three provinces expanded by 17.75 × 10⁴ hm², 7.31 × 10⁴ hm², and 8.77 × 10⁴ hm², respectively, with the increase in impermeable surfaces substantially enhancing surface runoff. In contrast, Sichuan Province exhibited a distinct mechanism: the 10.7 × 10⁴ hm² of grassland was mainly converted into forest land (3.53 × 10⁴ hm²) and unused land (6.04 × 10⁴ hm²). This approximately 20 × 10⁴ hm² land conversion contributes to the elevated rate of land use to water yield changes.

Regarding soil conservation and water purification, the Qinghai, Sichuan, Henan, and Shandong provinces exhibited distinct driving mechanisms. During the peak implementation of the “Qinghai Sanjiangyuan Nature Reserve Ecological Protection and Construction Master Plan” (2004–2010), Qinghai implemented grassland restoration and conservation measures. The conversion of unused land (−77 × 10⁴ hm²) into grassland (72 × 10⁴ hm²) significantly increased vegetation cover, thereby substantially enhancing soil conservation and water purification capacities. Although Sichuan experienced grassland degradation of 10.7 × 10⁴ hm², the soil conservation benefits from newly added forest land (3.53 × 10⁴ hm²) were approximately 2–3 times greater than that of grassland. The combined positive and negative effects resulted in an unusually high land-use contribution to soil conservation. Additionally, intense land-use conversion (approximately 20 × 10⁴ hm²) amplified the rate of land-use change in nitrogen and phosphorus emissions. In Henan and Shandong, infrastructure development during peak periods occupied large areas of degraded grasslands (reducing them by 4.76 × 10⁴ hm² and 5.67 × 10⁴ hm², respectively), temporarily increasing the land-use contribution to soil conservation. The expansion of construction land in these two provinces (7–9 × 10⁴ hm²) directly increased the land-use contribution rates to nitrogen and phosphorus emissions.

This study focuses on quantifying the relative contributions of climate and land use. The contribution rate was calculated using a standard relative contribution decomposition method commonly applied in ecosystem service attribution studies. The normalized framework (R_cli + R_lu = 100%) ensures comparability of results across different time periods. Notably, when climate and land use influence a service in opposite directions, the current method (based on absolute values) captures only the magnitude of each contribution rather than the net effect. At the basin scale, opposing influences of climate and land use were observed for water yield during 2000–2005 and 2010–2015, soil conservation during 2000–2005 and 2005–2010, and nitrogen output during 2005–2010 and 2010–2015 (Table S1). This indicates that climate and land use exerted mutually offsetting impacts on the three WES indicators in these periods.

3.2.3. Identification of Dominant Factors

Factor analysis was conducted to examine the explanatory strength of the different driving factors on the spatial differentiation of water yield, soil conservation, and water purification services in the YRB (Figure 7). The findings indicated that climatic conditions, topographic characteristics, and vegetation factors were the dominant drivers of spatial variation in water yield services. In contrast, anthropogenic influences played a comparatively minor role. Precipitation demonstrated the highest explanatory powers of 80.5% and 77.1%, respectively, followed by temperature. Precipitation and topographic relief showed relatively high explanatory power for soil conservation services. This indicates that climatic and topographic factors are the key determinants of spatial heterogeneity in water yield and soil conservation services. In contrast, human activities have a greater influence on water purification services. With the implementation of measures such as farmland-to-forest conversion, the adoption of precision fertilization techniques, and the improvement of wastewater treatment facilities, land-use factors explained 55.8% and 61.3% of the spatial heterogeneity in nitrogen and phosphorus outputs, respectively. In contrast, wastewater treatment facilities accounted for 45.4% and 52.2%, respectively. These findings demonstrate that, under the combined effects of agricultural and urban non-point-source pollution, land use remains the primary factor driving spatial heterogeneity in nutrient (N and P) output across the YRB.

Factor interaction effects on WES spatial heterogeneity were determined using interactive detection (Figure 8). All interactions displayed enhancement characteristics (bivariate or nonlinear), confirming that the combined drivers exerted stronger control over spatial patterns than any single factor. For example, the explanatory power of the interaction between annual average temperature and DEM on the spatial heterogeneity of the water yield service reached 54.7% in 2000, exceeding the maximum explanatory power of either individual factor (52.9%). The interaction between topographic relief and NDVI increased the explanatory power for soil conservation service to 31.7% (2020), which is 3.6% higher than the sum of the individual explanatory powers (14.5% + 13.6% = 28.1%).

Notably, the interactions between precipitation and other factors exhibited the most prominent enhancement in explanatory power for the spatial heterogeneity of water yield service. Additionally, interactions of precipitation and topographic relief with other factors respectively enhanced the explanatory power for the spatial heterogeneity of soil conservation service, whereas interactions between land use and other factors increased the explanatory power of spatial heterogeneity in nitrogen and phosphorus outputs. Although the GDP and population density had weak individual explanatory power for each service, their interactions with other factors enhanced the explanatory power, with q-values generally exceeding 10%. This indicates that these factors influence the spatial heterogeneity of WES through synergistic effects with natural factors rather than by acting independently. Therefore, climatic and topographic factors act as “enhancers” of spatial heterogeneity in water yield and soil conservation services. In contrast, land-use factors predominantly drive spatial heterogeneity in water purification capacity.

4. Discussion

4.1. Spatial Patterns and Driving Mechanisms

Overall, the water yield, soil conservation, and nitrogen and phosphorus outputs determined in this study exhibited a clear pattern of higher values in the southeast and lower values in the northwest (Figure 3), which is consistent with previous research. For instance, Fang et al. [46] showed that the water yield and water purification capacity in the YRB decreased from southeast to northwest in 2015. Fang et al. [41] also found that from 2000 to 2016, high soil conservation values in the YRB were distributed in the southwestern mountainous areas, with all the highest values located in steep mountainous regions. Moreover, Geng et al. [47] found that from 2000 to 2018, the dominant areas of soil conservation in the YRB were mainly distributed in the plateau region of the middle and upper reaches of the basin. Although the absolute values obtained in this study differ to some extent from those reported in the aforementioned literature, they remain within the same order of magnitude. More importantly, the main spatiotemporal patterns, including the “high in the southeast, low in the northwest” spatial gradient and the relative ranking among the three topographic zones, show strong consistency with previous findings, thereby supporting the reliability of the methods adopted in this study. The discrepancies in absolute values should be interpreted in the context of differences in data sources, processing methods (e.g., resampling resolution and parameter localization), and model parameter settings, rather than as conflicting results. Furthermore, the conclusions of the geodetection analysis in this study are consistent with those of Zhu et al. [40], indicating both that climatic and geographic factors are important factors affecting the water yield and soil conservation services and that land-use type is the most important driver of the water purification capacity.

This study confirms that climate change dominates the variation in water yield services in the YRB, which strongly corroborates findings from multi-scale studies. For instance, Yang et al. [43] attributed 96.68–100.48% of water yield changes in the basin to climate change during 1995–2018, and Wu et al. [48] found that precipitation exerted a far stronger influence on water yield than land use in the Wei River Basin (a tributary of the Yellow River) during 1985–2019. These studies collectively support the scientific understanding that climatic factors play a central regulatory role in water ecological processes at the watershed scale. This study extends the time series to 2020 and simultaneously covers three core services, thereby enhancing the generalizability and timeliness of the conclusions. Regarding land-use impact, the study revealed that the contribution rates of land use to various WES have been continuously rising, with its influence on water purification capacity intensifying significantly over time, aligning with cross-regional consensus findings by Hou et al. [21] in the Shanxi YRB and by Wang et al. [24] in the Wulie River Basin. All three studies indicated that agricultural non-point source pollution (fertilizer application) and land-use changes resulting from construction land expansion are key drivers of variations in nitrogen and phosphorus outputs. This shared conclusion highlights the consistent interference of land-use changes on water environments across different basins, underscoring that “controlling non-point source pollution through optimized land-use structure” has emerged as a central challenge in watershed ecological management.

From the perspective of driving mechanisms, the dominant role of climate change in WES in the YRB stems from its comprehensive, long-term impacts. Regarding water yield services, precipitation is the primary source of replenishment for regional water resources. Its spatial distribution, characterized by “more in the southeast, less in the northwest,” directly determines the spatial gradient of water yield in the basin. Regarding soil conservation, climate change affects the service capacity through three-dimensional interactions between climate, vegetation, and soil. The escalating frequency and magnitude of extreme precipitation events enhance the erosive force of surface runoff on soil, elevating erosion risks. However, climate warming, combined with the loose texture of loess soil, triggers soil cracking and accelerates the decomposition of organic matter, thereby reducing soil erosion resistance. Drought stress leads to decreased vegetation cover and weakened root-soil stabilization. In water purification, land use alters nitrogen and phosphorus sources (e.g., chemical fertilizer application), whereas climate fundamentally determines the scale of nitrogen and phosphorus migration and the efficiency of purification through hydrological processes. From the above analysis, we observe that although climate change plays a dominant role in WES in the YRB, the impact of land use should also be prioritized. Land use is not only a direct driver of changes in WES, with its influence growing increasingly stronger, but also a modulator of the effects of climate change (e.g., through vegetation–soil–hydrology feedback processes); thus, it serves as an important bridge in the dominant role of climate change. This is evidenced by the nonlinear enhancement observed in its interactions with other factors. For example, the interaction between precipitation and land-use type explained 85.1% of the spatial heterogeneity of the water yield service in 2000, exceeding the sum of their individual explanatory powers (80.5% for precipitation and 3.8% for land use, totaling 84.3% theoretically), indicating a nonlinear enhancement effect.

4.2. Contributions and Limitations

In this study, we overcome the limitations of previous research, which mostly focused on general ecosystem services (e.g., water yield, soil conservation, habitat quality, carbon storage, food production, etc.) in specific regions [30]. By focusing on WESs, we systematically assess three key services across the YRB: water yield, soil conservation, and water purification. A comprehensive assessment framework of “water quantity–water quality–water ecology” is established, which provides a more comprehensive understanding of the progression of water ecological functions across the basin. For the first time, it quantified the temporal dynamics of how factors such as climate and land-use change influence various WES across the entire YRB, clarifying a new pattern: continuously rising land-use contribution rates with more prominent impacts on water quality. Compared with climate change, land-use change is more readily driven by human production and lifestyle choices as well as land-use policies, making it more controllable. Therefore, considering research on the management of the water ecosystem in this basin, further studies on land use should be performed, for example, focusing on the different effects of management intensity within the same land-use type, including control measures, such as fertilizer application rates. Additionally, interactive detection using geographic detectors revealed the mechanism of “climate-terrain-land use” multi-factor synergy that amplifies spatial differentiation of services, addressing the shortcomings of previous single-factor analyses.

However, this study has some limitations that require further exploration in future research. First, while the InVEST model is widely used in ecosystem service studies, it has limitations in characterizing complex terrains. To meet the model’s operational requirements, we resampled all datasets to a 1 km resolution. Although this approach is reasonable for basin-scale analysis, it inevitably introduces aggregation effects. For example, important terrain features captured at finer resolutions are lost, especially in mountainous and highly dissected areas. This affects slope-related calculations and consequently soil conservation estimates. The same issue applies to hydrological routing and nutrient transport, where a coarse resolution may distort flow paths. Meanwhile, model parameters, such as vegetation evapotranspiration coefficients, rely on basin-wide averages, ignoring subregional differences that may affect the simulation accuracy. Moreover, the model inherently links water yield to climatic variables and explicitly states that it does not differentiate among surface flow, subsurface flow, and baseflow but instead assumes that all water yield from a pixel reaches the point of interest through one of these pathways. This implies that the contribution of land use to water yield in this study may be underestimated, as the model does not explicitly account for hydrological effects mediated through specific pathways (e.g., groundwater recharge or infiltration). However, this limitation falls beyond the analytical scope of the water yield model. Consequently, the conclusion that “climate plays a dominant role” raises certain doubts. Future studies must integrate the Soil & Water Assessment Tool (SWAT) model (specialized in the simulation of detailed hydrological processes) for multi-model coupling and validation and optimize the parameter spatial differentiation settings through sensitivity analysis. Second, the scenario simulation in this study follows established methods to separate the effects of climate and land use, focusing on the relative contribution rates of climate versus land use. Consequently, the method for calculating the contribution rates ignores the direction of change. Future studies should further investigate the influence of the direction of change on the contribution rates. Third, current assessments of water purification only quantify non-point-source nitrogen and phosphorus exports through surface runoff. Although this approach adequately captures phosphorus transport, it may underestimate nitrogen loading owing to subsurface mobility. Additionally, the effects of point-source pollution (industrial and domestic sewage) are not incorporated in this study, and the impact of aquatic biodiversity on purification capacity is not considered. Subsequent research should incorporate groundwater modules to address nitrogen pathways, integrate point sources to establish a dual-source assessment framework, and couple these with water quality models (e.g., dissolved oxygen) to expand the evaluation dimensions. While existing studies have identified the dominant roles of climate and land use, they often do not quantify the marginal impacts of distinct land-use transitions (e.g., converting farmland to forest or grassland to construction land) on services or clarify the thresholds for synergistic interactions between climatic factors (precipitation and temperature) and land-use factors. Future research could adopt a dual-scale refinement approach (sub-basin to county level) to quantify the service effects of land-use transitions using intensity analysis models while identifying critical thresholds for factor synergies through threshold detection. Fourth, to address the need for quantitative validation, we compared the simulated outputs with observed data from key hydrological stations (Figure S1). The simulated annual water yield showed a strong correlation (R = 0.85, p < 0.1) with measured runoff at Lijin Station (Yellow River estuary). Simulated actual soil erosion (USLE) was highly correlated with the observed sediment load at the same station (R = 0.94, p < 0.05). For water purification, simulated total nitrogen export exhibited a downward trend consistent with measured ammonia nitrogen concentrations at Luokou Station during 2005–2015. These comparisons confirm that the model simulations capture the dominant temporal dynamics of water yield, soil conservation, and water purification in the YRB, despite the inherent uncertainties discussed above. Future studies should conduct more detailed station-by-station calibration using higher-frequency data.

Notably, in this study, we used five discrete years (2000, 2005, 2010, 2015, and 2020) to represent two decades of change. Although this 5-year interval captures decadal trends, it may ignore interannual variability and extreme events (e.g., droughts or floods) that can disproportionately affect the water yield and soil conservation. Furthermore, the InVEST nutrient delivery module operates on an annual time step, which cannot capture the seasonal dynamics of fertilizer application and runoff. Future studies should consider higher temporal resolutions (e.g., monthly or daily) and incorporate climate extremes to better capture the full range of variability in WES.

4.3. Policy Implications

The pronounced spatial heterogeneity of WES across the three zones of the YRB calls for zone-specific management strategies. Zone I (headwater area) features high water yield and soil conservation but extremely low nutrient exports. This region is characterized by ecological vulnerability and high climate sensitivity (e.g., glacier retreat, local urban pollution, etc.). Therefore, the adjustment strategy for Zone I should prioritize climate-adaptive ecological protection, including the strict restriction of grazing and infrastructure expansion in alpine meadows and wetlands, establishment of long-term glacier-permafrost monitoring networks, and the control of urban non-point source pollution by improving sewage treatment and limiting impervious surface growth. Zone II (Loess Plateau) exhibits the lowest water yield and soil conservation and moderate nutrient outputs. The main issues in this region are severe soil erosion and prominent agricultural non-point source pollution. Therefore, the adjustment strategy for Zone II should mainly prioritize the continuous strengthening of soil and water conservation projects (terraces, check dams, afforestation, among others) in erosion-prone areas, implementation of precision fertilization and controlled-release fertilizers in core agricultural areas, optimization of the land-use structure by converting steep sloping farmland to forest or grassland, and construction of riparian buffer strips along tributaries to intercept agricultural pollutants. Zone III (lower plain) shows the highest water yield and nutrient outputs. This region is characterized by rapid urbanization, expansion of impervious surfaces, and increasing pressure on water purification. Therefore, the recommended strategy for Zone III should include the strict control of the expansion of construction land, promotion of the development of sponge cities, upgrading of the discharge standards of wastewater treatment plants, promotion of precision fertilization and controlled-release fertilizers, and restoration or construction of wetland parks along the Yellow River and its tributaries to enhance the natural purification capacity. At the basin scale, to address the risks of extreme precipitation and drought caused by climate change, the flood control and drought relief engineering system across the entire basin should be improved. Additionally, a cross-zonal coordination mechanism is required. Upstream regions (Zones I and II) that implement soil conservation and pollution control can be compensated by downstream Zone III through a water-quality trading or ecological compensation scheme.

5. Conclusions

Based on the InVEST model, scenario simulation, and the geodetector method, this study analyzed the spatiotemporal evolution of water yield, soil conservation, and water purification capacity in the YRB from 2000 to 2020 and quantified the individual and interactive effects of climate and land-use changes on these WES. Our findings provide a quantitative basis for differentiating climate versus land-use interventions in water resource management and designing spatially targeted ecological restoration policies in the YRB and similar regions worldwide that are sensitive to climate change, span multiple climatic zones, and experience significant human activities. The main conclusions are as follows.

(1)

From 2000 to 2020, water yield and soil conservation in the YRB showed generally increasing trends with interannual fluctuations, rising by 26.24% and 30.19%, respectively. In contrast, nitrogen and phosphorus outputs declined slightly by 4.82% and 3.08%, indicating a modest improvement in water purification. Spatially, all three services exhibited higher values in the southeast and lower values in the northwest, with clear distinctions among the three topographic zones. The multiyear average water yields in Zones I, II, and III were 434.50, 244.42, and 458.87 mm; the soil conservation amounts were 171.90, 61.64, and 118.91 t·hm⁻²; the nitrogen outputs were 1.98, 2.72, and 6.31 kg·hm⁻²; and the phosphorus outputs were 0.24, 0.27, and 0.65 kg·hm⁻², respectively.

(2)

Climate is a pivotal regulatory driver of water ecosystem change in the YRB. The contribution rates of climate change to variations in water yield, soil conservation, nitrogen export, and phosphorus export were 97.4–99.3, 94.5–98.3, 87.2–96.0, and 85.7–95.2%, respectively. Further analysis revealed that while the influence of climate on basin services is diminishing, land-use impacts are increasing over time, with particularly pronounced effects on nitrogen and phosphorus outputs. This highlights the necessity of giving greater attention and focus to land use in future research.

(3)

The interaction between different influencing factors increases the explanatory strength of individual factors in explaining the spatial heterogeneity of WES. Specifically, climate and topography, when coupled with other factors, markedly intensify spatial variation in water yield and soil conservation services. The interaction between land use and other factors further accentuated the spatial heterogeneity of the water purification capacity within the basin.

(4)

Regionally differentiated policy strategies are essential. In Zone I (headwater area), climate-adaptive ecological protection should be prioritized, including grazing restrictions, glacier-permafrost monitoring, and control of urban non-point source pollution. In Zone II (Loess Plateau), soil and water conservation projects, precision fertilization, land-use structure optimization, and riparian buffers should be prioritized. In Zone III (lower plain), management strategies must include the control of construction-land expansion, promotion of sponge city development, upgrading of wastewater treatment, implementation of precision fertilization, and restoration of wetland parks. Across the basin, a cross-zonal coordination mechanism (e.g., water quality trading or ecological compensation) and an improved flood-drought engineering system are required to address climate-induced extremes.