Spatiotemporal Dynamics and Trade-Off Analysis of Ecosystem Services in the Caijiachuan Watershed of the Loess Plateau

Abstract

As a typical reforested region of the Loess Plateau, the Caijiachuan watershed plays a vital role in ecological security and resource management. This study evaluates the spatiotemporal variations in key ecosystem services—namely soil retention, water yield, carbon storage, and habitat quality—between 2002 and 2024 using the InVEST model, calibrated with field-measured rainfall, carbon density, and high-resolution land use data derived from integrated remote sensing and field surveys. Statistical analyses based on the R language reveal dynamic trade-offs and synergies among these services. The results show that: (1) soil retention, carbon storage, and habitat quality have steadily improved, while water yield shows an overall upward trend with significant spatial heterogeneity; (2) a consistent and significant trade-off exists between carbon storage and water yield (average R² ≈ 0.28), while other ecosystem service interactions are relatively weak; (3) climatic variability, topographic heterogeneity (e.g., slope and elevation), and vegetation structure are key drivers of these trade-offs. This study provides scientific evidence to support ecological management and policy formulation in reforested areas of the Loess Plateau.

1. Introduction

Ecosystem services (ESs), which form a vital bridge between natural ecosystems and human well-being, have emerged as a central theme in contemporary research across ecology, geography, and sustainability science [1]. Investigating ESs not only enhances environmental quality but also offers effective solutions to pressing challenges such as resource scarcity [2,3,4]. Amid accelerating population growth, urbanization, and climate change, ecosystem functions are increasingly under pressure, leading to service degradation and a growing mismatch between supply and demand [5]. In ecologically fragile regions, human-induced disturbances to ecosystem structure and function often exacerbate complex trade-offs among multiple services [6]. Under these conditions, scientifically assessing the interactions among ESs and elucidating their spatial patterns and dynamic processes is essential for achieving regional ecological sustainability and informing effective policy design [7].

The Loess Plateau is one of China’s most ecologically fragile regions and a major hotspot for soil erosion [8]. For decades, it has grappled with severe challenges such as land degradation and ecological imbalance. In response, China initiated the Grain for Green Program (GGP) in 1999, which has achieved remarkable success across the region. The program has significantly increased vegetation cover, improved soil structure and microclimatic conditions, and enhanced a range of ecosystem services, including water retention, carbon sequestration, and biodiversity [9,10,11]. However, ongoing ecological restoration has also given rise to new environmental challenges, such as simplified forest structures, reduced water supply capacity, and localized biodiversity loss [12]. These emerging issues reflect fundamental trade-offs among ecosystem services, whereby the enhancement of one service may occur at the expense of others [13].

The trade-offs and synergies among ecosystem services exhibit significant spatial heterogeneity and scale dependence. For example, vegetation recovery in reforested areas of the Loess Plateau has been shown to improve carbon storage and soil retention, while potentially decreasing water yield as a result of increased evapotranspiration [14]. Moreover, the interactions among ecosystem services are shaped by both natural and anthropogenic factors, including land use patterns, slope gradients, soil types, and rainfall intensity [15]. Consequently, it is essential to conduct quantitative analyses and scenario-based simulations of multiple ecosystem services at the regional scale using spatially explicit modeling approaches, in order to provide scientific support for ecological spatial planning and functional zoning.

Currently, several mature tools are available internationally for the quantitative assessment of ecosystem services, including the Integrated Valuation of Ecosystem Services and Trade-offs (InVEST) model, ARIES, and SolVES. Among these, InVEST is widely applied in fields such as land use scenario assessment, ecological compensation, and spatial planning, owing to its modular structure, moderate data requirements, and strong spatial visualization capabilities [16]. In China, scholars have employed the InVEST model to analyze individual or combined services—such as carbon storage, water retention, and soil conservation—particularly in studies focused on the Loess Plateau. However, systematic evaluations of ecosystem multifunctionality and trade-off mechanisms remain limited, especially comprehensive quantitative analyses conducted at the county or watershed scale [17].

This study focuses on the Caijiachuan watershed, a representative Grain for Green region in the Loess Plateau, and utilizes the InVEST model’s carbon storage, water yield, habitat quality, and soil retention modules. By integrating field-measured rainfall and carbon density data, it conducts a comprehensive analysis of the spatiotemporal dynamics and trade-off mechanisms of ecosystem services from 2002 to 2024. The objectives of this study are to: (1) assess the spatiotemporal variations and spatial heterogeneity of carbon storage, water yield, soil retention, and habitat quality during 2002–2024; (2) examine the long-term trade-off relationship between carbon storage and water yield, along with its evolutionary trend during ecological restoration; (3) identify the combined effects of topography, climate, and vegetation structure on the carbon–water trade-off mechanism.

Unlike many previous studies that relied primarily on secondary data sources or coarse-scale land use classifications, this study integrates high-resolution remote sensing imagery, original afforestation planning maps, and extensive field survey data including 36 vegetation plots with detailed tree measurements and soil properties such as particle size distribution, organic matter content, and bulk density. This approach enables the construction of a fine-scale (over 3000 patches) land use vector map and allows for the localized parameterization of the InVEST model using interpolated field-measured rainfall and soil carbon density data. These methodological enhancements substantially improve the accuracy and regional relevance of ecosystem service assessments in the Caijiachuan watershed.

2. Materials and Methods

2.1. Study Area Overview

The Caijiachuan watershed (Figure 1) is located on the western slope of the Lüliang Mountains in the central Loess Plateau of China, in Jixian County in Shanxi Province. Its geographic coordinates are 36°14′27″–36°18′23″ N and 110°39′45″–110°47′45″ E. The watershed covers an area of approximately 39 km². The region is characterized by pronounced topographic variation, with elevations ranging from 950 m to 1700 m and a general slope descending from northwest to southeast. The area experiences a warm temperate semi-arid climate, with an average annual precipitation of about 540 mm, mostly occurring during the flood season from July to September. The mean annual temperature is around 10 °C. The soils in the region are classified as cinnamon soils developed from loess parent material and are generally alkaline in nature [18].

The Caijiachuan watershed serves as a typical example of the “Grain for Green” reforestation model implemented in the Loess Plateau during ecosystem restoration. Land use types in the watershed are primarily composed of forest land, grassland, and a small proportion of cropland. Among them, artificial forests dominate vegetation restoration, mainly consisting of Robinia pseudoacacia, Pinus tabuliformis, and Platycladus orientalis, while natural secondary forests are primarily composed of Quercus liaotungensis [19]. Since 2000, the implementation of the Grain for Green Program in this area has significantly altered the ecological landscape and enhanced ecosystem service functions. In recent years, with the continued advancement of ecological restoration measures, the region has exhibited positive trends in carbon storage, habitat quality, and soil retention, along with an improved water regulation capacity.

2.2. Vegetation Survey and Sample Collection

From June to August 2024, comprehensive field surveys were conducted across the Caijiachuan watershed to support ecosystem service modeling. The land use classification (Figure 2) comprised 12 representative types, including Robinia pseudoacacia, Pinus tabuliformis, Platycladus orientalis, and Quercus liaotungensis forests; mixed forests (Robinia–Platycladus, Robinia–Pinus); cropland; orchards; shrubland; grassland; check dams; and unused land. Among these, 10 dominant types were selected for sampling, while check dams and unused land were excluded due to their limited spatial extent and low ecological relevance.

To ensure spatial representativeness across major topographic and vegetation gradients, a stratified sampling approach was employed. Plot allocation was determined by the area and spatial distribution of each land use type. Due to field limitations and land cover fragmentation, a fully randomized design was not feasible. Instead, sampling intensity followed established protocols from comparable watershed-scale studies in the Loess Plateau [20].

Given their limited extent, mixed forests and orchards were each represented by three plots. Dominant pure forest types were sampled proportionally: 14 plots for Robinia pseudoacacia, 8 for Pinus tabuliformis, 5 for Quercus liaotungensis, and 3 for Platycladus orientalis. The plot sizes adhered to standard ecological survey protocols: 20 × 20 m for forest plots, 3 × 3 m for shrublands, and 1 × 1 m for herbaceous communities [21,22].

In each forest plot, species composition, diameter at breast height (DBH), basal diameter, and tree height were recorded. Above-ground biomass was estimated using species-specific allometric equations and converted to carbon using a standard factor of 0.5. Surface litter was collected from three randomly placed 50 × 50 cm subplots, oven-dried at 65 °C, and converted to carbon using the same factor [23,24,25].

Below-ground biomass was estimated by selecting one representative tree per plot. A 1 × 1 m pit (~1 m deep) was excavated around the stump, and roots were extracted, washed, oven-dried, and weighed. The biomass was converted to carbon using a factor of 0.5 [21,26,27,28,29].

In shrub plots, three individuals were randomly selected and excavated entirely. Above- and below-ground biomass were dried and weighed separately, and carbon was calculated using the same factor. Herbaceous biomass was assessed in three randomly placed 1 × 1 m quadrats per plot. Above-ground parts were clipped at ground level, and roots were excavated to ~30 cm depth. All samples were oven-dried and weighed, with carbon content calculated using a conversion factor of 0.45 [30].

At the same time, three standard soil profiles were excavated within each plot. Soil samples were collected at six depth intervals: 0–10 cm, 10–20 cm, 20–40 cm, 40–60 cm, 60–80 cm, and 80–100 cm. For each layer, undisturbed soil cores were obtained using a ring knife to determine bulk density, while disturbed samples were air-dried and passed through a 2 mm sieve for particle size analysis and organic matter content measurement. The soil organic carbon density (SOCD) of each layer was then calculated based on the measured soil organic carbon content and bulk density [31,32].

$${SOCD}_i={BD}_i\times D_i\times {SOC\%}_i\times 10$$

(1)

where ${BD}_i$ is the bulk density (g/cm³), $D_i$ is the thickness of the soil layer (cm), and SOC% is the soil organic carbon content (%) for the layer. The multiplication by 10 converts the units to Mg C ha⁻¹.

Total soil organic carbon density in the 0–100 cm profile was obtained by summing the carbon densities of all layers as follows:

$${SOCD}_{total}=\sum _{i=1}^6{SOCD}_i$$

(2)

This layered method provides a detailed and vertically integrated assessment of soil carbon storage, capturing the variability in both the soil properties and organic carbon content across depth.

Field-measured vegetation and soil data were used to parameterize the carbon storage and soil retention modules in the InVEST model. In addition, basic topographic metrics (elevation and slope) were recorded for each plot to support the spatial pattern analysis of ecosystem services. Complementing the 2024 field campaign, historical data from 2002 and 2012 were compiled from long-term ecological monitoring programs and land use records, including remote sensing imagery, climate observations, and previous vegetation surveys. These datasets provided consistent input parameters for InVEST simulations across all years, ensuring the temporal comparability of ecosystem service estimates.

2.3. Data Sources for Remote Sensing and Model Inputs

2.3.1. Land Use Data and Classification Approach

A total of 3000 micro-patches in the Caijiachuan watershed were delineated based on intensive field surveys, providing a detailed spatial foundation for land use classification. By comparing high-resolution remote sensing imagery (QuickBird for 2002 and 2012; Jilin-1 for 2024) with historical afforestation planning maps, it was observed that land use patterns exhibited limited change over time. Therefore, a unified land use classification scheme was adopted across all three years. Each micro-patch was assigned to one of twelve standardized land use categories, based on vegetation type and field-verified land use status. This approach ensured high thematic accuracy and temporal consistency, supporting robust input preparation for the InVEST model.

2.3.2. Digital Elevation Model (DEM)

A 1 m × 1 m resolution DEM was developed by integrating detailed topographic maps with high-precision field survey data. The derived terrain variables—elevation, slope, and aspect—were used as essential inputs for the InVEST soil retention and water yield models and facilitated the spatial interpretation of ecosystem service patterns across the heterogeneous landscape.

2.3.3. Meteorological Data Collection

Daily precipitation data were obtained from 18 long-term automated monitoring sites located within the Caijiachuan Field Observation Station (Jixian County, Shanxi Province) covering the period 2002–2024. These stations are part of an integrated monitoring network under the Shanxi Caijiachuan Field Scientific Observation and Research Station. To enhance temporal continuity and spatial representativeness, additional precipitation records were compiled from provincial statistical yearbooks and the peer-reviewed literature. This comprehensive dataset provided critical inputs for hydrological modeling, particularly for simulating the spatiotemporal dynamics of water yield.

2.3.4. Evapotranspiration

Evapotranspiration inputs were derived from gridded remote sensing products provided by the China Meteorological Administration (CMA). These products reflect the regional hydroclimatic conditions of the Loess Plateau and were used for model calibration and validation.

2.3.5. Soil Parameters

Plant-available water content (PAWC) was extracted from the Harmonized World Soil Database (HWSD v1.2), with regional values ranging from 0.14 to 0.18 mm/mm. These parameters were incorporated into the InVEST water yield model to characterize soil water-holding capacity, an essential determinant of runoff generation and ecosystem water regulation.

2.4. InVEST Model Assessment

The Integrated Valuation of Ecosystem Services and Trade-offs (InVEST) model, in combination with ArcGIS 10.8 and R (4.3.3) software, was used to assess habitat quality, carbon storage, water yield, and soil retention in the Caijiachuan watershed, along with their spatial distribution patterns. To enhance model accuracy and regional applicability, input parameters were calibrated using extensive field-measured data. Specifically, rainfall erosivity factors were derived through the spatial interpolation of observed precipitation data, while carbon densities for various land cover types were determined from plot-level vegetation surveys and soil carbon measurements. These calibrated parameters enabled a more localized and robust assessment of ecosystem service dynamics under the Grain for Green Program.

2.4.1. Carbon Storage

Carbon storage was simulated using the InVEST Carbon Storage and Sequestration module, which quantifies the carbon stocks associated with different land use types. Given the difficulty of accurately estimating atmospheric carbon fluxes, this study focused on four terrestrial carbon pools: above-ground biomass, below-ground biomass, soil organic carbon, and dead organic matter.

Field-derived carbon density values (Mg C ha⁻¹) for each pool were assigned to corresponding land use categories and integrated with land use maps to produce spatially explicit estimates of total carbon storage.

The total carbon storage was calculated using the following formula [33]:

$$C_{total}=C_{above-ground}+C_{below-ground}+C_{soil}+C_{dead-carbon}$$

(3)

Each carbon pool was estimated using standardized approaches informed by field measurements and published methodologies.

Above-ground carbon was calculated from tree, shrub, and herbaceous biomass using species-specific allometric equations, then converted to carbon using a standard carbon fraction. Below-ground carbon for 2002 and 2012 was estimated using root-to-shoot ratios specific to vegetation types, while the 2024 values were based on direct excavation and biomass measurements. Soil organic carbon was estimated from composite soil samples collected at 0–100 cm depth, averaged by land use category. Dead organic carbon was determined from surface litter and coarse woody debris sampling, then converted using a biomass-to-carbon factor.

All carbon stock estimates were mapped to their respective land use units, enabling the spatial analysis of carbon storage dynamics over time. Detailed sampling and calculation protocols are described in Section 2.2 (Vegetation Survey and Sample Collection).

2.4.2. Water Yield

• Model Overview

The InVEST water yield model is grounded in the Budyko hydrological framework, which simulates annual water yield by integrating climatic conditions and landscape characteristics. Specifically, the model calculates water yield at the pixel level as the difference between annual precipitation (P) and actual evapotranspiration (AET), using a modified Budyko curve. Core input parameters include plant-available water content (PAWC), the seasonality factor (Z), and root-restricting layer depth (RZD). The model estimates water yield using the following equation [33]:

$$Y_{xj}=(1-{\displaystyle \frac{{AET}_{xj}}{P_x}})\times P_x$$

(4)

where Y_xj and AET_xj represent the annual average water yield and actual evapotranspiration for the j-th land use type in the x-th grid cell, respectively; P_x is the annual precipitation of the x-th grid cell. Total watershed water yield is derived by aggregating values from all raster cells.

2.

Parameterization Strategy

Plant-available water content (PAWC) was defined as the difference between field capacity and wilting point, representing the fraction of soil water that can be absorbed by plant roots. PAWC values were extracted from the Harmonized World Soil Database (HWSD v1.2).

Root-restricting layer depth (RZD) was assigned based on vegetation-specific rooting characteristics to reflect differences in soil water uptake capacity. For instance, deep-rooted species such as Robinia pseudoacacia and Quercus liaotungensis were assigned RZD values of 2000–3300 mm and 1500–2000 mm, respectively [34]; shallow-rooted species, including grasslands and Platycladus orientalis, were assigned values of 300–500 mm and 800–1000 mm, respectively [35]; and mixed forests were assigned intermediate RZD values [36].

The seasonality parameter (Z) characterizes spatial variation in climate seasonality and vegetation water use patterns. A higher Z value indicates stronger seasonality, which typically results in lower runoff and higher evapotranspiration. This parameter allows the model to reflect differences in vegetation physiology, phenology, and soil moisture dynamics, enhancing the ecological realism of water yield simulations.

In this study, a Z value of 1.5 was adopted based on the recommendations of Donohue et al. (2012) [37], who suggest that moderately seasonal climates in semi-arid regions are best represented by Z values between 1 and 2. This value is consistent with existing water yield modeling studies in the Loess Plateau and northern China, where seasonal precipitation and deep-rooted vegetation types dominate the landscape [38,39]. The choice of Z = 1.5 ensures a realistic balance between evapotranspiration and runoff generation in the Caijiachuan watershed.

2.4.3. Soil Retention

To evaluate soil retention services in the Caijiachuan watershed, this study employed the Universal Soil Loss Equation (USLE) embedded within the InVEST model framework. The key parameters—rainfall erosivity (R), soil erodibility (K), topographic factor (LS), cover management factor (C), and support practice factor (P)—were primarily derived from field measurements and localized datasets, ensuring high model reliability under the region’s heterogeneous terrain and land use patterns.

Monthly precipitation data from 2002, 2012, and 2024 were used to calculate R values, fully reflecting the spatiotemporal variability in rainfall erosivity based on observed meteorological data. The LS factor was derived from a high-resolution 1 × 1 m Digital Elevation Model (DEM), which significantly improved the accuracy of slope length and steepness estimation in complex hill–valley terrain, in line with recent findings emphasizing the importance of high-resolution terrain data in erosion modeling [40]. The C factor was obtained from site-specific vegetation surveys and the Caijiachuan field monitoring dataset, while the P factor was determined from conservation practice records and verified through on-site field investigation [41].

Soil retention (ΔUSLE) was calculated by comparing potential soil loss under bare land conditions with actual soil loss using the following standard equation [42]:

$$\Delta USLE=RKLS-USLE=R\times (1-C-PS)$$

(5)

This approach, grounded in empirical field data and fine-scale topographic analysis, allows for a more realistic and spatially refined assessment of soil retention capacity in the ecologically fragile Loess Plateau region.

• Rainfall erosivity factor (R):

In this study, the R factor was calculated using monthly precipitation data from 18 long-term automated monitoring stations distributed across the Caijiachuan watershed. These stations are part of a regional hydrometeorological network established to capture localized rainfall dynamics with a high temporal resolution. This approach provides greater spatial accuracy than interpolated data from distant meteorological stations, which is essential for modeling soil erosion in the complex terrain and highly erodible soils of the Loess Plateau.

The R factor was computed using a modified empirical equation suitable for Chinese rainfall patterns, as proposed by Liu et al. [42], as follows:

$$R={\displaystyle {\sum }_{i=1}^{12}}\left[1.735\times {10}^{\left(1.5\times \log \left({pi}^2/p-0.8188\right)\right)}\right]$$

(6)

where pi = monthly precipitation (mm); $p$ = annual precipitation (mm); and the resulting unit of R is 100 ac·ha⁻¹, which follows the traditional unit system used in earlier erosion studies and is compatible with local empirical models and previous InVEST applications in China. For applications requiring international units such as MJ·mm·ha⁻²·h⁻¹·a⁻¹, a conversion factor of 17.02 can be applied if needed, but was not used in this study in order to maintain consistency with prior research using unconverted R values.

2.

Soil Erodibility Factor (K):

The soil erodibility factor (K) reflects the susceptibility of soil particles to detachment and transport by rainfall and runoff. In this study, the K factor was estimated using the empirical formula from the Erosion Productivity Impact Calculator (EPIC) model, which incorporates soil texture and organic carbon content (Sharpley & Williams, 1990) [43].

The K factor was computed using the following expression [42,43]:

$$\begin{gathered} k=\left\{0.2+0.3\times exp\left[-0.0256SAN\left(1-SIL/100\right)\right]\right\}\times {\left[SIL/\left(CLA+SIL\right)\right]}^{0.3}\times \\ \left\{{\displaystyle \frac{0.25C}{SOC+exp\left(3.72-2.95soc\right)}}\right\}\times \\ \left\{\begin{array}{c}1.-0.7\times \left(1-SAN/100\right)/\\ \left[1-SAN/100+exp\left[22.9\times \left(1-SAN/100\right)-5.51\right]\right]\end{array}\right\} \end{gathered}$$

(7)

where SIL, SAN, and CLA represent the percentages of silt, sand, and clay content, respectively and SOC represents the soil organic carbon content.

3.

LS Factor—Topographic Index

The LS factor was computed from a 1 × 1 m resolution DEM using ArcGIS, employing the Moore and Burch (1986) method to account for both slope length and gradient as follows [44]:

$$LS={\left({\displaystyle \frac{flow accumulation\times cell size}{22.13}}\right)}^{0.4}\times {\left({\displaystyle \frac{\mathit{sin}slope in degrees}{0.0896}}\right)}^{1.3}$$

(8)

4.

Cover Management Factor (C):

The C factor represents the effect of vegetation cover and management practices on soil erosion rates. Rather than relying on remote sensing proxies such as NDVI, this study employed the classical classification framework proposed by Wischmeier and Smith (1978), with modifications tailored to Chinese ecological conditions following the recommendations of Liu et al. (2002) [45,46].

Field measurements of canopy cover and understory vegetation density were conducted across different land use types in the Caijiachuan watershed, and used to calibrate C factor values to better reflect local vegetation structure and ecological management practices.

Specifically, C values were assigned based on observed vegetation characteristics as follows: Natural secondary forests—C = 0.001–0.01, depending on tree canopy closure and litter layer thickness. Plantations—C = 0.01–0.05, with a consideration of monoculture effects and species-specific canopy structures. Grasslands—C = 0.02–0.1, classified by the development degree of the grass mat and ground cover continuity [45,47].

This empirical calibration approach enhances the ecological realism of the model and improves its applicability in data-rich, fine-resolution studies such as this one.

5.

Support Practice Factor (P)

The support practice factor (P) accounts for the impact of soil conservation practices (e.g., terracing, contour farming, strip cropping). In this study, p values were derived from a combination of field investigation and land use records. Terraced agricultural lands were assigned values based on slope and practice type, following classification tables developed for the Loess Plateau by Liu et al. [42]. These values were verified through field validation in 2024 [42].

2.4.4. Habitat Quality Module

The habitat quality module of the InVEST model was used to assess the spatial patterns of ecological degradation and habitat integrity in the Caijiachuan watershed. This module integrates land use/land cover (LULC), anthropogenic threat factors, and habitat sensitivity to simulate relative habitat quality. The output is a continuous index ranging from 0 (severely degraded) to 1 (pristine condition), representing the landscape’s capacity to support biodiversity [33].

Two key input tables were developed for the model:

Sensitivity table (

Table 1. 2002 sensitivity of LULC types to threats.

LULC	Habitat	Roads	Farmland	Orchards	Silt Dam
Grassland	0.6	0.95	0.9	0.8	0.7
Platycladus orientalis	0.5	0.8	0.7	0.6	0.5
Mixed Cedar–Robinia forest	0.6	0.8	0.7	0.6	0.5
Robinia pseudoacacia	0.6	0.8	0.7	0.6	0.5
Mixed Robinia–Pine forest	0.65	0.8	0.7	0.6	0.5
Cropland	0.25	0.7	0.8	0.7	0.8
Shrubland	0.6	0.7	0.7	0.7	0.7
Orchard	0.35	0.6	0.8	0.9	0.7
Quercus liaotungensis	0.7	0.9	0.8	0.7	0.6
Unused land	0.1	0.4	0.4	0.6	0.8
Pinus tabuliformis	0.65	0.8	0.8	0.6	0.5
Check dam land	0.5	0.8	0.8	0.8	0.95

,

Table 2. 2012 sensitivity of LULC types to threats.

LULC	Habitat	Roads	Farmland	Orchards	Silt Dam
Grassland	0.7	0.9	0.8	0.7	0.6
Platycladus orientalis	0.6	0.7	0.6	0.5	0.4
Mixed Cedar–Robinia forest	0.7	0.7	0.6	0.5	0.4
Robinia pseudoacacia	0.7	0.7	0.6	0.5	0.4
Mixed Robinia–Pine forest	0.75	0.7	0.6	0.5	0.4
Cropland	0.35	0.6	0.9	0.8	0.7
Shrubland	0.75	0.6	0.6	0.6	0.6
Orchard	0.55	0.5	0.7	0.9	0.6
Quercus liaotungensis	0.8	0.8	0.7	0.6	0.5
Unused land	0.2	0.3	0.3	0.5	0.7
Pinus tabuliformis	0.75	0.7	0.6	0.5	0.4
Check dam land	0.5	0.7	0.7	0.7	0.9

and

Table 3. 2024 sensitivity of LULC types to threats.

LULC	Habitat	Roads	Farmland	Orchards	Silt Dam
Grassland	0.8	0.8	0.7	0.6	0.5
Platycladus orientalis	0.7	0.6	0.5	0.4	0.3
Mixed Cedar–Robinia forest	0.8	0.6	0.5	0.4	0.3
Robinia pseudoacacia	0.8	0.6	0.5	0.4	0.3
Mixed Robinia–Pine forest	0.85	0.6	0.5	0.4	0.3
Cropland	0.45	0.5	0.8	0.7	0.6
Shrubland	0.85	0.5	0.5	0.5	0.5
Orchard	0.65	0.4	0.6	0.8	0.5
Quercus liaotungensis	0.9	0.7	0.6	0.5	0.4
Unused land	0.3	0.2	0.22	0.4	0.6
Pinus tabuliformis	0.85	0.2	0.5	0.4	0.3
Check dam land	0.5	0.6	0.6	0.6	0.8

): This matrix defines the sensitivity of each LULC type to specific anthropogenic threats such as roads, farmland, orchards, and check dams. Each row corresponds to a LULC type, while each column represents a threat factor. The values range from 0 to 1, with higher values indicating greater sensitivity to a given threat.

Threat table (

Table 4. 2002 threat type.

Threat	Weight	Max Dist	Decay
Orchard	0.7	700	Exponential
Roads	0.8	2500	Exponential
Dam	0.6	2500	Linear
Farm	0.75	1500	Linear

,

Table 5. 2012 threat type.

Threat	Weight	Max Dist	Decay
Orchard	0.6	1000	Exponential
Roads	0.7	1700	Exponential
Dam	0.5	800	Linear
Farm	0.6	1000	Linear

and

Table 6. 2024 threat type.

Threat	Weight	Max Dist	Decay
Orchard	0.5	1500	Exponential
Roads	0.6	1000	Exponential
Dam	0.4	500	Linear
Farm	0.5	500	Linear

): This table specifies the main anthropogenic threats affecting habitat quality and includes the relative weight (0–1), maximum effective distance (in meters), and the distance decay function (linear or exponential) for each threat type. A higher weight indicates a stronger negative impact on surrounding habitats. The maximum distance parameter defines the spatial extent of the threat’s influence, while the decay type determines how rapidly its effect diminishes with distance.

All parameters were derived through a combination of field investigations, regional ecological knowledge, and expert consultations, and further validated using published literature. This parameterization ensures that the model inputs realistically reflect local land use practices and ecological conditions in the watershed.

2.5. Quantification of Ecosystem Service Trade-Offs

After extracting the four ecosystem services at the pixel level, the Pearson correlation method was applied to quantify the linear relationships among them. The Pearson correlation coefficient is calculated using the following formula [48]:

$$R={\displaystyle \frac{{\displaystyle {\sum }_{i=1}^n}(x_i-\overline{x})(y_i-\overline{y})}{\sqrt{{\displaystyle {\sum }_{i=1}^n}{(x_i-\overline{x})}^2\sqrt{{\displaystyle {\sum }_{i=1}^n}{(y_i-\overline{y})}^2}}}}$$

(9)

where x_i and y_i are the values of two ecosystem services at the same pixel, ($\overline{x}$) and ($\overline{y}$) are the mean values of the respective services, and n is the total number of pixels.

This method allows for identifying the synergies (positive correlations) and trade-offs (negative correlations) among the services across space.

3. Results

3.1. Soil Retention Function Evaluation

As shown in Figure 3, the soil retention in the Caijiachuan watershed has exhibited an increasing trend from 2002 to 2024. In 2002 (Figure 3a), the unit soil retention showed minimal variation, with the lowest value of 1.73 t/hm² and the highest value of 1839 t/hm². In 2012 (Figure 3b), the central and northeastern areas showed higher soil retention, while the northwestern region experienced minimal change. The minimum value was 0.67 t/hm², and the maximum value remained at 1839 t/hm². In 2024 (Figure 3c), there was a significant increase, with uniform growth observed across the entire region. The minimum value reached 9.15 t/hm², and the maximum value increased to 2532 t/hm² (the observed value reflects a local peak, mainly occurring in steep slope forest patches with dense vegetation coverage, aligning well with the ecological conditions of reforested zones under the Grain for Green Program in the Loess Plateau).

3.2. Water Yield Function Evaluation

As shown in Figure 4, the unit area water yield in the Caijiachuan watershed exhibited a gradual increase from 2002 to 2024. The eastern region showed minimal change, while the western region experienced more significant variations. In 2002 (Figure 4a), there was a large spatial difference in water yield, with the eastern region having a higher water yield (334.8 mm/hm²) and the western region a lower water yield (97.7 mm/hm²). By 2012 (Figure 4b), the spatial difference in water yield between the eastern and western regions had notably reduced, although the water yield in the eastern region remained higher (397.2 mm/hm²) compared with the western region (136.9 mm/hm²). In 2024 (Figure 4c), the water yield trend showed that the eastern and western regions had relatively higher water yields (291.2 mm/hm²), while the central region had a lower water yield (165.3 mm/hm²), reflecting a continued decrease in the overall regional disparity in water yield.

3.3. Carbon Storage Function Evaluation

The unit area carbon storage in the Caijiachuan watershed showed a steady increase from 2002 to 2024 (Figure 5). In 2002 (Figure 5a), the highest value was 9.632 t/hm², in 2012 (Figure 5b) it increased to 12.996 t/hm², and by 2024 (Figure 5c), the highest value reached 16.276 t/hm². Over the period from 2002 to 2024, the regions with higher unit area carbon storage were the eastern and western parts of the watershed, while the central region exhibited lower carbon storage values.

3.4. Habitat Quality Evaluation

Based on its spatial distribution characteristics, habitat quality in the Caijiachuan watershed exhibited a consistent upward trend over time (Figure 6). In 2002 (Figure 6a), the highest values were concentrated in the western and central regions, with a maximum index of 0.66. By 2012 (Figure 6b), although the overall spatial pattern remained largely unchanged, the index values increased, reaching a maximum of 0.76. In 2024 (Figure 6c), habitat quality continued to improve following the same spatial trend, with the highest index rising to 0.85.

3.5. Temporal Correlation Patterns of Ecosystem Services

Pearson’s correlation heatmaps were constructed based on the normalized pixel-level values of four key ecosystem service indicators: carbon storage, water yield, soil retention, and habitat quality (Figure 7). The color gradient ranges from light to dark blue, with darker shades indicating stronger positive correlations. Each cell displays the Pearson correlation coefficient, rounded to two decimal places.

Figure 7a (2002), Figure 7b (2012), and Figure 7c (2024) reveal several consistent interaction patterns over the study period. Carbon storage and water yield exhibited a moderate and stable negative correlation in all three years (r ≈ −0.50 to −0.56), suggesting a persistent trade-off between these two services. Although carbon storage and soil retention showed a statistically significant positive correlation (r = 0.08–0.09, p < 0.05), the effect size was extremely small (R² < 0.01), indicating a negligible practical relationship. This weak association, likely driven by the large sample size (n > 5000), offers limited value for informing ecosystem management strategies.

The correlation between carbon storage and habitat quality weakened significantly over time, from r = −0.25 in 2002 to r = −0.03 in 2024. This shift from a weak negative to near-zero correlation likely reflects reduced conflict between these services as forest stands matured and ecosystem structure stabilized. Improved vegetation layering and species coexistence may have promoted functional complementarity between carbon accumulation and habitat provisioning. Given that the habitat quality model and input parameters remained consistent across years, this trend is unlikely to result from methodological artifacts, but rather indicates a positive ecological response to long-term restoration.

These findings highlight a relatively stable spatial correlation structure among ecosystem services over the 22-year period, while also indicating subtle shifts in interaction strength that may be driven by land use change, ecological restoration, or climatic variability.

3.6. Pixel-Level Scatter Analysis of Ecosystem Service Relationships

To evaluate the pixel-level trade-offs and synergies among ecosystem services (ESs), we constructed scatterplot matrix diagrams for the years 2002, 2012, and 2024, based on high-resolution (1 × 1 m) raster datasets. The analysis focused on four key ES indicators: carbon storage, water yield, soil retention, and habitat quality. Given the non-normal and non-linear nature of the data, a Spearman’s rank correlation was used to quantify monotonic relationships. All p-values were adjusted using the False Discovery Rate (FDR) method to reduce the risk of false positives, and 95% confidence intervals (CIs) were estimated via bootstrapping (n = 1000). The resulting matrices (Figure 8, Figure 9 and Figure 10) display univariate distributions along the diagonal, bivariate scatterplots in the lower triangle, and FDR-adjusted correlation coefficients with significance levels and CIs in the upper triangle.

In 2002 (Figure 8), distinct trade-offs among ecosystem services began to emerge in the early phase of ecological restoration. Carbon storage showed a strong negative correlation with water yield (ρ = −0.43, 95% CI: −0.46 to −0.42, p < 0.001), suggesting a consistent and robust trade-off, likely driven by the increased water consumption of newly established vegetation. The relatively narrow confidence interval further reinforces the reliability of this negative relationship for ecological inference and management. Additionally, carbon storage was moderately negatively correlated with habitat quality (ρ = −0.34, 95% CI: −0.37 to −0.31), indicating potential early-stage conflicts between carbon accumulation and biodiversity. In contrast, the relationships involving soil retention were weak and slightly positive (e.g., carbon–soil: ρ = 0.08, 95% CI: 0.05 to 0.11), with wide intervals that suggest substantial uncertainty and limited practical significance.

These findings highlight that while some trade-offs—particularly between carbon and water—were already pronounced and statistically robust in 2002, other interactions remained weak and ecologically ambiguous. The inclusion of confidence intervals provides a more nuanced understanding of correlation strength and reliability, which is critical for assessing the consistency and management relevance of ecosystem service relationships.

By 2012 (Figure 9), the trade-offs among ecosystem services had intensified. Carbon storage maintained strong negative correlations with both water yield (ρ = −0.42, 95% CI: −0.45 to −0.40) and habitat quality (ρ = −0.44, 95% CI: −0.47 to −0.42), which were both highly significant (p < 0.001). These stable and narrow confidence intervals reinforce the persistence and reliability of the underlying trade-offs, likely driven by increased vegetation biomass and associated resource demands. The negative correlation between water yield and habitat quality also became more pronounced (ρ = −0.34, 95% CI: −0.37 to −0.31), reflecting heightened functional competition between water availability and habitat conditions as canopy cover expanded.

In contrast, soil retention remained weakly and positively correlated with the other services (e.g., carbon–soil: ρ = 0.07, 95% CI: 0.04 to 0.09), suggesting limited synergistic effects. The relatively wide confidence intervals and low effect sizes imply that soil retention benefits—while present—played a minor role in shaping ES interactions during this phase of restoration.

These patterns indicate that by 2012, vegetation recovery had progressed to a stage where competition for water and habitat space became more ecologically significant, and the dominant trade-offs between regulating and supporting services were firmly established. The inclusion of confidence intervals clarifies which relationships are consistent enough to inform ecosystem management and which remain uncertain or context-dependent.

By 2024 (Figure 10), the structure of the interactions among ecosystem services showed signs of transition. The negative correlation between carbon storage and habitat quality weakened significantly (ρ = −0.31, 95% CI: −0.34 to −0.28), while the carbon–water relationship remained stable and strong (ρ = −0.42, 95% CI: −0.45 to −0.40), suggesting persistent water-related constraints despite the maturation of vegetation. This stability highlights the long-term nature of the trade-off between carbon sequestration and water availability in restored landscapes.

Notably, weak but consistent positive correlations emerged between soil retention and both carbon storage (ρ = 0.04, 95% CI: 0.02 to 0.07) and habitat quality (ρ = 0.04, 95% CI: 0.01 to 0.06). While the effect sizes remain small, the directionality and statistical significance of these correlations indicate the onset of spatial synergies among regulating services. These emerging synergies likely reflect an enhanced vegetation structure, deeper root systems, and the cumulative effects of long-term ecological restoration and adaptive land management.

Overall, the 2024 patterns suggest a shift from initial trade-offs toward partial coordination among services, marking a potential transition point in ecosystem functioning. Incorporating confidence intervals allows for a more nuanced understanding of this evolution and enhances the reliability of ecosystem management inferences based on these trends.

The robustness of these findings is supported by rigorous statistical procedures and a large sample size (n > 5000 pixels per year), which provide a high spatial resolution and strong analytical power. The application of FDR correction and bootstrap-derived confidence intervals reduces the risk of false positives and improves the interpretability of correlation estimates. Together, these methodological safeguards lend confidence to the observed patterns, which reveal a temporal shift from pronounced early-stage trade-offs to gradually emerging synergies. This trajectory highlights the dynamic and evolving nature of ecosystem service interactions under long-term ecological restoration in the Loess Plateau.

3.7. Temporal Trends of Carbon Storage and Water Yield

To further explore the dynamic relationship between carbon storage and water yield over time, we extracted the mean values of these two ecosystem services for the years 2002, 2012, and 2024, and visualized their temporal evolution (Figure 11). The results show that both carbon storage and water yield exhibit an overall increasing trend during the study period. However, the growth rate of carbon storage is significantly higher than that of water yield, suggesting differentiated ecosystem responses under ecological restoration.

Specifically, carbon storage shows a marked increase from 2002 to 2024, consistent with the large-scale implementation of the Grain for Green Program and the expansion of forest cover in the Caijiachuan watershed. This rapid increase reflects the strong response of above- and below-ground biomass accumulation to vegetation restoration.

In contrast, water yield increases at a much slower pace, and the overall variation is relatively modest. This can be attributed to two competing processes: on the one hand, improved vegetation coverage enhances soil infiltration and reduces surface runoff, potentially decreasing immediate water yield; on the other hand, long-term improvements in soil structure and water-holding capacity may help stabilize hydrological regulation.

These trend patterns align well with the correlation analysis results (Section 3.6), where a significant negative correlation (ρ = −0.43, p < 0.001) between carbon storage and water yield was observed in 2002. This implies that the increase in carbon storage may be accompanied by a trade-off in hydrological services, a pattern particularly evident in forested catchments undergoing ecological restoration.

4. Discussion

Based on the above findings, we further explored the underlying mechanisms driving these patterns.

4.1. Spatial Characteristics and Driving Mechanisms of Carbon Storage Increase

Between 2002 and 2024, carbon storage in the Caijiachuan watershed increased significantly. This upward trend was not primarily driven by changes in land use type but rather by the continuous restoration and maturation of both planted and natural secondary forests. Forest coverage in the watershed has reached approximately 80%, dominated by high carbon sequestration species such as Robinia pseudoacacia, Pinus tabuliformis, and Populus davidiana, whose growth has substantially enhanced the region’s carbon storage capacity [49].

Additionally, the region’s deep loess-derived soils provide strong potential for carbon accumulation. Previous studies have demonstrated that thick loess profiles are highly effective in stabilizing organic carbon, thus enhancing soil carbon sequestration [50]. The long-term implementation of ecological restoration projects, such as the Grain for Green Program launched in the early 2000s, has further facilitated the sustained recovery of vegetation and the stability of carbon sinks [51]. Although major land use categories have remained relatively unchanged over the past two decades, microscale ecological processes—such as forest stand development, increasing stand age, and understory vegetation enrichment—have become key drivers of carbon storage enhancement [52].

4.2. Spatiotemporal Evolution and Driving Factors of Water Yield

The water yield in the Caijiachuan watershed exhibited a continuous upward trend from 2002 to 2024. Spatially, the distribution shifted from a pronounced east–west disparity in 2002 to a pattern characterized by higher yields in the eastern and western regions and lower yields in the central area. This spatial reorganization is closely linked to the watershed’s geomorphology: a narrow, elongated catchment oriented east–west, with a declining elevation gradient from west to east. The upstream region, characterized by steep slopes and rapid runoff, contrasts with the central section, which has a gentler terrain and a lower water yield efficiency [53].

Approximately 50% of the watershed area lies within the 15–35° slope range, a range conducive to both runoff generation and yield stability. These areas become particularly active during the primary rainfall season (May–October), contributing to enhanced water yield [54]. Despite the region’s high forest coverage, the dominance of coniferous plantations (e.g., Robinia pseudoacacia and Pinus tabuliformis) tends to intercept precipitation and reduce surface runoff [55]. However, as plantations mature and understory vegetation improves, the interception effects weaken [56]. Although the annual mean precipitation (~540 mm) remains relatively stable, the interannual coefficient of variation (0.23) indicates considerable variability in rainfall distribution [19], contributing to peak water yield events in the central and western regions during periods of intense rainfall [57].These localized peaks highlight the spatial heterogeneity of water provisioning services and further emphasize the role of climatic and topographic drivers in shaping trade-offs among ecosystem services in the Grain for Green region.

4.3. Topographic and Climatic Context of the Trade-Off Between Carbon Storage and Water Yield

Carbon storage and water yield consistently exhibited a significant trade-off (R² = 0.24–0.31) across the study period. This trade-off is primarily attributed to the combined effects of natural geography and ecological processes rather than land use conflicts. On one hand, the accumulation of arboreal biomass increases transpiration and canopy interception, thus reducing surface runoff and suppressing water yield [58]. On the other hand, the loess hill–gully terrain promotes rapid water convergence on steep slopes, limiting runoff generation in forested areas with dense root systems, particularly on slopes exceeding 25° [59].

Moreover, although frontal precipitation events contribute substantially to regional water yield, the long intervals between events allow for significant water loss through transpiration and soil evaporation, particularly in forested areas [60]. Given the region’s warm temperate continental climate with high evapotranspiration potential, this trade-off remains stable at the interannual scale [61].

4.4. Synergistic Effects of Topographic Heterogeneity and Ecological Patterns

The synergies and trade-offs among ecosystem services are largely shaped by the interactions between topographic heterogeneity and ecological patterns. The pronounced west–east elevation gradient in the Caijiachuan watershed creates distinct spatial patterns of slope and altitude. Medium- to high-slope areas serve as hotspots for both soil retention and carbon storage, as well as zones of continuous habitat quality improvement [62]. The synergy between soil retention and carbon storage (R² > 0.2) is primarily driven by dense root systems enhancing soil stability and the accumulation of soil organic matter through litterfall [63].

In contrast, services related to water yield are strongly influenced by runoff pathways and slope gradients. Although footslope areas accumulate water, their high forest density also accentuates the trade-off between carbon storage and water yield [64]. Additionally, areas with high carbon storage typically have dense forest stands but lower species diversity, leading to a trade-off between habitat quality and carbon storage [65]. Therefore, optimizing forest stand structures and enhancing plantation biodiversity are critical strategies for improving service synergies.

4.5. Ecosystem Service Trade-Offs, Agricultural Adaptation, and Rural Livelihoods

Although farmland now accounts for a relatively small proportion of the Caijiachuan watershed, the long-term conversion of cropland to forest under the Grain for Green Program has profoundly reshaped local land systems and resource availability. The observed trade-offs between carbon storage and water yield suggest that ecological restoration, while beneficial for carbon sequestration, may reduce water availability for agricultural and domestic use, especially in downstream areas. This could exacerbate water competition in the context of regional drought and seasonal water shortages.

These findings highlight the need to consider the water–carbon trade-off in the broader context of rural adaptation and sustainable resource management. Empirical research shows that vegetation-induced water limitation may significantly affect crop productivity and rural household resilience in reforested areas [66]. As a result, balancing ecological benefits and water use rights has become an urgent challenge in policy design for semi-arid regions [14]. Recent studies emphasize the importance of incorporating farmers’ adaptive strategies, such as mixed land use, agroforestry, and water-saving irrigation, into ecosystem governance frameworks [67].

Moreover, the long-term success of restoration projects depends not only on ecological performance but also on the socioeconomic sustainability of local communities. In the Loess Plateau, where livelihoods have historically depended on rainfed agriculture, the trade-off between water yield and carbon storage directly influences food security, labor allocation, and land use decisions [68]. Therefore, future research should investigate how ecosystem service trade-offs reshape rural livelihood strategies and identify win–win pathways that align carbon sequestration goals with sustainable agricultural development.

4.6. Interpreting Effect Sizes: Statistical Significance Versus Practical Relevance

While this study employed rigorous statistical procedures—including FDR-adjusted p-values and bootstrap confidence intervals—the interpretation of correlation strength must go beyond statistical significance, especially in large spatial datasets. Given our sample size (n > 5000 pixels per year), even very weak correlations (e.g., ρ = 0.04) may appear statistically significant, yet they may not reflect ecologically meaningful interactions.

We acknowledge that the present study did not systematically compare observed effect sizes with conventional thresholds or restoration-focused benchmarks. According to Cohen’s widely used guidelines (small: ρ = 0.10, medium: ρ = 0.30, large: ρ = 0.50), most of the weak positive correlations in our results fall below the threshold for even a small effect, suggesting limited practical relevance. In contrast, stronger and consistent negative correlations—such as between carbon storage and water yield (ρ ≈ −0.42)—may approach medium effect levels and deserve closer attention in future restoration planning.

Moving forward, we recommend that future research integrates effect size interpretation and cross-study comparisons more explicitly, particularly in the context of ecological restoration where trade-offs and synergies among ecosystem services must be weighed not just statistically, but in terms of their real-world implications. This will help distinguish patterns that are meaningful for management from those that may simply result from high-resolution data and large sample sizes.

5. Conclusions

This study examined the Caijiachuan watershed, a representative Grain for Green region on the Loess Plateau, to assess the spatiotemporal evolution and interaction mechanisms of multiple ecosystem services from 2002 to 2024. Using the InVEST model calibrated with extensive field observation data, we quantified the changes in soil retention, water yield, carbon storage, and habitat quality, with a particular focus on the long-term trade-off dynamics between carbon storage and water yield. The main conclusions are as follows:

5.1. Substantial Ecosystem Service Improvements with Pronounced Spatial Heterogeneity

Under the sustained influence of the Grain for Green Program, ecosystem services in the watershed have improved markedly. By 2024, carbon storage and soil retention showed significant increases compared with 2002, primarily driven by vegetation restoration and forest maturation. While water yield also improved overall, its spatial distribution exhibited strong heterogeneity, shaped by terrain configuration, land use, and forest structure.

5.2. A Persistent Trade-Off Between Carbon Storage and Water Yield

Regression analyses and scatterplot matrices consistently revealed a significant negative correlation between carbon storage and water yield (mean R² ≈ 0.28), indicating a stable trade-off across the study period. This finding highlights the inherent tension between vegetation-based carbon sequestration and hydrological provision, underscoring the need for balanced ecological restoration strategies. However, an R² value of 0.28 also implies that approximately 72% of the variation in water yield remains unexplained, suggesting that other factors—such as soil texture, slope gradient, and interannual climatic variability—likely play substantial roles. As such, while the carbon–water trade-off is relevant, it may not be sufficient alone to inform decisions such as planting density. To improve management precision, future strategies should integrate field-based hydrological observations and process-based models to better capture the complex drivers of water regulation in restored landscapes.

5.3. Trade-Off Mechanisms Driven by Topography, Climate, and Vegetation Dynamics

The trade-off between carbon and water services is jointly controlled by topographic gradients, climate variability, and vegetation structure. Steeper slopes and higher elevations are associated with higher carbon storage but reduced water yield due to interception and evapotranspiration. In contrast, low-lying areas with gentle slopes tend to support greater water yield. Interannual precipitation fluctuations further modulate these dynamics through their influence on soil moisture and vegetation growth.

5.4. Weak Synergies Among Other Ecosystem Services

Apart from the evident carbon–water trade-off, the synergies among other ecosystem services—such as habitat quality and soil retention—remain relatively weak and spatially inconsistent. This suggests that optimizing one service does not necessarily enhance others, and highlights the importance of place-based, multi-service management rather than one-size-fits-all ecological planning.

5.5. Ecosystem Service Trade-Offs Have Implications for Agricultural Adaptation and Rural Livelihoods

Although cropland currently occupies a minor proportion of the watershed, long-term reforestation has reshaped land use systems and resource availability. The carbon–water trade-off poses potential challenges for downstream water users, especially in seasons of water scarcity. This trade-off could constrain agricultural productivity and limit livelihood resilience in rural communities. Therefore, future ecological restoration efforts must incorporate adaptive agricultural strategies, such as agroforestry, mixed land use, and water-efficient practices, to reconcile ecological goals with socioeconomic sustainability.

In summary, the Caijiachuan watershed exemplifies the complex interplay between ecological restoration and ecosystem service trade-offs in semi-arid regions. Integrated land management strategies that account for spatial heterogeneity, ecological thresholds, and rural development needs are essential for achieving long-term restoration success and enhancing human–environment system resilience.