Spatiotemporal Simulation Prediction and Driving Force Analysis of Carbon Storage in the Sanjiangyuan Region Based on SSP-RCP Scenarios

Abstract

Global warming and rapid urban industrialization are profoundly transforming land-use patterns and carbon storage capacity in terrestrial ecosystems. A rigorous analysis of spatiotemporal variations in regional land-use changes and carbon storage dynamics provides critical insights for sustainable land-use planning and ecological security, particularly within the context of achieving carbon peaking and carbon neutrality targets. In this study, the PLUS-InVEST model was coupled with climate change and policy constraints to construct six future scenarios. We analyzed the characteristics of land-use evolution and the spatial and temporal changes in carbon storage in the Sanjiangyuan region from 2000 to 2020. We also predicted the potential impacts of land-use shift on carbon storage. The results show the following: (1) Land-use transitions exerted significant impacts on carbon stock. The Sanjiangyuan region experienced a net carbon stock reduction of 9.9 × 10⁶ t during 2000–2020, with the most pronounced decline (6.1 × 10⁶ t) occurring between 2000 and 2010. (2) Under the same climate scenario, the natural development (ND) scenario exhibited decreasing carbon reserves relative to 2020 baseline levels. Notably, land-use planning scenarios demonstrated spatially heterogeneous impacts, with the ecological protection (EP) scenario consistently maintaining higher carbon stocks compared to the ND scenario. (3) Multivariate driver interactions exerted stronger control over spatial carbon storage patterns than any individual factor. These findings inform targeted land-use management strategies to enhance regional carbon sequestration capacity, promote sustainable development, and support China’s carbon peaking and neutrality objectives.

1. Introduction

Anthropogenic activities have driven an unprecedented global warming trend over the past century, with climate projections indicating persistent temperature increases that will likely exacerbate the frequency and intensity of extreme climate events, thereby threatening human livelihoods and sustainable development [1]. The ongoing industrialization process has driven a fundamental transition in economic development priorities, shifting focus from agricultural production to specialized industrial and service sectors. Concurrently, rapid urbanization has induced substantial transformations in land-use/cover change (LUCC) patterns [2]. LUCC modifications exert profound influences on regional carbon stocks through ecosystem structural and functional alterations, thereby regulating the release of terrestrial carbon storage [3]. Consequently, land-use transitions constitute a critical driver of carbon stock variability in terrestrial ecosystems [4]. These ecosystems represent Earth’s largest carbon reservoirs and play a pivotal role in global climate regulation and carbon cycling processes [5]. The IPCC Sixth Assessment Report highlights that human activities have notably influenced terrestrial circulation systems by altering land-use patterns. Over the past 150 years, land-use changes have accounted for approximately 33% of global carbon emissions, second only to fossil fuel combustion [6]. As a significant contributor to global carbon emissions, China has committed to ambitious climate targets, including achieving carbon peaking by 2030 and carbon neutrality by 2060, as formally incorporated into its 14th Five-Year Plan [7]. This demonstrates China’s commitment to significantly contributing to global efforts to reduce carbon in the future. Therefore, comprehensive assessment of carbon stock dynamics induced by land-use transitions across multiple climate scenarios—both historical and projected—carries significant practical implications. This analysis will help optimize land-use planning in the region, reinforce carbon storage capability, and implement the “dual-carbon” strategic goal [8,9].

Land-use change has been demonstrated as the primary driver of alterations in terrestrial carbon stocks. Changes in land-use patterns fundamentally transform ecosystem composition and functionality, thereby regulating the carbon reserve capacity of both vegetation and soil compartments. Thus, land-use shifts significantly affect the level of district carbon stocks. Investigating land-use changes allows us to research the effects of land use on shifts in carbon stocks [10]. Currently, multiple models are employed to quantify carbon stock changes, including InVEST, ARIES, and ESM. Among these, the InVEST model is widely recognized for its operational simplicity, computational efficiency, and simulation accuracy. Notably, it facilitates the spatiotemporal assessment of carbon stocks across large-scale landscapes. This model has been extensively applied by researchers across diverse geographical and ecological contexts [11,12], demonstrating robust applicability in China [13,14,15]. For instance, Yue et al. utilized the InVEST model to examine temporal and regional variations in carbon reserves within Anhui Province [16]. Similarly, Babbar et al. employed InVEST to quantify and project carbon sequestration potential in their study area [9]. Furthermore, Li et al. utilized InVEST to evaluate ecosystem services in the Shennandong River Basin (Guizhou Province), revealing a marginal decline in carbon storage capacity between 2000 and 2020 [17]. Numerous studies have employed various modeling frameworks to quantify the relationship between land-use changes and carbon stock dynamics. Among these, the Advanced Patch Generation Land Use Simulation (PLUS) model has emerged as a robust analytical tool, retaining the adaptive inertial competition mechanism and spatial allocation accuracy of conventional land-use simulation models. The PLUS model utilizes a Random Forest (RF) algorithm to precisely simulate spatiotemporal land-use patterns. Within this framework, the Land Expansion Analysis Strategy (LEAS) module enables systematic identification of the underlying drivers of land-use change, while the Cellular Automata for Rule-Set Simulation (CARS) module enhances the spatial precision of land-use projections. Notably, the PLUS model facilitates data mining and dynamic simulation of land-use transitions, demonstrating direct applicability to real-world scenarios such as optimal land resource allocation and urban growth boundary delineation. A key advancement of this model lies in its ability to address multicollinearity among driving factors, significantly improving both spatial and temporal simulation accuracy [18]. These attributes have led to its widespread adoption in land-use change research. By creating scenarios involving the preservation of arable land, economic development, and ecological preservation, many studies [19,20,21] have carried out simulation studies on forecasting land-use patterns and carbon reserves distribution in various locations. The integrated application of the PLUS and InVEST models capitalizes on their complementary strengths, enabling comprehensive ecosystem service assessment coupled with dynamic land-use simulation. This combination offers a scientific foundation for a deeper comprehension of the connection among land-use change and carbon reserves and notably enhances spatiotemporal changing modeling and ecosystem service evaluation capabilities in carbon stock research [22,23].

Existing research demonstrates that LUCC shifts affecting ecosystem carbon stocks are jointly regulated by climatic factors and socioeconomic drivers [24,25]. The latest Coupled Model Intercomparison Project Phase 6 (CMIP6) provides researchers with future development scenarios under global climate change through the integration of Shared Socioeconomic Pathways (SSPs) with Representative Concentration Pathways (RCPs) [26,27]. An increasing number of studies are using SSP-RCP scenarios to project future land-use demand. For instance, recent research has developed a scenario-based analytical framework to simulate China’s land-use requirements and spatial allocation patterns [28]. Some scholars studied system dynamics (SD) models with the PLUS model, incorporating InVEST model assessments to establish a comprehensive analytical platform for simulating urban-scale land-use transitions and associated carbon stock dynamics [25]. Other studies have projected the global soil erosion rate and evaluated global soil regulation services from 2015 to 2070 under three SSP-RCP scenarios [26]. Additionally, some scholars have simulated spatiotemporal patterns of land use in Central Asia by integrating SSP-RCP scenarios with land-use transition models. They also conducted a thorough evaluation of the levels of ecosystem services in the study area [27]. Historical carbon density data were predominantly utilized in the majority of studies on projecting carbon stocks under various future scenarios without accounting for various future scenarios [8]. Terrestrial ecosystem carbon stocks are critically dependent on land-use transitions between historical and contemporary periods. A quantitative assessment of carbon stock variations induced by land-use dynamics is essential for guiding regional sustainable development, enhancing carbon sequestration potential, and facilitating climate change adaptation. Specifically, rigorous quantification of land-use transition impacts on carbon storage across historical and future time horizons represents a critical research priority.

Located in the eastern Qinghai-Tibetan Plateau, the Sanjiangyuan region is the source of the Yellow River, Yangtze River, and Lancang River. Ecologically, this place is crucial for ensuring water security and the ecological balance of China and Asia. As China’s largest water conservation and supply ecosystem service region, Sanjiangyuan also represents a key biodiversity conservation priority area [29]. The implementation of major strategies, such as the “Major Project for Ecological Protection and Restoration of the Sanjiangyuan Region,” has elevated this ecologically sensitive area to a priority conservation target [30]. Some scholars have also assessed regional carbon stocks in this critical ecosystem [31,32,33]. However, China’s rapid socioeconomic development over the past three decades, coupled with intensified human activity and fast city expansion, has significantly altered land-use patterns in the Sanjiangyuan region and significantly altered spatiotemporal changes in carbon stocks. Nevertheless, no study has encompassed the dynamic evolution of carbon stocks across the entire region, and research on contact between shifts in land-use types and carbon stocks in the Sanjiangyuan region remains ambiguous. A macro-scale understanding of these relationships can help protect the region’s biodiversity and support regional sustainable development and ecological security, which is central for maintaining the region’s ecology [34].

This study develops an integrated modeling framework combining the PLUS and InVEST models to simulate spatiotemporal patterns of LUCC in the Sanjiangyuan region. The simulations consider the needs of land-use demand, population, economics, and climate change scenarios under three Representative Concentration Pathways (RCPs) and Shared Socioeconomic Pathways (SSPs): SSP1-2.6, SSP2-4.5, and SSP5-8. Through quantitative assessment of carbon stock distributions and driver contributions across three SSP-RCP scenarios (SSP1-RCP2.6, SSP2-RCP4.5, and SSP5-RCP8.5), this study aims to advance understanding of regional land resource allocation, ecosystem management, and socioeconomic policy formulation in the Sanjiangyuan region.

2. Materials and Methods

2.1. Overview of the Study Area

The Sanjiangyuan region is located in the southern part of Qinghai Province (31°39′–37°10′ N, 89°24′–102°27′ E). As the hinterland of the Qinghai-Tibetan Plateau, it is the source of Yangtze River, Yellow River, and Lancang River. For this reason, it is widely called “Chinese Water Tower” (Figure 1). The Sanjiangyuan region is a vital ecologically sensitive and vulnerable location in China and around the world. The topography is complicated; northwest terrain is high and southeast is low. The region has an average elevation of approximately 4484 m [35]. Climate is characteristic of a plateau continental climate with long sunshine hours, intense solar radiation, and distinct wet and dry seasons. The range of average annual temperature is from −5.6 to 7.8 °C [36], and average annual precipitation is 470.7 mm [37]. The Sanjiangyuan region has many lakes and marshes; snowy mountain glaciers are also widely distributed. Lakes are mainly found at the sources of Yangtze and Yellow Rivers. The main vegetation types are alpine grassland and alpine meadow, covering 16.55% and 47.06% of total region, respectively [35].

2.2. Data Sources

Data sources and acquisition methods are shown in

Table 1. Data sources.

Number	Type	Name of Data	Resolution (m)	Data Sources
	Basic Data	Land use	30	National Geographic Information Resource Catalog Service System (https://www.webmap.cn/) (accessed on 15 March 2025)
X1	Environmental Data	Soil type	1000	Soil Science Database (http://www.resdc.cn/) (accessed on 16 March 2025)
X2		Average annual temperature	1000	National Earth System Science Data Center (http://www.geodata.cn/) (accessed on 16 March 2025)
X3		Average annual precipitation	1000
X4		DEM	30	Geospatial Data Cloud (http://www.gscloud.cn/search) (accessed on 16 March 2025)
X5		Slope	30	Obtained directly from DEM calculation in ArcGIS
X6	Socioeconomic Data	GDP	1000	Resource and Environment Science and Data Center (http://www.resdc.cn) (accessed on 16 March 2025)
X7		Population density	1000
X8		Nighttime light	1000
X9		Distance to railroads	30	Open Street Map (https://www.openstreetmap.org/) (accessed on 16 March 2025)
X10		Distance to national highways	30
X11		Distance to provincial roads	30
X12		Distance to county roads	30
X13		Distance to township roads	30
X14		Distance to county governments	30	National Geographic Information Resource Catalog Service System (https://www.webmap.cn/) (accessed on 16 March 2025)
X15		Distance to waters	30

. Three remote sensing images of land use in the Sanjiangyuan region of Qinghai Province were chosen from National Geographic Information Resource Catalog Service System for the years 2000, 2010, and 2020. Using ArcGIS 10.8 software, these images were classed into 6 key land-use categories: cultivated land, woodland, grassland, water, construction land, and unused land, with references to previous research and the Chinese Academy of Sciences’ primary classifications. Other data include environmental data (soil type, average annual temperature, average annual precipitation, DEM, slope) and socioeconomic data (GDP, population density, nighttime light, and distances to railroads, national highways, provincial roads, county roads, township roads, county governments, and waters). We use ArcGIS software to process DEM so we can gain slope data. Distances to roads and waters were obtained from vector data using Euclidean distance analysis in ArcGIS [38]. All data were rasterized with the same row and column numbers, as well as WGS_1984_UTM (EPSG:32646) projected coordinate system, as land-use data, resampling the data resolution to 30 m through Nearest Neighbor Allocation Method (NEAREST).

2.3. Research Protocol

The comprehensive research framework proposed in this research comprises three steps (Figure 2): (1) According to features of land-use changes in the Sanjiangyuan region from 2000 to 2020, we created land-use weight matrices under 6 different scenarios and forecasted region’s future land-use needs in many scenarios. (2) We used PLUS to compute geographical distribution of land in the Sanjiangyuan region in 2030, combining expected land-use demand and driving factor data from six different future development scenarios. We then undertook detailed research of spatiotemporal history of land in study region from 2000 to 2030. (3) InVEST was selected to combine carbon density and site data from comparable research sites in the historical literature to predict historical and future carbon stock patterns in the Sanjiangyuan under several SSP-RCP scenarios spanning 2000 to 2030.

2.4. Forecast of Future LUCC Based on PLUS

2.4.1. PLUS Model

The PLUS model demonstrates superior performance as a patch-level land-use simulation tool, offering key advantages over conventional modeling approaches. It can handle numerous types of land-use patch changes and refine simulation scale [39]. This model is made up of two major modules: land expansion analysis strategy (LEAS) model and change analysis, resilience, and sustainability (CARS) model [40]. This study’s calculations are on account of percentage in area transferred by every type of expansion in the 2010 and 2020 land-use data (

Table 2. Domain weight parameters of distinct land-use types.

Land-Use Type	Cultivated Land	Woodland	Grassland	Water	Construction Land	Unused Land
Domain weight	0.41	0.42	0.14	1.00	0.52	0.10

), and the formula is as follows:

$$W_{\mathrm{i}}={\displaystyle \frac{\left(X_i-X_{\min }\right)}{\left(X_{\max }-X_{\min }\right)}}$$

(1)

where W_i denotes value of domain weight of land category i; X_i denotes extension range of land class i; X_max denotes the maximum value of expansion area of whole land categories; and X_min denotes the minimum value of expansion area of whole land categories.

In terms of scenario setting, this study incorporates the study area’s situation and pertinent research results to create two simulation scenarios: natural development (ND) and ecological protection (EP). In ND scenario, water is classified as a limited conversion area. From 2010 to 2020, land-use shift change management governs development and transfer patterns for each land-use category, regardless of other policy restrictions. In EP scenario, the rise in ecological land area is aligned with the ecological protection measures recommended in the “Qinghai Provincial Land and Spatial Planning (2021–2035)” and “Three Rivers Source National Park Overall Plan (2018)”. Research area’s biodiversity, water conservation, and soil conservation are also taken into account. The Sanjiangyuan National Park’s borders and water are designated as restricted conversion areas. There is a 50% lower chance of converting grassland and woodland to development land. There is a 30% added likelihood of converting used land to grassland and woodland. There is a 30% decrease in likelihood of turning arable land into building land (

Table 3. Multi-scenario land-use transition matrix for 2030.

Land-Use Type	ND						EP
	A	B	C	D	E	F	A	B	C	D	E	F
A	1	1	1	1	1	1	1	1	1	1	1	0
B	1	1	1	1	1	1	0	1	0	0	1	0
C	1	1	1	1	1	1	0	1	1	1	1	0
D	1	1	1	1	1	1	0	0	0	1	0	0
E	1	1	1	1	1	1	1	1	1	1	1	0
F	1	1	1	1	1	1	1	1	1	1	1	1

A, B, C, D, E and F mean cultivated land, woodland, grassland, water, construction land and unused land, respectively; 1 represents mutual transfer, and 0 represents no mutual transfer.

).

2.4.2. Model Accuracy Verification

Regarding these results, PLUS was applied to simulate land use in 2020 based on land-use data from 2000 to 2010. The Kappa coefficient’s accuracy was then tested using both simulated and true land-use data from 2020 [38]. The Kappa coefficient was calculated to be 0.94, with an accuracy of 0.97. Kappa coefficient greater than 0.85 suggests that the model is quite accurate at forecasting future land use in study space [41].

2.5. SSP-RCP Coupling Scenario Setting

To address and assess future climate shifts, we focus on CMIP6 and SSPs, which are emission scenarios influenced by several socioeconomic factors. SSPs are important in researching climate change [42]. Three CMIP6 scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) were utilized in this work to forecast future switches in carbon stocks attributed to land-use shifts in Sanjiangyuan. SSP1-2.6 (Sustainable Green Development Pathway) represents long-term socioeconomic development and lower greenhouse gas emissions; SSP2-4.5 (Intermediate Pathway) represents centered socioeconomic development and gas emission patterns; and SSP5-8.5 (Forcing Path) represents highest socioeconomic development and gas emission patterns [43]. Two land-use planning scenarios (natural development and ecological protection) and three climate scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) that are expected to occur in 2030 were investigated concurrently. The outcome of climate and land-use switches on carbon reserves in Sanjiangyuan was quantitatively investigated using a combination of six future development scenarios. These scenarios were given the names ND126, ND245, and ND585 (natural development scenarios) and EP126, EP245, and EP585 (ecological protection scenarios). The future climate scenario dataset was created using Delta spatial downscaling scheme (MATLAB R2023b) and a large-resolution climate dataset from WorldClim (http://www.worldclim.org/) based on IPCC’s most recent SSPs (SSP1-2.6, SSP2-4.5, and SSP5-8.5) [44,45].

2.6. InVEST Model

InVEST model is a tool for estimating ecosystem carbon stocks and sink potential, as well as quantifying ecosystem carbon intake. It has been widely adopted [46]. The model classified terrestrial ecosystem carbon stocks as four basic pools: soil organic matter, aboveground, belowground, and dead organic matter. The carbon stocks from each land-use type are combined together to achieve overall carbon stock of study region. Following formula shows how to calculate full carbon stocks:

$$\begin{array}{l}{C}_{\mathrm{i}}=C_{above}+C_{\mathrm{b}elow}+C_{dead}+C_{soil}\\ C_{total}={\displaystyle \sum _i^n{C}_i\times A_i,(i=1,2,\mathrm{\dots },n)}\end{array}$$

(2)

where i represents i-th land-use type; C_i represents total carbon stock per unit region of each land-use type (t/ha); C_above, C_below, C_soil, C_dead, respectively, represent the aboveground biomass carbon density, underground biomass carbon density, soil organic carbon density, and dead organic carbon density; C_total represents total carbon stock of ecosystem(t); and A_i represents area of each land-use type.

The carbon density data for each land-use type in this research were primarily according to a past study. In China Land Ecosystem Carbon Density Dataset (2010) (http://www.nesdc.org.cn) [47], we gained carbon density data of cultivated land, woodland, grassland. Samples with similar latitudes and longitudes to the study area (90–102° E and 31–36° N) were selected from the dataset. This comprised 444 aboveground biomass carbon densities, 361 belowground biomass carbon densities, and 246 soil organic carbon densities. The average values of various land types were analyzed separately to obtain carbon densities of the aforementioned land types’ carbon pools [34,48]. Because of the fact that most of construction land is surface-hardened and most unused land in the Sanjiangyuan region includes highland deserts and bare land, the percentage of dead organic carbon density of these two land types is poor, and data are hard to gain, so the carbon density data of the above two lands refer to relevant previous studies [49,50]. Similarly, carbon density of diverse carbon pools in water can be negligible and was set to 0. The carbon density values assigned to construction and unused land categories in this study were derived from other places in China, necessitating localized calibration to account for potential spatial heterogeneity in carbon storage characteristics. Existing research has shown that annual precipitation is positively linked to biomass and soil organic carbon density [51]. Thus, we use precipitation to modify carbon density [52], which was calculated as follows:

$$\begin{array}{l}{C}_{BP}=6.789e^{0.005P}\\ C_{SP}=3.3968\times P+3996.1\end{array}$$

(3)

where C_BP means biomass carbon density of vegetation corrected for annual precipitation; C_SP means soil carbon density corrected for annual precipitation; P means average annual precipitation (mm).

$$\begin{array}{l}{K}_{BP}={\displaystyle \frac{{C^{\prime }}_{BP}}{{C^{\prime \prime }}_{BP}}}={\displaystyle \frac{e^{0.0054\times P_A}}{e^{0.0054\times P_B}}}\\ K_{SP}={\displaystyle \frac{{C^{\prime }}_{SP}}{{C^{\prime \prime }}_{SP}}}={\displaystyle \frac{3.3968\times P_A+3996.1}{3.3968\times P_B+3996.1}}\end{array}$$

(4)

where K_BP and K_SP are precipitation correction factors for vegetation biomass carbon density and soil carbon density, respectively; C′_BP and C″_BP are vegetation biomass carbon density data corrected according to annual precipitation in the Sanjiangyuan region and nationwide, respectively; C′_SP and C′_SP are soil carbon density data corrected according to annual precipitation in the Sanjiangyuan region and nationwide, respectively; P_A and P_B represent annual average precipitation in the Sanjiangyuan region and nationwide, respectively.

Table 4. Carbon density values of various land-use types in the Sanjiangyuan region (t/ha).

Land-Use Type	C_Above	C_Below	C_Soil	C_Dead
Cultivated Land	0.17	0.82	35.20	0.1
Woodland	39.70	12.57	88.71	0.14
Grassland	0.61	6.98	71.70	0.07
Water	0	0	0	0
Construction Land	0.03	0.28	0	0
Unused land	0.01	0	0.22	0

shows computed carbon density results. Because the fluctuation range of carbon storage data in the Sanjiangyuan is too wide, log10 conversion is used in mapping of carbon reserves for spatial distribution.

2.7. Geodetector

The Geodetector method detects spatial differences in geographic phenomena and their driving forces using indicators such as “factor force value” [53]. A variety of variables contribute to fluctuations in carbon stocks in the Sanjiangyuan region. Factor and interaction detection were used to investigate underlying causes of these changes. The value of the dependent variable, Y, represents carbon stock attributes within the Sanjiangyuan region’s spatial network. Independent variable X includes 15 factors, such as soil type, DEM, and average annual precipitation (Table 1). The factor detector and interaction detector assess the explanatory power of individual and interacting factors for carbon stock changes.

3. Results

3.1. Spatiotemporal Changes in Land Use of the Sanjiangyuan Area in Last 20 Years

For the past two decades, the land-use types in Sanjiangyuan have changed dramatically. Structurally, the land use is dominated by three primary types, grassland, unutilized land, and water bodies, which collectively account for >94% of the total study area (

Table 5. Area and percentage of distinct land-use types in the Sanjiangyuan region from 2000 to 2020.

Land-Use Type	2000		2010		2020
	Area/ha	Percentage/%	Area/ha	Percentage/%	Area/ha	Percentage/%
Cultivated Land	251,061.66	0.64	258,653.34	0.66	258,016.95	0.67
Woodland	1,676,811.69	4.30	1,676,730.87	4.30	1,676,716.74	4.30
Grassland	27,886,071.87	71.48	27,805,857.21	71.27	27,757,545.30	71.15
Water	2,081,886.84	5.34	2,093,165.37	5.37	2,194,290.81	5.62
Construction Land	19,629.45	0.05	20,167.56	0.05	38,848.05	0.10
Unused Land	7,096,691.70	18.19	7,157,578.77	18.35	7,085,025.45	18.16
Total	39,012,153.21	100.00	39,012,153.12	100.00	39,010,443.30	100.00

). In this region, grassland occupied 71.15% of the total area in 2020, so it is the principal land use. Unused land makes up 18.16% of the full area and is predominantly found in the western part of the Sanjiangyuan region, specifically in Zhiduo County, Tanggula Township, Qumalai County and Zaduo County. Large lakes, including Qinghai Lake, Zhaling Lake, and Eling Lake, contribute 5.63% of the water. Woodland is concentrated in the eastern region of Sanjiangyuan, which includes Guinan, Tongde, Gonghe, Guide, and Xinghai counties. It makes up 4.30% of Sanjiangyuan’s total area. Construction and cultivated lands account for less than 1% each. From 2000 to 2020, the land-use types have changed notably, with significant changes in several land uses. Specifically, construction land, water, and cultivated land increased throughout this time. Construction land increased by 19,218.60 ha, water by 112,403.97 ha, and cultivated land by 6955.29 ha, resulting in growth rates of 97.9%, 5.4%, and 2.8%. In contrast, grassland acreage declined by 128,526.57 ha (0.46%), while unused land decreased by 11,666.25 ha (0.16%).

Land-use transfer matrix analysis was conducted for the Sanjiangyuan region in two time periods, 2000–2010 and 2010–2020, producing information on the transferred areas of various land categories within each period. Figure 3 shows that between 2000 and 2010, land circulation areas occupied 0.26% and 2.03% of the whole region, respectively. From 2010 to 2020, the land circulation area rose significantly, with a maximum change of around 96,020.9 ha. Land-use changes were mostly driven by transfers between grassland, woodland, and water, with 188,278 ha of grassland transitioning to forest and water and 155,229 ha of woodland and water converting to grassland. Notably, grassland showed an overall decline. In contrast, cultivated land, construction land, and unused land showed mutual circulation with less noticeable alterations.

3.2. Spatiotemporal Diversity of Carbon Storage in the Sanjiangyuan Region

The carbon stock module of InVEST was used to count the carbon stocks of Sanjiangyuan in 2000, 2010, and 2020, accordingly. The results show that total carbon stocks were 24.60 × 10⁸ t, 24.54 × 10⁸ t, and 24.50 × 10⁸ t in 2000, 2010, and 2020, respectively. There has been an overall gradual loss in total carbon stocks over the past 20 years, with an entire reduction of 9.9 × 10⁶ t. The largest decrease in carbon stock occurred during the period from 2000 to 2010, along with an entire reduction of 6.1 × 10⁶ t. Figure 4a depicts obvious spatial heterogeneity in the carbon reserves distribution pattern in the Sanjiangyuan region from 2000 to 2020. The maximum carbon stock value for every unit area is 12.70 t/ha, while the minimum is 0. Low carbon stocks are sparsely distributed in the west, north, and northeast, while high carbon stocks are concentrated in the east and south. Overall, the spatial assignment pattern is “east is high, west is low, south is high and north is low.”

As demonstrated in Figure 4b, comparing the spatial allocation of carbon reserves changes in different periods reveals that, from 2000 to 2020, carbon stocks in this region fluctuated in a staggered manner. The northwestern region exhibited a slight increase, while the central and northeastern regions experienced significant decreases. In general, carbon stocks remained relatively stable in most areas during the two-decade period. However, due to the influence of land-use changes, carbon stock trends may differ in various time periods. Consequently, the areas experiencing changes in carbon stocks during the 2000–2020 period were smaller than in the two subperiods.

Land-use perspective analyses reveal significant heterogeneity in carbon storage across Sanjiangyuan’s dominant ecosystem types, with measurable interannual fluctuations in the relative carbon stock distribution during the 2000–2020 observation period. However, the overall ranking remained unchanged (Figure 4d). Grassland had the highest carbon stocks among all land types, followed by woodland. Unused and constructed land had the lowest carbon stocks. Sanjiangyuan’s land use is dominated by grassland, which is the most important carbon stocks. As a land type with high carbon density, grassland accounted for over 96% of the total carbon stock each year, making it the category that contributed the most to the region’s carbon stock. Therefore, changes in grassland and carbon stock over the last two decades have influenced the overall carbon stock in Sanjiangyuan to some extent. Woodland, like grassland, has a powerful carbon fixation capacity and a large area share among many types in the Sanjiangyuan region; thus, its carbon stock is also considerable. Despite its vast size, unused land has a weak carbon reserves capability. From 2000 to 2020, each category in the study region had shifting degrees of change in carbon storage: Grassland and forest showed a decreasing trend, with rates of 0.28% and 0.01%, respectively. Conversely, the value of cultivated land and construction land increased constantly, rising by 3.02% and 2.74%, respectively. These results demonstrate the progressive degradation of grassland ecosystems in the Sanjiangyuan region, concomitant with the expansion of construction and cultivated land driven by intensifying industrialization pressures. Figure 4d depicts the carbon stock of every carbon pool. The soil pool is the largest and most impactful in terms of regional carbon stock fluctuations. Carbon reserves in four primary pools (aboveground, belowground, dead organic matter, and soil) dropped by 8.9 × 10⁶ t, 8.0 × 10⁴ t, 8.9 × 10⁵ t, and 8.3 × 10³ t between the two decades, respectively. The soil carbon pool has declined the most in the last 20 years. From 2000 to 2010, the four major pools had the most dramatic reduction at 5.0 × 10⁴ t, 5.0 × 10⁴ t, 5.0 × 10⁵ t, and 8.3 × 10³ t, respectively. In terms of the magnitude of increase and decrease, the soil carbon pool decreased the most over most the phases of the study period, followed by the belowground carbon pool. The aboveground carbon pool experienced the smallest reduction.

3.3. Land-Use Change in the Sanjiangyuan Region in 2030 Under Multi-Scenario Simulation

3.3.1. Land-Use Simulation of the Sanjiangyuan Region in Various Development Scenarios of 2030

Land-use patterns in the Sanjiangyuan region were predicted under six scenarios using the PLUS model (

Table 6. Area of various land-use types under various scenarios in 2030 (ha).

Year	Scenarios	Cultivated Land	Woodland	Grassland	Water	Construction Land	Unused Land
2020		258,016.95	1,676,716.74	27,757,545.30	2,194,290.81	38,848.05	7,085,025.45
2030	EP126	256,657.32	1,677,436.38	27,786,553.56	2,309,501.16	44,622.18	6,935,672.70
	ND126	256,657.41	1,677,403.98	27,779,708.34	2,309,501.16	45,524.07	6,941,648.34
	EP245	261,654.21	1,677,061.53	27,717,851.61	2,289,721.77	42,403.68	7,021,750.50
	ND245	261,228.87	1,676,600.82	27,707,585.04	2,289,691.62	46,773.45	7,028,563.50
	EP585	268,174.89	1,684,883.97	27,709,794.00	2,288,403.90	37,348.29	7,021,838.25
	ND585	267,463.35	1,676,044.44	27,697,820.22	2,288,369.07	42,958.17	7,037,788.05
2020–2030	EP126	−1359.63	719.64	29,008.26	115,210.35	5774.13	−149,352.75
	ND126	−1359.54	687.24	22,163.04	115,210.35	6676.02	−143,377.11
	EP245	3637.26	344.79	−39,693.69	95,430.96	3555.63	−63,274.95
	ND245	3211.92	−115.92	−49,960.26	95,400.81	7925.40	−56,461.95
	EP585	10,157.94	8167.23	−47,751.30	94,113.09	−1499.76	−63,187.20
	ND585	9446.40	−672.30	−59,725.08	94,078.26	4110.12	−47,237.40

): ND126, ND245, ND585, EP126, EP245, and EP585. Future land-use changes under these scenarios will be significantly different from the 2020 baseline. All scenarios indicated changes in cultivated land compared to 2020. The EP126 and ND126 scenarios resulted in losses of 1359.63 ha (0.53%) and 1359.54 ha (0.53%), respectively. Conversely, the other scenarios showed an increase in cultivated land: EP245, ND245, EP585, and ND585 increased by 3637.26 ha (1.41%), 3211.92 ha (1.24%), 10,157.94 ha (3.94%), and 9446.4 ha (3.66%), respectively. There was no obvious shift in woodland except for the ND245 and ND585 scenarios, which decreased slightly by 115.92 ha (0.01%) and 672.3 ha (0.04%), respectively, and EP126, ND126, EP245, and EP585 scenarios, which increased slightly by 719.64 ha (0.04%), 687.24 ha (0.04%), 344.79 ha (0.02%), and 8167.23 ha (0.49%), respectively. None of the scenarios showed an obvious change in grassland. EP126 and ND126 increased by 29,008.26 ha (0.1%) and 22,163.04 ha (0.08%), respectively. Meanwhile, EP245, ND245, EP585, and ND585 decreased by 39,693.69 ha (0.14%), 49,960.26 ha (0.18%), 47,751.3 ha (0.17%), and 59,725.08 ha (0.22%), respectively. In all scenarios, the water increased consistently within the same climate scenario. There were increases in EP126, ND126, EP245, ND245, EP585, and ND585 of 115,210.35 ha (5.25%), 115,210.35 ha (5.25%), 95,430.96 ha (4.35%), 95,400.81 ha (4.35%), 94,113.09 ha (4.29%), and 94,078.26 ha (4.29%), respectively. Except for EP585, which decreased by 1499.76 ha (3.86%), the construction land showed the largest change. EP126, ND126, EP245, ND245, and ND585 all had an upward trend, with increases of 5774.13 ha (14.86%), 6676.02 ha (17.18%), 3555.63 ha (9.15%), 7925.4 ha (20.4%), and 4110.12 ha, respectively.

Land-use transition is primarily driven by the increase in construction land and water (Appendix A). Scenario SSP1-2.6 shows a decrease in arable and unused land, which transform primarily into grassland, water, and construction land. Scenario SSP2-4.5 depicts a considerable proportion of grassland and unused land transforming into water. Meanwhile, construction land also expands to some degree, primarily from unused land and woodland. Within the SSP5-8.5 scenario, cultivated land mainly expands from construction land and unused land, whereas water expands primarily from grassland and unused land. Considering the spatial distribution, grassland is predicted to be the most common land use in Sanjiangyuan, followed by unused land and water. In future, the enormous unused land in the western part of Sanjiangyuan, primarily in Zhiduo County and Golmud City, will be converted into water. Sporadic areas are going to be converted into cultivated land and woodland. Grassland will mainly be converted to water, primarily in Golmud City. Gonghe County has many types of land-use conversion in the lagoon area, while Guide County has a limited portion of construction land transformed into cultivated land. The expansion of construction land is concentrated around cities and encroaches on previously unused land.

3.3.2. Analysis of Driving Factors of Land Use

By using the model (Figure 5), the main driver for cultivated land was average annual temperature, while the secondary factors were population density and distance to country roads. The main driver for woodland was DEM, and the secondary factors were average annual temperature and nighttime light. Grassland is influenced by multiple factors. The primary driver is average annual precipitation, with secondary factors of slope, distance to waters, DEM, and nighttime light. Average annual precipitation has the greatest impact on water, with soil type and distance to country government as secondary factors. The primary driver for construction land was nighttime light, with the secondary factors being distance to country roads and distance to county government. The primary drivers for unused land were DEM and average annual precipitation. Secondary factors were average annual temperature, distance to county roads, and distance to township roads.

3.4. Carbon Storage Change of the Sanjiangyuan Area in 2030 Under Multi-Scenario Simulation

3.4.1. Simulation of Carbon Storage in the Sanjiangyuan Region Within Various Scenarios for 2030

Climate change will have a substantial impact on the spatial assignment of carbon stocks in the study area in the future, with carbon stocks changing to variable degrees from 2020. The study results illustrate that the spatial assignment of carbon stocks in the study area is significantly variable, and the spatial distribution of carbon stocks is the same in the ecological protection and natural development scenarios but not identical to that in 2020. The three climate scenarios under two land-use scenarios basically show the same trend of changes in the eastern, southern, and southeastern edges, with high values, while the western and southwestern edges have low values. It also shows a spatial pattern in which high carbon storage values are scattered, while low values are distributed in patches. Under ecological protection scenarios, SSP1-2.6 has maximum carbon stock, followed by SSP5-8.5, while SSP2-4.5 is minimum. All of the natural development scenarios show a progressive decline in carbon stocks across climate scenarios. SSP1-2.6 has the highest carbon stock, followed by SSP2-4.5. SSP5-8.5 has the lowest carbon stock. When the SSP scenarios are the same, the consequences of various land-use scenarios are mainly localized. Compared with the natural development scenario, ecological protection has higher carbon stocks. To quantify scenario-specific carbon storage across land-use types in the Sanjiangyuan region for 2030, we conducted district-level carbon stock assessments under all six development scenarios (

Table 7. Sanjiangyuan region’s carbon storage in various land-use types in multi-scenario simulation (×10⁶/t).

	EP126	EP245	EP585	ND126	ND245	ND585
Cultivated Land	9.315	9.496	9.733	9.315	9.481	9.707
Woodland	236.718	236.665	237.769	236.714	236.600	236.522
Grassland	2204.940	2199.489	2198.849	2204.397	2198.674	2197.899
Construction Land	0.014	0.013	0.011	0.014	0.014	0.013
Unused Land	1.605	1.625	1.625	1.607	1.627	1.629

). Future developments will incorporate expanded scenario analyses. Future developments will include additional scenarios. The amount of land under each scenario is similar, so the difference in the carbon stock capacity of each category under six scenarios is minimal. The carbon reserves capacity of various land-use types within each development scenario is as follows: grassland > woodland > cultivated land > unused land > construction land. Grassland and woodland have high vegetation coverage and a stronger carbon storage function; therefore, the overall carbon stock is larger. Conversely, cultivated land, unused land, and construction land have lower vegetation coverage, resulting in weaker carbon stock capacity. Water has almost no vegetation coverage and, hence, cannot store carbon. Woodland, cultivated land, and unused land stored the most carbon under SSP5-8.5. However, woodland had the lowest carbon storage in SSP2-4.5, while cultivated land and unused land had the lowest in SSP1-2.6. In contrast, grassland and construction land had the highest carbon reserves in SSP1-2.6 and the lowest in SSP5-8.5. In the natural development scenario, carbon stocks on cultivated land and unused land were highest in SSP5-8.5 and the lowest in SSP1-2.6. Meanwhile, woodland, grassland, and construction land had the lowest carbon stocks under SSP5-8.5. Under similar climate scenarios, grassland, woodland, and cultivated land have higher carbon stocks in the ecological protection scenario than in natural development. In contrast, unused land and construction land have higher carbon stocks in the natural development scenario.

A comparison of carbon stock results in future and in 2020 (

Table 8. Carbon storage change in 2030 under multi-scenario simulation (×10⁶/t).

Year	Scenarios	Grassland	Woodland	Cultivated Land	Unused Land	Construction Land	Total
2020		2202.639	236.617	9.364	1.640	0.012	2450.271
2030	ND126	2204.397	236.714	9.315	1.607	0.014	2452.047
	ND245	2198.674	236.600	9.481	1.627	0.014	2446.396
	ND585	2197.899	236.522	9.707	1.629	0.013	2445.770
	EP126	2204.940	236.718	9.315	1.605	0.014	2452.593
	EP245	2199.489	236.665	9.496	1.625	0.013	2447.289
	EP585	2198.849	237.769	9.733	1.625	0.011	2447.988

) reveals that the carbon stock in both ND126 and EP126 in 2030 is slightly higher than in 2020, while the carbon stock in other scenarios is lower. SSP5-8.5 in the natural development scenario and SSP2-4.5 in ecological protection result in the greatest carbon stock reductions among land-use planning scenarios. For the same climate scenario, natural development reduces carbon stocks more than ecological conservation.

3.4.2. Driving Factors of Spatial Diversity in Carbon Storage

In this study, the geodetector tool was used to choose 15 natural and socioeconomic factors after comprehensive consideration. These factors include soil type (X1), DEM (X2), average annual temperature (X3), average annual precipitation (X4), slope (X5), GDP (X6), population density (X7), nighttime light (X8), distance to railroads (X9), distance to national highways (X10), distance to provincial roads (X11), distance to county roads (X12), distance to township roads (X13), distance to county government (X14), and distance to waters (X15). Except for X8, the p-values of the remaining 14 driving factors are less than 0.001 in geodetector single-factor detection results (Figure 6a). These factors passed the significance test and can explain fluctuations in carbon stock in the Sanjiangyuan region. Under single-factor detection conditions, the social factor with the strongest explanatory power was distance to county roads (0.173). The subsequent factors with the strongest explanatory power were distance to county governments (0.166), distance to township roads (0.161), and distance to provincial roads (0.145). The natural factors with the strongest explanatory strength were annual average temperature (0.139) and DEM (0.139), followed by average annual precipitation (0.092). The factors with the weakest explanatory strength were GDP (0.006) and distance to waters (0.020).

The findings of the interaction factor detection analysis (Figrue 6b) show that the interaction of either factors has a bigger influence on carbon stocks than the discovery of one component. All interaction factors show nonlinear and two-factor amplification effects, demonstrating that the assignment of carbon stocks is influenced by a number of factors. The top three interaction explanatory powers are X12 and X13 (0.229), X2 and X3 (0.227), X3 and X12 (0.227). This indicates that interaction between these factors has the maximum effect on switches in carbon stocks.

4. Discussion

4.1. Effect of Land-Use Shift on Ecosystem Carbon Storage

Terrestrial ecosystems constitute one of the planet’s most significant carbon reservoirs, where carbon stocks demonstrate strong correlations with land-use types and carbon densities [54]. Changes in soil and vegetation carbon sinks caused by transitions in land use will result in changes in the region’s carbon stock capacities [55,56], emphasizing the value of land-use dynamics in understanding the spatiotemporal dynamics of carbon cycling on earth ecosystems [57]. An evaluation of land-use transitions in the research region over the past two decades revealed that cultivated land, water, and construction land generally increased; however, grassland and unused land continued to decline. This conclusion is consistent with the findings of other scholars [58,59,60]. Even if ecological conservation initiatives like managing pastures and converting pasture to grassland have been implemented, industrialization is still accelerating. The Sanjiangyuan region’s water area has expanded more quickly due to the melting of glaciers caused by global warming and humidification, which has reduced the size of the original grassland ecosystem. An increase in high-carbon-density land-use types or a decrease in low-carbon-density land-use types can add local carbon stock, and vice versa. Our analysis demonstrates that carbon stock dynamics are principally governed by transitions between land-use types of differing carbon densities. This study reveals a persistent decline in regional carbon stocks from 2000 to 2020, with high-carbon-density grasslands consistently comprising over 96% of the total carbon storage throughout the observation period. These grasslands, which are crucial components of the Tibetan Plateau’s carbon sink, exhibited a decline during this time, directly contributing to reduced carbon storage in the Sanjiangyuan region. Several studies have confirmed that carbon stock in the Tibetan Plateau region has continuously decreased [61]. The decline in carbon stock was larger between 2000 and 2010 than between 2010 and 2020. While China implemented policies like “returning farmland to forests” in 1995, policy implementation has exhibited temporal lags in achieving desired conservation outcomes [62]. The lasting extension of construction and cultivated land has resulted in a decline in grasslands. Given that these land-use types exhibit lower carbon storage capacity compared to native grasslands, their areal increase fails to offset the carbon sequestration losses from grassland conversion, ultimately leading to continuous carbon stock depletion in the Sanjiangyuan region. Therefore, the reduction in carbon reserves of the Sanjiangyuan region was more significant over this period than during 2010–2020. Our results of spatial distribution analysis revealed that the carbon stock in the research region decreased generally from southeast to northwest. These findings align with previous studies by Shen et al. [63] and Li et al. [64], further validating the consistency of reported spatial patterns across multiple investigations.

Changes in carbon stocks are mediated by multiple interacting drivers, including climate, population growth, economic development, and environmental policy interventions. Our comprehensive modeling framework significantly advances our predictive understanding of future carbon storage variations. Previous studies of the Sanjiangyuan and Tibetan Plateau regions ignored socioeconomic–climate frameworks. This study systematically integrates socioeconomic indicators (e.g., GDP and population density), policy limitations (e.g., ecological red lines), and climate change data (e.g., CMIP6) from SSP scenarios into multiple policy scenarios for conserving the Sanjiangyuan region under the national “dual-carbon” target. This framework significantly improves the spatial accuracy of future LUCC projections for the region. The future LUCC map for the Sanjiangyuan region is more consistent with the scenario’s logic. This study demonstrates that the estimated carbon storage in the Sanjiangyuan region varies significantly depending on the future scenario. SSP1-2.6 emphasizes continuous development and has low carbon emissions, so the carbon stocks of the two land-use planning scenarios are higher than in 2020. SSP2-4.5 maintains the current social development trend, with continuous population growth and rapid social development. In stark contrast, SSP5-8.5 exhibits the most substantial carbon stock depletion. This development pathway prioritizes regional socioeconomic growth at the expense of extensive grassland degradation. Cultivated land and unused land expanded, making the carbon stocks of these two land-use types significantly larger than in other scenarios. Urbanization-driven expansion of construction land leads to extensive land surface sealing and vegetation removal, significantly altering local carbon stocks. Under future scenarios, the alteration of high-carbon-density grasslands to low-carbon-density land-use types, like water and construction land, is projected.

Construction land expansion due to accelerated urbanization results in large areas being covered with concrete and trees being cut down. Under future scenarios, an alteration from high-carbon-density grasslands to low-carbon-density land-use types, like water and construction land, is likely. While the transformation to water and unregulated lands contributes to carbon stock variations, its overall impact remains relatively minor compared to other land-use transitions. The results indicate that, under identical climate scenarios, the natural development scenario has significantly lower carbon stocks than ecological protection in the research area. As a critical component of the Tibetan Plateau’s ecosystem services and a global ecological zone, enhancing this area’s carbon sequestration capacity can promote the Tibetan Plateau’s ecosystem recovery and reconstruction. Therefore, long-term and sustained ecological protection work is necessary in the area. To reconcile ecological conservation with carbon storage optimization in the study region, it is more appropriate to prioritize the creation of ecological protection scenarios. This approach enables systematic comparative scenario analysis to identify optimal land-use configurations while safeguarding critical carbon sink regions. Methodologically, most land-use-based carbon stock studies rely on carbon density data from literature surveys; this study utilizes carbon density data from National Science & Technology Infrastructure’s carbon density dataset and validated from the literature. The national-scale carbon density was calibrated using a reasonably credible precipitation model. However, because of the lack of accurate geographic density, timeliness, and uniformity of carbon density survey data, results from different literature sources frequently differ greatly [65]. As a result, land-use carbon density calculated in this research for the Sanjiangyuan region may differ from real values and is simply a relative reference value.

Land-use carbon stocks in the Sanjiangyuan region exhibit dynamic fluctuations driven by complex natural–socioeconomic interactions. Single factor analysis results show that, among social factors, distance to county road and distance to county government influence spatial carbon stock differentiation. Urbanization agglomeration causes construction land expansion, which reduces regional carbon storage capacity and amplifies spatial heterogeneity. Natural factors affecting spatial differences in carbon reserves are mainly average annual precipitation, DEM, and average annual temperature. These findings are consistent with previous reports by Jiang et al. [66]. The study region exhibits a pronounced southeast–northwest precipitation gradient, with the southeastern sector receiving significantly higher rainfall. Thus, another important factor affecting regional variations in ecosystem carbon stock is average annual precipitation. The geography of Sanjiangyuan is quite undulating, with significant climatic heterogeneity. Elevation and temperature changes can have a more noticeable effect on carbon storage by influencing soil formation and hydrological processes. The findings of interaction detection indicate that the distinction in regional carbon storage is better explained by the interplay between natural and human influences. Consistent with Zhang’s findings, the synergistic interactions between various driving elements significantly enhance the explanatory power for carbon stock patterns [67]. These findings imply that factor interaction should be taken into account in real-world applications. A thorough assessment of how human and natural causes affect local diversification of carbon reserve is crucial. To improve the regional carbon stocks, for instance, varied heights should be used to develop diverse land types. The synergistic interactions between average annual temperature, distance to county roads, and other environmental variables amplify their individual effects on spatial heterogeneity in ecosystem carbon storage. This demonstrates that multivariate interactions constitute as the dominant factors determining spatial patterns of ecosystem carbon stocks in the Sanjiangyuan ecosystem. The spatial heterogeneity and temporal dynamics of carbon stocks in the Sanjiangyuan ecosystem arise from synergistic interactions among multiple environmental drivers. A single factor is insufficient to cause changes in the ecosystem’s carbon stock pattern.

Meanwhile, our analysis of land-use change and carbon stock dynamics identified strong interdependence between two key natural drivers: average annual precipitation and DEM. These factors showed strong mutual correlation and internal consistency in the analysis of land flow and carbon stock change. Precipitation and topography constrain land cover patterns, which have a direct impact on land-use changes. For example, precipitation thresholds determine the lower elevational boundary of forest ecosystems, whereas topography restricts lowland development. They have a significant influence on regional assignments of carbon stores by altering the spatial patterns of land use. For example, carbon stocks are high in forested areas but low in massively human areas. Furthermore, they have a direct influence on carbon stocks via ecosystem progress (like net primary productivity and decomposition rates). Our results establish a mechanistic pathway, whereby precipitation regimes and topographic gradients govern LULC patterns, which, in turn, drive spatial heterogeneity in carbon storage. This demonstrates that precipitation and topography serve as dual determinants of both land-use distributions and carbon stock variability in the study region. This consistency improves our understanding of carbon cycle processes within coupled human–environment systems while demonstrating the fundamental role of natural constraints in mediating land-use–carbon storage relationships. Future land resource management and carbon sink augmentation methods must take into account the spatial restrictions and possibilities for precipitation and topography.

4.2. Land-Use Policy Recommendations

This study systematically evaluates spatiotemporal changes in land use, as well as the primary factors in the Sanjiangyuan region from 2000 to 2020. It was found that rapid expansion of water, construction land, and cultivated land during a 20-year period led to the conversion of a significant amount of ecological land. Based on land-use-type trends and policy constraints, we predict and analyze 2030 land-use distributions under six scenarios. The sustainability-focused SSP1-2.6 scenario demonstrates particular efficacy in grassland restoration, while ecological protection scenarios consistently outperform natural development scenarios in grassland conservation across all climate projections. This is mainly because ecological protection restrains the extension of non-ecological land and encourages its conversion into ecological land. Based on these findings, research provides recommendations for land-use proposals and sustainable development policies in this place.

Ecological protection should occupy a central position in the process of urbanization [68]. Environmental degradation can be effectively slowed with scientific and reasonable urban planning and infrastructure design. To achieve an optimal balance between urban expansion, agricultural production, and ecosystem preservation, rigorous spatial delineation of urban growth boundaries is essential to constrain development within ecologically sustainable limits [69]. During urban border expansion, intrusion into ecological protection and national critical ecological function areas should be strictly avoided, and permanent basic farmland occupation should be minimized [70]. Establishing the ecological red line concept is necessary, and nature reserves and ecosystem restoration areas must be clearly defined to strengthen ecological protection and regional sustainable development [71]. Protected area limits must be carefully established for different ecosystems, such as woodland and grassland. Degraded and damaged original ecological lands must be protected and restored through revegetation and ecological restoration techniques. Targeted ecological restoration projects should be implemented for lands with serious ecological degradation. Long-term ecological restoration efforts, particularly in the Sanjiangyuan region, should prioritize safeguarding regions with high ecological sensitivity and frequent human activity, like cultivated land, woodland, and grassland. It may prevent their conversion into construction land or unused land [72]. Also, it will increase carbon sinks and effectively improve the ecological environment.

5. Conclusions

(1) Between 2000 and 2020, the extent of grassland shrank dramatically, while construction land and water expanded, primarily encroaching on unused land and grassland. EP scenarios maintain grassland more successfully than ND scenarios in terms of expected land use for 2030. Under SSP1-2.6, cultivated and unused land decline, primarily converting to grassland and water. Under SSP5-8.5, cultivated land grows dramatically, while water and building land continue to increase.

(2) From 2000 to 2020, the total carbon reserves in the Sanjiangyuan region continued to decline. Total carbon stock decreased by 9.9×10⁶ t due to a regular decrease in grassland. Grassland is the most vital form of carbon storage in the Sanjiangyuan region and accounts for over 96% of the total carbon stock. The spatial assignment of carbon stock exhibits a type of “east is high, west is low, south is high and north is low.”, with higher carbon density in eastern and southern regions. Carbon stocks in the Sanjiangyuan region are expected to diminish in 2030 compared to 2020 under natural development, although the impact of various land-use planning scenarios is most visible in localized locations when climate scenarios are the same. The carbon stock under ecological protection is larger than in natural development.

(3) The linkages between natural and socioeconomic variables had a substantial impact on the spatial distribution of carbon stocks in Sanjiangyuan. For natural factors, average annual temperature and DEM had the greatest effect on carbon stocks. In terms of social factors, distance to county roads and distance to county governments had the greatest effect. The results of the interaction test showed that combined effect of driving power was more noticeable than the effect of a single factor on spatial variation in carbon stocks.

(4) This study employs an integrated PLUS-InVEST modeling framework to project land-use dynamics in the Sanjiangyuan ecosystem under coupled climate change and policy constraints. The modeling approach provides novel insights into regional carbon stock mechanisms. SSP1-2.6 and ecological protection scenarios encourage grassland growth, implying that policies effectively limit non-ecological land-use increase while promoting restoration. In the face of various future climate scenarios, an ecological conservation policy can significantly enhance grassland acreage. This shows that active management, such as converting agriculture to grassland and limiting development, is more appropriate for ecological restoration than natural development. To ensure the Sanjiangyuan region’s long-term development, it is critical to follow the ecological protection policy, establish an ecological red line, and the spatial layout of land use should be reasonably optimized. Nature reserves and ecosystem restoration zones require precise spatial delineation, with ecological rehabilitation systematically incorporated into regional development frameworks to maintain economic growth within ecological carrying capacity limits. As the science and technology develop, we expect artificial intelligence and machine learning techniques will enable more detailed studies of the Sanjiangyuan region. For example, we may employ machine learning methods to optimize model parameters and improve simulation accuracy or use deep learning to evaluate accurate-resolution remote sensing data and improve forecasts of land-use change.