Spatio-Temporal Variations and Drivers of Carbon Storage in the Tibetan Plateau under SSP-RCP Scenarios Based on the PLUS-InVEST-GeoDetector Model

Abstract

Enhancing carbon storage in terrestrial ecosystems has become a key strategy for mitigating climate change. The Tibetan Plateau holds a pivotal position in achieving carbon neutrality, with the structural pattern of its land use types directly impacting the region’s ecosystem carbon storage capacity. However, there is still a lack of understanding of the spatial distribution of carbon storage in their ecosystems. This study targeted the Tibetan Plateau, utilizing land use data from 2000 to 2020, and employed the Patch-generating Land Use Simulation (PLUS) model to project land use patterns for 2030. By integrating future climate change projections, this study forecasted land use under SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios. The Integrated Valuation of Ecosystem Services and Tradeoffs (InVEST) model was employed to quantify carbon storage from 2000 to 2030, while the GeoDetector model was used to explore the driving influences of factors such as the Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Leaf Area Index (LAI), Net Primary Productivity (NPP), population density, and road network density on carbon storage. The results revealed that: (1) Grassland predominated the land use types on the Tibetan Plateau, with most types having a stability of over 70%, whereas significant changes were observed in the western Tibet Autonomous Region and southern Xinjiang Uygur Autonomous Region. (2) Carbon storage on the Tibetan Plateau generally followed a tendency towards an initial decrease followed by an increase, with an average annual reduction of 50,107,371.79 Mg. The SSP1-2.6 scenario demonstrated the most substantial increase in carbon storage, being 18 times the natural trend, while the SSP5-8.5 scenario indicated the largest decrease. (3) Over the two decades, NDVI emerged as the most influential driver of carbon storage on the Tibetan Plateau, which was maintained at around 0.4, with the interaction between NDVI and NDWI exerting the strongest driving force, which was maintained at around 0.45. The conversion to forestland and grassland was the primary factor accounting for the change in carbon storage. Based on these results, despite the absence of empirical carbon density data, the SSP1-2.6 scenario could be regarded as a reference pathway for carbon storage changes on the Tibetan Plateau. Solely focusing on enhancing carbon storage by converting low-carbon land uses to high-carbon land uses is misguided; sustainable development represents the optimal approach for augmenting carbon storage on the Tibetan Plateau.

1. Introduction

Since the 1760s, the atmospheric concentrations of highly heat-trapping greenhouse gases, such as carbon dioxide, have progressively increased, exacerbating global warming [1]. This trend has not only significantly affected human safety and socio-economic activities but has also disrupted the balance of terrestrial ecosystems and the global carbon cycle to varying extents [2]. Presently, mitigating global warming and enhancing terrestrial ecosystem carbon storage have become focal points of international concern. The fluctuation in carbon storage in terrestrial ecosystems results from a multitude of influences. Beyond the effects of climate change, key determinants include NDVI, NDWI, LAI, NPP, population density, and road network density. However, the predominant factor remains the land use types [3]. Annually, changes in land use account for approximately 30% of global CO₂ emissions, 70% of CH₄ emissions, and 90% of N₂O emissions [4]. Consequently, investigating the spatio-temporal variations in regional carbon storage based on different land use types, elucidating the driving forces behind spatial differentiation, and examining their interrelationships are of substantial practical importance for enhancing carbon storage capacity and optimizing national land use planning [5]. Additionally, these studies provide critical guidance for natural resource management.

The construction of land use types necessitates initial field measurements, which are then extrapolated to broader regions using machine learning algorithms or change simulation models [6]. Various machine learning algorithms, including Support Vector Machines [7], Random Forests [8], and Convolutional Neural Networks [9], have been shown to be effective in delineating land use types. Existing land use change simulation models, such as CA-Markov [10], CLUE-S [11], and PLUS [12], mitigate the limitations of machine learning algorithms by forecasting future scenarios and accurately representing land use types. Nevertheless, many models face challenges in dynamically capturing the spatio-temporal evolution of land patches. The PLUS model, integrating land expansion strategies and multiple random seed types, markedly achieves an optimal land use pattern that closely approximates actual landscape configurations [13]. This model is predominantly applied in multi-scenario simulations at provincial [14], municipal [15], and county scales [16], but its application in constructing land use types within finer geographical subdivisions remains relatively underexplored. Methods for assessing ecosystem carbon storage can be broadly categorized into three approaches: field surveys, remote sensing inversion, and model simulations [17]. Field surveys offer high accuracy but have limited spatial applicability and an inability to investigate historical dynamic processes [18]. Remote sensing inversion can capture large scale spatio-temporal variations but focuses on specific ecosystems, leading to relatively lower accuracy [19]. Compared to the aforementioned methods, model simulation methods provide readily accessible, highly operable data that reflect the interactions between natural and socio-economic factors [20]. These methods are advantageous for assessing the dynamic evolution of regional carbon storage across various scales, effectively addressing the lengthy sampling periods and substantial workloads associated with traditional carbon storage estimation methods [21]. Among these, the InVEST model, which integrates multi-source remote sensing data to assess carbon storage across different land use types, offers rapid computation [22]. It facilitates spatial quantification and can accurately simulate, predict, and assess carbon storage across diverse scales. In recent years, model simulation methods have gained prominence, with an increasing scholarly focus on carbon storage across different activities, scenarios, and scales [23]. The interplay between natural and socio-economic factors is intricate and significantly influences carbon storage dynamics. Optimal temperatures and sufficient precipitation facilitate vegetation growth and enhance carbon storage, whereas extreme climatic conditions can result in vegetation degradation and diminished carbon storage [24]. Urbanization and agricultural expansion typically lead to the reduction of forestland and grassland. Regions with high population densities, characterized by intensified economic activities and infrastructure development, impose greater stress on natural environments, thereby potentially reducing carbon storage capacity [25]. However, analyzing carbon storage variations solely through the lens of land use type changes while elucidating the spatio-temporal evolution patterns and future trends of carbon storage necessitates enhanced quantitative analysis of the coupled interactions between natural and socio-economic factors [26]. Exploring the influence of multiple factors on the dynamic changes in regional carbon storage is imperative for a comprehensive understanding of the driving mechanisms underlying regional ecosystem carbon storage.

The Tibetan Plateau, situated as the “Third Pole”, stands as a critical ecological security barrier for China and the wider Asian region, distinguished by its pronounced carbon storage capabilities [27]. Extant research underscores that during the recent 40 years, the Tibetan Plateau has undergone a warming rate approximately double the global average, with the severity of climate warming continuously escalating [28]. Under the influence of natural and socio-economic factors, the ecosystem of the Tibetan Plateau has experienced significant transformations. For example, climate warming can induce the transition of glacier, grassland, and desert ecosystems into wetland ecosystems [29,30]. Although many studies have been carried out, the complexity and diversity of the Tibetan Plateau result in an unclear understanding of the effects and driving mechanisms of land use types on ecosystem carbon storage. A more comprehensive evaluation, incorporating multiple natural and socio-economic factors, is required to address these uncertainties. Therefore, it is essential to explore the spatio-temporal patterns and elucidate the driving mechanisms of carbon storage in the Tibetan Plateau. The investigation aimed to probe the dynamics of carbon storage across the Tibetan Plateau, elucidating the spatio-temporal variations therein by integrating diverse natural and anthropogenic factors. Moreover, leveraging forthcoming climate change projections, this study forecasted carbon storage alterations by 2030 under both natural progression and three distinct scenarios. These endeavors aspire to furnish scholarly insights conducive to informing future land use management and bolstering ecological security on the Tibetan Plateau. Furthermore, the findings were poised to furnish a scholarly basis for advancing ecological preservation and fostering sustainable development agendas.

2. Materials and Methods

2.1. Study Area

The Tibetan Plateau, located within the geographical coordinates of 25.98–39.82° N and 73.43–104.67° E, spans from the eastern Qinling Mountains to the western Pamir Plateau and from the northern Kunlun Mountains–Alataw Mountains–Qilian Mountains to the southern Himalayas [31]. Administratively, it comprises Qinghai Province and the Tibet Autonomous Region, alongside portions of the southern Xinjiang Uygur Autonomous Region, southwestern Gansu Province, western Sichuan Province, and northwestern Yunnan Province, encompassing an estimated area of 2,574,000 km² [32]. Due to its geographical positioning and topographical characteristics, the plateau exhibits a distinct plateau climate, characterized by a mosaic of climatic zones, including subtropical and temperate regions, featuring intricate hydrothermal conditions [33]. From the southeast to the northwest, mean annual temperatures and rainfall steadily diminish. In the majority of regions, the mean annual temperature is less than 5 °C, and the mean annual rainfall is less than 400 mm [34]. Renowned for its rich ecosystems and biodiversity resources, the Tibetan Plateau serves as the headwaters for major river systems, earning it the epithet of the “Asian Water Tower” (Figure 1). The socio-economic conditions of the Tibetan Plateau are characterized by its unique geographic and climatic challenges, which influence the predominantly pastoral and agricultural livelihoods of its inhabitants. The region’s economy relies heavily on animal husbandry, traditional farming, and increasingly on tourism, driven by its rich cultural heritage and natural landscapes. Despite these opportunities, the area faces significant developmental challenges, including limited infrastructure, scarce arable land, and a relatively low population density, which together contribute to lower average income levels and slower economic growth compared to other regions in China.

2.2. Data Sources and Descriptions

Land use types, classified into six categories: cropland, forestland, grassland, water body, built-up land, and unused land using the reclassification tool; mean annual temperature and mean annual rainfall, using the kriging interpolation tool based on point data; nighttime light (used to characterize the intensity of population activity), NDVI, NDWI, LAI, NPP, and population density, with de-clouded mean processing on the Google Earth Engine cloud platform; distance to road, distance to railway and distance to waterway, using the Euclidean distance tool based on road, railway, and waterway data; and road density, using the kernel density tool based on road data. All the data were collected on 24 April 2024 (

Table 1. Data sources in this study.

Data	Type	Year	Source
			Name	Link
Tibetan Plateau Boundary	Vector	2020	Resource and Environment Science and Data Center	http://www.resdc.cn/
Land use types	Raster	2000–2020	Resource and Environment Science and Data Center	http://www.resdc.cn/
GDP	Raster	2020	Resource and Environment Science and Data Center	http://www.resdc.cn/
Future climatic projections	Raster	2030	Resource and Environment Science and Data Center	http://www.resdc.cn/
DEM	Raster	2020	Earthdata	http://www.earthdata.nasa.gov/
Slope	Raster	2020	Earthdata	http://www.earthdata.nasa.gov/
Mean annual temperature	Vector	2020	Chinese meteorological station	http://data.cma.cn/
Mean annual rainfall	Vector	2020	Chinese meteorological station	http://data.cma.cn/
Nighttime light	Raster	2020	Google Earth Engine	http://code.earthengine.google.com/
NDVI	Raster	2000–2020	Google Earth Engine	http://code.earthengine.google.com/
NDWI	Raster	2000–2020	Google Earth Engine	http://code.earthengine.google.com/
LAI	Raster	2000–2020	Google Earth Engine	http://code.earthengine.google.com/
NPP	Raster	2000–2020	Google Earth Engine	http://code.earthengine.google.com/
Population density	Raster	2000–2020	Google Earth Engine	http://code.earthengine.google.com/
Road	Vector	2000–2020	OpenStreetMap	http://www.openstreetmap.org/
Railway	Vector	2020	OpenStreetMap	http://www.openstreetmap.org/
Waterway	Vector	2020	OpenStreetMap	http://www.openstreetmap.org/

). All the rasters were meticulously resampled to a standardized resolution of 500 m × 500 m and meticulously projected onto the WGS_1984_World_Mercator coordinate system for consistency and comparability (Figure 2).

2.3. Research Framework

The methodology employed in this study encompasses three primary components: Firstly, utilizing land use type data spanning from 2000 to 2020, segmented into five distinct periods, the PLUS model was invoked to forecast the spatio-temporal land use type dynamics for 2030. This prognostication was conducted under three delineated scenarios: natural progression and anticipated future climate alterations. These forecasts were formulated while accounting for nine distinct constraints, including nighttime light and GDP. Secondly, leveraging land use type data spanning from 2000 to 2030 across six discrete periods under the three aforementioned climate scenarios, the InVEST model was deployed to evaluate the spatio-temporal evolutions of ecosystem carbon storage from 2000 to 2030. Noteworthy refinements were integrated into this assessment, accounting for mean annual temperature and rainfall variations. Lastly, utilizing ecosystem carbon storage data from 2000 to 2020, segmented into five periods, the GeoDetector model was employed to unravel the underlying mechanisms driving shifts in ecosystem carbon storage dynamics. This exploration incorporated an analysis of six pivotal driving factors, namely NDVI, NDWI, LAI, NPP, population density, and road network density (Figure 3).

2.4. Methods

2.4.1. PLUS Model

The Geographic CA model serves as a platform for delineating the nuanced dynamics of land use types [35]. Conventional CA models have demonstrated limited efficacy in discerning the root drivers of land use change, particularly in dynamically simulating patch-level variations across diverse land use types such as forestland and grassland [36]. However, the PLUS model significantly mitigates these deficiencies through the integration of the Land Expansion Analysis Strategy (LEAS) rule mining framework and the Class-Assigned Random Seed (CARS) cellular automaton [37]. LEAS facilitates the derivation of development likelihoods for distinct land use types and delineates the contributions of driving factors to the expansion of each land use type. Meanwhile, CARS, integrating random seed generation and threshold decay mechanisms, governs local land use competition dynamics to meet future land use demands. Constrained by development probabilities, it dynamically simulates patch generation in both temporal and spatial domains [38]. The formulations are delineated below.

$$P_{i,j}^c\left(x\right)=\frac{{{\displaystyle \sum }}_{n=1}^{N}I\left(T_n\left(x\right)=d\right)}{N},$$

(1)

$$O{P}_{i,j}^{c,y}=\left\{\begin{array}{c}{P}_{i,j}^c\left(x\right)\times \left(r\times {\mu }_j\right)\times D_j^y if {\Omega }_{i,j}^y=0 and r<P_{i,j}^c\left(x\right)\\ P_{i,j}^c\left(x\right)\times {\Omega }_{i,j}^y\times D_j^y all others\end{array}\right.,$$

(2)

where $i$ denotes the unit; $j$ denotes the type; $x$ denotes the array of factors; $c$ denotes the transition of another type to $j$; $P_{i,j}^c\left(x\right)$ denotes the development likelihood; $n$ denotes the decision tree; $N$ denotes the amount of decision trees; $I$ denotes an indicator function; $T_n\left(x\right)$ denotes the anticipated type; $y$ denotes the year; $O{P}_{i,j}^{c,y}$ denotes the overall development likelihood; $r$ denotes a random value; ${\mu }_j$ denotes the threshold value; $D_j^y$ denotes the influences of future demand; ${\Omega }_{i,j}^y$ denotes the neighborhood effects.

Based on land use type data from 2010 to 2020 in the Tibetan Plateau, and informed by relevant literature [39], this study selected five environmental factors, including DEM, slope, mean annual temperature, mean annual rainfall, and distance to the waterway, as well as four socio-economic factors, including nighttime light, GDP, distance to the road, and distance to the railway. The proportional expansion areas of land use types over two time periods served as domain weights for all land use types (

Table 2. Domain weights for each land use type.

Type *	CL	FL	GL	WB	BL	UL
parametric	0.030	0.167	0.412	0.121	0.009	0.261

* CL denotes cropland; FL denotes forestland; GL denotes grassland; WB denotes water body; BL denotes built-up land; UL denotes unused land.

). This study predicted land use patterns in the Tibetan Plateau for the year 2020 with the help of a Markov model. The resultant Kappa coefficient reached 0.89, the overall accuracy achieved 0.93, and the FOM reached 0.77. Under the condition of achieving satisfactory results, predictions for the land use types in 2030 were made according to the actual data from 2020. Future scenarios are explored using the CMIP6 dataset, including SSP1-2.6, SSP2-4.5, and SSP5-8.5. The SSP1-2.6 scenario was a sustainable development pathway, where the conversion likelihood of unused land to forestland, grassland, and water body increased by 0.2, the conversion likelihood of forestland and water body to other land use types decreased by 0.2, and the rate of expansion of built-up land was appropriately increased; the SSP2-4.5 scenario was an intermediate pathway, where the conversion likelihood of grassland, and unused land to cropland and built-up land increased by 0.2, while the conversion likelihood of cropland and built-up land to other land use types decreased by 0.2, appropriately reducing the rate of expansion of forestland and grassland; the SSP5-8.5 scenario was the traditional fossil fuel pathway, where the conversion likelihood of other land use types to cropland and built-up land increased by 0.4, and the conversion likelihood of cropland and built-up land to other land use types decreased by 0.8, greatly increasing the rate of expansion of cropland and built-up land (

Table 3. Transfer matrix between different land use types.

*	SSP1-2.6						SSP2-4.5						SSP5-8.5
	CL	FL	GL	WB	BL	UL	CL	FL	GL	WB	BL	UL	CL	FL	GL	WB	BL	UL
CL	1	1	1	1	1	0	1	0	0	0	1	0	1	0	0	0	0	0
FL	0	1	0	0	0	0	0	1	0	0	0	0	1	1	1	0	1	0
GL	0	1	1	1	0	0	1	1	1	1	1	0	1	1	1	0	1	0
WB	0	0	0	1	0	0	0	0	0	1	0	0	0	0	0	1	0	0
BL	1	0	0	0	1	0	1	0	0	0	1	0	0	0	0	0	1	0
UL	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1	1

* 0 means that converting is not permitted, and 1 means that converting is permitted.

).

2.4.2. InVEST Model

Terrestrial ecosystem carbon storage is primarily categorized into three fundamental compartments: biomass carbon storage, soil organic carbon storage, and dead organic carbon storage [40]. Within biomass carbon storage, a further distinction is made between above-ground biomass carbon and below-ground biomass carbon. In instances where empirical data are lacking, the InVEST model emerges as the optimal approach for calculating terrestrial ecosystem carbon storage [41]. Its methodological simplicity and modest data requirements render it adept at providing accurate reflections of carbon storage dynamics. Particularly, the Carbon Storage and Sequestration component operates under the assumption of static carbon storage, impervious to temporal fluctuations. Leveraging ecosystem typologies and integrating them with carbon density metrics, this component facilitates the spatial delineation of carbon storage distributions [42]. Regarding the land use types of the Tibetan Plateau, average carbon densities for various carbon pools were computed for each land use type. Consequently, the total carbon storage of all land use types was derived by multiplying their respective areas by their carbon density and summing the results. The formulation is delineated below.

$$C_{total-j}=A_j\times \left(C_{above-j}+C_{below-j}+C_{soil-j}+C_{dead-j}\right),$$

(3)

where $j$ denotes the type; $C_{total-j}$ denotes the total carbon storage; $A_j$ denotes the area; $C_{above-j}$ denotes aboveground biomass; $C_{below-j}$ denotes belowground biomass; $C_{soil-j}$ denotes soil organic matter; $C_{dead-j}$ denotes dead matter.

As per the model user manual, carbon density stands as an indispensable parameter for carbon storage computation. Due to the dearth of empirical data, carbon density values for various land use types were sourced from prior studies, with reference to literature capturing Chinese-level metrics [43]. Nevertheless, its applicability to the Tibetan Plateau region remains uncertain. Leveraging extant scholarship, carbon density may be rectified based on climatic variables, given the robust correlation between biotic, soil organic, and dead carbon densities and annual mean rainfall and temperature. The selection of widely applicable carbon density adjustment formulations with climatic congruence was imperative [44]. Interpolated meteorological observation data revealed that the Tibetan Plateau experienced an annual mean temperature of −1.08 °C and annual rainfall of 374.53 mm in 2020. In contrast, the corresponding national averages for China were 7.34 °C and 609.07 mm; respectively. Thus, calibration coefficients were derived accordingly, as per the following equation: The product of the Chinese carbon density data and the calibration coefficients yields carbon density estimates tailored to the Tibetan Plateau (

Table 4. Four carbon density values for each land use type.

Type	Aboveground Carbon Density	Belowground Carbon Density	Soil Carbon Density	Dead Carbon Density
Cropland	2.484	0.374	95.722	0.971
Forestland	41.656	10.367	110.344	1.845
Grassland	0.402	4.110	88.816	0.077
Water body	0.280	0.990	17.846	1.194
Built-up land	1.793	4.483	73.174	0.262
Unused land	1.214	1.933	22.464	0.932

).

$$K_B=1/2\times \left(\frac{-0.4\times T^{\prime }+43.0}{-0.4\times T^{″}+43.0}+\frac{0.03\times P^{\prime }+14.4}{0.03\times P^{″}+14.4}\right),$$

(4)

$$K_S=1/2\times \left(\frac{-3.4\times T^{\prime }+157.7}{-3.4\times T^{″}+157.7}+\frac{0.07\times P^{\prime }+79.1}{0.07\times P^{″}+79.1}\right),$$

(5)

$$K_D=1/2\times \left(\frac{-0.03\times T^{\prime }+2.03}{-0.03\times T^{″}+2.03}+\frac{0.001\times P^{\prime }+0.58}{0.001\times P^{″}+0.58}\right),$$

(6)

where $K_B$ denotes the correction coefficient for biotic carbon density; $K_S$ denotes the correction coefficient for soil organic carbon density; $K_D$ denotes the correction coefficient for dead carbon density; $T^{\prime }$ denotes the annual mean rainfall of the Tibetan Plateau; $T^{″}$ denotes the Chinese annual mean temperature; $P^{\prime }$ denotes the annual mean rainfall of the Tibetan Plateau; $P^{″}$ denotes the Chinese annual mean rainfall.

2.4.3. GeoDetector Model

The GeoDetector model is a statistical model rooted in spatial statistics and spatial autocorrelation theory, employed to probe the spatial heterogeneity and driving mechanisms underlying target objects [45]. It scrutinizes the impact strength of various perturbing factors on the target variable. Its fundamental principle revolves around assessing whether the cumulative variance of each factor within sub-levels is less than that of the overarching level. In such instances, the perturbing factor exhibits discernible spatial heterogeneity across sub-levels. Furthermore, if two perturbing factors demonstrate congruent spatial distributions, they are deemed statistically correlated [46]. Over the years, the GeoDetector model has found comprehensive application in remote sensing image analysis and related domains. Through factor detection and interaction detection, the GeoDetector model was skillfully used to dissect the driving factors underpinning carbon storage variations within the Tibetan Plateau, unraveling their latent influencing characteristics.

The Factor Detector elucidates the spatial pattern characteristics of dependent variables, namely ecosystem carbon storage variation, and unveils diverse driving factors associated with different independent variables while showcasing their explanatory power regarding the changes in ecosystem carbon storage. The magnitude of their influence is typically quantified using z values. The formulation is delineated below. The Interaction Detector not only reveals the explanatory power of individual independent variables but also delves into the combined effect of two independent variables, thereby illustrating the degree of their joint impact on the dependent variable, ecosystem carbon storage variation [47]. The interaction effects are broadly categorized into nonlinear attenuation, single-factor nonlinear attenuation, and two-factor enhancement (

Table 5. The decision principles of the Interaction Detector.

Basis of Judgment	Interactive Relationship
$z\left(X1\cap X2\right)<min\left[z\left(X1\right),z\left(X2\right)\right]$	Nonlinear weakening
$min\left[z\left(X1\right),z\left(X2\right)\right]<z\left(X1\cap X2\right)<max\left[z\left(X1\right),z\left(X2\right)\right]$	Single-factor nonlinear attenuation
$z\left(X1\cap X2\right)>max\left[z\left(X1\right),z\left(X2\right)\right]$	Two-factor enhancement
$z\left(X1\cap X2\right)=z\left(X1\right)+z\left(X2\right)$	Mutually independent
$z\left(X1\cap X2\right)>z\left(X1\right)+z\left(X2\right)$	Nonlinear enhancement

). Ecosystem carbon storage is not a static value, it may fluctuate because of variations in area and carbon density. Although carbon density cannot directly reflect these fluctuations, its spatial distribution patterns, influenced by different driving factors, may affect ecosystem carbon storage. Extensive research indicates that environmental factors may exert direct or indirect impacts on ecosystem carbon storage, particularly in relation to climatic conditions, physiological parameters, and anthropogenic activities [48]. Hence, this study employed a per-pixel approach to calculate NDVI, NDWI, LAI, NPP, population density, and road network density as independent variables. By exploring the explanatory power of these six factors on carbon storage variations in the Tibetan Plateau over time and space, this study sought to delineate the effects of climatic conditions, physiological parameters, and anthropogenic activities on regional ecosystem carbon storage dynamics.

$$z=1-\left({{\displaystyle \sum }}_{n=1}^N{E}_n{\sigma }_n^2\right)/E{\sigma }^2,$$

(7)

where $z$ denotes the explanatory power of the independent variables concerning carbon storage; $n$ denotes the independent variables; $N$ denotes the count of independent variables; $E$ denotes the unit; ${\sigma }^2$ denotes the overall variance.

3. Results

3.1. Changes in Land Use Types from 2000 to 2030

Notably, certain factors exhibited pronounced influence on specific land use types: for cropland, the primary contributors were found to be annual mean temperature, DEM, and GDP; for forestland, significant impacts were attributed to annual mean rainfall, DEM, and annual mean temperature; and grassland was notably affected by GDP, annual mean temperature, and nighttime light. Similarly, water body demonstrated considerable sensitivity to slope, DEM, and annual mean temperature. Furthermore, built-up land showcased a substantial influence from nighttime light, GDP, and DEM, while unused land was notably affected by annual mean rainfall, slope, and GDP. This comparison underscores the significant impact of natural environmental factors on cropland, forestland, water body, and unused land, whereas socio-economic factors exhibited robust correlations with grassland and built-up land. Remarkably, nighttime light emerged as the predominant contributor in built-up land, accounting for over half of the overall impact (Figure 4).

Between 2000 and 2030, the Tibetan Plateau’s land use types predominantly featured grassland, constituting approximately 48.43% of the total land area. Subsequently, unused land accounted for roughly 33.87%, followed by forestland at approximately 11.38%. Conversely, cropland, water body, and unused land represented relatively minor proportions, with built-up land encompassing a mere 0.11% of the overall land area. Cropland and built-up land exhibited comparable distributions, primarily concentrated in the northeastern and southern sectors of the Tibetan Plateau, encompassing the eastern reaches of Qinghai Province, the southern reaches of the Tibet Autonomous Region, and the southwestern reaches of Gansu Province. Forestland was principally concentrated in the southeastern sectors of the plateau, spanning the eastern reaches of the Tibet Autonomous Region, the southwestern reaches of Gansu Province, and the western reaches of Sichuan Province. Grassland was predominantly concentrated in the eastern and southern locales, spanning the southern reaches of Qinghai Province, the central reaches of the Tibet Autonomous Region, and the western reaches of Sichuan Province. Water body was primarily situated in the northeastern and southwestern regions, encompassing the central reaches of Qinghai Province, the central reaches of the Tibet Autonomous Region, and the southern reaches of the Xinjiang Uygur Autonomous Region. Unused land was chiefly concentrated in the northern and southwestern zones of the Tibetan Plateau, spanning the northwestern reaches of Qinghai Province, the western reaches of the Tibet Autonomous Region, and the southern reaches of the Xinjiang Uygur Autonomous Region. Predominantly influenced by natural environmental determinants, the Tibetan Plateau manifested limited extents of cropland and built-up land, with the majority of its regions characterized by natural land use types such as forestland and grassland.

Between 2000 and 2030, alterations in land utilization patterns transpired within the Tibetan Plateau. Across the entirety of the period, cropland, forestland, water body, built-up land, and unused land demonstrated an ascending trajectory, notably with built-up land exhibiting the most pronounced surge at approximately 87.51%. Sequentially, water body, unused land, forestland, and cropland followed suit, while grassland exhibited a discernible decrement of about 16.08%. In terms of comprehensive shifts, cropland evinced a stability quotient of 75.51%, primarily transitioning to grassland, predominantly distributed in the southern reaches of the Tibet Autonomous Region and the southern extents of Gansu Province. Forestland manifested a steadiness level of 82.78%, denoting the most gradual evolution among land use types across the Tibetan Plateau, with principal transitions occurring towards grassland, notably concentrated in the southeastern zones of the Tibet Autonomous Region and the western zones of Sichuan Province. Grassland displayed a stability quotient of 70.19%, evidencing the most pronounced transformations among land use types in the Tibetan Plateau, predominantly shifting towards unused land, notably distributed in the western zones of the Tibet Autonomous Region and the southern reaches of the Xinjiang Uygur Autonomous Region. Water body exhibited a stability quotient of 74.18%, primarily transitioning towards unused land, predominantly located in the southeastern zones of the Tibet Autonomous Region and the southern extents of the Xinjiang Uygur Autonomous Region. Built-up land demonstrated a stability quotient of 74.25%, predominantly transitioning towards cropland, notably concentrated in the eastern zones of Qinghai Province and the southeastern zones of Gansu Province. Unused land demonstrated a stability quotient of 72.68%, primarily transitioning towards grassland, notably distributed in the southern extents of Qinghai Province and the southern reaches of the Xinjiang Uygur Autonomous Region. Despite variances in the stability of land use types, the prevailing trend remained consistently above 70%, indicative of a proclivity towards stability in land use types across the Tibetan Plateau. This was largely attributed to its geographical location within the southwestern expanse of China, where socio-economic conditions were relatively less developed, engendering diminished population densities. Furthermore, amidst circumstances wherein the natural milieu remained predominantly undisturbed, land use types tended to maintain their incumbent status quo on a macroscopic scale. Nonetheless, owing to the extensive expanse of the Tibetan Plateau, transitions between distinct land use types persisted and were primarily concentrated within the Tibet Autonomous Region (Figure 5).

Examining the scenarios projected for 2030, it was discerned that under the SSP1-2.6 scenario, minimal expansion was evident in cropland, built-up land, and unused land, while the SSP5-8.5 scenario exhibited the least extent of forestland, grassland, and water body. Conversely, the SSP1-2.6 scenario demonstrated a pronounced increase in forestland, grassland, and water body, whereas the SSP5-8.5 scenario depicted a substantial expansion in cropland, built-up land, and unused land. This delineated a discernible trend where natural land use types were progressively dwindling while anthropogenic land use types were correspondingly expanding, notably attributable to escalating carbon dioxide emissions. In deviation from conventional trajectories, the SSP1-2.6 scenario notably witnessed a marked augmentation in forestland, grassland, and water body, alongside a gradual ascent in built-up land, predominantly concentrated along the confluence of Qinghai Province, Tibet Autonomous Region, and Xinjiang Uygur Autonomous Region. The salient characteristic of the SSP2-4.5 scenario lied in the marginal increase in cropland and built-up land, juxtaposed with a gradual elevation in forestland and water body, albeit with grassland undergoing a diminishing trend, primarily prevalent in the western fringes of Qinghai Province and the northern precincts of the Tibet Autonomous Region. Conversely, the SSP5-8.5 scenario predominantly manifested a substantial expansion in cropland and built-up land, concomitant with a significant reduction in forestland, grassland, and water body, predominantly distributed across the eastern realms of Qinghai Province, the southern territories of the Tibet Autonomous Region, and the southwestern regions of Gansu Province. These regions undergoing transformation typically exhibited harsher natural environments or heightened population densities, indicative of more pronounced anthropogenic interventions. Overall, the tripartite scenarios exhibited distinct advantages and disadvantages: the SSP1-2.6 scenario emerged as a protracted aspiration, emphasizing augmented anthropogenic land utilization while safeguarding natural land; the SSP2-4.5 scenario was envisioned as a provisional ambition, accentuating heightened anthropogenic land utilization while compromising only select portions of natural land; and the SSP5-8.5 scenario served as a cautionary paradigm, spotlighting the hazards entailed in prioritizing regional economic expansion at the expense of sustainable ecological equilibrium (Figure 6).

3.2. Changes in Carbon Storage from 2000 to 2030

Based on the outcomes derived from the InVEST model, the carbon storage dynamics of the Tibetan Plateau manifested a temporal trajectory characterized by an initial decline followed by a subsequent increase, eventually converging towards a state of relative stability. Overall, there was a discernible decrement in total carbon storage, amounting to approximately 1,503,221,153.67 Mg, spanning the temporal spectrum from 2000 to 2030. Notably, the nadir of carbon storage, approximately 507.75 Mg, was predominantly localized within the southwestern and northern extents of the Tibetan Plateau, encompassing regions such as the northwestern Qinghai Province, western Tibet Autonomous Region, and southern Xinjiang Uygur Autonomous Region, typified by prevalent land use types of water body and unused land. Conversely, the apogee of carbon storage, reaching approximately 4105.30 Mg, predominantly emanated from the southeastern fringes of the Tibetan Plateau, encompassing locales such as the southeastern Qinghai Province, eastern Tibet Autonomous Region, and northern Sichuan Province, characterized by land use types predominantly comprised of forestland and grassland. The conspicuous fluctuations in carbon storage observed during the interval from 2005 to 2010 were primarily ascribed to the extensive conversion of grassland in the Tibet Autonomous Region into water body and unused land, underscored by the substantially higher carbon density inherent in grassland compared to the cumulative density of water body and unused land.

Between 2000 and 2030, fluctuations in carbon storage across the Tibetan Plateau were markedly influenced by fluctuations in land use types. Under the condition of uniform pixel dimensions, whereby land use extents remained constant, transitions from land use types characterized by low carbon densities to those with elevated carbon densities invariably yielded an augmentation in the aggregate carbon storage. Typically observed conversions encompassed the transformation of cropland into forestland, grassland into cropland or forestland, water body into alternative land use types, and built-up land into cropland, forestland, or grassland. Conversely, transitions from land use types typified by high carbon densities to those exhibiting lower densities invariably precipitated a reduction in overall carbon storage. Such transitions commonly involved conversions from cropland into grassland, water body, built-up land, or unused land; forestland into alternative land use types; grassland into water body, built-up land, or unused land; built-up land into water body or unused land; and unused land into water body. These transitions were predominantly concentrated in the southern expanse of Qinghai Province, the eastern precincts of the Tibet Autonomous Region, the southern reaches of the Xinjiang Uygur Autonomous Region, and the northwestern reaches of Sichuan Province for the former scenario, while for the latter scenario, they were primarily concentrated in the western expanse of the Tibet Autonomous Region, the southern precincts of the Xinjiang Uygur Autonomous Region, and the northern precincts of Sichuan Province (Figure 7).

In the SSP1-2.6 scenario, the projected carbon storage of the Tibetan Plateau for 2030 stood at 28,197,875,296.83 Mg, indicating an increase of 135,636,488.14 Mg compared to 2020. Deviating from historical trends, this augmentation represented a substantial elevation of 128,104,939.13 Mg, a notable eighteenfold surge. This underscored an ascending trajectory in carbon storage under this sustainable development pathway, with a spatial distribution generally adhering to an east-west, south-north gradient. In contrast to prior trends, conspicuous decrements were observed in select regions such as the Haixi Mongolian-Tibetan Autonomous Prefecture, Yushu Tibetan Autonomous Prefecture, Nagqu City, Ali Prefecture, and Kashgar Prefecture, while discernible increments manifested in areas such as the Haibei Tibetan Autonomous Prefecture, Chamdo City, Nyingchi City, Shannan City, and Shigatse City. In the SSP2-4.5 scenario, the anticipated carbon storage of the Tibetan Plateau in 2030 was estimated at 28,106,342,770.92 Mg, presenting an uptick of 44,103,962.23 Mg compared to 2020. Diverging from past trends, this elevation represented a substantially higher surge of 36,572,413.22 Mg, marking a sixfold amplification. This suggested an ascendant trend in carbon storage under this intermediate trajectory, albeit without the pronounced effects observed in the sustainable development scenario. Spatial distribution conformed to the east-west, south-north gradient. Unlike historical trends, notable decrements occurred in locales such as Xining City, Haixi Mongolian-Tibetan Autonomous Prefecture, Yushu Tibetan Autonomous Prefecture, Nagqu City, and Ali Prefecture, while evident increments manifested in areas namely Chamdo City, Nyingchi City, Shannan City, Aba Tibetan-Qiang Autonomous Prefecture, and Garze Tibetan Autonomous Prefecture. In the SSP5-8.5 scenario, the estimated carbon storage of the Tibetan Plateau in 2030 was projected to reach 28,066,810,563.32 Mg, indicating an increase of 4,571,754.63 Mg compared to 2020. In contrast to historical trends, this represented a notably lower elevation of 2,959,794.38 Mg, markedly attenuating the trend of carbon storage augmentation. This indicated that under this fossil fuel-intensive pathway, while the carbon storage of the Tibetan Plateau exhibited a growth trend, it gradually diminished in the long term, signifying potential hazards. The spatial distribution conformed to the east-west, south-north gradient. Deviating from prior trends, prominent decrements occurred in locales such as Hainan Tibetan Autonomous Prefecture, Haixi Mongolian-Tibetan Autonomous Prefecture, Yushu Tibetan Autonomous Prefecture, Nagqu City, and Lhasa City, while marked increments manifested in areas such as Ali Prefecture, Hotan Prefecture, Aba Tibetan-Qiang Autonomous Prefecture, Garze Tibetan Autonomous Prefecture, and Deqen Tibetan Autonomous Prefecture. Overall, regions of augmented carbon storage were chiefly concentrated in the southeast, while areas of diminished carbon storage were primarily concentrated in the northwest of the Tibetan Plateau. This phenomenon was attributed to superior natural and anthropogenic conditions in the southeast compared to the northwest, with the latter facing more pronounced governance challenges and complexities (Figure 8).

From the perspective of the three scenarios in 2030, the SSP1-2.6 scenario exhibited the highest total carbon storage value, while the SSP5-8.5 scenario showed the lowest total carbon storage value. The most notable disparities among the scenarios manifested in the eastern, southern, and western regions of the Tibetan Plateau. Specifically, in the eastern segment of the plateau, alongside prevalent natural land use types such as forestland and grassland, there was a notable presence of cropland and built-up land. Amid anthropogenic perturbations, inevitable transitions occurred within natural land use types, rendering the preservation of their pristine state challenging. The extension of cropland and built-up land altered the ambient land use landscape, consequently influencing their carbon storage. The southern sector of the plateau was primarily characterized by forestland and grassland, albeit devoid of dense human habitation. Nevertheless, owing to significant factors such as slop, mean annual temperature, and rainfall patterns, its ecological stability was compromised, prompting the conversion of forestland and grassland into alternative land use types and thus shifts in carbon storage dynamics across different scenarios. Conversely, the western expanse of the plateau, sparsely populated, is contended with extreme environmental exigencies typified by elevated terrain, rugged topography, and diminished mean climatic parameters. As a consequence, certain locales might be unsuitable for sustained vegetative growth, fostering rapid degradation of natural land use types such as forestland and grassland, thereby transmuting towards unused land. This process engendered commensurate increments or decrements in the degraded zones, occasioned by alterations in land use types (Figure 9).

3.3. Drivers of Ecosystem Carbon Storage

Carbon storage within terrestrial ecosystems exhibits dynamic variability, contingent upon differences in both area coverage and carbon density [49]. Given the inherent limitations of carbon density as a sole metric for capturing such fluctuations, the resultant carbon storage manifests heterogeneous spatial distribution patterns, indicative of the diverse array of driving forces at play. Environmental quality, characterized by climatic conditions, physiological parameters, and anthropogenic activities, yields varying degrees of direct and indirect impacts on carbon storage. Considering the relatively sparse population density and developmental status of the Tibetan Plateau, anthropogenic influences were discounted, while a pixel-by-pixel methodology was employed to compute key explanatory variables such as NDVI, NDWI, LAI, NPP, population density, and road network density. Thus, spanning both temporal and spatial dimensions, the efficacy of these aforementioned factors in elucidating shifts in carbon storage across the Tibetan Plateau was scrutinized, shedding further light on the nuanced interplay between climatic conditions, physiological parameters, and regional ecosystem carbon dynamics.

Notably, the interactional effects of factors surpassed those of their individual counterparts, demonstrating a bifold augmentation modality. In terms of factor detectors, over the past two decades, NDVI has exhibited the highest explanatory power, characterized by an initial decline followed by an increase. In contrast, NDWI and road network density showed the weakest explanatory power. NDWI followed a pattern of initial decline, subsequent increase, and final decline, whereas road network density demonstrated a consistently increasing trend. LAI and NPP displayed comparable explanatory power; the former showed an initial increase followed by a decrease, while the latter exhibited a persistent upward trend. Meanwhile, population density remained relatively stable, consistently around 0.09 (Figure 10). This thematic progression reflected the preponderance of grassland as the dominant land utilization category across the Tibetan Plateau, distinguished by copious vegetative coverage. Consequently, NDVI adeptly delineated growth status, with elevated indices indicative of optimal vegetative vitality and heightened carbon accumulation, thereby correlating with elevated carbon storage. In terms of interaction detectors, over the past two decades, the interaction between NDVI and NDWI has demonstrated the highest strength. In contrast, the interactions between population density and road network density were the weakest in 2000 and 2005, while those between NDWI and population density were the weakest in 2010, 2015, and 2020 (Figure 11). This could be attributed to the relatively humid climate and high annual rainfall found on the Tibetan Plateau, which encouraged dense vegetation to grow. Under the premise that NDVI held the strongest explanatory power, the interaction between NDVI and NDWI emerged as a more effective driving force for the region. However, the inherent explanatory power of NDWI, population density, and road network density was relatively low, generally below 0.1. Consequently, aside from NDVI, the interactions of other factors with these indices were not markedly significant. In summation, NDVI emerged as the quintessential driving force underlying carbon storage dynamics in the Tibetan Plateau, advocating for intensified ecological conservation and amelioration endeavors within the region.

4. Discussion

4.1. Explanation of the Drivers of Carbon Storage and Their Suggestions

It was apparent that land-use types exert a predominant influence on carbon storage. In comparison with other studies [50], the trends and spatial distribution patterns of carbon storage in the Tibetan Plateau observed in this research were consistent. Furthermore, the carbon density values employed in this study have been corrected for temperature and rainfall, thereby improving their accuracy. Typically, when the collective carbon density of land use types is high—encompassing aboveground biomass carbon density, belowground biomass carbon density, soil organic carbon density, and dead carbon density—the ecosystem’s carbon storage in the region tends to be substantial; conversely, it is diminished [51]. However, the magnitude of carbon storage did not invariably denote an optimal condition, as delineated by the hierarchy of carbon density values: forestland > cropland > grassland > built-up land > unused land > water body. Presently, built-up land constituted a negligible proportion of the Tibetan Plateau, and notwithstanding its elevated carbon density value, it could not substantiate the conversion of other land use types having lower carbon density values into built-up land. It was imperative to holistically consider ecological environments and other constraining factors. Under judicious policy planning, the extension of cropland, built-up land, and other natural land use types was indispensable but not without constraints. Similarly, land use types with diminished carbon density values, such as water body, warrant attention. Despite their relatively marginal role in carbon storage, they epitomized the fundamental constituents of sustainable development and were indisputably significant. Concurrently, natural land use types such as forestland and grassland, characterized by heightened carbon density values, necessitate prioritization [52]. Leveraging the distinctive geographical locale of the Tibetan Plateau, targeted management strategies should be devised based on environmental parameters such as temperature and rainfall to curtail their conversion into unused land and ensure the perpetuation of their ecological functionalities, such as carbon storage and hydrological regulation. Unused land, typified by reduced carbon density values, could be repurposed into other land use types through afforestation endeavors.

Taking into account the driving factors, natural elements such as NDVI exhibited a significant influence on carbon storage. It was crucial to enhance the role of NDVI in carbon storage to sustain the carbon reserves in the southeastern region of the Tibetan Plateau [53]. Strategic focus should be directed towards the northwestern region, where environmental management interventions could fortify stability and subsequently augment carbon storage. Conversely, anthropogenic factors, including population density, demonstrated a relatively minor driving effect. Human activities and associated ecological policies could impact land use patterns and carbon storage, but these effects were predominantly observed in densely populated urban areas. The Tibetan Plateau encompassed a limited number of such urban regions, for instance, Lhasa and Xining, while the majority of the area remained sparsely inhabited [54]. Therefore, the influence of population density and road network density, as well as their interactions with other factors, was generally limited. Overall, considering prospective scenarios of carbon storage, the Tibetan Plateau ought to tailor strategies commensurate with local exigencies, adhering to the dictum of afforestation where appropriate, grassland rehabilitation where fitting, agricultural cultivation where viable, and pastoralism where applicable, while respecting scientific precepts, averting degradation of forestland, grassland, and water body, fostering the orderly extension of cropland and built-up land, and implementing holistic management encompassing mountains, rivers, forests, fields, lakes, and grasslands, thereby realizing high-caliber development across the Tibetan Plateau.

4.2. Limitations and Future Perspectives

In the context of the PLUS model computations, the anticipation of forthcoming scenarios pertaining to land use types entailed a subjective element. With a data resolution set at 500 m, the processing of data posed considerable challenges, necessitating manual adjustments to parameters such as transition matrices, neighborhood weights, and transition probabilities within the PLUS model to attain the desired precision [55]. Within the purview of the InVEST model computations, it was noteworthy that carbon density values were not directly measured but rather derived from refined formulas aimed at enhancing previous outcomes, thereby requiring further refinement to enhance accuracy. Moreover, it was essential to underscore that the carbon density data in the InVEST model were static values, which might exhibit fluctuations in reality due to diverse factors such as climate variability, vegetation composition, and soil attributes, without accounting for temporal seasonality changes [56]. If one considered the possibility of varying carbon density across different terrestrial ecosystems at smaller spatial scales, it became apparent that such densities could not remain constant. Nonetheless, it was imperative to acknowledge that numerous studies had validated the suitability of this model for large-scale estimation of carbon storage, thus affirming the acceptability of the outcomes derived from this study. In subsequent research endeavors, a continuous and dynamic monitoring approach to assess carbon density in the Tibetan Plateau should be pursued, with measured data being incorporated into adjustments of carbon density values to mitigate assessment uncertainties. Furthermore, the inclusion of additional driving factors as independent variables in carbon storage analyses can provide deeper insights into their respective mechanisms influencing ecosystem carbon storage dynamics in the Tibetan Plateau.

5. Conclusions

This study utilized the PLUS-InVEST-GeoDetector model to conduct an assessment of carbon storage on the Tibetan Plateau from 2000 to 2030. It investigated the variations in carbon storage under different future climate scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5), elucidating the influential roles of NDVI, NDWI, LAI, and NPP, population density, and road network density. The ensuing conclusions were delineated as follows:

Across the temporal span from 2000 to 2030, grassland predominated as the primary land use type across the Tibetan Plateau, maintaining its hegemony consistently. Noteworthy expansions were discerned in cropland and built-up land, with all land use types exhibiting a trend towards stabilization, albeit marked by significant regional transformations, particularly in the western expanse of the Tibet Autonomous Region and the southern precincts of the Xinjiang Uygur Autonomous Region. Informed by the dynamics of land use types, carbon storage on the Tibetan Plateau exhibited an initial phase of decline followed by subsequent increments. Under the auspices of all three prospective scenarios, carbon storage escalated, conforming to a spatial pattern with elevated values in the southeast and subdued values in the northwest. However, the SSP1-2.6 scenario was envisaged as a protracted trajectory (+128,104,939.13 Mg), SSP2-4.5 as an interim course (+36,572,413.22 Mg), and SSP5-8.5 solely as a precautionary vector (−2,959,794.38 Mg), underscoring the imperative of prioritizing ecological preservation over the singular pursuit of augmenting carbon storage. Additionally, NDVI emerged as the most potent driver of carbon storage on the Tibetan Plateau, while NDWI and road network density evinced the least influence, underscoring the primacy of accentuating natural land use types such as forestland and grassland. Future endeavors should maintain a focus on ecological preservation and aspire towards low-carbon development, with particular attention to the northwestern region of the Tibetan Plateau. It is imperative to safeguard vulnerable grassland ecosystems by employing ecological compensation mechanisms to rigorously regulate livestock population densities.