Spatio-Temporal Evolution and Multi-Scenario Modeling Based on Terrestrial Carbon Stocks in Xinjiang

Abstract

The increase in atmospheric CO₂ leads to global warming and ecological environment deterioration. Carbon storage modeling and assessment can promote the sustainable development of the ecological environment. This paper took Xinjiang as the study area, analyzed the spatial and temporal evolution of land use in four periods from 1990 to 2020, explored the spatial relationship of carbon stocks using the InVEST model, and coupled the GMOP model with the PLUS model to carry out multiple scenarios for the future simulation of land use in the study area. We found (1) Over time, the types with an increasing area were mainly impervious and cropland, and the types with a decreasing area were grassland, snow/ice, and barren; spatially, the types were predominantly barren and grassland, with the conversion of grassland to cropland being more evident in the south of Northern Xinjiang and north of Southern Xinjiang. (2) The evolutionary pattern of terrestrial carbon stocks is increasing and then decreasing in time, and the carbon sink areas are concentrated in the Tarim River Basin and the vicinity of the Ili River; spatially, there are differences in the aggregation between the northern, southern, and eastern borders. By analyzing the transfer in and out of various categories in Xinjiang over the past 30 years, it was obtained that the transfer out of grassland reduced the carbon stock by 5757.84 × 10⁴ t, and the transfer out of Barren increased the carbon stock by 8586.12 × 10⁴ t. (3) The land use layout of the sustainable development scenario is optimal under the conditions of satisfying economic and ecological development. The reduction in terrestrial carbon stocks under the 2020–2030 sustainable development scenario is 209.79 × 10⁴ t, which is smaller than the reduction of 830.79 × 10⁴ t in 2010–2020. Land optimization resulted in a lower loss of carbon stocks and a more rational land-use layout. Future planning in Xinjiang should be based on sustainable development scenarios, integrating land resources, and achieving sustainable economic and ecological development.

1. Introduction

Terrestrial ecosystems are an important component of the global carbon pool and play a key role in CO₂ uptake, carbon cycling, and climate regulation [1]. Vegetation, soil, and dead organic matter are the main sources of carbon in terrestrial ecosystems [2]. The uptake of atmospheric CO₂ through vegetation and soil reduces its concentration in the atmosphere and increases the carbon sink capacity of carbon sources [3,4]. Therefore, investigating carbon storage in terrestrial ecosystems (hereinafter referred to as “terrestrial carbon storage/carbon stock”) and enhancing carbon sequestration capacity have become hot study topics for solving global climate problems.

Land use/cover change (LUCC) changes the land use pattern and ecosystem structure. It affects the circulation of materials and energy flow in the ecosystem and the capacity of atmospheric carbon dioxide storage [5]. There is a great difference in the carbon sink capacity of carbon sources among different land classes; for example, agricultural land in China produces 15% of the total carbon emissions from carbon dioxide [6], whereas forest absorbs 80% of the country’s total carbon sequestration [7]. Therefore, assessing the terrestrial carbon stock and quantifying its relationship with the LUCC is conducive to optimizing and adjusting the LUCC structure and the low-carbon development of sink enhancement. It can regulate regional climate, guide regional low-carbon development, and accelerate the realization of China’s “3060 goal”. Therefore, more researchers are studying the relationship between the LUCC and land reserves.

Foreign researchers recognized in the early 20th century that LUCC study was vital to solving global change problems [8]. With the development of the study on global change, LUCC has led to changes in the global land surface, causing changes in the ecological environment. Early foreign researchers explored LUCC mainly through field surveys and land class classification to carry out this study, consuming considerable manpower and material resources. Domestic LUCC studies have been relatively late compared to international development. Still, with the progress of science and technology and the accelerated development of urbanization in China, the LUCC study achieved good results. For example, Lin et al. simulated the land use and land cover (LULC) changes in a major city (Korla) around the Tarim River [9]. Land use studies in China mainly analyze LUCC based on the degree of land use change, land use transformation characteristics, and land use change driving forces [10].

To date, many researchers have studied the spatial prediction of land use [11], such as the Cellular Automata (CA) model, Conversion of Land Use and its Effects (CLUE) model, Future Land Use Simulation (FLUS) Model, and patch-generating land use simulation (PLUS) model [12,13,14,15]. The PLUS model simplifies the analysis of land use change while supporting the ability of multi-type and complex land use change and better simulating the growth of multi-land class patches for land structure optimization.

Researchers from various fields have studied terrestrial carbon stocks from multiple perspectives, including carbon stock modeling, estimation methods, influencing factors, and their values [16]. Terrestrial carbon stocks have been studied abroad for a long time. In the early stages, soil carbon stock was mainly estimated by the acquired soil carbon density and then combined with mathematical statistics to construct the global soil carbon stock [17]. When using the Integrated Valuation of Ecosystem Services and Trade-offs (InVEST) model to carry out terrestrial carbon stock study, it can study the vegetation carbon stock and compensate for the lack of soil carbon stock study in the Carnegie–Ames–Stanford Approach (CASA) model, which is easy to operate, spatialize, and visualize [18,19]. Some researchers have combined the InVEST model with LUCC to analyze changes in terrestrial carbon stocks in LUCC results [20,21,22]. For example, Li et al. combined the Slope, Land Use, Exclusion, Urban, Transportation, and Hillshade (SLEUTH) models with the InVEST model to assess the impact of urban growth on carbon stocks in Wuhan [23]. Zhu et al. used the CA-Markov model to predict LUCC, combined with the InVEST model to assess carbon stocks in arid areas, analyze the impact of LUCC changes on carbon stocks, and make predictions for the future [24]. This paper is one of the few studies that combine all three modules of LUCC, carbon stock, and multi-scenario modeling. Moreover, the Xinjiang Uyghur Autonomous Region (hereinafter referred to as Xinjiang) is taken as the study area and analyzed in pieces to make the study more detailed.

Xinjiang is in Northwest China and accounts for approximately one-sixth of the country’s total area. Low average annual precipitation, uneven distribution of water resources, and irrational exploitation of land there have led to serious ecological problems. For example, the region has a fragile ecological environment, low land-carrying capacity, obvious changes in land-use types, longstanding land desertification, soil erosion, and grassland degradation. These problems can also lead to the region being affected by droughts and sandstorms. Soil erosion caused by the ravages of sandstorms results in low carbon sequestration capacity, with indirect or direct impacts on carbon stocks. These problems are not conducive to protecting the ecological environment or future sustainable development [25,26]. Since the 13th Five-Year Plan, ecological problems in Xinjiang have become a focus of the government’s attention. The autonomous region uses remote sensing image data to comprehensively evaluate the ecological environment status of the entire territory, key regions, and watersheds. The management of mountain, water, forest, field, lake, grass, and sand systems and the protection of the ecological environment in the entire region is carried out through methods of investigation, monitoring, and assessment [27]. Therefore, it is of great significance to conduct a study on land-use change and carbon stocks in Xinjiang, which has abundant land resources and an underdeveloped economy.

This paper takes Xinjiang as the study area and focuses on “land use change—spatial and temporal evolution of terrestrial carbon stocks—scenario simulation—policy recommendations”. Firstly, the spatiotemporal evolution and transformation characteristics of the LUCC in the Xinjiang region were analyzed. Secondly, based on the InVEST model, the spatial and temporal evolution of terrestrial carbon stocks and the impact of the transfer of each land type on terrestrial carbon stocks were analyzed using the LUCC and corrected carbon density data. Finally, Grey Multi-objective Programming (GMOP) and genetic algorithms were used to set up multiple scenarios to simulate the future land-use distribution in Xinjiang for different needs. The above analyses of land-use changes and terrestrial carbon stocks in multiple scenarios provide a basis for land-use optimization and layout in Xinjiang. This study will help optimize land space and ecological environmental protection in the ecologically fragile and complex Xinjiang and provide a reference value to realizing China’s “dual-carbon” goal.

2. Study Area Overview

Xinjiang Uygur Autonomous Region is the hinterland of the Asia–Europe continent, located in Northwestern China (73°40′ E–96°23′ E, 34°25′ N–49°10′ N). With more than 5600 km of land borders and eight neighboring countries, Xinjiang’s strategic position is very important. Historically, it has been recognized as an important Silk Road corridor and a necessary route for the second “Asia–Europe Continental Bridge”. The study area is 1950 km wide from east to west and 1550 km long from north to south. It is the largest province in China in terms of land area, accounting for about one-sixth of the country’s total land area (Figure 1). The topographic features of Xinjiang are referred to as “three mountains and two basins”; i.e., mountain ranges and basins are arranged one after the other. This topography divides Xinjiang into northern and southern halves, with the Junggar Basin to the north and the Tarim Basin to the south. There are clear spatial differences in elevation in Xinjiang, leading to large differences in precipitation and fragile ecological environments in the region. Xinjiang is in the arid northwestern part of China, where the temperature differences between the coldest and hottest months reach more than 30 °C. The sunshine situation is better, with less precipitation, higher evaporation rates, and an average annual precipitation of 199.6 mm across the region (according to the 2020 Statistical Yearbook). Xinjiang has more than 500 rivers distributed in the northern and southern basins of the Tianshan Mountains. It is home to the famous Ten Great Lakes, which account for approximately 0.6% of its total area and has beautiful natural scenery. The glaciers in the region are well developed, accounting for about two-fifths of the total glacier area in China, and are, therefore, known as “natural solid reservoirs”. Xinjiang is the largest province in China, with a population of 25.9 million as of 2020, comprising 56.53% urban population and 57.76% ethnic minorities. Xinjiang’s gross domestic product (GDP) of 137,975,800,000,000 yuan is mainly dominated by the tertiary service sector, accounting for 51.2% of the total. In 2020, Xinjiang achieved a comprehensive victory in the fight against poverty.

3. Data Sources and Methodology

3.1. Data Sources

This paper incorporated diverse datasets for its research, encompassing LUCC, driving factors, and other data sources. The LUCC dataset was derived from the CLCD dataset from Professor Huang Xin’s team at Wuhan University [28]. Considering the socioeconomics, traffic location, topography, and geomorphology of the Xinjiang region and the study needs, there are 14 driving factor datasets that mainly include climate factors [29], soil factors, topography and geomorphology factors, socioeconomic factors, traffic location factors, and ecosystem factors. Drivers are selected primarily by reference to the principles of accessibility, consistency, and comprehensiveness. The climate factor includes average annual temperature and average annual precipitation; the soil factor is mainly soil organic carbon data; the terrain factor data include elevation (DEM) and slope; socioeconomic factors include population density data, Gross Domestic Product (GDP) data, and luminous light remote sensing data; the traffic location factor includes highway, railway, and waterway data, and the ecosystem factor refers to the driving process simulation of the normalized difference vegetation index. This study considers the local economic situation and planning policies when implementing multi-scenario settings for future land use. It was also necessary to collect data from the Xinjiang Uyghur Autonomous Region Statistical Yearbook, the Xinjiang Uyghur Autonomous Region Land Use Master Plan, and the Xinjiang Uyghur Autonomous Region Territorial Spatial Planning. All the data and sources are shown in

Table 1. Data and data sources.

Type	Data		Spatial Resolution	Source
Land use	LUCC		30 m	The 30 m annual land cover dataset and its dynamics in China from 1990 to 2019 [28]
Driving factor	Climactic	Temperature	1 km	National Tibetan Plateau Science Data Centre [29]
		Precipitation
	Soil	Soil organic carbon content	1 km	National Science and Technology Resources Sharing Service Platform
	Terrain	Elevation	30 m	Geospatial Data Cloud Platform
		Slope
	Socio- Economic	Population density	100 m	WorldPop Grid Population Dataset
		GDP	1 km	Resource and Environment Science and Data Centre, Chinese Academy of Sciences
		Luminous remote sensing	500 m	Earth Observation Group’s VIIRS DNB dataset
	Traffic Location	Distance to road	1 km	National Geographic Information Resources Catalogue Service System
		Distance to railway
		Distance to water system
	Ecosystems	NDVI	1 km	National Science and Technology Resources Sharing Service Platform
Other data	Xinjiang Statistical Yearbook		-	Official Website of National Bureau of Statistics
	Xinjiang Land Use Master Plan
	Xinjiang Territorial Spatial Planning
	2010s China Terrestrial Ecosystem Carbon Density Dataset		-	National Science and Technology Resources Sharing Service Platform

.

3.2. Methodology

This study conducted a spatial and temporal land-use evolution analysis from 1990 to 2020 to determine the characteristics of land-use changes in Xinjiang over the past 30 years. The spatial and temporal distribution characteristics of terrestrial carbon stocks over the past 30 years were simulated based on the InVEST model ensemble using the LUCC and corrected carbon density data. For multiple scenarios, the coupled GMOP and PLUS models predicted spatial and temporal land-use changes and terrestrial carbon stock changes in Xinjiang in 2030. The technology roadmap is shown in Figure 2.

3.2.1. The InVEST Model

The InVEST model, referred to as the Ecosystem Services Valuation and Trade-off Model, was developed by a US Natural Capital Project team to provide an intuitive visual representation of the functional value of ecosystem services over space [30]. The model simulates various LUCC scenarios, analyses changes in ecosystem values, and weighs the impact of people’s production and livelihoods on ecosystems. It can provide a meaningful reference for decision-makers and is widely used in spatial planning, risk assessment, and ecological restoration. The carbon storage and sequestration module in InVEST first estimates four basic carbon pools—aboveground biomass, belowground biomass, soil organic matter, and dead organic matter—and then integrates them to calculate the carbon stock (in which aboveground biomass and belowground biomass are collectively referred to as plant carbon stock). The model requires the input of LUCC and carbon density data and calculates terrestrial carbon stocks using the following expression:

$$C_i=C_{i\_ab}+C_{i\_b}+C_{i\_s}+C_{i\_d}$$

(1)

where C_i represents the integrated carbon density of land class i, and C_{i_ab}, C_{i_b}, C_{i_s}, and C_{i_d} represent the carbon densities of the aboveground biomass, belowground biomass, soil organic matter, and dead organic matter of land class i, respectively.

$$C_{tot}={{\displaystyle \sum }}_{i=1}^n{C}_i\times S_i$$

(2)

where C_tot is the total terrestrial carbon stock in the study area; S_i is the area of land class i, and n is the total number of LUCC types studied.

The carbon densities of different land types in China and Northwest China were obtained through a literature analysis [24,31,32,33,34,35]. Tree species, stand density, and hydrological vegetation cover have a greater impact on biological and soil carbon density [35], and temperature and precipitation are the main factors affecting these factors [36]. Therefore, Tang et al. established the relationship between soil organic carbon and biomass organic carbon and temperature and precipitation in different climates based on 14,371 measured points, showing that carbon density negatively correlated with temperature and positively correlated with precipitation [35]. These results have been confirmed by other researchers and are widely applied in arid regions [32,37,38,39]. Because the main source of carbon density in Xinjiang is the results of the carbon density study in the northwest arid region and even in China, there are some differences from the actual data in the study area, which will have an impact on the simulation results of the carbon stock. Therefore, temperature and precipitation data were selected to correct the initial carbon density and obtain carbon density data for the study area. However, owing to the large spatial differences in precipitation and temperature in Xinjiang, Xinjiang was divided into three parts for correction, i.e., north, south, and east, using the following specific correction algorithms:

(1)

Regression models considering precipitation

When MAT ≤ 10 °C:

$$\left\{\begin{array}{c}{C}_{SP}=79.1+0.07\times MAP\\ C_{BP}=14.4+0.03\times MAP\end{array}\right.$$

(3)

When MAT > 10 °C:

$$\left\{\begin{array}{c}{C}_{SP}=45.3+0.03\times MAP\\ C_{BP}=5.78+0.02\times MAP\end{array}\right.$$

(4)

(2)

Regression model considering air temperature

When MAT ≤ 400 mm:

$$\left\{\begin{array}{c}{C}_{ST}=100-5.8\times MAT\\ C_{BT}=16.7+1.3\times MAT\end{array}\right.$$

(5)

When MAT > 400 mm:

$$\left\{\begin{array}{c}{C}_{ST}=157.7-3.4\times MAT\\ C_{BT}=43.0-0.4\times MAT\end{array}\right.$$

(6)

where C_SP, C_BP, C_ST, and C_BT stand for soil carbon density and biomass carbon density corrected for mean annual precipitation, and soil carbon density and biomass carbon density corrected for mean annual air temperature, respectively; MAT refers to mean annual air temperature (in °C). MAP stands for mean annual precipitation (in mm).

The annual temperature and precipitation in the north, south, and east of Xinjiang and China were substituted into the above equation to solve the correction coefficients for different regions of Xinjiang and calculate the carbon stock density in the north, south, and east of Xinjiang as follows:

$$\left\{\begin{array}{c}{K}_B=Average\left(K_{BP},K_{BT}\right)=Average\left({\displaystyle \frac{C_{BP}^{\prime }}{C_{BP}^{″}}},{\displaystyle \frac{C_{BT}^{\prime }}{C_{BT}^{″}}}\right)\\ K_S=Average\left(K_{SP},K_{ST}\right)=Average\left({\displaystyle \frac{C_{SP}^{\prime }}{C_{SP}^{″}}},{\displaystyle \frac{C_{ST}^{\prime }}{C_{ST}^{″}}}\right)\end{array}\right.$$

(7)

where K_B is the correction factor for biomass carbon density, and K_S is the correction factor for soil carbon density; K_BP and K_BT refer to the correction factor for biomass carbon density for the precipitation factor and temperature factor, respectively; K_SP and K_ST refer to the correction factor for soil carbon density for the precipitation factor and temperature factor, respectively; and C′ and C″ refer to the carbon density data of Xinjiang region (Northern Xinjiang, Southern Xinjiang, and Eastern Xinjiang) and the whole country, respectively. Finally, the carbon density of the study area was determined (

Table 2. Corrected carbon intensity of different land types (t/hm²) *.

Land Type	North Xinjiang			South Xinjiang			Eastern Xinjiang
	Ci_ab	Ci_b	Ci_s	Ci_ab	Ci_b	Ci_s	Ci_ab	Ci_b	Ci_s
Cropland	1.72	2.04	53.63	0.47	0.56	26.14	0.37	0.44	24.69
Forest	18.35	5.42	75.48	5.01	1.48	36.79	3.98	1.18	34.75
Grassland	0.29	2.55	52.89	0.08	0.7	25.78	0.06	0.55	24.35
Water	0.38	0.27	0	0.1	0.07	0	0.08	0.06	0
Snow/Ice	0	0	0	0	0	0	0	0	0
Barren	0.27	0.51	26.99	0.07	0.14	13.15	0.06	0.11	12.42
Impervious	1.34	0.86	0	0.37	0.24	0	0.29	0.19	0
Wetland	1.1	0.88	148.9	0.3	0.24	72.57	0.24	0.19	68.55

* hm stands for hundred meters, and hm² means hundred meters squared, i.e., 10,000 square meters, i.e., 1 hectare.

).

3.2.2. Overview of GMOP Models and Genetic Algorithms

The GMOP model was obtained by combining a grey model with a multi-objective linear programming model. A genetic algorithm was proposed in 1969 by Professor Holland in the United States [40], who mainly modeled the problem as a process of biological evolution. Holland generated the next generation of solutions in different ways and iterated continuously to eliminate solutions with low function values, thus obtaining the optimal solution for the objective function [41]. This method solves the problem of obtaining an optimal solution under a multi-objective function, compensates for the subjectivity of solving the optimal solution using the linear weighting method, and plays an important role in solving the optimal solution of the multi-objective function.

According to the needs of the multi-scenario simulation, this study used the grey prediction model to predict the future economic benefits of the Xinjiang region and combined it with a multi-objective planning model to carry out an optimization study on the quantity structure of land use in the Xinjiang region. The mathematical model can be expressed using Equations (8) and (9):

$$\max \left(\min \right)f_k\left(x\right)={{\displaystyle \sum }}_{i=1}^n{e}_i{x}_i$$

(8)

$$s.t.=\left\{\begin{array}{c}{{\displaystyle \sum }}_{i=1}^n{c}_{ij}{x}_i=\left(\ge ,\le \right)d_i, \left(j=1,2,\cdots ,m\right)\\ x_i\ge 0, \left(i=1,2,\cdots ,n\right)\end{array}\right\}$$

(9)

where k is the number of objective functions; f(x) is the specific objective function; e_i is the coefficient of the i-th decision variable; x_i is the i-th decision variable; c_ij represents the coefficient corresponding to the j-th constraint of decision variable i; d_i is the constraint value of the i-th decision variable, where e_i is mainly predicted by the GM(1,1) model. The relevant policies mainly delineate the constraints.

According to the GMOP model construction to obtain the multi-objective function and the corresponding constraints, the genetic algorithms are used to obtain the optimal solution simultaneously to meet the multi-objective, that is, the Pareto solution set, and then optimize the land-use layout of the Xinjiang region. Currently, most researchers use the NSGA-II algorithm to solve land use multi-objective problems, and its application effect is better [42,43].

The construction of GMOP is mainly based on the current land use situation and development needs of Xinjiang. In this paper, the objective function is established with the objective of maximizing economic, ecological, and comprehensive benefits. Eight decision variables (cropland, forest, grassland, water, snow/ice, barren land, impervious land, and wetland) and 10 constraints (total land area constraints for each land use type, their respective land area constraints and other constraints of the model) are set to generate optimization scenarios for the economic development scenario (ED), the ecological preservation scenario (EP), and the sustainable development scenario (SD).

3.2.3. PLUS Modeling Principles

The PLUS model is a patch-generated land-use model developed by the School of Geomatics and High-Performance Spatial Computational Intelligence Laboratory of the China University of Geosciences (Wuhan). It consists of a Land Expansion Analysis Policy Module (LEAS) and a CA model (CARS) based on multiple types of stochastic patch seeds and threshold-decreasing mechanisms [44]. Compared with other models such as CLUE, CA, and FLUS, the PLUS model is more inclusive of data, does not require higher-quality data, is not overly restrictive in running on large-scale data, has higher model accuracy, is faster, and is more convenient to use.

(1) LEAS framework

LEAS analyzes land-use data for two periods to obtain the probability of development for each land type and the extent to which the driving factors contribute to different land types. The formula is as follows:

$$P_{i,k}^d\left(x\right)={\displaystyle \frac{{{\displaystyle \sum }}_{n=1}^{M}I\left(h_n\left(x\right)=d\right)}{M}}$$

(10)

where $P_{i,k}^d\left(x\right)$ represents the development probability of land type k in cell i; h_n(x) is the prediction type of the decision tree n for vector x; I(∙) represents the indicator function of the set of decision trees; d represents whether or not it is possible to switch from other land types to land type k, and when d = 1 it means yes, while when d = 0 it means no, and M represents the total number of decision trees;

(2) CARS module

The CARS module, based on the development probability of the LEAS module, uses a CA model based on multiclass stochastic seed generation to simulate the spatiotemporal dynamics. The simulation is mainly performed through the adjustment of adaptive coefficients so that the land use meets the future land use quantity demand and, thus, determines the future spatial distribution of land use. The total transformation probability ($O{P}_{i,k}^{d=1,t}$) of land-use type k is calculated as follows:

$$O{P}_{i,k}^{d=1,t}=\left\{\begin{array}{c}{P}_{i,k}^{d=1}\times \left(r\times u_k\right)\times D_k^t if {\Omega }_{i,k}^t=0 and r<P_{i,k}^{d=1}\\ P_{i,k}^{d=1}\times {\Omega }_{i,k}^t\times D_k^t all other\end{array}\right.$$

(11)

where $P_{i,k}^{d=1}$ is the growth probability of land use type k on cell i; u_k is the threshold value of land patch type k; r is a random value ranging from 0 to 1; ${\Omega }_{i,k}^t$ represents the neighborhood effect of cell i of the raster, i.e., the share of land type k under the range of the domain; and $D_k^t$ is the inertia coefficient of land type k at the moment t.

The simulation results of LUCC in 2020 were calibrated quantitatively and spatially using Kappa, OA (Overall accuracy) and Fom (Figure of Merit) values, respectively. The quantitative calibration yields a Kappa value of 0.846 and an OA value of 92.33%, indicating that the accuracy is guaranteed. The Fom value of the spatial calibration is 0.112, and we know from the references that the Fom value of most scholars’ experimental results is within 0.3, with 0.1–0.2 results being the most common [45,46,47]. Therefore, all the tests are within a reasonable range, indicating that the model and parameter settings can be used for future land use prediction simulations.

4. Results and Analyses

4.1. Analysis of the Spatial and Temporal Evolution of Land Use

4.1.1. Changes in the Temporal Pattern of Land Use

In this study, the spatial superposition of four periods of land-use data in the Xinjiang region in 1990, 2000, 2010, and 2020 was used to obtain the land-use transfer matrix (

Table 3. Transfer matrix of land types in Xinjiang region, 1990–2020 (hm²).

1990	2020								Transfer out
	Cropland	Forest	Grassland	Water	Snow/Ice	Barren	Impervious	Wetland
Cropland	-	11,725	768,775	31,925	0	28,500	100,500	150	941,575
Forest	2375	-	225	25	0	0	0	0	2625
Grassland	2,621,050	752,650	-	109,425	105,650	4,411,675	189,525	30,800	8,220,775
Water	15,925	5400	29,575	-	2350	25,125	11,300	75	89,750
Snow/Ice	0	16,200	60,900	37,100	-	617,950	25	0	732,175
Barren	1,281,350	1275	4,711,925	280,850	491,675	-	154,650	75	6,921,800
Impervious	25	0	50	200	0	25	-	0	300
Wetland	2550	175	1825	300	0	125	0	-	4975
transfer in	3,923,275	787,425	5,573,275	459,825	599,675	5,083,400	456,000	31,100	16,913,975

) and Sankey diagram (Figure 3).

We can visualize the number of transfers in and out and the total change in all types of land from 1990 to 2020 in Table 3. The table shows that the total amount of land change over the past 30 years is 1691.40 × 10⁴ hm², which is 10.33% of the area of the entire study area, indicating that the area has been transformed more frequently between land classes in the past 30 years. Regarding a single land category, the net transformation of cropland reached 298.17 × 10⁴ hm²; impervious reached 45.57 × 10⁴ hm², and barren reached 183.84 × 10⁴ hm² over the past 30 years. This phenomenon shows that an increase in population has increased the demand for food and housing, and the opening of agricultural land has taken up a large amount of the barren land, whereas overgrazing and reclamation have degraded the grassland and reduced its area.

The changes in different land uses in each phase were analyzed in terms of land type (Figure 3). The results showed that grassland was always a net transfer out during the three stages, with the net transfer out of land use reaching the largest area in 2010–2020, accounting for 50.03% of the total net transfer out, mainly to barren and cropland. The barren land changed from net transfer out to net transfer in 2010, indicating that the transfer of grassland degradation in this stage caused land desertification, and the barren area increased. Cropland and impervious increased in all stages, in which the net transfer of impervious was 7.35 × 10⁴ hm², 18.40 × 10⁴ hm², and 19.82 × 10⁴ hm² in all stages, respectively. This result indicates that impervious development accelerates and slows down after gradually satisfying the production and life requirements. The net transfer area of cropland at each stage was 48.70 × 10⁴ hm², 156.46 × 10⁴ hm^2, and 93.02 × 10⁴ hm², respectively, indicating that the phenomenon of over-cultivation of cropland is gradually decreasing as the awareness of ecological protection and the demand for cropland is gradually satisfied.

4.1.2. Changes in Spatial Patterns of Land Use

According to the map of current land use in Xinjiang (Figure 4), the region is dominated by barrens and grasslands. This paper analyzes the spatial land changes in Xinjiang by dividing the region into three parts: north; south; and east, according to the differences in temperature and precipitation in Xinjiang (the average annual temperatures are 8.10, 12.95, and 13.39 °C, respectively; the average annual rainfall is 250.45, 80.93 and 29.03 mm, respectively).

The land types on the northern frontier are dominated by grasslands and barrens, which account for more than 80% of the region’s total area. The area of grassland expansion has gradually degraded over the past 30 years, and the forest has grown, mainly interspersed with grassland. It is distributed in the mid-mountainous zones of mountain ranges. The cropland area continued to increase, mainly at the junction of grassland and barren land, and the increase was mainly concentrated at the edge of the Junggar Basin. Impervious continued to expand, concentrating in Urumqi in the southern part of the northern border.

The land type in Southern Xinjiang is predominantly barren, with barren lands accounting for more than 70% of the region’s total area. Over the past 30 years, the region’s grasslands have been seriously degraded with the rise in temperature and the development of cropland. However, the degradation rate has slowed, and the forest area has increased because of ecological and environmental protection policies. This phenomenon indicates that policy implementation led to improvements in the ecological environment. As a result of economic and social development, the area of cropland in the region has continued to grow, and impervious has expanded, indicating that urbanization in the southern border region has been more rapid.

The eastern frontier consists mainly of the Turpan and Hami regions, and barren is the main land category in the region, accounting for a minimum of 87.03% of its total area in 2000. Changes in land types are not as pronounced as those at the northern and southern borders, where the barren area has gradually decreased over the past 30 years and has been replaced by cropland and impervious lands. Regarding spatial distribution, forest land and water areas are steadily increasing. As temperatures rise, snow/ice melts into grasslands and barrens.

4.2. Spatial and Temporal Evolution and Impacts of Terrestrial Carbon Stocks

4.2.1. Temporal and Spatial Patterns of Change in Terrestrial Carbon Stocks

Based on the InVEST model, terrestrial carbon stocks in Xinjiang were simulated at different times to assess changes in the total carbon stocks and carbon stocks in different land categories. According to Figure 5, the overall trend of terrestrial carbon stocks from 1990 to 2020 is increasing by 1.02%. The average annual changes in terrestrial carbon stocks during the three periods were as follows: an increase of 1.63 × 10⁶ t, an increase of 2.74 × 10⁶ t, and a decrease of 0.61 × 10⁶ t, respectively.

In terms of land type, barren was the main carbon pool, accounting for approximately 45% or more of the total carbon stock, followed by grassland (approximately 43%), cropland (approximately 8%), and forest (approximately 4%), among others. Although the carbon density of barren lands is lower than that of grasslands and croplands, and the carbon sequestration capacity is weaker, barren lands are the main land category in Xinjiang, covering a larger and wider area. They are, thus, a vital carbon pool in the region. The overall change in the carbon stock ratio structure of each land category from 1990 to 2020 was not significant, and the changes in the carbon stock ratios of cropland, forest, and grassland increased by 3.22, 1.92, and 3.98 percentage points, respectively. Cropland carbon stock increased significantly from 2000 to 2010, increasing by 67.09 × 10⁶ t. The carbon stock in forest land increased most significantly from 1990 to 2000, with an increase of 38.92 × 10⁶ t. The decrease in carbon stock in grassland was mainly concentrated from 2010 to 2020, with a decline of 62.72 × 10⁶ t.

Local autocorrelation analysis was performed using GeoDA V1.14.0.0 software to obtain the local Moran’s I scatter plot, and a LISA clustering map of the terrestrial carbon stock in Xinjiang was produced using ArcGIS 10.6 and GeoDA V1.14.0.0 software (Figure 6). The spatial positive autocorrelation of land carbon stocks in Xinjiang was concentrated in the north, whereas the positive autocorrelation of low-carbon stocks was focused in the south. From 1990 to 2020, the number of “high–high” clusters decreased by 24, mainly in the eastern part of the northern border, and most of them were converted to “insignificant” types. The number of “low–low” clusters decreases by 81, mainly in the southern and eastern parts of the southern border, and most of them are transformed into “insignificant” and “high–low” types. The types of “low–high” clusters remain unchanged and are mainly concentrated in the central Xinjiang region. The number of “high-low” aggregation types increased from 0 to 16, most converted from “low–low” types. This indicates that the carbon stock increased in some grids, which was higher than the value of the terrestrial carbon stock in the neighboring regions. The spatial aggregation of terrestrial carbon stocks in Xinjiang over the past 30 years showed large spatial differences between north and south; however, the overall characteristics were relatively stable.

4.2.2. Impacts of Land-Use Change on Terrestrial Carbon Stocks

Due to the differences in the transfer area, soil carbon density, and vegetation carbon density among land types, there are differences in the number of carbon stock increases and decreases caused by changes in different land uses, and the specific terrestrial carbon stock changes are shown in

Table 4. Changes in carbon stocks due to land use change in Xinjiang, 1990–2020.

Change in LUCC		Change in Area		Change in Carbon Stock (×104 t)
Transfer out	Transfer in	Area (hm2)	Percentage (%)	C_VEG	C_SOIL	Subtotal
Cropland	Forest	11,725.00	5.57	21.46	24.08	−653.39
	Grassland	768,750.00		−46.82	−43.80
	Water	31,925.00		−7.53	−141.92
	Barren	28,500.00		−4.99	−53.88
	Impervious	100,500.00		−9.53	−391.38
	Wetland	150.00		−0.01	0.94
Forest	Cropland	2375.00	0.02	−4.42	−4.93	−10.44
	Grassland	225.00		−0.39	−0.45
	Water	25.00		−0.06	−0.19
	Barren	0.00		0.00	0.00
	Impervious	0.00		0.00	0.00
	Wetland	0.00		0.00	0.00
Grassland	Cropland	2,620,975.00	48.60	155.76	145.57	−5757.84
	Forest	752,650.00		1416.54	1579.76
	Water	109,425.00		−14.08	−409.37
	Snow/Ice	105,550.00		−17.30	−391.48
	Barren	4,409,600.00		−437.98	−7244.94
	Impervious	189,525.00		−6.51	−678.53
	Wetland	30,800.00		−0.76	145.47
Water	Cropland	15,925.00	0.53	2.58	56.40	246.98
	Forest	5400.00		10.21	35.53
	Grassland	29,575.00		2.97	96.12
	Snow/Ice	2350.00		−0.06	0.00
	Barren	25,125.00		0.16	41.63
	Impervious	11,300.00		0.89	0.00
	Wetland	75.00		0.00	0.54
Snow/Ice	Forest	16,200.00	4.32	38.16	121.50	1360.27
	Grassland	60,775.00		8.86	210.93
	Water	37,050.00		1.32	0.00
	Barren	616,650.00		19.10	960.39
	Impervious	25.00		0.01	0.00
Barren	Cropland	1,281,325.00	40.93	211.70	2340.76	8586.12
	Forest	1275.00		2.25	5.18
	Grassland	4,711,425.00		400.57	7136.57
	Water	280,850.00		−1.79	−471.90
	Snow/Ice	490,925.00		−14.36	−744.09
	Impervious	154,650.00		12.67	−292.21
	Wetland	75.00		0.01	0.76
Impervious	Cropland	25.00	0.00	0.00	0.07	0.21
	Grassland	50.00		0.00	0.13
	Water	200.00		−0.01	0.00
	Barren	25.00		0.00	0.03
Wetland	Cropland	2550.00	0.03	0.12	−11.84	−24.71
	Forest	175.00		0.10	−0.63
	Grassland	1825.00		0.05	−9.28
	Water	300.00		−0.01	−2.18
	Barren	125.00		−0.01	−1.06
Total		16,908,950.00	100.00	1738.87	2008.33	3747.20

.

The total terrestrial carbon stock increased by 3747.42 × 10⁴ t from 1990 to 2020, of which the vegetation carbon stock increased by 1738.87 × 10⁴ t and soil carbon stock increased by 2008.33 × 10⁴ t. The increase in the total carbon stock is mainly due to the conversion of a large barren area to a more carbon-dense grassland and cropland, which enhances the carbon sequestration capacity of the area and facilitates carbon sinks. The area of barrens transferred out accounted for 40.93% of the total number of lands transferred out, and the transfer of barrens increased the total terrestrial carbon stock by 8586.12 × 10⁴ t, with the main source of the increase being the soil carbon stock. The main reason for decreased carbon stocks over the last 30 years was the change in grassland transfer, which accounted for 48.46% of the total transfer. Grassland degradation reduced the total terrestrial carbon stock by 5757.84 × 10⁴ t, of which the vegetation carbon stock increased by 1095.68 × 10⁴ t, but the soil carbon stock decreased by 6853.52 × 10⁴ t. This degradation was mainly due to converting grasslands to barren, impervious, and water lands with lower carbon densities. Although the transfer of land with higher carbon density increased the carbon stock, the increase was much smaller than the decrease, resulting in an overall decrease in the grassland carbon stock. Negative ecological evolution, such as converting grassland and forest to barren and impervious, was more common in Xinjiang from 1990 to 2020. However, there were also positive ecological evolutions, such as the conversion of barren to grassland and cropland, which increased the total carbon stock by 7537.14 × 10⁴ t and 2552.46 × 10⁴ t, respectively, and the total terrestrial carbon stock in the region showed an overall growth trend.

4.3. Multi-Scenario Modeling of Terrestrial Carbon Stocks

4.3.1. Multi-Scenario Land-USE Change

The spatial distribution of land use and the area of each land-use category in Xinjiang under different scenarios were obtained using the PLUS and GMOP models (Figure 7).

In the Inertia Development (BS) scenario, an increase in cropland, grassland degradation, and land desertification still occurred. Compared to 2020, the BS scenario LUCC in 2030 shows an increase of 716,950 hm² of cropland, mainly distributed in the southern part of the northern border and northern part of the southern border, where grassland and barrens are the main land types. Impervious area increased by 52,050 hm², mainly occupying grassland and cropland, and expanded from 2020 impervious area.

In the Economic Development (ED) scenario, the optimized design focuses on the Economic Development of the economy. As a result, impervious areas expand at an accelerated rate of 70% in the ED scenario. The impervious area in this scenario expanded in all directions compared to the original BS scenario, with barren lands and grasslands being the main sources of expansion. The rate of grassland degradation decreased in the ED scenario, with the rate of change decreasing from 3.4% to 2.7%, mainly because of the reduction in grassland occupation because of barren development.

The Ecological Protection (EP) scenario focuses on ecological development and protection. The cropland and impervious areas were consistent with those in 2020, and the non-increase in cropland greatly improved grassland degradation, with its degradation rate decreasing from 3.4% to 0.13%. In this scenario, the reduction in grassland degradation contributes to the weakening of land desertification, and no expansion of barren areas occurs. These phenomena indicated that the ecosystem was protected under this scenario.

A Sustainable Development (SD) scenario is the result of land-use optimization, which aims to reconcile urban sprawl with ecological conservation. Cropland in this scenario expanded slower than in the BS scenario, with a growth rate of 0.57%. Impervious increased with the BS scenario, but the area decreased with the ED scenario. This indicates that this scenario protects and restores ecological land while satisfying impervious development.

The BS and ED scenarios show a faster expansion of cropland and forest compared to the EP and SD scenarios, with the area expanding by 7 × 10⁵ hm² and 3 × 10⁴ hm². The fastest expansion of grassland is in the EP scenario, followed by the SD scenario, which proves that the EP and SD scenarios have a better role in ecological protection. The ED and SD scenarios have a faster expansion of impervious than the BS and EP scenarios, indicating that economic development is enhanced in these two scenarios. Through the above comparison between the scenarios, it is found that the SD scenario satisfies the rapid economic development and also plays a positive role in ecological protection.

4.3.2. Multi-Scenario Terrestrial Carbon Stock Changes

Based on the multi-scenario LUCC data above, we imported different scenario land data for 2030 into the carbon module of the InVEST model to simulate changes in terrestrial carbon stocks under the four future scenarios (Figure 8).

The total amount of terrestrial carbon stock in Xinjiang under the BS scenario in 2030 is 369,476.56 × 10⁴ t, an increase of 7.61 × 10⁴ t compared with the terrestrial carbon stock in 2020. The carbon stocks of barrens, grasslands, and croplands dominated the terrestrial carbon stocks in the study area. The terrestrial carbon stocks of cropland and barren in the BS scenario in 2030 increased compared with those in 2020, with an increase of 3075.66 × 10⁴ t and 432.03 × 10⁴ t, respectively. In contrast, the terrestrial carbon stock of grassland decreased by 4444.96 × 10⁴ t. This change was mainly caused by the continuation of the rate and direction of land type change from 2010–2020 and the continued degradation of grasslands with higher carbon sequestration capacity. Therefore, the carbon sequestration capacity in the BS scenario will be stronger in 2030 than in 2020.

The total terrestrial carbon stock under the ED scenario 2030 is 339,199.59 × 10⁴ t, the smallest among the four scenarios, and 269.36 × 10⁴ t lower than that in 2020. The ED scenario for 2030 needs to vigorously increase impervious and cropland to maximize economic benefits; therefore, a large amount of grassland is occupied by cropland and impervious, reducing the grassland area and weakening carbon sequestration capacity. Although the carbon density of cropland is greater than that of grassland, the increase in cropland is smaller, and the decrease in grassland area is greater than in cropland area; therefore, the carbon stock is weakened in this scenario.

The total terrestrial carbon stock under the EP scenario 2030 is 369,717.78 × 10⁴ t. Compared to the stock in 2020, the terrestrial carbon stock under this scenario increases by 248.83 × 10⁴ t, mainly due to the protection of ecological lands, such as forests and grasslands, by restricting the expansion of impervious. The carbon stocks of cropland, impervious land, and barren land obtained from the simulation of the EP scenario are consistent with those of 2020. This phenomenon indicates no impervious and cropland expansion under this scenario and that the barren no longer expands and degrades. Analyzing the single land category, the carbon stock value of grassland under the EP scenario reaches a maximum value of 147,464.58 × 10⁴ t under the four scenarios, accounting for 39.86% of the total carbon stock. The carbon stock of forest under this scenario is 17,472.35 × 10⁴ t, an increase of 549.14 × 10⁴ t compared with 2020.

The 2030 SD scenario simulates the spatial structure of land use to meet the maximum sum of future land use’s ecological and economic benefits. The carbon stock of the forest under this scenario is 17,269.92 × 10⁴ t, which is under the EP scenario and higher than the BS and ED scenarios. And the carbon stock of impervious is 115.29 × 10⁴ t, which is the highest among the four scenarios. It is demonstrated that the SD scenario satisfies both ecological and economic maximization.

The total terrestrial carbon stock under the SD scenario was 369,259.16 × 10⁴ t, and the value of the carbon stock was between those of the ED and EP scenarios. Compared with the carbon stock in 2020, the carbon stock is reduced by 209.79 × 10⁴ t, but the reduction is smaller than the value of carbon stock reduction from 2010 to 2020. This indicates that when ensuring that the sum of ecological and economic benefits is maximized, the direction of land development moves towards low-carbon and green development, although the terrestrial carbon stock decreases.

5. Discussion

Xinjiang is a northwestern arid region, and this paper focuses on “land use change—spatial and temporal evolution of terrestrial carbon stocks—scenario modeling—policy recommendations”. First, the spatial and temporal evolution of land use in the four periods from 1990 to 2020 was analyzed using geospatial analysis. Second, terrestrial carbon stocks were simulated based on the corrected carbon density using the InVEST model, and the spatial relationship of carbon stocks was investigated using the spatial autocorrelation analysis method. The impacts of the LUCC on terrestrial carbon stocks were also analyzed by combining the transfer in and out of the LUCC. The GMOP and PLUS models were coupled to simulate future land use in the study area. Based on the simulation results, we analyzed the land-use changes in multiple scenarios and the changes in terrestrial carbon stocks under different scenarios. Furthermore, we provided policy advice for future land-use planning in Xinjiang.

Comparative analyses of this study’s data results suggest that the above land use and carbon stock changes in Xinjiang are due to the following reasons: (1) The increase in population over the past 30 years has increased the demand for food and housing, the degradation of grasslands, and a significant reduction in grasslands due to the clearing of cropland and overgrazing by human beings; (2) With the awareness of ecological protection and the demand for cropland gradually being satisfied, the phenomenon of excessive agricultural cultivation has gradually reduced. Since the 13th Five-Year Plan, the ecological problems in Xinjiang have become the focus of the government’s attention; (3) The autonomous region uses remote sensing image data to carry out a comprehensive evaluation of the ecological environment status of the whole territory and key regions and watersheds and carries out the governance of the mountain, water, forest, field, lake, grass, and sand systems and the protection of the ecological environment in the whole region through investigation, monitoring, and assessment [27].

From the results of the multi-scenario simulation, we found that the Sustainable Development (SD) scenario results from land use optimization that maximizes economic and ecological benefits under certain conditions. The main reasons for this result are as follows: (1) The expansion of cropland in this scenario was slower than that in the BS scenario, and the area of impervious increased compared to the BS scenario and decreased compared to the ED scenario; (2) Compared to 2020, the areas of forest, water, and wetland increased, indicating that this scenario protected and restored ecological land while meeting the development of impervious. Therefore, land use in the SD scenario meets the needs of low-carbon development while considering the economy and ecology, meeting the requirements of the Xinjiang regional master plan, and providing certain reference suggestions for land planning in the region.

Many researchers have studied land use, carbon stock, and multi-scenario modeling [48,49,50,51,52,53,54]. Yang et al. analyzed the spatiotemporal change pattern of land use patterns and its driving mechanism using the Google Cloud Computing Platform and GeoProbe [50]. Rolinski et al. simulated the carbon balance from 1900 to 2100 using a dynamic vegetation model to assess the impacts of observed land use changes and future climate and land use change scenarios [51]. Chen et al. simulated and analyzed the land use of the Raohe River Basin in 2035 based on the Markov-FLUS coupled model to meet the optimal allocation of national land space under the orientation of different development goals in the basin [54]. Many researchers have studied and analyzed carbon stocks in various regions of Xinjiang [55]. For example, Fu et al. used the coupled PLUS-InVEST model to assess changes in land use and carbon stock in the Tarim River Basin from 1980 to 2020, predicted the trends of land use and carbon stock in the study area by 2030, and explored the impacts of land use changes on carbon stocks [55]. Using the carbon stock module of the InVEST model, Lu et al. estimated the carbon stock and its spatial distribution pattern in the Tianshan Mountains of Xinjiang from 1990 to 2020 and analyzed the impact of land use changes on carbon stock [32]. However, previous studies have analyzed carbon stocks in the Xinjiang region, and few researchers have combined the three modules of land use, carbon stock, and multi-scenario simulation. Therefore, based on previous studies, this study analyses the results in more detail by considering the entire Xinjiang region as the study area and concludes that the sustainable development (SD) scenario best meets the requirements of the Xinjiang regional master plan. Fu et al. concluded that the ecological protection (EP) scenario could significantly enhance the carbon stock in Xinjiang more than other scenarios, which is consistent with the findings of this paper [56]. In this study, the sustainable development (SD) scenario maximizes both ecological and economic benefits, so the SD scenario is most consistent with the Xinjiang master plan.

This study’s revised carbon density values of the Xinjiang region are more reasonable and reliable than terrestrial carbon stocks simulated directly using national carbon density data. By simulating land use under different scenarios, our quantitative and visual analyses combined with terrestrial carbon stocks provide a reference for ecological, environmental protection, and sustainable development in Xinjiang. However, because sampling data consumes a lot of time and personnel, most of the carbon density data in the study area are summarized from previous study results and sampling results and then corrected according to climate and precipitation. This study divided Xinjiang into north, south, and east Xinjiang according to topography and landscape to carry out the simulation, improving the accuracy while calculating carbon stocks in the large-scale region. If other researchers are not limited by time and personnel, they can consider combining field surveys and long-term monitoring of carbon density over different periods and land types to improve the results of terrestrial carbon stock assessments. As a next step, the different scenarios can also be studied and compared two by two to find the optimal scenario that best suits Xinjiang.

6. Conclusions

This study analyzed the spatial and temporal evolution of land use in four periods from 1990 to 2020, explored the spatial relationship of carbon stocks using the InVEST model, and coupled the GMOP model with the PLUS model to carry out multiple scenarios for the future simulation of land use in the study area. The main conclusions of this study are as follows:

(1) The spatial and temporal changes in the LUCC in Xinjiang from 1990 to 2020 were remarkable. These changes are dominated by the rapid expansion of impervious and grassland degradation and barren land exploitation. The growth rates of impervious and cropland were 1093.46% and 52.75%, respectively, and the proportions of grassland and barren areas decreased by 1.62% and 1.4%, respectively. The LUCC change from 2000 to 2010 was relatively flat, while the LUCC change from 2010 to 2020 was larger, mainly caused by the active change in the LUCC due to the mutual transformation of each category after 2010;

(2) Terrestrial carbon stocks in Xinjiang increased and decreased, with significant differences between the north and south. Modeling the terrestrial ecosystems in Xinjiang through modified carbon density in the region revealed that the terrestrial carbon stock showed an increasing trend from 1990 to 2010 and a decreasing trend from 2010 to 2020. The spatial autocorrelation analysis shows a large spatial variability of carbon stocks in Xinjiang, with the northern part of the region dominated by “high–high” aggregation and the southern part dominated by “low–low” aggregation;

(3) The terrestrial carbon stock decreased in the sustainable development scenario, but the sequestration potential increased compared to the previous stage. The projections show that in 2030, impervious areas will continue to grow, and the grassland areas will still be degraded, but the degradation rate will decrease. Although the total terrestrial carbon stock decreased under this scenario, the decrease was less significant than that observed in the previous phase.

Through the analysis of the spatial and temporal evolution of terrestrial carbon stocks and multi-scenario simulation in this study, it was found that the land use of the sustainable development scenario considers the economy and ecology, meets the needs of low-carbon development, and meets the requirements of the Xinjiang regional master plan. Therefore, Xinjiang should aim to maximize economic and ecological benefits in its future development and achieve sustainable development through the strategy of optimizing oases and stabilizing deserts.